LINGUISTICS AND THE FORMAL SCIENCES

The formal sciences, particularly mathematics, have had a profound influence

on the development of linguistics. This insightful overview looks at techniques
that were introduced in the fields of mathematics, logic, and philosophy during
the twentieth century, and explores their effect on the work of various linguists.
In particular, it discusses the foundations crisis that destabilised mathematics
at the start of the twentieth century, the numerous related movements which
sought to respond to this crisis, and how they influenced the development
of syntactic theory in the 1950s. This book provides a ground-breaking and
detailed reassessment of Chomsky’s early work, and concludes by discussing
the resulting major consequences for current syntactic theory. Informative
and revealing, this book will be invaluable to all those working in formal
linguistics, in particular those interested in its history and development.

m a r c u s t o m a l i n is a Fellow of Downing College, University of

Cambridge. His academic interests are wide-ranging and include syntactic
theory, the history of linguistics, mathematical models of linguistic theory
development, and the modelling of syntactic structure in large vocabulary
speech recognition systems. He publishes regularly on these diverse topics in
various international journals.
CAMBRIDGE STUDIES IN LINGUISTICS
General Editors: p. a u s t i n, j . b r e s n a n, b. c o m r i e ,
s . c r a i n, w. d r e s s l e r , c . j . e w e n, r . l a s s ,
d . l i g h t f o o t, k . r i c e , i . r o b e r t s , s . r o m a i n e ,
n. v. s m i t h

Linguistics and the Formal Sciences

The Origins of Generative Grammar
 In this series
71 k n u d l a m b r e c h t: Information structure and sentence form: topic, focus, and the
mental representation of discourse referents
72 l u i g i bu r z i o: Principles of English stress
73 j o h n a . h aw k i n s: A performance theory of order and constituency
74 a l i c e c . h a r r i s and ly l e c a m p b e l l: Historical syntax in cross-linguistic
perspective
75 l i l i a n e h a e g e m a n: The syntax of negation
76 p a u l g o r r e l: Syntax and parsing
77 g u g l i e l m o c i n q u e: Italian syntax and universal grammar
78 h e n r y s m i t h: Restrictiveness in case theory
79 d . r o b e r t l a d d: Intonational morphology
80 a n d r e a m o r o: The raising of predicates: predicative noun phrases and the theory
of clause structure
81 r o g e r l a s s: Historical linguistics and language change
82 j o h n m . a n d e r s o n: A notional theory of syntactic categories
83 b e r n d h e i n e: Possession: cognitive sources, forces and grammaticalization
84 n o m t e r t e s c h i k - s h i r: The dynamics of focus structure
85 j o h n c o l e m a n: Phonological representations: their names, forms and powers
86 c h r i s t i n a y. b e t h i n: Slavic prosody: language change and phonological theory
87 b a r b a r a d a n c y g i e r: Conditionals and prediction
88 c l a i r e l e f e b v r e: Creole genesis and the acquisition of grammar: the case of
Haitian creole
89 h e i n z g i e g e r i c h: Lexical strata in English
90 k e r e n r i c e: Morpheme order and semantic scope
91 a p r i l m c m a h o n: Lexical phonology and the history of English
92 m at t h e w y. c h e n: Tone Sandhi: patterns across Chinese dialects
93 g r e g o r y t. s t u m p: Inflectional morphology: a theory of paradigm structure
94 j o a n b y b e e: Phonology and language use
95 l a u r i e b a u e r: Morphological productivity
96 t h o m a s e r n s t: The syntax of adjuncts
97 e l i z a b e t h c l o s s t r a u g o t t and r i c h a r d b. d a s h e r: Regularity in
semantic change
98 m aya h i c k m a n n: Children’s discourse: Person, space and time across languages
99 d i a n e b l a k e m o r e: Relevance and linguistic meaning: The semantics and
pragmatics of discourse markers
100 i a n r o b e r t s and a n n a r o u s s o u: Syntactic change: a minimalist approach to
grammaticalization
101 d o n k a m i n k o va: Alliteration and sound change in early English
102 m a r k c . b a k e r: Lexical categories: verbs, nouns and adjectives
103 c a r l o ta s . s m i t h: Modes of discourse: the local structure of texts
104 r o c h e l l e l i e b e r: Morphology and lexical semantics
105 h o l g e r d i e s s e l: The acquisition of complex sentences
106 s h a r o n i n k e l a s and c h e r y l z o l l: Reduplication: doubling in morphology
107 s u s a n e d wa r d s: Fluent aphasia
108 b a r b a r a d a n c y g i e r and e v e s w e e t s e r: Mental spaces in grammar:
conditional constructions
109 m at t h e w b a e r m a n, d u n s ta n b r ow n and g r e v i l l e g . c o r b e t t: The
syntaxmorphology interface: a study of syncretism
110 m a r c u s t o m a l i n: Linguistics and the formal sciences: The origins of generative
grammar
Earlier issues not listed are also available
LINGUISTICS AND THE
FORMAL SCIENCES
THE ORIGINS OF GENERATIVE
GRAMMAR

MARCUS TOMALIN
Downing College, Cambridge
  
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo

Cambridge University Press

The Edinburgh Building, Cambridge  , UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521854818

© Marcus Tomalin 2006

This publication is in copyright. Subject to statutory exception and to the provision of

relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.

First published in print format 2006

- ---- eBook (NetLibrary)

- --- eBook (NetLibrary)

- ---- hardback

- --- hardback

Cambridge University Press has no responsibility for the persistence or accuracy of s
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
since feeling is first
who pays any attention
to the syntax of things
will never wholly kiss you;
e. e. cummings
 Contents

Acknowledgments page xi
List of mathematical symbols xii
List of abbreviations xiii

1 Introduction 1
2 The consequences of analysis 21
2.1 Chapter overview 21
2.2 Calculus: doubts and disputes 22
2.3 Rigour, arithmetic, and axioms 26
2.4 Set theory and paradoxes 29
2.5 Logicism 32
2.6 Formalism 38
2.7 Intuitionism 45
2.8 Evangelism and pedagogy 47

3 Mathematical linguistics 54
3.1 Chapter overview 54
3.2 Axiomatics 55
3.3 Recursive definitions 60
3.4 Logical systems 67
3.5 Constructional system theory 73
3.6 Constructive nominalism 84
3.7 Formal linguistic theory 88
3.8 New directions 106

4 Systems of syntax: 1951–1955 108

4.1 Chapter overview 108
4.2 Biography and influences 109
4.3 Simplicity and grammar 112
4.4 Constructive nominalist syntax 121
4.5 Logic and linguistic theory 125

ix
x Contents

5 Transforming generative grammar: 1955–1957 140

5.1 Chapter overview 140
5.2 Stochastic processes and autonomous grammar 140
5.3 From discovery to evaluation 149
5.4 Constructional levels 156
5.5 Transforming transformations 159
5.6 Recursive rules 168
5.7 Formal syntax 174

6 Conclusion 183

Notes 201
Bibliography 209
Index 220
 Acknowledgments

I began tentatively to explore some of the topics presented here in November

1995, and, over the intervening years, the central issues have become increas-
ingly distinct. As usual, this process of gradual clarification has involved numer-
ous people, many of whom cannot be mentioned here. Explicit thanks, though,
are due to Peter Matthews and Ian Roberts, who encouraged me to pursue this
research at an early stage. For similar reasons, I owe a debt of gratitude to
Neil Smith and Ted Briscoe, whose comments and advice were of significant
value. More recently, I have benefited greatly from discussions with Theresa
Biberauer and Fiorien Bonthuis: over the past two years I have repeatedly sub-
jected them both to countless ill-formed questions and incoherent musings,
and they have invariably responded with humour, intelligence, and acuity. Less
obviously, perhaps, the disparate members of the Cambridge Syntax Reading
Group should also be mentioned, since they have contributed to this project
considerably, without necessarily realising that they were assisting. In addition,
I have been enriched by my connections with various members of Downing
College, especially Cathy Phillips, who has watched over me for many years
now, and Will Poole, whose restless polymathic brilliance always leads me
towards deeper insight.
Obviously, I am indebted to Cambridge University Press for agreeing to
publish this book. In particular, Andrew Winnard has been involved in the
process of manuscript preparation from the very beginning; he has been my
guide throughout.
Heterogeneously, I acknowledge Phil Woodland, for offering me so many
opportunities; Gary Thorne, for inexpressible friendship and inspiration; and
members of my various families, both the living and the dead, for laughter,
confusion, and love. Finally, though, last and most, my accomplice, Sarah, for
everything.
While it is hoped that all solecisms, inexactitudes, and stupidities have been
unremittingly excised from this text, the remaining errors exist to remind us of
our impefections.
This book is dedicated to Arbutus Cove, Victoria B.C., where, turning the
corner, I glimpsed the sea.
xi
 Mathematical symbols

‘∀’ indicates universal quantification: ‘∀x[φ(x)]’ means ‘for all x it is the case
that φ(x) holds’.
‘∃’ indicates existential quantification: ‘∃x[φ(x)]’ means ‘there exists an x for
which φ(x) holds’.
‘¬’ indicates negation: ‘¬φ(x)’ means ‘it is not the case that φ(x) holds’.
‘→’ indicates implication: ‘ p → q’ means ‘if p, then q’.
‘∧’ indicates conjunction: ‘ p ∧ q’ means ‘ p and q’.
‘∨’ indicates disjunction: ‘ p ∨ q’ means ‘ p or q’.
‘N’ indicates the set of natural numbers: N = {1, 2, 3, ...}.
‘∈’ indicates ‘is a member of’: ‘x ∈ A’ means ‘x is a member of set A’.
‘∈’
/ indicates ‘is not a member of’: ‘x ∈
/ A’ means ‘x is not a member of set
A’.

xii
 Abbreviations

Books/Book-length Manuscripts/Theses

LCW Carnap, R. (1928), Der logische Aufbau der Welt [The Logical
Construction of the World]
LPV Quine, W. V. O. (1953), From a Logical Point of View
LSL Carnap, R. (1937[1934]), The Logical Syntax of Language
LSLT Chomsky, N. (1975[1955]), The Logical Structure of Linguistic
Theory
MMH Chomsky, N. (1979b[1951]), Morphophonemics of Modern Hebrew
MP Chomsky, N. (1995), The Minimalist Program
MSL Harris, Z. S. (1951), Methods in Structural Linguistics
PM Whitehead, A. N. and Russell, B. A. W. (1925[1910]), Principia
Mathematica [The Principles of Mathematics]
SA Goodman, N. (1951), The Structure of Appearance
SS Chomsky, N. (1957), Syntactic Structures

Papers

‘LSS’ Chomsky, N. (1955a), ‘Logical Syntax and Semantics: Their

Linguistic Relevance’
‘SCN’ Goodman, N. and Quine, W. V. O. (1947), ‘Steps Towards a
Constructive Nominalism’
‘SSA’ Chomsky, N. (1953), ‘Systems of Syntactic Analysis’
‘TMDL’ Chomsky, N. (1956), ‘Three Models for the Description of
Language’

xiii
1 Introduction

The emergence of Transformational Generative Grammar (TGG) in the 1950s

is an event in the history of linguistics that has been recounted many times, in
many different ways, by many different people, and since this book is primarily
concerned with the development of TGG, some sort of apologia is required in
order to justify retelling the same story yet again. Accordingly, it is hoped that
this introduction will provide the requisite justification, and, in summary expla-
nation, it can be stated at the outset that the main motivation for the particular
narration offered here is dissatisfaction – specifically, dissatisfaction stemming
from the conviction that none of the existing versions of the TGG narrative
provide sufficient information concerning the influence of contemporaneous
advances in the formal sciences upon the development of linguistic theory in
the twentieth century. If indeed it is the case that this aspect of TGG history has
been neglected in the past, then this neglect is certainly surprising, since the
earliest proponents of TGG have never disguised the fact that the theory derived
considerable inspiration from the formal sciences. For instance, in 1995 (to con-
sider just one example) Chomsky stated explicitly that ‘[g]enerative grammar
can be regarded as a kind of confluence of long-forgotten concerns of the study
of language and mind, and new understanding provided by the formal sciences’
(Chomsky 1995: 4), and the scientific nature of TGG itself has often been noted
over the years. Indeed, it has sometimes been claimed that TGG was a supe-
rior linguistic theory primarily because it was more ‘scientific’ (whatever that
means) than the syntactic theories that preceded it. For instance, this was the
basic claim made by Robert Lees in his influential 1957 review of Syntactic
Structures, the text that was partly responsible for inculcating a widespread
interest in TGG in the late 1950s. Although Lees’ review is well known, it is
worth quoting the relevant passage in full:
Chomsky’s book on syntactic structures is one of the first serious attempts
on the part of a linguist to construct within the tradition of scientific theory-
construction a comprehensive theory of language which may be understood
in the same sense that a chemical, biological theory is ordinarily understood

1
2 Introduction

by experts in those fields. It is not a mere reorganisation of the data into a new
kind of library catalogue, nor another speculative philosophy about the nature
of Man and Language, but rather a rigorous explication of our intuitions about
our language in terms of an overt axiom system, the theorems derivable from
it, explicit results which may be compared with new data and other intuitions,
all based plainly on an overt theory of the internal structure of languages; and
it may well provide an opportunity for the application of explicit measures of
simplicity to decide preference of one form over another form of grammar.
(Lees 1957: 377–378)

It is significant that Lees’ assessment of TGG’s status as a scientific theory

concentrates upon its use of ‘an overt axiom system’, suggesting that it is
this aspect of the approach that enables the linguistic theory developed to be
‘rigorous’. The implication here is that, before TGG, linguistic theories did
not use this kind of methodology (or at least not so extensively), and that they
provided instead either ‘a mere reorganization of the data’, or else a ‘speculative
philosophy’ which (presumably) could not be validated empirically. Clearly,
then, TGG was perceived by some of its first adherents to be a more scientific
theory than its predecessors partly because it employed the axiomatic-deductive
method and so obtained ‘explicit results’, and, given this perception, it is curious
that the relationship between TGG and the formal sciences that influenced it
has not been explored more fully.
Before assessing the existing historiographical studies of TGG, though, it
is necessary to clarify some of the terminology used above. For instance, the
phrase ‘formal sciences’ has already been deployed several times without com-
ment and, since it will haunt the ensuing discussion, an initial consideration of
its meaning is already overdue. In particular, it should be confessed from the
outset that, in the following chapters, this phrase will be used in a rather broad
sense. As expected, it will be understood to include various branches of pure
mathematics and symbolic logic, but, in addition, it will also be stretched to
include various kinds of applied mathematics and logic – an extension that is not
free from controversy.1 For example, specific attempts to use the methodology
of classical logic to analyse processes of knowledge acquisition (e.g., logical
empiricism) will be considered as instances of ‘formal science’, as will efforts to
explore the logical relationship between language and the real world (e.g., truth-
theoretical semantics). Consequently, the phrase will come to denote a rather
heterogeneous collection of related theories drawn mainly from mathematics
and philosophy, and, at times, it may seem as if it denotes so much that ulti-
mately it denotes nothing at all. However, it is important to recognise that the
theories grouped together beneath this umbrella term all utilise some form of the
 Introduction 3

axiomatic-deductive method and that, therefore, despite their many differences,

they all involve the deduction of consequences (i.e., theorems) from a small set
of intuitively obvious axioms or assumptions, and, as a result, they can be
viewed as being unified by the same basic scientific method. In the light of
this observation, it should be remembered that not all intellectual enterprises
(specifically, not even all sciences) can be pursued by means of this method. In
order for an axiomatic-deductive system to be constructed at all, it is necessary
to be able to state initial assumptions, to identify primary elements of some
kind, and to make valid deductive inferences from these assumptions and ele-
ments. There are many areas of research that are not understood with sufficient
precision to permit an axiomatic-deductive analysis. However, the ‘formal sci-
ences’ all attempt to utilise this methodology, and it is one of their characteristic
features. Other similarities will be mentioned as and when they become relevant
to the discussion sustained in the rest of this book, but, from henceforth, the
cautionary quotation marks will be dropped, enabling the ‘formal sciences’ to
become merely the formal sciences.
Another terminological problem that was avoided in the first couple of para-
graphs concerns the phrase ‘Transformational Generative Grammar’, and since
this phrase appears frequently in this book, it is imperative that its meaning
is clear. It is already the case (in the foregoing discussion of Lees’ review)
that ‘TGG’ has been referred to as if it were a single, identifiable, unchang-
ing, monolithic entity. However, if the work presented in the following chapters
demonstrates anything, it demonstrates that it is alarmingly anachronistic to use
the term ‘TGG’ broadly to refer to the various theoretical stances and projects
associated with Noam Chomsky (b. 1928) during the years c. 1951–c. 1956,
since, although the various lines of research eventually fused into an identi-
fiable theory, a coherent picture did not begin to emerge until c. 1957, when
Chomsky began to draw together his ideas concerning the structure of linguistic
theory, the validity (or otherwise) of statistical models in syntactic research, the
benefits (or otherwise) of logical syntax and semantics, and so on. However,
this convergence of ideas was of ephemeral duration, and by the early 1960s
the theory had begun to change once more. Perhaps, instead of ‘TGG’, the term
‘Transformational Analysis’ should be used to refer to the theory presented in
Chomsky’s 1955 text The Logical Structure of Linguistic Theory, since this was
the title he gave to his 1955 Ph.D. thesis? Unfortunately, though, the range of
work that Chomsky accomplished in his earliest publications and manuscripts
is sufficiently wide that any single term would cause difficulties and eventually
demonstrate its own inadequacy. Consequently, for the purposes of the follow-
ing discussion, the term ‘TGG’ will be understood to refer to the syntactic
4 Introduction

theory that Chomsky expounded in The Logical Structure of Linguistic Theory

(c. 1955) and, less formally (but in a broader context), in Syntactic Structures
(1957). As a result, ‘TGG’ can become TGG again. It should be added, though,
that since the intention is to show how the formal sciences gradually began
to influence syntactic theory in general, ultimately resulting in the creation of
TGG, the focus of the discussion will often be upon various papers, theses, and
manuscripts concerning linguistic theory that appeared during the years 1900–
1957 and not just on Chomsky’s work from the 1950s. The task is to situate
TGG more securely (if possible) in the scientific context of the time.
Accordingly, having provided at least an initial discussion of some of the
difficulties, terminological and otherwise, that are inevitably encountered in a
study of this kind, it is necessary to assess existing publications concerning the
genesis of TGG in order to motivate more particularly the need for a serious
consideration of the influence exerted by the formal sciences upon the devel-
opment of TGG. The following discussion of the existing historiographical
literature makes no claim to be exhaustively comprehensive. Indeed, the body
of secondary literature concerning the development of TGG has started to accu-
mulate at such a rate that a detailed summary of all the posited interpretations,
assessments, revisions, and disagreements would leave little time or space for
a discussion of primary sources. Instead, the overview offered here is intended
generally to indicate how historiographical studies of TGG have developed
since the 1960s, with the perceived deficiencies in the existing studies being
highlighted. One final word of warning, though. As is usually the case, this
introduction was written after the rest of the book was complete. Therefore,
some of the topics mentioned in passing here may seem richer, deeper, less
arbitrary, and less idiosyncratic after the main chapters of the following text
have been perused. Consequently, although the rest of this introduction effec-
tively constitutes a survey of related literature, like all such surveys it should
be re-evaluated after the rest of the book has been read.
The earliest historical studies of the development of TGG can be found in the
first textbooks devoted to the theory, and these began to appear in the early 1960s.
Although these books often contained some sort of overview of the development
of the theory, the summaries offered were usually brief and generally gave the
impression that the origins of TGG could be traced back to Syntactic Structures,
and no further. A fine example of this kind of pedagogic text is Emmon Bach’s
An Introduction to Transformational Grammars (1964). The plural of the title
is revealing: the syntactic theory Bach presented had not yet developed into
the study of a single, genetically embedded Universal Grammar, consequently,
there could still be lots of TGGs. In his more historically inclined passages,
 Introduction 5

Bach gives particular emphasis to the influence of the formal sciences upon the
development of TGG. For instance, he states explicitly that the theory ‘has taken
its inspiration from modern logic’ (Bach 1964: 9), and he goes on to suggest
that mathematics, logic, and linguistics have moved closer together during the
last one hundred years (i.e., 1864–1964):

In the last century a great deal has been learned about the structure of deductive
systems (systems of logic, mathematics, axiom systems for various sciences).
Logicians and mathematicians have been concerned more and more with
studying various ‘language systems’ or ‘calculi’ from an abstract point of
view. At the same time, modern linguistics has tended towards describing lan-
guages as abstract formalised systems. In many ways, the theory of language
presented here may be considered the result of a convergence between these
two currents. The grammars that we shall study are attempts to state the prin-
ciples by which sentences of a language may be constructed, in much the same
way that a formalised mathematical theory may be used to construct theorems.
(Bach 1964: 9–10)

Like Lees before him, therefore, Bach states clearly that TGG makes use of
the same kind of axiomatic-deductive systems utilised by various branches
of mathematics, and, in addition, he suggests that linguistics and mathematics
have been converging for at least a century. Later, he observes more specifically
that the use of techniques derived from ‘modern logic and mathematics’ in
TGG may constitute ‘the most lasting result of the linguistic research of the last
decade’ (Bach 1964: 143). Unfortunately, though, Bach feels that ‘to document
this bit of cultural history in detail would take us well beyond the bounds of
this introduction’ (Bach 1964: 144). Consequently, he leaves this topic largely
undeveloped and does not return to it later in the book.
As the 1960s progressed it became increasingly obvious to the international
linguistics community that TGG was significantly more than an ephemeral fad,
and consequently the issue of the historical roots of the theory began to inspire
more interest. Chomsky himself contributed to this general trend when he pub-
lished Cartesian Linguistics: A Chapter in the History of Rationalist Thought
(1966), a somewhat disingenuous attempt to demonstrate that historical prece-
dents existed for concepts such as deep structure and surface structure – concepts
he was then in the process of elaborating. Specifically, Chomsky declared that
TGG could be viewed as ‘a reawakening of interest in questions that were, in
fact, studied in a serious and fruitful way during the seventeenth, eighteenth and
early nineteenth centuries’ (Chomsky 1966: 1), and he sought to argue his case
by focusing on such works as the Port-Royal Grammar (1660) and by reassess-
ing the work of linguists such as Wilhelm von Humboldt. Although (much to his
6 Introduction

annoyance) Chomsky’s book has never really been taken seriously by linguistic
historiographers, who tend to classify it as a work of ideological propaganda
rather than as an objective historical assessment of the development of syntactic
theory, it certainly seems to have inspired an interest in the task of situating
TGG securely within the history of ideas.2 However, since Chomsky was pri-
marily concerned with linguistic research that had been completed before the
mid-nineteenth century, he did not consider the actual development of TGG
itself, and therefore Cartesian Linguistics offers few insights into the emer-
gence of generative grammar. Curiously, a more revealing contemporaneous
insight into the nature of this development can be gleaned from the writings of
the post-Bloomfieldian Charles Hockett (1916–2000), who, by the late 1960s,
had become a rather lonely figure in the world of international linguistics. Dur-
ing the previous decade Hockett had been generally recognised as one of the
dominant linguists of his generation, but, as the years passed, linguistic the-
ory started to take a distinctly Chomskyan turn, and Hockett found himself
marginalised; a prematurely redundant relic of a bygone era. His response was
to publish The State of the Art (1968), a book in which he expressed his dis-
satisfaction with Chomsky’s general approach to syntactic theory, and it was
while he was preparing the ground for his robust critique of Chomskyan syntax
that Hockett recalled the circumstances surrounding the development of TGG
in the 1950s:

Lacking any explicit guidance as to where to turn for a broadened basis for
linguistic theory, Chomsky was forced on his own resources and tastes, and
turned towards the abstract fields of logic, mathematics, and philosophy, rather
than to science. If Harris’ work suggested either of these directions, it was the
former. Indeed, a number of us at the time, in our search for ‘rigor’, were gazing
longingly towards mathematics . . . the move was reasonable, since linguistics
(or language) surely has interconnections of various sorts with these scholarly
endeavours just as it has with anthropology, psychology and biology. (Hockett
1968: 36)

Certainly Hockett cannot be presented as an impartial historian of syntactic the-

ory, mainly because he himself was an active participant in the various debates
of the 1950s, and was specifically interested in utilising techniques derived from
mathematics in linguistic research. Nevertheless, his comments are perhaps of
some significance.3 It is revealing, for instance, that, while recalling the heady
days of the previous decade, when linguists (and not just those closely associ-
ated with TGG) were ‘gazing longingly’ towards mathematics, Hockett uses a
first-person plural pronoun, suggesting that a number of linguists were turning
to mathematics as a possible source of analytical techniques because they felt
 Introduction 7

that particular methodologies employed by the formal sciences (i.e., for Hock-
ett, ‘abstract’ sciences such as logic and mathematics) could enable linguistic
theory to be endowed with greater rigour. As will be shown in section 2.3, the use
of the term ‘rigour’ in this context is replete with significance. Unfortunately,
though, like Bach before him, Hockett does not provide a detailed account of
precisely which branches of mathematics he considered to have been especially
influential, nor does he indicate how the linguists managed to acquire a working
knowledge of contemporaneous developments in mathematics.
During the 1970s, as it became increasingly clear that Chomsky’s place in
the history of linguistics was secure, his early work began to be scrutinised
more closely by linguistic historiographers. For instance, John Lyons discussed
various aspects of TGG in his relatively non-technical book Chomsky (Lyons
1970), and while he certainly argues in this text that Chomsky’s work had
ushered in a new period of linguistic research (describing Syntactic Structures
as a ‘short but epoch-making book’ (Lyons 1970: 36)), Lyons is keen also to
emphasise the similarities between TGG and the type of grammatical research
conducted by the post-Bloomfieldians. For example, at one point he remarks that

Chomsky’s general views on linguistic theory as presented in Syntactic Struc-

tures are in most respects the same as those held by other members of the
Bloomfieldians school, and notably by Zellig Harris. In particular, it may be
noted that there is no hint, at this period, of the ‘rationalism’ that is so charac-
teristic a feature of Chomsky’s more recent writing. His acknowledgement of
the influence of the ‘empiricist’ philosophers, Nelson Goodman and Willard
Van Orman Quine, would suggest that he shared their views; but there is no
general discussion, in Syntactic Structures, of the philosophical and psycho-
logical implications of grammar. (Lyons 1970: 36)

It is striking that Lyons should emphasise the apparent absence of rationalism

in Chomsky’s early work, and that he should associate this absence with the
influence of the ‘empiricist’ philosophers Nelson Goodman (1906–1998) and
Willard Van Orman Quine (1908–2000). However, although Lyons returns to the
topic of ‘the evolution of Chomsky’s thought from empiricism to rationalism’
(Lyons 1970: 38) later in the book, he does not discuss the influence of Goodman
and Quine’s constructive nominalism upon Chomsky’s early work, nor does he
assess Chomsky’s later rejection of nominalistic techniques. Instead, the focus
of Lyons’ discussion is exclusively upon Chomsky’s better-known 1959 critique
of behaviourism.4
Another authoritative (and now sadly neglected) study of Chomsky’s work,
which appeared in the early 1970s, was Finngeir Hiorth’s Noam Chomsky: Lin-
guistics and Philosophy (1974). As the title suggests, Hiorth was as interested
8 Introduction

in the philosophical implications of Chomsky’s research as he was in its conse-

quences for linguistics, and, as a result, his book discusses the origins of TGG
in greater detail than most previous (and subsequent) studies. For instance,
Chomsky’s frequently ignored 1953 paper is discussed in (comparatively) con-
siderable detail, and Hiorth concludes that it ‘does not necessarily show that
Chomsky ever had great faith in the importance of formal, purely symbolic,
analyses for the purpose of empirical linguistic research’ (Hiorth 1974: 35). In
this context Hiorth goes on to consider the influence of constructive nominalism
upon the young Chomsky, observing that
In 1952–1953 Chomsky not only had a firm belief in the power of formal
methods; he also believed in an ‘inscriptional nominalistic’. . . approach. The
trend in his thinking at that time was due to the influence of Nelson Goodman
and W. V. Quine. The occurrence of the term ‘nominalistic’ here did not
reflect any mature belief in philosophical nominalism. To my knowledge in
Chomsky’s later work there is no trace of nominalism. (Hiorth 1974: 37)

This (brief) mention of Goodman and Quine is unusual in the historiographi-

cal literature devoted to TGG, where virtual silence is the norm. Regrettably,
though, Hiorth does not trace the precise nature of the influence of constructive
nominalism, and it is suggested that after 1953 Chomsky rejected nominalistic
methods altogether – a rather misleading claim, as the discussion in chapters 4
and 5 of this book seeks to demonstrate. Hiorth’s interest in the influence of
mathematics and logic upon TGG is also manifest, though, when he consid-
ers Chomsky’s complex attitude towards logical syntax and semantics and the
nature of their relationship to syntactic theory. The focus of his discussion is
upon the exchange between Yehoshua Bar-Hillel (1915–1975) and Chomsky in
the mid-1950s, but there is a brief mention of Rudolf Carnap (1891–1970), par-
ticularly his influence upon Bar-Hillel. After this the bulk of Hiorth’s text con-
centrates upon Syntactic Structures and Chomsky’s publications from the 1960s.
The year after Hiorth’s text appeared, Chomsky himself published his own
account of the genesis of TGG. His overview was contained in his introduction
to The Logical Structure of Linguistic Theory, a truncated version of which was
published for the first time in 1975. Chomsky’s discussion is detailed, and the
section that concerns his own intellectual development during the late 1940s
and 1950s contains the following passage:
At Harris’s suggestion I had begun to study logic, philosophy, and founda-
tions of mathematics more seriously as a graduate student at the University
of Pennsylvania, and later at Harvard. I was particularly impressed by Nel-
son Goodman’s work on constructional systems. In its general character, this
work was in some ways similar to Harris’s, and seemed to me to provide the
 Introduction 9

appropriate intellectual background for the investigation of taxonomic pro-

cedures that I then regarded as central to linguistic theory. But Goodman’s
ongoing critique of induction seemed to point in a rather different direction,
suggesting the inadequacy in principle of inductive approaches. Goodman’s
investigation of the simplicity of systems also suggested (to me at least) possi-
bilities for nontaxonomic approaches to linguistic theory. Quine’s critique of
logical empiricism also gave some reason to believe that this line of enquiry
might be a plausible one. Quine argued that the principles of scientific theory
are confronted with experience as a systematic complex, with adjustments
possible at various points, governed by such factors as general simplicity.
(Chomsky 1975[1955]: 33)

This revealing summary suggests that during the early 1950s Chomsky
perceived a close association between the methodology of constructional
system theory and the techniques employed by the post-Bloomfieldians,
especially Harris. Unfortunately, as will be shown below, Chomsky’s reference
to ‘Goodman’s ongoing critique of induction’ seems to have convinced certain
linguistic historiographers that Goodman was a rationalist who rejected
empirical procedures when, actually, the opposite was in fact the case. Indeed,
as is shown in section 5.3 below, Goodman’s apparent critique of induction
was merely the prelude to a robust defence of the same, and Chomsky himself
was well aware of this. Quine, who started out as an adherent of Carnapian
logical empiricism, shifted his position more than Goodman, so that by the
1950s he was certainly disillusioned with Carnap’s approach, and, as indicated
in the above passage, his writing from this period influenced Chomsky
directly. Chomsky’s recollections also indicate the direct association between
Goodman’s ideas concerning the simplicity of constructional systems and
his own early preoccupation with the concept of grammatical simplicity. It
should be noted, though, that, while it is illuminating to read Chomsky’s own
reflections upon these aspects of his early work, his reminiscences do not
really reveal anything that was not already implied by certain comments and
footnotes in his original papers and manuscripts.
Another significant passage in Chomsky’s 1975 recollections occurs when
he considers some of the particular branches of mathematics that influenced his
thinking during the formative years in the early 1950s.

Perhaps a word might be usefully added on the general intellectual climate

in Cambridge at the time when [The Logical Structure of Linguistic Theory]
was written. Interdisciplinary approaches to language communication and
human behaviour were much in vogue . . . Oxford ordinary language analysis
and Wittgenstein’s later work were attracting great interest. The problem of
reconciling these approaches (if possible) with Quine’s provocative ideas on
10 Introduction

language and knowledge troubled many students. Mathematical logic, in par-

ticular recursive function theory and metamathematics, were becoming more
generally accessible, and developments in these areas seemed to provide tools
for a more precise study of natural language as well. All of this I personally
found most stimulating. (Chomsky 1975[1955]: 39)

This passage supports Hockett’s description of linguists ‘gazing longingly’

towards mathematics in the 1950s. Recursive function theory and metamath-
ematics are specifically highlighted, but, since no details are given, questions
remain. Which aspects of recursive function theory and metamathematics inter-
ested linguists at this time? Where was information concerning these subjects
obtained? Who was making these ideas accessible? Precisely which linguists
were interested in these topics? There are many unresolved issues. Indeed,
although the above passage helpfully confirms the nature of the influences that
are detectable in Chomsky’s early publications and manuscripts, it also frus-
trates since it raises numerous further queries without providing answers.
In the same year that Chomsky published his authoritative account of
the development of TGG, detailing the manner in which he had reformu-
lated linguistic theory by rejecting the ‘taxonomic’ approaches of the post-
Bloomfieldians, another text appeared that seemed to question this interpre-
tation of linguistic history. This text, American Structuralism, was written by
Dell Hymes and John Fought, and their detailed re-evaluation of structuralist
research in the 1940s and 1950s undermined the (then) generally accepted view
that Chomsky’s work had inspired a complete revolution in syntactic theory.
Hymes and Fought’s main argument was that the post-Bloomfieldians could not
usefully be grouped together as a single group that pursued a coherent research
programme, since, in truth, they constituted a disparate collection of individuals
who were responsible for ‘a variety of alternative conceptions, and individual
directions of interest’ (Hymes and Fought 1981[1975]: 156). Therefore, Hymes
and Fought argued that it was misleading to view Chomsky’s work as a reaction
against a homogeneous generation of linguists, and, in a passage reminiscent of
Lyons, they provocatively suggest that ‘in 1957 Chomsky’s work might appear
to be to a great extent a victory for one Bloomfieldian approach to syntax (that
of Harris) as against another (that of Trager, Smith, et al.)’ (Hymes and Fought
1981[1975]: 155). However, with specific reference to the origins of TGG,
Hymes and Fought do not explore the influence of the formal sciences upon the
work of the post-Bloomfieldians in any great detail, and they largely neglect
the gradual mathematisation of syntactic theory that occurred during the 1940s
and 1950s, with the result that their account of the genesis of TGG, though
impressive, is, in this respect, incomplete.
 Introduction 11

As indicated above, Hymes and Fought’s work was certainly provocative,

and by 1980 Frederick Newmeyer was sufficiently provoked to publish his
Linguistic Theory in America: the First Quarter-Century of Transformational
Generative Grammar. This work is still generally considered to be the first major
historical study of the whole development of (pre-1980) generative grammar.
Indeed, Newmeyer himself claimed (rather grandly) that ‘what I have written is,
to my knowledge, the only work that combines a comprehensive account of the
forging of modern linguistic theory with a detailed elaboration and explanation
of its development’ (Newmeyer 1980: xi). However, Newmeyer devotes most
of his attention to the post-1957 development of TGG, with only two out of
eight chapters considering research into syntactic theory published during the
1940s and early to mid-1950s; and his approach to this research is strikingly
robust. For instance, he entirely rejects Hymes and Fought’s suggestion that
the so-called ‘structuralists’ never really constituted a coherent group, argu-
ing instead that structural linguistics provided an homogeneous context for the
‘Chomskyan Revolution’ that was inaugurated by the publication of Syntactic
Structures. For instance, we are told that The Logical Structure of Linguistic
Theory ‘completely shattered the prevailing structuralist conceptions of lin-
guistic theory’ (Newmeyer 1980: 35), although these conceptions are never
exhaustively described in their full complexity. In this context, Newmeyer dis-
cusses behavioural psychology in a rather cursory manner, suggesting that the
sort of ‘empiricist assumptions’ (Newmeyer 1980: 11) associated with it were
also a characteristic feature of structural linguistics. However, no attempt is
made to associate the ‘empiricist assumptions’ of structural linguistics with the
type of logical empiricism that was championed by Carnap and Goodman, and
which exerted such a profound influence over Chomsky’s earliest work. Indeed,
although Goodman is mentioned in his capacity as one of Chomsky’s teachers,
and as a person who was influential in enabling Chomsky to obtain a Junior
Fellowship at Harvard in 1951, there is no discussion of the nature of his intel-
lectual influence upon Chomsky. Similarly, although statistical methods for
modelling of language are briefly discussed with reference to Hockett (i.e.,
Newmeyer 1980: 2), no attempt is made to explore the connections that existed
between various sub-branches of mathematics, logic, and syntactic theory dur-
ing the 1940s and 1950s.
Unfortunately, Newmeyer’s discussion of Chomsky’s early work is as selec-
tive as his discussion of the structuralists. Perhaps it is not surprising that
the main emphasis falls upon Syntactic Structures, even though Newmeyer
acknowledges that most of the text was ‘excerpted’ from The Logical Structure
of Linguistic Theory (Newmeyer 1980: 30). Also, though a brief consideration of
12 Introduction

the notion of simplicity in relation to the task of grammar evaluation is offered,

Newmeyer inexplicably neglects to mention the influence of Goodman’s work
concerning the basal simplicity of constructional systems upon the young
Chomsky; and the influence of Goodman and Quine is further scanted when
Chomsky’s first published paper, which appeared in 1953, fails to receive
detailed consideration. It is almost as if Chomsky’s flirtation with construc-
tive nominalism (which inevitably came with considerable empirical/empiricist
baggage) is being intentionally excised from the history of TGG. In a similar
fashion, Newmeyer does not discuss Chomsky’s ideas concerning the rela-
tionship between linguistic theory and logic, ideas which were expressed in
two papers during the years 1954–1955; and further opacities in Newmeyer’s
version of events include the lack of a detailed consideration of the gradual
appropriation of transformation rules from Carnapian logical syntax and the
neglect of Bar-Hillel’s influence upon Chomsky during the mid-1950s, par-
ticularly in his exploration of recursive rules in syntactic analysis. A revised
edition of Newmeyer’s text appeared in 1986, but the main changes were to
chapters 6–8, which were rewritten in order to provide an insight into the most
recent developments in syntactic theory (i.e., Principles and Parameters, Gener-
alised Phrase Structure Grammar, Lexical-Functional Grammar, and the like),
and no attempt was made substantially to alter the account of TGG’s early
development.
It is ironic that in the very same year the first protracted statement of the
‘Chomskyan Revolution’ interpretation of TGG history appeared, the validity
of this very interpretation was questioned again when Stephen Murray sug-
gested in his article ‘Gatekeepers and the Chomskian Revolution’ (Murray
1980) that the popular notion of Chomsky fighting to have his radical new ideas
accepted by a discouraging and dismissive dominant linguistic establishment
was in need of fairly drastic revision. Rather than merely asserting a different
interpretation, though, Murray managed to obtain contemporaneous documents,
including correspondence from Bernard Bloch (1907–1965), who was the edi-
tor of the influential journal Language during the 1950s. The correspondence
Murray unearthed indicated that, rather than persecuting Chomsky, Bloch had
actively encouraged him to publish his work, and (not surprisingly) this evi-
dence generated considerable interest since it seemed at odds with Chomsky’s
own account of events, which he had provided in his 1975 recollections, as
discussed above. Consequently, Murray’s paper suggested that TGG was more
closely connected to post-Bloomfieldian syntactic theory than was generally
acknowledged at the time, and therefore his research contributed to the gath-
ering revisionist movement, which sought to explore the development of TGG
 Introduction 13

with greater accuracy. Nevertheless, despite Murray’s provocative research, the

main focus in the 1980s was often upon post-1957 developments of the the-
ory. For instance, in a 1989 paper ‘Philosophical Speculation and Cognitive
Science’, George Lakoff explored the different assumptions that underlie inter-
pretivist interpretations of TGG and the offshoot of the theory that became
known as Generative Semantics. As shown below, research into this subject
blossomed in the 1990s when the topic became something of an obsession
among certain historians of linguistics.
The various historiographical controversies mentioned above inspired a flurry
of research into the development of TGG in the 1990s, and one of the most
significant studies of this period was Peter Matthews’ book Grammatical The-
ory in the United States from Bloomfield to Chomsky (1993). As is clear from
Matthews’ humorous introduction, his basic intention was to question the valid-
ity of the caricatured history of TGG that had become commonplace by the
early 1990s, largely due to the influence of Newmeyer’s widely read account.
Indeed, Matthews refers to Newmeyer’s 1980 book as ‘the official history of
transformational-generative grammar’ (Matthews 1993: 208), and, if it is indeed
the case that Newmeyer’s version of events is generally accepted as the standard
interpretation, then Matthews’ own text can be viewed as an insightful alterna-
tive history of TGG. While Newmeyer had been keen to (over)emphasise the
difference between Chomsky’s work and that of his predecessors, Matthews
was equally keen to redress the balance by stressing some of the similarities
and connections that existed between post-Bloomfieldian research and TGG.
As a result, TGG is often presented as if it were the logical consequence of cer-
tain research traditions pursued by leading post-Bloomfieldian linguists, and,
in this sense, Matthews’ work can be classified as belonging to the revisionist
Lyons–Hymes–Fought–Murray tradition. To take but one example, Matthews
draws particular attention to the fact that Hockett was discussing the predictive
nature of syntactic theory as early as 1948, and demonstrates that TGG largely
fulfilled this particular research goal.5 However, despite this emphasis on con-
tinuity, Matthews, like Newmeyer before him, still focuses primarily upon the
relationship between TGG and the other linguistic theories that preceded it.
Consequently, the full complexity of the intellectual environment out of which
TGG emerged is not fully portrayed. For example, there is a detailed discus-
sion of the relationship between linguistic form and meaning in the work of
both Leonard Bloomfield (1887–1949) and Chomsky, yet there is no attempt
to consider these issues with reference to Hilbertian Formalism, the metamath-
ematical movement that dominated large areas of scientific discourse during
the 1930s, 1940s, and 1950s, and which fascinated many linguists, including
14 Introduction

Bloomfield and Chomsky (as discussed at length in chapters 3 and 4 below).

Also, although such topics as grammatical simplicity are considered, there is
no mention of constructional system theory, and Matthews does not trace the
influence of Goodman (and/or Quine) upon the young Chomsky. Without such
assessments, though, any discussion of the topic of simplicity criteria in TGG
is inevitably impoverished.
As already implied above, during the first half of the 1990s various researchers
began to focus on later developments within the TGG tradition rather than upon
the question of origins, and the fragmentation of the ‘generativists’ camp in
the 1960s received considerable attention. For instance, Randy Harris’ Lin-
guistic Wars appeared in 1993, closely followed by Geoffrey Huck and John
Goldsmith’s Ideology and Linguistic Theory: Noam Chomsky and the Deep
Structure Debates (1995), and both texts considered the rise and fall of Gen-
erative Semantics in the 1960s and 1970s. However, despite this focus on later
events, the development of the initial TGG formalism is discussed in both pub-
lications. Unfortunately, Harris’ book, which is self-professedly written in the
‘popular science’ (Harris 1993: vii) tradition, is rather too swift and glib in its
treatment of the origins of TGG to be taken seriously. There is an attempt to
consider the implications of the term ‘formal’, but no mention either of David
Hilbert (1862–1943) or of metamathematics, and therefore no real exploration
of the consequences of formalising syntactic theory. By contrast, Huck and
Goldsmith’s assessment of TGG’s origins is more convincing, and it includes
the following evaluation of Goodman’s influence upon Chomsky:
Another of Chomsky’s teachers [i.e., in addition to Harris] who had a signif-
icant influence on him was the philosopher Nelson Goodman. Having taken
several courses with Goodman at the University of Pennsylvania, Chomsky
was struck by the similarity between Harris’s perspective on language and
Goodman’s perspective on philosophical systems generally, what Goodman
called ‘constructional systems’. The important feature of constructional sys-
tems in this regard is that there exist objective criteria for evaluating how
simple or complex they are. Thus, if a theory of language were a construc-
tional system, then that theory would be subject to the principles that govern
such systems; that is, it could be evaluated according to criteria of simplicity
and economy. Furthermore, as a constructional system, the theory of language
should be formalizable; in fact, if language is to be evaluated as a construc-
tional system, it must be formalized, because only in that way can its simplicity
relative to other theories be measured. (Huck and Goldsmith 1995: 13)

Although it is always refreshing to encounter even a brief discussion of the

influence of constructional system theory in an historiographical study of TGG,
the above account is inadequate in a number of respects. For instance, it does not
 Introduction 15

consider the implications of the empiricist assumptions of the constructional

systems that Chomsky explored in his earliest work. Also, there is no mention
of Chomsky’s direct use of one of Goodman’s own constructional systems in his
1953 paper. Consequently, Huck and Goldsmith’s discussion tantalises rather
than satisfies, though, in their defence, it should be remembered that their main
concern was with the ideological linguistic wars of the 1960s and 1970s rather
than with the philosophical foundations of pre-1957 syntactic theories.
The year 1994 saw the publication of the collection of articles and papers
Noam Chomsky: Critical Assessments, edited by Carlos Otero. This publica-
tion fully acknowledges the influence of Chomsky upon contemporary society
and culture, and constitutes a significant step towards his official canonisa-
tion as one of the great intellectuals of the twentieth century. Various papers
in the collection touch upon aspects of the origins of TGG, but perhaps the
most comprehensive overview is contained in Otero’s introduction to volume I,
‘Chomsky and the Cognitive Revolution of the 1950s: The Emergence of Trans-
formational Generative Grammar’, the stated purpose of which is to place
Chomsky’s work in ‘a more general context’ (Otero 1994a: 1). Consequently, a
wide range of scientific topics is discussed, including relativity theory, quantum
mechanics, computer science, and others. However, Otero’s discussion is often
so brief that misleading statements intrude. For instance, he claims that White-
head and Russell’s exploration of the logical foundations of mathematics ‘led
directly to the research programme of the formalists’ (Otero 1994a: 10), when,
in fact, the first of Hilbert’s papers concerning the metamathematical approach
to foundations problems (the methodology that would later be referred to as
Formalism) appeared in 1904, six years before the first volume of Whitehead
and Russell’s Principia Mathematica was published. While it is certainly true
that Hilbert’s later work was directly influenced by Principia Mathematica,
Otero’s statement clearly gives a false impression of the chronology of events.
Another problem is that, since Otero seeks to provide ‘a preliminary’ (Otero
1994a: 1) discussion of some of the topics considered by the papers in volume
I, he does not himself make direct connections between the advances in the
formal sciences that he recounts and the development of TGG. For instance,
although the work of Kurt Gödel (1906–1978), Alonzo Church (1903–1995),
Stephen Kleene (1909–1994), Emil Post (1897–1954), and others is (briefly)
discussed, no attempt is made to consider the way in which aspects of this work
were knowingly used by linguists working in the 1940s and 1950s. Indeed, dis-
cussion of the post-Bloomfieldians in general is kept to a minimum, giving the
false impression that Chomsky was one of the first linguists to concern himself
with mathematical techniques; and since these issues are not addressed in any
16 Introduction

of the papers and essays contained in the volume, this false impression is never
corrected.
Another 1994 publication that contains a detailed consideration of the genesis
of TGG is Stephen Murray’s Theory Groups and the Study of Language in North
America: A Social History. While primarily a work of anthropology rather than
a straightforward historiographical study, Murray’s book extends his research
into the development of syntactic theory in the twentieth century that he had
begun in the early 1980s. With specific reference to the sections of the book
that deal with the period 1951–1957, Murray’s research once again involves
digging up forgotten letters and documents that provide insights into the actual
reception of TGG in the 1950s. However, although he robustly renews his attack
on Newmeyeresque interpretations of linguistic history, he does not provide an
account of the relationship between linguistics and the formal sciences during
the 1950s, though the book does emphasise the closeness of the connections that
existed between the post-Bloomfieldians and the proto-generativists, reinforc-
ing the idea that TGG was largely a continuation of existing research concerns.
Indeed, perhaps Murray’s most stimulating and thought-provoking contribution
in his book stems from his interest in generative grammar as a scientific theory.
Influenced by the well-known work of Thomas Kuhn (1922–1996), particularly
The Structure of Scientific Revolutions (1962), Murray discusses the status of
generative grammar as a scientific theory, and while happy to acknowledge that
TGG does appear to meet some of the criteria for an authentic scientific revolu-
tion, he suggests that later versions of the theory fall alarmingly short. Indeed,
the sketch that Murray provides of MIT-based research into generative grammar
in the post-1980 period is remarkable for its condemnatory tone. Describing
Chomsky as ‘an ageing dictator’, Murray summarises the way in which syco-
phantic ‘cohorts’ compete for favour and seek to emulate ‘Chomsky’s contemp-
tuous rhetoric’, and he condemns the general ‘MIT ethos of imperviousness to
criticism’ (Murray 1994: 445). In Murray’s opinion, the main consequence of
this is that linguistics at MIT is more akin to a dictatorship than a scientific
research centre, and, since the influences of the formal sciences upon syntactic
theory is a central theme of this book, some of these issues are of relevance and
will be discussed again later.
Partly prompted by the new revisionist accounts of linguistic history, such as
those by Matthews and Murray, that had appeared in the early 1990s, Newmeyer
collected together a number of his most recent essays and reviews and published
them in 1996 as Generative Linguistics: A Historical Perspective. Clearly, gen-
eral academic interest in linguistic historiography had increased since 1980, for,
in 1996, Newmeyer felt able to claim (somewhat solecistically) that ‘the origin
 Introduction 17

and development of generative grammar are now a hot topic’ (Newmeyer 1996:
1). However, the chapter of Newmeyer’s new book that dealt primarily with the
question of origins largely restated his well-known interpretation of the history
of TGG in terms of a Chomskyan revolution, and his main purpose appears to
have been to refute the recent criticism of his earlier work. Compared to the
studies due to Matthews, Huck, Goldsmith, and Murray, Newmeyer’s arguments
are often disappointingly assertive and superficial. In chapter 2, for instance,
he attempts to provide a thorough account of the development of TGG, but the
discussion contains numerous obscurities and errors. For instance, he claims
that Chomsky’s 1951 Master’s thesis was ‘the first to point out that the proce-
dures of American descriptivist linguists can be likened to the programme of
Carnap’s Der logische Aufbau der Welt’ (Newmeyer 1996: 15), yet Carnap’s
work is not explicitly mentioned in either the submitted or published version
of Chomsky’s thesis. Consequently, Newmeyer’s claims are misleading, and
they sometimes seem designed to support a simplistic, pre-existing interpreta-
tion of linguistic history rather than to constitute an intense investigation of the
bewildering complexity of actual events.
As the 1990s approached their conclusion, a number of significant contri-
butions to TGG historiography appeared. For instance, in 1999 a two-volume
collection of papers by various scholars was collected together and published
in honour of E. F. K. Koerner as The Emergence of the Modern Language
Sciences. This selection contains a number of papers that specifically discuss
the development of TGG. One of the most stimulating is Danny Steinberg’s
elaborately entitled ‘How the Anti-Mentalist Skeletons in Chomsky’s Closet
Make Psychological Fiction of his Grammars’. As the title leads one to expect,
Steinberg’s basic argument is that during the 1950s Chomsky was ‘a fervent
formalist and anti-mentalist’ (Steinberg 1999: 267), who, although he began
to shift towards a rationalist stance after 1959 (the basic shift that Lyons had
observed in 1970), never managed entirely to relinquish the ideological vestiges
of his early empiricism. Astonishingly, although Steinberg is keen to argue that
the pre-1959 Chomsky favoured an anti-mentalist approach to syntactic theory,
he does not even mention the influence of Goodman and Quine’s empiricism
upon Chomsky’s early work, nor does he discuss Chomsky’s use of construc-
tional systems and the empirical assumptions inherent in such an approach to
syntactic analysis.
Since the start of the new millennium a number of studies have been pub-
lished that are concerned either entirely or in part with the origins of TGG.
For instance, Giorgio Graffi’s 2001 publication 200 Years of Syntax: A Critical
Survey contains a whole section (specifically, section 8.4) that considers ‘The
18 Introduction

Emergence of Generative Syntax’. Graffi emphasises the ‘scientific formation’

of the young Chomsky, and although an attempt is made to assess the influence
of Harris, Goodman, Quine, and Bar-Hillel, the treatment is all too brief. For
instance, Goodman and Quine are disposed of in a single sentence in which it
is remarked that ‘their criticism of induction (Goodman) and of the “dogmas of
empiricism” (Quine)’ (Graffi 2001: 331) were the main aspects of their influ-
ence upon the young Chomsky. Clearly this is little more than a rather timid
paraphrase of Chomsky’s own 1975 account (quoted above), and contributes
nothing to a detailed exploration of these issues. In particular, there is no men-
tion of Goodman’s preoccupation with simplicity criteria and the influence
this had (and continues to have) upon Chomsky. Further, although there is a
discussion of the Chomsky versus Bar-Hillel debate concerning the relation-
ship between linguistic theory and logical syntax and semantics, the analysis
essentially consists of a brief summary of the main arguments, and there is no
exploration of the relationship between the views expressed in that debate and
other contemporaneous attitudes towards mathematics, logic, and the formal
sciences in general. In addition to Graffi’s account, Peter Matthews returned
to the topic of the origins of TGG in his 2001 publication A Short History of
Structural Linguistics, and though this text continues the process of exploring
the relationship between the young Chomsky and the post-Bloomfieldians that
was begun in Matthews’ 1993 book, it does not contribute anything of signifi-
cance to the discussion of the influence of the formal sciences upon the genesis
of TGG.
As the above summary of the existing historiographical studies demonstrates,
previous work has not paid sufficient attention to the influence of the formal
sciences upon the development of TGG, providing only occasional, cursory
overviews at best. Accordingly, the discussion sustained in the following chap-
ters is intended to rectify this situation, and the basic structure of the book
can be summarised as follows. Chapter 2 provides an overview of the devel-
opments within mathematics that resulted in the so-called ‘foundations crisis’
of the early twentieth century. After a brief consideration of the illogicalities
associated with early forms of the calculus, the rigorisation movement of the
nineteenth century is described, along with the emergence of Cantorian set the-
ory. The paradoxes of set theory are assessed, and the crisis in the foundations
of mathematics is discussed. Each of the three main foundational schools –
Formalism, Logicism, and Intuitionism – are considered in some detail; and,
finally, several influential textbooks that spread the gospels of Logicism and
Formalism are described. With the intellectual context securely established,
chapter 3 traces the gradual influence of the formal sciences upon syntactic
 Introduction 19

theory during the first half of the twentieth century. The main purpose is to set
the stage for the detailed consideration of TGG that is contained in chapters 4
and 5. A number of separate but related strands are followed. The main top-
ics considered are the interest of pre-TGG linguists in the axiomatic-deductive
method, the use of recursive definitions in logic and pre-TGG linguistics, the
development of constructional system theory, the advent of constructive nomi-
nalism, the increasing preoccupation of linguists with logical syntax and logical
semantics, and the gradual perception that, during the 1940s and 1950s, linguis-
tics had become more mathematical. By contrast, chapters 4 and 5 constitute a
sustained assessment of the influence of the formal sciences specifically upon
the development of TGG. Initially, the focus is upon Chomsky’s pre-1956 work,
and the main issues considered are his preoccupation with simplicity criteria,
his early belief in the utility of a constructive nominalist approach to syntactic
theory, and his complex attitude towards the use of techniques derived from
logic in linguistic analysis. The discussion then shifts to Chomsky’s work from
the years 1955–1957, and numerous topics are considered, including his nega-
tive appraisal of stochastic techniques, his advocation of evaluation procedures
over discovery procedures, his rejection of the empiricist implications of con-
structional system theory, his use of constructional techniques in his own work,
his early notion of transformation rules, his deployment of recursive rules, and
the axiomatic-deductive character of 1957-style TGG.
Although this introduction has now reached its natural conclusion, it is nec-
essary to make one final point. While such words as ‘historiography’, ‘post-
Bloomfieldians’, and phrases like ‘the 1950s’ have figured prominently in the
preceding paragraphs, just as they will feature conspicuously in the ensuing
chapters, it would be entirely incorrect to assume that this book only deals with
issues that have become quaint and intriguing exhibits in a dusty museum that
seeks to preserve now extinct linguistic theories. On the contrary, the implica-
tions of the issues addressed in this book for contemporary generative grammar
are potentially profound. Ever since the advent of the Minimalist Program (MP)
in the early 1990s, most leading researchers within the field have attempted,
in various ways, to reduce the theory of generative grammar to its essential
elements, rejecting all theoretical constructs that can be reinterpreted in terms
of more fundamental components; and, intriguingly, many of the elements that
have now come to be viewed as essential and irreducible were originally incor-
porated into TGG when they were adapted from the formal sciences in the 1950s.
For instance, such techniques as the axiomatic-deductive method and formal
recursion, as well as general concerns with theory-internal simplicity, were all
associated, in different guises, with the earliest versions of generative grammar,
20 Introduction

and therefore their continued presence within contemporary versions of the the-
ory is of considerable interest. While these issues are discussed at greater length
in the conclusion, it is important here simply to stress the current relevance of
the topics considered in this book. Indeed, perhaps it could be claimed that to
consider the aims and goals of contemporary generative grammar, without first
attempting to comprehend something of the intellectual context out of which
the theory developed, is to labour in a penumbra of ineffectual superficiality.
These words of caution complete this introduction; the requisite apologia for
yet another retelling of an old story has been offered, and a detailed exploration
of the complex relationship between linguistic theory and the formal sciences
can now commence.
2 The consequences of analysis

2.1 Chapter overview

The purpose of this chapter is to summarise some of the movements within the
formal sciences that occurred during the nineteenth and twentieth centuries,
and which were ultimately to exert a profound influence over the development
of TGG. Since the origins of TGG are the main focus of the following chapters,
and since any search for origins necessarily entails an infinite regress if taken
to an extreme, an arbitrary beginning is required, and the starting point chosen
for this discussion is the emergence of the calculus as an identifiable set of
algorithmic procedures in the late seventeenth century. Accordingly, the advent
of the calculus is discussed in section 2.2 and some of the disputes associated
with its appearance are reviewed, along with the main subsequent advances that
led to the creation of the branch of mathematics known as ‘analysis’.1 In section
2.3 various attempts to provide a more secure foundation for analysis are briefly
assessed, with particular attention being given to the endeavour to derive the
calculus from the rudiments of number theory. The development of set theory,
which grew out of the need to secure the basis of arithmetic, is summarised in
section 2.4, and some of the resultant paradoxes are explored. The remaining
sections of the chapter discuss the three main theories that emerged at the end
of the nineteenth century and the beginning of the twentieth century in direct
response to the foundations crisis prompted by set theory. These are standardly
known as Logicism (section 2.5), Formalism (section 2.6), and Intuitionism
(section 2.7). As a postlude to this chapter (and a prelude to the next), section 2.8
considers some of the main textbooks which appeared during years 1911–1955,
and which made many of the topics considered in this chapter available for the
first time to a wider audience. In particular, the texts considered in this section
were instrumental in enabling ideas developed in various sub-fields of the formal
sciences to be utilised within the context of linguistics. It is crucial to recognise
throughout that the topics considered here all point directly towards the discus-
sion of syntactic theory that is developed in chapters 3, 4, and 5. While it may

21
22 The consequences of analysis

not always be immediately obvious how some of the topics assessed in the fol-
lowing sections relate to syntactic theory, the many associations will eventually
emerge.

2.2 Calculus: doubts and disputes

Although the mathematical techniques associated with the calculus have their
roots in the work of René Descartes, Buonaventura Cavalieri, Pierre de Fermat,
Blaise Pascal, Isaac Barrow, and many others, a significant date that can be
used (somewhat arbitrarily) to denote the commencement of the public his-
tory of the calculus is 1684, the year in which Gottfried Wilhelm Leibniz
(1646–1716) published his paper ‘A New Method for Maxima and Minima’.
The significance of this paper lies in the fact that (i) it contains a clear state-
ment of the basic procedure of differentiation, (ii) it introduces the product
rule, the quotient rule, and the power rule, and (iii), as the title of the paper
suggests, it provides a new practical operation for obtaining maxima and min-
ima. Consequently, by introducing his ‘remarkable type of calculus’2 (Leibniz
1863c[1684]: 467) to the wider mathematical community, Leibniz’s 1684 paper
inaugurated a new stage in the development of modern mathematics. Two years
later, he published ‘Concerning a Deeply Hidden Geometry’, which introduced
the process of integration, and demonstrated that it was the inverse of differen-
tiation; an observation of such importance that it is now usually referred to as
‘the fundamental theorem of the calculus’. As is well known, while these papers
constituted a significant contribution to the advancement of mathematics, Leib-
niz’s conclusions had been largely anticipated by Isaac Newton (1643–1727). In
particular, in several unpublished papers written during the years 1664–1671,
Newton had also formulated the processes of differentiation and integration,
using his own notational system, and he went on to employ these techniques
extensively in The Mathematical Principles of Natural Philosophy (1687), his
seminal work of mathematical physics. Despite the fact that both Newton and
Leibniz obtained similar basic results in their respective versions of the calcu-
lus, a plethora of unforeseen consequences eventually resulted from the fact
that they both provided different conceptual justifications for the procedures
they introduced. Revealingly, though, neither man was able to construct a con-
vincing foundation for the calculus. For instance, in his 1665 paper ‘A Method
for Finding Theorems’, Newton used the symbol ‘o’ to indicate ‘an infinitely
little distance’, adding (in parenthesis!) that this manoeuvre ‘cannot in this case
bee understood to bee good unlesse infinite littleness may bee considered geo-
metrically’ (Newton 1967[1665]: 282). Newton required numbers such as ‘o’
 Calculus: doubts and disputes 23

because, when calculating certain kinds of derivatives, he obtained formulae

such as3
a
v =o+ +x (2.1)
2
in which the term containing ‘o’ ‘must be blotted out’ leaving the reduced
equation
a
v= +x (2.2)
2
In order to accomplish this, Newton argued (dubiously) that, in equations such
as (2.1), the ‘o’s became infinitely small and could therefore simply be ignored
(Newton 1967[1665]: 273). However, as the initial quotation from his 1665
paper demonstrates, Newton himself had profound doubts about the validity of
such reasoning, since he was not convinced that the notion of ‘infinite little-
ness’ had any validity in an algebraic context that was entirely removed from
geometric presentations.
Newton never managed to eradicate his early uncertainties concerning
infinitely small numbers, and, in the 1670s, he started to approach the problem
in a different way by claiming that differentiation actually involved quantities
that were continuously flowing rather than static, infinitesimally small numbers
of the sort he had used in his earlier work. In his 1670 paper ‘Concerning the
Method of Series and Fluxions’, for example, Newton presented the infinitesi-
mal increments as being in a state of flux, describing them as ‘quantities [that]
increase during each infinitely small instant of time’4 (Newton 1969[1670]: 80),
and such passages suggest that Newton was starting to think of his infinitely
small quantities as constantly varying, hence his use of the term ‘fluxions’.
However, the definitions and explanations that he offered in order to clarify his
notion of a ‘fluxion’ were no less opaque than his remarks concerning infinites-
imals, and, needless to say, such vague statements did little to elucidate this
crucial aspect of the new calculus.5 However, Newton appears not to have been
satisfied with his use of fluxions since several years later, in The Mathemati-
cal Principles of Natural Philosophy, he adopted yet another approach to the
calculus and, instead of referring to infinitely small quantities or fluxions, he
spoke instead of ultimate ratios and limits, observing that

these ultimate ratios [ultimae rationes] in which quantities disappear are not
actually the ratios of ultimate quantities, but limits [limites] towards which
the ratios of quantities decreasing without limit [sine limite] always converge,
and to which they approach nearer than by any given difference, but actu-
ally never go beyond [nunquam vero transgredi], nor in effect reach until
24 The consequences of analysis

the quantities are infinitely diminished [diminuatur in infinitorum]. (Newton

1972[1687/1726]: 88)

Although Newton’s inchoate notion of a limit would eventually be devel-

oped by a later generation of mathematicians, and would ultimately provide
a secure basis for the calculus, Newton himself did not manage to clarify the
vagueness inherent in the approach, as outlined in the above quotation. As a
result, his definitions of the techniques he proposed remained imprecise, and
he continued to justify his use of these poorly defined mathematical proce-
dures primarily by means of geometric (rather than purely algebraic) argu-
ments, and by emphasising the undoubted utility of the experimental results he
obtained.
If Newton’s vague definitions were inadequate, Leibniz’s attempts at expla-
nation and justification for his version of the calculus were no more convincing.
In ‘A New Method for Maxima and Minima’, for instance, Leibniz had sim-
ply avoided the issue by asserting that his work would be understood by those
familiar with such topics:

The demonstration of all this will be easy to one who is experienced in these
matters and who considers the fact, until now not sufficiently emphasised,
that d x, dy, dv, dw and dz can be taken as proportional to the momen-
tary differences [differentiis . . . momentaneis proportionales haberi posse],
either increments or decrements, of the corresponding x, y, v, w, z. (Leib-
niz 1863c[1684]: 223)

No explanation of these ‘momentary differences’ was forthcoming, though,

and, without pausing for further discussion or clarification, Leibniz immedi-
ately began to introduce the basic theory of differentiation. Not surprisingly,
this kind of optimistic appeal to sympathetic specialists in tacit possession of
shared knowledge did not exactly convince the sceptics, and Leibniz’s meth-
ods swiftly came under probing attack. For instance, in a 1695 paper, Bernhard
Nieuwentijdt criticised the use of infinitesimal numbers, and Leibniz responded
(rather inadequately) simply by rebuking his overly scrupulous critics.6
Predictably, given the parallel development of the calculus and the inevitable
priority debates that ensued, distinct traditions began to establish themselves
during the late seventeenth century, and these were broadly associated either
with the Newtonian or with the Leibnizian version of the calculus. Essentially,
the divisions were national, with leading British mathematicians such as John
Landen, Brook Taylor and Colin Maclaurin developing Newton’s fluxional
treatment, while continental mathematicians such as Jakob Bernoulli, Johann
Bernoulli and Leonhard Euler steadily extended Leibniz’s work.7
 Calculus: doubts and disputes 25

However, despite the various divisions and disagreements, the calculus

swiftly became the dominant research topic in certain mathematical circles,
and it was during the eighteenth century that the branch of mathematics known
as analysis began to emerge, largely as a result of significant advances made
in the theories of differential equations, power series, and the calculus of vari-
ations. These advances were often associated with impressive experimental
results, which were frequently taken as sufficient evidence of the validity of
the underlying methods, though the basic theory still lacked clear foundations.
Although the more pragmatic practitioners of the new analytical techniques
seemed content with this situation, many leading intellectuals were under-
standably dismayed by the fact that analysis was based upon such insecure
foundations, and the attacks upon its credibility, which had started in the late
seventeenth century, increased in intensity. Perhaps the most vehement critique
was delivered by George Berkeley (1685–1753). In his magisterially polemical
discourse The Analyst (1734), Berkeley incisively lampooned those mathemati-
cians who uncritically availed themselves of the procedures provided by the
calculus, accusing them of being too willing to accept the sort of logical absur-
dities that they were swift to ridicule in Christianity. He concentrated his attack
upon the poorly defined infinitesimals, memorably referring to them as ‘Ghosts
of departed Quantities’ (Berkeley 1992[1734]: 199). Although Berkeley’s crit-
icism may have been the most trenchant, and possibly the most influential, it
was certainly not unique, and similar negative assessments were articulated by
Pierre Gassendi and Pierre Bayle, amongst others.
Gradually, as the criticisms increased in potency and number, it became
obvious that the intellectual credibility of the most admired branch of modern
mathematics was at stake. Not surprisingly, the leading mathematicians of the
age were dissatisfied with this situation, and they responded by attempting (yet
again) to demonstrate the validity of analysis. Various approaches were adopted.
In his Foundations of Differential Calculus (1755), for instance, Leonhard Euler
(1707–1783) endeavoured to remove the geometric basis of the calculus (which
was part of the Newtonian legacy) by founding it instead upon the theory of
functions. As a result, his approach was more systematic than that of his pre-
decessors, since he undertook a thorough study of all the elementary functions
and their respective derivatives and integrals. Unfortunately, though, he was still
unable to deal adequately with infinitesimals, referring to them as the ratio of
‘infinitely small quantities [quantitates infinite parva]’ (Euler 1912[1755]: 69),
without being able to state clearly what this meant. Infinitesimals continued to
pose non-trivial problems throughout the second half of the eighteenth century,
but a significant advance was largely anticipated by Jean le Rond D’Alembert
26 The consequences of analysis

(1717–1783), who revived Newton’s vague notion of a limit and argued in a

1754 article ‘Differential’ that the use of such limits, when clearly defined,
could provide a more secure foundation for the whole of analysis. While this
idea would eventually inspire a whole generation of mathematicians in the nine-
teenth century, it was not recognised as especially significant by D’Alembert’s
contemporaries, and indeed Joseph Louis Lagrange (1736–1813) rejected both
Leibnizian infinitesimals and D’Alembertian limits when presenting his ver-
sion of the calculus. Lagrange articulated his approach most coherently in his
‘Essay Concerning a New Method’ (1760), a work in which he attempted to
present the calculus of variations (introduced by Euler) in a purely algebraic
framework. The final years of the eighteenth century were characterised by
a flurry of alternative proposals, all seeking to situate analysis upon a stable,
logical basis; but the existence of so many different proposals only served to
increase the general sense of disquiet.

2.3 Rigour, arithmetic, and axioms

As indicated in section 2.2, the various disagreements and disputes associ-
ated with the development of analysis in the eighteenth century were directly
inspired by the dubious logical status of its conceptual foundation. Conse-
quently, at the start of the nineteenth century, a number of mathematicians
renewed the attempt to establish the validity of analysis in an unambiguous
fashion. Of these, Augustin-Louis Cauchy (1789–1857) proved to be the most
influential reformer. His general approach to the problem was first outlined
in his Algebraic Analysis (1821) before being elaborated in his Lesson Con-
cerning the Infinitesimal Calculus (1823) and other publications. According to
Birkhoff, this work ‘marked the dawn of a new era’ (Birkhoff 1973: 1) in the
development of the calculus, and Cauchy’s proposals certainly proved to be
extremely influential. In essence, Cauchy believed that analysis could be ren-
dered secure by constructing it upon the foundation provided by number theory,
and a characteristic feature of his work was the repeated desire for rigour. For
instance, in the preface to his Lesson Concerning the Infinitesimal Calculus,
Cauchy stated that his main aim was ‘to reconcile rigour [rigueur] . . . with the
simplicity which results from the direct consideration of infinitesimals [quantité
infiniment petite]’ (quoted in Birkhoff 1973: 1). In this context, as Grabiner has
shown, in the early nineteenth century mathematical ‘rigour’ generally required
a tripartite methodology:

First, every concept of the subject had to be explicitly defined in terms of con-
cepts whose nature was held to be already known . . . Second, theorems had
 Rigour, arithmetic, and axioms 27

to be proved, with every step in the proof justified by a previously proved the-
orem, by a definition, or by an explicitly stated axiom . . . Third, the definitions
chosen and the theorems proved, had to be sufficiently broad to support the
entire structure of valid results belonging to the subject. (Grabiner 1981: 5)

In essence, this indicates that one of Cauchy’s most significant contributions

was to focus attention upon the the axiomatic-deductive method, and, as will
be discussed in later sections, it was this concern with deduction and valid
proof techniques that resulted in the late nineteenth-century preoccupation with
axiomatics, which, in turn, would influence the development of linguistics in
the twentieth century.
In brief, then, Cauchy’s basic strategy was to guarantee the security of anal-
ysis by basing it upon number theory, and to provide rigorous proofs for the
basic theorems. A number of consequences were to follow from this general
approach. For instance, since Cauchy emphasised the role of proofs in math-
ematical arguments, he appeared to undermine the long-standing yet dubious
tradition of justifying the use of specific theoretical techniques by emphasising
the obvious utility of the experimental results obtained: in the Cauchian frame-
work, analysis had to be justified without reference to practical utility. Another
consequence concerned the perceived relationship between analysis and logic.
Given the increased rigour demanded by Cauchy’s project, he frequently had
to seek sufficient and necessary conditions for the theorems he was endeavour-
ing to prove, and, since the task of determining such conditions was primarily
accomplished by means of logical deduction, it is not surprising that his work
suggested a relationship between logic and calculus, indirectly encouraging
mathematical rather than purely philosophical research into logical systems
(although these disciplines clearly overlap).8 In passing it should be noted that
although Cauchy’s emphasis on rigour was (to some extent) new, he never-
theless drew heavily upon the work of his most illustrious predecessors. For
instance, following D’Alembert he used the concept of a limit in order to con-
struct definitions of differentiation and integration, and this in turn enabled him
to unite the theory of functions, advocated by Euler, with theories concerning
the convergence of infinite series that had been explored by Taylor and others.9
In other words, Cauchy largely succeeded in synthesising various approaches
to analysis that had been presented as rival alternatives during the eighteenth
century, thus paving the way for greater agreement amongst the members of
the international mathematical community.
As mentioned above, Cauchy’s work proved to be extremely influential
throughout the nineteenth century, and his most able successors included Karl
28 The consequences of analysis

Weierstrass (1815–1892), Richard Dedekind (1831–1916), and Georg Cantor

(1845–1918). These men were all involved in the movement that is sometimes
referred to as ‘the arithmetisation of analysis’; that is, the research programme
(begun by Cauchy) that attempted to ensure the security of the foundations of
analysis by basing it upon the rudiments of number theory. Weierstrass, in par-
ticular, was instrumental in achieving Cauchy’s ambition to remove illogicality
and vagueness from the core of the calculus, and in a series of lectures delivered
during the academic year 1858/1859 at the University of Berlin he began his
exploration of the foundations of analysis that would ultimately provide the
basic  − δ definition of a limit, and the corresponding definitions of differen-
tiation and integration that are essentially those still used today.10 However, as
Weierstrass was well aware, the task of founding analysis upon number theory
obviously presupposed the security of the latter, and, as a result, a considerable
amount of research in the mid- to late nineteenth century was devoted to the
task of exploring the foundations of arithmetic. Often during this period it was
assumed (somewhat dangerously) that the real numbers were well defined and
that it was the irrational and complex numbers that required coherent specifi-
cation. Accordingly, in 1872, Dedekind published a booklet entitled Continuity
and Irrational Numbers in which he undertook to provide an exhaustive defini-
tion of the irrational numbers. The method he employed required the creation
of a technical procedure he called a ‘cut’ (‘Schnitt’), which was used to divide
the continuous number line into two classes, a method that would eventually
became a standard procedure in number theory.11
During the 1870s and 1880s Dedekind continued his probing exploration of
various aspects of number theory, focusing increasingly on the rational num-
bers, and in 1888 he published what was to become his best-known paper, ‘What
are Numbers and What Should they Be?’ In this work Dedekind outlined an
axiomatic approach to the task of defining the rational numbers, and this kind
of methodology was entirely in keeping with the movement towards greater
rigour that had been initiated by Cauchy, since, as mentioned earlier, during the
nineteenth century axiomatic systems were used increasingly by mathemati-
cians concerned with foundational questions. Dedekind’s research into number
theory was extended by Giuseppe Peano (1858–1932), who, in his Principles
of Arithmetic (1889), derived the rational numbers from axioms concerning the
positive whole numbers. With Peano’s work it seemed as if the foundations of
number theory were close to being rendered secure.
Given the preoccupations of his contemporaries as summarised above, it is
not surprising that Cantor was interested in the foundations of number theory
at the start of his career. Indeed, his first papers were entirely devoted to the
 Set theory and paradoxes 29

task of exploring and extending work in this area already largely accomplished
by Dedekind and others.12 However, during the 1870s he began to develop new
ideas, inspired by his number-theoretical investigations, and so began the intel-
lectual odyssey that led him to formulate the rudiments of modern set theory.
Since set theory was largely responsible for triggering the crisis of foundations
which shook mathematics to its core in the early twentieth century, it is necessary
briefly to summarise the basics of the conceptual scheme Cantor introduced.

2.4 Set theory and paradoxes

As implied above, Cantor’s Mengenlehre (his name for the collection of defi-
nitions and operations that eventually developed into modern set theory) was
initially inspired by research into trigonometric series that was accomplished by
Dedekind, Bernhard Riemann (1826–1866), and Eduard Heine (1821–1881).
In particular, as part of a research endeavour the results of which were pub-
lished in 1867, Riemann had examined functions that possessed an infinity
of turning values (i.e., maxima and minima) and/or discontinuities (Riemann
1990[1867]), and this work raised many unresolved issues. In a similar manner,
in 1870 Heine considered various problems associated with the convergence of
trigonometric functions, and his work required the notion of an infinite num-
ber of points on the real number line (Heine 1870). Cantor was intrigued by
these number-theoretical concerns, and in his pre-1872 papers he extended and
refined some of these ideas. However, he swiftly realised that the notion of
an infinite collection of mathematical objects, which was required if the real
number line was to be adequately defined, had been treated rather informally
by number theorists in the past; even by those associated with the Cauchy–
Weierstrass rigorisation programme. Therefore, in a series of papers published
between the years 1872 and 1895 Cantor gradually developed an approach to
number theory that was grounded in the concept of a set (‘Menge’). His ideas
evolved gradually, and there were a number of significant moments when his
thinking coalesced. For instance, the publication of the first paper primarily
devoted to set theory in 1874 was a significant occurrence, as was the appear-
ance of his monumental Foundations of a General Theory of Manifolds (i.e.,
for ‘manifolds’ read ‘sets’) in 1883, which summarised the main aspects of
the theory to date. However, perhaps his most influential paper was ‘Contri-
butions to the Foundations of Transfinite Set Theory’, which was published in
1895, and which was subsequently dispersed widely throughout the interna-
tional mathematical community. Since an exhaustive discussion of the gradual
development of set theory is neither feasible nor desirable here, the following
30 The consequences of analysis

summary is intended to introduce only those aspects of the theory that relate to
topics considered at length in later sections.13
In the Cantorian tradition a set is viewed as a collection of objects; a given
set constitutes a single entity, and the elements it contains are referred to as
the ‘members’ of that set. For instance, using modern notation, the statement
‘A = {α, β}’ indicates that A is a set that contains two members, the elements α
and β. Sometimes sets can be defined by specifying properties of their members.
For example, the statement ‘C = {x | x is an even number}’ indicates that C is
the set of even numbers, with the symbol ‘|’ being understood to mean ‘such
that’. Set membership is indicated using the symbol ‘∈’ (i.e., ‘α ∈ A’ indicates
that α is a member of the set A), while the symbol ‘∈’ / is defined as the inverse
of ‘∈’ (i.e., ‘γ ∈/ A’ indicates that γ is not a member of the set A). Cantor
introduced the basic operations of set theory that are now referred to as union
(‘∪’), intersection (‘∩’) and difference (a.k.a. complement) (‘\’). In addition,
he defined the notion of a subset: if A and C are both sets, then A is a proper
subset of C if every member of A is also a member of C but it is not the case that
every member of C is also a member of A (i.e., in symbols, ‘A ⊂ C’). Cantor
also introduced the notion of a power set: given a set, A, the power set of A
contains all possible subsets of A and is is denoted symbolically as ‘P(A)’.
Although these various definitions and operations seem innocuous in the finite
realm, one of Cantor’s motivations for developing his theory was to devise a
method of dealing with the infinite sets of points on the real number line. Since
he considered a set to be a self-contained whole (i.e., a collection of individuals
that could be treated as an individual entity itself), he was able to explore the
notion of an infinite set simply by extending the concept of a finite set. For
instance, if A = {x | x is an even number}, then clearly the set A contains an
infinite number of elements, since it contains all the even numbers. Eventually,
following this kind of reasoning, Cantor reached the startling conclusion that
infinite sets are not all the same size. In order to illustrate this idea he used the
notion of one-to-one correspondence. For instance, to give an example of the
correspondence principle being used to indicate that two sets are the same size,
since the elements in the set of whole numbers, I = {1, 2, 3, . . .}, can be put
into one-to-one correspondence with the elements in the set of even numbers,
E = {2, 4, 6, . . .}, these sets can be considered to be the same size, which is
clearly counter-intuitive since E ⊂ I . This result suggests that, in set theory at
least, a part can be equal to the whole.
Cantor’s research into infinite sets created the branch of number theory known
as ‘transfinite arithmetic’. As indicated above, his exploration of the basic
set-theoretical formalism caused him to consider the possibility that there were
 Set theory and paradoxes 31

infinite sets of different sizes, and, in order to examine this more fully, he intro-
duced transfinite cardinal and ordinal numbers. Given a finite set A, where
A = {α, β}, the cardinality of A is 2 (in modern notation, ‘|A| = 2’), since A
contains two members. The same ideas can be extended to infinite sets and
Cantor introduced the symbol ‘ℵ0 ’ (aleph-null) to represent the cardinal num-
ber associated with the set of whole numbers (i.e., if I = {1, 2, 3, . . .}, then
|I | = ℵ0 ). By the mid-1890s Cantor had explored various properties of the
transfinite cardinals and, in particular, he was able to demonstrate that the power
set of a given set must have a larger cardinal number than the original set itself,
and this seemingly harmless theorem was to have unforeseen consequences
during the early years of the twentieth century. The transfinite ordinal numbers
(which Cantor also introduced) are similar in principle to transfinite cardinals,
the difference being that they can only be obtained for sets the elements of which
have been organised in a pre-defined sequence. Accordingly, Cantor used the
symbol ‘ω’ to represent the transfinite ordinal associated with the set of whole
numbers. Since, in Cantorian arithmetic, numbers are created by a generative
procedure that involves adding ‘1’ to an existing number, Cantor realised that
an infinity of ordinal numbers could be obtained simply by applying the same
procedure to ω. In this fashion, the infinite sequence
1, 1 + 1, . . ., ω, ω + 1, . . ., ω2 , . . ., ωω , . . . (2.3)
can be produced. Since these ordinals denote the size of corresponding sets, it
follows that an infinity of differently sized infinite sets could be constructed.
Cantor’s Mengenlehre was an undeniably provocative theory, and it quickly
inspired controversy. Nevertheless, it appealed to those mathematicians con-
cerned with the task of arithmetising analysis, since it seemed to offer a plausible
framework that could be used to secure the foundations of number theory once
and for all. Consequently, Cantor’s work was elaborated by other researchers,
and it began to be used as a theoretical basis in many different sub-branches of
mathematics. Without doubt, the most important post-Cantorian extension of
the theory was due to Ernst Zermelo (1871–1953) and Adolf Fraenkel (1891–
1965), who managed to provide an axiomatic foundation for set-theoretic con-
cepts. Indeed, the so-called ‘ZF’ axiom set they provided is still standardly used
as the basis for modern classical set theory, and it has enabled the discipline to
blossom into a highly sophisticated branch of twentieth-century mathematics.14
However, despite the prevalent enthusiasm for set theory that characterised the
mid-1890s, Cantor himself soon began to identify apparent weaknesses in his
work, and the problems seemed to cluster around the transfinite aspects of the
theory. For instance, in a series of letters to Dedekind written in 1899, Cantor
32 The consequences of analysis

considered the logical validity of the set of all sets and argued that, since this
set contains all other sets, its transfinite cardinal must be larger than any other
transfinite cardinal. However, since the set of all subsets of a given set must
have a larger cardinal than the set itself (as he had previously demonstrated), it
follows that there is a larger cardinal number than the largest cardinal number
(Cantor 1937[1899]). This paradox caused him to recommend that the set of all
sets and its associated cardinal number should not be included in discussions of
the subject since the concept was not coherent. Accordingly, he began to speak
(somewhat vaguely) of consistent and inconsistent sets, the latter including
those that result in paradoxes.
Alarming as the discovery of this paradox was, Cantor later realised that
the problems were not confined to transfinite cardinals, and he began to dis-
cover unexpected inconsistencies in the theory of transfinite ordinal numbers.
As mentioned above, a basic theorem of transfinite arithmetic states that the
ordinal number of the set of all ordinal numbers up to and including n is larger
than n, and the ordinal number associated with the set of ordinal numbers
{1, 2, 3, . . ., ω} is ω + 1. Therefore the set of all ordinals should be associated
with an ordinal that is larger than the largest in the set. But this is a contradic-
tion, since the set contains all ordinals. Cantor communicated this concern to
Hilbert in c. 1896, and it later became known as the Burali-Forti Paradox after
Cesare Burali-Forti (1861–1931), who discussed the problem in a later paper.
These problems, perhaps ‘paradoxes’, were a cause of real unease when they
became known to the mathematical community: how could set theory possibly
provide a secure foundation for number theory if it clearly contained inherent
illogicalities itself? Indeed, the difficulties resulting from transfinite arithmetic
largely inspired the various proposals for securing the foundations of mathe-
matics that coalesced into the three dominant ideologies that came to be known
(rather too simplistically perhaps) as Logicism, Formalism, and Intuitionism.
Since these philosophical-mathematical movements ultimately influenced the
development of linguistics in the twentieth century, it is necessary to consider
them separately in some detail.

2.5 Logicism
Although logic has its roots in antiquity, the development of modern symbolic
logic was given significant impetus in the seventeenth century when Leibniz
began to consider the possibility of a universal symbolic language that could
be used to clarify all arguments and resolve all disputes, as proposed in his
Dissertation Concerning the Art of Combinations (1666). Leibniz’s ideas were
 Logicism 33

extended only spasmodically by his immediate successors, and it was not until
the start of the nineteenth century that more focused and ambitious treatises
concerning logic began to appear. In England, Augustus De Morgan (1806–
1871), George Boole (1815–1865), and Francis (F. H.) Bradley (1846–1924)
made significant contributions to the logic of relations and explored the con-
nections between logic, algebra, and probability theory; in Germany, Hermann
Grassmann (1809–1877) and Ernst Schröder (1841–1902) advanced the theory
of logical operators and clarified the nature of the identity relation; while, in
America, Charles Peirce (1839–1914) fused the work of Boole and De Mor-
gan, thus creating a Boolean logic of relations which required him to adopt the
basic principles of the propositional calculus.15 However, while these various
research trends served to augment the range and power of symbolic logic, it
was Gottlob Frege (1848–1925) who first proposed that logic could provide a
secure foundation for arithmetic and (by extension) mathematics in general. In
his first work devoted entirely to logic, Concept-Script (i.e., Begriffsschrift),
which was published in 1879, Frege presented a sophisticated logical system
that included the truth-functional propositional calculus, the analysis of propo-
sitions as functions and arguments (rather than the traditional subject-predicate
decomposition), the basic theory of existential and universal quantification, the
use of derivations based entirely on the expression-form of statements, and
numerous other procedures and techniques that have since become a standard
part of classical logic. However, it was not until his next major work, The
Foundations of Arithmetic, that Frege explicitly suggested that the type of log-
ical system he had outlined in his Concept-Script could be used as a basis
for arithmetic. This bold claim, which he went on to explore more thoroughly
in The Fundamentals of Arithmetic (1893), effectively initiated the Logicism
movement that was to dominate research into symbolic logic during the early
twentieth century.16
While Frege was developing his logico-mathematical philosophy during the
1880s, Peano was busy in Italy extending the work of Boole, Grassmann, and
Schröder, thereby establishing his own school of symbolic logic. In his ambi-
tious Principles of Arithmetic (mentioned in section 2.3 above) he argued (like
Frege) that arithmetic could be constructed upon the foundation provided by
logic. To this end, he set about devising a consistent notational system that would
enable him to axiomatise mathematics using a logical symbolic language. In
addition, one of Peano’s most significant contributions was to move logic away
from the purely algebraic tradition that had dominated in the mid-nineteenth
century, and he achieved this (in part) by examining the relationship between
logic and Cantorian set theory.17 Gradually Peano’s work became known to
34 The consequences of analysis

a wider international audience, helped by his involvement in the creation of

the journal Rivista di matematica (i.e., Mathematical Review) in 1891, which
enabled his ideas (and that of his growing number of students) to be dissemi-
nated with greater ease.
The movement to derive the whole of arithmetic (and therefore mathematics)
from the parsimonious essentials of symbolic logic culminated in the work of
Alfred Whitehead (1861–1947) and Bertrand Russell (1872–1970). Whitehead
had been awarded a fellowship at Trinity College, Cambridge in 1884 after
submitting a (now lost) dissertation on James Clerk Maxwell’s theory of elec-
tromagnetism, and his interest in symbolic logic developed during the 1890s
while he was working on his book A Treatise on Universal Algebra with Appli-
cations, which eventually appeared in 1898. During this time he was involved in
examining a dissertation by Russell on the foundations of geometry, so the two
future collaborators first met as fellow and student. Russell’s early interest in
the philosophy of mathematics was inspired by his initial attempt to axiomatise
the rudiments of geometry, and he did not begin seriously to devote himself to
the study of symbolic logic until 1897.18 His interest in the subject had been
stimulated by his study of Bradley’s The Principles of Logic (1883), and after
1897 he made swift and purposeful progress in the task of turning himself into
a fully-fledged logician. During the years 1897–1910, working in conjunction
with Whitehead, he began to synthesise the work of Cantor, Frege, and Peano
(amongst others) in an ambitious attempt to derive the whole of mathematics
from a small collection of self-evident logical axioms, in direct pursuance of
what came to be known as the Logicist programme, and, crucially, many of
Whitehead and Russell’s contributions to the theory of symbolic logic were
motivated by the various paradoxes associated with set theory.
Russell had first encountered Cantor’s set theory in 1895, when he obtained a
pamphlet copy of the Foundations of a General Theory of Manifolds; he studied
Cantor’s work intensely from 1896 to 1897, and, by 1899, was largely persuaded
of the validity of his basic approach. But problems began to emerge. In Novem-
ber 1900 Russell discovered a ‘fallacy’ in Cantor’s work (now generally known
as ‘Russell’s Paradox’), which can be succinctly expressed as the following
question: given the set, S, which is the set of all sets that are not members of
themselves, is S a member of itself or not? The answer to this question is para-
doxical since, if S is a member of itself, then (obviously) it follows that it is not
a member of itself – and vice versa. Disturbed by this illogicality, Russell began
assiduously to collect similar paradoxes during the early 1900s, often giving
them names (i.e., ‘The Liar’s Paradox’, ‘Berry’s Paradox’, ‘The Burali-Forti
Paradox’, and so on), and part of the task he set for the research programme
 Logicism 35

that he and Whitehead were pursuing was to extirpate fully these paradoxes
from set theory. The first clear statement of Russell’s intent was contained in
his Principles of Mathematics, which appeared in 1903. In the introduction to
this text, Russell articulated his ‘general doctrine’ (which he associated with
Leibniz) that ‘all mathematics is deduction by logical principles from logical
principles’ (Russell 1938[1903]: 5), and, in the ensuing text, he went on to
articulate his convictions more precisely:

The connection of mathematics with logic . . . is exceedingly close. The fact

that all mathematical constants are logical constants, and that all the premises
of mathematics are concerned with these, gives, I believe, the precise statement
of what philosophers have meant in asserting that mathematics is à priori
[sic]. The fact is that, when once the apparatus of logic has been accepted, all
mathematics necessarily follows . . . From what has now been said, the reader
will perceive that the present work has to fulfil two objectives, first, to show
that all mathematics follows from symbolic logic, and secondly to discover,
as far as possible, what are the principles of symbolic logic itself. (Russell
1938[1903]: 8–9)

In accordance with these objectives, during the years 1903–1910 Russell and
Whitehead collaborated and published a series of papers in which they consid-
ered various difficulties attendant upon the task of seeking (i) to discover the
most parsimonious axiomatic logical system, and (ii) to derive the whole of
mathematics from this basis in such a way as to avoid all ‘fallacies’. Although,
as mentioned above, much of their work involved synthesising the research of
their predecessors and contemporaries, they also made numerous significant
theoretical contributions themselves. Of these, perhaps the most controversial
was the ‘theory of logical types’. Whitehead and Russell had observed that the
paradoxes of set theory invariably involved self-reference of one kind or another,
hence their tendency to refer to them as ‘vicious-circle fallacies’. Accordingly,
the theory of logical types was designed to delimit the extent of permissible
self-reference in an attempt to avoid the paradoxes. As they later explained,

An analysis of the paradoxes to be avoided shows that they all result from
a certain kind of vicious circle. The vicious circles in question arise from
supposing that a collection of objects may contain members which can only
be defined by means of the collection as a whole . . . The principle which
enables us to avoid illegitimate totalities may be stated as follows: ‘whatever
includes all of a collection must not be one of the collection’. (Whitehead and
Russell 1925[1910]: 37)
36 The consequences of analysis

Although an arbitrary and rather elaborate ‘principle’, the theory at least pro-
vided a practical way of avoiding paradoxes while developing set-theoretical
concepts from the axioms of a given logical system.19
Whitehead and Russell’s decade of research into the feasibility of their Logi-
cist project resulted in the publication of Principia Mathematica (hereafter
PM) during the years 1910–1913. The influence of PM upon the development
of logic during the twentieth century cannot really be overestimated. Although
it contains inevitable inconsistencies and obscurities, this work still constitutes
the most profound attempt to reduce mathematics to logic. Consequently, it is
necessary to consider the form and content of PM in some detail.20
The various chapters of PM were subdivided in sections which were further
subdivided into ‘numbers’, indicated by an asterix and a numeral (i.e., *1),
which served to facilitate cross-references. In numbers *1–*5, the so-called
‘theory of deduction’ was developed, and this effectively constituted the propo-
sitional calculus. Accordingly variables were introduced to denote elementary
propositions (e.g., p, q) and the fundamental logical operators negation (‘¬’),
disjunction (‘∨’), conjunction (‘∧’), and implication (‘→’) were presented and
defined along with primitive propositions. This latter group includes informal
statements of the type ‘Anything implied by a true proposition is true’, and
these were used to enable more complex propositions to be derived. The logi-
cal operators, listed above, were not all assumed to be primitive. Rather, only
disjunction and negation were defined directly, and the definitions of the other
operators were constructed from these. For instance, using the symbol ‘=df ’ to
denote definition, implication can be defined as
p → q =df ¬ p ∨ q (2.4)
indicating that the statement ‘p implies q’ and the statement ‘either not-p or q’
are functionally equivalent. From this minimal basis, essential non-primitive
propositions were then derived. For example, the ‘principle of tautology’ is
asserted (i.e., ‘’) as
 ( p ∨ p) → p (2.5)
and this proposition is associated with the symbol ‘Taut’ in order to facilitate
future reference.
Having established the basic propositional calculus in this manner, White-
head and Russell then introduced the predicate calculus, and this was accom-
plished in numbers *9–*14, where the propositions established for the proposi-
tional calculus were simply extended so that they could be applied to functions
taking variables as arguments (i.e., φ(x)). In addition, the universal operator,
 Logicism 37

‘∀’, was introduced, and statements of the form ‘∀x[φ(x)]’ were taken to mean
‘For all x it is the case that φ(x) is true’. Using the existing formalisms, the
existential operator was then defined in terms of the universal operator and
negation as follows
∃x[φ(x)] =df ¬[∀x [¬φ(x)]] (2.6)
where, as previously, ‘ = df’ indicates that the definiendum is being defined.
With the basics of the predicate calculus established for 1-place propositional
functions, Whitehead and Russell extended the framework so that n-place func-
tions could be handled, and, having introduced the predicate calculus in this
manner, number *20 prepared the ground for many of the later technical devel-
opments by introducing the calculi of classes and relations. ‘Class’ is the word
that is used in PM for Cantorian sets. This term derives from pre-Cantorian
theories of aggregates, and, in the first years of the 1900s, Whitehead and Rus-
sell began to use it as the English equivalent of Cantor’s term ‘Menge’.21 As
defined in PM, classes are understood to be groups of variables that satisfy
propositional functions of the form ‘φ(x)’. Class membership can therefore be
defined as
 x ∈ ẑ[φ(z)] ≡ φ(x) (2.7)
where ‘≡’ indicates bidirectional implication. Here ‘ẑ’ constitutes the class
determined by the function φ, so (2.7) indicates that the statement ‘x is a member
of the class determined by the function φ’ implies the statement ‘φ(x) is true’
(and vice versa). The definition of relation follows on from the function-based
definition of classes since, in PM, relations are understood to specify a ‘class
of couples’. In other words, the variable pair (x, y) is a member of the class of
couples associated with the relation R, so long as the statement ‘x R y’ holds
for x and y. Since any 2-place propositional function determines a relation,
there is a clearly defined connection between functions and relations, and this
connection is captured by the equivalence
 R = x̂ ŷφ(x, y) ≡ x R y ≡x,y φ(x, y) (2.8)
Upon this theoretical basis, the calculus of relations was developed in numbers
*23–*38, during which time various essential propositions were offered in con-
venient shorthand notations. For instance, if the relation R implies the relation
S, then the notation
R ⊂ S = x R y →x,y x S y (2.9)
was used to prevent laborious use of unwieldy symbolism.
38 The consequences of analysis

As PM continued, the central task of deriving the rudiments of mathematics

(i.e., the essentials of number theory) was accomplished. Finite arithmetic was
obtained first (*100–*106) before the leap to the transfinite realm was made
(*118–*126). Eventually the general theory of series was presented (*200–
*276), which led to the introduction of vector families (*330–*375). The whole
project had begun as a simple attempt to provide a revised version of Russell’s
Principles of Mathematics and, after ten years of continuous labour, it had
resulted in the publication of three monumental volumes.
The general reception of PM was complex, and some specific aspects of its
influence and subsequent development will be discussed in chapter 3. It is only
necessary here to indicate that during the 1920s and 1930s many researchers
attempted to simplify and/or extend the system Whitehead and Russell had con-
structed. After PM and the foundational debates of the 1920s (discussed in the
following sections), the desire to derive the whole of mathematics from logic
waned rather, though research into logic as an independent discipline continued
unabated. From the 1950s onwards various non-classical logics such as condi-
tional logic, fuzzy logic, and quantum logic began to receive more attention, and
there were numerous attempts to use techniques derived from modern symbolic
logic to analyse natural language (as discussed in chapter 3).22 Nevertheless,
despite the countless developments and changes that have occurred since its
first appearance, PM remains an astonishing testament to the ambition of the
Logicist movement in the early twentieth century.

2.6 Formalism
The foundational movement that came to be known as Formalism, and which
came to be viewed as an alternative to Logicism, was associated primarily with
David Hilbert, one of the most influential mathematicians of the late nineteenth
and early twentieth centuries.23 Hilbert’s early work was devoted to a range of
topics in number theory, analysis, and algebra, and his interest in foundational
issues was not signalled until the publication of his Foundations of Geometry
in 1899. In this booklet, Hilbert attempted to provide a viable axiomatic foun-
dation for geometry, just as Russell had attempted to do two years previously.
General dissatisfaction with the existing axiomatic-deductive geometrical sys-
tems (especially Euclid’s Elements) had been provoked during the nineteenth
century by the proliferation of non-Euclidean geometries which substantially
undermined the role of spatial intuition as a means of validating geometri-
cal arguments. By contrast with the classical Euclidean methodology, Hilbert
endeavoured to remove all latent remnants of geometric intuition by exploiting
 Formalism 39

the correspondence between geometry and arithmetic. He argued that geometric

relations could be interpreted as arithmetic relations, in which case the valid-
ity of axiomatic-deductive geometrical systems could be guaranteed without
the need for intuition-based arguments, assuming (of course) that arithmetic
itself was constructed upon a secure basis. This kind of relativistic foundational
approach is reminiscent of the attempts to secure analysis by means of number
theory that were discussed in section 2.3.
Given the ultimate dependence of geometry upon arithmetic (or, more specif-
ically, the theory of real numbers), as advocated by Hilbert in Foundations of
Geometry, it was perhaps inevitable that he should have begun to explore the
axiomatic basis of number theory itself in the early 1900s, and, accordingly, he
considered the implications of such a task explicitly in his 1900 paper ‘Con-
cerning the Concept of Number’. However, it was not until 1904 that he began
to address the issue of mathematical foundations in general, and he seems to
have been galvanised into action primarily by the paradoxes of set theory that
had been collected and discussed by Russell in his Principles of Mathematics,
though Hilbert had been familiar with some of the difficulties since the late
1890s.24 While agreeing with Russell that the paradoxes undermined set the-
ory as currently formulated, Hilbert dismissed the assertion that they could be
eliminated only by deriving mathematics from a small set of logical axioms.
The Logicist research programme was misguided, Hilbert maintained, primar-
ily because logic utilises various mathematical concepts that are later derived
from it, thus inducing a fatal circularity:

Arithmetic is often considered to be part of logic, and the traditional fun-

damental logical notions are usually presupposed when it is a question of
establishing a foundation for arithmetic. If we observe attentively, however,
we realise that in the traditional exposition of the laws of logic certain funda-
mental arithmetic notions are already used, for example, the notion of set and,
to some extent, also that of number. Thus we already find ourselves turning
in a circle, and that is why a partly simultaneous development of the laws of
logic and of arithmetic is required if paradoxes are to be avoided. (Hilbert
1967b[1904]: 131)

This quotation is taken from Hilbert’s 1904 paper ‘Concerning the Foundations
of Logic and Arithmetic’, which is often regarded as the earliest statement of his
Formalist manifesto, and there is no doubt that in this paper Hilbert introduced
several of the key ideas that were to dominate his mature foundational work.
During the 1910s Hilbert was enchanted by PM and started to write more
enthusiastically about logic as a result. In particular, he came to admire the
powerful symbolic language that Whitehead and Russell had developed in
40 The consequences of analysis

order to facilitate their logical deductions.25 Despite his augmented appreci-

ation, though, Hilbert continued to maintain that the Logicist movement was
flawed due to the aforementioned circularity inherent in the strategy it adopted.
However, during this period he felt compelled not only to demonstrate the
weaknesses of the renewed Logicist agenda, but also to invalidate the Intu-
itionist arguments (discussed in section 2.7) that were beginning to permeate
the consciousness of the international mathematical community. Goaded into
activity, therefore, by these alternative foundational movements, Hilbert began
to present, with greater clarity, his own proposal for salvaging classical mathe-
matics from the paradoxes of set theory. As a result, in a series of publications
that appeared during the years 1918–1934, frequently aided by his assistant Paul
Bernays (1888–1977), Hilbert developed his ‘Beweistheorie’ (i.e., ‘proof the-
ory’), which was intended explicitly to define his Formalist position concerning
the question of foundations. As Hilbert’s theory evolved over the years, many
of the technical details altered, but the underlying principles remained fairly
constant. Therefore, rather than attempting to provide a superficial overview of
the complete life-cycle of the theory, one particular mature expression of it will
be considered in some detail here in order to convey Hilbert’s main aims and
strategies. The version of the theory discussed will be that presented in the 1927
paper ‘The Foundations of Mathematics’. The exposition Hilbert offered in this
paper is comparatively lucid and reveals many of the abiding concerns that were
later to be distorted and exaggerated in countless more extreme accounts, some
of which will be considered in section 2.8.
‘The Foundations of Mathematics’ begins with a clear statement of intent
that effectively constitutes a non-technical overview of the method developed
in the whole paper:

I should like to eliminate once and for all the questions regarding the foun-
dations of mathematics, in the form in which they are now posed, by turning
every mathematical proposition into a formula that can be concretely exhibited
and strictly derived, thus recasting mathematical definitions and inferences in
such a way that they are unshakable and yet provide an adequate picture of
the whole science. (Hilbert 1967a[1927]: 464)

This passage clearly indicates that Hilbert’s proof theory involved two related
tasks. First, a procedure was required that enabled ‘every mathematical propo-
sition’ to be converted into a ‘formula’, then it must be demonstrated that the
formulae obtained could be ‘strictly derived’. The first task stipulates that math-
ematical statements must be formalised (i.e., converted into strings of precisely
defined symbols) so that mathematics as a whole can be viewed simply as
 Formalism 41

‘an inventory of formulae’ (Hilbert 1967a[1927]: 465), and more will be said
about the process of formalisation later. The second task involves the deriva-
tion of the formulae within a given system. The overriding concern here is
with the nature of the proof techniques that are utilised, hence Hilbert’s use
of the compound noun ‘Beweistheorie’. Obviously, since this task involves the
manipulation of strings of symbols that represent mathematical propositions, it
can be said to be characterised by a certain (not necessarily vicious) circularity:
proof-theoretical mathematical techniques are used to determine the validity of
(suitably encoded) mathematical propositions. It is this apparent self-reference
that caused the second of Hilbert’s tasks to be referred to as ‘metamathematics’;
that is, mathematics about mathematics.
Having delineated his basic intentions at the start of the paper, Hilbert imme-
diately proceeds to introduce the fundamental machinery he requires, and the
three main components he presents are a set of logical operators, a general
proof schema, and a set of axioms. The logical operators are unremarkable,
and they include symbols for implication, conjunction, disjunction, and nega-
tion, as well as universal and existential operators. These are all common to
PM-style systems, though (as shown later) Hilbert adopts definitions for the
last two that differ from those offered in PM. The general proof schema Hilbert
presents, however, is of greater interest. In the paper a mathematical proof is
informally defined (rather unsatisfactorily) as ‘an array that must be given as
such to our perceptual intuition’ (Hilbert 1967a[1927]: 465), and (more help-
fully) it is stated that a proof consists of a sequence of inferences. The specific
proof schema Hilbert presents takes the form

G
G→F
(2.10)
F

and it defines the type of inference that is permitted in proof construction.

The symbols G and F in schema (2.10) indicate formulae (i.e., mathematical
proposition converted into the specified formal symbolism) which are either
(i) axioms in the system, (ii) propositions derived from (i) by substitution,
(iii) the end formulae of a previous valid proof, or (iv) propositions derived from
(iii) by means of substitution. In schema (2.10), G and the implication G → F
are given, with the result that F can be inferred. A formula, F, therefore, is said
to be provable if it is an axiom in the system, or if it is the final formula in a proof.
This methodology enables proofs to be viewed as sequences of logical infer-
ences that enable formulae to be derived within a given axiomatic system. It is
crucial for Hilbert’s project that the procedural definition of a proof, as outlined
42 The consequences of analysis

above, is clear and unambiguous, since, as he states later in the paper, it is

imperative that ‘a formalised proof, like a numeral, is a concrete and surveyable
object’ (Hilbert 1967a[1927]: 471). It is the property of being ‘surveyable’
that is so important: if a proof cannot be checked in an infallible manner, then
mathematics cannot be raised upon a secure proof-theoretical foundation.
The axioms, mentioned above, that Hilbert introduces in his paper are sub-
divided into six main categories:
r Group I: Axioms of Implication (e.g., A → (B → A))
r Group II: Axioms of Conjunction and Disjunction (e.g., ( A ∧ B) →
A)
r Group III: Axioms of Negation (e.g., ¬¬A = A)
r Group IV: The -axiom: A(a) → A((A))
r Group V: Axioms of Equality (e.g., a = a)
r Group VI: Axioms of Number (e.g., a  = 0, where ‘a  ’ means ‘the
number following a’)

The axioms in groups I–IV are referred to as ‘the logical axioms’, while those
in groups V–VI are called ‘mathematical axioms’ since they involve number-
theoretic concepts. Once again, this highlights the difference between Formal-
ism and Logicism: Hilbert assumes that certain mathematical objects, such
as the numeral ‘1’, are pre-theoretical, existing in the intuition as a thought-
object (‘Gedankeding’), while Whitehead and Russell seek to derive even such
basic objects from the principles of logic. The axiom group that demands most
attention is group IV, which contains the -axiom. This axiom is required in
order to enable transfinite arithmetic to be incorporated within the basic proof-
theoretical framework. More practically, it also permits the vague notions of
‘all’ and ‘there exists’ to be defined. As shown above, the axiom takes the form
‘A(a) → A((A))’, where (A) denotes an object for which the proposition
A(a) holds, if it holds for any proposition at all. Consequently, the universal
and existential operators can be defined as follows:

∀x[A(x)] ≡ A((¬A)) (2.11)

∃x[A(x)] ≡ A((A)) (2.12)

These definitions should be compared with those used by Whitehead and Russell
in PM, and included in (2.6) above. As mentioned previously, Hilbert’s use of
the -axiom (which uses the function (x)) enabled him to construct proofs for
transfinite arithmetic in accordance with the strictures of proof theory. However,
while the -function facilitated the construction of such proofs, Hilbert often
 Formalism 43

sought to eliminate the function at a later stage in a given derivation. In order

to accomplish this he made extensive use of recursion and recursive functions.
Since this aspect of Hilbert’s work was partly responsible for the advancement
of recursive function theory in the early twentieth century, it will be considered
in more detail in that context in section 3.3.
Armed with his set of operators, his proof schema, and his axioms, Hilbert
was now able to address the issue of proof construction. The central task was
to construct a metamathematical proof that would demonstrate the validity of
a given axiom set. For Hilbert, the validity of such a set was always closely
related to the technical notions of completeness and consistency. The require-
ment of completeness simply demands that all well-formed formulae, derived
within a given system, can be shown to be either true or false. As for the require-
ment of consistency, from a proof-theoretical perspective, a given axiom set is
considered to be consistent if no formulae taking the form ‘a = a’ can ever be
derived. In other words, a consistent axiom set will never allow contradictions to
be proved. The task of proof theory in part, therefore, is to secure the axiomatic
system underlying the whole of mathematics by establishing its consistency.
The nature of this task, for Hilbert at least, was very different from the task
of converting mathematical propositions into formal strings of symbols. As he
states in the 1927 paper,

To prove consistency we therefore need only show that 0 = 0 cannot be

obtained from our axioms by the rules in force as the end formula of a proof,
hence that 0 = 0 is not a provable formula. And this is a task that funda-
mentally lies within the province of intuition, just as much as does√ in con-
tentual number theory the task, say, of proving the irrationality of 2 (Hilbert
1967a[1927]: 471)

Statements such as this are not atypical. Hilbert repeatedly emphasised the
contentual nature of the metamathematical aspects of proof theory. For instance,
in a 1922 paper, while providing an overview of proof theory, he observes:

In addition to this proper mathematics, there appears a mathematics that is to

some extent new, a metamathematics which serves to safeguard it by protecting
it from the terror of unnecessary prohibitions as well as from the difficulty of
paradoxes. In this metamathematics – in contrast to the purely formal modes of
inference in mathematics proper – we apply contentual inference; in particular,
to the proof of the consistency of the axioms. (Hilbert 1998[1922]: 212)

The emphasis here is absolutely clear: although formal (i.e., meaning-less)

methods may be used in mathematics proper, such methods cannot be used
during the metamathematical stage of analysis, indicating that, for Hilbert at
44 The consequences of analysis

least, proof theory was considerably more than a game involving the manipu-
lation of meaningless symbols. Statements such as the above, with their focus
upon the differences between formalisation and metamathematical analysis,
should be recalled when the nature of Hilbertian Formalism is considered.
A common misconception presents Hilbert as wanting to reduce the whole of
mathematics to a contentless exercise in symbol manipulation that is performed
in accordance with clearly defined rules. From this perspective, in the Formal-
ist game it is the relationship between the strings of symbols that is crucial,
and the meaning either of the symbols themselves or of the strings they form is
deemed to be irrelevant. This misconstrual of Hilbert’s programme is partly due
to the practice of extracting certain of his comments from out of their imme-
diate context. For instance, as mentioned above, part of Hilbert’s contribution
in his Foundations of Geometry was to demonstrate that the meaning of the
geometrical objects he considered need not be accommodated in order to anal-
yse them coherently. In other words, statements about lines, points, and planes
could just as readily be interpreted as statements about arithmetic objects, or,
as Hilbert allegedly put it ‘tables, chairs and beer-mugs!’ (quoted in Grattan-
Guinness 2000: 208). However, this conventional misinterpretation of Hilbert’s
programme is also the result of his distinction between the formalisation process
and the metamathematical process being ignored. On numerous occasions, for
instance, Hilbert emphasised that the task of converting mathematical propo-
sitions into a formal symbolic language was a mechanical procedure that did
not rely upon considerations of meaning. In his 1927 paper, for example, he
states that ‘in my theory, contentual inference is replaced by the manipulation
of signs according to rules’ (Hilbert 1967a[1927]: 467). Although this obser-
vation refers only to the pre-metamathematical stage of analysis, comments
such as these (when extracted out of context) appear to suggest that it is the
formal relationships between strings of symbols that matter, not the meaning
of the strings themselves, even during the metamathematical manipulations
of these strings. It was the (mis)perceived extremity of this emphasis on the
formal properties of mathematical statements that caused Brouwer to refer to
Hilbert’s programme as ‘formalisme’. However, as demonstrated above, Hilbert
was never so extreme in his own brand of Formalism, and this observation has
caused some commentators to recommend the avoidance of the term when dis-
cussing Hilbert, or at least to insist upon an accurate definition.26 However, as
discussed in section 2.8, it was a caricatured version of Hilbert’s original theory
that was popularised throughout North America and Europe during the 1930s
and 1940s, and which ultimately influenced the development of syntactic theory
in the twentieth century.
 Intuitionism 45

The fate of Formalism, or, more specifically, of proof theory, is well known.
In 1931 the young Kurt Gödel published an incompleteness theorem which
demonstrated that, if a formal system is strong enough to prove theorems from
basic arithmetic, then there will always be theorems that are true but which
cannot be proved within the system. In other words, Gödel demonstrated that
the criterion of completeness was a chimera, and this proof appeared to invali-
date the Formalist approach to the foundations problem. Nevertheless, despite
Gödel’s results, a number of mathematicians have continued to work within
the general framework of proof theory and, as shown in chapter 3, the philos-
ophy behind the theory has exerted a profound influence over many different
disciplines.

2.7 Intuitionism
The third major foundational movement, which became known as Intuition-
ism, was originally associated with the Dutch mathematician Luitzen Brouwer
(1881–1966), and it was intended to provide a valid alternative both to Logi-
cism and (incipient) Formalism.27 Brouwer signalled his preoccupation with
foundational issues as early as 1907, when he submitted a doctoral thesis on the
subject, Concerning the Foundations of Mathematics. Although he later refined
considerably the ideas presented in this early work, the thesis nevertheless out-
lines the basic preoccupations that remained remarkably constant throughout
his long career. For instance, in chapter 2 of his thesis, Brouwer explicitly rejects
the assumption that mathematical objects and the symbols used to express them
are equivalent, arguing instead that thought and language are largely separate.
So crucial was this assumption to the whole Intuitionist enterprise that Brouwer
was later to refer to it as ‘the first act of Intuitionism’ (Mancosu 1998: 8–9). The
inevitable result of this act was that he came to view mathematics primarily as a
process of isolated mental construction, accomplished by the individual math-
ematician, which must then be (imperfectly) communicated to others by means
of language, either a natural language or, most frequently in the mathematical
literature, some kind of formal symbolic language. Consequently, intuition is
understood to provide the foundation for all of mathematics, and the Logicist
and Formalist programmes, with their various ways of prioritising linguistic
forms, are deemed to be utterly misguided since they consistently fail to recog-
nise this crucial fact.
While Brouwer’s characteristic approach to the foundations problem was
outlined in his doctoral thesis, he seems to have felt that he had to establish
himself as a leading mathematician if his nascent Intuitionist philosophy were
46 The consequences of analysis

ever to be taken seriously by the international mathematical community. Conse-

quently, during the years 1908–1912 he swiftly turned himself into the leading
expert in algebraic topology, publishing around forty papers in the space of
four years. As a result, his reputation as a leading mathematician was quickly
secured.28 Now, with his professional career established, he promptly returned
to his abiding preoccupation with the task of elaborating an intuition-based the-
ory of mathematics, and this return to foundational considerations marked the
start of his sustained attempt to develop Intuitionism as a valid philosophical
position. The public renewal of the Intuitionist programme effectively began
on 14 October 1912, when Brouwer delivered a paper entitled ‘Intuitionism
and Formalism’. This inaugural address largely restated the case as outlined in
his doctoral thesis, though, to some extent it anticipated future developments
by focusing more of the discussion upon set-theoretical concerns. The research
programme (re)initiated by the presentation of this paper continued unabated
until 1928, and, as it gathered momentum, it began to arouse significant disquiet
and, occasionally, acrimonious controversy. The main cause of complaint was
that Brouwer’s reformulation of the foundations problem jeopardised the valid-
ity of significant portions of classical mathematics. Certain individuals began
to sense that the old mathematical establishment was beginning to crumble,
and, if Hermann Weyl (one of Brouwer’s early converts) was to be believed,
‘Brouwer, das ist die Revolution!’29 (quoted in van Stigt 1990: 71).
During the 1920s Brouwer began to articulate how, in his markedly solipsistic
philosophy, mathematics could be viewed as the task of creating objects in the
individual consciousness by means of constructive procedures, and he accom-
plished this mainly by devising specific techniques that could be used within a
strict Intuitionistic framework. For instance, although his attitude towards logic
had been wholly negative during the 1910s, he became more accepting during
the 1920s, though he still refused to countenance full-blown PM-style classical
logic. Perhaps his most provocative stance in this respect was characterised by
his rejection of the so-called ‘principle of the excluded middle’. This principle
had played a fundamental role in logic since Aristotle, and, essentially, it claims
that all propositions are either true or false, a claim which in turn implies that
all mathematical statements can be shown to be either true or false. Brouwer
rejected this assumption and, as a result, developed what has since become
known as ‘intuitionist logic’.30
In addition to his reformulation of classical logic upon intuitionistic prin-
ciples, Brouwer also concerned himself with the nature of the real number
continuum, and came to despise the standard set-theoretic definition that had
emerged from the work of Weierstrass, Dedekind, and Cantor. For Brouwer,
 Evangelism and pedagogy 47

the continuum was a primitive concept that could not be constructed from more
elementary entities. In particular, it could certainly not be built up from sets
of discrete points, and therefore was not an arithmetic manifold of real num-
bers as Cantor (and others) had supposed. Consequently, in a series of papers
published in the 1920s, he began to devise an intuitionistic version of set the-
ory based upon a dynamic conception of continuity in which points on the
real number line were characterised as convergent sequences of nested inter-
vals that were generated by the free choice of the individual consciousness. In
order to achieve this ambitious project, Brouwer sometimes extended existing
set-theoretical techniques and sometimes devised his own radical new meth-
ods, many of which outraged various sections of the conservative mathematical
establishment.31
While Brouwer’s labours impressed many, the unwieldy nature of the con-
structive procedures demanded by an unremitting adherence to Intuitionistic
principles dissuaded all but a small coterie of devoted followers (the foremost
of whom was Arend Heyting (1898–1980)) from adopting Brouwer’s meth-
ods as practical tools, and there is no doubt that Brouwer was discouraged
by this lack of general acceptance. His disquiet increased in the 1920s when
the Formalism–Intuitionism debate became a personal feud between himself
and Hilbert, resulting in his dismissal by the latter, in 1928, from the edito-
rial board of the journal Mathematische Annalen (Mathematical Annals). After
these final disputes Brouwer retreated into relative obscurity, from which he
did not emerge until the 1950s, when he toured the world as an itinerant lec-
turer, still preaching the Intuitionist gospel. These lectures seem to have been
received by his audiences with considerable warmth, although he was mainly
viewed as a curious relic, a remarkable remnant of an earlier age when anxieties
about the foundations of mathematics were obsessively debated by professional
mathematicians, logicians, and philosophers alike.

2.8 Evangelism and pedagogy

Having considered the work of the main participants in the foundational debates
of the early twentieth century, it is now necessary briefly to discuss some of
the later publications that communicated many of the central arguments sum-
marised in the preceding sections to a wider audience. The survey offered
here makes no claim to be exhaustively comprehensive. Indeed, such was
the proliferation of books and articles concerning the perceived foundations
crisis that appeared during the years 1910–1960, that several sizable tomes
could be devoted solely to the task of exploring the way in which the work
48 The consequences of analysis

of Russell, Hilbert, Brouwer, and others was interpreted, reformulated, modi-

fied, and utilised for evangelical and pedagogic purposes by later generations.
Consequently, some kind of selection process is required, and the texts referred
to in this section simply happen to be those that are cited most frequently in
linguistics papers published during the first half of the twentieth century. As
a result, the following discussion intentionally anticipates some of the main
themes of chapter 3, functioning simultaneously as a postlude to the current
chapter and a prelude to the next.
One of the first textbooks that attempted to provide a detailed summary of
pre-1910 developments in various branches of mathematics was John Young’s
(1879–1932) Fundamental Concepts of Algebra and Geometry. The text was
published in 1911, but it was based upon a series of lectures that had been
delivered at the University of Illinois in 1909. Consequently, Young was not
able to consider the implications of PM, the first volume of which appeared
in 1910, but he did provide a thorough introduction to a wide range of topics,
including Euclidean and non-Euclidean geometry, logic, set theory, number
theory, and numerous other subjects. He openly declared that his primary aim
was to provide ‘an elementary account of the logical foundations of algebra
and geometry’ (Young 1911: v), a remark that possibly indicates some kind of
sympathy with the Logicist movement, and he repeatedly stresses the fact that
mathematical propositions are ‘logically connected’ (Young 1911: 1). However,
he also admits that throughout the book he has adopted a ‘formal point of
view’ (Young 1911: v), and certainly his knowledge of Hilbert’s proto-Formalist
work is revealed in chapters 13 and 14, when he discusses Hilbert’s axiomatic
approach to geometry in some detail. In this context it is striking that by 1911
the task of providing a logical foundation for specific branches of mathematics
was already associated with the Formalist programme.
Although, as mentioned above, Young’s text was not able to provide a discus-
sion either of PM or of Hilbert’s mature proof theory, other textbooks eventually
appeared that covered these topics in considerable detail. Often these publica-
tions were aimed at advanced undergraduates, or postgraduates in their first
years of research, and they attempted primarily to summarise existing results in
symbolic logic and metamathematics in a coherent fashion. An early example of
this type of textbook was Quine’s Mathematical Logic, which first appeared in
1940 (reprinted in 1951) and which provided a clear overview of the rudiments
of PM-style symbolic logic, including chapters on statements, quantification,
the theory of classes, relations, numbers, and logical syntax. As will be shown
in section 3.7, the theory of logical syntax developed considerably during the
1930s and 1940s, largely due to the influence of Carnap, who refined ideas
 Evangelism and pedagogy 49

that Hilbert had introduced. Quine’s interest in the syntax of formal languages
is apparent throughout his book; he devoted a whole chapter to the topic, and
his presentation is conspicuously influenced by Formalism. For instance, while
discussing the syntax of formal logical systems, he introduces an alphabet of
primitive symbols and adds, with reference to strings formed from this basic
symbol set, ‘all these characterizations are formal systems, in that they speak
only of the typographical constitution of the expressions in question and do
not refer to the meanings of these expressions’ (Quine 1940: 283), and it is
this emphasis on the form, rather than the content, of symbolic expressions that
reveals the influence of Hilbert’s proof theory (as mediated by Carnap). As will
become apparent below, such observations became the norm in introductory
texts of this kind during the 1940s and 1950s.
Another presentation of the rudiments of symbolic logic appeared when
Church published his Introduction to Mathematical Logic: Part 1 in 1944 – a
text that (re)appeared in a revised and expanded form in 1956. Like Quine,
Church presented the same basic topics which, by the mid-1950s, were rapidly
becoming an incantational mantra (i.e., primitive symbols, variables, quanti-
fiers, propositional calculus, first-order predicate calculus, second-order predi-
cate calculus, and so on), and his Formalist agenda is expressed unambiguously
when he observes that ‘traditionally, (formal) logic is concerned with the anal-
ysis of sentences or of propositions and of proofs with attention to the form in
abstraction from the matter’ (Church 1956: 1). It is revealing that the task of
using logic to analyse a proof could already be described as traditional, since
this indicates the speed with which Hilbert’s metamathematical programme
had been accepted. Church goes on to discuss the specific topic of the syntax of
logical systems in a subsection of his introduction, and he draws a distinction
between natural and formal languages. Like Russell, Church felt that natural
languages were far more complicated than formal (i.e., artificial) languages
since they had ‘evolved over a long period of history to serve practical pur-
poses of facility of communication’ (Church 1956: 3). Consequently, when
Church used the term syntax, he was referring specifically to the syntax of for-
mal languages, rather than natural languages, and he made a further distinction
between ‘elementary’ and ‘theoretical’ syntax. Elementary syntax is concerned
with ‘setting up the logistic system and with the verification of particular well-
formed formulas, axioms, immediate inferences, and proofs’, while theoretical
syntax constitutes ‘the general mathematical theory of a logistic system or sys-
tems and is concerned with all the consequences of their formal structure (in
abstraction from the interpretation)’ (Church 1956: 59). Clearly, this approach is
rooted in Hilbert’s proof theory, since the distinction is essentially that between
50 The consequences of analysis

mathematics proper (i.e., proving theorems in a logical system) and metamath-

ematics (i.e., proving theorems about a logical system). Note, though, that, for
Church, the ‘interpretation’ of the system is irrelevant even during the meta-
mathematical stage, while, as indicated in section 2.6, for Hilbert at least, mean-
ing was involved in metamathematical considerations. Church’s recommended
methodology is an example of the sort of extreme Formalism that came to domi-
nate in the 1940s, at the expense of Hilbert’s more cautious and subtle approach.
It is also significant that, with reference to formal systems, Church remarks that
‘like any branch of mathematics, theoretical syntax may, and ultimately must,
be studied by the axiomatic method’ (Church 1956: 59), since, as will be shown
in section 3.2, linguists (particularly syntacticians) became increasingly inter-
ested in the axiomatic-deductive method during the first decades of the twentieth
century, and the method was ultimately to influence the development of TGG
considerably.
Although, as indicated above, Church’s text draws upon Formalism, while
introducing the rudiments of symbolic logic, it is not primarily intended as an
introduction to proof theory. However, an authoritative exposition of the theory
did appear in 1952, when Kleene published his Introduction to Metamathemat-
ics. Kleene’s debt to Hilbert is clear from the title alone, and, since the book is
intended to be an introduction to proof theory in its entirety, it covers more top-
ics than the introductions to logic discussed above. For instance, in addition to
the obligatory chapters on propositional and predicate calculus, it also contains
sections on set theory, formal system theory, and the general theory of recursive
functions. As expected, the need to avoid meanings, or ‘interpretations’, when
manipulating strings of symbols in a formal system, is repeatedly stressed. For
instance, while introducing the notion of a formal system, Kleene remarks that
it must be ‘described and investigated, by finitary methods and without making
use of an interpretation of the system’ (Kleene 1952: 69). Later, the above advice
is restated in a more elaborate (and rather ungainly) form when the primitive
symbols of the system are introduced:

We reiterate that the interpretations are extraneous to the description of the

formal system as such. It must be possible to proceed regarding the formal
symbols as mere marks, and not as symbols in the sense of symbols for
something which they symbolize or signify. (Kleene 1952: 70)

So, the symbols in a formal system must be manipulated without worrying

about that which they denote, and, just in case there could be any lingering
doubts, on the very next page Kleene stresses again that ‘in metamathematics
we must treat the formal symbols as meaningless’ (Kleene 1952: 71). These
 Evangelism and pedagogy 51

passages indicate that, like Church’s presentation, Kleene’s text can be viewed
as yet another example of the more extreme Formalist position that came to
characterise pedagogic texts during the 1940s and 1950s.
One other book, also published in 1952, that deserves comment in this context
is Raymond Wilder’s Introduction to the Foundations of Mathematics. While
Church and Kleene were concerned with presenting the technical apparatus of
symbolic logic and proof theory respectively, reserving an historical overview
of the development of these theories only for footnotes and asides, Wilder
wanted to provide a detailed historical survey of the whole foundations crisis.
Consequently, in addition to the expected topics, his text includes chapters on
the axiomatic-deductive method, set theory, number theory along with separate
sections on Logicism, Intuitionism, and Formalism. It is of particular interest
that Wilder’s book was directly inspired by Young’s text, discussed above. As
Wilder explains in his introduction,

In a general way, the idea of the book is similar to that which motivated J.
W. Young’s Fundamental Concepts of Algebra and Geometry, first published
in 1911. In 1932 I discussed with Professor Young the desirability of a book
such as this one; he agreed thoroughly that it was desirable to write it, if only
to have available a book on foundational concepts that will take into account
the great strides that have been made in Foundations since the publication of
his book. (Wilder 1952: vi)

Clearly, then, by the early 1950s there was an interest not only in the math-
ematical techniques that emerged from the foundations crisis, but also in the
cultural and intellectual history of the period, and Wilder’s influential text was
consciously designed to cater for this need.
The texts mentioned so far are all either mathematical or logical textbooks
that attempted to provide insights into fundamental results in logic and meta-
mathematics. However, other publications began to appear during the 1940s and
1950s that emphasised the wider utility of some of the techniques developed
in these areas. For instance, a number of texts attempted to argue that logic
could be used to facilitate the analysis of natural language. Two influential texts
of this type are Hans Reichenbach’s Elements of Symbolic Logic (1947) and
Paul Rosenbloom’s Elements of Mathematical Logic (1950). Reichenbach’s
text is a curious mixture of conventional expository sections that introduce
standard topics, and idiosyncratic sections that are mainly concerned with his
own research interests. The most significant chapter from the viewpoint of
linguistic theory is chapter 7, entitled ‘Analysis of Conversational Language’,
in which Reichenbach outlines his plans for ‘a logistic grammar’, which stems
52 The consequences of analysis

from ‘the desire to connect logic with the natural use of language’ (Reichenbach
1947: vi). Although the details of Reichenbach’s system need not be considered
here, it is important to note that his basic intention was to create a grammar
for the English language based upon propositional functions. For instance, he
observed that the declarative sentence ‘x loves y’ could be represented by
the propositional function ‘ f (x, y)’, where ‘ f ’ corresponds to the verb ‘love’,
and his interest in this sort of function-theoretic grammar was motivated by
a dissatisfaction with traditional subject-predicate analysis. Indeed, Reichen-
bach claimed that the standard subject-predicate analysis ‘does violence to the
structure of the sentence’ (Reichenbach 1947: 252). As an example, he con-
siders the sentence ‘Peter is taller than Paul’ and claims that this structure
is inadequately analysed by classifying ‘Peter’ as subject and the rest as the
predicate, since clearly, in this example, there is some kind of functional sim-
ilarity between ‘Peter’ and ‘Paul’, and therefore this similarity (as well as the
obvious functional differences) should be captured by a sufficiently detailed
analysis. He goes on to claim that a grammar based on propositional functions
like those mentioned above could capture these structural similarities more
accurately than subject-predicate analysis. Although Reichenbach’s work was
never developed extensively by professional linguists, his claim that natural
language and logic are more compatible than was commonly supposed at the
time intrigued mathematically minded linguists during the 1950s, as will be
shown in section 3.4.
While not as radical as Reichenbach’s text, Rosenbloom’s small volume
also implied that the syntax of formal logical systems and the syntax of natural
language are essentially identical. His main discussion of this issue is contained
in chapter 4, entitled ‘The General Syntax of Language’, and the definition of
a language that he offers is certainly sufficiently broad to include both formal
and natural languages.

A language consists of certain signs, and certain strings of these signs. Its
syntax consists of rules for classifying and transforming these strings. The
alphabet of a language consists of certain basic signs, usually in finite number
[sic]. By a string we mean a finite sequence of signs. A string is exhibited by
writing its signs in linear order from left to right. (Rosenbloom 1950: 152)

Once again the Formalist emphasis on inscription (i.e., the writing of finite
sequences of signs) is apparent here, and, in this respect, the definition given
above is not particularly unusual. What is of interest, though, is the way in
which Rosenbloom goes on to discuss the differences between formal and
 Evangelism and pedagogy 53

natural languages. For instance, he implies that these two types of language
differ in degree rather than in kind:
As in all natural languages, including Esperanto, the rules of word and sen-
tence formation in English are so complicated and full of irregularities and
exceptions that it is almost impossible to get a general view of the structure
of the language, and to make generally valid statements about the language.
It is for this reason that mathematicians and logicians prefer to work with
languages like L3 [a formal language defined earlier in the text] with very
simple and regular structures. (Rosenbloom 1950: 153)

The implication, then, is that, although English and other natural languages are
bafflingly complex, and although mathematicians and philosophers find it easier
to work with artificially constructed languages, there is no reason in principle
why natural languages cannot be subjected to the same kinds of formal analysis
as their artificial cousins. As will be shown in chapter 3, this is exactly the sort
of approach that certain linguists gradually adopted in the the 1940s and 1950s,
and it is no coincidence that Chomsky used Rosenbloom’s text as one of his
main sources of information concerning formal syntax during the early 1950s.
3 Mathematical linguistics

3.1 Chapter overview

The purpose of this chapter is to summarise some of the developments, asso-
ciated with the formal sciences, that grew out of the foundations debates of
the early decades of the twentieth century, and to examine the way in which
they exerted an influence upon the formulation of linguistic theory. The basic
strategy is to focus upon particular techniques or theories that were ultimately
to be involved in the creation of TGG. Consequently, it should be remembered
throughout that this chapter is necessarily selective, and that it does not attempt
to provide an exhaustive coverage of all the associations between mathematics
and linguistics that were mooted during the first half of the twentieth century.
The first subject, discussed in section 3.2, is the use of the axiomatic-deductive
method, and detailed attention is given to the work of Bloomfield, Bloch, and
Harwood. Recursive function theory is considered next, in section 3.3, with
specific reference to the work of Gödel, Kleene, Post, and Bar-Hillel. In sec-
tion 3.4 the work of the Lvov-Warsaw school of logicians is assessed, and
particular emphasis is placed upon the way in which some of Ajdukiewicz’s
research into logical systems was revived by Bar-Hillel in the early 1950s.
The evolution of constructional system theory is considered in section 3.5,
and the associations between Carnap and Goodman are explored, while, in
section 3.6, the extreme philosophical stance that came to be known as con-
structive nominalism is presented. The crucial topic of formal syntax is exam-
ined in section 3.7, and the main focus is on the work of Carnap, Bloomfield,
Hjelmslev, and Harris. Finally, in section 3.8 the diverse strands of the chapter
are brought together by considering the general response of linguists work-
ing in the early 1950s to the gradual mathematisation of large parts of their
discipline.

54
 Axiomatics 55

3.2 Axiomatics
As indicated in chapter 2, the debates concerning the foundations of mathemat-
ics that raged during the early 1900s generated considerable interest at the time,
and the details of the various foundational strategies proposed by the main par-
ticipants were discussed in introductory texts such as those by Young, Quine,
Church, Kleene, and Wilder (as summarised in section 2.8), all of which were
accessible to mathematically inclined linguists. Without doubt the most signifi-
cant linguist actually to follow the progress of the foundational debates closely
at the time was Leonard Bloomfield (1887–1949), and the first of Bloomfield’s
publications to reveal the extent of his preoccupation with mathematics was
his 1926 paper ‘A Set of Postulates for the Science of Language’. In this short
paper Bloomfield suggested that linguists should start to use the same basic
axiomatic-deductive method which had transformed the study of arithmetic
and geometry in the nineteenth century. Bloomfield uses the term ‘postulates’
instead of axioms, and, at the start of his paper, he explains why a postulational
approach could benefit linguistics:
The method of postulates (that is, assumptions or axioms) and definitions
is fully adequate to mathematics; as for other sciences, the more complex
their subject-matter, the less amenable they are to this method, since, under
it, every description or historical fact becomes the subject of a new postu-
late . . . Nevertheless, the postulational method can further the study of lan-
guage, because it forces us to state explicitly whatever we assume, to define
our terms, and to decide what things may exist independently and what things
are interdependent. (Bloomfield 1926: 153)

As far as Bloomfield was concerned, then, the axiomatic-deductive method was

of value since it could introduce new rigour (in the Cauchian sense) into lin-
guistics, just as it had been used to render mathematics more exact during the
nineteenth century. The comparison with Cauchy’s rigorisation programme is
not vacuous, since the emphasis in the above passage is upon stating assumptions
‘explicitly’, and determining which aspects of a given theory are ‘interdepen-
dent’ and which can be treated ‘independently’. In this way, Bloomfield appears
to be recommending a reformulation of linguistics that is similar both in spirit
and methodology to the reformulation of the calculus that Cauchy proposed at
the beginning of the nineteenth century (as discussed in section 2.3). In order
to clarify how this new rigorisation programme for linguistics might be accom-
plished, Bloomfield explicitly states later in the paper that, by the judicious
use of axioms, definitions, and deduction, ‘certain errors can be avoided or
corrected by examining and formulating our (at present tacit) assumptions and
56 Mathematical linguistics

defining our (often undefined) terms’ (Bloomfield 1926: 153). In other words,
when compared to more fully developed formal sciences (such as mathemat-
ics), linguistics appeared to be infested with errors that could be avoided if
an axiomatic-deductive approach were adopted, and, in accordance with this
proposal, Bloomfield introduced a set of postulates that could provide a secure
foundation for the whole of linguistics. The particular postulates he introduced
included definitions and assumptions such as1

Definition: An act of speech is an utterance.

Assumption: Within certain communities successive utterances are alike or
partly alike.

It is significant that, although Bloomfield recommended the use of a basic pos-

tulational methodology because it could make linguistics more precise, as these
examples indicate, he did not attempt to introduce a formal language (à la PM)
that would enable the axioms of linguistics to be converted into unambiguous
sequences of precisely defined symbols. However, as will be shown below, this
extension was accomplished by a later generation of linguistics.
The text that Bloomfield cites as the main source of his information con-
cerning the axiomatic-deductive method is Young’s 1911 text Lectures on the
Fundamental Concepts of Algebra and Geometry (discussed in section 2.8), and,
given Bloomfield’s knowledge of this work, it is reasonable to suppose that, by
1926 at least, he was broadly familiar with all the pre-1911 topics summarised
in chapter 2 above, since they are all discussed in Young’s book. In addition
to this direct inspiration from the pedagogical mathematical literature, Bloom-
field was persuaded by the work of the psychologist Albert Weiss (1879–1931)
that mathematical procedures could be usefully employed in the mind-based
sciences. For instance, in a 1925 paper, Weiss had proposed a set of postulates
for psychology, and this attempt at axiomatisation partly inspired Bloomfield’s
proposal for the reform of the methodology of linguistics.2 Clearly, therefore, by
1926 Bloomfield was intrigued by the precise nature of the relationship between
mathematics and cognitive phenomena such as natural language, and, far from
being an ephemeral fad, his interest in this topic seems to have increased during
the years following 1926. For instance, there are various comments concerning
the relationship between language and mathematics in his most famous and
influential book, Language, which appeared in 1933. To take one example, at
one point he refers to mathematics as ‘the ideal use of language’ (Bloomfield
1933: 29), and later declares (rather provocatively) that one of the tasks con-
fronting the practising linguist is to ‘reveal the verbal character of mathematics’
 Axiomatics 57

(Bloomfield 1933: 507). Although Bloomfield does not state explicitly in

Language how such a task could best be accomplished, this remark certainly
suggests that by the early 1930s he had begun to consider the possibility of using
techniques from linguistics in order to analyse mathematics, rather than merely
using mathematical procedures to explore fundamental properties of language.3
Despite Bloomfield’s great influence upon research into natural language, his
acknowledged enthusiasm for the axiomatic-deductive method and, in partic-
ular, his proposed set of postulates for linguistics did not immediately inspire
other linguists to employ similar techniques. Indeed, it was not until the late
1940s that other researchers began to develop the basic methodology he had
outlined in his 1926 paper. For instance, in 1948 Bernard Bloch (1907–1965)
published ‘A Set of Postulates for Phonemic Analysis’, in which he referred
directly to Bloomfield’s earlier work:
Leonard Bloomfield was the first to state explicitly some of the assumptions
that underlie the methods of linguistic science; his formulation of these axioms
. . . has remained for more than twenty years the only attempt of its kind. We
may find it necessary now, in the light of recent theoretical discussion, to
make certain changes of detail in his list of assumptions; but the importance
of his article as a contribution to linguistic theory is undiminished. Whoever
undertakes, in future, to apply the postulational approach to linguistics will
find his task made easier by the model that Bloomfield has provided. (Bloch
1948: 3)

This passage indicates that Bloch’s own attempt to provide a set of axioms
for phonemic analysis was partly inspired by Bloomfield’s work, although his
observation that due to ‘recent theoretical discussion’ the exact system Bloom-
field proposed would have to be altered, indicates that he was aware of the
limitations of the 1926 postulate set. As a result, Bloch himself made several
‘changes of detail’ to Bloomfield’s system. For instance, he rejected Bloom-
field’s assumption that phonemes are actually present in sound waves, adopting
instead the (then) contemporary view that phonemes are abstract linguistic
units. Nevertheless, despite these alterations, the basic postulational method-
ology Bloch used was identical to that proposed by Bloomfield. In particular,
Bloch’s postulates are expressed as statements in English, and these are intended
unambiguously to define all phenomena associated with phonemic analysis. For
instance, Bloch’s first postulate and first definition take the form4

1.1 Postulate 1: There are communities of human beings who interact partly
by the use of conventional auditory signs.
1.2 Definition: Such a community is a speech-community.
58 Mathematical linguistics

As these examples indicate, like Bloomfield, Bloch made no attempt to convert

his postulates into a precisely defined formal language, and was content with a
natural language exposition of his system.
Bloch’s 1948 postulate set for phonemic analysis heralded a renewed inter-
est in the axiomatic-deductive method amongst the North American linguistics
community, and the intensity of this renewal is indicated by the rapid increase in
the number of axiomatic-deductive linguistic theories that began to appear dur-
ing the late 1940s and early 1950s. To consider just one example,5 in 1955 F. W.
Harwood (dates unknown) published a paper entitled ‘Axiomatic Syntax: The
Construction and Evaluation of a Syntactic Calculus’, and it is revealing that,
in this paper, Harwood cites neither Bloomfield’s nor Bloch’s work directly,
declaring instead that the methodology outlined constitutes an attempt to apply
‘ideas from the field of mathematical logic’ in a linguistic context (Harwood
1955: 409). This remark suggests that, while Bloomfield had been inspired pri-
marily by work in geometry (and psychology), and while Bloch had consciously
sought to revive Bloomfield’s method, Harwood had been influenced directly by
advances in symbolic logic, though he does also mention Bar-Hillel explicitly,
demonstrating that he was aware of the contemporaneous interest in adapting
techniques derived from mathematics in order to facilitate syntactic analysis.
The extent of Harwood’s indebtedness to research into logic is apparent from the
kind of terminology and formal machinery utilised in the paper. For instance, his
basic intention is to discuss ‘methods for presenting syntactic information in the
form of a calculus, and for measuring its goodness of fit to a language’ (Harwood
1955: 409). More precisely, he proposes a system that has access to a finite word
list, and which combines the words in this list in various principled ways in
order to generate sentences in a given language. Harwood views sentences as
‘a serially ordered set of positions, 1, 2, . . ., p, each filled by one of the words
on our list’ (Harwood 1955: 410). Consequently, the sentence ‘John discov-
ered the path’ is a 4-positional sentence. Having defined these terms, Harwood
continues:

Now a set of p positions may be filled from a finite list of r words,

w1 , w2 , . . ., wr , in r p ways; we shall call the N possible p-positions sequences
of w1...r . Let L p be a subset of the N possible p-positional sequences of w1...r
which occur as sentences in the language L and −L p (read not L p) the remain-
der. (Harwood 1955: 410)

This formulation indicates that, in Harwood’s scheme, the set of sentences in

language L can be subdivided into two disjoint subsets; namely, the set of
 Axiomatics 59

grammatical sentences L p and the set of ungrammatical sentences −L p. He

continues:

An axiomatic system S operating on w1...r divides the N possible p-positional

sequences of w1...r into a subset K p which are derivable in S, and −K p the
remainder which are not derivable. (Harwood 1955: 410)

In other words, given the binary grammaticality assumption introduced above, it

is the task of an axiomatic syntactic system such as S to divide the set of possible
sentences into two distinct subsets; namely, K p, the set of derivable sequences,
and −K p, the set of non-derivable sequences. Obviously, the more closely
K p approximates to L p, defined above, the better the system S performs. In
practice, Harwood’s theory requires the following elements in order to achieve
its purpose (see Harwood 1955: 410):
r A finite set of words, w1...r , which are divided into classes, C1...m , so
that C1...m can be used as variables of which the words in these classes
are the values.
r An initial set of sequences of C1...m .
r A set of procedure rules for deriving further sequences from the initial
set.

As an example of a procedure rule, Harwood offers the following: ‘Given Ca Ci

as an initial sequence and Ci → C j as a procedure rule we may derive Ca C j ’
(Harwood 1955: 410). This example illustrates the way in which one sequence
can be derived from an existing sequence, and it obviously resembles Hilbert’s
proof schema given as (2.10) above since it uses logical implication in order to
derive a particular sequence. In passing, it should be noted that, while Harwood
discusses the fact that K p should approximate as closely as possible to L p,
he is compelled to address the issue of compactness; that is, he suggests that
the system S should be constructed as economically as possible. Specifically,
he states that ‘compactness is an important and measurable feature’ (Harwood
1955: 411), and this concern for compact grammars will become the focus of
attention in section 4.3 when simplicity criteria are discussed in relation to TGG.
As this brief overview clearly demonstrates, by 1955 expositions written
in natural language outlining axiomatic-deductive systems that could be used
to analyse natural languages were no longer considered adequate by certain
researchers. Taking Harwood as an example, as the above quotations indicate,
he specifically introduced a formal symbolic language that enabled him to state
his initial assumptions more precisely than either Bloomfield or Bloch, with
the result that the degree of ambiguity in the system he develops is significantly
60 Mathematical linguistics

reduced. In addition, it is important to recognise that one of the most revealing

characteristics of Harwood’s system is the assumption that all sentences (i.e.,
word sequences) in a given language can be classified as belonging either to
the subset K p (i.e., the set of derivable sentences) or the subset −K p (i.e.,
the set of all non-derivable sentences). This classificatory system demonstrates
that one of the natural consequences of adopting a strict axiomatic-deductive
approach to syntactic theory is the assumption that grammaticality is a binary,
set-theoretical concept: given the set of all sentences in a given language, each
sentence can be shown to be either grammatical (i.e., derivable) or ungrammat-
ical (i.e., non-derivable), and consequently a given sentence must belong to one
of these two disjoint subsets. Some of the issues involved here will be consid-
ered further in chapters 4 and 5. The main point is that, as Harwood’s work
demonstrates, by the mid-1950s the use of the axiomatic-deductive method in
linguistics, particularly syntax, had been revived by certain linguists, following
Bloomfield’s initial 1926 proposal, and made more precise by the use of formal
symbolic languages, ultimately derived from systems of logic. This advance had
profound consequences for the development of TGG, and, as will be shown in
section 5.7, it was the axiomatic-deductive nature of TGG that bedazzled some
of its earliest commentators.6

3.3 Recursive definitions

As indicated in section 3.2, one consequence of the influence of the axiomatic-
deductive method upon linguistics was the emergence in the 1950s of systems of
‘axiomatic syntax’ (to use Harwood’s phrase), and, given the language-based
emphasis of Hilbert’s axiomatic-deductive proof theory (which in turn drew
upon Whitehead and Russell’s language-based axiomatic-deductive Logicism),
it was perhaps inevitable that other proof-theoretical techniques would eventu-
ally be employed in a linguistic context. As mentioned briefly in section 2.6,
one of the most distinctive features of Hilbertian Formalism was the emphasis
it placed upon the use of finite methods during the process of mathematical
proof construction, and recursive functions came to be viewed as important
tools for this purpose. Consequently, numerous researchers advanced recursive
function theory during the 1930s and 1940s, and in the early 1950s recursive
components began to be explicitly introduced into syntactic theory. Since the
initial TGG formalism utilised such components, it is necessary to consider the
manner in which these ideas migrated from proof theory to linguistics.7 Before
providing explicit definitions of various kinds of recursive functions, though,
it is necessary first to outline the lineage of the main ideas considered in the
 Recursive definitions 61

following paragraphs, since the development of recursive function theory is

convoluted and problematic.
In summary, then, the basic concept of a recursive function had been intro-
duced in the nineteenth century, and such functions were deployed by Dedekind
during his investigations into the nature of the real numbers. In turn, Dedekind’s
use of recursive functions influenced Peano, who used such techniques while
defining mathematical induction (as in Peano 1959a[1889]), and a number
of leading mathematicians and logicians developed the theory further before
Hilbert began to advocate its use within the context of Formalism. However, the
basic theory was endowed with especial significance when Gödel used primi-
tive recursive functions in his celebrated incompleteness theorem of 1931, and
again when he introduced general recursive functions in 1934. Gödel’s work was
advanced by Church, Kleene, Alan Turing (1912–1954), and Emil Post (1897–
1954) during the 1930s and 1940s, and, specifically, it was demonstrated that
recursive function theory, λ-calculus, and computability theory provided alter-
native, but equivalent, formal expositions of the informal notion of effective
calculability. Crucially, the notion of recursively enumerable sets was intro-
duced by Church in 1936, before being reformulated with renewed impetus by
Post in 1944.
While the above overview provides a (very) brief summary of the basic
development of recursive function theory from the late nineteenth century to
the 1940s, it perhaps insufficiently stresses that the history of the theory from
the late 1930s to the present is peculiarly convoluted, since it came to be intri-
cately intertwined both with the notion of λ-definability and with the theory of
computability. While some of the consequences of this conflux are discussed
in the conclusion, the following paragraphs will not consider the relationship
between recursion, λ-calculus, and computability in detail, since the main focus
must be upon Gödelian recursive function theory. The explanation for this is
simple: the notions of induction by definition and recursively enumerable sets,
which were both primarily associated with Gödelian recursive function theory,
influenced the development of syntactic theory most profoundly in the 1950s,
and therefore these aspects of the theory require focused consideration. Having
specified the limits of the ensuing discussion in this manner, it is necessary now
to define recursive functions more precisely.
If a given recursive function is viewed as a number-theoretic function that is
defined upon the domain of non-negative integers, then, informally, the charac-
teristic feature of such a function is that each value it calculates is specified in
terms of previous values that the same function has already calculated. In other
words, recursive functions are associated with a particular kind of self-reference,
62 Mathematical linguistics

and this can be cited as a defining property. In the context of number-theoretic

functions and predicates, then, recursive definitions consist of a pair of equa-
tions, the first of which determines the condition that terminates the recursion,
and the second of which constitutes the recursive step. For instance, consider
the following pair of equations:

φ(0) = q (3.1)

φ(y ) = χ (y, φ(y)) (3.2)

Here q, y, y  ∈ N, where y  indicates the successor of y (i.e., y  = y + 1). In

this pair of equations (3.1) defines the termination condition since, when the
function φ is called with 0 as its argument, it will simply return the natural
number q. By contrast, equation (3.2) defines the recursive step since, if the
function φ is called with the natural number y  as its argument, then the value
for φ(y  ) is calculated by calling the function χ which takes two arguments –
namely, the natural number y (i.e., y  − 1) and the function φ(y). To consider a
concrete example, if add(x, y) = x + y and add(0, x) = x, and ‘add’ is called
in order to sum the numbers 4 and 1, then the value returned by the function is
calculated as follows:

1. add(4, 1) = add(3, 1) + 1
2. add(3, 1) = add(2, 1) + 1
3. add(2, 1) = add(1, 1) + 1
4. add(1, 1) = add(0, 1) + 1
5. add(0, 1) = 1
6. add(1, 1) = 1 + 1 = 2
7. add(2, 1) = (1 + 1) + 1 = 3
8. add(3, 1) = ((1 + 1) + 1) + 1 = 4
9. add(4, 1) = (((1 + 1) + 1) + 1) + 1 = 5

In this example, the initial function call, add(4, 1) triggers a sequence of four
recursive function calls which only terminates in line 5 when add(0, 1) is
reached. Once the recursion has been halted in this manner, the values that
are returned by each recursive call are calculated. Equation pairs of the kind
defined in (3.1) and (3.2) were used by Peano in order to define mathematical
induction, and they have always been at the core of recursive function theory.
As indicated in section 2.6, Hilbert’s interest in primitive recursive functions
was primarily due to the fact that they could replace -functions in mathematical
proofs, and in order to appreciate why this might be desirable, it is worth empha-
sising the relationship between recursive functions and proof by induction.
In mathematics an inductive proof attempts to prove a given theorem by
 Recursive definitions 63

establishing that the theorem holds for the first case and for the n + 1th case.
Since the proof is obtained for both an initial instance and a general succes-
sor, then the proof is understood to hold for all cases, since any case can be
derived from the initial instance by repeated applications of the inductive step.
Consequently, although an inductive proof is finite, it covers an infinite number
of cases. Primitive recursive functions permit the same type of iterated infer-
ence and, consequently, Hilbert believed that such functions could guarantee
the validity of proofs which previously had to utilise -functions. Accordingly,
the notion of recursion was established at the core of Formalism and, as a result,
it received considerable attention from other mathematicians. Indeed, the rela-
tionship between recursive functions and induction is extremely close, to the
extent that recursion can be viewed as definition by induction (e.g., Kleene
1952: 217). Further, the fact that recursive functions enable finite proof tech-
niques to be employed was noted by many researchers, and this was often cited
as a desirable and characteristic property. For example, Gödel stated explicitly
that ‘[r]ecursive functions have the important property that, for each given set
of values of the arguments, the value of the function can be computed by a
finite procedure’ (Gödel 1986b[1934]: 348). One obvious question that arises
from considerations of this kind can be expressed as follows: which classes of
number-theoretic functions can be defined recursively? The answer to this ques-
tion was developed during the 1930s, primarily by Gödel, and later elaborated
by his successors, especially Kleene. One crucial notion that was advanced as
a result of this research was that of primitive recursion. Formally, a function is
classified as being a primitive recursive function if it is definable by a series of
applications of the following equations:8
φ(x) = x  (3.3)
φ(x1 , x2 , . . ., xn ) = q (3.4)
φ(x1 , x2 , . . ., xn ) = xi (3.5)
φ(x1 , x2 , . . ., xn ) = ψ(χ1 (x1 , x2 , . . ., xn ), . . ., χm (x1 , x2 , . . ., xn ))
(3.6)
φ(0) = q (3.7)
φ(y  ) = χ(y, φ(y)) (3.8)
φ(0, x2 , . . ., xn ) = ψ(x2 , . . ., xn ) (3.9)

φ(y , x2 , . . ., xn ) = χ(y, φ(y, x2 , . . ., xn ), x2 , . . ., xn ) (3.10)
The equations above define number-theoretic functions when n and m are
positive integers, i is an integer such that 1 ≤ i ≤ n, q is a natural number,
and φ, ψ, χ , χ1 . . . χm are number-theoretic functions that take the indicated
64 Mathematical linguistics

numbers of arguments. In Kleene’s terminology, a function that satisfies the

equation in (3.3) is called the successor function; a function that satisfies (3.4)
is called a constant function; a function that satisfies (3.3–3.4), or (3.5) is called
an initial function; a function that satisfies (3.6), (3.7–3.8), or (3.9–3.10) is
called an immediate dependent of other functions (e.g., the χ functions). Con-
sequently, a function φ is called a primitive recursive function if there is a finite
sequence of occurrences of functions, φ1 , φ2 , . . ., φk , where k ≥ 1, such that
each function of the sequence is either an initial function, or an immediate
dependent of preceding functions of the sequence, and the last function φk is
the function φ.
While the equations given above can be used to define primitive recursive
functions, Gödel soon realised that there were certain effectively calculable
functions that were not primitive recursive (for instance, functions that were
defined by induction in respect to two variables simultaneously), and, conse-
quently, he defined a wider class of functions called general recursive functions.
In essence, a function φ is a general recursive function with respect to the func-
tions ψ1 , . . ., ψn if there is a system of equations which defines φ recursively
from ψ1 , . . ., ψn .9
Given the definitions of primitive and general recursive functions, a number
of researchers began to consider the consequences of recursive function theory.
For instance, in 1936 Church introduced the idea of a recursively enumerable
set, and this basic idea was influentially expanded by Post in the early 1940s
(especially Post 1944). In essence, a set of non-negative natural numbers, A, is
recursively enumerable if there is a general recursive function φ which enumer-
ates the members of the set. In other words, the set A is recursively enumerable
if φ(0) ∈ A, φ(1) ∈ A, φ(2) ∈ A, . . . where φ is a general recursive function,
and where the sequence constitutes an enumeration of the members of A. From
the perspective of syntactic theory, it is of considerable interest that Post used
the verb ‘to generate’ in order to describe the process of obtaining members
of sets using recursive devices (such as general recursive functions), and was
therefore able to speak of ‘generated sets’ (Post 1944: 286). Critically, Post
viewed the task of enumerating the members of a recursively enumerable set as
a generational problem. The implications of this will be considered when TGG
is discussed in section 5.6.
As with the axiomatic-deductive approaches discussed in section 3.2, during
the 1950s recursive function theory began to be investigated by mathematically
minded linguists who were keen to explore the methodology of the formal
sciences in the hope of finding techniques that could make linguistic research
more rigorous. One of the first explicit discussions of the benefits of recursive
 Recursive definitions 65

definitions in linguistic theory was contained in Bar-Hillel’s short paper ‘On

Recursive Definitions in Empirical Science’ which appeared in 1953.10 As
the title suggests, Bar-Hillel’s main intention in this work was to argue that
recursive definitions need not only be used in formal mathematical situations,
but could also be used beneficially in the context of the empirical sciences. As
an example of elementary recursion, Bar-Hillel considers Peano’s Immediate-
Successor (IS) function, which he had encountered in Kleene’s Introduction
to Metamathematics,11 and, following Kleene, Bar-Hillel provides a two-part
definition of this function:

Definition 3.1: The IS Function (Recursive)

1. a + 1 = a 
2. a + n  = (a + n)

Definition 3.1 can be directly compared with the equation pair (3.1)–(3.2) above,
since, in both cases, the recursion is induced by the second step. Having con-
sidered the basic implications of adopting such definitions in a mathematical
context, Bar-Hillel goes on to suggest that, in the empirical sciences, defini-
tions have often included statements that are actually recursive, even though
this property has not been explicitly acknowledged. In order to illustrate his
point he states that, ‘for a change’ (Bar-Hillel 1953a: 163), he will consider
several examples from linguistics. As implied, this manoeuvre is rather unex-
pected, since recursive definitions had usually been considered in relation to the
formal sciences (especially proof theory), and had not been explicitly utilised
previously in the context of linguistic theory. In his discussion of this subject,
though, Bar-Hillel adopts a purely language-based approach by using English
as a metalanguage and by employing it to analyse his chosen object language,
French. At the beginning of his analysis, he introduces the following definition
(see Bar-Hillel 1953a: 163):12

Definition 3.2: Sentence (Recursive in Disguise)

x will be called a sentence (in French) if (and only if) x is a sequence of a
nominal and a (intransitive) verbal, or a sequence of a nominal, a (transitive)
verbal and a nominal or, . . ., or a sequence of a sentence, the word ‘et’, and a
sentence, or, . . .

In this definition, the terms ‘nominal’ and ‘verbal’ can be understood to mean
‘noun phrase’ and ‘verb’ respectively, and it is important to note that this def-
inition seeks to define sentences which can be infinitely long, in terms of sub-
components combined by means of the conjunction ‘et’. Having introduced
66 Mathematical linguistics

definition 3.2, Bar-Hillel proceeds to demonstrate that although it does not

explicitly manifest the two-part structure of a recursive definition, it is neverthe-
less ‘recursive in disguise’, and he accomplishes this by restating the definition
of a sentence as follows (see Bar-Hillel 1953a: 163):

Definition 3.3: Sentence (Recursive)

1. x is a sentence1 (a simple sentence) =df x is a sequence of a nominal and
a (intransitive) verbal or a sequence of a nominal, a (transitive) verbal, and a
nominal, or . . .

2. x is a sentencen+1 (a compound sentence of the n + 1th order) =df x is a

sequence of a sentence p , the word ‘et’, and a sentencem , where either p or m
(or both) are equal to n and none is greater than n, or . . .

Expressed in this fashion, the definition of a sentence now takes the familiar
form of ‘a pair of simultaneous recursive definitions’, and, Bar-Hillel goes
on to claim that it is ‘simple’ to check that a given compound sentence is ‘a
proper French sentence’, since the structure can be iteratively broken down
into smaller units until the basic constituent units (i.e., clauses) are obtained
(Bar-Hillel 1953a: 164). If this can be achieved, then the sentence is ‘proper’,
if not then (presumably) it is improper (i.e., ungrammatical). Once again it is
clear that one of the consequences of a mathematical (in this case, recursive)
approach to syntax is that grammaticality is most naturally expressed in binary
terms: if a sentence can be analysed in accordance with the posited recursive
definition, then it is grammatical, if not, then it is ungrammatical. Bar-Hillel
ends his short paper with a rallying cry which, though expressed in very general
terms, was bound to intrigue any linguists who were even remotely interested
in the relationship between the formal sciences and linguistic theory:

In conclusion, let me say that, in view of the role played by recursive definitions
in concept formation in empirical science, it is the task of the methodologists
to dedicate time and effort to the evaluation of their precise import in different
fields of inquiry and the task of the scientists to become acquainted with the
recent investigations on recursive definitions to a degree, at least, that would
free them from the misconceptions that have so frequently been connected
with their occurrence in disguise. (Bar-Hillel 1953a: 165)

The basic thrust of the above passage is that certain fields of empirical enquiry
were infested with ‘misconceptions’ primarily because they failed to realise
that they were concerned with phenomena that could most easily be defined
recursively. Given the topic of Bar-Hillel’s own discussion – namely, natural
 Logical systems 67

language – it would be difficult to avoid the conclusion that linguistics is one

of the empirical sciences that was bedevilled by such misconceptions. With its
emphasis on greater explicitness and the avoidance of misrepresentation, it is
striking how similar the tone of this passage is to Bloomfield’s 1926 plea for
the avoidance of ‘error’ in linguistic research, and certainly Bar-Hillel’s desire
for precision was swiftly becoming a characteristic feature of the work of those
linguists interested in making their discipline more mathematical. This attitude
has already been identified in the work of Bloomfield, Bloch, and Harwood,
and these are only a few of the researchers who were actively involved in
the process of rigorisation. Others will be mentioned later. Certainly, though,
Bar-Hillel’s use of recursive definitions to analyse the structure of sentences in
natural language can be viewed as one manifestation of this pervasive desire for
the mathematisation of syntactic analysis, which became such a characteristic
feature of certain kinds of linguistic research in the mid-twentieth century.
Significantly, Bar-Hillel’s ideas intrigued Chomsky in the early 1950s, and,
in section 5.6, the influence of recursive function theory upon TGG will be
considered directly.

3.4 Logical systems

As mentioned briefly in section 2.5, after the publication of PM the develop-
ment of symbolic logic became complex and fragmented as new groups of
researchers emerged, often gathered around particular academic institutes, and
often inspired by the desire of improving or extending or replacing the logi-
cal system offered by Whitehead and Russell. While the work of one of these
groups, the Vienna Circle, will be discussed in section 3.5, in this section some
of the research associated with another post-PM group of logicians – namely,
the Lvov-Warsaw school – will be considered.
The Lvov-Warsaw school developed in the early twentieth century under the
leadership of Kazimierz Tawardowski (1866–1938) who moved from Vienna to
Lvov in 1895. Although trained mainly in philosophy and psychology, Tawar-
dowski started lecturing on logic soon after being appointed, and gradually
a whole generation of Polish logicians matured under his guidance. Domi-
nant amongst these were Jan Lukasiewicz (1878–1956), Stanislow Leśniewski
(1886–1939), Kazimierz Ajdukiewicz (1890–1963), and Alfred Tarski (1902–
1983). Although the individual members of the school pursued many different
research interests, much of their work was inspired by perceived limitations
and weaknesses in the logical system presented in PM. For reasons that will
68 Mathematical linguistics

become apparent later, only the work of Leśniewski and Ajdukiewicz will be
considered here.13
Leśniewski studied under Tawardowski, and was appointed to the position of
Chair of Philosophy at Warsaw in 1915. His most significant contribution to the
development of logic was the tripartite logical system he devised, and which
was intended to provide a more secure basis for mathematics than the system
contained in PM. The ambitious nature of this work alone is revealing, since
the very fact that Leśniewski attempted such a project indicates that, during the
1920s, the Logicist movement was still very much alive. His main doubts about
the PM system concerned the imperfections inherent in the kind of formal sym-
bolic language it employed, and he was particularly concerned by the dubious
validity of Whitehead and Russell’s assertion operator. For instance, Leśniewski
juxtaposed the proposition ‘ p’ and the assertion ‘ p’ and asked whether the
two propositions have the same meaning or not. Indeed, he even asked whether
the latter was actually a proposition at all (Leśniewski 1992c[1927]: 181–196).
As for the controversial theory of logical types, Leśniewski argued that it was
unclear whether the theory defined an ontological or a semantic hierarchy, and,
besides, he was convinced that the theory was too intricate to serve as a basic,
intuitive part of a logical system (Leśniewski 1992b[1914]: 115–128). Accord-
ingly, motivated by these concerns, Leśniewski gradually developed his own
alternative to PM during the years 1914–1939, and his system was charac-
terised by a tripartite structure, the main components of which he referred to as
‘protothesis’, ‘ontology’, and ‘mereology’.14 Protothesis and ontology, taken
together, provided the logical foundation of the whole system, the former being
a generalised sentential calculus based upon the notion of equivalence, the latter
being a nominal calculus with the function ‘∈’ (i.e., a stylised form of the 
that appears at the start of the word ‘` σ τ ί’ [Greek: ‘it is’]) as its primitive term.
The logical core provided by these components was equal in power to that con-
tained in PM. The third component of Leśniewski’s system, mereology, was an
extralogical theory based upon the single primitive non-logical constant ‘Part’.
In essence, mereology was a reformulation of Cantorian set theory in which
set membership was transitive rather than intransitive. For instance, in classical
set theory, given the fact that (i) x ∈ A and (ii) A ∈ B, it does not necessarily
follow that x ∈ B, while in mereology, if (i) and (ii) hold, then x ∈ B follows.
One of the reasons for preferring this kind of set theory was that it
enabled Russell’s paradox to be avoided when deriving set theory from log-
ical axioms without having to resort to elaborate and artificial methods. How-
ever, Leśniewski was also motivated to construct his transitive alternative to
classical set theory by his nominalistic beliefs. These beliefs were manifest
 Logical systems 69

in his assertion that only individuals exist and that, therefore, aggregates or
collections of any kind should not be manipulated as independent entities, but
should rather be defined in terms of their individual elements (see Leśniewski
1992c[1927]: 229–230, especially Axiom II). Consequently, classical sets must
be consistently rejected. This extreme stance can be seen as a direct reaction
to the paradoxes of set theory: if the theory as standardly formulated permitted
contradictions, then it must be reformulated, and nominalism, with its focus on
individual entities and its refusal to accept the abstract notion of a set, consti-
tutes an intellectually plausible alternative framework. With specific reference
to formal symbolic languages, Leśniewski’s nominalism caused him to view
them as finite sequences of inscriptions that could be combined in various ways
to produce longer sequences, and these methods largely anticipated the kind of
approach espoused by the architects of constructive nominalism in the 1940s,
as will be shown in sections 3.5 and 3.6.
As indicated above, perceived flaws in PM inspired Leśniewski to devise
a viable alternative, and Whitehead and Russell’s work also provoked
Ajdukiewicz, who was one of Leśniewski’s young contemporaries at War-
saw. Rather than focusing on the Logicist foundational programme, though,
Ajdukiewicz was more concerned with the task of exploring the syntax and
semantics of formal logical languages of the type synthesised by Whitehead
and Russell. In particular, he was interested in the differences that existed
between such languages and natural languages. Therefore, in a series of
papers, starting with ‘On the Meaning of Expressions’ in 1931, he consid-
ered various technical aspects of semantics and syntax. Often in his published
work he chose to focus on well-defined formal languages simply because,
by contrast, natural languages suffered from ‘vagueness and lack of clarity’
(Ajdukiewicz 1978b[1931]: 26). Nevertheless, he was always keen to consider
his conclusions in relation to natural languages. In particular, in much of this
work, Ajdukiewicz repeatedly emphasised the close connection between syn-
tax and semantics, and he sought to specify various sets of ‘meaning-rules’
(Ajdukiewicz 1978a[1934]: 57) for natural language. Since he often advocated
a relativistic view of language, with a pronounced concentration on situational
interpretation, his work in this area can be viewed (to some extent) as a kind of
proto-pragmatics.15
Ajdukiewicz’s main contribution to syntactic theory, though, was ‘Syntactic
Connexion’ (1936), and, at the very start of this paper, he indicated that his inter-
est in the syntax of formal languages was inspired by the paradoxes of set theory
and the Logicist movement that sought to eliminate them: ‘the discovery of the
antinomies, and the method of their resolution, have made problems of linguistic
70 Mathematical linguistics

syntax the most important problems of logic’ (Ajdukiewicz 1978a[1936]: 118).

Consequently, he was motivated by the paradoxes of set theory ultimately to
devise a system of syntactic analysis that would facilitate the analysis of ‘syntac-
tic connexion’. Towards the end of the paper, Ajdukiewicz provides a reasonable
definition of ‘connexion’, but, initially he illustrates the concept by means of
an example: the sequence of words ‘John loves Ann’ possesses the property
of syntactic connection, whereas the sequence ‘Perhaps horse of will however
shine’ does not (Ajdukiewicz 1978a[1936]: 118). As this example indicates,
Ajdukiewicz’s notion of ‘connexion’ is associated with the traditional idea of
grammaticality, and the basic task for the type of analytic syntactic system
he develops in the paper is to demonstrate that the first sentence is connected
(i.e., grammatical), while the second is not. It is important to note, though, that
syntactic connection of this kind does not correspond directly to the Formalist
notion of syntactic well-formedness, since, for Ajdukiewicz, syntactic connec-
tion is developed as a theory of semantic categories, while well-formedness is
determined without reference to meaning. In elaborating his ideas, Ajdukiewicz
states that he was primarily influenced by the work of his compatriot Leśniewski,
but that he was also motivated by Russell and Whitehead’s theory of types. As
will be discussed below, the theory of syntactic connection that Ajdukiewicz
propounded in turn came to inspire the development of the theory of Categorial
Grammar in the 1950s.
The basic methodology that Ajdukiewicz outlined in his main presentations
of his theory involved a hierarchy of semantic categories which were represented
by indices of the form

s s s s n n n n
s, n, , , . . ., , , . . . , , . . ., , , . . . (3.11)
n nn s ss n nn s ss

When this analytic strategy was applied specifically to natural language, each
basic part of speech was associated with one of these indices. For instance, the
definite article ‘the’ was associated with the index ‘n/n’, while a noun such as
‘lilac’ was associated with the index ‘n’. Consequently, in this manner, syntactic
analysis can be accomplished by means of manipulating indices. It is crucial to
note that, although Ajdukiewicz developed his system of syntactic analysis for
formal languages, he clearly indicated that, as far as he was concerned, such
techniques could usefully be applied to natural languages. This is yet another
indication of the fact that, during the 1930s and 1940s, the supposed distance
that was perceived to separate formal and natural languages appeared gradually
to be diminishing.
 Logical systems 71

Although the work of Ajdukiewicz was discussed by logicians, it was ignored

by linguists until, in a 1953 paper entitled ‘A Quasi-Arithmetical Notation for
Syntactic Description’, Bar-Hillel revived and adapted Ajdukiewicz’s basic
scheme. Bar-Hillel’s initial statement of intent is of considerable interest:

The purpose of this paper is to present the outline of a method of syntactic

description that is new insofar as it combines methods developed by the Polish
logician Kasimir [sic] Ajdukiewicz on the one hand and by American structural
linguists on the other. (Bar-Hillel 1953b: 47)

The particular ‘structural linguists’ that Bar-Hillel cites in a footnote are Charles
Fries (1887–1967) and Zellig Harris (1909–1992), both of whom can be broadly
classified as post-Bloomfieldians. It is perhaps unnecessary to stress that Harris’
work in particular was extremely influential during the 1940s and early 1950s.
Indeed, his Methods in Structural Linguistics (hereafter MSL) (1951), with its
emphasis on distributional discovery procedures, had helped to define the basic
methodology of structural linguistics.16 As will be shown in later chapters,
Bar-Hillel’s conviction that techniques derived from the formal sciences could
be combined with methods developed by structural linguists is one that was
voiced frequently, by various researchers, during the 1950s, and some of the
reasons for the perceived association, or compatibility, between mathematical
techniques and the methodology of structural linguists are considered in section
3.7. Suffice it to say here that, for Bar-Hillel, the connection was due to the fact
that certain techniques derived from logic met the structuralist demand for
analytic procedures that functioned in a mechanical fashion. However, in order
fully to appreciate Bar-Hillel’s work, it is crucial also to note that throughout
his paper the emphasis is upon the description of sentential structure and not
primarily upon the type of mechanical discovery procedures sought by Harris
(and others). As Bar-Hillel puts it himself,

We are not interested here in developing a method which a linguist might use to
ARRIVE at an analysis of a linguistic corpus, but only in a new way in which
he could PRESENT the results of his investigations. (Bar-Hillel 1953b: 47)

One possible implication of this statement is that, for Bar-Hillel, Harris-style

discovery procedures are not of primary importance, and such an interpretation
is reasonable since, as will be shown when TGG is considered in section 5.3, this
lack of concern for discovery procedures was starting to become more prevalent
amongst the generation of young linguists reaching intellectual maturity in the
early 1950s. In particular, the rejection of such procedures was perceived to be
one of the most radical aspects of mature TGG.
72 Mathematical linguistics

Whatever his underlying motivations may have been, though, once Bar-
Hillel had indicated how Ajdukiewicz’s approach to syntactic analysis could
be combined with the methodology of structural linguistics, he started to out-
line the basic theory which (in linguistic circles) would later be referred to as
Categorial Grammar. The basic approach involves the definition and manipula-
tion of fundamental syntactic categories. For instance, given the sentence ‘Poor
John sleeps’, following Ajdukiewicz, Bar-Hillel claims that this structure can
be exhaustively analysed using the two categories ‘n’ and ‘s’, which he defines
as
r n: the category of ‘name-like strings’.
r s: the category of sentences.

With these definitions in place, he then introduces the following derived cate-
gories:
r n/[n]: the category of those strings with an n to their right form a
string that belongs to category n.
r s/n: the category of the string that with an n to the left forms a string
belonging to the category s.

Given the above definitions, a categorial analysis of the sentence ‘Poor John
sleeps’ simply requires the sentence to be associated with the index sequence
n s
n (3.12)
[n] n
In which ‘n/[n]’ denotes the adjective ‘poor’, ‘n’ denotes the noun ‘John’, and
‘s/n’ denotes the verb ‘sleeps’. Consequently, the subsequence ‘n/[n]n’ can
be reduced to n, giving the ‘first derivative’ of (3.12)
s
n (3.13)
n
which is turn can be reduced to ‘s’, forming the ‘second and last derivative’ of
the initial index sequence in (3.12). Since the last derivative of the string belongs
to the category s, the sentence can be considered grammatical since it consti-
tutes a legitimate sentence. As Bar-Hillel observes, and as the title of his paper
indicates, the process sketched above is ‘something like ordinary mathematical
multiplication of fractions’ (Bar-Hillel 1953b: 48), and, as suggested previously,
it is the mechanical nature of this process that allows it to be fused with struc-
turalist linguistic methods: using Bar-Hillel’s analytical procedure, sentences
of natural language can be converted into strings of well-defined symbols in
 Constructional system theory 73

a formal language, and then manipulated (without reference to meaning) as if

they were part of an algebraic system.
As with the formal axiomatic approaches to syntactic theory that began to
emerge in the 1950s (discussed in section 3.2), Bar-Hillel’s work implies that
the type of formal languages used in symbolic logic (particularly the techniques
developed by Ajdukiewicz) are closely related to natural languages and that,
therefore, methods developed to analyse the former can be readily adapted in
order to provide analyses of the latter. Although the type of approach to syntactic
analysis proposed by Bar-Hillel was never explicitly incorporated into the TGG
framework, it clearly demonstrated that techniques from logical syntax could
be utilised for the purpose of linguistic analysis, and this is of considerable
interest since, as will be shown in chapter 5, TGG itself drew heavily upon the
methodology of symbolic logic.

3.5 Constructional system theory

In the previous sections of this chapter the influences of three particular aspects
of the formal sciences upon linguistics were considered – namely, the influence
of the axiomatic-deductive method, the influence of recursive function theory,
and the influence of logical syntax. In this section constructional system theory
is described, since it too exerted a lasting influence over syntactic theory in the
1950s. The theory, which was initially proposed by Carnap, grew out of attempts
to use the logical techniques of PM in order to solve traditional philosophical
problems. In particular, the theory was to influence Chomsky directly while he
was developing TGG, consequently it is necessary to consider its genesis in
some detail.
Carnap published his first book, The Logical Construction of the World (from
henceforth LCW), in 1928.17 The intellectual odyssey that had led him towards
the philosophical position adopted in this work had been protracted and tortuous.
As a student at Jena and Buchenbach from 1910 to 1926 he had studied experi-
mental physics, mathematical logic (with Frege), and philosophy. Although, as
a result of these eclectic interests, he initially struggled to devise a research pro-
posal that was acceptable to one single department, he was awarded his Ph.D. in
1921 and immediately began work on the LCW. The bulk of the text was written
during the years 1922–1925 and publication occurred just before he travelled
to Vienna, where he subsequently became one of the most prominent members
of the heterogeneous collection of intellectuals and academics now known as
the Vienna Circle. The Vienna Circle had its origins in the early 1900s when
the Viennese students Hans Hahn (1879–1934), Otto Neurath (1882–1945), and
74 Mathematical linguistics

Phillip Frank (1884–1966) started discussing various ideas concerning the rela-
tionship between logic, mathematics, philosophy, and science. In 1921 Hahn
returned to Vienna as a lecturer and managed to arrange for Moritz Schlick
(1882–1936) to be appointed to the chair of inductive philosophy. Schlick was
a physicist by training, but was preoccupied with various topics in epistemology
and ethics. He began to lead coffee-house discussion groups in the early 1920s,
and the Vienna Circle emerged out of these gatherings. Other key members of
the group were Karl Menger (1902–1985) and Gödel, and (to put it simply)
this disparate group was unified by a common desire to rescue philosophy from
the clutches of the metaphysicians by making it as precise and as exacting as
mathematics or physics. In particular, the members of the Circle sought to pro-
vide a secure foundation for the sciences, to reject metaphysics, and to utilise
logical analysis for the purposes of philosophical enquiry. Consequently, given
Carnap’s interests, it is no surprise that, when he joined the Circle in 1925, he
quickly established himself as one of its defining figures. Indeed, although LCW
was completed before he travelled to Vienna, in later years the book came to be
viewed as one of the seminal texts of the so-called logical positivist movement
that was associated with the Vienna Circle, and, since the text exerted a pro-
found influence over the work of Goodman and Quine (both of whom in turn
influenced Chomsky), it is necessary briefly to summarise the philosophical
project that the book describes.18
Writing in his ‘Intellectual Autobiography’ of 1963, Carnap recalled that
during the 1920s he had made ‘numerous attempts at analysing concepts of
ordinary language relating to things in our environment and their observable
properties and relations, and at constructing definitions of these concepts with
the help of symbolic logic’ (Carnap 1963: 16). It is intriguing that, from the out-
set, Carnap was interested in ‘concepts of ordinary language’, since this research
was the start of his exploration of logical epistemology, which was ultimately
to influence the development of syntactic theory. His general approach was to
employ the logical system developed in PM as a tool for considering ques-
tions of knowledge acquisition. This basic project was certainly influenced by
Russell’s provocative 1914 publication Our Knowledge of the External World,
which had considered the possibility of such an approach to epistemological
questions.19 Specifically, in LCW, Carnap was explicitly concerned with the
task of creating a Konstitutionssystem, or ‘constructional system’, which he
described as follows:

The present investigations aim to establish a ‘constructional system’, that is,

an epistemic-logical system of objects or concepts . . . Unlike other conceptual
 Constructional system theory 75

systems, a constructional system undertakes more than the division of con-

cepts into various kinds and the investigation of the differences and mutual
relations between these kinds. In addition, it attempts a step-by-step derivation
or ‘construction’ of all concepts from certain fundamental concepts, so that a
genealogy of concepts results in which each one has its definite place. It is the
main thesis of construction theory that all concepts can in this way be derived
from a few fundamental concepts, and it is in this respect that it differs from
most other ontologies. (Carnap 1967[1928]: 5)

More precisely, as Carnap goes on to explain, a constructional system is one

which contains a set of basic elements and a set of basic relations which, taken
together, constitute the basis of the whole system, with the consequence that
all other objects (‘Gegenstanden’) in the system are ultimately defined in terms
of this basis. In other words, the task confronting the philosopher working
within the broad framework of logical epistemology involves the generation of
a definitional genealogy for all the objects in the universe of discourse, starting
from a basis of irreducible primitives and the fundamental relations that exist
between them.
Before embarking upon the task of actually creating a epistemic-logical con-
structional system in LCW, Carnap considers at length various formal difficulties
that must be confronted before any such system can be attempted. In particu-
lar, he identifies four main problems which he refers to as the problems of (i)
basis, (ii) ascension form, (iii) object form, and (iv) system form. He succinctly
summarises these problems as follows:

To begin with, a basis must be chosen, a lowest level upon which all others
are founded. Secondly, we must determine the recurrent forms through which
we ascend from one level to the next. Thirdly, we must investigate how the
objects of various types can be constructed through repeated applications
of the ascension forms. The fourth question concerns the over-all form of the
system as it results from the stratified arrangement of the object types. (Carnap
1967[1928]: 47)

Carnap acknowledges that problems (ii), (iii), and (iv) are closely related to
problem (i), since the choice of basis largely determines the form of the objects
created and the manner of their creation, and these two characteristics largely
determine the nature of the system as a whole. Since the technical issues asso-
ciated with this cluster of interconnected problems are of such fundamental
importance to Carnap’s logical-epistemic approach, and since they were also
explored by his most influential successors, it is necessary to discuss some of
them briefly here.
76 Mathematical linguistics

As mentioned above, in LCW, Carnap adopts the logical framework presented

in PM. In other words, his system utilises logical constants (e.g., negation,
disjunction, and so on), variables, logical operators (e.g., the universal and
existential operators), n-place functions, and all the other machinery of symbolic
logic as developed by Whitehead and Russell, and summarised in section 2.5.
Consequently, the problem of the basis can be classified as an extralogical
problem since it concerns the choice of the basic elements and relations that must
be assumed in addition to the logical basis. The basic elements that Carnap uses
in his system are unanalysable elementary experiences (‘Elementarerlebnisse’).
This choice illustrates the particular fusion of phenomenology and symbolic
logic that characterises the project outlined in LCW, and, in his autobiography,
Carnap indicated that it was this phenomenological concern that distinguished
this work from his previous efforts.

A change in approach occurred when I recognised, under the influence of the

Gestalt psychology of Wertheimer and Köhler, that the customary method of
analyzing material things into separate sense-data was inadequate – that an
instantaneous visual field and perhaps even an instantaneous total experience is
given as a unit, while the allegedly simple sense-data are the result of a process
of abstraction. Therefore, I took as elements total instantaneous experience
rather than single sense-data. I developed a method called ‘quasi-analysis’,
which leads, on the basis of the similarity-relation among experiences, to the
logical construction of those entities which are usually conceived as com-
ponents. On the basis of a certain primitive relation among experiences, the
method of quasi-analysis leads step by step to the various sensory domains –
first to the visual domain, then to the positions in the visual field, the colors, and
their similarity system, the temporal order, and the like. (Carnap 1963: 16–17)

The ‘similarity-relation’ referred to above constituted the only extralogical rela-

tion that was incorporated in the basis of the system. This relation, ‘Rs’, is
paraphrased as ‘the recollection of similarity’. In other words, the relational
statement ‘x Rs y’ indicates ‘x and y are elementary experiences which are
recognised as part similar through the comparison of a memory image of x with
y’ (Carnap 1967[1928]: 127). From this single extralogical primitive, Carnap
was able to derive more complex relations such as ‘part-identity’. Although, as
indicated above, these relations are all extralogical constructs, it is clear both
from the notation used and the manner in which they are defined that Carnap
had been profoundly influenced by the theory of relations that Whitehead and
Russell had presented in PM (especially numbers *23–*38, briefly considered
in section 2.5 above).
 Constructional system theory 77

Assuming this minimal basis, then, Carnap proceeds to construct larger

objects by means of definition; that is, a new object is constructed by demon-
strating that statements about the new object can be transformed into statements
either about the basic elements in the system or else into statements about
objects that have already been constructed from the basic elements themselves.
Such definitions are called explicit. If, however, a concept is introduced that
cannot be specified by means of explicit definition, then a process of implicit
definition is required which permits ascension to a new constructional level.20
Consequently, the various constructional levels (which are ultimately deter-
mined by the extralogical basis) largely determine the structure of the system
created. Therefore, as suggested earlier, the problem of system form is directly
related to the problem of the basis, and the precise nature of this relationship
was explored at length by certain of Carnap’s immediate successors, including
Quine.
Quine appears to have encountered Carnap’s LCW for the first time in the
early 1930s when the book was discovered by one of his fellow Harvard graduate
students. Although Quine’s Ph.D. research, which was supervised by Whitehead
himself, had been devoted to simplifying the methodology of PM, his interest
in the work of the Vienna Circle in general, and of Carnap in particular, was
sufficient to prompt him to take the joint advice of two associates and apply
for a scholarship that would enable him to spend some time in Vienna. His
application was successful and he arrived in the city in September 1932. As a
result of this expedition, Quine became one of the conduits through which the
work of the Vienna Circle was conveyed to North America; a development that
was (eventually) to have significant consequences for syntactic theory.
Quine remained in Vienna until May 1933 (his stay punctuated by frequent
trips to other European cities) and during this period he came into close contact
with various members of Schlick’s Circle. With his rapidly improving German,
he attended the group’s weekly meetings and even presented a paper outlining
his own doctoral research. On arriving in the city, he had been disappointed
to find that Carnap had recently left for Prague, but a meeting between the
two men was arranged to coincide with Carnap’s trip to Vienna in 1933. So
began a relationship that Quine was later to describe as a ‘sustained intellectual
engagement’ (Quine 1985: 98). In February of the same year, Quine travelled to
Prague and was entertained by the Carnaps for several weeks, during which time
he attended Carnap’s lectures on his yet-to-be-published The Logical Syntax
of Language (from henceforth LSL). Carnap encouraged Quine to read the
typescript of the book, which was then in the process of preparation, and, as a
result, Quine acquired a detailed knowledge of Carnap’s most recent work.21
78 Mathematical linguistics

Back at Harvard, Quine was keen to communicate some of the ideas he had
encountered in Europe to his colleagues. Consequently, he lectured on Carnap’s
latest work in November 1934, and in 1935 he led a series of ‘informal seminars’
concerning the LSL of which ‘the participants were a professor, [David] Prall,
an instructor, Henry Leonard, and some graduate students, including Nelson
Goodman, Charles Stevenson and John Cooley’ (Quine 1985: 122).22
Carnap himself was able to join the group in December 1935, since he was
in America to spend the winter in Chicago. During his stay at Harvard he lec-
tured on the LSL, with Goodman and Quine (along with Prall and Leonard)
functioning as his intellectual bodyguards: ‘we moved with Carnap as hench-
men through the metaphysicians’ camp’ (Quine 1985: 122). These interactions
appear to have been mutually rewarding, and, to some extent, they determined
the course of the research undertaken by all the main participants for the next
few years. In particular, Quine became increasingly concerned with technical
difficulties associated with the syntax of formal languages. His syntactic inter-
ests are apparent both in his pedagogic publications from this time (especially
Mathematical Logic (Quine 1940), which was discussed in 2.8) and certain
journal papers. For instance, inspired by the work of Tarski and Hans Her-
mes (b. 1912), Quine devoted a whole paper to an exploration of the role of
concatenation in formal language theory, suggesting that the concatenation
relation could provide a viable foundation for the whole of arithmetic (Quine
1946). As shown in section 5.4, some of these ideas would later resurface
in TGG.
It was during these seminars devoted to the LSL, therefore, that Goodman and
Quine first came into regular contact, and, given their respective philosophical
predilections, it is no surprise that they quickly became friends and collab-
orators. Although two years older than Quine, Goodman’s academic career
advanced more slowly, but, by the time the two men met, his intellectual inter-
ests had already begun to coalesce. Indeed, he was already actively working
with Henry Leonard on a research project that explored aspects of Carnapian
logical epistemology and which would ultimately provide material for his 1941
Ph.D. thesis. Appropriately, Goodman and Leonard first confessed their secret
project to Quine while the three men were travelling back from Cambridge
(where they had just left Carnap) to Baltimore, and Quine’s recollection of the
conversation indicates that the context of the research was clear to him even at
the time:

We talked in our hotel room until four in the morning. They were concerned
with constructing a systematic theory of sense qualities, and their effort had
 Constructional system theory 79

much in common with Carnap’s Logische Aufbau der Welt. As an auxiliary

they had developed a logic of the part–whole relation, which I recognised as
Leśniewski’s so-called mereology. They had been meeting fortnightly on the
project, and I happily joined them in subsequent meetings. Leonard was called
away but Goodman and I continued to meet. This project flowered in Nelson’s
dissertation, which he revised and published as The Structure of Appearance.
(Quine 1985: 124)

Leśniewski’s ‘mereology’ was discussed in section 3.4, and the nominalistic

nature of this work was emphasised there. Although Goodman (in conjunction
with Leonard) had developed his notion of the part–whole relation without being
aware of Leśniewski’s research, as will be shown later, there is no doubt that both
approaches share the same kind of preoccupation with nominalism. In general,
though, Quine’s observation that Goodman and Leonard’s project was directly
inspired by LCW was entirely accurate. Indeed, when Goodman’s Ph.D. thesis,
A Study of Qualities, was eventually submitted, it was generally perceived to
constitute an attempt to improve upon the type of constructional system that
Carnap had developed in LCW. In fact, in later years Carnap himself came to
recognise Goodman as one of the first philosophers to have responded directly
to some of the issues he had raised in his early work.23 Further, the extent
of Goodman’s appreciation of Carnap’s LCW reveals itself in the frequent
references to this text in his publications which appeared during the 1940s
and 1950s. For instance, chapter 5 of his book The Structure of Appearance
(from henceforth SA), published in 1951, comprises an extended summary and
critique of LCW, and this fusion of admiration and dissatisfaction is typical
of Goodman’s attitude towards Carnap’s work since, as he claimed in SA, ‘the
purpose of my critical scrutiny is not to disparage [Carnap’s] accomplishments
but to determine just where the remaining problems lie and perhaps pave the
way for their solution’ (Goodman 1951: 114). Twelve years later, writing in his
1963 essay ‘The Significance of Der logische Aufbau der Welt’, Goodman was
still keen to defend the work from critical attacks, even attacks from Carnap
himself, and, accordingly, the paper articulates a robust defence of Carnapian
logical epistemology.
There is no doubt, then, that the research project undertaken by Goodman and
Leonard, and which eventually involved Quine, was directly inspired by Car-
nap’s LCW. Equally, there is no doubt that this project provided the main focus
for Goodman’s academic work during the late 1930s, since it resulted in his first
published paper, ‘The Calculus of Individuals and its Uses’, which appeared
in 1940 and which was co-authored with Leonard. The main purpose of this
paper was to facilitate the use of multigrade relations in constructional systems.
80 Mathematical linguistics

Multigrade relations are predicates that take classes of various magnitudes for
their arguments and which cannot therefore be analysed in terms of, say, dyadic
relations which ultimately define the association between primitive elements in
the system. For instance, as Goodman and Leonard demonstrate, the predicate
‘met with’ is a multigrade relation since any number of people could meet with
any number of other people, and the event of meeting cannot accurately be
analysed as the sum of individual meetings which involve only two people. For
these reasons, and partly motivated by what Goodman and Leonard refer to as
‘considerations of economy’ (Goodman and Leonard 1940: 51), philosophers
working within a broadly logical-epistemic framework had been forced to reject
multigrade relations as primitives in their constructional systems. Accordingly,
Goodman and Leonard were keen to encourage the use of such relations by
introducing the ‘part–whole’ relation, ‘x | y’, indicating ‘x and y have no part
in common’, and by demonstrating that this relation can be used to analyse these
relations more intuitively. As indicated previously, this is the aspect of Goodman
and Leonard’s work that Quine recognised as an independent reformulation of
Leśniewski’s mereology. It is of particular interest that Goodman and Leonard
cite ‘consideration of economy’ as one of the reasons why multigrade rela-
tions had been rejected as appropriate predicates for extralogical bases of con-
structional systems, since such considerations of economy were increasingly to
dominate Goodman’s academic research for the next two decades. More specif-
ically, partly motivated by his exploration of the part–whole relation defined in
his first paper, Goodman came to believe that ‘simplicity’ and ‘economy’ were
crucial aspects of system construction. However, he also recognised that these
related notions were currently being used inconsistently, and that their precise
meaning needed to be clarified. Consequently, during the 1940s and 1950s he
endeavoured to provide this clarification.
Although the term ‘simplicity’ has intuitive appeal, Goodman realised that a
formal definition was required if the centrality of basal simplicity in construc-
tional system theory was to be fully acknowledged. He also perceived that the
existing definitions were conspicuously inadequate. The intuitive assumption
underlying his own approach was that a constructional system possessing a sim-
pler, more economical, basis was preferable to an alternative system that utilised
a more complex basis. Initially, the easiest way of determining the relative sim-
plicity of two given bases appeared to involve merely counting the number
of extralogical primitives, and, certainly, with this end in view, Goodman was
keen to develop ways of reducing the number of such primitives. For instance,
in his 1940 paper ‘Elimination of Extra-Logical Postulates’, which was writ-
ten with Quine, Goodman devised a definitional methodology that effected the
 Constructional system theory 81

elimination of basal relations. For example, if a given basis contains the primi-
tive transitive relation ‘Pt’, which indicates ‘is a spatial part of’, then ‘x Pt y’
states that ‘x is a spatial part of y’, and this relation can be determined by the
extralogical postulate

∀x∀y∀z[x Pt y ∧ y Pt z → x Pt z] (3.14)

In this case, rather than taking ‘Pt’ as a primitive element in the basis, the more
general relation ‘O’, which indicates spatial overlapping, could be introduced
instead with the result that the transitivity of the relation ‘Pt’ becomes a theorem
rather than a primitive notion:

x Pt y =df ∀z[x O z → y O z] (3.15)

In other words, (3.14) has become unnecessary since it is merely an abbrevi-

ation, according to (3.15), which can be derived using the logical component
of the system. In this manner various extralogical primitives in the basis can
be replaced by ‘mere definition’ (Goodman and Quine 1940: 104), and, con-
sequently, their number can be reduced. Indeed, Goodman and Quine go on
to demonstrate that such primitives are ‘ordinarily eliminable’ (Goodman and
Quine 1940: 109), and this prompts them to suggest that existing ideas concern-
ing the relative economy of constructional system bases should be reassessed.
In particular, they argue that a distinction must be made between ‘real and
apparent economy’ (Goodman and Quine 1940: 109), and in subsequent work
they both went on to explore this distinction further. Quine returns to the notion
of theory-internal simplicity, for instance, throughout his 1953 book From a
Logical Point of View (from henceforth LPV), and distinguishes consistently
between the ‘economy of practical expressions’, which is merely a notational
convenience, and ‘economy in grammar and vocabulary’, which actively ‘sim-
plifies theoretical discourse’ (Quine 1953: 26). Using the terminology intro-
duced above, the former concern here is ‘apparent’, while the latter is ‘real’.
In a similar fashion, Goodman argued that only apparent economy is achieved
if, for example, one simply concatenates all extralogical primitives in a given
basis, for the basis of any system could easily be reduced to one in this manner,
thus invalidating this method of counting the number of extralogical primitives
as a means of determining the relative simplicity of various systems. Clearly,
a more subtle approach to the problem of basal economy was required. Good-
man’s musings concerning this topic are revealing and so are worth quoting in
full.
82 Mathematical linguistics

What we apparently need is a way of gauging the relative simplicity and

complexity of ideas . . . We do not need to be able to determine whether a given
idea is simpler than another in every way, but only whether it is simpler in those
respects which are relevant to economy. Naturally, any criterion of simplicity
must, within announced limitations, conform to our stronger and justifiable
intuitions concerning simplicity in general. But beyond this, the crucial test for
our present purposes will be whether the replacement of a given idea by one
that is simpler does generally require – and therefore signify – the application
of some special knowledge concerning the ideas involved. It is this that makes
an economy significant or genuine. If, through the application of some purely
automatic procedure like that by which the number of extralogical primitives
in a basis can always be reduced to one, an idea that is less simple by the
criterion in question can always or even usually be replaced by another that is
more simple, then that criterion in unsatisfactory for our purposes. (Goodman
1943: 108)

A number of the issues raised here are worth highlighting. For instance, it is
important to recognise that, for Goodman, simplicity considerations are only
valid if they involve the replacement of one idea by another, and if the replace-
ment is motivated by ‘some special knowledge concerning the ideas involved’.
Consequently, some kind of simplicity criterion is required that would be able
to quantify the degrees of genuine economy associated with the extralogical
primitives in the respective bases of two given constructional systems and,
in his 1949 paper ‘The Logical Simplicity of Predicates’ Goodman proposed
one possible criterion for nominalistic constructional systems. He begins by
introducing a complexity measure that assigns a value to the predicates in the
basis of a given system. Initially, the measure merely computes the complexity
value for an n-place predicate using the formula 2n − 1. For instance, a 2-place
predicate, a 3-place predicate, and a 5-place predicate are associated with the
values 3, 5, and 9 respectively. This complexity measure reveals that a basis
containing a single 5-place predicate is more complex than a basis that contains
a 2-place predicate and a 3-place predicate, since, in this case, the bases have the
complexity values 8 and 9 respectively. Goodman proceeds to consider other
aspects of predicate structure, including the number of joints and segments,
and, ultimately, the complexity value for a given predicate is determined by
computing the value of the initial complexity measure and adding the number
of joints and segments.24 He also emphasises that these numbers are dependent
upon the available information: if it is not known how many joints and segments
are associated with a given predicate, then (obviously) these values cannot be
incorporated into a measure of complexity.
 Constructional system theory 83

Goodman elaborated his approach to complexity measurement in several

later papers as well as in SA. For instance, in 1950 he proposed a superior way
of counting the joints associated with a predicate (Goodman 1950), and in 1955
modified his whole approach to the problem by attempting to place the task
of complexity measurement on an axiomatic foundation (Goodman 1955). In
addition, in his 1952 paper ‘New Notes on Simplicity’ he considered the issue
of choosing between two extralogical bases that have the same complexity
value, and his heuristic decision criteria are specified as follows (see Goodman
1952: 190):

If A and B are alternate bases of equal computed complexity, we choose A

rather than B if:
(i) each consists of one predicate, and A has fewer places; or
(ii) A, in number of predicates and of places in each, is exactly like some
basis that consists of some but not all the predicates of B; or
(iii) after the above rules have been applied, A and B have the same total
number of places but A has the more predicates and thus the lower ratio of
places to predicates.
These criteria enable one basis to be preferred in cases of equal complexity
and they contribute to the task of reducing basis analysis and selection to an
essentially mechanical procedure.
Despite his own conviction that basal economy and complexity measures
were crucial to the task of constructional system creation, Goodman recog-
nised that the importance of this aspect of system theory was not necessarily
immediately apparent. One obvious counter-argument was that the inherent
simplicity of the basis of a constructional system was merely an aesthetic con-
sideration that had no significant consequences for the resultant system itself.
In his 1943 paper ‘On the Simplicity of Ideas’, Goodman explicitly refuted this
argument.

The motives for seeking economy in the basis of a system are much the same
as the motives for constructing the system itself. A given idea A need be left
as primitive in a system only so long as we have discovered between A and the
other primitives no relationship intimate enough to permit defining A in terms
of them; hence the more the set of primitives can be reduced without becoming
inadequate, the more comprehensively will the system exhibit the network of
interrelationships that comprise its subject-matter. Of course we are often
concerned less with an explicit effort to reduce our basis than with particular
problems as to how to define certain ideas from others. But such special
problems of derivation, such problems of rendering certain ideas eliminable
84 Mathematical linguistics

in favor of others, are merely instances of the general problem of economy.

Thus it is quite wrong to think of the search for economy as a sort of game,
inspired by an abnormal love of superficial neatness. Some economies may
be relatively unimportant, but the inevitable result of regarding all economy
as trivial would be a willingness to accept all ideas as primitive at the outset,
making a system both unnecessary and impossible. (Goodman 1943: 107)

As it is presented here, far from being a manifestation of ‘an abnormal love of

superficial neatness’, the general theory of simplicity is a metalogical concern
which impinges upon the validity of a given constructional system. In essence, a
‘simpler’ system is a better system, so long as it does not become ‘inadequate’;
and ‘better’ in this context means a more economical system, since such systems
are understood to provide more profound insights into the phenomena analysed.
By 1955 Goodman apparently sensed that he had managed to win this particular
battle, since he felt able to declare:

No longer do we need to take seriously the idea that simplicity is something

to be sought only if there is time after truth has been attained. To seek truth is
to seek a true system, and to seek such system [sic] at all is to seek simplicity.
(Goodman 1955: 709)

Theory-internal simplicity is here unequivocally associated with some kind of

philosophical truth, and as will be shown later, it is significant that Goodman
was exploring these ideas and engaging in this kind of inspiring polemic in the
late 1940s, around the time he began to teach philosophy at the University of
Pennsylvania to a promising young undergraduate called Noam Chomsky. The
precise nature of the influence of simplicity considerations and other aspects of
constructional system theory upon TGG is explored at length in sections 4.3,
4.4, and 5.4.

3.6 Constructive nominalism

As indicated in the preceding section, Goodman began his philosophical career
as a logical empiricist in the LCW tradition, and, during the 1940s, his empirical
stance became even stricter. Once again Carnap’s influence was significant, for
he returned to Harvard during the academic year 1940/1941 and was involved
in discussions with both Goodman and Quine concerning various aspects of
logic. Ever since completing the LSL, Carnap’s academic work had focused on
problems associated with formal language theory and, as he later recalled, the
discussions at Harvard enabled him to refine some of his ideas.
 Constructive nominalism 85

My thinking on these problems received fruitful stimulation from a series of

conversations which I had with Tarski and Quine . . . when I was at Harvard;
later Nelson Goodman participated in these talks. We considered especially
the question of which form the basic language, i.e., the observation language,
must have in order to fulfil the requirement of complete understandability. We
agreed that the language must be nominalistic, i.e., its terms must not refer to
abstract entities but only to observable objects or events. (Carnap 1963: 79)

The ideas promulgated and debated in these meetings clearly influenced the
thinking of all the participants, and Goodman and Quine eventually articulated
their response in their 1947 paper ‘Steps Towards a Constructive Nominalism’
(from henceforth ‘SCN’). The basic purpose of this work was to implement pre-
cisely the type of nominalistic formal language that had been proposed during
the Harvard discussions, and this project was pursued within the broad frame-
work of constructional system theory. The polemically nominalistic stance of
the paper is declared in its notorious opening sentence (‘We do not believe in
abstract entities’ (Goodman and Quine 1947: 105)), and this declaration pro-
vided the philosophical foundation for the entire scheme. In order to motivate
their extreme stance, Goodman and Quine explain their refusal to tolerate such
notions as follows:

Fundamentally this refusal is based on a philosophical intuition that cannot

be justified by appeal to anything more ultimate. It is fortified, however, by
certain a posteriori considerations. What seems to be the most natural principle
for abstracting classes or properties leads to paradoxes. Escape from these
paradoxes can apparently be effected only by recourse to alternative rules
whose artificiality and arbitrariness arouse suspicion that we are lost in a
world of make-believe. (Goodman and Quine 1947: 105)

This statement indicates that, for Goodman and Quine (as for Leśniewski before
them) the main motivation for a nominalistic approach to constructional system
theory was the belief that, by rejecting abstract entities, the infamous paradoxes
that had disrupted all attempts to place mathematics on a secure (set-theoretical)
foundation might be avoided. Consequently, in this paper, Goodman and Quine
make a clear distinction between ‘platonistic logic’ (i.e., logic that admits
abstract entities such as sets) and ‘nominalistic logic’ (i.e., logic that does
not admit abstract entities such as sets), and the first sections of the paper are
devoted to outlining some of the strategies that must be employed in order to
convert platonistic statements into nominalistic statements. For instance, the
platonistic statement ‘Class A has three members’ assumes the existence of a
finite class (i.e., set), which is an abstract entity. However, if there are three
distinct objects x, y, and z, such that an object is in A if and only if it is x
86 Mathematical linguistics

or y or z, then this abstract set-theoretical statement can be converted into the

nominalistic statement

∃x∃y∃z[x = y ∧ y = z ∧ x = z ∧ ∀w[w ∈ A
≡ w = x ∨ w = y ∨ w = z]] (3.16)

Clearly, (3.16) assumes the existence of nothing other than concrete individuals
in addition to the basic logical operators.
Having outlined their basic approach in this fashion, Goodman and Quine pro-
ceed to develop in detail the nominalistic syntax they require in order to remove
all ‘taint’ of platonism from their system (Goodman and Quine 1947: 107). The
syntax they create is inscriptional by design since it is ultimately concerned with
strings of physical marks rather than with abstract entities of any kind. Conse-
quently, as the foundation of the extralogical basis of their system, Goodman
and Quine define six so-called ‘shape-predicates’. These are predicates that
take concrete individuals as arguments and which specify the shape of a par-
ticular character. For instance, the predicate ‘Vee x’ means ‘the object x is a
vee (i.e., a v-shaped inscription)’ (Goodman and Quine 1947: 112). The other
characters defined in this way are: ’, (, ), |, and . In addition, a concatenation
relation, C, is assumed where ‘x C yz’ is understood to mean that ‘x and y
are composed of various characters of the language . . . and that the inscription
x consists of y followed by z’ (Goodman and Quine 1947: 112). The role of
concatenation is crucial in this system. As mentioned in section 3.5, Quine had
already explored the idea that the whole of arithmetic could be founded upon
the basis of concatenation alone, and, the inclusion of the C predicate in ‘SCN’
suggests that the notion of concatenation was considered to be fundamental also
to the formal syntax of a nominalistic object language. The final two predicates
used are introduced as ‘Part’ and ‘Bgr’, and are taken to mean ‘is part of’ and
‘is bigger than’ respectively. More precisely, ‘x Part y’ means ‘x, whether or
not it is identical with y, is contained entirely within y’, while ‘x Bgr y’ means
‘x is spatially bigger than y’ (Goodman and Quine 1947: 112–113).
Having presented the basic syntax of their object language, Goodman and
Quine then provide definitions of auxiliary predicates that are constructed in a
purely nominalistic manner from their initial predicate set. For instance, a four-
part predicate can be defined using the existential operator, logical conjunction,
and the concatenation operator:

x y C zw = ∃t[x C yt ∧ t C zw] (3.17)

 Constructive nominalism 87

Ultimately, it is demonstrated that variables, quantification, formulae, and (ulti-

mately) proofs can all be obtained within the unremittingly nominalistic frame-
work outlined in the paper.
Interestingly, their joint work on constructive nominalism eventually led
Goodman and Quine in very different directions. Goodman continued to advo-
cate the strict, individual-based type of logical epistemology associated with
nominalistic constructional system theory, and, as was mentioned earlier, he
revised his ideas concerning simplicity measures for extralogical bases so that
the methods he devised could be used in the more extreme nominalist frame-
work presented in ‘SCN’. Further, in his 1954 book Fact, Fiction and Forecast,
Goodman highlighted some of the problems that had bedevilled traditional jus-
tifications of the inductive processes required by logical epistemology of the
LCW variety, before attempting to resolve some of these problems by develop-
ing a ‘theory of projection’ that was designed to distinguish between valid and
invalid inductive procedures. By contrast, during the late 1940s Quine began
increasingly to reject the type of logical epistemology that had so entranced
him previously. His most perceptive rebuttal of this whole philosophical move-
ment was succinctly articulated in ‘Two Dogmas of Empiricism’ (1953), which
remains a seminal text of North American analytic philosophy. In this artful
paper, Quine endeavoured to demonstrate that two basic assumptions that have
often characterised ‘modern empiricism’ (Quine 1953: 20), as manifest in Car-
napian logical epistemology, were essentially invalid. These two assumptions
can be summarised as (i) the Kantian belief that there is a qualitative differ-
ence between analytic and synthetic truths,25 and (ii) the reductionist belief that
each meaningful statement can be exhaustively analysed as a logical construct
ultimately derived from sensory experience. It is essential to note here that con-
sideration (i) was a topic of much discussion in the 1950s. For instance, this
particular issue was considered at length by Morton White (b. 1917) in his paper
‘The Analytic and the Synthetic: an Untenable Dualism’ (1952). In this article,
White referred to ‘a revolt’ against the general acceptance of the conventional
Kantian distinction, and he indicated that this revolt was led by himself and cer-
tain ‘fellow revolutionaries’ (White 1952: 272) – namely, Goodman and Quine.
Presumably in order to further the cause of the revolution, White indicated that
his purpose in the paper was to ‘present some of the reasons for this decline
of faith’ (White 1952: 272). While the specific details of White’s argument are
not necessary here, it is crucial to recognise that his attempt to destabilise the
analytic-synthetic distinction depended upon problems inherent in the notion of
meaning in natural language. For instance, (following Quine) White considered
the sentence
88 Mathematical linguistics

Every man is a man (3.18)

and argued (with some justification) that (3.18) embodies a tautology and there-
fore can be classified as an analytic truth. He then presents the sentence
Every man is a rational animal (3.19)
and suggests that sentence (3.19) can only be classified as analytical if the
synonymy of ‘man’ and ‘rational animal’ can be irrefutably established. In the
ensuing assessment of synonymy, White acknowledges that artificial languages
(e.g., L1 ) could be created that would resolve all problems of synonymy, but he
goes on to declare that
these languages are the creatures of formal fancy; they are dreamed up by a
logician. If I ask: ‘Is “All men are rational animals” analytic in L1 ?’, I am
rightly told to look up the rule-book of language L1 . But natural languages
have no rule-books and the question of whether a given statement is analytic
in them is much more difficult. (White 1952: 277)

Clearly the implication here is that natural languages and logical languages
differ considerably, at least as far as considerations of meaning are concerned,
and this view should be situated in the context of the general move towards
the further integration of natural and logical languages that has already been
discussed with reference to the work of Bloomfield, Ajdukiewicz, Harwood,
and Bar-Hillel. More specifically, White argues that issues of meaning are
fundamentally problematic in natural languages, though they can (usually)
be resolved in logical languages. As will be shown in sections 4.4 and 4.5,
Quine’s assault upon, and Goodman’s defence of, logical empiricism, along
with Quine and White’s questioning of the analytic–synthetic dualism, all stim-
ulated Chomsky while he was in the process of developing TGG, and it is crucial
to recognise that these developments were largely provoked by the constructive
nominalist research programme that Goodman and Quine had propounded in
the 1940s.

3.7 Formal linguistic theory

In the previous section the influence of Carnap’s LCW upon North American
philosophy was discussed and in chapters 4 and 5 it will be shown that the work
of Goodman and Quine directly inspired various aspects of TGG. However,
Carnap’s later work concerning logical syntax (particularly LSL) also fascinated
linguists during the first half of the twentieth century. Consequently, in this
 Formal linguistic theory 89

section, some distinctive features of formal syntactic theory are considered,

and their influence upon linguistics is assessed.
As indicated in section 3.4, the fact that a formal syntactic theory devised by
a logician (i.e., Ajdukiewicz) was later extended by a linguist (i.e., Bar-Hillel)
in the 1950s indicates that, by the middle of the twentieth century, symbolic
logic had begun to influence certain areas of linguistics directly. However, dur-
ing the 1930s and 1940s other developments in the theory of logical syntax
occurred, which were ultimately to have profound implications for linguis-
tic research, and the starting point was usually Hilbert’s proof theory, which
seemed to imply that meaning-less syntactic manipulations could suffice to
resolve a whole range of epistemological problems. As mentioned in section
2.6, Hilbert’s work had influenced the Vienna Circle directly, and one of the
most distinctive developments associated with Vienna-based research into the
nature of logical systems in the 1930s was the shift of focus away from the
structure of logic itself towards a concentration upon the structure of the lan-
guage in which logical arguments are expressed. As mentioned previously, the
most influential work of this type was due to Carnap, whose LSL appeared in
1934, with an English translation of the German original following in 1937, and
it should be recalled that this is the text which had fascinated both Goodman
and Quine, and which had ultimately pushed them towards constructive nomi-
nalism. However, Carnap’s own agenda, as expressed in LSL, was not explicitly
nominalistic, and his main intention was simply to explore the nature of logi-
cal syntax with greater precision. In particular, he emphasised the fact that all
logical arguments, as expressed in conventional publications, usually consist of
various formal deductions that are connected by sentences written in a natural
language. Since these natural languages are external to the formal logical system
being used, sentences written in them are not precisely defined, and therefore
constitute weak links in any logical argument. Consequently, Carnap’s intention
was to provide a coherent logical system that could be used to analyse these
sentences about sentences. In other words, just as Hilbert had created metamath-
ematics (mathematics about mathematics), so Carnap was keen to construct a
more general metalanguage that could be used to define and describe any given
language.
Given the importance of Carnap’s undertaking, it is essential to determine
precisely what he meant by the term ‘language’. In parts 1, 2, and 3 of LSL
Carnap develops two artificial formal languages in an attempt to clarify the
relationship between logic and the language of science, and he states his central
aim as follows:
90 Mathematical linguistics

The aim of the logical syntax is to provide a system of concepts, a language,

by the help of which the results of logical analysis will be exactly formulable.
Philosophy is to be replaced by the logic of science – that is to say, by the
logical analysis of the concepts and sentences of the science, for the logic of
science is nothing other than the logical syntax of the language of science.
(Carnap 1937[1934]: xiii)

This passage suggests that, at this stage in his career, Carnap was convinced
that philosophy could be replaced by the systematic study of logical syntax.
However, in part 4 he broadens his basic approach and attempts to outline
the formal syntactic structure of language in general. It is crucial to note that
Carnap consistently views artificial languages as forming a well-defined subset
of natural languages, though he makes it clear that his intention is not to describe
the syntax of natural language.

In consequence of the idiosyncrasies and logically imperfect structure of the

natural world-languages (such as German or Latin), the statement of their
formal rules of formation and transformation would be so complicated that
it would be hardly feasible in practice . . . Owing to the deficiencies of the
world-languages, the logical structure of a language of this kind will not be
developed. (Carnap 1937[1934]: 2)

The types of rules that Carnap refers to as ‘formation’ and ‘transformation’ rules
will be discussed at some length below, but it is necessary to stress that the above
passage reveals the domain of Carnap’s enquiry: instead of focusing on natural
languages, he concentrates on the task of defining and constructing formal
artificial languages, and, in order to accomplish this, he utilises assumptions
and techniques derived from (Hilbertian) Formalism. Despite this emphasis,
Carnap does not entirely preclude the possibility that the techniques he develops
in LSL may be of relevance for theories of natural language. Indeed, he states
explicitly that

The method of syntax which will be developed in the following pages will not
only prove useful in the logical analysis of scientific theories – it will also help
in the logical analysis of the world-languages. Although here, for the reasons
indicated above, we shall be dealing with symbolic languages, the syntactical
concepts and rules – not in detail but in their general character - may also be
applied to the analysis of the incredibly complicated world-languages. (Carnap
1937[1934]: 8)

As will be shown in chapters 4 and 5, this basic belief that ‘symbolic’ and
‘world’ languages could (in principle) be analysed using the same fundamental
 Formal linguistic theory 91

techniques exerted a potent influence over certain linguists during the following
years.
It was suggested above that LSL reveals the influence of (Hilbertian) Formal-
ism, and Carnap’s indebtedness to Formalism is explicitly acknowledged several
times; indeed, the influence of Hilbert pervades the entire book. For instance,
Carnap specifically credits Hilbert with developing the theory of formal lan-
guages, observing that ‘[t]he point of view of the formal theory of languages
(known as ‘syntax’ in our terminology) was first developed for mathematics by
Hilbert’ (Carnap 1937[1934]: 1), and he utilises Hilbert’s notion of a formal
system extensively. For instance, at the very beginning of part 1 Carnap offers
the following definition of the term ‘formal’:

A theory, a rule, a definition, or the like is to be called formal systems when

no reference is made in it either to the meaning of the symbols (for example,
the words) or to the sense of the expressions (e.g., the sentences), but simply
and solely to the kinds and order of the symbols from which the expressions
are constructed. (Carnap 1937[1934]: 1)

and, obviously, this quotation evinces the same concern for meaning-less sym-
bol manipulation as the various pedagogic texts considered in 2.8, with the
result that Carnap’s formal languages consist of meaning-less formulae derived
ultimately from primitive symbols by means of rules in a characteristically
Formalist manner; that is, the formal languages are defined solely in terms
of the syntactic structure of the sentences they produce, and the meanings of
the resulting formulae and primitive symbols are not considered. In order to
emphasise this point, Carnap considers the sentence ‘Pirots karulize elatically’
and states that this sentence can be parsed accurately as a Noun+Verb+Adverb
sequence even though the words are all unfamiliar (Carnap 1937[1934]: 2),
thus demonstrating that sentences can be exhaustively analysed solely in terms
of their formal syntactic structure even if the meaning of the individual words
is not known. This type of argument, which (according to Carnap) affirms the
separation of meaning and syntax, proved to be influential, and it will reappear
when TGG is discussed in section 5.7.
There is another aspect of the theory Carnap develops in LSL that is relevant to
the development of TGG. As mentioned above, Carnap distinguishes between
formation rules and transformation rules, and he introduces these terms in a
passage that considers the possibility of reducing logic to ‘syntax’:

Logic will become syntax, provided that the latter is conceived in a sufficiently
wide sense and formulated with exactitude. The difference between syntacti-
cal rules in the narrower sense and the logical rules of deduction is only the
92 Mathematical linguistics

difference between formation rules and transformation rules, both of which

are completely formulable in syntactical terms. Thus we are justified in desig-
nating as ‘logical syntax’ the system which comprises the rules of formation
and transformation. (Carnap 1937[1934]: 2)

As this passage indicates, in the framework Carnap outlines in LSL, a distinc-

tion can be made between rules that are used to construct strings of symbols
from (ultimately) primitive elements, and the rules of inference that enable one
string of symbols to be inferred from another string of symbols. Specifically,
the former are classified as formation rules, while the latter are referred to as
transformation rules. For example, in part 1 of LSL, logical conjunction is cat-
egorised as a formation rule since it enables a sentence to be constructed from
smaller units. In this case, the sentences G1 and G2 can be conjoined to create
the sequence

G1 ∨ G2 (3.20)

where ‘∨’ is a two-termed junction symbol (Carnap 1937[1934]: 19), and

this type of conjoining is classified as a formation process. By contrast, the
rule of implication is categorised as a transformation rule in Carnap’s system,
since it enables sentences to be inferred from other sentences. Therefore, given
the premises G1 and G2 , the rule of implication is stated as follows (Carnap
1937[1934]: 32)

G 2 , G 1 ⊃ G3 (3.21)

where ‘⊃’ is the symbol Carnap uses to indicate implication. This rule states
that sentence G3 can be obtained from the sentences G1 and G2 by means
of implication, although G3 does not constitute a simple combination of G1
and G2 . Consequently, (3.21) constitutes a transformation rule, and it should
be obvious that this rule is simply a restatement of Hilbert’s proof-theoretical
schema, given as (2.10) above.
Transformation rules are used extensively in LSL, and Carnap repeatedly
emphasises their centrality. The following passage is typical:

For the construction of a calculus the statement of the transformation rules,

as well as of the formation rules, as given for language I, is essential. By
means of the former we determine under what conditions a sentence is a
consequence of another sentence or sentences (the premises). But the fact that
G2 is a consequence of G1 does not mean that the thought of G1 would be
accompanied by the thought of G2 . It is not a question of psychological but
of a logical relation between sentences. (Carnap 1937[1934]: 27)
 Formal linguistic theory 93

Once again, this passages stresses the fact that in Carnap’s framework trans-
formation rules are purely rules of logical inference that indicate a ‘logical
relation’ between particular sentences, with one sentence following as a con-
sequence of another sentence (or group of sentences). In case this summary
gives the impression that Carnap was devoid of any qualms concerning the
nature of logical inference, it is necessary to indicate that he was fully cog-
nizant of the many difficulties that beset his methodology. Indeed, Carnap was
happy to acknowledge that the notion of ‘consequence’ had to remain unde-
fined in his system, though the more restrictive notion of derivability could be
established.

It is impossible by the aid of simple methods to frame a definition for the

term ‘consequence’ in its full comprehension. Such a definition has never yet
been achieved in modern logic (or, of course, in the older logic) . . . At present,
we shall determine for Language I, instead of the term ‘consequence’, the
somewhat narrower term ‘derivable’ . . . For this purpose, the term ‘directly
derivable’ will be defined, or – as it is more commonly expressed – rules of
inference will be laid down. (Carnap 1937[1934]: 27)

As will be demonstrated in section 5.5, Carnap’s presentation of his formation

and transformation rules, with their associated terminology and philosophical
preoccupations, were subsequently adapted in different ways by several lin-
guists during the years following the publication of LSL, and, as will be shown,
the nature of these rules was debated in various discussions and disputes that
were sustained while TGG was being developed.
There is no doubt that formal approaches to syntax, such as that outlined
by Carnap in LSL, captured the imagination of many linguists during the first
half of the twentieth century, and once again Bloomfield emerges (perhaps
unexpectedly) as a prophetic figure. Bloomfield’s interest in the axiomatic-
deductive method was discussed in section 3.2, and it was shown there that he
was familiar with recent developments in modern mathematics. Indeed, during
the 1930s Bloomfield seems to have become preoccupied with the nature of
the relationship between mathematics and linguistics, and his understanding of
this topic was directly influenced by the work of the Vienna Circle. The full
extent of Bloomfield’s familiarity with the main publications associated with
the leading member of Schlick’s Circle can be gauged from his 1936 paper
‘Language or Ideas?’ In this paper, Bloomfield cites five works written by Neu-
rath and seven by Carnap, including LSL. This suggests that Bloomfield was
entirely familiar with Carnap’s provocative ideas concerning logical syntax,
and, as will become apparent, this fact is of some significance. In this context,
94 Mathematical linguistics

then, Bloomfield pursued his interest in the relationship between mathematics

and linguistics, and his most extended consideration of this subject was con-
tained in a long essay which he contributed to the International Encyclopedia
of Unified Science in 1939 (five years after LSL appeared). The Encyclopedia
was a forum for assessing the methodology of scientific research, and many of
the contributors were associated either directly or indirectly with the Vienna
Circle. In particular, Carnap was on the board of editors that read and assessed
the contributions, including Bloomfield’s paper. This paper, ‘Linguistic Aspects
of Science’, was based on a 1935 article that he had published, and it served
several purposes. For instance, it summarised various ideas and techniques
employed in linguistic research in the early decades of the twentieth century,
and, in this respect, it can be viewed in part as a brief informal summary of
Bloomfield’s 1933 book Language. However, in addition, the essay considers
the relationship between linguistics and mathematics, or, as Bloomfield puts
it himself, ‘the relation of linguistics to logic and mathematics’ (Bloomfield
1955[1939]: 273). In the light of the foundations debates (and given Bloom-
field’s knowledge of the same), this statement should be read very carefully,
since it appears to imply that, for Bloomfield, mathematics and logic were
separate fields of research. In other words, by stating his interest in this way,
Bloomfield may be consciously avoiding the extreme Logicist viewpoint that
would consider mathematics to be entirely derivable from logic. Whether this
is so or not, having outlined his basic intention, Bloomfield then considers var-
ious aspects of the broad topic he has broached. For example, he declares that
‘logic is a branch of science closely related to linguistics, since it observes how
people conduct a certain type of discourse’ (Bloomfield 1955[1939]: 273–274),
and this leads him to suggest in turn that logical arguments can be analysed
specifically as linguistic discourses. Such statements certainly imply a close
correspondence between linguistics and logic, and they reinforce the sugges-
tion (made in section 3.2) that during the 1930s Bloomfield had started to think
of mathematics as a form of language that would be amenable to linguistic
analysis.
Clearly, then, Bloomfield was fascinated by the relationship between logic
and natural language, and one of the most revealing aspects of the discussion
he offered in his 1939 paper was his enthusiastic acceptance of Hilbert’s basic
Formalist creed; the very creed that had inspired Carnap’s LSL. For example,
Bloomfield makes a clear distinction between formal and informal scientific
discourse, describing the former as a manner of communication that ‘uses a
rigidly limited vocabulary and syntax and moves from sentence to sentence only
within the range of conventional rules’ (Bloomfield 1955[1939]: 261), and he
 Formal linguistic theory 95

later argues that, in considering the ‘characters’ (i.e., symbols) used in logical
and mathematical discourse, he has not ‘left the domain of language’ since

In general, to be sure, the separate characters have been agreed upon as sub-
stitutes for specific words or phrases. In many cases, however, we manage
best by ignoring the values and confining ourselves to the manipulation of
the written symbols; systems of symbolic logic, especially, may be viewed,
in a formal way, as systems of marks and conventions for the arrangement
of these marks . . . our formal systems serve merely as written or mechanical
mediations between utterances of language. (Bloomfield 1955[1939]: 262)

This passage, which could easily have come from one of the textbooks discussed
in 2.8, suggests that Bloomfield was essentially convinced of the validity of the
Formalist approach to mathematics. At the very least, it implies that Bloomfield
accepted the Formalist dictum that ‘we manage best’ (to use his own words) if
we focus on syntactic manipulation and ignore considerations of meaning. The
implications of this aspect of Bloomfield’s work are considerable and have never
been adequately discussed. Indeed, an exhaustive exploration of Bloomfield’s
appreciation of Formalism could well help to explain why so many young lin-
guists in the 1950s found that the techniques of structural linguistics appeared to
be compatible with the techniques employed by the formal sciences. In essence,
as the above passage demonstrates, Bloomfieldian linguistics and the formal
sciences were both shaped by Formalism during the 1930s, and the effects of this
influence are readily apparent in Bloomfield’s work. For instance, to consider
one example, it is well known that Bloomfield repeatedly expressed scepticism
concerning the role of meaning in linguistic theory. A standard expression of
this mistrust, taken from Language, runs as follows: ‘The statement of meaning
is . . . the weak point in language-study, and will remain so until human knowl-
edge advances very far beyond its present state’ (Bloomfield 1933: 140). In the
past, attempts to account for this scepticism have focused upon ideas concern-
ing syntax and semantics within linguistics, and upon the relationship between
linguistics and psychology.26 While there is no doubt that linguistics and psy-
chology were both responsible for determining the direction of Bloomfield’s
thought in many ways, it is certainly possible that some of his ideas concerning
the role of meaning in linguistic theory were directly influenced by his knowl-
edge of Formalism (and/or vice versa), which appeared to advocate the manip-
ulation of meaning-less symbols extracted from their semantic context. While
it would be needlessly excessive to claim that Bloomfield mistrusted linguistic
meaning primarily because he was intrigued by the methodology of Formal-
ism, it might well have been the case that his understanding of the foundational
96 Mathematical linguistics

debates within mathematics confirmed his initial misgivings about semantics

in linguistic research, causing him to marginalise the role of meaning in his
own work, and therefore unwittingly to pave the way for the type of ‘formal’
syntactic theories that began to emerge in the late 1940s and early 1950s.
Although Bloomfield’s 1939 essay is detailed and authoritative, it was not
his most extensive discussion of the relationship between linguistics and math-
ematics, for, two years previously, he had submitted a 300-page manuscript
on the topic to the Committee on Research of the Linguistic Society. The
proposed monograph was called The Language of Science and it constituted
an elaborate attempt to analyse large portions of modern mathematics from a
linguistic perspective.27 Indeed, this extraordinary document appears to have
constituted an attempt to resolve the foundations crisis by approaching the para-
doxes of set theory from the perspective of linguistic theory. With becoming
humility, the linguists on the committee considered themselves to be unequal
to the task of assessing the value of the manuscript, so it was passed on to sev-
eral professional mathematicians, including the prominent Formalist Haskell
Curry (1900–1982). Since the manuscript contained a few mathematical errors,
Curry advised against publication, but, despite his technical reservations, he
was impressed by Bloomfield’s approach and offered general advice as to how
the manuscript could be improved. However, Bloomfield decided that he would
not be able to revise the text so as to make it acceptable either to mathemati-
cians or to linguists, and, instead, he used the reverse sides of the pages as scrap
paper. Consequently, only a few fragments survive, but (fortunately) enough
remains to reveal the ambitious nature of the work. For instance, it is known
that there were chapters dealing with such topics as infinite classes, recur-
sion, logical vocabulary and syntax, and many other topics from (then) modern
mathematics. These glimpses of Bloomfield’s text hint at the full extent of his
familiarity with issues concerning the nature and function of logical syntax, and
it is particularly tantalising that several of the techniques he considered (i.e.,
those associated with recursive function theory and logical syntax) were to be
incorporated into TGG in the 1950s.
Despite Bloomfield’s discussion of the various topics mentioned above,
though, his main intention was not simply to provide an expository overview
of contemporaneous mathematics, but rather to solve some of the problems of
self-reference that had undermined set theory. His interest in the problem of
self-reference had been signalled in his 1935 paper when he had discussed the
differences between natural and mathematical languages. In particular, in that
paper he had identified two stages in the process of scientific activity which he
characterises as follows:
 Formal linguistic theory 97

The linguist naturally divides scientific activity into two phases: the scientist
performs ‘handling’ actions (observation, collecting of specimens, experi-
ment) and utters speech (report, classification, hypothesis, prediction). The
speech-forms which the scientist utters are peculiar both in their form and in
their effect upon hearers. (Bloomfield 1935: 499)

He later clarifies the nature of this peculiarity by observing that the language of
mathematics can only be understood after ‘severe supplementary training’, and
that utterances in such a language have the curious effect of causing the hearers
to ‘respond uniformally and in a predictable way’ (Bloomfield 1935: 499).
Clearly, therefore, the language of science differs significantly from natural
language, and the speech-forms of scientific language appear to constitute ‘a
highly specialized linguistic phenomenon’ (Bloomfield 1935: 500). It is at this
point that Bloomfield’s ambitious agenda starts to reveal itself. The following
passage is crucial:

To describe and evaluate this phenomenon is first and foremost a problem for
linguistics. The linguist may fail to go very far towards the solution of this
problem, especially if he lacks competence in the branches of science other
than his own. It is with the greatest diffidence that the present writer dares
to touch upon it. But it is the linguist and only the linguist who can take the
first steps towards its solution; to attack this problem without competence in
linguistics is to court disaster. The endless confusion of what is written about
the foundations of science or of mathematics is due very largely to the author’s
lack of linguistic information. (Bloomfield 1935: 500)

The central idea here is transparent: the complex and acrimonious arguments
that had come to characterise the mathematical foundations crisis debates in
the 1920s and 1930s could be resolved if only the participants were able to
view the problem from a linguistic perspective. Indeed, in Bloomfield’s words,
‘the linguist and only the linguist’ can intervene in order to resolve the dis-
putes. Obviously, this is a bold and startling claim, hence Bloomfield’s self-
confessed ‘diffidence’, but the proposal is serious nonetheless. Since (infuriat-
ingly) Bloomfield does not cite specific sources in his discussion, the precise
causes of his dissatisfaction with existing proposed solutions to the foundations
crisis can only be guessed. It should be recalled, though, that, as mentioned
previously, introductory texts such as Young’s Lectures on the Fundamental
Concepts of Algebra and Geometry predated the main foundational debates,
and consequently did not contain detailed discussions of the main disagree-
ments, suggesting that Bloomfield acquired his knowledge of these debates
from primary sources. As mentioned in section 3.5, some foundational issues
were addressed in certain works produced by the members of the Vienna Circle,
98 Mathematical linguistics

and Bloomfield certainly knew some of these publications. However, questions

concerning specifics inevitably remain. Had Bloomfield read the main publica-
tions associated with Hilbert or Russell? If so, precisely which publications had
he read? Certainly, references in Carnap’s Logische Syntax der Sprache (which
Bloomfield had read) would have provided him with information concerning
Hilbert’s most significant pre-1934 articles, and, by the mid-1930s, Russell and
Whitehead’s work, especially Principia Mathematica, had already become a
common starting point for most contemporary work in symbolic logic, and
was therefore hardly an obscure and unobtainable text. Whatever the precise
sources of his knowledge, though, it is clear that Bloomfield was well aware
of the fact that the paradoxes which had provoked the foundations crisis in the
early decades of the twentieth century were associated with specific kinds of
self-reference.28 Indeed, it is this aspect of the whole foundations debate that
seems to have intrigued Bloomfield most, since, as he was keen to demon-
strate, the basic problem of self-reference can be approached from a linguistic
perspective. His particular concerns are manifest in the following footnote in
which he reflects upon Kurt Grelling’s (1886–1942) well-known heterological
paradox.29
An adjective which describes itself is autological (e.g., short is autological,
since the adjective short is actually a short word). An adjective which is not
autological is heterological (e.g., long is not a long word). Is the adjective
heterological heterological? If it is heterological, it describes itself and is
therefore autological. If it autological, it does not describe itself and is therefore
heterological. (Bloomfield 1935: 500.n3)

Before continuing with the footnote it is worth pausing to clarify the discus-
sion. As should be apparent, Grelling’s heterological paradox is closely related
to Russell’s paradox (discussed in section 3.4 above), the main difference being
that, rather than being situated in the context of set theory, Grelling’s paradox
enables Bloomfield to view the problem from the perspective of natural lan-
guage so that it can be assessed from a different standpoint. However, a mere
restatement of a known difficulty is one thing, while a specific proposal for its
resolution is quite another, yet, as the footnote continues, this is precisely what
Bloomfield attempts:
The fallacy is due to misuse of linguistic terms: the phrase ‘an adjective which
describes itself’ makes no sense in any usable terminology of linguistics; the
example of short illustrates a situation which could be described only in a
different discourse. E.g.: We may set up, without very rigid boundaries, as to
meaning, various classes of adjectives. An adjective which describes a pho-
netic feature of words is morphonymic (e.g., short, long, monosyllabic). A
 Formal linguistic theory 99

morphonymic adjective which describes a phonetic feature of itself is auto-

logical. A morphonymic adjective which is not autological is heterological.
The adjectives autological and heterological designate meanings of adjectives
and not phonetic features; hence they are not morphonymic. – Contrast the
following sensible discourse: A hakab is a word that ends in a bilabial stop
(p, b). A word that is not a hakab is a cowp. The word hakab and cowp are
hakabs. (Bloomfield 1935: 500.n2)

Although this discussion is necessarily sketchy, constituting as it does a brief

footnote, the basic outline of Bloomfield’s proposal is clear: his intention was
to avoid the problem of direct self-reference by reanalysing the categorical allo-
cation of the words involved. In this simple example, by introducing the notion
of morphonymic adjectives, Bloomfield suggests that linguistic categories can
be redefined in order to exclude the type of direct self-reference that engenders
paradox. It is important to note that, for Bloomfield, this was specifically a
linguistic solution for a pervasive problem.
The radical stance of the unpublished 1937 MS is clear from the observa-
tion that it could be characterised as ‘contradicting the beliefs of mathemati-
cians’, though Bloomfield’s self-doubt is also apparent. Ultimately, his doubts
seem to have predominated, since, as mentioned above, he did not resubmit the
manuscript, and only fragments now survive. Thankfully, though, Bloomfield’s
basic motivation for writing the text is clearly articulated in one of the surviving
passages. For instance, after observing that no ‘student of human speech’ has
ever made an extensive study of mathematics, he continues:

Having made the attempt, the present writer has reached the conclusion that
such a study, apart from its linguistic interest, leads to the solution of cer-
tain problems that have baffled non-linguistic attack – the problems which
concern the foundations of mathematics. If this conclusion is justified, the
following pages should be of wider than linguistic interest. (Bloomfield
1970a[1937]: 335)

This is an extraordinary statement. As indicated above, in his 1935 article

Bloomfield had observed that certain problems of self-reference within mathe-
matics could be avoided if a linguistic approach were adopted. In the light of this
remark it becomes apparent that the now lost 1937 MS constituted an extended
attempt actually to provide a linguistic-based solution to the foundations cri-
sis. Although it is no longer possible to reconstruct Bloomfield’s arguments in
exhaustive detail, some kind of revivification can be accomplished. For instance,
as mentioned previously, a partial chapter list has survived, and, consequently,
it is known that the MS contained sections dealing with such topics as ‘infinite
classes’, ‘recursion’, ‘logical vocabulary and syntax’, and other subjects that
100 Mathematical linguistics

were active areas of contemporaneous mathematical research. The reference to

a chapter concerning ‘infinite classes’ is of especial interest since Bloomfield
delivered an (unpublished) paper on this topic to the Annual Meeting of the
Linguistic Society in 1936, and it was clearly a subject that preoccupied him at
the time.30 Given his familiarity with the foundations debates, this preoccupa-
tion is not surprising since, as mentioned in section 2.4, many of the paradoxes
of mathematics were understood to be associated with the notion of an infi-
nite set, and, therefore, any valid solution to the foundations crisis must either
reconsider the implications of such sets, or else must reformulate this aspect of
set theory in such a way that such sets were precluded.31 Indeed, the extant MS
fragments suggest that, in his 1937 text Bloomfield focused primarily upon the
task of naming infinite sets. For instance, he considers various methods that can
be used to define irrational numbers, and criticises the use of summation series.

The members of the summation series can be obtained one by one, but we
have no finite formula for the direct naming or recognition of these members.
To prescribe the naming, in this form, of an irrational number, is to insist that
our hearers complete the recitation of an infinite class of speech-forms. This
fallacy is still current among mathematicians; we shall return to it in Chapter
22. (Bloomfield 1970a[1937]: 337)

Unfortunately, Chapter 22 no longer exists, so Bloomfield’s discussion of this

perceived fallacy cannot be completely revived. However, his analysis of the
use of limits as a means of defining irrational numbers has survived, as has a
short section of his discussion of the  class. Bloomfield defines the  class
using linguistic notions associated with naming. He defines three activities:
(1) Say decimal point
(2) recite any sequence of digits or none
(3) name a second sequence of digits, not all zeros as a circulating
sequence
where a ‘circulating sequence’ is any sequence of numbers that repeats itself,
and Bloomfield concludes by asserting that ‘any speech-form of the shape
(1)-(2)-(3) or of the shape (1)-(3) is a member of the class ’ (Bloomfield
1970a[1937]: 338). With this definition in place, Bloomfield proceeds to con-
sider the implications of naming infinite sets.

Given the class , together with a formula for well-ordering it . . . we can

define, as functions of , infinite classes of speech-forms of the type N. For
instance, we add 1 to the kth digit of the kth R [Rs are defined earlier as ‘thing-
nouns’], except that when the sum is 10 we replace it by 1. We thus obtain
the infinite class of speech-forms N1 , the non-circulating decimals whose first
 Formal linguistic theory 101

ten digits are .5471111117. This formula for naming N1 , is stated in terms of
 and its well-ordering: a digit of N1 can be named only if one first names
k digits of the kth R. Hence to calculate and recite digits of N1 to the end of
one’s patience is not to name a number: it is only the formula N1 , interpreted
as above, which names a number. (Bloomfield 1970a[1937]: 338)

Although this remnant of a larger discussion is opaque in places, the basic thrust
of the passage is clear: the act of enumerating the members of an infinite class
(i.e., set) is not the same as naming the set, and, presumably, in the remaining
chapters of the MS, Bloomfield sought to demonstrate that the paradoxes of set
theory could be obviated if this kind of linguistic distinction were systematically
observed.
When the remaining MS fragments were collected by Hockett in 1970 for
inclusion in A Leonard Bloomfield Anthology (which he was then editing), he
commented concerning the destruction of the manuscript:

I cannot refrain from expressing my regret at this loss. Had he [i.e., Bloom-
field] lived to rework the topic, benefiting from Professor Curry’s suggestions
(even if not accepting them all), some of his successors, who have concerned
themselves with the inter-relations of language and mathematics, might have
been helped to avoid various stupid errors. (Bloomfield 1970b: 334)

Unfortunately, Hockett does not name the linguists who had been guilty of
making ‘stupid errors’, nor does he indicate which particular mistakes he has
in mind. There can be little doubt, though, that some of the names that fea-
ture prominently in the remaining chapters of this book may well have been
Hockett’s intended targets.
Bloomfield’s ideas concerning the relationship between mathematics and lin-
guistics and, in particular, his Formalist tendencies (whether overtly or covertly
expressed) which emphasised the primacy of syntactic (rather than seman-
tic) considerations, exerted a profound influence over a whole generation of
linguists that came to maturity in the 1940s and 1950s. It is important to recog-
nise, though, that this desire for a more mathematical approach to linguistic
analysis was not solely confined to North America. For instance, to consider
just one European example, Bloomfield’s interest in recent advances in math-
ematics was shared by the heterogeneous group of researchers that constituted
the Copenhagen Circle, and especially by Louis Hjelmslev (1899–1965). In his
1943 book Prolegomena to a Theory of Language, for example, Hjelmslev cited
Bloomfield’s 1926 paper in a footnote when he refers to ‘transcendent kinds
of linguists’ who have attempted to construct ‘systems of axioms’ (Hjelmslev
1961[1943]: 6). Hjelmslev not only shared Bloomfield’s interest in axioms and
102 Mathematical linguistics

deduction, though, he was also inspired by the methodology of Formalism, with

its emphasis on syntactic manipulation. In particular, according to Hjelmslev,
a description of a language should be ‘free of contradiction (self-consistent),
exhaustive, and as simple as possible’, linguists should assume ‘as few pre-
misses as possible’, and all definitions constructed should be ‘strictly formal’
(Hjelmslev 1961[1943]: 20). This explicit injunction to avoid contradictions,
along with the emphasis placed on formal structure, imply an informed
awareness of Hilbert’s proof theory, and Hjelmslev later explicitly discusses
Hilbert’s work in relation to the Lvov-Warsaw school, Carnap, and Saussurian
linguistics.

The logistic theory of signs finds its starting-point in the metamathematics of

Hilbert, whose idea was to consider the system of mathematical symbols as a
system of expression-figurae with complete disregard of their content, and to
describe its transformation rules in the same way as one can describe the rules
of a game, without considering possible interpretations. This method is car-
ried over by the Polish logicians into their ‘metalanguage’ and is brought to its
conclusion by Carnap in a sign-theory where, in principle, any semiotic is con-
sidered as a meaning expression system without regard for the content . . . The
sign-theory of linguistics, on the other hand, has deep roots in the tradition
according to which a sign is defined by its meaning. It is within this tradition
that Saussure struggles with the problem. (Hjelmslev 1961[1943]: 110–111)

This passage indicates that, for Hjelmslev, there was a clear difference between
the concerns of metamathematics and linguistics. The two disciplines were asso-
ciated by their status as sub-branches of semiotics, but linguistics could never
be subsumed by metamathematics because semantic considerations dominate
in the former while they are non-existent in the latter. Nevertheless, the above
passage clearly indicates that Hilbertian Formalism passed from the realm of
mathematics into the realm of linguistics in Europe as well as North America,
and that, for Hjelmslev (at least), the transmission of ideas was mediated by
‘the Polish logicians’ and Carnap.
While certain European linguists, such as Hjelmslev, were gradually devel-
oping a more formal approach to linguistic theory during the 1930s and 1940s,
similar developments were occurring in North America, and since TGG is part
of the story of North American linguistics, the focus of this discussion must
once again fall upon the United States. As already indicated, the influence of
Formalism is perhaps most apparent in the work of the post-Bloomfieldians
– a disparate group of linguists that included Bloch, Hockett, Yuen Ren Chao
(1892–1982), Rulon Wells (b. 1919), Martin Joos (1902–1978) and, of course,
Harris32 - who were, in many different ways, directing their research towards the
 Formal linguistic theory 103

task of formalising the discovery procedures that were believed to be required

for the sort of distributional linguistic analysis that had developed in the tradition
of Edward Sapir (1884–1939) and Bloomfield. Increasingly, during the 1940s,
the researchers working in this general area began to emphasise the desirabil-
ity of greater formal clarity. For instance, as partial justification for his 1947
development of Immediate Constituent analysis, Wells stated that his aim was
‘to replace by a unified, systematic theory the heterogeneous and incomplete
methods hitherto offered’ (Wells 1947: 81). Similarly, in the previous year, Har-
ris had claimed that the techniques introduced in his paper ‘From Morpheme
to Utterance’ constituted a ‘formalised procedure for describing utterances dis-
tinctly in terms of sequences of morphemes rather than of single morphemes’
(Harris 1946: 161). Indeed, the notion of a ‘formalised procedure’ was still suf-
ficiently new in 1946 to prompt the following footnote apologia from Harris:

In view of the fact that methods as mathematical as the one proposed here have
not yet become accepted in linguistics, some apology is due for introducing
the procedure. However, the advantage which may be gained in explicitness,
and in comparability of morphologies, may offset the trouble of manipulating
the symbols of this procedure. (Harris 1946: 161.n1)

The phrase ‘manipulating the symbols’ suggests the Formalist tendency of Har-
ris’ thinking (i.e., morphemic analysis is essentially viewed as an exploration
of symbol permutations), and Harris’ apology for using this kind of ‘mathemat-
ical’ procedure suggests that he himself believed that the nature of linguistic
research had begun to change. In the same way, many of the post-Bloomfieldian
papers written during the 1940s and 1950s reveal fragments of their intellectual
heritage which serve to illuminate the various influences that were prompting
these changes. However, it was only with hindsight that Harris himself compre-
hensively acknowledged the multifarious influences that had inspired his own
research during this period:

The expectation of useful mathematical description of the data of language

stems from developments in logic and the foundations of mathematics during
the first half of the twentieth century. One main source was the growth of
syntactic methods to analyse the structure of formulas, as in Skolem normal
form and Löwenheim’s theorem, and in the Polish School of logic (as in the
treatment of sentential calculus in J. Lukasiewicz, and the categorial gram-
mar of S. Leśniewski, and later K. Ajdukiewicz . . .), and in W. V. O. Quine’s
Mathematical Logic (Norton, New York) of 1940. Another source is in the
post-Cantor paradoxes constructionist views of L. E. J. Brouwer and the Intu-
itionist mathematicians, and in the specific constructionist techniques of Emil
104 Mathematical linguistics

Post and Kurt Gödel, in recursive function theory, and from a somewhat dif-
ferent direction in the Turing machine and automata theory . . . In linguistics,
the ‘distributional’ (combinatorial) methods of Edward Sapir and Leonard
Bloomfield were hospitable to this approach. Cf. also Nelson Goodman, The
Structure of Appearance . . . (Harris 1991: 145)

This catalogue of influences indicates that Harris was well aware of the major
developments in the philosophy of mathematics that had dominated the first
decades of the twentieth century, and that he believed these ideas had directly
influenced his own approach to the study of language. Also, once again we
encounter the claim that techniques derived from the formal sciences were
considered to be compatible with the distributional methods employed by the
post-Bloomfieldians. For Harris, the ‘distributional methods’ used in linguistics,
and originally associated with Bloomfield and Sapir, seemed to be ‘hospitable’
to the techniques of logical analysis, and this recalls Bar-Hillel’s assertion that
Ajdukiewicz’s system of logical syntax could be combined with mechanical dis-
covery procedures; and, as suggested above, Bloomfield’s Formalist tendencies
may well have been responsible for preparing the ground for this compatibil-
ity. In addition, it is worth noting that, in the same paragraph as that partly
quoted above, Harris goes on to refer explicitly to two contemporaneous text-
books, Kleene’s Introduction to Metamathematics and Church’s Introduction
to Mathematical Logic, citing them as sources that were used by linguists at
the time. Once again, therefore, the influence of these pedagogic texts upon
the linguistics community is apparent. Indeed, the relationship between logi-
cians and linguists was growing closer during the 1940s, and, in the context of
this rapprochement, it is significant that the manuscript of Church’s book was
proofread by the philosopher-cum-linguist Rulon Wells (Church 1944: vi).
Since a consideration of Carnap’s LSL has already occupied part of this
section, and since the text was to influence the development of TGG directly,
it is worth emphasising the fact that, although Carnap is not one of the authors
Harris cited explicitly in his 1991 reflections, several publications from the
1950s reveal the extent of Harris’ knowledge of LSL. For instance, in MSL,
Harris considers the general problem of linguistic analysis and observes in a
footnote:

It is widely recognised that forbidding complexities would attend any attempt

to construct in one science a detailed description and investigation of all the
regularities of language. Cf. Rudolf Carnap, Logical Syntax of Language 8,
‘Direct analysis of (languages) must fail [. . .]’. Linguists meet this problem
differently than do Carnap and his school. Whereas the logicians have avoided
the analysis of existing languages, linguists study them; but instead of taking
 Formal linguistic theory 105

parts of the actual speech occurrences as their elements, they set up very simple
elements which are merely associated with features of speech occurrences.
(Harris 1951: 16.n17)

Obviously, this passage indicates that Harris was familiar with Carnap’s work.
However, more than this, the content of the passage indicates that, far from
wholeheartedly advocating the adoption of techniques from logic for the pur-
poses of analysing natural languages, Harris was keen to stress the differences
that distinguish linguistics and logic. Essentially, the main difference appears
to be that linguists deal with natural languages, while logicians create artificial
languages. Nevertheless, this does not imply that, for Harris, logical languages
and natural languages were entirely different kinds of systems, rather that differ-
ent types of researchers (i.e., linguists and logicians) focus on different aspects
of the problem.
As suggested above, formal syntactic theories (especially the theory outlined
by Carnap in his LSL) exerted an influence on Bloomfield, Hjelmslev, Wells,
Harris, and many other linguists, particularly during the 1940s and 1950s. How-
ever, perhaps the most enthusiastic adherent of Carnapian syntactic theory in
linguistic circles was the irrepressible Bar-Hillel. Bar-Hillel’s undergraduate
training had been in mathematics and philosophy (not an uncommon route to
the theoretical study of language in the days when there were few undergraduate
linguistics courses on offer) and, in his preface to Language and Information
(1964), he provided a detailed overview of his own intellectual development.
Once again there is a polymathic cascade of names – Quine, Tarski, Bloomfield,
Reichenbach, Ajdukiewicz, and so on – but, in particular, he recalls the influ-
ence of Carnap. He first encountered LSL during the academic year 1936/1937
and,

For the next couple of years, I was seldom seen without a copy of this book
under my arm. My fellow students dubbed it ‘Bar-Hillel’s Bible’. It was doubt-
less the most influential book I read in my life, and a good part of my work is
directly or indirectly related to it. (Bar-Hillel 1964: 2)

Clearly, then, Bar-Hillel’s peers considered his devotion to LSL to have been
equivalent to a religious conviction, and some of the main tenets of Bar-Hillel’s
logico-syntactic faith are discussed in relation to TGG in section 4.5. It is of
interest, though, that Bar-Hillel had also encountered Bloomfield’s 1939 essay
concerning the scientific status of linguistics, and that this text too influenced the
development of his thinking. Once again his own comments are illuminating:
106 Mathematical linguistics

I think that the only work by a modern professional linguist I had studied
in some depth before these talks [i.e., talks with Harris in the early 1950s]
was Bloomfield’s little contribution to the Encyclopedia of Unified Science,
published in 1939. This booklet showed a surprising convergence between
ways of thinking of at least certain circles of American linguists and those of
say, Carnap, and I made a mental note to pursue this issue further sometime.
But only in 1951 did I find the time to do so. (Bar-Hillel 1964: 4)

This passage indicates that, for Bar-Hillel at least, Bloomfield’s work suggested
the possibility of integrating techniques from Carnapian logical syntax with the
methodology of linguistics. As already indicated, this kind of project was cer-
tainly not unique since, by the early 1950s, there was a general tendency in
certain linguistic circles to seek formal linguistic theories (particularly syntac-
tic theories) which advocated meaning-less symbol manipulation rather than
contentual, semantics-based analysis, and, as discussed above, Carnap’s LSL
was a prominent text that appeared to provide formal analytic techniques that
could be usefully transferred from the realm of logical syntax into the sphere
of linguistics.

3.8 New directions

To summarise the discussion so far, the topics considered in this chapter have
included the use of the axiomatic-deductive method in linguistic research, the
use of recursive definitions to analyse sentences in natural languages, the use
of logical categories to analyse the notion of grammaticality and to construct
syntactic descriptions, the creation of constructional system theory as an exten-
sion of logical analysis, the need for simplicity measures, the development
of formal linguistic theories, and the influence of such theories upon syntac-
tic research. The various connections between mathematics, metamathematics,
logic, and linguistics implied by the research discussed in the preceding sections
largely motivated the interest in formal analytical procedures that became such
a characteristic feature of syntactic research in the 1950s, and, not surprisingly,
these developments did not go unnoticed. In 1949, for instance, Carl Borgström
(1909–1986) acknowledged that there had been a recent shift away from the
analysis of particular linguistic phenomena towards a preoccupation with the
general analytical discovery procedures themselves, and he made an explicit
distinction between ‘basic-research’, which concerned itself with the analysis
of language, and ‘meta-research’, which concerned itself with the analysis of
the analysis of language:
 New directions 107

Meta-research is a kind of research in which the processes (secondary expe-

riences) of previous research are turned into objects of research (primary
experiences) by being subjected to new processes (e.g., by being named and
described); the previous research is the basic research. (Borgström 1949: 4)

The strongly Hilbertian character of this terminology is immediately apparent.

In discussing this recent shift, though, Borgström did not fail to recognise the
similarities between the work of the post-Bloomfieldians and the work of the
Copenhagen Circle, particularly Hjelmslev, and he opined that both traditions
were becoming interested in ‘form’ over ‘substance’ (Borgström 1949: 9). By
the early 1950s the similarities between the two movements were so conspic-
uous that they were being discussed in detail. In 1951, for example, Einar
Haugen (1906–1994) noted the recent interest in ‘metalanguage’ amongst lin-
guists, indicating that it had its roots in Formalist-inspired logical analysis, and
commenting that the post-Bloomfieldians and the Copenhagen Circle were
both attempting to give a mathematical formulation to linguistic statements.
Harris has described his syntactic analyses as ‘mathematical’; Hjelmslev
declares his purpose is the creation of a ‘linguistic algebra’. Harris expressly
points out that his analysis is purely formal; Hjelmslev describes his theory as
being based on ‘an exclusively formal set of postulates’. (Haugen 1951: 213–
214)

Haugen concludes by claiming that the formal procedures advocated by the

post-Bloomfieldians do not constitute ‘a linguistics as we have known it, but
rather a metalinguistics’ (Haugen 1951: 212). Linguistics papers now ‘bristle
like a page of symbolic logic’, and Haugen admits that, although he would not
change the current emphasis, he misses ‘the leisurely, even charming quality
of the traditional grammars’ (Haugen 1951: 222). Nevertheless, the charming
traditional grammars were never to return since, as will be demonstrated in
chapters 4 and 5, it was this new, complex, interdisciplinary, intellectual culture,
with its drive towards ever-increasing rigour and precision of formal exposition,
that directly inspired the creation of TGG.
4 Systems of syntax: 1951–1955

4.1 Chapter overview

The main purpose of this chapter is to consider Chomsky’s research from the
years 1951–1955 in the light of the topics presented in the previous chapters,
and, to renew the words of warning in the introduction, it is at this stage of
the discussion that the convenient abbreviation ‘TGG’ ceases to be especially
helpful. As indicated previously, the main problem is that, during this period,
Chomsky explored a number of different approaches to syntactic theory and,
though various aspects of these approaches were maintained and further devel-
oped, others were swiftly discarded. As a result, it is not possible simply to use
the term ‘TGG’ with reference to Chomsky’s work of the early 1950s as if it
denoted a clear and consistently identifiable grammatical theory. Consequently,
it is necessary either to accept that TGG must be viewed as a fluid concept that
altered continually during this period, or else to avoid using the term altogether
(whenever possible). In this chapter and the next the latter course is adopted,
since it necessitates more specific reference to the various ideas that Chomsky
explored during this period and so avoids confusion.
The main sections of this chapter focus upon the influences that shaped Chom-
sky’s earliest work. The basic approach is to identify the presence of a particular
influence, and then to trace its development as his research gradually matured
during the 1950s. Consequently, the intention is not simply to provide a strict,
consistently chronological study of the gradual evolution of TGG as a whole.
Rather, a number of distinct (but fundamentally related) themes are considered,
with the result that, as the various topics are presented, a more complete pic-
ture of the overall development of TGG starts to emerge. This type of approach
enables the continuities and discontinuities to be revealed more clearly. It should
be noted, incidentally, that the task is not only to identify those techniques and
ideas that Chomsky actually adapted and used in his own linguistic research,
but also to consider some of the techniques that he rejected, since a negative
assessment can often be as informative as a positive advocation. With this in

108
 Biography and influences 109

mind, the basic chapter plan can be summarised as follows. In section 4.2 some
biographical information about Chomsky is presented, with particular attention
given to his contact with some of the individuals mentioned in chapter 3. This
enables direct connections between his work and that of his most influential pre-
decessors to be established. In section 4.3 the influence of Goodman is explored
by examining the way in which Chomsky reformulated the notion of simplicity
in syntactic terms. This consideration of simplicity naturally invites an assess-
ment of other aspects of Goodman’s influence upon the young Chomsky, and
section 4.4 explores his brief flirtation with constructive nominalism. Finally,
in section 4.5 some of the arguments concerning the relationship between syn-
tactic theory and the formal sciences are considered. In particular, the debate
between Bar-Hillel and Chomsky concerning the relationship between logic
and linguistic theory that took place during the years 1954–1955 is assessed,
since it provides valuable insights into Chomsky’s whole approach to syntactic
theory.

4.2 Biography and influences

The basic facts concerning Chomsky’s early life and intellectual development
are well known and a comprehensive account will not be attempted here. Instead,
the following discussion will focus on Chomsky’s relationships with those
linguists and philosophers who appear to have exerted the greatest influence
over his early work. More information concerning the issues considered in the
following discussion can easily be found in the standard texts.1
Chomsky was born on 7 December 1928 and the whole of his early life was
centred in Philadelphia, Pennsylvania. His father, William Zav Chomsky (1897–
1977), a Hebrew teacher, had left Russia with his wife for the United States
in 1913 and, in time, he became a noted scholar, publishing numerous books
on a range of topics including linguistics and education policy. He specialised
in medieval Hebrew and one of his most important publications, Hebrew, the
Eternal Language, appeared in 1957. Chomsky’s mother, Elsie Simonofsky
(1903–1972) also taught Hebrew and is believed to have been instrumental in
inculcating a strong sociopolitical conscience in her eldest son. Whatever the
exact nature of his mother’s influence upon his intellectual development, from
1930 to 1940 Chomsky attended Oak Lane County Day School, an experimental
institution that eschewed the more formal aspects of conventional education.
Consequently, it was not until he began to attend the Central High School in
Philadelphia that he experienced traditional teaching methods for the first time.
In 1945, at the age of 16, he enrolled in a general undergraduate programme
110 Systems of syntax: 1951–1955

at the University of Pennsylvania, but disliked the experience and considered

leaving in order to join one of the flourishing kibbutzim in Israel. However,
in 1947 he encountered Zellig Harris, who encouraged him to continue with
his formal education, and there is no doubt that Harris was instrumental in
encouraging Chomsky’s interest in theoretical linguistics. Indeed, Chomsky
would claim later that his introduction to linguistics occurred when Harris
asked him to proofread the manuscript of MSL. Chomsky submitted his B.A.
thesis, The Morphophonemics of Modern Hebrew in 1947, and, in the same year,
entered the Graduate School at Pennsylvania. Two years later he completed his
M.A. thesis, which (curiously) was simply a revised version of his B.A thesis,
and, during this period, prompted by Harris, he began to study philosophy
with Goodman and White, as well as mathematics with Nathan Fine (1916–
1994). The association with Goodman in particular proved to be advantageous
for, in 1951, Goodman helped Chomsky secure a Junior Fellowship in the
Society of Friends at Harvard. The fellowship lasted until 1955 and, during this
period, Chomsky came into close contact with Quine and the other members
of the Harvard intellectual elite. The Society was instituted in 1933 by Abbott
Lawrence Lowell (1856–1943), who was President Emeritus of Harvard at the
time. When Chomsky joined in 1951, the fellowship included (amongst others)
Kuhn and Marvin Minsky (b. 1927), in addition (of course) to Goodman and
Quine, and, in general, the Harvard community was strongly influenced by
James Bryant Conant (1893–1978), who was president from 1933 to 1953.
Despite the attractions of this environment, the fellowship did not come with
tenure, and in 1955 Chomsky left Harvard for the Massachusetts Institute of
Technology, and he has remained on the academic staff there ever since.
Some of Chomsky’s later observations concerning his education and his
former teachers merit consideration here, since they constitute his own reflec-
tions upon some of the issues considered in the main chapters of this book.
For instance, he has suggested that Harris was never particularly interested in
linguistic theory per se, concerning himself instead with practical methods:

[Harris] thought of linguistics as a set of procedures for organising texts, and

was strongly opposed to the idea that there might be anything real to discover.
He did think that the methods of linguistic analysis could be used for analysis
of ideology, and most of my actual graduate courses were devoted to that . . .
(quoted in Barsky 1997: 52)

This perception of Harris’ work seems to have been commonplace. For exam-
ple, in the 1960s Hockett had referred to Harris’ ‘theoretical nihilism’ (Hockett
1968: 35). However, despite Harris’ early interest in Chomsky, the two men
 Biography and influences 111

drifted apart as Chomsky matured. To quote a portion of Barsky’s rather syco-

phantic summary (which is useful mainly because it draws heavily upon Chom-
sky’s own recollections),

even at this early stage [i.e., early 1950s], Chomsky was producing highly
original work, which diverged fundamentally from Harris’. In his B.A. thesis
he was doing things that were, in his own words, ‘radically at odds with
everything in structural linguistics . . . which is why [both the thesis and The
Logical Structure of Linguistic Theory] were published only 30 years later’.
The thesis was ‘as different from structural linguistics as anything could be’,
which was why ‘Harris never looked at it and no one in the field reacted to it’.
(Barsky 1997: 53)

In a later passage Chomsky suggests that Harris considered the nascent theory
of TGG to be ‘crazy’ (Barsky 1997: 54), although no evidence to support this
claim is provided. Strangely, the basic pattern of Chomsky’s relationship with
Harris (or at least Chomsky’s own account of it) is identical to the basic pattern
of his relationship with Goodman. At first Goodman encouraged and supported
Chomsky, helping him to obtain a Junior Fellowship at Harvard, but then the
association turned sour. The following passage is also taken from Barsky’s
account, mainly because (once again) it weaves together some of Chomsky’s
own words. After mentioning (with no details) that Chomsky began to study the
work of Carnap, Frege, Whitehead, and Wittgenstein (no mention of Russell?),
Barsky continues,

the readings Chomsky now undertook gave him a fresh perspective that his
teacher Nelson Goodman considered to be ‘completely mad’. When Goodman
found out about Chomsky’s work in the mid-1960s, he apparently ended their
friendship, even though, as Chomsky says, they’d ‘been quite good friends
until he learned about this, which he regarded somehow as a personal betrayal’.
(Barsky 1997: 54)

There are various problems with this account. First, it is not at all clear why
Goodman would have been offended by any ideas Chomsky might have gleaned
from Carnap, Whitehead, and Wittgenstein (or even Russell). Second, as will be
shown in section 5.3, if anything caused Chomsky to find a ‘fresh perspective’
during the early 1950s, it was Goodman’s own work. Third, it is not clear
why it took Goodman ten years to find out about Chomsky’s research into
syntactic theory. By the mid-1960s Chomsky’s name was already well known
and Goodman must have been living as a hermit not to have heard something
about TGG before c. 1965. Curiously, though, Chomsky’s memories of his
friendship with Quine outline a similar pattern: at first Quine had encouraged
112 Systems of syntax: 1951–1955

him (for instance, by reading his 1953 paper before it was submitted), but then
‘lost interest’ (quoted in Barsky 1997: 93) in Chomsky’s work.
This general pattern of initial closeness followed by sudden separation is
clearly of interest and it raises various questions: why did Chomsky come to be
spurned by his early mentors, and precisely which aspects of his work offended
them? Whatever the truth of Chomsky’s recollections, though, as will be shown
in the next two chapters, there can be no doubt that, during the early 1950s,
Chomsky was working closely with Harris, Goodman, Quine, and others, and
that he was directly influenced by some of their ideas. Accordingly, the task of
revealing traces of these influences in his work from the 1950s is accomplished
in the following sections, and it may be the case that an exploration of this
kind proves to be of greater interest than a protracted attempt at retrospective
psychological analysis.

4.3 Simplicity and grammar

As indicated in the introduction, the importance of the notion of ‘simplicity’ in
Chomsky’s early syntactic research has never really been adequately discussed,
even though Chomsky himself frequently acknowledged the source of his ideas
at the time, and even though he has recently suggested that this aspect of his early
work has been revived (albeit in a different form) in his most recent research.2
In this instance, even the arduous nature of the requisite intellectual archae-
ology cannot truthfully be cited as a valid reason for the neglect, since, as was
shown in section 3.5, the task of devising a simplicity measure for extralogical
bases in constructional systems was an abiding preoccupation for Goodman
throughout the 1940s and 1950s, during which time he was lecturing both at
Pennsylvania and Harvard; and since it is known that Chomsky studied under
Goodman directly during this time, it is no real surprise that similar consider-
ations figure prominently in Chomsky’s earliest known research. Indeed, there
are many traces of Goodman’s thinking in Chomsky’s work from the 1950s,
but, as will become apparent below, even at this early stage Chomsky was not
simply borrowing techniques from one field and mindlessly employing them,
unmodified, in another area. Rather, he was keen to shape and adapt the meth-
ods he had encountered in an attempt to facilitate the specific task of syntactic
analysis, and this process of adaptation can be identified in his 1951 Master’s
thesis The Morphophonemics of Modern Hebrew (hereafter MMH).
In the introduction to MMH, Chomsky states that a grammar constructed for
a given language must meet two kinds of adequacy criteria. First, it must ‘cor-
rectly describe the “structure” of the language’; that is, it must ‘isolate the basic
 Simplicity and grammar 113

linguistic units and, in particular, must distinguish and characterise just those
utterances which are considered “grammatical” or “possible” by the informant’
(Chomsky 1979b[1951]: 1). Second, the grammar must either meet criteria that
are imposed by the ‘special purpose’ for which the grammar was created (e.g.,
pedagogic utility), or else, if there is no such purpose, it must meet require-
ments of ‘simplicity, economy, compactness, etc.’ (Chomsky 1979b[1951]: 1).
From the very beginning, then, Chomsky seems to have been persuaded that
considerations of simplicity were intimately involved in the processes of gram-
mar construction, and the task of unearthing the sources of these ideas is fairly
simple, mainly because Chomsky never attempted to conceal his intellectual
debts. For instance, in MMH he explicitly cites Goodman’s 1943 paper, ‘On the
Simplicity of Ideas’, and paraphrases Goodman’s argument against simplicity
being considered merely an aesthetic luxury in a constructional system (quoted
in full in section 3.5 above). With particular reference to syntactic theory, one
of the implications of Goodman’s views concerning the critical importance of
simplicity criteria in constructional systems is that the reasons for wanting a
grammar to be as simple as possible are the same as for wanting a grammar
at all. Crucially, though, it should be noted that this focus upon theory-internal
economy considerations was not encouraged by Harris, who was the other main
influence upon the form and content of MMH. For instance, in MSL Harris had
clearly stated that ‘it is a matter of other than descriptive purpose how compact
and convenient the formulation is, or what other qualities it may have’ (Harris
1951: 9), and he had gone on to claim that

It therefore does not matter for basic descriptive methods whether the system
for a particular language is so devised as to have the least number of elements
(e.g., phonemes) or the least number of statements about them, or the greatest
over-all compactness, etc. These different formulations differ not linguistically
but logically. They differ not in validity but in their usefulness for one purpose
or another (e.g., for teaching the language, for describing its structure, for
comparing it with genetically related languages). (Harris 1951: 9.n8)

For Harris, then, practical utility, rather than considerations such as simplicity,
was the guiding principle of grammar construction, and this may well be a good
example of Harris’ aforementioned ‘theoretical nihilism’ (Hockett 1968: 35).
Whether this is the case or not, the clear implication of the above passage is
that Harris deemed issues such as grammatical ‘compactness’ to be extraneous
to the task of linguistic analysis, or, at the very least, he was convinced that
such considerations should only be assessed with reference to the utility of the
grammar. In other words, if simplifications facilitate the use of the grammar as
114 Systems of syntax: 1951–1955

a pedagogic tool, then they are valuable; if not, then they are of no real interest.
The clear juxtaposition of Goodman’s repeated emphasis on the importance of
theory-internal economy and Harris’ apparent indifference to the same must
have intrigued Chomsky at the time, and, indeed, MMH can be viewed partly
as an attempt to introduce Goodman-style simplicity criteria into a Harris-
style analytic framework. Already, therefore, Chomsky’s remarkable ability to
reconcile and synthesise existing techniques is clearly apparent in the perceived
analogy between syntactic analysis and constructional system theory.
Having outlined the basic problem in this general fashion, Chomsky goes on
to distinguish between ‘discovery’ processes (e.g., Harris’ distributional proce-
dures) and processes of ‘description’. The distinction between these processes
is clear: discovery processes can be used to determine the set of grammatical
sentences in a corpus, while processes of description can be used to analyse the
form of those utterances. In this context, it is worth recalling that Bar-Hillel’s
revival of Ajdukiewicz’s system of syntactic analysis (discussed in section 3.4)
was intended to provide a method of ‘description’ (Bar-Hillel 1953b: 47) and
was not intended to constitute a discovery procedure. In the early 1950s, then,
the distinction that Chomsky’s draws was certainly not unique. However, he pro-
ceeds to extend the argument by suggesting that, although this distinction might
be easy to make when a grammar is constructed for a ‘special purpose’, it is less
useful when it is created without a particular use in mind, for, in this case, there
are no arbitrary constraints imposed upon the grammar. Consequently, in the
latter case, the grammar must be constructed ‘solely in accordance with consid-
erations of elegance’ (Chomsky 1979b[1951]: 2). Indeed, Chomsky goes on to
suggest that the very notion of ‘elegance’ itself threatens to undermine the basic
distinction between discovery and description processes, since such considera-
tions are fundamental to the task of discovery and to the task of description. In
the light of this observation, it is worth emphasising that, from the very begin-
ning, Chomsky’s focus was upon syntactic analysis for the sake of syntactic
analysis. He appears never to have been especially interested in the practical
(i.e., pedagogical) applications of the sort that so preoccupied Harris.
Despite the fact that Chomsky explicitly places the issue of grammatical sim-
plicity at the core of the theoretical framework he outlines in MMH, he does not
provide an exhaustive discussion of the topic. He observes that merely counting
the number of rules in the grammar is an inadequate way of measuring sim-
plicity, and he refers somewhat obliquely to an unpublished paper of his own
that considers the topic, but he does not summarise the contents of that paper in
his thesis.3 Nevertheless, Chomsky maintained his interest in simplicity mea-
sures long after his M.A. thesis was submitted, and the most comprehensive
 Simplicity and grammar 115

discussion of the subject was offered in chapter 4 of his manuscript The Logical
Structure of Linguistic Theory (hereafter LSLT), the first draft of which was
completed by early 1955.4 In this work the importance of grammatical simplic-
ity is stressed repeatedly and (once again) it is considered to play a fundamental
role in the task of grammar construction.

In linguistic theory we face the problem of constructing this system of levels

in an abstract manner, in such a way that a simple grammar will result when
this complex of abstract structures is given an interpretation in actual linguistic
material. (Chomsky 1975[1955]: 100)

In other words, it is desirable that a general theory of linguistic structure is

designed so as to permit the creation of simple grammars for given languages.
As Chomsky puts it later in the same chapter, the ‘reduction of the complexity of
grammar is one of the major motivations behind level construction’ (Chomsky
1975[1955]: 100).5 Given this general goal of grammatical complexity reduc-
tion, it is obvious that the notion of ‘simplicity’ requires adequate elucidation
in the specific context of linguistic theory. However, before attempting to clar-
ify these terms, Chomsky is keen to demonstrate why the notion of simplicity
is of considerable importance when a general theory of linguistics is being
devised. In particular, he argues that grammatical simplicity is significant pri-
marily because it can provide a way of choosing between various competing
grammars. In other words, if the degree of grammatical simplicity could be
measured consistently, then, given a corpus and various grammars all of which
exhaustively generate all the grammatical utterances in the corpus, a simplicity
measure of some kind could provide a viable evaluation procedure that would
enable one grammar to be preferred above all the others, and the centrality of
decision criteria that can be used to select one from amongst several competing
grammars is stressed throughout LSLT.6

Linguistic theory must enable us to choose among proposed grammars, and

every consideration relevant to the choice must be built into the theory. So far,
the only consideration we have placed on the grammar is that the system of
levels it determines . . . must be of the form required by linguistic theory. But
this feature of grammatical construction leaves out one of the most important
and characteristic features of grammar construction. In careful descriptive
work, one almost always finds that one of the considerations involved in
choosing among alternative analyses is the simplicity of the resulting grammar.
If we can set up elements in such a way that very few rules need to be given
about their distribution, or that the rules are very similar to the rules for other
elements, this fact certainly seems to be a valid support for the analysis in
question. It seems reasonable, then, to inquire into the possibility of defining
116 Systems of syntax: 1951–1955

linguistic notions in the general theory partly in terms of such properties of

grammar as simplicity. (Chomsky 1975[1955]: 114)

This passage neatly summarises the core motivation for an assessment of sim-
plicity in the context of linguistic theory: simpler grammars are preferred
because they capture more general linguistic patterns than more complex gram-
mars. Clearly, this assumption echoes Goodman’s pro-simplicity arguments,
which were discussed in detail in section 3.5. For instance, Goodman had
asserted that ‘the more the set of postulates can be reduced without becom-
ing inadequate, the more comprehensively will the system exhibit the network
of interconnections that comprises its subject matter’ (Goodman 1943: 107),
and this observation is clearly related to various passages in LSLT such as that
quoted above. Specifically, in linguistic terms, a measure of grammatical sim-
plicity is required in order to permit the ‘best’ grammar for a given corpus to
be selected from among all other possible grammars, and the task of creating
such a simplicity measure is precisely the one that Chomsky sets for himself
in chapter 4 of LSLT. Before discussing his treatment of this topic in detail,
though, it is worth highlighting his belief (already expressed in MMH) that the
technical issue of grammatical simplicity is not merely a superficial adjunct to
linguistic theory. Indeed, Chomsky bluntly asserts that the notion of simplicity
is of fundamental importance, and he admits that his thinking concerning this
issue has been guided by the work of both Goodman and Quine.

It is important, incidentally, to recognise that considerations of simplicity are

not trivial or ‘merely esthetic’. It has been remarked in the case of philosophical
systems that the motives for the demand for economy are in many ways the
same as those behind the demand that there is a system at all. See Goodman,
‘On the simplicity of ideas’, where the reference is to economy in the basis of
primitives. It seems to me that the same is true of grammatical systems, and of
the special sense of simplicity that will concern us directly. See Quine, From
a Logical Point of View for recent discussion of the role of simplicity in the
choice of scientific theories. (Chomsky 1975[1955]: 114.n2)

This footnote is actually an extended, more specific version of footnote 1 in

MMH and the main difference between the two notes is that in the 1951 ver-
sion only Goodman’s work is cited, while in the above version Quine’s recent
arguments concerning the role of simplicity in scientific theories in general are
enlisted as additional support. Clearly, the basic thrust of Chomsky’s argument
is that, since simplicity plays a crucial role in the creation of any theoretical
system, it is necessarily involved in the task of grammar construction. Conse-
quently, it is expedient to assess the way in which the notion of simplicity is
 Simplicity and grammar 117

presented in LSLT in some detail, since it reveals the manner in which Chomsky
modified some of Goodman’s and Quine’s ideas so that they could be used to
facilitate the specific task of syntactic analysis.
At the outset of his discussion of grammatical simplicity in chapter 4 of LSLT
Chomsky openly confesses that his current thinking on this topic is still at a
preliminary stage and that, therefore, it is ‘sketchy and incomplete’ (Chomsky
1975[1955]: 116). Despite this, he is sufficiently confident about the issues
involved to be able to propose a series of consolidation rules that converts a
given grammar into a maximally condensed form. Once again, this emphasis on
rule-driven processes recalls Harris’ interest in automatic discovery procedures
(mentioned in section 3.7), as well as Bar-Hillel’s enthusiasm for syntactic
categories that enable a sentence to be analysed ‘mechanically’ (discussed in
section 3.4). In the present context, then, Chomsky’s desire for an algorithmic
approach to simplicity measures is in keeping with the general drift of syntactic
research in the 1950s. In addition, though, the search for a mechanical procedure
that would permit an automatic measurement of grammatical simplicity again
echoes Goodman who, in his 1943 paper, advocated the avoidance of intuition
when determining the respective simplicity of various bases, favouring instead
an ‘automatic procedure’ (Goodman 1943: 108). Given the fact (discussed at
length in section 5.4) that the definition of a linguistic level offered in LSLT
draws heavily upon constructional system theory, it would be reasonable to sus-
pect that, following Goodman, Chomsky would focus on the task of reducing
the number of primitive relations associated with a particular syntactic level.
However, since each level of the general system outlined in LSLT uses con-
catenation as its sole primitive relation, reducing the number of such relations
was not a feasible option. Consequently, Chomsky realised that, if a measure of
grammatical simplicity were ever to function as a valid evaluation procedure, it
must operate upon something other than the extralogical bases of the linguistic
levels.
Accordingly, developing the scheme outlined in MMH, Chomsky argued that
the basic formal components of the grammatical theory developed in LSLT that
must be involved in any technical definition of simplicity are the rewrite rules,
called ‘conversions’ in the early TGG literature. In the standard TGG notation,
rewrite rules take the form ‘X → Y ’, and they are used to permit the analysis
of larger syntactic units into smaller sub-components. For instance, a standard
P-level (i.e., phrase-level) conversion is ‘Sentence → NP VP’, which states
(obviously) that a sentence can be rewritten as a noun phrase followed by a
verb phrase. It is reasonable that such rules should be involved in the techni-
cal definition of a simplicity measure, since a measure of any kind necessarily
118 Systems of syntax: 1951–1955

involves some sort of quantification, and the conversions in a given grammar

can be counted easily. Indeed, the assumption that conversions could provide
a viable foundation for a simplicity measure underlies Chomsky’s whole dis-
cussion of the subject in LSLT, and he states clearly that ‘it is tempting . . . to
consider the possibility of devising a notational system which converts consider-
ations of simplicity into considerations of length’ (Chomsky 1975[1955]: 177).
In other words, as an initial proposal, Chomsky felt that it was reasonable to
assume that grammars containing fewer conversions (i.e., shorter grammars)
should be preferred over grammars containing a larger number of conversions.
This basic approach, with its emphasis on the trivial enumeration of key com-
ponents, is reminiscent of Goodman’s preliminary statements concerning the
relative simplicity of extralogical bases in constructional systems: systems with
fewer extralogical predicates are preferred over systems with a larger number of
predicates. However just as Goodman swiftly realised that such a trivial measure
was too crude, so Chomsky quickly rejected the trivial approach to grammatical
simplicity outlined above (just as he had done in MMH), and the motivation for
the modifications he goes on to propose comes from the insight (again related
to Goodman’s work on constructional systems) that prioritising shorter gram-
mars indirectly prioritises grammars that attain greater generality; and such
generality is highly valued: if a particular sub-sequence of conversions can be
replaced by a single conversion, then the grammar more accurately captures
a general fact about the corpus being analysed. It should be noted that, while
stressing the importance of generality and the relationship between generalising
and enumerating conversions, Chomsky was keen to avoid the misconception
that generality and length are exactly equivalent concepts in this context. Conse-
quently, he drew a clear distinction between ‘real’ simplicity and all other kinds.
As an example, he observed that a complex grammar, represented schemati-
cally as Q(a1 , . . ., an ), could be defined as f (x1 , . . ., xn ), with the result that
f (a1 , . . ., an ) provides a concise (i.e., one-line) representation of the complex
initial grammar (Chomsky 1975[1955]: 118). However, as Chomsky demon-
strated, this modification is nothing more than a notational sleight of hand that
effects no actual simplification of the grammar, and this type of argument recalls
Goodman and Quine’s distinction between the ‘real’ and ‘apparent’ economy
of extralogical bases (discussed in section 3.5). For Chomsky then, following
Goodman and Quine, the task is to avoid superficial simplifications that do not
actually contribute to the reduction of system complexity.
In order, therefore, for a coherent simplicity measure to be constructed, it
is necessary to ensure that a given grammar is maximally ‘consolidated’, and
that the consolidation constitutes a real reduction in the complexity of the
 Simplicity and grammar 119

grammar. Accordingly, Chomsky introduces a number of ‘notational transfor-

mations’ (Chomsky 1975[1955]: 118) which essentially define a normal form
for the grammar. In other words, given the set of notational transformations, any
grammar can be reduced to a maximally condensed format, which permits the
direct comparison of competing grammars. For instance, adopting the notation
‘{}’ to indicate positional equivalence and ‘−’ to indicate the null element, it
follows that the three strings

adeg (4.1)
bdg (4.2)
cd f g (4.3)

can be replaced by the single consolidated string

{a, b, c}d{e, −, f }g (4.4)

It is clear that (4.4) is more general than the separate strings in (4.1)–(4.3),
and that it is not simply a vacuous tautological restatement of pre-existing
facts, since it makes more particular claims about the distributional properties
of the formal language used in the example. As Chomsky observes, since this
kind of consolidation can be applied to both sides of a given conversion rule,
these notational manipulations enable a maximally consolidated grammar to be
constructed in such a manner that real complexity reduction is assured.

The notational devices we have introduced permit certain selected features

of similarity among statements of the grammar (i.e., each type of partial
generalisation) to effect a decrease in length, so that grammars whose rules
have these features are more highly valued. Thus these constructions can be
understood as offering an analysis for certain aspects of simplicity. (Chomsky
1975[1955]: 123)

Having devised, in this fashion, a coherent measure of simplicity based upon

conversion consolidation, Chomsky goes on to discuss the issue of conversion
ordering. It is obvious that conversions in a formal grammar cannot simply be
listed in an ad hoc fashion: rules converting phrases into morphemes, for exam-
ple, must apply before morphemes can be converted into phonemes. Indeed, it
is this observation that causes Chomsky to refer to the conversions in a given
grammar as constituting a ‘sequence’ rather than a set, since the order of ele-
ments in a set is irrelevant, while a sequence is determined precisely by the order
of its elements. Consequently, the issue of conversion ordering is considered to
be sufficiently important to merit explicit specification within a general theory
of linguistic structure.
120 Systems of syntax: 1951–1955

There is a great advantage in giving the principle of ordering once and for all
in the general theory. Otherwise, the gain in economy resulting from ordering
of rules in a particular grammar will be much reduced because of the need
to specify, within that grammar, the order of application of rules. The most
favourable situation is one in which we can linearly order the conversions in
such a way that all derivations can be formed by running through this sequence
from beginning to end. (Chomsky 1975[1955]: 125)

The implication here is that the conversions that constitute the grammar should
be so arranged that all grammatical utterances in a given corpus can be cor-
rectly derived simply by applying the various conversions in sequence. This
observation leads Chomsky to propose certain ‘optimality conditions’, which
he introduces as follows:

Putting it roughly, a grammar will meet these conditions if, when the rules are
given in a maximally condensed form, it is possible to arrange the resulting
statements in a sequence in such a way that:
(i) we can form all derivations by running through the sequence of rules from
beginning to end;
(ii) no conversion X → Y will appear twice in the sequence (i.e., no rule need
be repeated in several forms at various places in the grammar);
(iii) each conditioning context is developed to exactly the extent relevant for
the application of the rule in which it appears. (Chomsky 1975[1955]: 125)

These practical conditions define a heuristic procedure similar to that proposed

by Goodman for extralogical bases in his 1952 paper ‘New Notes on Simplicity’
discussed in section 3.5. In practice, these optimality conditions should be
applied at each linguistic level and they are designed to ensure (i) the linearity
of the grammar, (ii) the avoidance of redundancy, and (iii) the validity of context-
dependent rules. Given this set of conditions, it becomes possible to resolve the
problem of choosing between formal grammars: given two grammars, both
of which generate all the grammatical sentences in a given language, choose
the grammar that is simpler in accordance to the optimality conditions stated
above. As is obvious, this concentration on simplicity necessitates a detailed
consideration of the formal properties of the proposed grammars, and therefore
motivates many of the innovations associated with the development of TGG.
For example, as shown in section 5.5, grammars that used transformational rules
were generally perceived to be ‘simpler’ than phrase structure grammars, and
this perception was of crucial importance since in the 1950s it was not known
whether the formal machinery of a phrase structure grammar was actually
insufficient to enable it to generate all the sentences of a language.7 However,
as was claimed at the time, the difference between standard phrase structure
 Constructive nominalist syntax 121

grammars and transformational grammars was that the latter were able to deal
with unbounded dependencies more successfully than the former, while being
able to generate complex sentences, such as active–passive constructions, more
economically, and considerations such as these emphasised the centrality of
simplicity criteria in the earliest TGG research.

4.4 Constructive nominalist syntax

As the preceding section has emphasised, the influence of Goodman is clearly
manifest (and explicitly acknowledged) in Chomsky’s earliest work. However,
Chomsky borrowed more from Goodman than merely some of his ideas con-
cerning simplicity measures for constructional systems. Indeed, the extent of
Chomsky’s preoccupation with Goodman’s work can be gauged most accu-
rately by assessing the contents of his first published paper, ‘Systems of Syn-
tactic Analysis’ (hereafter ‘SSA’), which appeared in 1953.8 Once again, this
paper demonstrates that, at an early stage, Chomsky was concerned with the
non-trivial task of combining various existing techniques in order to develop a
more ‘mechanical’ approach to linguistic analysis.
It was shown above that in both MMH and LSLT Chomsky had sought to
reconcile Harris’ distributional techniques with some of Goodman’s ideas con-
cerning theory-internal simplicity criteria. In ‘SSA’ a different kind of synthesis
was attempted which involved reconciling the methodology of nominalistic con-
structional system theory, as expounded by Goodman and Quine in ‘SCN’, with
the sort of syntactic discovery procedures advocated by Harris. Consequently,
‘SSA’ constitutes an initial attempt to develop a constructional system that can
automatically assign morphemes in a given corpus of utterances to syntactic
categories using distributional information. It is essential to stress that, as usual,
the intellectual origins of this research are freely acknowledged by Chomsky
throughout the paper itself. For instance, having observed the recent trend in
syntactic theory to develop formal methods of analysis that do not require ref-
erence to semantic information, Chomsky outlines his own particular project
as follows:

It is of interest to inquire seriously into the formality of linguistic method

and the adequacy of whatever part of it can be made purely formal, and to
examine the possibilities of applying it . . . to a wider range of problems. In
order to pursue these aims it is first necessary to reconstruct carefully the
set of procedures by which the linguist derives the statements of a linguis-
tic grammar from the behavior of the language users, distinguishing clearly
between formal and experimental in such a way that grammatical notions,
122 Systems of syntax: 1951–1955

appearing as definienda in a constructional system, will be formally derivable

for any language from a fixed sample of linguistic material upon which the
primitives of the system are experimentally defined. (Chomsky 1953: 242)

This passage clearly demonstrates that Chomsky considered ‘SSA’ to be as

much an exercise in applied constructional system theory as a contribution
to linguistics, and it should be noted that the paper appeared in the Journal
of Symbolic Logic rather than in an established linguistics journal, such as
Language or Word. The Formalist emphasis of Chomsky’s paper is clear from
the stated desire to make syntactic analysis (or at least part of it) ‘purely formal’,
and the basic methodology of the paper is indicated in the above quotation:
‘grammatical notions’ are created by means of a process of iterated definition
within a constructional system, thereby associating each created grammatical
object with a definitional genealogy. In the footnotes that accompany the above
passage, Chomsky reveals the sources of his ideas and he explicitly cites Harris’
MSL, Goodman’s SA, and Goodman and Quine’s ‘SCN’ as particular influences.
In addition, Chomsky thanks the three aforementioned individuals personally
for their offered suggestions and criticisms, indicating that he had discussed
his research with them before publishing his paper. The precise nature of the
relationship between the proposed method of syntactic analysis and the type of
constructive nominalism adopted in ‘SCN’ is further clarified when Chomsky
explains why he considers an ‘inscriptional nominalistic framework’ to be of
value when performing syntactic analysis:

The inscriptional approach seems natural for linguistics, particularly in view of

the fact that an adequate extension of the results of this paper will have to deal
with the problem of homonymity . . . It will appear below that the calculus of
individuals can often supply quite simple solutions to constructional problems
that seem on the surface to require a set-theoretic solution, thus removing
all necessity for an involved hierarchy of types and increasing the overall
workability of the system. (Chomsky 1953: 243)

This passage suggests that the main advantage of a constructive nominalist

approach to syntactic analysis is that it obviates the need for set-theoretical
assumptions, and consequently avoids the sort of type hierarchies (i.e., artifi-
cial approaches such as Whitehead and Russell’s theory of logical types) which
had fallen into disrepute in the realm of mathematical logic. It is no coincidence,
of course, that these kinds of set-theoretical constructs are precisely those that
had been rejected by Goodman and Quine in ‘SCN’, since, as discussed in
section 3.6, the need to reject such entities largely motivated their whole con-
structive nominalist project. The influence of Goodman in particular is revealed
 Constructive nominalist syntax 123

in the passage quoted above, since the ‘calculus of individuals’ mentioned there
is that developed by Goodman and Leonard during the mid-1930s, and Chom-
sky makes direct use of this system throughout his own paper, employing the
particular version of the calculus that appeared in Goodman’s SA. The implica-
tions of using this type of constructional system for the purposes of syntactic
analysis are many. In particular, the ‘inscriptional’ emphasis determines that
the sentences in a language must be manipulated as finite strings of symbols,
as in Formalist mathematics. Also, the rejection of such abstract entities as sets
ensures that all larger structures must be exhaustively analysed in terms of their
constituent elements.
Given the above, and given the freely confessed nominalist agenda of ‘SSA’,
Chomsky’s claim that a constructional approach to syntactic analysis could
help to resolve certain fundamental problems associated with the phenomenon
of homonymity is intriguing, if somewhat premature, since (as he himself
acknowledges) such problems do not manifest themselves in the simple for-
mal languages discussed in the paper, and (so far) he has never explored these
issues in subsequent work.9 Indeed, it is important to emphasise that through-
out ‘SSA’ Chomsky is solely concerned with formal languages, and openly
declares that ‘the present system as given here is not adequate for the analysis
of natural languages’ (Chomsky 1953: 243). The particular language he uses in
order to illustrate his theoretical points consists merely of ‘utterances’ such as
‘ab’ and ‘axd’, where each alphabetic character represents a morpheme in the
language. Obviously, this focus on such simple languages is in marked contrast
to Chomsky’s later work, which was predominantly concerned with problems
of natural language analysis; an issue that is further discussed in section 4.5.
As indicated above, in the details of its structure, the type of nominalistic con-
structional system that Chomsky develops in sections 2 and 3 of ‘SSA’ is merely
an implementation of the system presented by Goodman in SA, with the majority
of the definitions, axioms, and theorems presented being borrowed unaltered.
The paper’s main contribution, in fact, is simply the attempt to apply construc-
tional system theory to the task of syntactic analysis. Accordingly, following
Goodman, the core of the system outlined in the paper contains a logical basis
(i.e., standard symbolic logic of the PM variety) and an extralogical basis con-
sisting of primitive elements (i.e., inscriptions which are the basic morphemes)
and five primitive relations. It should be remembered that (as mentioned above)
the primitive elements are all considered to be ‘experimentally defined’, thus
revealing the strong empirical nature of the proposed analytic method.10 The
primitive relations include such predicates as ‘O’, where ‘a O b’ means a ‘over-
laps’ b (i.e., there exists an inscription that is part of both a and b), and these
124 Systems of syntax: 1951–1955

relations are all taken, unmodified, from SA, as Chomsky readily admits. He
then proceeds to construct a set of ‘elementary notions’ such as ‘SEG’, where
‘a SEG b’ means ‘a is a segment of b’, and is defined in terms of the logical
and extralogical bases as
a SEG b = ∀x[x O a → x O b] (4.5)
which indicates that a is a segment of b if, for all x, x overlaps both a and
b. Once again these definitions are Goodman’s (indeed, (4.5) is identical with
Goodman’s definition of his ‘Part’ relation, given as (3.15) above), and they
enable further axioms and theorems to be constructed. Finally, in section 3 the
manner in which Goodman’s constructional system can be utilised for the task
of syntactic analysis is outlined. The basic methodology is to construct ‘an
indefinite series of similarity relations’ of the form ‘Sn ’ (Chomsky 1953: 249),
where the subscript n ranges over numerals and indicates the length of the
morpheme sequence. These similarity relations can be used to group the mor-
phemes encountered in a given corpus into syntactic classes. To illustrate how
the process works, Chomsky uses a corpus of six sentences, ab, cb, de, f e,
axd, and cy f , and he shows that the morphemes a and c would be related by
the similarity relation S1 , as would the morphemes d and f , while a..d and
c.. f would be related by S2 , with x and y being related by S3 .
As the above summary indicates, the general methodology offered in ‘SSA’
suggests that the nominalistic constructional system presented by Goodman in
SA could be used in the context of syntactic theory as a discovery procedure
that would reduce the task of syntactic class assignment to an automatic pro-
cess, thus illustrating the potential correspondence between Harris’ discovery
procedures and constructional system theory. As mentioned in the introduc-
tion, though, Chomsky’s 1953 paper has generally been neglected by linguistic
historiographers, and, given the above summary, it is not difficult to see why
it has been forgotten (or perhaps intentionally ignored?). When viewed in a
cursory manner, the paper appears to be only tenuously connected to the type
of syntactic theory Chomsky was soon to develop in LSLT and which would
eventually become known as TGG. For a start, as mentioned above, ‘SAA’ is
solely concerned with simple formal languages, and, as shown in the next sec-
tion, Chomsky was soon to question the validity of arguments involving such
languages when used in the context of linguistic theory (as opposed to pure log-
ical research). Also, drawing as it does upon Goodman’s work in constructional
system theory, ‘SA’ appears to have a strong empiricist bias (though Hiorth, for
one, has doubted this (Hiorth 1974: 35)), since it focuses upon the automatic
assignment of morphemes in a specific corpus using distributional information.
 Logic and linguistic theory 125

Nevertheless, despite these clear differences, various aspects of the approach to

syntactic analysis proposed in ‘SSA’ were to manifest themselves in the mature
theory of TGG, although often in subtly different guises. For instance, as dis-
cussed in section 5.4, the machinery of constructional system theory was not
entirely discarded after 1953; rather, parts of it were included in the definition
of a linguistic level expounded in LSLT. Similarly, Chomsky’s preoccupation
with mechanical axiomatic-deductive analytic methods did not wane during the
mid-1950s; rather, it remained constant, eventually manifesting itself in terms
of conversion-based deductions in mature TGG. However, this is not to suggest
that significant aspects of the stance adopted in ‘SSA’ were not rejected later.
Indeed, it is clear that this was the case. For instance, ‘SSA’ clearly constitutes
yet another attempt to automate Harris-style discovery procedures, and there is
no doubt that by the mid-1950s Chomsky had started to question the validity
of such procedures, favouring instead evaluation procedures, as discussed in
section 5.3. The purpose of the above discussion, therefore, has been to argue
neither for complete uniformity nor for complete disseverance, but rather to
illustrate the full complexity of the situation. It is crucial, though, to emphasise
that ‘SSA’ is not merely of marginal interest to the history of TGG, and that it is
completely mistaken to view it as a bizarre, isolated, and self-evidently imma-
ture foray into the realm of empiricist philosophy. Rather, the paper is absolutely
central to the genesis of TGG, since it reveals the course of Chomsky’s intellec-
tual development, and so provides an early insight into several analytical tech-
niques and preoccupations that would later become standard features of TGG.

4.5 Logic and linguistic theory

The previous sections of this chapter have demonstrated that, from the very
beginning, Chomsky was involved in the task of combining the methodology
of structural linguistics with techniques derived from constructional system the-
ory, and his emphasis on the role of simplicity criteria in the creation of formal
grammars, along with his attempt to adapt Goodman’s ‘calculus of individuals’
for the purposes of syntactic analysis, reveals the extent of his preoccupation
with these issues. However, it is not the case that Chomsky recommended this
kind of interdisciplinary approach indiscriminately, and the complexity of his
views can best be explored by considering his attitude towards the relation-
ship between linguistics and logic. His most explicit statements concerning this
topic are contained in a 1955 paper that was written in response to an article by
Bar-Hillel in which the latter had advocated the further integration of linguistics
and logic. It is not remarkable that such a subject should have been debated in a
126 Systems of syntax: 1951–1955

leading linguistics journal of the time for, as chapter 3 and the earlier sections of
this chapter have demonstrated, during the 1940s and 1950s the precise nature
of the relationship between logic and natural language was being reassessed
by various logicians and linguists, and certainly Bar-Hillel and Chomsky were
both active participants in this process of reassessment. Their exchange in the
mid-1950s began with the publication of Bar-Hillel’s paper ‘Logical Syntax
and Semantics’ (1954) in Language and continued when Chomsky published
‘Logical Syntax and Semantics: Their Linguistic Relevance’ (1955) (hereafter
‘LSS’) in the same journal. While considering the various arguments put for-
ward by Bar-Hillel, it is essential to remember that, as discussed in section 3.4,
he had recently revived the work of Ajdukiewicz, and had suggested that tech-
niques adapted from logical syntax and logical semantics could be used to
analyse natural language, thus laying the foundations for the function-based
syntactic theory now known as Categorial Grammar. Clearly, therefore, Bar-
Hillel was fully persuaded that logic could provide valuable insights into the
nature of linguistic structure and, though he later confessed that these ideas
were ‘deplorably naı̈ve’ (Bar-Hillel 1964: 3), they certainly inspired his 1954
paper, which must now be considered at length.
In general terms, Bar-Hillel’s paper constitutes a ‘plea for the reintroduction
of semantics into the theatre of operations of descriptive linguistics’ (Bar-Hillel
1954: 235–236), and his primary motivation seems to have been a profound dis-
satisfaction with the type of analytical methodology advocated by Harris in his
MSL. Revealingly, in order to articulate his frustration with distributional proce-
dures more precisely, Bar-Hillel quotes a lengthy passage from the 1937 English
translation of Carnap’s LSL, a book with which he had become obsessed.11 The
quoted passage is taken from the general introduction in which Carnap consid-
ers the relationship between syntax and logic, and in which he distinguishes
between formation and transformation rules. Carnap observes that ‘the prevalent
opinion’ (i.e., in 1934) was that syntax and logic are ‘fundamentally theories of
a very different type’, yet he goes on to challenge this presupposition by assert-
ing that both disciplines are ‘equally concerned with the relation of meaning
between sentences’, which in turn leads him to prophesy that ‘logic will become
part of syntax’ (Carnap 1937[1934]: 1–2). By means of this conviction, Car-
nap is able to speak coherently of ‘logical syntax’, a hybrid term that gestures
towards the interrelation of the two disciplines.
Having presented Carnap’s views in this direct manner, Bar-Hillel then con-
siders the way in which these ideas had been (and could be) developed by
linguists. He emphasises the fact that formation rules (in the Carnapian sense)
had been explored in some detail, citing Harris’ MSL and Fries’ The Structure
 Logic and linguistic theory 127

of English (1952) as key examples of works in the broad tradition of distri-

butional structural linguistics (just as he had cited them in his 1953 paper in
which he had revived Ajdukiewicz’s work), but he bewails the corresponding
neglect of transformation rules (again in the Carnapian sense).12 In other words,
in Bar-Hillel’s view, structuralist linguistic theories are solely concerned with
constructional grammatical processes that enable larger linguistic complexes to
be analysed in terms of smaller constituent elements (e.g., phonemes or mor-
phemes), and such theories pay insufficient regard to the logical relations that
exist between linguistic expressions.13 Bar-Hillel attributes this neglect partly
to the fact that logicians (such as Carnap) devote their time primarily to the
task of creating and analysing artificial formal languages that are much simpler
than natural languages. Indeed, as shown in section 3.7, Carnap’s pessimism
concerning the possibility of analysing natural languages in detail using the for-
mation and transformation rules that he developed was clearly stated in LSL and
certainly Bar-Hillel was aware of Carnap’s opinion on this matter (see Bar-Hillel
1954: 231). However, Bar-Hillel points out that linguists, not logicians, are the
people that concern themselves most directly with natural language, and that,
therefore, the burden of responsibility for exploring the applicability of transfor-
mation rules in this context falls upon them. Indeed, he claims that (in his view)
Carnap’s ideas might be ‘linguistically sound’, but, since Carnap himself is not
‘a linguist proper’, he concludes that it behoves practising linguists carefully to
scrutinise and assess the implications of these ideas (Bar-Hillel 1954: 231).
Having identified a lack of interest in Carnapian transformation rules on the
part of linguists, Bar-Hillel then considers the sort of rules that are neglected
in standard structuralist distributional analyses, but which could prove to be of
value to linguistics, and, as an example, he discusses the problem of synonymy.
The starting point for his discussion is Harris’ claim in MSL that distributional
procedures can provide a cogent analysis of linguistic meaning. In particular,
Harris had stated that

distributional procedures, once established, permit, with no extra trouble, the

definite treatment of those marginal cases which meaning considerations leave
indeterminate or open to conflicting opinions. (Harris 1951: 8.n7)

and Bar-Hillel interpreted this to mean that linguistic units associated with
different distributions inevitably have different meanings. Consequently, he
concludes that, for Harris, units with identical distributions are considered to
be synonymous. This is admittedly a rather strong interpretation of Harris’
statement, but the inference is not entirely unreasonable. However, Bar-Hillel
rejects the idea that synonymy can be fully analysed in terms of distribution,
128 Systems of syntax: 1951–1955

arguing that, while a distributional analysis might be able show that (i) ‘oculist’,
(ii) ‘eye-doctor’, and (iii) ‘dentist’ can all appear in the same syntactic positions,
it can never indicate that (i) and (ii) have identical meanings, while this is
not the case either for (i) and (iii) or (ii) and (iii). This failure, according to
Bar-Hillel, is caused by the neglect of the truth-conditions that would provide
information about the meaning of the words. Consequently, since distributional
procedures ignore such semantic considerations, they provide an incomplete
basis for linguistic analysis.
Bar-Hillel’s conviction that semantic considerations can be incorporated into
syntactic analyses without the latter succumbing to ‘an infestation by meaning’
(Bar-Hillel 1954: 234) clearly has its roots in his well-attested interest in the
work of Carnap and the Lvov-Warsaw school of logicians. As discussed in sec-
tion 3.7, Carnap’s research into logical syntax had been influenced by Hilbert’s
proof theory and consequently, he had defined a number of formal languages
that would enable him to create logical systems that were free from semantic
considerations. However, in turn (and rather ironically), this work caused Car-
nap to reflect upon the relationship between formal linguistic systems of the
type presented in the LSL and the external world. In other words, his interest in
formal syntactic systems that attempted to avoid semantic considerations incul-
cated a profound interest in semantics, and for the next twenty years Carnap
focused his research upon this broad topic. In 1942 he published his Intro-
duction to Semantics while, in the following year, his Formalization of Logic
appeared, and both books explored various aspects of semantics in relation
to formal systems. The main emphasis in both texts is upon truth-conditional
interpretations. In other words, given a statement in a formal language, in order
to explore its meaning, Carnap argued that it was first necessary to know how
the universe would have to be configured in order for the statement to be true.
In Carnap’s treatment, the rules for semantic interpretation constitute a separate
and isolated part of the full linguistic system he develops: the syntactic com-
ponent outputs sentences in a formal language that are subsequently subjected
to semantic interpretation. In order to clarify his position, Carnap offers the
following definition of a semantic system:

A semantical system is a system of rules which states truth-conditions for

the sentences of an object language and thereby determine [sic] the meaning
of these sentences. A semantical system S may consist of rules of formation
defining ‘sentence in S’, rules of designation defining ‘designation in S’, and
rules of truth defining ‘truth in S’. The sentence in the metalanguage ‘Gi is true
in S’ means the same as the sentence Gi itself. This characteristic constitutes
a condition for the adequacy of definitions of truth. (Carnap 1942: 22)
 Logic and linguistic theory 129

As this passage indicates, Carnap’s basic approach to logical semantics was

truth-conditional, and his thinking in this respect had been influenced by Tarski,
one of the members of the Lvov-Warsaw school. Appropriately, Carnap explic-
itly acknowledged his debt to his younger contemporary.14 Tarski had studied
with Ajdukiewicz, and since the latter considered syntax and semantics to be
closely related, it is not surprising that Tarski should have been convinced that
formal languages required a system of semantic interpretation in order to be
complete. In a 1931 paper,15 Tarski emphasised the relationship between truth
and meaning, and, while seeking an adequate definition of truth, he introduced
the schema, (T ), which takes the form

(T ) X is true if, and only if, p (4.6)

The basic idea behind this schema is that any sentence will be classified as an
‘equivalence’ of the form (T ) if the letter ‘ p’ above can be replaced by a sen-
tence, and the letter ‘X ’ by the name of the sentence. Once again, the Hilbertian
notion of a metalanguage is dominant here since there is a clear distinction
between metalanguage and object language. The most famous example of a
sentence-name pair that is equivalent to (T ) is the sentence ‘Snow is white if
and only if snow is white’, and it is the concept of a metalanguage alone that
rescues this statement from tautology: ‘X ’ (i.e., ‘snow is white’) is true if and
only if it is actually the case that p holds (i.e., snow is white). In a paper writ-
ten several years later,16 Tarski again emphasised the relationship that exists
between statements in a formal language and the external world:

We shall understand by semantics the totality of considerations concerning

those concepts which, roughly speaking, express certain connexions between
the expressions of a language and the objects and states of affairs referred to
by these expressions. (Tarski 1956b[1936]: 401)

Consequently, his understanding of meaning was based upon his understanding

of the extension of a predicate expression. Essentially, the extension of such an
expression constitutes the class denoted by the expression, while the intension
of the expression constitutes the property (or properties) that distinguishes the
members of the class denoted by the expression. As he expressed it in a later
paper, ‘A sentence is true if it designates an existing state of affairs’ (Tarski
1944: 343).
As mentioned in section 3.6, during the early 1950s some of these develop-
ments in the theory of logical semantics caused various philosophers to explore
aspects of the traditional (Kantian) distinction between analytic and synthetic
truths. For instance, in 1952 Carnap published a paper in which he attempted to
130 Systems of syntax: 1951–1955

provide a new ‘way of explicating the concept of analyticity’ (Carnap 1952: 66).
The basic approach outlined in the paper involved sentences of the type ‘If Jack
is a bachelor, then he is not married’ being formally represented using meaning
postulates of the type

B(a) → ¬M(a) (4.7)

where, in this case, ‘B’ and ‘M’ indicate ‘is a bachelor’ and ‘is married’ respec-
tively. Since this implication is necessarily valid, Carnap suggested that the
following meaning postulate could be adopted

P1 : ∀x[B(x) → ¬M(x)] (4.8)

and the rest of the paper considered the ways in which adopting such postulates
can simplify the task of semantic analysis. As already implied above, this revival
of logical semantics in the 1930s and 1940s, which was largely engineered by
Ajdukiewicz, Tarski, and Carnap, can be viewed as a reaction against the more
extreme varieties of Formalism that attempted to exclude meaning entirely from
logical syntax.
To return to Bar-Hillel’s paper in the light of this summary, it is clear that
he was entirely familiar with the developments sketched above. For instance,
he refers to ‘the Warsaw-Lvov school’ explicitly, and mentions Ajdukiewicz
and Tarski by name. In addition, while considering these topics, he points to
similarities between the work of Bloomfield and Carnap concerning the role
of meaning. Since the relevant passage appears not to be well known, it merits
being quoted in full.

It is an interesting fact, deserving the attention of sociologists of science, that

at approximately the same time, but in complete independence of each other,
Bloomfield and Carnap were fighting the psychologism that dominated their
respective fields, linguistics and logic. They both deplored the mentalistic
mud into which the study of meaning had fallen, and tried to reconstruct
their fields on a purely formal-structural basis. I think it is correct to say
that the difference between the structural linguist and the formal logician is
one of stress and degree rather than of kind. Both are essentially attempting
to construct language systems that stand in some correspondence to natural
languages – though most linguists would say that they are just describing
the latter. But whereas for the linguist the closeness of the correspondence
is the criterion by which he will judge the adequacy of the language system
he is setting up, which alone entitles him to consider himself as describing
a given natural language, the logician will look primarily for other features
of his system, such as simplicity of handling, fruitfulness for science, and
ease of deduction and computation, with close correspondence to a natural
 Logic and linguistic theory 131

language as only a secondary desideratum. Constructed language systems are

judged by the linguist according to the degree to which they approximate
a natural language; natural languages are judged by the logician according
to the degree to which they approximate efficient, well-constructed language
systems. (Bar-Hillel 1954: 234–235)

There are several issues here that merit comment. For instance, the parallel
between Bloomfield and Carnap is yet another instance of the association that
was felt to hold between logic and linguistics in the 1950s, and this helps to
explain why certain syntacticians borrowed techniques from formal logic (and
related disciplines) and used them to facilitate (or so they hoped) the analysis
of natural language. Bar-Hillel’s own advocation of logical syntax and recur-
sive definitions has already been considered, as has Chomsky’s adaptation of
constructional system theory. As Bar-Hillel acknowledges, one of the reasons
for this perceived closeness was the rejection of meaning that had characterised
Bloomfield’s and Carnap’s early work – an association that has already been
mentioned several times in earlier chapters (especially section 3.7). In the con-
text of this observation, it should be noted, though, that Bar-Hillel’s claim that
Bloomfield and Carnap worked ‘in complete independence’ is not entirely true.
As mentioned in section 3.7, Bloomfield had read LSL and in 1939 he wrote
an article for an Encyclopedia of which Carnap was one of the editors, though
admittedly it is difficult to gauge the full extent of their contact during this
period. Whatever the precise nature of their interaction, though, there is no
doubt that Bloomfield’s work seemed to make certain assumptions that were
similar to those made by Carnap in his pre-1935 publications, and, as noted
previously, to a later generation of linguists this apparent similarity appeared to
provide a basis for future developments. At the very least, it seemed to licence
a free borrowing of the techniques of formal logic for the task of linguistic
analysis.
While the picture painted by Bar-Hillel gives the impression of being clear
and comprehensive, it is (not surprisingly) rather too simplistic as an account
of the real situation. For instance, not all linguists shared his perception of this
close correspondence. As already indicated, Harris, for one, was convinced
that logicians and linguists did different things, and he was not forced into this
conclusion by ignorance of recent developments, for his familiarity with LSL
(at least) is well attested by the various references to Carnap’s book in MSL (as
discussed in section 3.7). Indeed, the difference between Harris’ and Bar-Hillel’s
views concerning logic and natural language could not be more striking, since,
according to the latter, linguists and logicians differ in ‘degree’ rather than in
‘kind’, while the former had stated that ‘linguists meet the problem [i.e., that of
132 Systems of syntax: 1951–1955

describing the structure of language] differently than do Carnap and his school’
(Harris 1952: 16.n17). The main difference between linguists and logicians
from Bar-Hillel’s perspective was that the latter were often more concerned
with the form of the systems they created than with the task of validating said
systems by means of empirical investigation. By contrast, linguists were forced
to compare the results of their theoretical investigations with utterances in actual
corpora. Therefore, issues such as ‘simplicity of handling’ (whatever that might
be exactly) and ‘ease of deduction and computation’ are primarily a concern for
logicians. However, as was shown in section 4.3, the notion of theory-internal
simplicity had already begun to infiltrate Chomsky’s conception of syntactic
theory largely due to his interest in Goodman’s work. Also, around this time,
Chomsky was starting to reconsider the usefulness of corpus-based discovery
procedures, and would eventually introduce a different kind of research method
that was not so deeply rooted in empirical validation. In both these respects, then,
Bar-Hillel’s description of linguistic investigation was soon to become outdated.
Nevertheless, the conception of linguistics and logic that he develops in his paper
is fundamental to Bar-Hillel’s view of their potential inter-connectivity.
As indicated above, Chomsky’s response to Bar-Hillel’s paper is of consider-
able interest for various reasons. His primary intention is to refute Bar-Hillel’s
suggestion that advances in the theory of logical syntax and semantics can
facilitate the analysis of meaning in natural language, a task of real significance
since, by the mid-1950s, Chomsky was fully persuaded that considerations of
meaning had no place in the study of syntax. For instance, to take just one
quotation from his other 1955 paper, ‘Semantic Considerations in Grammar’,
which also considers the role of meaning in linguistic theory, Chomsky had
plainly stated (in true Bloomfieldian fashion) that

Meaning is a notoriously difficult notion to pin down. If it can be shown that

meaning and related notions do play a central role in linguistic analysis, then
its results and conclusions become subject to all the doubts and obscurities that
plague the study of meaning, and a serious blow is struck at the foundations
of linguistic analysis. (Chomsky 1955b: 141)

The implication here is clear: meaning is a pernicious, troublesome aspect of

natural language that should be excluded (if possible) from linguistic analysis. In
the light of this observation, and before considering in some detail the argument
Chomsky constructed to refute Bar-Hillel, it is worth outlining the sources upon
which he drew. As ever, this task is fairly easy to accomplish due to Chomsky’s
willingness to specify both the nature and extent of his indebtedness. In this
case, he openly confesses that, in preparing his response, he has ‘borrowed freely
 Logic and linguistic theory 133

from various critical accounts of the theory of meaning’ (Chomsky 1955a: 36).
More specifically, he explicitly names Quine’s LPV (especially chapters 2, 7,
and 8), as well as White’s paper ‘The Analytic and the Synthetic: An Untenable
Dualism’, which first appeared in 1950 (while Chomsky was still attending
White’s lectures?), but which was later published in the collection Semantics and
the Philosophy of Language edited by Leonard Linsky. The appearance of this
book in 1952 seems to have been partly responsible for provoking Bar-Hillel’s
and Chomsky’s exchange in the mid-1950s, since it is referred to in both papers
and raises many questions about the status of meaning in philosophical theories
of language. In particular, Quine’s and White’s attacks upon the traditional
Kantian distinction between analytic and synthetic truths, which were briefly
summarised in section 3.6, appear to have persuaded Chomsky, and his use of
their ideas in his own arguments implies his acceptance of their views. In this
context, the details of his argument will now be considered below.
Appropriately enough, Chomsky begins his paper by summarising Bar-
Hillel’s main points and immediately attacks the idea that Carnapian transfor-
mations can be profitably used by linguists in order to analyse natural language.
His central argument is that, since Carnap assumes such relations as ‘formal
consequence’ and ‘synonymy’ as primitives in the logical systems he creates,
his work offers no means of explaining or clarifying these notions, and therefore
the use of such relations cannot benefit linguistics except in a ‘trivial’ fashion. If
such relations were adopted in a linguistic theory, Chomsky claims, the valid-
ity of the resulting inferences could only be assessed by listing all possible
options as postulates of the given language system, an argument that Chomsky
had borrowed directly from White’s paper since, as mentioned in section 3.6,
White had argued that a ‘rule-book’ (White 1952: 277) of synonyms would be
required in order to deal with analytic truths in natural language, and (of course)
no such book exists. With considerable scorn, then, Chomsky declares that this
constitutes nothing more than ‘an ad-hoc approach to the problem of classi-
fication and characterisation of elements in particular languages’ (Chomsky
1955a: 38), and (in his opinion) the apparent failure of Carnap’s transforma-
tion rules to contribute anything of substance to the task of analysing natural
language phenomena such as synonymy entirely undermines Bar-Hillel’s trum-
peted belief in the efficaciousness of such rules in a linguistic context.
Having proceeded thus far in his demolition of Bar-Hillel’s argument, Chom-
sky pauses to consider the term ‘formal’, which ‘has played a rather crucial
role in this discussion’ (Chomsky 1955a: 39). In particular, Chomsky is keen
to reconsider the consequences of using such terminology while discussing
linguistic systems. For instance, Bar-Hillel had claimed in his paper that the
134 Systems of syntax: 1951–1955

active–passive relation that converts ‘plays’ into ‘is played by’ is a relation
of formal consequence. Chomsky does not accept this, since the meaning of
the term ‘formal’ has ‘misleading connotations’ in this context. Consequently,
he offers a (somewhat opaque) definition of ‘formal’ in which a relation is so
classified if ‘it holds between linguistic expressions’ (Chomsky 1955a: 39),
and while agreeing with Bar-Hillel’s statement concerning the active–passive
relation, he adds that the relation ‘longer by three words’ is also formal (in
the above sense), since it too holds for the expression pair ‘John did not come
home’ and ‘John came’. Since logical syntax of the Carnapian variety is unable
to determine which of these examples is an instance of formal consequence,
Chomsky is forced to conclude that relations of formal consequence are of no
use when analysing a natural language. Rather, the ‘systematic investigation
of linguistic expressions alone’ (Chomsky 1955a: 39) is required in order to
specify the sets of expressions for which such relations hold. Presumably, in
this context, the phrase ‘systematic investigation’ denotes some kind of Harris-
style distributional procedure that can be used to group expressions in a given
language, or possibly the type of constructional system-based approach devel-
oped in ‘SSA’. It is crucial to recognise, though, that such a procedure would
still be ‘formal’, since no semantic considerations would be permitted in such
a scheme. Chomsky concludes this section of his discussion with the unam-
biguous declaration that ‘logical syntax and semantics provide no grounds for
determining synonyms and consequence relations’ (Chomsky 1955a: 39).
The various arguments (summarised above) that Chomsky marshals in the
first part of his paper are partly designed to undermine Bar-Hillel’s claim that
linguistics could benefit from those aspects of Carnap’s work in logical systems
that deal with logical syntax. However, Chomsky is not convinced that logical
semantics has much to offer either, and, by rejecting this possibility, he reveals
the full extent of his dissatisfaction with contemporaneous accounts of linguistic
meaning. The causes of his dissatisfaction are established when he goes on
to assess the difference between logical implication and similar phenomena
in natural language. He considers several examples (including one borrowed
from Goodman) and easily demonstrates that natural language inferences are
unexpectedly complex, and therefore cannot be glibly analysed in terms of the
standard logical implication operator. Once again this provides him with an
ideal opportunity to stress the divide that separates natural language and the
type of artificial languages that Carnap creates in his ‘logical laboratory’.

The question as to the nature of inference in natural languages can scarcely

be intelligibly put now, since we have almost no systematic knowledge about
 Logic and linguistic theory 135

inference or meaning in ordinary linguistic behavior, and no study of new and

deeper foundations for mathematics can be expected to tell us more about this.
(Chomsky 1955a: 40)

The advice here (directed towards Bar-Hillel and like-minded logician-linguists,

no doubt) is to resist the temptation unthinkingly to appropriate techniques
developed in the course of research into the foundations of mathematics, since
it does not follow that they will necessarily facilitate the analysis of natural
language. As is shown below, towards the end of the paper Chomsky does not
exclude the possibility that such techniques could prove to be useful tools in the
context of linguistic research; his point is simply that the utility of such methods
is not guaranteed. Nevertheless, it is fascinating to juxtapose statements of this
kind with the sort of optimism encountered in ‘SSA’, written just two years
before. As mentioned in section 4.4, in his 1953 paper, Chomsky actively used
Goodman’s ‘calculus of individuals’ (a logico-philosophical system) to analyse
linguistic structures. More to the point, the language he analysed in that paper
was a simple formal language, and he openly stated that the methodology
presented could not be used to analyse natural language. Clearly, sometime
during the years 1953–1955, Chomsky had changed his mind concerning the
feasibility, and perhaps the validity, of the type of project attempted in ‘SSA’.
It is important to note, therefore, in the light of his exchange with Bar-Hillel,
that Chomsky was not discussing these matters solely from the perspective
of abstract linguistic theory, since, in addition to his pronounced theoretical
concerns, he had considerable practical experience of working both with logical
systems and with artificial languages.
Not content, however, with having undermined Bar-Hillel’s statements con-
cerning the usefulness of logical syntax and semantics, Chomsky also questions
Bar-Hillel’s understanding of these topics, and he does this by focusing on the
theory of meaning. After asserting that Tarski’s main contributions were to
the theory of reference and not to the theory of meaning (Bar-Hillel had claimed
the latter), Chomsky considers the recent work of Quine and concludes that, far
from extending Carnap’s notion of intension (as Bar-Hillel had argued), Quine
had actively sought to demonstrate the parlous state of meaning in logical sys-
tems theory. In particular, Chomsky cites Quine’s celebrated paper ‘Two Dog-
mas of Empiricism’, referring to it as ‘Quine’s most trenchant attack on current
formulations of the theory of meaning’ (Chomsky 1955a: 41), and he goes on to
suggest that, for Quine, this theory ‘remains in pretty much the state that repelled
Bloomfield’ (Chomsky 1955a: 40). In addition to Quine’s, White’s influence
can also be detected here, since his discussion of analytic and synthetic truths
136 Systems of syntax: 1951–1955

(cited explicitly by Chomsky at the start of his paper) also served to destabilise
logical semantics. It is apparent, therefore, that Chomsky’s main intention in this
passage is twofold. First, he is keen to correct Bar-Hillel’s alleged misreading
of recent developments in logical semantics. However, second, he is also eager
to demonstrate that none of the recent advances cited by Bar-Hillel had enabled
meaning to be considered as a viable foundation for linguistic theory. Conse-
quently, as far as Chomsky is concerned, the post-Bloomfieldian emphasis on
the manipulation of meaning-less linguistic forms still provides the most secure
basis for syntactic theory, and, although this conclusion is not new, Chomsky’s
argument certainly demonstrates the extent to which his thinking on this issue
had been influenced by contemporaneous discussions in the field of analytic
philosophy, an influence that has not been adequately recognised in the past.
With his basic position now established, Chomsky seeks to render it more
secure by dismissing Carnap’s suggestion (endorsed by Bar-Hillel) that the
study of artificial languages can provide insights into the nature of natural lan-
guage. Consequently, Bar-Hillel’s assertion that logicians and linguists differ
only in ‘degree’ is shown to be mistaken: if one draws conclusions about nat-
ural language from logical systems constructed to explore the foundations of
mathematics ‘one might as well argue that a science-fiction writer or an artist is
doing roughly the same thing as a physicist’ (Chomsky 1955a: 42). The anal-
ogy is polemical, but Chomsky is able to conclude that ‘artificial languages
are neither special cases nor idealised versions of natural language’ (Chomsky
1955a: 42), and yet again the purpose here is to convey the impression that a
significant divide separates the disciplines of logic and linguistics, an attitude
that has already been identified in the work of Harris, suggesting that Chom-
sky was here dutifully following Harris’ lead. In general, Chomsky’s aim is to
acknowledge and emphasise the independence of linguistics as an intellectual
discipline. If the association between it and logic were accepted too rapidly
and too extremely, the former would simply be annexed to the latter, with (pre-
sumably) dire consequences. Once again, though, it is worth emphasising that
Chomsky himself had attempted to develop a system of linguistic analysis in
‘SSA’ that had concentrated on simple artificial languages, and the above pro-
nouncements must have been influenced to some extent by his own personal
experience of this type of approach.
In characteristic fashion, having arrived at the strident position outlined in
the above paragraph, Chomsky then immediately proceeds to reveal the full
complexity of his views. Although he remains adamant that ‘logical syntax
and semantics can bring the linguist no nearer to an adequate conception of
synonymy and transformations’ (Chomsky 1955a: 41), he does not rule out
 Logic and linguistic theory 137

the possibility that logic can be fruitfully employed in a linguistic context, and
he specifically mentions Bar-Hillel’s work involving recursive definitions as a
recent positive example. Aware that such pronouncements could seem para-
doxical when juxtaposed with the previously articulated scepticism, Chomsky
seeks to clarify his position:

The correct way to use the insights and techniques of logic is in formulating
a general theory of linguistic structure. But this does not tell us what sort of
systems form the subject matter for linguistics, or how the linguist may find it
profitable to describe them. To apply logic in constructing a clear and rigorous
linguistic theory is different from expecting logic or any other formal system
to be a model for linguistic behavior. (Chomsky 1955a: 45)

This single passage reveals the full intricacy of Chomsky’s attitude towards the
use of techniques derived from formal symbolic logic in linguistic analyses,
and, despite some of the more extreme comments quoted above, it emerges
that he is neither simply for, nor simply against, the use of logic in linguistic
research. Rather, he is opposed to the unthinking assumption that logic will
necessarily provide insights into the structure of natural language. In the event,
some techniques derived from logic may prove to be useful when a linguistic
theory is being constructed, while others may offer no benefits. The task of
the linguist is sensibly to assess the validity of the various techniques in the
context of natural language analysis. As indicated above, Chomsky is seem-
ingly convinced that the methods of logic can certainly be usefully employed
when ‘a general theory of linguistic structure’ is developed, and, as with his
remarks concerning the use of artificial languages in linguistic research, his
comments here are not abstract musings since, during early 1955, Chomsky
was busy completing the first draft of LSLT, the main text in which he outlined
his own general theory of linguistic structure. The full title of this work alone
emphasises the point made at length above: The Logical Structure of Linguistic
Theory, with its conscious nod towards (the English translation of) Carnap’s
The Logical Structure of Language,17 implies that the structure of linguistic
theory must be logical. However, it does not follow from this that the task of
analysing natural language can be reduced in a trivial fashion to an exercise in
logical manipulation. In choosing his title, therefore, Chomsky seems simul-
taneously to be aligning himself with, and distancing himself from, Carnap’s
work. Consequently, it is appropriate that many of the insights into the nature
of the relationship between logic and linguistics, which can be gleaned from
his 1955 paper, can be considered in relation to various aspects of LSLT, some
of which will be explored in detail in chapter 5.
138 Systems of syntax: 1951–1955

As a coda to the above discussion, it is worth mentioning that, eight years

after Chomsky’s paper appeared, some of the issues raised in the exchange were
reconsidered by Bar-Hillel and (even more intriguingly) by Carnap himself.
Bar-Hillel contributed an essay entitled ‘Remarks on Carnap’s Logical Syntax
of Language’ to the 1963 volume of Schilpp’s Library of Living Philosophers,
which was devoted to an assessment of Carnap’s work. This paper shows that
Bar-Hillel’s basic views had changed little over the intervening years, though he
did acknowledge that Chomsky’s 1955 defence of Harris had been ‘well taken’
(Bar-Hillel 1963: 542). Responding to Bar-Hillel’s paper, Carnap himself con-
siders the relationship between linguistics and logic and declares that he has
‘full sympathy’ with the suggestion that the former could profit from the tech-
niques borrowed from the latter, although he does advise caution when applying
them. With specific reference to the Bar-Hillel versus Chomsky exchange in
the mid-1950s, Carnap observes,
It is always difficult to build a bridge between two fields of knowledge which
have developed their methods and techniques separately, so that even elemen-
tary connections are not easy. Bar-Hillel’s paper of 1954 . . . seems to have
found little echo among linguists so far, although his paper, in contrast to
my publications, is written in a generally comprehensible language, is pub-
lished in a linguistics periodical, and makes direct reference to the work of the
structural linguists. I am not surprised to find that Chomsky in his reply to Bar-
Hillel’s article does not agree with Bar-Hillel’s views; I think that Chomsky
is to some extent right, because Bar-Hillel claims too much when he speaks
about the immediate importance of my investigations for linguistics. But, on
the other hand, I have the impression that Chomsky failed to grasp the meaning
of Bar-Hillel’s appeal and also the aims and nature of my theories of syntax
and semantics, and this shows the great difficulty of communication between
the two fields. (Carnap 1963: 941)

Unfortunately, Carnap does not go on to discuss the precise nature of Chom-

sky’s perceived misunderstanding either of Bar-Hillel’s proposals or of his own
work, and it is no real surprise to find that a discussion which partly concerns
the role of meaning in linguistic theory should conclude with the suggestion
that certain lines of reasoning had been misinterpreted. Nevertheless, the issues
raised by Bar-Hillel, Chomsky, and Carnap concerning the relationship between
natural and logical languages, and, in particular, the role of semantics in linguis-
tic analysis, remained topics of debate throughout the 1960s. Indeed, as is well
known, the more extreme integrationist position adopted by Bar-Hillel reached
its apotheosis in the late 1960s and early 1970s in the work of Richard Mon-
tague (1930–1971). Although the large body of work devoted to ‘Montague
Semantics’ is (alas) outside the scope of this book, it is worth emphasising that,
 Logic and linguistic theory 139

for Montague, natural and logical languages were identical, or, as he expressed
it in the trenchant beginning of his 1970 paper ‘English as a Formal Language’,
‘I reject the contention that an important theoretical difference exists between
formal and natural language’ (Montague 1970: 188). Consequently, Montague
was able to develop a detailed formalism for a truth-theoretical analysis of
natural language which was explicitly intended to provide an alternative to
generative grammar, indicating that the old divisions of the 1950s were still
clearly motivating research twenty years later.18
5 Transforming generative
grammar: 1955–1957

5.1 Chapter overview

The main purpose of this chapter is to continue to reassess the development
of TGG in the light of advances in the formal sciences. Consequently, Chom-
sky’s writings from the years 1955–1957 will be the focus of the discussion,
although, as previously, connections will be made between Chomsky’s work and
that of both his predecessors and his contemporaries. To this end, in section 5.2
Chomsky’s rejection of stochastic grammars is considered as part of his asser-
tion that syntax can be studied autonomously. In section 5.3 his redefinition of
syntactic research is assessed, particularly the recommended shift away from
discovery procedures and towards evaluation procedures. Since certain of the
arguments that caused Chomsky to reconsider the role of discovery procedures
in linguistic research were initially articulated by Goodman and Quine, it is
necessary to explore the influence of constructional system theory upon the
approach to syntactic analysis outlined in LSLT and Syntactic Structures (here-
after SS), and this is achieved in section 5.4 where it is shown that Chomsky’s
definition of a linguistic level is derived from constructional system theory.
Given the name of the syntactic theory discussed in this book, ‘Transformational
Generative Grammar’, it is necessary to discuss both the concept of ‘transfor-
mation’ and the process of ‘generation’. Accordingly, the complex evolution of
grammatical transformations in syntactic theory is traced in section 5.5, while
in section 5.6 the generative role of recursive definitions in TGG is considered.
In both these sections the connections that exist between Chomsky’s work and
that of Bar-Hillel and others are emphasised. Finally, in section 5.7, the influ-
ence of Formalism upon TGG is considered, and the proof-theoretical character
of the theory is discussed.

5.2 Stochastic processes and autonomous grammar

Before considering the type of syntactic theory that Chomsky developed during
the years 1955–1957, it is first necessary to assess his reasons for claiming that

140
 Stochastic processes and autonomous grammar 141

the grammar of a language could be considered in isolation from all other aspects
of language (especially semantics). In section 4.5 the various arguments that
Chomsky developed in the mid-1950s to justify his rejection of logical semantics
were discussed at length, and one consequence of this was that, following the
tradition of Bloomfield and Harris, Chomsky became convinced that syntax
could be studied separately from semantics. In fact, he claimed specifically
that syntax provides a basis for semantics.1 However, Chomsky not only felt
compelled to rescue syntax from the hands of the logicians, but he was also
keen to preserve it from the clutches of statisticians. In particular, he responded
negatively to the concept of a stochastic grammar, which had been proposed
by Shannon and Weaver in the 1940s, and which had been enthusiastically
welcomed by certain post-Bloomfieldian syntacticians. Consequently, before
discussing Chomsky’s rejection of these ideas, it is necessary to summarise
them in some detail.
In the first decades of the twentieth century the statistical properties of nat-
ural language became an active area of research. For instance, Andrei Markov
(1856–1922) introduced stochastic processes, called finite state automata, and
used them to model letter frequencies in Russian poetry (see Sheynin 1988).
Also, George Zipf (1902–1950) published two books, The Psycho-Biology of
Language (1935) and Human Behavior and the Principle of Least Effort (1949),
in which he explored frequency counts for linguistic units such as phonemes,
syllables, and words, observing that they were invariably distributed in a char-
acteristic fashion, with a small group of frequently occurring units and a long
tail of infrequent units. This type of distribution, which can be represented as
a linear plot when the natural logarithm of the frequency is plotted against the
natural logarithm of the unit number, was later called a Zipfian distribution,
and research of this kind suggested that there were patterns hidden in natural
language that could best be revealed by detailed statistical analysis. In turn, this
implied that stochastic models of linguistic behaviour constituted a branch of
statistical theory that was well worth exploring. This general project was given
significant impetus by the publication of Claude Shannon (1916–2001) and
Warren Weaver’s (1894–1978) The Mathematical Theory of Communication in
1949.2 This text, which became one of the seminal works of information theory,
provided a clear introduction to a number of comparatively recent advances in
the theory of stochastic processes, thus making these techniques available to a
wide, non-specialist audience for the first time.
From the viewpoint of linguistic theory, the most significant mathematical
model discussed by Shannon was the finite state machine; that is, the stochastic
process introduced by Markov which, by the 1940s, had come to be known as
142 Transforming generative grammar: 1955–1957

a Markov model. Assuming the existence of a finite set of discrete symbols

(i.e., an alphabet), a Markov model is a stochastic process that generates a
string of symbols in a probabilistic fashion. Each symbol, Si , can be associated
with a probability, p(Si ), and the probability of symbol Si being followed by
symbol S j is given by the transition probability p(Si , S j ). If the history of the
symbol sequence is taken into account, then conditional probabilities of the
form p(S j | S1 , S2 , . . . , S j−1 ) can be defined which consider the probability
of the current symbol to be conditionally related to the preceding sequence of
symbols. With these probabilities established, a Markov model can generate a
string of symbols in an automatic fashion, and it is important to note that, in
Shannon and Weaver’s discussion of the topic, they explicitly state that Markov
models can be used to approximate natural languages such as English. More
precisely, they define an ordered series of such approximations which can be
summarised as follows:

r zero-order approximation: all symbols are independent and equi-

probable.
r first-order approximation: all symbols are independent and associated
with their respective frequencies in English.
r second-order approximation: the symbol probabilities are dependent
upon the previous two symbols (i.e., p(S j | S j−2 , S j−1 ))
r n th -order approximation: the symbol probabilities are dependent upon
the previous n symbols (i.e., p(S j | S j−n+1 , . . . , S j−1 )).

The potential benefit of Markov modelling to the task of grammar construction

is explicitly emphasised by Weaver later in the book, though it is made clear
that the approach recommended is far from idealistic since, when dealing with
natural language, one can only hope to achieve approximate results.

Language must be designed (or developed) with a view to the totality of things
that man may wish to say; but not being able to accomplish everything, it too
should do as well as possible, as often as possible. This is to say, it too should
deal with its task statistically. (Shannon and Weaver 1949: 117)

Accepting that to do ‘as well as possible, as often as possible’ is all that syntactic
theory can hope to achieve, the suggestion here is that language must be studied
as some kind of stochastic information source, since only a model of this type
will permit the sort of approximations which (it is supposed) are required when
natural language is modelled mathematically. However, as will be discussed
below, this humble desire was not shared by the more idealistic members of
 Stochastic processes and autonomous grammar 143

the current generation of linguists, who wanted to accomplish something more

than mere approximation.
In the years immediately following the publication of Shannon and Weaver’s
text, there was a flurry of books and papers that explored some of the implica-
tions of finite state grammars and stochastic processes, and the general response
was one of interest and enthusiasm. For instance, George Miller (b. 1920)
devoted a whole chapter of Language and Communication (1951) to ‘The Sta-
tistical Approach’. In this chapter he summarises Shannon’s use of stochastic
approximations as a model of linguistic behaviour, and emphasises that ‘the
question under consideration is the extent to which any particular verbal unit
is determined by the other verbal units that surround it’ (Miller 1951: 81). It
is important to note that Miller did not only see possible value in Shannon’s
approximations for the modelling of syntax, but was keen also to suggest that
stochastic processes could be used to model any aspect of natural language
structure, including phonetic and phonological structure. The use of stochastic
techniques to analyse the phonetic structure of language was further explored by
Colin Cherry (1928–1981), Morris Halle (b. 1923), and Roman Jakobson (1896–
1982) in their joint 1953 paper ‘Toward the Logical Description of Languages
in their Phonemic Aspect’, in which it was unambiguously stated that the main
purpose of the research presented there was to ‘contribute to a logical descrip-
tion of the phonemic structure of language, employing some of the elementary
concepts of statistical communication theory’ (Cherry, Halle, and Jakobson
1953: 34). In a footnote the authors explicitly cite Shannon and Weaver’s text,
and state clearly that ‘in the description that follows, language will be treated
as a Markoff [sic] process’ (Cherry, Halle, and Jakobson 1953: 36). However,
perhaps the most enthusiastic advocate of finite state process for the modelling
of syntactic phenomena was Hockett, who at that time was one of the leading
post-Bloomfieldians. In 1953, for instance, he reviewed Shannon and Weaver’s
book in the journal Language, and responded positively, stating that Shannon’s
ideas ‘must be investigated’ (Hockett 1953: 70). Further, he felt that some of
the techniques associated with stochastic theories of language shared similar-
ities with the sort of methods standardly employed by structural linguists (yet
another example of the commonly mooted mathematics-linguistics correspon-
dence), and, after his discussion of Shannon’s ideas in relation to linguistic
theory, he declared that ‘we have demonstrated that a certain number of prob-
lems of concern to linguists can be phrased in the terminology of information
theory’ (Hockett 1953: 89). As indicated above, this belief in the compatibility
of linguistics and statistics can be compared with Bar-Hillel’s conviction that
the methodology of language-based structural analysis could be combined with
144 Transforming generative grammar: 1955–1957

certain types of logical systems, as well as with Chomsky’s attempt to fuse

Harris-style discovery procedures and constructional system theory. Clearly
these perceived connections were exciting at the time and, for Hockett, his
perception of this apparent compatibility was associated with a sense of ‘exhil-
aration’, although he was quick to warn against the unthinking acceptance of a
‘misleading analogy’ (Hockett 1953: 89).
Hockett’s 1953 review of Shannon and Weaver’s book was followed two years
later by a detailed statement of his belief in the utility of statistical methods
in linguistic analysis which appeared in his A Manual of Phonology (1955).
In this work, Hockett uses engineering-style flow diagrams to represent the
way in which information is communicated by the speaker to the hearer in a
conversational situation. In order to clarify the stages in the process, he posits the
existence of a ‘Grammatical Headquarters’ (or ‘G.H.Q’), which is responsible
for generating the sentences that are spoken, and his description of this stage
of the process indicates the extent to which he had come to accept Shannon’s
basic approach to syntax.

A unit regarded as purely a source of a discrete flow of signals can be mathe-

matically characterized in a complete fashion on the basis of the statistics of
the signal-flows which it emits; the technique for doing this was developed
by Claude Shannon. We imagine that G.H.Q. can be in any number of a very
large number of different states. At a given moment it is necessarily in one
or another of these states. Associated with each state is an array of probabili-
ties for the emission of various morphemes of the language: a certain relative
probability that the morpheme and will next be emitted, a certain relative
probability that the morpheme tackle will next be emitted, and so on. When
some morpheme is actually emitted, G.H.Q. shifts to a new state. Which state
the new one is depends in a determinate way (not just probabilistically) on
both the preceding states and on what morpheme has actually been emitted.
(Hockett 1955: 7)

This is a strong statement of the stochastic grammar creed: the sequence of

linguistic units emitted by the G.H.Q can be determined ‘in a complete fashion’
by the Markov process. The basic assumption appears to be that finite state
machines are entirely adequate for the task of modelling the grammar of natural
language. Hockett’s belief in the theoretical validity of this assumption (even
though he recognises that such an approach could entail practical difficulties)
is indicated when he states that

By making an enormous count of relative frequencies of occurrence of all the

morphemes and many morpheme-sequences in natural English, followed by
an enormous amount of computation, and by writing very small entries on an
 Stochastic processes and autonomous grammar 145

enormous sheet of paper, the entire grammatical structure of English could be

portrayed . . . (Hockett 1955: 10)

Once again, therefore, Hockett stresses his conviction that stochastic processes,
at least in theory, provide an adequate model for the grammar of a natural
language, since, as indicated above, he believed that ‘the entire grammatical
structure of English’ could be analysed using statistical techniques.
In the light of this support for stochastic techniques amongst certain lin-
guists, particularly Hockett, it is revealing to consider Chomsky’s assessment
of these techniques,3 and the most complete presentation of his ideas concern-
ing stochastic grammars can be found in his 1956 paper ‘Three Models for
the Description of Language’ (hereafter ‘TMDL’) which (it should be noted)
appeared in the I.R.E. Transactions for Information Theory. In fact, the paper
was delivered as part of a series that focused on various aspects of information
theory, and the event took the form of a conference at MIT in 1956. Such a gath-
ering was hardly a conventional forum for a discussion of the syntax of natural
language, and it is revealing that this particular conference should be consid-
ered by some to indicate the emergence of fully-fledged cognitive science.4 As
the title of his 1956 paper suggests, Chomsky’s main purpose was to explore
three different models (i.e., grammars) that provided different descriptions of
a given language. Specifically, the three types of model he considers are finite
state grammars, phrase structure grammars, and transformational grammars.
Various aspects of his presentation of the latter two types will be considered
in the other sections of this chapter, therefore, for now, the emphasis will be
exclusively upon his assessment of finite state grammars.
As expected, Chomsky’s presentation of finite state machines was derived
primarily from Shannon and Weaver’s 1949 book. Consequently, following
them, he states that a finite state grammar, G, for a given language, L consists
of

r a finite number of states, s0 , . . . , sq

r a set, A, of transition symbols where A = {ai j | 0 ≤ i, j ≤ q}
r a set, C, of connected state pairs where C = {(s , s )}
i j

In order to clarify the nature of the set C, it is best to consider an example: if

state s j can be reached from state si , then the pair of states (si , s j ) constitutes a
connected pair. Sentences are formed from the grammar by concatenating the
symbols that are generated. Consequently, a sentence, S, in L takes the form
S = a1  a2  . . .  an , where ai ∈ A for 1 ≤ i ≤ n.
146 Transforming generative grammar: 1955–1957

Having summarised Shannon and Weaver’s Markov process formalism in this

fashion, Chomsky then states that he intends to consider ‘the absolute limits’
(Chomsky 1956: 115) of the type of finite state languages he has just defined. In
other words, he is keen to determine precisely which languages can be generated
by finite state grammars, and which cannot be generated by such machines. In
order to accomplish this, he introduces the concept of ‘dependency’. Assuming
that language L and a sentence S are defined as above, then (i, j)-dependency
for sentence S with respect to L can be said to hold only if the following
conditions are met:

1. 1 ≤ i < j ≤ n
2. there are symbols bi , b j ∈ A with the property that S1 is not a sentence
of L, and S2 is a sentence of L, where S1 is formed from sentence S
by replacing the i th symbol of S (namely, ai ) by bi , and S2 is formed
from S1 by replacing the j th symbol of S1 (namely, a j ) by b j .

The above conditions rely upon the concept of replacement, which, though
not defined, is understood to correspond to substitution. In this way, the
type of technique Chomsky uses in order to investigate ‘the absolute lim-
its’ of finite state grammars, which exploit the distributional properties of
symbol pairs, are closely related to some of the methods employed by Har-
ris and other post-Bloomfieldians. The basic idea behind replacement is that
two symbols, ai and a j , are dependent if, when ai is replaced by bi (where
ai = bi ), a j must also be replaced, this time by the symbol b j (where a j = b j ).
To consider a specific example, given the sentence S = a1  a2  a3 
a4 , where S ∈ L, if the following sentences can be obtained by means of
replacement,

S1 = b1  a2  a3  a4 (5.1)
S2 = b1  a2  a3  b4 (5.2)

where S1 ∈ / L and S2 ∈ L, then a1 and a4 are dependent and the sentence S

has (1,4)-dependency. Armed with this technical definition, Chomsky then pro-
ceeds to define a dependency set, D, such that D = {(α1 , β1 ), . . . , (αm , βm )}.
The set D is a dependency set if and only if the following conditions are
met:
r (i) for 1 ≤ i ≤ m, S has a (α , β )-dependency with respect to L
i i
r (ii) for each i, j, αi < β j
r (iii) for each i, j such that i = j, αi = α j and βi = β j
 Stochastic processes and autonomous grammar 147

If the first element in a connected pair is referred to as a ‘determining’ element

and the second element as a ‘determined’ element, then the above conditions
ensure that all determining elements precede all determined elements and that
no two determiners and no two determined elements are identical. In short, the
conditions guarantee that every two dependencies are distinct. Since the number
of elements in D is m, it follows that, for a given sentence S, the corresponding
finite state grammar must contain at least 2m states. This conclusion leads
Chomsky to state a necessary condition for any language, L, generated by a
finite state grammar:
r Condition 1: There is an m such that no sentence S of L has a depen-
dency set of more than m terms in L
The crucial function of this condition is to delimit the range of languages that can
be generated by a finite state grammar. It achieves this by indicating that, for any
such grammar, there must be an upper bound on the number of terms contained
in the dependency set. This condition then enables Chomsky to create three
simple languages (an infinite number are possible) that cannot be generated by
a finite state grammar. For instance, if the language L uses a two-letter alphabet
(i.e., A = {a, b}), then some typical grammatical sentences in L might be

S1 = a  a (5.3)
S2 = a  b  b  a
S3 = a  a  b  b  a  a
The dependency set for L can be given as D = {(1, 2m), (2, 2m −
1), . . . , (m, m + 1)} for any m, and since D can contain more than any fixed
number, m, of terms, it violates condition 1 above.
In the light of Chomsky’s doubts concerning the usefulness of artificial lan-
guages when analysing natural language (discussed in section 4.5), it is of
interest that, as in ‘SSA’, he here makes full use of simple toy languages in
order to determine the limitations of finite state grammars. However, having
determined these limits, Chomsky is eager to explore these findings in relation
to natural language. Consequently, in order to prove that finite state grammars
are not sufficiently powerful to provide a valid grammatical model for a given
natural language, he simply needs to demonstrate that natural languages contain
mirror-image structures (for instance) such as those presented in (5.3) above.
He accomplishes this by considering structures of the type
if S1 , then S2 (5.4)
148 Transforming generative grammar: 1955–1957

and by observing that there is a dependency between ‘if’ and ‘then’ in (5.4), since
the clause indicated by S1 could itself contain structures of the form ‘if . . . then’,
thus producing an English sentence with the mirror-image property assessed
previously. Consequently, Chomsky concludes that English fails condition 1
above and that therefore ‘no finite-state Markov process that produces symbols
with transitions from state to state can serve as an English grammar’ (Chomsky
1956: 113).
Having demonstrated that finite state grammars are not sufficiently powerful
to generate all the possible sentences of English, Chomsky then dismisses the
idea that n th order approximations, of the type proposed by Shannon and Weaver,
can ever be used to generate the set of grammatical sentences in English. The
thrust of his argument is that grammaticality and frequency are not related
notions, and to indicate that this is so he considers the sentences

Colorless green ideas sleep furiously (5.5)

Furiously sleep ideas green colorless (5.6)

Chomsky claims that these sentences (which were later made famous in SS),
occur equally infrequently in English, yet sentence (5.5) is grammatical while
sentence (5.6) is not. Therefore, he is obliged to conclude that frequency reveals
nothing about grammaticality. The only possible conclusion, therefore, is that
n th order approximations must be rejected since ‘as n increases, an n th order
approximation to English will exclude (as more and more improbable) an ever-
increasing number of grammatical sentences, while it still contains vast numbers
of completely ungrammatical strings’ (Chomsky 1956: 116).
As ever, it is essential to emphasise the full complexity of Chomsky’s argu-
ment. Indeed, there are, in fact, general parallels between his rejection of
stochastic techniques in ‘TMDL’ and his rejection of certain tools from logic in
‘LSS’. In particular, just as Chomsky had advised linguists not to apply blindly
the methodology of logic while attempting to analyse the structure of natural
language, but rather to select and adapt specific techniques if they seem to be
useful, so his advice to statistically minded linguists was actually expressed as
caution rather than prohibition. Certainly, as indicated above, he was convinced
that the grammar of natural language simply cannot be comprehensively mod-
elled as a stochastic process (despite the initial claims of Hockett and others),
but this did not lead him to advocate banishing statistical methods from lin-
guistics entirely. Indeed, on the contrary, he freely confessed that the form and
structure of natural language could be usefully explored using statistics, but his
scepticism concerning syntactic modelling remained.
 From discovery to evaluation 149

Given the grammar of a language, one can study the use of the language sta-
tistically in various ways; and the development of probabilistic models for
the use of language (as distinct from the syntactic structure of language) can
be quite rewarding . . . One might seek to develop a more elaborate relation
between statistical and syntactic structure than the simple order of approxi-
mation model we have rejected. I would certainly not care to argue that any
such relation is unthinkable, but I know of no suggestion to this effect that
does not have serious flaws. (Chomsky 1957b: 17.n4)

In the excised section above, Chomsky cites recent work by Herbert A. Simon
(1916–2001) (specifically, Simon 1955) and Benoit Mandelbrot (b. 1924)
(specifically, Mandelbrot 1954) as specific examples of revealing statistical
studies, and, as the above passage indicates, he is willing to accept that statisti-
cal analyses of linguistic structure can highlight significant patterns. However,
he emphasises that these studies are primarily descriptive in intention and do not
make elaborate claims about the actual mechanisms of language. For Chomsky,
these sorts of concerns must be left for the linguists to discuss, since linguists are
more inclined (than statisticians) to focus upon the actual properties of natural
language, and therefore they are better able to resist the temptation to become
obsessed with peripheral non-linguistic considerations.

5.3 From discovery to evaluation

As mentioned in the previous section, Chomsky’s repeated assertion that gram-
mar is autonomous was at least partly motivated by his belief that semantics and
statistical considerations should not be involved in the definition of grammati-
cality, and it is this belief that enabled him to focus his attention exclusively upon
syntactic phenomena during the 1950s. In this respect, Chomsky’s work cannot
really be distinguished from that of certain post-Bloomfieldians since, as dis-
cussed at length in section 3.7, several generations of linguists, from Bloomfield
to Harris, had sought to provide syntactic analyses of grammatical utterances
without recourse to considerations of meaning. In the context of this tradition,
therefore, the tenor of Chomsky’s research was unremarkable. Nevertheless,
although Chomsky’s focus on syntax was conventional, the manner in which he
came to formulate the task of syntactic analysis in the mid-1950s was certainly
provocative. As is well known, the basic development that Chomsky advocated
was a general shift away from (overly restrictive) discovery procedures of the
type favoured by Harris, towards evaluation procedures, such as the simplicity
criteria discussed in section 4.3, which were designed to facilitate the task of
selecting one of many competing grammars rather than to permit the creation of
150 Transforming generative grammar: 1955–1957

a grammar from scratch given a specific corpus. This basic development is well
known and has been discussed extensively.5 However, Chomsky’s switch from
discovery to evaluation procedures has never really been adequately considered
in the light of his changing response to the kind of logical empiricism advocated
by Goodman (in particular). Consequently, these aspects of Chomsky’s work
will be discussed here.
As mentioned in section 4.4, Chomsky’s first published paper, ‘SSA’,
appeared to suggest that he was contentedly working within a post-
Bloomfieldian paradigm, since the point of that paper was to explore the pos-
sibility of utilising Goodman’s calculus of individuals as a mechanical dis-
covery procedure that could automate grammatical analysis. However, by the
mid-1950s Chomsky’s understanding of the general purpose and function of
linguistic theory had clearly begun to change with the result that, between the
appearance of ‘SSA’ in 1953 and the completion of the first draft of LSLT some
time in early 1955, the emphasis of his research had shifted away from tax-
onomic discovery procedures towards the type of theory that would later be
referred to as TGG. Chomsky’s own account of this change in direction is well
known but is still worth quoting in this context.

By 1953 I came to the same conclusion [i.e., as Halle]: if the discovery proce-
dures did not work, it was not because I had failed to formulate them correctly
but because the entire approach was wrong. In retrospect I cannot understand
why it took me so long to reach this conclusion – I remember exactly the
moment when I finally felt convinced. On board ship in the mid-Atlantic,
aided by a bout of seasickness, on a rickety tub that was listing noticeably – it
had been sunk by the Germans and was now making its first voyage after
having been salvaged. It suddenly seemed that there was a good reason –
the obvious reason – why several years of intense effort devoted to improv-
ing discovery procedures had come to nought, while the work I had been
doing during the same period on generative grammar and explanatory theory,
in almost complete isolation, seemed to be consistently yielding interesting
results. (Chomsky 1979a: 131)

Whatever the truth of this story (and whatever ‘the obvious reason’ might actu-
ally have been), the underlying causes of this general shift in perspective were
many and various, though, certainly, the fact that ‘several years of intense effort’
had been (seemingly) wasted on discovery procedures could have provided the
main motivation for the change. In addition, though, there is no doubt that the
prioritising of evaluation over discovery procedures was partly prompted by
Chomsky’s evolving appreciation of the work of Goodman and Quine. In partic-
ular, during the early 1950s the type of hard-line Carnapian logical empiricism
 From discovery to evaluation 151

espoused by Goodman, and (seemingly) championed so assertively by Chomsky

in ‘SSA’, ceased to impress him, and his increasing disillusion was apparently
bolstered by his sympathy for the trenchant questioning of reductionist logi-
cal empiricism offered by Quine in ‘Two Dogmas of Empiricism’. Chomsky’s
1975 recollection of the general progress of his thinking during the formative
years 1947–1955 was quoted in the introduction, but it will be reproduced here
for ease of reference.6

At Harris’s suggestion I had begun to study logic, philosophy, and founda-

tions of mathematics more seriously as a graduate student at the University
of Pennsylvania, and later at Harvard. I was particularly impressed by Nelson
Goodman’s work on constructional systems. In its general character, this work
was in some ways similar to Harris’s, and seemed to me to provide the appro-
priate intellectual background for the investigation of taxonomic procedures
that I then regarded as central to linguistic theory. But Goodman’s ongoing cri-
tique of induction seemed to point in a rather different direction, suggesting
the inadequacy in principle of inductive approaches. Goodman’s investiga-
tion of the simplicity of systems also suggested (to me at least) possibilities
for nontaxonomic approaches to linguistic theory. Quine’s critique of logical
empiricism also gave some reason to believe that this line of enquiry might be
a plausible one. Quine argued that the principles of scientific theory are con-
fronted with experience as a systematic complex, with adjustments possible
at various points, governed by such factors as general simplicity. (Chomsky
1975[1955]: 33)

If these remarks can be trusted, then this passage is of particular interest since it
reveals a considerable amount about the underlying philosophical motivations
that prompted the development of TGG. Crucially, it identifies some of the
causes for the dramatic shift away from the sort of empirical discovery pro-
cedures that had characterised post-Bloomfieldian linguistics, and which had
provided the focus for Chomsky’s own early work. Specifically, it is perhaps
surprising that, in the above account of the development, Chomsky’s motivation
seems to have been primarily philosophical in origin rather than purely linguis-
tic. As was discussed in section 3.5, ever since Carnap had formulated construc-
tional system theory in the LCW, the use of such systems had been strongly
associated with Carnapian logical empiricism, which, as a theoretical approach
to the problem of knowledge acquisition, was (as the name suggests) funda-
mentally empirical, with objects in a given system (i.e., the ‘Gegenstanden’)
being ultimately derived, by means of definition, from sensory experience.
However, the above passage suggests (ironically) that Goodman’s attempts in
the early 1950s to justify the type of inductive procedures required by logi-
cal empiricism, combined with Quine’s probing scepticism about empiricism
152 Transforming generative grammar: 1955–1957

in general, caused Chomsky to question the validity of the sort of inductive

assumptions inherent in the standard taxonomic discovery procedures proposed
by the post-Bloomfieldians. Although Chomsky does not give specific details,
it is likely that he had Goodman’s Fact, Fiction and Forecast (1954) in mind.
As mentioned briefly in section 3.6, in this work Goodman discussed the role of
induction in philosophical research and, consequently, was obliged to consider
some of the difficulties associated with inductive procedures. In particular, the
emphasis in his discussion was upon the task of prediction, and it should be
recalled that Chomsky’s nascent theory of TGG was specifically designed to
accomplish the task of predicting all grammatical sentences in a given natural
language, hence the relevance of Goodman’s discussion to Chomsky’s work.
For Goodman (following David Hume (1711–1776)), one of the main issues
that must be addressed was ‘the question of how predictions are related to
past experience’ (Goodman 1954: 60), and he observes that ‘the problem of
justifying induction has called forth as much fruitless discussion as has any
half-way respectable problem of modern philosophy’ (Goodman 1954: 61).
From the outset, then, the tone of Goodman’s summary is negative, and he
continues

Understandably, then, more critical thinkers have suspected that there might
be something wrong with the problem we are trying to solve. Come to think
of it, what precisely would constitute the justification we seek? If the problem
is to explain how we know that certain predictions will turn out to be correct,
the sufficient answer is that we don’t know any such thing. If the problem is to
find some way of distinguishing antecedently between true and false predic-
tions, we are asking for prevision rather than for philosophical explanation.
(Goodman 1954: 62)

The clear implication of this overview is that inductive procedures are beset
with real difficulties and, to provide a contrast, Goodman then considers ‘non-
inductive inferences’ (i.e., deductive processes) and, in this case, the situation
is less complex.

How do we justify a deduction? Plainly by showing that it conforms to the

general rules of deductive inference. An argument that so conforms is justified
or valid, even if its conclusion happens to be false. An argument that violates
a rule is fallacious, even if its conclusion happens to be true. To justify a
deductive conclusion therefore requires no knowledge of the fact it pertains
to. (Goodman 1954: 63)

Although Goodman then goes on to discuss the relationship between induc-

tion and deduction, and although the main point of the essays in Fact, Fiction
 From discovery to evaluation 153

and Forecast is to attempt to resolve some of the difficulties associated with

inductive procedures, it is certainly possible that Goodman’s critique of induc-
tion is sufficiently powerful to induce scepticism concerning its validity as a
research method in the empirical sciences. Consequently, the above extracts
(and other similar passages) from Goodman’s argument, which were intended
simply to provide the motivation for the development of an epistemic system
of projections, could certainly appear to constitute an alarming condemnation
of empiricism as a practical philosophy.
So much for Goodman’s ‘ongoing critique of induction’. As for Quine, his
growing scepticism concerning logical empiricism was well expressed in LPV.
In ‘Two Dogmas of Empiricism’ (the text Chomsky explicitly cited as an influ-
ence in 1975), for instance, Quine considers the nature of the relation between
statements in natural language and the experiences that are involved in the
confirmation of said statements. In this context, he continues,
The most naı̈ve view of the relation is that it is one of direct report. This is
radical reductionism. Every meaningful statement is held to be translatable
into a statement (true or false) about immediate experience . . . So stated, the
doctrine [i.e., of radical reductionism] remains ambiguous as between sense
data as sensory events and sense data as sensory qualities; and it remains
vague as to the admissible ways of compounding. Moreover, the doctrine is
unnecessary and intolerably restrictive in its term-by-term critique which it
imposes . . . (Quine 1953: 38)

Having outlined his basic concerns in this general fashion, Quine then launches
a blistering attack on logical empiricism, as practised by Carnap in LCW (and
Goodman in SA, and Chomsky in ‘SSA’), and his basic criticism is that the
whole approach was ‘left in a sketchy state’ (Quine 1953: 40) since Carnap
never adequately indicated how statements such as ‘Quality q is at point-instant
(x, y, z, t)’ could ever be translated into the parsimonious initial language that
consisted solely of logical axioms, extralogical primitives, and sense data. As a
result, the whole methodology was flawed. It is important to emphasise, though,
that Quine’s purpose in his paper was to propose a modification of empirical
philosophy, not entirely to reject it (‘As an empiricist I continue to think of the
conceptual scheme of science as a tool, ultimately, for predicting future expe-
rience in the light of past experience’ (Quine 1953: 44)), and, in this respect,
his position is similar to Goodman’s concerning induction: both men desta-
bilise aspects of empiricism and then proposed ways of rendering them more
secure. As indicated in this brief summary, the critique of induction and logical
empiricism that was developed in the work of Goodman and Quine would cer-
tainly have interested Chomsky in the early 1950s, given his experiments with
154 Transforming generative grammar: 1955–1957

constructional systems, and, as he himself acknowledged in 1975, Goodman’s

and Quine’s arguments were partly responsible for propelling him towards the
(now) more characteristic rationalist stance that he began publicly to adopt in
the early 1960s. One consequence of the (often neglected) influence of Good-
man and Quine upon Chomsky’s work is that Chomsky’s later well-known
criticism of behaviourism should be reassessed. Indeed, it seems most likely
that the robust rejection of behaviourism, which became one of the defining
characteristics of the general philosophy of mind that underpinned research
into generative grammar in the 1960s, actually had its roots in Chomsky’s
changing response to logical empiricism of the Carnapian variety.7 Gradually,
therefore, during the years 1953–1955 the focus of Chomsky’s research began
to shift away from pure post-Bloomfieldian methodologies and, as a result, he
felt obliged to reassess the fundamental aims and purpose of linguistic theory
in general.8 Specifically, and in accordance with his rejection of inductive pro-
cedures, Chomsky was keen to simplify the basic task that linguists had set for
themselves. To this end, in SS he defined three different approaches to the task
of syntactic analysis, which he described as follows:

1. Discovery Procedure: ‘the theory must provide a practical and mechan-

ical method for actually constructing the grammar, given a corpus of
utterances’ (Chomsky 1957b: 50)
2. Decision Procedure: ‘the theory must provide a practical and mechan-
ical method for determining whether or not a grammar proposed for a
given corpus is, in fact, the best grammar’ (Chomsky 1957b: 50–51)
3. Evaluation Procedure: ‘given a corpus and given two grammars G 1
and G 2 , the theory must tell us which is the better grammar of the
language from which the corpus is drawn’ (Chomsky 1957b: 51)

and, having provided these definitions, Chomsky went on to state that

The point of view adopted here is that it is unreasonable to demand of linguistic

theory that it provide anything more than a practical evaluation procedure for
grammars. That is, we adopt the weakest of the three positions described above.
As I interpret most of the careful proposals for the development of linguistic
theory, they attempt to state methods of analysis that an investigator might
actually use, if he had the time, to construct a grammar of a language directly
from the raw data. I think that it is very questionable that this goal is attainable
in any interesting way, and I suspect that any attempt to meet it will lead to
a maze of more and more elaborate and more complex analysis procedures
that will fail to provide answers for many important questions about linguistic
structure. (Chomsky 1957b: 52–53)
 From discovery to evaluation 155

The thrust of Chomsky’s argument, then, was that while the mechanical pro-
duction of a grammar by means of discovery procedures may be a desirable
accomplishment, it may simply be impossible to achieve, whereas a less ambi-
tious approach could actually provide ‘answers’. As Chomsky indicates, the
fundamental difference between the three positions can be simply stated as
the privileging of evaluation procedures (i.e., the selection of one grammar
from amongst many, given a particular corpus) over either discovery or deci-
sion procedures (i.e., either the automatic inferring of a grammar from a given
corpus, or the automatic assessment of the validity of a grammar), and the
advice is to focus upon the relativistic exploration of the validity of com-
peting grammars, rather than upon the notion of absolute correctness. As
noted in section 4.3, Goodman’s research into simplicity measures for con-
structional system theory partly motivated this reassessment of the linguist’s
task by considering the relative simplicity of the extralogical components in
various competing systems. In addition, it should also be obvious that the
task Chomsky came to advocate was the least empirical of the three tasks
defined; a change that can be viewed as a manifestation of his response
to Goodman’s and Quine’s critiques of induction and logical empiricism
respectively.
Whatever the precise nature of the influences that prompted his revisions,
there is no doubt that Chomsky actively set about redefining the goal of linguis-
tics (and especially syntactic theory) in LSLT, a work in which the rejection
of inductive discovery procedures is clearly stated. The avoidance of a purely
nominalistic constructional methodology is signalled early in the book when
Chomsky explains that, although such an approach is ‘more natural’ than the
analytic framework adopted in LSLT, it will not be further developed because
it necessitates the use of ‘somewhat more elaborate constructions’ (Chom-
sky 1975[1955]: 110); a remark that recalls Quine’s complaint concerning the
‘unnecessarily and intolerably restrictive’ (Quine 1953: 38) methods required
by the more extreme forms of logical empiricism. Nevertheless, despite this
unambiguous change of direction, it is equally apparent that LSLT is securely
rooted in constructional system theory. Indeed, it is a curious aspect of the
development of TGG that, although the basic philosophy of inductive logical
empiricism was spurned during the years 1953–1955, the technical devices
associated with constructional system theory continued to be utilised exten-
sively. Somehow, the rejection of the philosophy underpinning the theory did
not invalidate the tools it provided, and, as will be shown in section 5.4, Chom-
sky continued to use and develop the core formal procedures, adapting them to
the task of linguistic analysis.
156 Transforming generative grammar: 1955–1957

5.4 Constructional levels

It was suggested in the preceding section that Chomsky’s changing understand-
ing of the implications of Goodman’s strict logical empiricism, and his appre-
ciation of Quine’s critique of the same, caused him to reconsider the empirical
aspects of syntactic analysis. Accordingly, he came to reject the standard tax-
onomic methodologies, favouring instead evaluation procedures that enabled a
linguist automatically to choose between competing grammars. This rejection
of Goodman-style empiricist concerns, and the corresponding disenchantment
with the type of approach to syntactic analysis outlined in ‘SSA’, could have
caused Chomsky to eradicate all vestiges of his early (and brief) construc-
tive nominalist phase from his later work. In the event, this did not occur. For
instance, in section 4.3 it was shown that considerations of grammatical simplic-
ity continued to feature prominently in his work during the years 1953–1957,
although his ideas concerning the purpose and function of grammatical analysis
changed throughout this period. As another example, in this section the basic
definition of a linguistic level as presented in LSLT is considered, and it will be
shown that the influence of constructional system theory is still apparent even
in his post-1954 work. If the following discussion demonstrates anything, it
demonstrates Chomsky’s remarkable ability to detach a useful procedure from
its philosophical context and employ it with few subsequent qualms.
Throughout LSLT it is emphasised that there are three main tasks that confront
the ‘descriptive linguist’ (Chomsky 1975[1955]: 77). These three tasks can be
summarised as (i) the task of creating a general theory of linguistic structure,
(ii) the task of constructing grammars for particular languages, and (iii) the task
of justifying and validating experimental results, and they provide the main
motivation for the type of theoretical approach to linguistic analysis adopted in
the book. It is crucial to recognise (as Chomsky repeatedly asserts throughout the
text) that, although the above tasks are distinct, they are not entirely independent,
and therefore impinge upon one other to a considerable extent. For instance, the
general theory of linguistic structure adopted for analysis will largely determine
the nature of the resultant grammar that is ultimately constructed; and vice versa.
With this interconnection in mind, Chomsky describes the fundamental problem
of general linguistic theory as follows:
A language is an enormously complex system. Linguistic theory attempts to
reduce the immense complexity to manageable proportions by the construction
of a system of linguistic levels, each of which makes a certain descriptive
apparatus available for the characterisation of linguistic structure. (Chomsky
1975[1955]: 63)
 Constructional levels 157

In other words, the essential task is to create an hierarchical system, composed

of various interrelated linguistic levels, that permits the analysis of natural lan-
guage into basic units of some kind. The assumption is that simple regularities
underlie the apparent surface complexity of all natural languages and, clearly,
this type of method for analysing linguistic structure is a form of reductionism.
Indeed, Chomsky freely acknowledges that the particular methodology devel-
oped in LSLT is characterised by ‘a strong kind of reductionism’(Chomsky
1975[1955]: 85, italics added). However, as the above passage suggests, the sort
of system required in order to reduce the full complexity of natural language
to single components must be fundamentally hierarchical. This fact stresses
the difference between the project proposed in LSLT and that outlined in ‘SSA’
(and discussed in section 4.4): while the earlier work was concerned exclusively
with one level of linguistic analysis, the morphemic level, the latter is concerned
with the attempt to develop a comprehensive analytic framework for the whole
of language, hence the need for multiple linguistic levels. While acknowledging
that the exact number of levels required in an exhaustively hierarchical linguis-
tic theory is a contentious issue, Chomsky maintains that such a theory needs
a minimum of six levels. These are abbreviated as Pn (the phonemic level), M
(the morphemic level ), W (the word level), C (the syntactic category level), P
(the phrase-structure level), and T (the transformational level). In an abstract
presentation, the schematic representation of each of the above is the same,
since any linguistic level, L, can be defined as an independent system which
contains the following components:
L = [L , , R1 , . . . , Rm , µ, , φ1 , . . . , φn ] (5.7)
where
r L is a set of primes (i.e., level-specific primitive elements).
r  is the concatenation relation.
r R1 , . . . , Rm are classes and relations defined within L.
r µ is a set of L-markers (i.e., elements created at level L).
r  is a mapping which, in particular, maps µ into the set of grammatical
utterances.
r φ1 , . . . , φn are relations that express the relationship between L and
all the other n levels.
It is clear from the above definition that at the core of any linguistic level
there is a set of primitive elements (the primes) and various sets of relations.
Of these relations, though, only the concatenation relation is primitive, and
therefore Chomsky later observes that ‘we have provided only one way of
158 Transforming generative grammar: 1955–1957

constructing elements in L’ (Chomsky 1975[1955]: 107). Since mathematical

logic is assumed as the basis of any general theory of linguistic structure (just
as it is usually assumed as the basis of any scientific theory (see Chomsky
1975[1955]: 87)), the various primes, and the primitive concatenation relation
associated with a given linguistic level, Ln , can be viewed as the extralogical
basis of a constructional system. Therefore, the general theory of linguistic
structure proposed in LSLT extensively utilises an hierarchy of independent
(though interrelated) constructional systems. Objects created at a given level,
Ln , using the level-specific primes and relations can be associated with objects
on other levels by means of the mapping function, , but the various levels
are essentially distinct since objects existing at a specific level are created at
that level. To consider a specific example, the linguistic level Pn contains a
finite and fixed set of primes, which are phonetic symbols ( p1 , . . . , pn ), and
various relations including the basic concatenation relation ‘’. Therefore,
if p1 and p2 are two primes in Pn, then the object ‘ p1  p2 ’ is constructed
by concatenating the two primes to form a sequence of primes. In this fashion,
larger objects, or L-markers, can be constructed at each level in accordance with
the basic principles of constructional system theory. It is revealing, incidentally,
to note the presence of ‘classes’ (i.e., sets) in the definition of a level. The
inclusion of such abstract entities once again underlines the difference between
the nominalistic methodology adopted in ‘SSA’ and the freer, less restrictive,
linguistic theory presented in LSLT, which willingly embraces abstract concepts
of certain kinds.
In passing, it is worth emphasising that the use of concatenation as the basic
primitive relation for the linguistic levels reveals the direct influence of Quine.
As mentioned in section 3.5, Quine had explored the role of concatenation
in formal language theory in the 1940s, and had claimed that the whole of
arithmetic could be founded upon this single relation. He had made his ideas
available in chapter 7 of his text Mathematical Logic (especially the revised
second edition, published in 1951), and this text provided Chomsky with an
important source of information concerning the core techniques of formal logic
(as did chapter 4 of Paul Rosenbloom’s Elements of Mathematical Logic). How-
ever, Chomsky borrowed more than concatenation from Quine. For example,
the definition of ‘occurrence’ given in LSLT is taken unaltered from Quine’s
book (compare Chomsky 1975[1955]: 109 and Quine 1951: 297), and these
borrowings indicate the full extent to which Quine’s presentation of symbolic
logic was influencing the type of formal linguistic theory that Chomsky was in
the process of developing. Once again, this fact should prompt a reconsideration
of Chomsky’s views concerning logic and natural language. In section 4.5 it
 Transforming transformations 159

was shown that, following Harris, he discouraged the identification of linguis-

tics with logic, questioning Bar-Hillel’s advocation of Carnap’s observation
that linguists and logicians differ in ‘degree’ rather than in ‘kind’. However,
as his adaptation of Quinean concatenation and occurrence indicate, he was
happy to make use of techniques from logic that were deemed to facilitate the
analysis of natural language in its own terms. Indeed, as should be clear by
now, constructional system theory itself was securely rooted in PM-style logic,
so Chomsky’s use of techniques from this theory in LSLT indicates the way
in which he was willing to adapt logic-based procedures and incorporate them
into the system he was constructing in order to analyse natural language.

5.5 Transforming transformations

The preceding section indicated that the influence of Goodman and Quine
was still very much present in Chomsky’s work from the mid-1950s despite
his apparent rejection of Goodmanesque logical empiricism. However, other
aspects of TGG can also be considered in relation to some of the develop-
ments in the formal sciences discussed in chapter 3. One such notion is the
use of so-called ‘transformations’. As mentioned in section 5.4, the ‘trans-
formational’ part of TGG was one of its most influential characteristics, and
this alone necessitates a consideration of the various ways in which the term
‘transformation’ was used in Chomsky’s work from the period 1955–1957. A
considerable amount of attention has been devoted to this topic over the years,
and the main emphasis has usually been upon establishing the priority either of
Harris’ or Chomsky’s use of the term.9 While such investigations are undeniably
entertaining, a more stimulating assessment can be provided if other sources
of influence are considered. Therefore, while Harris’ use of the term will be
mentioned below, the issue of priority will not really be addressed. Rather, the
intention is to examine the way in which Carnap’s ‘transformation rules’ were
transformed into Chomsky’s ‘transformations’ during the 1950s. Carnap’s use
of formation and transformation rules was discussed at length in section 3.7,
and it should be recalled that, in Carnap’s system, the former are involved in
creating sentences while the latter are essentially rules of logical inference that
enable one sentence to be inferred from another. However, as will be shown
below, the term came to be used in linguistics literature in a bewildering range
of different ways.
Chomsky first uses the term ‘transformation’ in MMH, and, in particular, he
makes extensive use of ‘transformation statements’ (Chomsky 1979b[1951]: 6),
which constitute a key component of the type of grammar constructed in his
160 Transforming generative grammar: 1955–1957

thesis. Indeed, one component of the grammar presented there is described as

‘a series of morphological and morphophonemic statements transforming any
grammatical sequence of morphemes into a sequence of phonemes’ (Chomsky
1979b[1951]: 4). So important are these statements to the grammatical formal-
ism that Chomsky develops that he later defines a grammar entirely in terms
of such statements: ‘the grammar, then, will be a set of transformation state-
ments each of which transforms a given representation of a sentence into a more
specific one’ (Chomsky 1979b[1951]: 6). Admittedly, as a definition of a gram-
mar, this statement is rather vague, and, in recognition of this fact, Chomsky
immediately starts to introduce the formal notation that he uses throughout the
thesis. The basic transformation statements take the form ‘α → β’, where α
and β ‘contain no notational elements but are simply sequences of the elements
set up to represent parts of sentences (phonemes, morphemes etc.)’ (Chomsky
1979b[1951]: 6). In addition, conditions can be imposed upon these transforma-

tions in order to restrict their scope. For instance, if α = α1 β1 γ and β = α1 β1 γ ,
then the corresponding transformation rule would be
α1 β1 γ → α1 β1 γ (5.8)
and this conversion could be written in the form
β1 → β1 , in environment α1 − γ (5.9)
where the context that determines the modification of β1 is indicated in (5.9) as a
general condition. In this way, key relationships between sequences of elements
can be precisely stated in a concise fashion. It is of interest that some of the
symbols Chomsky used in his transformation statements, particularly his use
of the symbol ‘→’, recall some of the notations used in standard expositions of
Hilbert’s proof theory. For instance, although Hilbert himself usually employed
the symbol ‘⊃’ for implication, the symbol ‘→’ was commonly used for this in
the sort of textbooks discussed in section 2.8, and, as a result, there is an apparent
association, both in terms of symbolism and basic concept, between Chomskyan
transformation statements and Hilbert’s proof-schema given as (2.10) above.
Essentially, in both cases, a string of symbols is obtained from another string
of symbols by means of a formal mapping procedure. Consequently, in the fol-
lowing discussion the Chomskyan transformations will sometimes be referred
to as rules of (quasi-)inference.
Harris’ public use of the term ‘transformation’ apparently dates from 1952
when, in his paper ‘Discourse Analysis’, he included a section that consid-
ered the topic of ‘grammatical transformations’. In this part of his discussion,
Harris introduces two types of transformation, those that operate upon classes
 Transforming transformations 161

of morphemes and those that operate upon single morphemes, and both types
are presented as ways of incorporating information from outside the text into
the task of analysing a discourse. Harris’ intention was basically to define
equivalence classes in order ‘to discover patterned (i.e., similar or partly sim-
ilar) combinations of these classes in successive intervals of the text’ (Harris
1952: 18–19). Consequently, the notation Harris uses is different from that
which Chomsky had employed since, rather than employing a symbol that
denotes some kind of (quasi-)inference, Harris simply uses the equality oper-
ator to enable him to state equivalences. For instance, in Harris’ formalism,
the equivalence of active–passive pairs enables sentences such as ‘Casals plays
the cello’ to be considered equal to sentences such as ‘The cello is played by
Casals’, and, in formal notation, Harris defines this equivalence as follows:

N1 V N2 = N2 is V − en by N1 (5.10)

Equivalences such as (5.10) define grammatical transformations of various

kinds, which Harris then uses to accomplish the task of discourse analysis.
So far, then, three different types of transformation have been considered (i.e.,
Carnap’s logical inferences, Chomsky’s context-dependent (quasi-)inferential
statements, and Harris’ equivalences) and the complex relationship between
these different types was recognised during the 1950s by other linguists. As
ever, Bar-Hillel’s interdisciplinary background enabled him to perceive some
of the connections that other linguists, less familiar with recent developments
in logical analysis, may have missed. For instance, in his 1954 paper (discussed
at length in section 4.5), Bar-Hillel explicitly considers Carnap’s formation
and transformation rules and regrets the fact that, while the former have been
studied by linguists with reference to natural language, the latter have been
almost entirely neglected. He refers to Harris’ work in particular and notes that,
although Harris uses the term ‘transformation’, the equivalence statements that
he employs are actually formation rules, in the Carnapian sense:

A recent attempt by Harris to reduce the transformational part of syntax to its

formational part is based on a series of equivocations in the terms language,
equivalent, commutable and their cognates, and so is without foundation.
(Bar-Hillel 1954: 237)

The precise nature of the supposed ‘equivocations’ that Bar-Hillel identified

in Harris’ work is not of primary concern here, nor indeed is it discussed in
Bar-Hillel’s paper. What is of greater interest is Bar-Hillel’s general identifica-
tion of the type of distributional equivalence statements that Harris employed
with Carnapian formation rules. For Bar-Hillel, at least, Harris’ transformations
162 Transforming generative grammar: 1955–1957

were not the same as Carnapian transformations, since, as mentioned above, the
latter are essentially rules of logical inference, while Harris’ statements cannot
obviously be classified as rules of this kind. Since it is known that Harris was
familiar with Carnap’s LSL, and since (presumably) he understood the structure
of the logical system presented in that text, it is either to be assumed that Harris
actually believed that his grammatical transformations defined rules of infer-
ence (which is unlikely), or else that he was consciously adapting a term taken
from logical syntax and redefining it within the context of linguistic analysis.
The latter alternative seems to be the most reasonable option, partly because (as
indicated in section 3.7) Harris was adamant that logicians and linguists studied
different things, and partly because (as stated above) Harris’ transformations are
not presented as if they were rules of inference. Also, revealingly, Harris never
cites Carnap as a direct source for his use of the term ‘transformation’. Indeed,
he generally implied that his own use of the concept was developed to facilitate
the analysis of utterances in natural language. For instance, writing in 1957,
he observed that ‘the study of transformations arose out of the attempt to con-
struct a method for analysing language samples longer than a sentence’ (Harris
1957: 283.n1), implying that a concern with discourse structure analysis pro-
vided the main inspiration. Obviously, this does not exclude the possibility that
he knowingly adapted terminology that he had first encountered in the logical
syntax literature, but his silence concerning this matter is curious nonetheless.
Bar-Hillel’s interpretation of Carnap’s transformation rules was reinforced
in Chomsky’s 1955 response (also discussed at length in section 4.5), provid-
ing further evidence to support the idea that, as far as linguists familiar with
recent developments in logic were concerned, Carnapian transformations did
indeed constitute rules of logical inference. For instance, while summarising
Bar-Hillel’s argument, Chomsky himself considers the nature of Carnapian for-
mation and transformation rules. He observes that, for Carnap, the notion of
formal consequence is given as a primitive in the system he constructs and,
while indicating that such an assumption would not contribute to the task of
analysing natural language, he comments that ‘the transformation rules of logic
are rules of valid inference; hence in this discussion the terms “inference” and
“transformation” will be used interchangeably’ (Chomsky 1955a: 37.n4). The
phrase ‘transformation rules of logic’ is revealing, since the implication is that
Chomsky was already thinking of linguistic transformations as being different
kinds of formal devices from those encountered in logical syntax, and it should
be remembered that, as mentioned above, the ‘transformation statements’ used
in MMH were certainly not strictly rules of valid inference. A slight confu-
sion occurs, though, when Chomsky goes on to consider Harris’ grammatical
 Transforming transformations 163

transformations. Since he has declared that the word ‘transformation’ will be

used interchangeably with the word ‘inference’, does it follow that he under-
stands Harris’ equivalence statements to be rules of inference, or is it the case
that the adjective ‘grammatical’ is intended to indicate that linguistic, and not
logical, transformations are being discussed here? Ambiguities such as these
are prevalent in the various papers from this period that discuss transformations.
What is clear, however, is that, by the mid-1950s, the active–passive relation
had come to be the focus of attention when grammatical transformations were
being discussed, and certainly this relation would continue to be associated with
transformations in the TGG framework outlined in LSLT and SS.
So far, the topic of transformations has been discussed with reference to the
work of Harris, Bar-Hillel, and Chomsky. However, it is crucial to recognise
that the notion of linguistic transformations had begun to spread throughout the
linguistics community by the mid-1950s, and this dissemination of ideas seems
to have been partly inspired by Bar-Hillel’s 1954 paper, which had advocated
the exploration of Carnapian transformation rules within the context of natural
language analysis. Not surprisingly, this suggestion appears to have resonated
with mathematically minded linguists in particular. For instance, in his 1955
paper (already discussed in section 3.2 in a different context), Harwood explic-
itly responded to Bar-Hillel’s plea for Carnapian transformations to be used in
linguistic research, and, at the start of this paper, he unambiguously outlined
his intentions as follows:

Given a morpheme list for a language, the aim of a syntactic system is to

tell us how to put together the sequences of morphemes which are used as
sentences in the language. Such directions we shall call the formation rules.
Additionally, works on syntax usually give a certain amount of information
about the equivalences between some sequences and others, e.g. that John
discovered the path = The path was discovered by John. We shall call such
statements transformation rules. Only a small part of the transformation
rules is covered in syntax; some others are discussed in mathematical logic, e.g.
Russell’s theory of descriptions, procedures of generalisation and abstraction.
So far in mathematical logic, most attention has been directed to developing
formation and transformation rules of artificial languages, and no complete
treatment has yet been made of the transformation rules of natural language.
(Harwood 1955: 409)

The passage raises a number of issues that have already been discussed, and
it indicates how these issues were considered to be deeply interconnected at
the time. For instance, in the above passage we encounter the (by now famil-
iar) belief that techniques from logic can be usefully employed when natural
164 Transforming generative grammar: 1955–1957

language is analysed, as well as the basic Carnapian distinction between for-

mation and transformation rules. However, for Harwood, Harris’ equivalences
are indeed considered to be transformations of the Carnapian type, contrary
to Bar-Hillel’s assertion discussed above. Indeed, the basic difference between
the two kinds of rules, as far as Harwood is concerned, is that formation rules
are used to construct sentences, while transformation rules are used to mod-
ify the form of those sentences. There are clear parallels between this view of
syntactic analysis and the mature theory of TGG, with its deep structure and
surface structure levels, that Chomsky was then in the process of constructing.
Once again, therefore, this serves to emphasise the extent to which particular
concepts and techniques that eventually came to be associated with TGG were
already spreading throughout the linguistics community, albeit in an imprecisely
articulated form, by the mid-1950s.
Given the above discussion, it is necessary to consider the way in which
Chomsky began to use the term ‘transformation’ between the years 1955–1957,
and it is worth noting that he used it conspicuously, since (as indicated earlier)
chapter 7 of LSLT, which he submitted as his Ph.D. thesis in 1955, was called
‘Transformational Analysis’. As is well known, the transformational level of
syntactic analysis presented in LSLT was motivated by perceived limitations in
phrase structure analysis. For instance, Chomsky explicitly discusses particular
constructions that the phrase structure component struggles to deal with, and
these include (for example) yes-or-no questions and active–passive pairs.10 The
central idea behind the transformational component, as developed in LSLT, was
that the transformational rules would operate upon the lowest-level strings of
symbols output by the phrase structure component (i.e., the set P̄ which consists
of morphological heads and syntactically functioning affixes). In the sections on
phrase structure analysis (i.e, chapter 7), Chomsky has already introduced ρ1 -
derivations and P-markers; the former constitute the set of grammatical strings
that consist of terminal symbols, and the latter encode a phrase structure analysis
of the grammatical strings. Consequently, at the start of chapter 9, Chomsky
summarises the fundamentals of transformational analysis as follows:

We are thus led to develop a new level of syntactic analysis, the level T of
transformations, and to assign T-markers to strings of words as markers of their
‘transformational history’. That is, the T-markers of a string of words will tell
us how this string is derived from a certain kernel of sentences which have ρ1 -
derivations and P-markers. In terms of previous levels we can represent each
sentence as a string of phonemes, words, syntactic categories, and, in various
ways, as strings of phrases. Now we will be able to represent a sentence as
a sequence of operations by which this sentence is derived from the kernel
 Transforming transformations 165

of basic sentences, each such sequence of operations corresponding to a T-

marker. (Chomsky 1975[1955]: 306)

This passage emphasises the fact that, as already indicated above, the trans-
formations used in TGG operate upon existing strings of symbols (i.e., kernel
sentences) and modify their structure in precisely specified ways. It is the need
to restrict the applicability of the various transformations that leads Chomsky
to specify three basic conditions - namely, C1, C2, and C3 – which apply to the
set of grammatical transformations. These conditions can be defined as follows
(compare Chomsky 1975[1955]: 311):

C1 A grammatical transformation T is defined on ordered pairs (Z , K ) where

(a) Z and T (Z , K ) are strings in P̄, (b) K is a set of strings in P, and Z is a
member of K .
Thus T operates on a string Z of P, with the analysis given by K , and produces
a new string of P which is denoted ‘(T (Z , K ))’.
C2 T (Z , K ) is unique; that is, T is a single-valued mapping.
C3 The domain of each T is limited to strings of a certain structure. This
limitation can be effected by associating with each T a finite restricting class
Q of sequences of strings.

The purpose of each of these conditions is broadly as follows: C1 specifies the

elements that can be contained in the domain and range of a given transforma-
tion; C2 ensures that only a single-valued mapping from a given ordered pair
(i.e., (Z , K )) in the domain to a unique string in the range (i.e., in the set P̄)
is allowed; and C3 ensures that each transformation applies only to a particu-
lar subsection of the potential domain. In summary, then, a given grammatical
transformation, T , operates upon a string, Z , which has constituent interpreta-
tion K (which may or may not be a P-marker), and it converts it into a string,
Z  , which has interpretation K  . The transformation T is restricted so that it
only operates upon a subset of the possible sets of pairs (Z , K ), and this subset
is determined by the restricting class Q.
Turning from abstract theory to the actual mire of English, in chapter 10
of LSLT Chomsky attempts to demonstrate that transformations can facilitate
the analysis of English grammar. In particular, he identifies certain ‘elementary
transformations’ (Chomsky 1975[1955]: 404) which involve specific modifica-
tions to kernel strings. For instance, the deformation transformation, δ, either
deletes or adds a constant string to certain terms in a given string. Similarly,
the permutation transformation, π, simply alters the order of the terms in a
given string. In order to clarify how these elementary transformations operate,
consider the case of π (and compare the discussion in Chomsky 1975[1955]:
166 Transforming generative grammar: 1955–1957

404–405). Given a string Y1 − . . . − Yn , the transformation π can be repre-

sented as
π  (Y1 , . . . , Yn ) = Ya1  . . .  Yan (5.11)
where (a1 , . . . , an ) is a permutation of the integers (1, . . . , n). In addition, for
each i, the following equivalence holds
(Y1 , . . . , Yi ; Yi , . . . , Yn ) = Yai (5.12)
Consequently, π is completely characterised by the sequence of integers
B = (a1 , . . . , an ). To give a concrete example, if Y1 − Y2 − Y3 indicates the
morpheme sequence ‘he - can - come’, then the permutation transformation
π  (Y1 , Y2 , Y3 ) = Y2  Y1  Y3 enables the string ‘can - he - come’ to be
obtained. As this example indicates, the transformations in TGG essentially
map P-markers into P-markers in a principled manner, and, as is well known,
in his subsequent discussion Chomsky develops specific transformations that
facilitate the analysis of a wide range of basic structures in English (e.g., yes-
or-no questions, wh-questions, passives and so on).
Although transformations have been discussed primarily in an abstract fash-
ion above, and a single example of a permutation transformation has been
considered, it should be noted that the passive transformation receives consid-
erable attention both in LSLT and SS, where it is discussed in relation to the
phrase structure kernel.11 While the basic difference between Chomsky’s trans-
formation rules and Harris’ equivalence statements was discussed above, it is
worth mentioning that during the years 1955–1957 Chomsky was keen to stress
the difference between his approach and that of Harris; and it is particularly
important to emphasise this since Harris had also used his transformations to
effect an analysis of active–passive sentence pairs. As an example of Chom-
sky’s desire to indicate the difference between his approach and that adopted by
Harris, in SS he explicitly observes that Harris’ use of transformation rules con-
stituted ‘a somewhat different approach to transformational analysis’ (Chomsky
1957b: 44), though he refrains from exhaustively enumerating the differences
between the two theories.
While assessing the deployment of transformations in TGG, and given the
discussion of simplicity criteria in section 4.3, it is necessary to emphasise that,
in LSLT, Chomsky repeatedly stresses the fact that a grammar with a trans-
formational component is inherently simpler than a grammar that lacks such a
component. In other words, as far as Chomsky is concerned, considerations of
simplicity partly motivate the use of transformational rules. For instance, at the
start of his formal development of transformational analysis, Chomsky states:
 Transforming transformations 167

I will try to show that a theory of transformations can provide a unified and quite
natural approach to all the problems mentioned in [earlier sections], and that
it can result in syntactic description which is considerably more economical
and revealing. (Chomsky 1975[1955]: 307)

As ever, the key word here is ‘economical’: in the event, Chomsky is keen
to demonstrate that the use of transformations greatly simplifies the proposed
grammar since it enables all types of recursion to be eliminated from the phrase
structure component. This important topic is discussed in greater detail in
section 5.6.
As the foregoing discussion indicates, the confusion concerning the nature
and purpose of grammatical transformations was widespread, and the relation-
ship between the transformations employed by Harris, Chomsky, Harwood, and
others, and those utilised by Carnap and his fellow logicians provided the main
focus for discussions of such topics throughout the 1950s. By the early 1960s
the situation seems to have become slightly clearer, mainly because Chomsky’s
detailed grammatical formalisms provided secure definitions of the terms as
used in the mature theory of TGG. For instance, writing in 1963, Bar-Hillel
indicated that, while he was personally convinced of the importance of TGG-
style transformation rules, he felt obliged to re-emphasise the fact that they were
not ‘transformation’ rules in the Carnapian sense, but rather ‘formation’ rules
(again in the Carnapian sense). Indeed, as far as Bar-Hillel was concerned, they
were a new kind of formation rule which permitted the analysis and descrip-
tion of ‘many more linguistic facts than I had originally thought’ (Bar-Hillel
1963: 542). However, as a result of the terminological confusion, Bar-Hillel felt
obliged to offer the following words of caution:

those formation rules that can handle these facts are of a type which has only
recently been analysed and understood . . . It is interesting that this novel
type of formation rules has been called by Harris and Chomsky (and is now
being called by everyone else) ‘rules of transformation’, and the reader should
beware of the confusion. (Bar-Hillel 1963: 543)

Confusion is perhaps an understatement. Throughout the period 1951–1957 the

word ‘transformation’ is used in a wide variety of ways, and it is not always a
simple task to determine the precise connotations that the term was intended
to imply. Somehow, from out of this general haze, TGG emerged in the late
1950s as a coherent, identifiable theory which used well-defined rules that
were called ‘transformations’. As the years passed, the relationship between
these rules and those proposed by Carnap became a less pressing issue, and the
original terminological complexities were forgotten. The purpose of the above
168 Transforming generative grammar: 1955–1957

discussion, however, has been partly to revive something of the original chaos.
To summarise therefore, it seems likely that the transformations developed
by Harris and Chomsky were related to the transformation rules presented in
(the English translation of) Carnap’s LSL. However, since Harris’ grammatical
equivalences and Chomsky’s transformations are not obviously rules of valid
inference, they must be viewed as formation rules from a Carnapian perspective
(the perspective consistently adopted by Bar-Hillel). However, as Harwood’s
work shows, during the 1950s transformation rules came generally to refer
to rules that operated upon complete sentences that had been generated by
other rules, with the transformations causing modifications to the order of the
elements in the existing sentences. Obviously, it was this kind of transformation
that came to characterise Chomsky’s work in the years 1955–1957 (and beyond).

5.6 Recursive rules

As shown in the previous section, the origins of the technical devices that
came to be referred to as ‘transformations’, and which played such a crucial
role in the development of TGG, can be traced back to advances in logical
syntax that were largely due to Carnap, who, in turn, had been profoundly
influenced by (Hilbertian) Formalism. Appropriately, the same is also true of
the ‘generative’ aspects of TGG. It was mentioned in section 3.3 that recursive
functions became central to Hilbert’s proof theory when it was recognised
that they were essential elements in finitistic proofs, and that research into
the nature of recursion resulted in the general theory of recursive functions,
with associated developments such as the theory of recursively enumerable
sets. Similarly, in the same section, it was shown that Kleene’s discussion of
recursive definitions was adapted by Bar-Hillel and considered in the context
of grammatical analysis in the early 1950s. The main consequence of this work
was that recursive definitions came to be explored within the broad framework
of syntactic theory, and, eventually, they came to provide a core component
of the TGG formalism. The course of this gradual development must now be
considered.
Not surprisingly, recursion is a crucial factor in TGG for the same reason it
is a crucial factor in finitistic proofs; namely, it provides a way of analysing the
(potentially) infinite in terms of the finite. Specifically, during the 1950s syn-
tacticians began to re-emphasise the fact that, although the grammar of a given
natural language must itself necessarily be finite, it must also be able to produce
a potentially infinite number of sentences, and this observation prompted the
basic conceptual shift away from post-Bloomfieldian corpora-based analytical
 Recursive rules 169

discovery procedures that iteratively decomposed given linguistic complexes

into their constituent parts, towards a synthetic or ‘generative’ grammar that
iteratively produced all the grammatical sentences of a given language using
a given lexicon, a finite set of initial assumptions, and certain formal devices.
Consequently, the problem for syntactic theory became one of constructing a
potentially infinite number of sentences using finite procedures and, clearly, the
theory of recursive functions offered a simple mechanism for achieving this. It is
important to realise, however, that the post-Bloomfieldian tradition prepared the
way for such a development by emphasising the predictive nature of the gram-
mars obtained from analytical procedures. In other words, discovery procedures
were often only considered valid if the resultant grammar could be successfully
transferred to new corpora, and Hockett, in particular, emphasised this require-
ment.12 Once again, therefore, from this perspective, TGG can be viewed as
the logical outcome of one particular branch of the post-Bloomfieldian research
programme. Accordingly, in ‘TMDL’ Chomsky emphasised the desirability of
predictive formal grammars, and stated plainly that a TGG-style approach to
syntactic analysis requires some kind of recursive component in order to con-
struct the full range of possible grammatical sentences: ‘If a grammar has no
recursive steps . . . it will be prohibitively complex . . . If it does have recursive
devices, it will produce infinitely many sentences’ (Chomsky 1956: 116). In
particular, Post’s theory of recursively enumerable sets, described in section 3.3,
provided a useful framework that could be adapted for the purposes of theo-
retical syntax. As mentioned previously, the intuition behind Post’s theory was
that all the elements belonging to a recursively enumerable set are in the range
of a general recursive function, and, in a 1959 paper, Chomsky explained why
such sets were important to the basic TGG framework.

Since any language in which we are likely to be interested is an infinite set, we

can investigate the structure of L only through the study of the finite devices
(grammars) which are capable of enumerating its sentences. A grammar of L
can be regarded as a function whose range is exactly L. Such devices have
been called ‘sentence-generating grammars’. (Chomsky 1959a: 137)

This passage indicates that, in mature TGG, the set of grammatical sentences,
L, is considered to be an infinite set of sentences, the elements of which can be
generated by the function, g, where g constitutes the grammar of the language
L. It is worth noting that (as indicated in section 3.3) even the terminology used
here reveals the connection between the theories: Post frequently used the verb
‘generate’ in order to describe how a recursively enumerable set is obtained from
its associated general recursive function, and Chomsky explicitly acknowledges
170 Transforming generative grammar: 1955–1957

this terminological source, suggesting that he adapted these techniques directly

from Post’s work. In a footnote following his use of the phrase ‘sentence-
generating grammars’, Chomsky adds:

Following a familiar technical use of the term ‘generate’ cf. Post (1944).
The locution has, however, been misleading, since it has erroneously been
interpreted as indicating that such sentence-generating grammars consider
language from the point of view of the speaker rather than the hearer. Actually
such grammars take a completely neutral point of view. Compare Chomsky
([1957b], p.48). We can consider a grammar of L to be a function mapping the
integers onto L, order of enumeration being immaterial (and easily specifiable,
in many ways) to this purely syntactic study . . . (Chomsky 1959a: 137.n1)

This passage indicates that Post’s work concerning recursively enumerable sets
was ‘familiar’ to Chomsky, and that the latter knowingly adapted the terminol-
ogy that the former had deployed. This fact should be borne in mind throughout
the following discussion since it is crucial to recognise that Chomsky was famil-
iar with Post’s own research and did not simply obtain his ideas concerning
recursively enumerable sets from later expositions of the theory.
The above discussion indicates that TGG, as it was presented in the 1950s,
required ‘recursive devices’ (Chomsky 1956: 116) in order to generate an infi-
nite number of sentences from a finite grammar, and, although Post provided
some kind of impetus for this assumption, Chomsky’s ideas concerning these
topics were also directly influenced by the work of Bar-Hillel. For instance, as
noted previously, Chomsky had remarked in ‘LSS’ that

At one point, Bar-Hillel suggests that recursive definitions may be useful in

linguistic theory; whether this turns out to be the case or not, I agree in this
instance with the spirit of his remarks. (Chomsky 1955a: 45)

It is no surprise, then, that recursive techniques were used extensively in TGG,

and the recursive aspects of the theory are meticulously presented in LSLT. In
order to clarify the following discussion, two different types of recursion in
TGG will be considered. The first type involves the successive application of
the sequentially ordered rules of the grammar, while the second type involves
the inclusion of rules within the grammar which are themselves recursive by
definition. Both types will now be assessed in turn.
In chapter 7 of LSLT Chomsky elaborates his basic theory of the phrase struc-
ture component of the grammar; in particular, he proposes that this component
can be specified as a sequence of conversions of the familiar form ‘X → Y ’,
and he refers to this type of formulation as ‘a linear grammar’ (Chomsky
 Recursive rules 171

1975[1955]: 194). However, Chomsky is quick to note that the finite nature
of a linear grammar requires the sequence of rules to be applied iteratively.

The linear grammar is a sequence of conversion statements S1 , . . . , Sn , where

each Si is of the form X i → Yi . We can produce derivations from this lin-
ear grammar by applying the conversions Si (interpreted as the instruction
‘rewrite X i as Yi ’) in sequence. Among the Si we distinguish between obliga-
tory conversions that must be applied in the production of every derivation, and
permissible conversions that may or may not be applied. There are only a finite
number of ways to run through the linear grammar, applying all obligatory
and some permissible conversions; hence only a finite number of derivations
can be produced by the linear grammar Si , . . . , Sn . This was not a difficulty
on earlier levels [e.g., the phonemic level, the syntactic categories level, etc.],
but we know that infinitely many sentences must be generated by some mech-
anism in the grammar. We can permit this infinite generation on the level P by
allowing the possibility of running through the linear grammar S1 , . . . , Sn an
indefinite number of times in the production of derivations. If the derivation
formed by running through the sequence of conversions does not terminate
with a string in P [the set of strings composed of morphological heads and
syntactically functioning affixes], then we run through the sequence again.
Thus we can understand the linear grammar to be the sequence of conver-
sions S1 , . . . , Sn , S1 , . . . , Sn , S1 , . . . , Sn , . . . . We then say that a derivation
D is recursively produced by the linear grammar S1 , . . . , Sn . We define a
proper linear grammar as a linear grammar which is so constructed that it
is impossible to run through it over and over again vacuously. (Chomsky
1975[1955]: 194–195)

This passage indicates that the iterative application of the conversion statements
in the linear grammar was viewed (by Chomsky) as a recursive procedure, and
this type of recursion was explicitly introduced in order to enable an infinite
number of sentences to be produced by the grammar. In the light of the dis-
cussion of simplicity criteria in section 4.3, it is worth considering the recur-
sive application of conversion statements from a grammar-internal perspective.
Specifically, it is important to note that, for Chomsky, the issue of recursive rule
application was closely connected to the issue of determining the simplicity of
the resultant grammar, and the following passage is crucial:

The recursive extension which permits infinite production of derivation might

pose a problem for the determination of simplicity. Thus we may ask whether
a grammar is simpler if it is necessary to run through it fewer times to set up
derivations. There are several possible ways to place conditions on grammars
that will eliminate a conflict between this consideration and the criterion of
simplicity already established. (Chomsky 1975[1955]: 200–2001)
172 Transforming generative grammar: 1955–1957

This passage is of interest because it indicates that, while constructing TGG,

Chomsky was forced to struggle with various technical problems that resulted
from his attempts to adapt both simplicity criteria and recursion for his current
purposes. However, Chomsky does not develop in detail the particular ‘consid-
erations’ that could eliminate the apparent conflict, for, in chapter 10 of LSLT,
he proposes that all recursive procedures should be removed entirely from the
phrase structure part of the grammar, a proposal that is considered in detail
below.
As indicated previously, if the repeated application of conversions constitutes
one type of recursive procedure introduced in LSLT, another type manifests
itself in certain rules contained within the phrase structure part of the grammar.
Initially (as mentioned above), this kind of recursion was introduced in the form
of rewrite rules such as ‘X → Y X ’, where the category to the left of the arrow,
‘X ’, also appears on the right-hand side of the rule, ‘Y X ’. Chomsky’s initial
presentation of the topic at the start of chapter 7 of LSLT clarifies the manner
in which these rules were used in TGG:
When we turn to the level of phrase structure, we find that certain rules may
have a recursive character. Thus Noun Phrase (NP) might be analyzed in such
a way that one of its components may be a NP as in such sentences as ‘the man
who made the discovery is my brother’, which might be derived by means of
a set of such conversions as

(i) NP → NP1  who  VP

VP → V  NP3
Conversions of this sort will permit the generation of infinitely many sentences
by that part of the grammar that deals with phrase structure . . . (Chomsky
1975[1955]: 171–172)

As should be obvious, the formalism that Chomsky gives here is (to use Bar-
Hillel’s term) ‘recursive in disguise’ (Bar-Hillel 1953: 163) since he does not
provide the definitions of NP that are required in order to terminate the recursion,
yet these are provided in the main discussion of the topic that follows later in
the chapter. The basic point is clear, though, nonetheless: the phrase structure
component of the type of grammar developed in LSLT contains recursive rules.
In the above example, the rule indicates that NPs can be rewritten as larger
structures (containing relative clauses) which in turn contain NPs, and so on ad
infinitum, at least in theory if not in practice. The rest of chapter 7 constitutes an
exploration of the abstract theory underlying level P (i.e., the phrase structure
level of the grammar), and when Chomsky actually begins to develop this
component for English in chapter 8, he introduces a number of recursive rewrite
 Recursive rules 173

rules. Sometimes these take the form of potentially infinite expansions. For
example, when discussing the structure of verb phrases, Chomsky proposes the
rules

Sentence → N P V P
V P → V PA  V P1
V P1 → < V PB > V P2
V PA → V PA1 < V PA2 >
V PA1 → {C, ed} < M >
V PA2 → < have  en >< be  ing >
V PB → Z 1 < Z 2 < . . . < Z n >>>, where each Z i is one of the forms
Vc  to, Vγ  ing
(Chomsky 1975[1955]: 249–250)

where V PA and V PB both enable certain kinds of auxiliary verb constructions

to be generated, with the difference that V PB is defined so as to permit an
unlimited number of Z i s. In order to illustrate how this particular group of rules
works, the sentence ‘John wants to read the book’ can be obtained from the
given conversions as follows:

Sentence
NP  VP
N P  V PA  V P1
N P  V PA  V PB  V P2
N P  C  V PB  V P2
N P  C  want  to  V P2
J ohn  C  want  to  r ead  the  book
(Chomsky 1975[1955]: 250)

As these examples indicate, the recursive rules were initially introduced into
the phrase structure component of TGG; that is, they appeared in the part of
the grammar that was responsible for generating the kernel of basic sentences,
enabling infinite structures to be generated. However, as Chomsky begins to
elaborate his theory of transformations in chapters 9 and 10 of LSLT, he is
compelled to reduce the recursive elements in the phrase structure component.
More specifically, as the theory develops, the recursive rules are moved out of
the phrase structure component entirely, and are placed in the transformational
component of the grammar instead. In addition, Chomsky suggests that the first
type of recursion discussed above, which permits the conversion statements to
be applied recursively in sequence, should also be removed. Ultimately, these
developments led Chomsky to make the following claim towards the end of
chapter 10:
174 Transforming generative grammar: 1955–1957

In the course of this analysis we have found that much of the recursive part of
the grammar of phrase structure . . . has been cut away. It seems reasonable
to place the formal requirement that no recursions appear in the kernel gram-
mar. Specifically we rule out such statements as [V PB → Z 1 < Z 2 < . . . <
Z n >>>], and we drop the constructions . . . that permit running through
the grammar indefinitely many times. As far as I can determine, this formal
requirement on P does not exclude anything that we would like to retain in P;
nor does it impose any artificial or clumsy limitation on the actual statement
of the grammar corresponding to P, now that the transformational analysis
presents an alternative way of generating sentences . . . Now that the higher
level of transformational analysis has been established, it is no longer neces-
sary to require that generation by the grammar of phrase structure is infinite.
As the level T has been formulated, the process of transformational derivation
is recursive, since the product of a T-marker can itself appear in the P-basis
of a T-marker. (Chomsky 1975[1955]: 516–518)

The requirement that the phrase structure component should not contain recur-
sive elements ensures that the recursive parts of the grammar must all be imple-
mented in the transformational component; a development which, at the time,
no doubt emphasised the potential importance of Chomskyan transformations.
Nevertheless, as the above extract indicates, whether they are situated in the
phrase structure kernel or in the transformational component, TGG requires
recursive rules since such rules enable a grammar, g, to generate all the sen-
tences of the recursively enumerable language L, and, as suggested above, such
rules clearly have their roots in recursive function theory.

5.7 Formal syntax

The preceding sections of this chapter have considered various aspects of TGG
that can be associated with specific techniques used in the formal sciences.
In this final section the general correspondences between TGG and Formal-
ism will be discussed. Section 3.7 attempted to indicate how key concepts
of the (caricatured) Formalist creed were disseminated amongst mathemati-
cally inclined linguists during the first decades of the twentieth century, and
there is no doubt that many of the characteristic features of TGG were derived
from the type of sources mentioned in section 2.8, since specific references to
named texts (such as Quine’s Mathematical Logic and Rosenbloom’s Elements
of Mathematical Logic) are frequent in Chomsky’s publications from the years
1953–1957. To some extent the discussion in sections 5.4–5.6 has anticipated
the following analysis, since it has already been established that specific meth-
ods strongly associated with proof theory, such as the use of recursive rules,
 Formal syntax 175

were incorporated into TGG at a comparatively early stage. However, the full
influence of Formalism upon the theory has yet to be illustrated.
As discussed at length in section 3.7, the need to formalise linguistic theory
became a pressing concern for particular groups of researchers in the late 1940s,
and this task had been partially accomplished by certain post-Bloomfieldians
(especially Bloch, Wells, Chao, Hockett, and Harris) by the mid-1950s. As
indicated previously, Harris was aware of contemporaneous developments in
logic and metamathematics and therefore it is no surprise that he encouraged
the young Chomsky to attend lectures on mathematics and philosophy. Once
again, therefore, although the following passage from Chomsky’s 1975
recollections was quoted in the introduction, it merits being reproduced here
since it provides a fascinating insight into the eclectic intellectual climate at
MIT in the early 1950s.13

Perhaps a word might be usefully added on the general intellectual climate

in Cambridge at the time when it [i.e., LSLT] was written. Interdisciplinary
approaches to language communication and human behavior were much in
vogue . . . Roman Jakobson’s work was well known and influential. Oxford
ordinary language analysis and Wittgenstein’s later work were attracting
great interest. The problem of reconciling these approaches (if possible) with
Quine’s provocative ideas on language and knowledge troubled many stu-
dents. Mathematical logic, in particular recursive function theory and meta-
mathematics, were becoming more generally accessible, and developments
in these areas seemed to provide tools for a more precise study of natural
language as well. All of this I personally found most stimulating. (Chomsky
1975[1955]: 39)

The explicit references here to ‘mathematical logic’, ‘recursive function the-

ory’ and ‘metamathematics’ clearly indicate that, as shown in chapter 3, during
this period linguistics and mathematics were sufficiently closely connected to
enable developments in logic and metamathematics to become a potent source
of inspiration for young linguists, although, as indicated in section 4.5 and else-
where, the precise nature of the relationship between logic and natural language
was a topic that was frequently debated. However, even if, like all recollections,
the above comments are not to be trusted entirely, there is plenty of contempo-
raneous evidence which suggests that Chomsky inherited from his immediate
post-Bloomfieldian predecessors the belief that techniques adapted from meta-
mathematics could provide useful tools for syntactic theory. As mentioned
previously, with particular reference to Chomsky’s own work, it is important to
emphasise the fact that the very title of his weighty LSLT manuscript (i.e., ‘The
Logical Structure of Linguistic Theory’) appeared explicitly to associate his
176 Transforming generative grammar: 1955–1957

research with Carnap’s research into logical syntax (i.e., in English translation,
Carnap’s The Logical Structure of Language), the main difference being that,
while Carnap had signalled his concern with the logical structure of language,
Chomsky was keen to stress his interest in the logical structure of linguistic
theories devised to analyse natural language. It is crucial to note the difference
in emphasis here: Chomsky’s title implies that, while it may not be possi-
ble to demonstrate that the structure of natural language can be reduced to
logic, it is certainly possible to ensure that the structure of a linguistic theory
is elaborated in a logical manner. Clearly, though, despite this difference in
emphasis, Chomsky seems to have been happy to associate himself (to some
extent) explicitly with the logico-philosophical Formalist tradition (as manifest
in Carnap at least), even if the association is intended primarily to illustrate
the distinctive character of his own approach; and, as the previous sections
have shown, such an association was a natural extension of the more formal
approach to the methodology of linguistics advocated by certain subsets of the
post-Bloomfieldians.
Given that developments in metamathematics were accessible to linguists in
the 1950s, and given that certain post-Bloomfieldians had sought to develop a
more systematic approach to linguistics by adopting a more (meta)mathematical
methodology, it is no surprise that, when TGG began to coalesce in the
mid-1950s, considerable emphasis was placed upon its status as a formal
theory. Indeed, Chomsky addresses this very issue in chapter 2 of LSLT,
and the Formalist nature of the project he is in the process of outlining is
clear.

In the strict sense of the word, an argument, a characterization, a theory,

etc. is ‘formal’ if it deals with form as opposed to meaning, that is, if it
deals solely with the shape and arrangement of symbols. In this sense, any
distributional theory or argument is formal. But the word ‘formal’ has mis-
leading connotations, implying ‘rigorous’, ‘clear’, etc. Suppose that we use
the word ‘formalized’ when we have the latter sense in mind. A formal-
ized theory, then, is one that is formulated in accordance with clear canons
of rigor and precision: definitions are given explicitly in such a way that
defined terms are always eliminable, and the axioms and methods of proof
are precisely stated. We can thus have a formalized theory purporting to be
about form or about meaning. A formalized theory is, of course, not neces-
sarily an acceptable or enlightening theory. And the fact that a certain subject
matter is stated to constitute the intended interpretation of a presented for-
malized theory does not confer any desirable properties upon this subject
matter, or upon the theory. It is possible to construct a formalized theory with
no interesting interpretation, or with intuition, ghosts, etc., in its intended
 Formal syntax 177

interpretation. What concerns us here is the possibility of a formalized theory

of linguistic form, and the problems involved in constructing such a theory.
(Chomsky 1975[1955]: 83)

This passage (or at least sections of it) could have been taken from any of the
textbooks discussed in section 2.8, and clearly the basic approach to syntactic
theory proposed here is entirely in keeping with Bloomfield’s recommended
axiomatic-deductive approach to theoretical linguistics. Indeed, the foregoing
sections of this book should already have illuminated the way in which the
above passage is richly embedded in a specific cultural tradition – namely, the
drive towards greater ‘rigour’ that was revived by Cauchy in the nineteenth
century in order to secure the foundations of the calculus, but which spread
to many other scientific disciplines during the twentieth century, largely due
to the influence of Logicism and Formalism. Before continuing, though, it is
necessary to emphasise Chomsky’s attitude towards the task of formalising a
linguistic theory.
As the above passage indicates, for Chomsky, formalisation for the sake of
formalisation is not the point: any theory can be converted into a ‘formalized’
theory, but this does not guarantee that the theory is ‘enlightening’. Conse-
quently, the motivation for propounding a formal linguistic theory is not to
spare the linguist the arduous task of analysing complicated and confusing data;
rather, the process of formalisation is desirable primarily because it enables spe-
cific conjectures concerning particular data to be stated more precisely. Indeed,
for Chomsky, formalisation was never a soulless activity that could be accom-
plished only when a theory was already complete; rather, it was a fundamental
aspect of the process of theory development. Indeed, as he observes elsewhere in
LSLT, ‘formalization can play a very productive role in the process of discovery
itself’ (Chomsky 1975[1955]: 59).
In the passage concerning formalisation quoted at length above, Chomsky
refers only in very general terms to the use of ‘axioms and methods of proof’.
However, the axiomatic nature of TGG is clear from the various presentations of
the basic theory that appeared in the mid-1950s. Once again, though, it is neces-
sary to make a basic distinction between the way in which Chomsky presented
his expositions of the basic theory, and the way in which language was analysed
within that theory. For example, the main chapters of LSLT conspicuously reveal
the axiomatic-deductive nature of Chomsky’s exposition. To take one example
virtually at random, when Chomsky introduces the ‘relation of representation’,
ρ, which he uses to specify relations such as ‘ρ(N P, the  man)’, indicating
that ‘the  man’ is an N P, he provides specific axioms and definitions, and
178 Transforming generative grammar: 1955–1957

then proceeds to deduce theorems from these. For instance, the axioms and
definitions take forms such as14

Axiom 1: ρ is irreflexive, asymmetrical, transitive and nonconnected.

Definition 2: P is the set of strings X such that for Y , ρ(X, Y ).

and the theorems are constructed using these axioms and definitions in the
standard axiomatic-deductive fashion. As is obvious even from these two
examples, unlike Bloomfield and Bloch (discussed in section 3.2), Chomsky
was not content solely with natural language descriptions of his various
technical entities and procedures; consequently, he introduces a rich symbolic
metalanguage that enables the various axioms and definitions to be presented
as precisely as possible.
While the presentation of TGG as a linguistic theory evinces a predilection
for the axiomatic-deductive method, it is important to emphasise the fact that
the basic sentence construction mechanism, as viewed from the perspective of
TGG, can also be classified as an axiomatic-deductive procedure. To consider
just one presentation, the rudiments of the theory were discussed explicitly
in ‘TMDL’, and, in this presentation, Chomsky seems to promote a distinctly
axiomatic-deductive interpretation of his theory since he describes the core
components of the phrase structure component as follows:

A phrase-structure grammar is defined by a finite vocabulary (alphabet) V p ,

a finite set  of initial strings in V p , and a finite set F of rules of the form:
X → Y , where X and Y are strings in V p . (Chomsky 1956: 117)

This passage recalls Hilbert’s specifications for metamathematical formal sys-

tems within the proof theory tradition, and, accordingly, it implies that the
phrase structure grammar component can be viewed as a particular kind of for-
mal system. This interpretation is implied by the description of the grammar as
consisting of a finite alphabet of primitive elements, V p , which are combined
using rules of the form, X → Y , in order to create well-formed formulae (i.e.,
grammatical sentences). For example, a simple phrase structure grammar of
this type could take the form (compare Chomsky 1957b: 26):

(i) Sentence → NP + VP
(ii) NP → T + N
(iii) VP → Verb + NP
(iv) T → the
(v) N → man, ball, etc.
(vi) Verb → hit, run, etc.
 Formal syntax 179

and, in this example, every well-formed formula (i.e., grammatical sentence) is

ultimately derived by means of rules (i.e., conversions) from the initial ‘Sen-
tence’ symbol. This suggests that Chomsky initially considered the task of
generating grammatical sentences to be comparable to an exercise in finitistic
proof theory, and the use of terminology such as ‘well-formed formulae’, which
was directly borrowed from the metamathematics literature and used in early
TGG papers, underlines the extent of the correspondence. Just in case there
could be any doubt concerning the analogy between sentence generation and
proof construction, Chomsky himself explicitly acknowledged this connection
on several occasions in the mid-1950s. For instance, when discussing the way in
which sentences are derived in a formal TGG-style grammatical framework, he
stated in LSLT that the process of derivation is ‘roughly analogous to a proof’
(Chomsky 1975[1955]: 67), and this seemingly innocent analogy between the
derivation of a grammatical syntactic structure and the construction of proof in
a metamathematical framework became one of the more characteristic features
of TGG-style syntactic theory.
As indicated above, in the main expositions written during the years 1955–
1957, TGG-style grammars were presented as if they were formal systems, and
the process of sentence generation was deemed to be similar to the task of proof-
theoretical deduction. These assumptions had profound consequences for the
type of syntactic theory that TGG advocated. For instance, as indicated earlier,
the role of meaning in Hilbertian proof theory was necessarily reduced to a
minimum (indeed, it was entirely negated by the extremists) since the construc-
tion of a valid proof was considered to be a meaning-less task accomplished by
a sequence of purely syntactic manipulations. The advantage of this approach
was that it provided a general axiomatic-deductive framework that enabled valid
derivations to be obtained within any formal system irrespective of the specific
domain of application. In other words, the abstract generality of the theory
appeared to eliminate relativism. Therefore, given the explicit analogy between
TGG-style syntactic analysis and proof theory, it is not surprising that, within
the broad TGG framework (as in the Bloomfield–Harris tradition) the role of
semantics was minimised, while abstract syntactic structure was considered to
be of fundamental importance. For instance, in the preface to LSLT (which was
added to LSLT after SS had been written), Chomsky specifically discussed the
distinction between syntax and semantics.

Linguistic theory has two major subdivisions, syntax and semantics. Syntax
is the study of linguistic form . . . Semantics, on the other hand, is concerned
with meaning and reference of linguistic expressions . . . Syntax and semantics
180 Transforming generative grammar: 1955–1957

are distinct fields of investigation. How much each draws from the other is not
known, or at least has never been clearly stated. The subject of investigation
in the following pages will be syntactic structure, and we shall study it as
an independent aspect of linguistic theory . . . In part, our desire to place
no reliance on meaning in systematic development is motivated by a feeling
that the theory of meaning fails to meet certain minimum requirements of
objectivity and operational verifiability. (Chomsky 1975[1955]: 57)

None of the above is startling given the aforementioned Formalist slant of

TGG, and, as indicated on numerous occasions already, this preoccupation with
‘form’ rather than ‘meaning’ had been a characteristic feature of certain kinds of
linguistic research since Bloomfield (at least). Therefore it is no surprise that a
comparable tendency manifests itself in the type of syntactic theory outlined in
LSLT and SS, where grammatical sentences are exhaustively defined in terms
of their formal structure without reference to semantics. Consequently, it is
possible to view the explicit association between TGG and proof theory as yet
another manifestation of the Formalist tendency of syntactic theory during the
first half of the twentieth century, which resulted in the separation of syntax
from semantics.
In the light of the above, it is worth considering the arguments Chomsky
used at the time in order to justify the purely syntactic definition of grammat-
icality that his research espoused. Standardly, his basic assumption appears to
have been that grammaticality, which is repeatedly associated with the notion
of ‘well-formedness’, can be completely determined without any reference to
meaning:

the notion ‘grammatical’ cannot be identified with ‘meaningful’ or ‘signifi-

cant’ in any semantic sense. Sentences (1) and (2) are equally nonsensical, but
any speaker of English will recognise that only the former is grammatical.
(1) Colorless green ideas sleep furiously.
(2) Furiously sleep ideas green colorless.
Such examples suggest that any search for a semantically based definition of
‘grammaticalness’ will be futile. (Chomsky 1957b: 15)

This argument essentially restates that already presented in ‘TMDL’, and, once
again, the influence of the logico-philosophical Formalist tradition is clear. For
instance, Chomsky’s celebrated examples and general argument in the above
passage seem consciously to echo those introduced by Carnap in his LSL. As
mentioned in section 3.7, Carnap constructed the sentence ‘Pirots karulize elati-
cally’ in order to demonstrate that well-formedness could be analysed in purely
syntactic terms without reference to semantic considerations, and, obviously
 Formal syntax 181

Chomsky is employing a similar strategy in order to make the same basic point.
Indeed, it is worth noting that Carnap’s example is actually more extreme than
Chomsky’s since the use of neologisms complicates the situation even more
from a semantic perspective, but there is no doubt that the main thrust of both
arguments is essentially the same.
It is clear that the type of argument summarised above is understood to apply
to syntactic research at a very general level; that is, the main recommendation
is that any sensible and self-respecting syntactic theory will distance itself
entirely (if possible) from considerations of meaning. However, while it is
sufficiently important to consider this kind of general recommendation, it is
equally revealing to explore the way in which Chomsky handles the problem of
meaning in his detailed developments of TGG. To give just one example of this,
in LSLT, when he is in the process of motivating the use of transformations in
addition to the conversions in the phrase structure component of the grammar,
Chomsky states:

We do not want a transformation to depend on the ‘content’ of the particular

strings into which Z [i.e., a string produced by the kernel] is decomposed for
the purposes of transformation, but only on the number and order of these
substrings. Only such transformations will reflect general structural relations
between classes and strings. Once we have settled on some fixed way of charac-
terizing grammatical transformations in terms of certain structural properties,
it will be possible to take the complexity of this fixed form of characteriza-
tion into account as a feature in the evaluation of given grammars. (Chomsky
1975[1955]: 315)

The emphasis of this passage should already be familiar given the previous
discussion: Chomsky wishes to ensure that the transformations do not need
to take semantic information into account, and that they should consider only
the ‘number and order’ of substrings, since only a system of this kind will
enable the transformations to reflect ‘general structural relations’. The impli-
cation of this (and similar passages scattered throughout LSLT and SS) is clear:
Chomsky assumed that syntactic form could be exhaustively analysed inde-
pendently of meaning and, in assuming this, the type of theory he proposed
in the mid-1950s reveals something of its Formalist inspiration. At the very
least, it demonstrates how far syntactic theory had come since Hjelmslev dis-
tinguished between meaning-less metamathematics and meaning-full natural
language analysis in the 1940s. Indeed, Chomsky’s position is more extreme
even than that of second-generation Formalists, such as Carnap in the 1930s,
since, as indicated above, Carnap maintained that, due to their complexity, nat-
ural languages could not be analysed adequately using techniques that were
182 Transforming generative grammar: 1955–1957

devised to analyse formal languages, while, by the mid-1950s Chomsky had

begun to challenge such views by adapting various techniques from mathemat-
ics, logic, and metamathematics, and deploying them for the purposes of natural
language analysis. In order to accomplish this, it was necessary to assume that
natural languages and formal languages were amenable to similar analytical
techniques. Given the apparent diversity and idiosyncrasy of human language,
this assumption necessitates the development of an abstract syntactic theory
that seeks to identify underlying regularities lurking beneath apparent surface
irregularities, and a research programme of this type has largely determined the
course of generative grammar from the 1950s to the present day.
6 Conclusion

As indicated in the introduction, and as demonstrated in the subsequent chapters,

this book manifestly constitutes an exercise in historiography, and the general
purpose has been to reveal the nature and extent of the influence of the formal
sciences upon the development of linguistic theory in the twentieth century, with
the main emphasis falling upon the advent of TGG. At this stage, therefore, it
is probably worth summarising the main issues considered, in order to review
the central conclusions of the various intertwined investigations presented in
the foregoing chapters.
One of the main tasks attempted in several sections of this book has been the
re-evaluation of certain aspects of Bloomfield’s work in the light of his interest
in the foundations crisis and (specifically) in Formalism, aspects which appear
to anticipate some of the preoccupations that obsessed mathematically minded
linguists in the 1950s. In particular, it was suggested that Bloomfield’s well-
known mistrust of semantic considerations in the study of natural languages
may have been influenced by his knowledge of Hilbertian proof-theoretical
techniques which (at least in their popularised form) recommend the avoidance
of semantic considerations in (meta)mathematical proofs in favour of non-
contentual syntactic manipulations. In general, recent research has neglected
Bloomfield’s knowledge of the foundational debates of the 1920s and 1930s,
as well as his professional interest in the relationship between linguistics and
mathematics. However, if indeed it was the case that part of Bloomfield’s own
research into natural language was influenced by Formalism in the manner pro-
posed in chapter 3, then full recognition of this fact may help to explain why
the distributional procedures that grew out of his research, and which were
developed by the post-Bloomfieldians, were considered to share similar char-
acteristics with various methods employed in the formal sciences in the 1940s
and 1950s. The conclusion offered in this book is that the formal sciences and
post-Bloomfieldian linguistics appeared to approach their different tasks in a
similar fashion partly because the methodologies used in both disciplines were
directly influenced by Formalism during the first half of the twentieth century.

183
184 Conclusion

In addition (and more specifically), Bloomfield was an early proponent (possi-

bly the earliest?) of the use of the axiomatic-deductive method in linguistics, an
approach that was revived first by Bloch in the 1940s, and then by Bar-Hillel,
Harwood, and others in the 1950s, and which gradually became the dominant
method of syntactic analysis after the appearance of SS. Consequently, the anal-
ogy between syntactic theory and proof theory, that was explicitly articulated
by Chomsky in LSLT, has its linguistic roots in Bloomfield’s 1926 set of pos-
tulates for the study of language, and, more generally, in his fascination with
contemporary mathematics.
If one of the accomplishments of the preceding chapters has been the reha-
bilitation of the mathematical aspects of Bloomfield’s linguistic research, in a
similar manner the work of Bar-Hillel may also have been revived. As men-
tioned previously, Chomsky has stated that he considers Bar-Hillel to have
been a ‘constructive participant’ (quoted in Kasher 1991: 6) in the develop-
ment of TGG, and it is hoped that the various explorations of different aspects
of Bar-Hillel’s work that have been presented here have largely validated this
epithet. In particular, it has been shown that Bar-Hillel responded directly to
some of Bloomfield’s ideas concerning the relationship between mathematics
and natural language, and that he was instrumental in exploring the relation-
ship between logic and linguistic theory. His extension of Ajdukiewicz’s work,
which alerted the linguistics community to the existence of the Lvov-Warsaw
school of logicians, and which eventually resulted in the advent of Categorial
Grammar, should be recognised as an impressive early attempt to incorpo-
rate techniques from logic into systems of syntactic analysis. To some extent,
therefore, Bar-Hillel was one of the main researchers who made the fusion of
linguistics and logic appear a real possibility in the early 1950s. Further, with
his appropriation of recursive definitions for the purposes of syntactic analysis,
and with his claim that natural and artificial languages are essentially identical
types of linguistic systems, Bar-Hillel influenced and provoked Chomsky, com-
pelling him to formulate his own ideas more precisely. Sometimes he agreed
with Bar-Hillel (e.g., concerning the use of recursive definitions in syntactic
theory), sometimes he disagreed with him (e.g., concerning the use of tech-
niques from logical semantics in natural language analysis), but, either way,
he invariably felt obliged seriously to consider the implications of the various
ideas that Bar-Hillel introduced. The closeness of their association and the par-
ticular details of their differences have not been fully considered in the past,
with the result that Bar-Hillel has faded rather from the official histories of
TGG. It is hoped that, to some extent, this book has revealed the centrality of
his role.
 Conclusion 185

If Bloomfield and Bar-Hillel have been dusted off a little in this book, then
perhaps a third rehabilitation has also been achieved – or perhaps even a third
and a fourth, depending on whether Goodman and Quine are sufficiently strong
collocators to count as a single entity or not. However that may be, one of the
aims of this research has been to explore the nature of the influence that the work
of Goodman and Quine exerted upon the development of TGG, particularly
since this topic has been so often neglected. As mentioned earlier, this neglect
is rather curious, since Chomsky himself has invited a consideration of these
issues. In the preface to SS, for instance, after dutifully thanking Harris for
his assistance, he adds that ‘in less obvious ways, perhaps, the course of this
research has been influenced strongly by the work of Nelson Goodman and
W. V. Quine’ (Chomsky 1957b: 6). Therefore, to this end, certain sections of
this book have simply explored this influence in greater detail than has been
accomplished in the past, and, as a result, various aspects of TGG have been
(re)assessed. For example, the use of simplicity criteria in Chomsky’s earliest
publications can be viewed as a direct adaptation of Goodman’s research into the
basal simplicity of constructional systems. Further, it has been demonstrated that
‘SSA’, Chomsky’s first published paper, is clearly an attempt to fuse Harris-style
discovery procedures with the particular constructional system that Goodman
had developed in SA. These early (i.e., pre-c. 1954) projects suggest that, at
this stage, Chomsky was still viewing linguistics as an empirical science that
required the elaboration of automatic discovery procedures. However, at some
point during the period 1953–1954 he appears to have rejected some of the
philosophical implications associated with the type of constructive nominalism
he had learnt from Goodman (and, to a lesser extent, Quine), and one of the
many consequences of this rejection was the rejection (in turn) of the use of
discovery procedures in linguistic theory in favour of evaluation procedures
designed to choose between competing grammars (or were the rejections in
the inverse order? or were they effectively simultaneous?). As indicated earlier,
this profound shift in Chomsky’s understanding of the purpose and function
of linguistic methodology has never really been assessed before in the light
of his changing attitudes towards constructive nominalism. Since Chomsky is
universally categorised these days as a self-professed ‘mentalist’, it is sometimes
difficult to accept that the pre-c. 1954 Chomsky seems to have been a logical
empiricist of the Carnap–Goodman variety, but the evidence contained in his
first publications and manuscripts certainly seems to point towards this con-
clusion. However, it should be emphasised that, even if Chomsky’s rejection
of logical empiricism was involved in his rejection of discovery procedures,
the evaluation procedures (i.e., simplicity criteria) that he recommended as
186 Conclusion

a preferable alternative also have their roots in the work of Goodman and
Quine. Indeed, it is a strange fact that, while Chomsky apparently came to
view constructive nominalism as having a deleterious effect upon linguistics,
the same theory nevertheless provided technical procedures that were involved
in his redefinition of the methodology of syntactic analysis. This conclusion
is admittedly rather complex, but it is to be hoped that the complexity offered
here is closer to the truth than the simplifications (or, worse, eerie silences) that
have characterised historiographical studies of the genesis of TGG in the past.
In addition to the three broad topics summarised above, a number of smaller
insights have been gained. For instance, with reference to non-TGG topics,
some forgotten linguists have been resuscitated, most notably Harwood, whose
early attempt to provide a formal axiomatic approach to syntactic theory seems,
unfortunately, to have fallen out of the collective consciousness. With reference
to TGG, though, the influence of White’s critique upon Chomsky’s rejection of
methods from logical semantics has been explored in a preliminary fashion, and
this association has not been fully recognised previously. Indeed, the treatment
of this topic (i.e., the influence of Quine–White style analytic philosophy from
the early 1950s upon TGG) is only sketchily accomplished in this book, and
the White–Chomsky relationship in particular still awaits comprehensive con-
sideration. In addition, there is the influence of Carnap’s research into logical
syntax upon the development of TGG, an association that has been generally
acknowledged in the past, but never discussed in depth. Specifically, in this
book, the difference between Carnapian and Chomskyan transformations has
been investigated, and it is hoped that some of the terminological complexities
have been elucidated. In general, though, perhaps the main achievement of
this book has been to associate TGG with both Formalism and Logicism, two
intellectual movements that profoundly influenced scientific methodology in
the early twentieth century. Indeed, this general issue seems to have been the
single destination towards which the various paths of enquiry have led. With
its focus on syntax as opposed to semantics, with its use of a logic-based
notational system, with its identification of the analogy between a proof and
the generation of a grammatical sentence, and with its use of such procedures
as recursive definitions and axiomatic deduction, TGG unambiguously reveals
its associations with the formal sciences.
Having summarised the main topics that this book has addressed, it is nec-
essary briefly to emphasise the main issues that have not been presented here.
In particular, it should be recalled that, as stated in the introduction, this study
of the development of TGG effectively ceases when the year 1957 has been
reached (this being the year in which SS was published), and one reason for
 Conclusion 187

choosing this date as the terminus of the current study is to illustrate the void at
the core of the various discussions of the history of TGG that take 1957 as their
starting point. However, as a result of adopting 1957 as an upper-bound, there
has been no attempt in the present study to discuss Chomsky’s research into
formal language theory, which constituted one of his main academic preoccupa-
tions between the years 1958–1965. In particular, the work he published jointly
with both Miller and Marco Schützenberger (1920–1996) has not been assessed,
although such a re-evaluation is obviously necessitated by the various topics
addressed in the foregoing chapters. Similarly, with 1957 as the chosen cut-off
point, there is no space here for discussing the emergence of the innateness
hypothesis in exhaustive detail, the reason being that, although this hypothesis
was eventually to become one of the most characteristic features of generative
grammar, the initial expositions of this hypothesis were not presented until the
late 1950s and early 1960s. Given the subsequent dominance of this hypothesis
in the generative grammar tradition, this work clearly merits reconsideration
(in particular, Chomsky’s interactions with Eric Lenneberg (1921–1975) await
discussion), but such a task is outside the scope of the present study. In addition
to this nexus of interrelated topics, there are fascinating sociopolitical aspects of
the development of TGG that demand further exploration. For example, in the
late 1950s and early 1960s research into generative grammar was often funded
by the US defence department. However, as the 1960s advanced, such funding
sources ceased to be available for such work, although other kinds of research,
such as automatic speech recognition and machine translation, saw no diminu-
tion in their financial support during these years. The reasons for this shift in
perception from the institutions that initially financed TGG research have never
been discussed, barely even mentioned; and, alas, they remain neglected in this
book too, although the omission has (at least) been acknowledged. The ques-
tions are enticing, though. For instance, surely it can only be a coincidence
that the defence department funding of generative grammar research should
diminish in the late 1960s; that is, around the time that Chomsky started to
promulgate political views that were severely critical of US foreign policy.
Having provided this overview of the genesis of TGG in the light of various
developments in the formal sciences, it would be possible to terminate the dis-
cussion here, safe in the realm of conventional, retrospective historiography.
However, since it is always necessary to resist the artificial sense of complete-
ness that is inevitably associated with the conclusion of an investigation of
this kind, the discussion will be extended a little longer. The motivation for
continuing can be simply explained, and (perhaps predictably) it is prompted
by contemporary preoccupations concerning the nature of generative grammar.
188 Conclusion

Possibly such considerations seem out of place in a study of this kind, but it
should be remembered that, although TGG in its initial guise has long passed
into the historical twilight, more modern variants of the theory are still very
much in the full glare of daylight. Indeed, despite the many changes that have
occurred since the late 1950s (i.e., the influential reconstruction(s) of the 1960s
and 1970s that came to be referred to as the Extended Standard Theory (EST),
the advent of the Principles and Parameters (P&P) model in the 1980s, and
the subsequent inauguration of the controversial Minimalist Program (MP)),
the essential elements of the syntactic theory broadly known as ‘generative
grammar’ have remained remarkably constant. Consequently, certain aspects
of the theory that were originally bold assumptions (some of which were
ultimately derived from the formal sciences) have crystallised and have become
incontrovertible components. As ever, though, consequences eventually follow
from assumptions, and it is worth pausing to assess some of the implications
that are inherent in the framework adopted by generative grammar.
One aspect of TGG that has been emphasised throughout the foregoing chap-
ters is the conspicuously axiomatic-deductive (quasi-)proof-theoretical char-
acter of the theory. As demonstrated in sections 3.2 and 5.7, the axiomatic-
deductive method fascinated certain linguists during the first half of the twen-
tieth century, and it clearly provided the conceptual foundation for the type of
computational syntactic theory that Chomsky outlined in LSLT. Indeed, to take
this discussion back to where it (almost) began, it should be recalled that Lees
had noted, in his 1957 review of SS, that Chomsky’s theory utilised ‘an overt
axiom system’ (Lees 1957: 378), suggesting that this was one of the characteris-
tic features of the theory that made it ‘rigorous’ (Lees 1957: 378); and certainly
the axiomatic-deductive framework (with all its attendant logico-philosophico-
mathematical baggage) has remained a fundamental component of generative
grammar ever since. In his 1995 presentation of the MP, for example, Chomsky
sketched the basic structure of C H L , the computational component of Universal
Grammar (UG), and the proof-theoretical nature of this system is clear. For
instance, recursion is still explicitly utilised (i.e., ‘the operations of C H L recur-
sively construct syntactic objects’ (Chomsky 1995: 226)), and derivations are
‘generated’ in a derivational (as opposed to a representational) fashion, start-
ing with primitive elements (i.e., ultimately lexical and formal features) and
resulting in phonological form and logical form pairings (i.e., (π, λ)) (Chom-
sky 1995: 219ff.). Although it is not possible exhaustively to consider all aspects
of the various proof-theoretical features of contemporary generative grammar
here, it is revealing to probe the implications of certain assumptions that have
been retained; and, to this end, the formulation of recursion within contemporary
 Conclusion 189

versions of the theory will be discussed, primarily because the importance of

recursion within the generative grammar framework has increased dramatically
in recent years.
As indicated above, in The Minimalist Program (from henceforth MP)
Chomsky states that the operations of C H L ‘recursively construct’ syntactic
objects, and he later defines such objects as follows (see Chomsky 1995: 243):

Definition 6.1: Syntactic Objects

1. lexical items
2. K = {γ {α, β}}, where α, β are objects and γ is the label of K

Chomsky explicitly notes that clause 2 of definition 6.1 provides the ‘recursive
step’ (Chomsky 1995: 243):
Suppose a derivation has reached state  = {α, β, δi , ..., δn }. Then application
of an operation that forms K as in [definition 6.1, clause 2] converts  to
  = {K , δi , ..., δn } including K but not α, β. (Chomsky 1995: 243)

This summary is rather brief, and it is useful to work through a specific example
in order to appreciate the framework that is being proposed. For instance, if it is
assumed that the only operation that creates K is Merge, then derivation creation
can be viewed primarily as a process involving the repeated application of this
operation, which terminates when the initial numeration is mapped to a single
syntactic object. Schematically, if α1 , α2 , α3 , and α4 are the lexical items in a
given numeration, then for a derivation that begins in state  = {α1 , α2 , α3 , α4 },
one possible series of subsequent steps can be explicitly represented as
follows:

Derivation: Example 1
Given  = {α1 , α2 , α3 , α4 }:
Step 1: K 1 = Merge(α1 , α2 ) and   = {K 1 , α3 , α4 }
Step 2: K 2 = Merge(K 1 , α3 ) and   = {K 2 , α4 }
Step 3: K 3 = Merge(K 2 , α4 ) and   = {K 3 }

Chomsky stresses that, during the course of derivation generation, ‘no

new objects are added . . . apart from rearrangements of lexical properties’
(Chomsky 1995: 228), and therefore the whole process is determined by the
operation Merge and the features associated with the lexical items. In essence,
as Chomsky formulates it, the ‘recursive step’ constructs syntactic objects that
possess hierarchical structure. More precisely, Merge causes two syntactic
190 Conclusion

objects, α and β, to be combined in a principled manner in such a way that

a label derived from either α or β can be determined and associated with the
resulting hierarchically structured syntactic object. In later work, the need for
labels was questioned (for example, see Collins 2002 and Chomsky 2000a),
but even if a label-free system is proposed, the essential constructional proce-
dure remains the same. It is of interest, though, that, rather than associating
recursion explicitly with self-reference and other possible characteristic prop-
erties discussed at length in section 3.3, Chomsky associates it primarily with
finite computations: a finite set of primitive elements (i.e., lexical items) and
a finite set of operations (i.e., Merge) are used to generate a potentially infi-
nite set of hierarchical structures (i.e., sentences). Therefore, the ‘recursive’
aspect of this process, as Chomsky presents it, appears primarily to involve an
operation (i.e., Merge) that takes its own output (e.g., the K s in example 1) as
its own input. In 1995, then, a ‘recursive’ component, which enabled infinite
structures to be created using finite means, was still posited within MP-style
generative grammar, even though (inevitably) the exact details of its expression
had altered since such a component was first incorporated into TGG in the
1950s.
Although, as indicated above, a ‘recursive step’ was included in the MP in
1995, in recent years Chomsky has suggested that, rather than merely constitut-
ing a useful constructional procedure that was initially derived from the formal
sciences and incorporated into formal grammars, recursion may actually consti-
tute a unique language-related aspect of human cognitive function. For instance,
in a 2002 paper, ‘The Faculty of Language: What Is It, Who Has It, and How Did
It Evolve’, which Chomsky co-authored with Marc Hauser and Tecumseh Fitch,
the authors discuss the Faculty of Language in the Narrow sense (FLN) and
they indicate that this term denotes ‘the abstract linguistic computational system
alone, independent of the other systems with which it interacts and interfaces’
(Hauser, Chomsky, and Fitch 2002: 1571). Further, they state explicitly that ‘a
core property of FLN is recursion’, before clarifying this succinct statement by
explaining that ‘FLN takes a finite set of elements and yields a potentially infinite
arrangement of discrete expressions’ (Hauser, Chomsky, and Fitch 2002: 1571).
So far, none of this is especially unusual within the tradition of generative gram-
mar. However, in addition to these familiar claims, Hauser, Chomsky and Fitch
go on to hypothesise that FLN is ‘uniquely human’ and that it consists primar-
ily of a recursive component and procedures for mapping syntactic objects to
the interfaces (Hauser, Chomsky, and Fitch 2002: 1573). More specifically, the
claim (referred to from henceforth as the ‘FLN hypothesis’) is expressed as
follows:
 Conclusion 191

we suggest that FLN – the computational mechanism of recursion – is recently

evolved and unique to our species . . . we propose in this hypothesis that FLN
comprises only the core computational mechanisms of recursion as they appear
in narrow syntax and the mappings to the interfaces. If FLN is indeed this
restricted, this hypothesis has the intriguing effect of nullifying the argument
from design, and thus rendering the status of FLN as an adaptation open to
question. (Hauser, Chomsky, and Fitch 2002: 1573)

Clearly, then, within the context of contemporary generative grammar, recursion

has begun to play a crucial role. In the 1950s recursive definitions were simply
viewed as useful procedures derived from the formal sciences which (as Bar-
Hillel had indicated in 1953) could be incorporated into grammars in order
to facilitate linguistic analyses. However, if the above hypothesis proves to be
correct, then, in the most recent versions of the theory, recursion will have to be
viewed as a fundamental and unique species-specific property that is particularly
associated with the faculty of language. Obviously, given the magnitude of this
claim, it is essential to establish precisely what ‘the computational mechanism of
recursion’ means. Unfortunately, though, the discussion of recursion presented
by Hauser, Chomsky, and Fitch in their 2002 paper is simply too vague to
enable the details of the FLN hypothesis to be clarified. For instance, no formal
examples of particular recursive procedures are provided, though it can probably
be safely assumed that recursive definitions such as definition 6.1 (discussed
above) are intended. However, while the uncertainty is caused in part by a lack
of detailed examples, it is also partly due to the fact that the term ‘recursion’,
as it is currently used in the formal sciences, is fundamentally ambiguous. As
mentioned briefly in section 3.3, from the late 1930s onwards recursive function
theory has been associated both with λ-definability and with computability
theory, and some of the problems caused by this association must now be
addressed.
As discussed in section 3.3, Gödel was one of the leading researchers respon-
sible for advancing recursive function theory, especially in his influential 1931
and 1934 publications. However, in 1936 two mathematicians provided alter-
native definitions of effective calculability. First, Alonzo Church published a
paper which claimed that there was a close association between recursive func-
tion theory and λ-definability since general recursive functions could be shown
to be λ-definable functions (and vice versa). Second, Alan Turing published his
precocious paper ‘On computable numbers, with an application to the Entschei-
dungsproblem’, in which he introduced the notion of a computing machine (i.e.,
the mathematical entity now generally known as a Turing machine), thus ini-
tiating computability theory, an independent branch of mathematics that was
192 Conclusion

characterised by its own procedures and techniques (e.g., Turing machines,

Turning computable functions, the Turing Thesis, and the like) and which even-
tually provided the theoretical foundations for many advances in electronic
computing in the 1950s. The relationship between recursive function theory
and computability theory was recognised from the very beginning, since both
theories attempted to define finite computable procedures, and Gödel himself
was quick to realise that Turing’s work effectively subsumed his own research
into recursive function theory by providing a more general theoretical frame-
work. Inevitably, as a result of the perceived associations between these three
distinct theories, recursive function theory, λ-calculus, and computability the-
ory became conflated in the late 1930s, and one undesirable consequence of this
was that research undertaken within the framework of (say) computability the-
ory (i.e., research which explicitly used Turing machines, Turing computable
functions, and so on) was often formally expressed in terms of recursive func-
tion theory – and vice versa. This situation is clearly undesirable since (for
example) while all primitive recursive functions can be re-expressed as Turing
computable functions, it is certainly not the case that all Turing computable
functions can be re-expressed as primitive recursive functions, and, therefore,
if one speaks of the former while using terminology associated with the lat-
ter, misunderstandings inevitably abound. However, such unhelpful practices
have become common in the literature concerning recursion and computabil-
ity, and the mathematician Robert Soare has referred to this as the ‘Recursion
Convention’ (Soare 1996: 28).

The Recursion Convention has brought ‘recursive’ to have at least four differ-
ent meanings . . . This leads to some ambiguity . . . Worse still, the Convention
leads to imprecise thinking about the basic concepts of the subject; the term
‘recursion’ is often used when the term ‘computability’ is meant. (By the term
‘recursive function’ does the writer mean ‘inductively defined function’ or
‘computable function’?) Furthermore, ambiguous and little recognized terms
and imprecise thinking lead to poor communication both within the subject
and to outsiders, which leads to isolation and a lack of progress within the
subject, since progress in science depends upon the collaboration of many
minds. (Soare 1996: 29)

This is a disquieting passage, and it certainly emphasises the current problematic

state of the notion of recursion within the formal sciences.
In the light of the remarks quoted above, it is necessary to return to the work
of Hauser, Chomsky, and Fitch. Obviously, given the FLN hypothesis, the first
task is to attempt to determine which type of recursion is implicated. From
their informal discussion of the subject, it appears to be the case that Hauser,
 Conclusion 193

Chomsky, and Fitch are explicitly associating ‘recursion’ with the process of
creating an infinite arrangement of discrete expressions from a finite set of
elements (Hauser, Chomsky, and Fitch 2002: 1571, quoted above). This is
reasonable, though it is essential to note that this is a far more general procedure
than that associated with other interpretations of ‘recursion’. For instance, if this
elusive term is understood to mean ‘definition by induction’ (the interpretation
that Bar-Hillel recognised in 1953 when he proposed that recursive definitions
could be used in formal grammars), then in addition to the property of generating
infinite structures from finite means, ‘recursion’ would also involve some kind of
explicit self-reference. Since self-reference is not stressed in their paper, though,
Hauser, Chomsky, and Fitch may be using the term ‘recursion’ to indicate
something like ‘effectively calculable’ or ‘specified by a finite algorithm’, and
if this were the case, then it would suggest that Chomsky’s current use of the
term is more closely related to the interpretation of ‘recursion’ that associates it
with computability theory rather than with Gödelian recursive function theory.
However, as mentioned previously, since the presentation offered in the Hauser,
Chomsky, and Fitch paper is both brief and informal, it is simply not possible to
determine with certainty which interpretation of ‘recursion’ is actually intended.
Clearly, this is not the place in which to attempt to resolve some of these
problems, and the preceding discussion is presented merely to indicate that it is
impossible insightfully to assess the linguistic/biological role of the recursive
components within contemporary generative grammar without first reflecting
upon the tortive development of recursive function theory and its complex
associations with λ-calculus and computability theory; and explorations such
as that in section 3.3, which attempt to identify the way in which recursive
definitions were incorporated into syntactic theories in the 1950s, simply seek
to provide some kind of clarification.
If recursion is one topic that requires careful and dextrous consideration,
then the notion of theory-internal simplicity is certainly another. As discussed
(possibly ad nauseam) in sections 3.5 and 4.3, the characteristic concern with
simplicity criteria in early TGG was directly motivated by Chomsky’s inter-
est in Goodman’s research into the basal simplicity of constructional systems,
and related concerns have haunted generative grammar ever since its inception.
From the mid-1960s onwards, these concerns were often discussed in terms
of explanatory, as opposed to descriptive, adequacy, and the opposing tensions
caused by these guiding principles are well known: while descriptive adequacy
requires the full complexity of those aspects of natural language that fall within
the specified domain of enquiry to be captured by a formal linguistic theory,
explanatory adequacy essentially demands that the theory developed should not
194 Conclusion

become unnecessarily complex. Due to the fact that the introduction of explana-
tory adequacy as a theoretical prerequisite roughly coincided with the explicit
adoption of the innateness hypothesis, the notion has conventionally been pre-
sented as a psychological/biological requirement in the generative grammar
literature. In general, then, during the years c. 1965–1990 the main focus in
mainstream generative grammar research was upon the task of improving the
basic theory so that it could account more accurately for a larger number of lin-
guistic phenomena, and the task of further developing the ‘sketchy’ ideas con-
cerning simplicity criteria outlined in LSLT (Chomsky 1975[1955]: 116) was
largely abandoned. Nevertheless, during this 25-year period of theory develop-
ment, a basic sense of the importance of theory-internal simplicity was retained,
and there were certainly times when the eradication of certain unnecessary the-
oretical complexities was openly acknowledged as a profound advance. For
example, the development of X-bar theory in the early 1970s, which consid-
erably simplified the orthodox theory of generative grammar by providing a
unified analytical framework for phrasal projections, was widely recognised
as a significant improvement. However, despite such cases, it is reasonable to
claim that, in pre-c. 1990 generative grammar, considerations of ‘economy’ or
‘theory-internal simplicity’ never received the type of principled and sustained
scrutiny that they began to receive when the MP emerged as an identifiable
research programme in the early 1990s.
As is well known, the MP outlines an ambitious agenda, the main purpose
of which is to determine the extent to which the human language faculty can
be viewed as a ‘perfect’ solution to the problems posed by external constraints
(i.e., the interfaces that connect the syntactic component with the phonological
and semantic components). Put simply, one of the main goals of the MP is to
reduce to a minimum the essential machinery required to represent the faculty
of language, or, as Chomsky himself has expressed it recently,

The minimalist program is the attempt to explore these questions [i.e., the
questions ‘what are the properties of language’ and ‘why are they that way’]. Its
task is to examine in detail every device (principle, idea, etc.) that is employed
in characterizing languages to determine to what extent it can be eliminated in
favor of a principled account in terms of general conditions of computational
efficiency and the interface condition that the organ must satisfy for it to
function at all. Put differently, the goal is to determine just what aspects of
the structure and use of language are specific to the language faculty, hence
lacking principled explanation at this level. (Chomsky 2004: 106)

Clearly, although it is important to stress that, in contrast to the use of sim-

plicity criteria to select amongst competing grammars in TGG, in the MP the
 Conclusion 195

economy considerations determine the way in which operations such as Merge

enable derivations to be generated by C H L ; nevertheless the basic concern with
constructing a ‘principled account’ that is regulated by considerations of ‘com-
putational efficiency’ can be directly associated with the exploration of the
simplicity of grammars in the 1950s. Indeed, Chomsky himself has claimed
that the MP effectively heralds a return to some of the concerns that originally
motivated the creation of TGG. Writing in 1995, for example (and explicitly
referring to chapter 4 of LSLT), Chomsky emphasised the distinction between
‘an imprecise but not vacuous notion of simplicity that enters into rational
enquiry generally’ and ‘a theory-internal measure of simplicity that selects
among I-languages’ (Chomsky 1995: 8), and he continues:
The former notion of simplicity has nothing special to do with the study of
language, but the theory-internal notion is a component of UG, part of the
procedure for determining the relation between experience and I-language; its
status is something like that of a physical constant. In early work, the internal
notion took the form of an evaluation procedure to select among proposed
grammars (in present terms, I-languages) consistent with the permitted format
for rule systems. The P&P approach suggests a way to move beyond that
limited though nontrivial goal to address the problem of explanatory adequacy.
With no evaluation procedure, there is no internal notion of simplicity in the
earlier sense . . . Nevertheless, rather similar ideas have resurfaced, this time
in the form of economy considerations that select among derivations, barring
those that are not optimal in a theory-internal sense. The external notion of
simplicity remains unchanged: operative as always, even if only imprecisely.
(Chomsky 1995: 8-9)

In the terms of the MP, then, the concern with issues of theory-internal econ-
omy is formulated as a set of economy considerations that prevent non-optimal
derivations converging, and (as indicated in the above quotation) Chomsky him-
self acknowledges that there is a noticeable similarity between this approach
and the TGG focus upon simplicity criteria that selected between competing
grammars. The implication is that the MP can be viewed (in part) as an attempt
seriously to respond to certain questions concerning the theory-internal ‘sim-
plicity’ of formal grammars that were asked in early work on TGG, but which
have remained largely unanswered during the intervening decades. The fact that
such concerns have ‘resurfaced’ is perhaps surprising, but it certainly indicates
that research concerns that were responsible for prompting the emergence of
TGG in the 1950s are closely related to the research concerns that prompted
the advent of the MP, and such associations surely merit detailed considera-
tion. Obviously, the task of determining the nature of theory-internal economy
is non-trivial, and this is indicated by the various ways in which the topic is
196 Conclusion

approached in standard collections of MP-focused papers such as Epstein and

Hornstein 1999 and Epstein and Seely 2002. In these two collections alone,
issues of economy and/or theory-internal simplicity are adduced in discussions
of interpretable features and Case (Martin 1999), the copy theory of movement
(Nunes 1999), multiple spell-out (Uriagereka 1999), ‘elegant’ syntax (Brody
2002), syntactic objects and labels (Collins 2002), and crash-proof derivations
(Frampton and Gutmann 2002), to cite just a few instances. At various places,
these discussions all utilise arguments involving some kind of notion of econ-
omy or theory-internal simplicity in order to justify the theoretical modifications
or analyses that are presented; and this in turn suggests that such considerations
are currently perceived to be central to the task of linguistic theory develop-
ment within the MP tradition. Given the pervasive interest in such issues, it
is necessary to emphasise that the centrality of such concerns has profound
consequences for the methodology of syntactic research. For instance, it is an
assumption to claim that considerations of simplicity (or ‘economy’, or ‘ele-
gance’, or ‘compactness’, or any other related variant) are central to the task
of linguistic analysis, whether they are expressed in terms of conversions in a
given grammar itself or whether they are included as explicit conditions that
delimit the types of permissible derivations generated by a particular grammar.
Indeed, as Paul Feyerabend (1924–1994) argued in 1975, it is equally reason-
able to assert that such issues are never necessary considerations for scientific
research of any kind, and that the tendency to devote too much attention to
these topics can be misleading (at best) and positively detrimental (at worst).
In support of this provocative position, Feyerabend noted that the physicist
Niels Bohr would ‘dismiss the usual considerations of simplicity, elegance or
even consistency with the remark that such qualities can only be properly judged
after the event’ (Feyerabend 2001[1975]: 14).1 Feyerabend adds that ‘science is
never a completed process, therefore it is always “before” the event. Hence sim-
plicity, elegance and consistency are never necessary conditions of (scientific)
practice’ (Feyerabend 2001[1975]: 15.n1). Statements such as these impinge
directly upon syntactic theory as pursued within the MP tradition, suggesting
that base-less assertions concerning the centrality of economy considerations in
syntactic research (particularly unprincipled assertions which are used primarily
in order to justify a proposed modification to the theory) are entirely insufficient.
Consequently, it is not surprising that the role of ‘simplicity’ in contemporary
generative grammar is necessarily a contentious issue. Once again, a detailed
understanding of the origins of the initial concern with such issues, as they were
expressed within the TGG framework, reveals the sources of these preoccupa-
tions, and so provides a foundation for a more insightful discussion of the role
 Conclusion 197

of economy considerations in contemporary versions of the theory; and such

an understanding can only be achieved once the role of simplicity criteria in the
work of Goodman (and Quine) in the 1940s and 1950s has been assessed.
If theory-internal considerations of simplicity and the use of recursive devices
are all still central concerns within the MP, as they once were within TGG
(although now manifest in very different ways), then it could also be claimed
that the MP essentially (re)adopts a (constructive) nominalistic methodology
for syntactic theory. This tendency manifests itself in the various developments
that involve the elimination of abstract structures and/or processes that were
formerly accepted within the theory in favour of a simpler lexicalist account, a
development that is closely related to the topic of theory-internal simplicity. For
instance, as mentioned above, one of the boldest developments that Chomsky
proposed in MP was the rejection of X-bar theory in favour of bare phrase struc-
ture. As noted earlier, the X-bar framework, which had been a part of generative
grammar since the 1970s, specified a unified framework for phrasal projections,
thus providing a well-defined abstract hierarchical structure in advance of the
actual construction process itself. However, in the bare phrase structure frame-
work, all syntactic entities are defined in terms of the most primitive elements
posited in the system – namely, the various features associated with the items in
the lexicon, and formal features such as the EPP feature that are associated with
the core functional categories. One consequence of this approach is that abstract
structures such as VPs and N’ projections are effectively eliminated from the
construction process (though in practice they are often retained in expositions
as a form of shorthand notation). Similarly, in his most recent work, Chomsky
has suggested that the Spec-Head relation, which has played a central role in
standard syntactic analyses since the 1970s, should be rejected since this rela-
tion cannot justifiably be unduly privileged, leading to the thesis that ‘[a]pparent
Spec-H relations are in reality head-head relations involving minimal search’
(Chomsky 2004: 113).
As mentioned above, in the MP, the process of syntactic object construction is
viewed as a process in which ‘no new objects are added . . . apart from rearrange-
ments of lexical properties’ (Chomsky 1995: 228), and therefore all syntactic
objects constructed can be viewed as particular configurations of irreducible
features. Clearly, this view of the process of grammatical construction is remi-
niscent of the approach to constructional system theory proposed by Goodman
and Quine in the late 1940s; a framework in which they rejected abstract notions
such as sets, defining instead a calculus of individuals (i.e., irreducible funda-
mental elements) that avoided the ‘taint’ of Platonism (Goodman and Quine
1947: 107). Consequently, the process of deriving a grammatical sentence in
198 Conclusion

the MP framework is a feature-driven process that does not involve abstract

X-bar theoretical structures. Although it is not possible here to examine this
issue in detail, there are clearly parallels between the parsimonious analytical
methodology advocated in the MP literature and that advocated by Goodman
in his calculus of individuals, which Chomsky had considered in his earliest
work. To some extent, the rejection of certain kinds of abstract entities, which is
a characteristic feature of constructive nominalism and the MP alike, is related
to the emphasis placed on simplicity criteria: if the task is to develop a max-
imally economical system, then there is a valid motivation for eliminating all
non-essential components of the theory, and abstract structures are an obvious
target, especially if they can be defined in terms of smaller, more fundamental
units and processes.
As the above paragraphs imply, the relationship between TGG and the MP is
a fascinating one that has yet to be explored in detail, and, obviously, the brief
tour offered above is not intended to rectify this situation by providing deep
insights into the nature of this relationship. Rather, the intention is simply to
highlight some of the similarities, and to indicate that certain characteristics of
1950s-style TGG have remained constant throughout the evolution of generative
grammar during the past fifty years, while others have reappeared with renewed
prominence in the MP. These developments merit serious consideration, and a
detailed exploration of the origins of TGG can enable us to view more clearly
the various assumptions and analytical techniques associated with the MP; and
if a reassessment of contemporary generative grammar is considered to be an
important task, then it is necessary also to reconsider one of the words that
has resonated throughout this book as it has resonated throughout the history
of twentieth-century linguistics. The word (of course) is ‘science’. Bloomfield,
Sapir, Lees, Hockett, Chomsky,2 and many more have all written about the
scientific status (or the lack thereof) of linguistics, and the fundamental question
that has been asked and reasked in so many different ways can be succinctly
paraphrased as the cogent query: ‘Is linguistics a science?’ This question, of
course, begs numerous other questions none of which can be discussed here.
However, with specific reference to generative grammar, the whole issue of the
scientific status of linguistics has been raised in a number of recent publications,
and arguments both for and against have been constructed. To start with an
example of a negative assessment, with reference to the validity of generative
grammar, a disillusioned Paul Postal commented in 1995 that Chomsky3

has good reason for being insecure because he cannot fail to have noticed that
he has few substantial results in the sense that these are understood in more
 Conclusion 199

serious fields such as logic, mathematics, computer science, or physics. And

it is striking how elements of his position which were once considered to be
profound contributions now have vanished or become enormously marginal-
ized. Where are syntactic rule ordering, the principle of cyclic application,
the A-over-A Principle etc? [today one could add ‘Where are deep-structure,
surface-structure, X-bar theory, etc?’] Many of the principles and accomplish-
ments touted in recent years are almost embarrassing in their inadequacy and
shoddiness . . . The significant point, then, is that there is an extraordinary con-
trast between the paucity of genuine results in Chomskyan linguistics and the
forests of paper which have been, and continue to be, devoted to the linguistic
ideas involved. (quoted in Huck and Goldsmith 1995: 141–142)

In Postal’s view, then, generative grammar does not deserve to be classified as

a science (or at least as one of the ‘serious’ sciences) mainly because it has pro-
duced so few ‘genuine results’. The various theoretical constructs that appeared
to be ‘profound contributions’ at an earlier period have since faded or even
vanished completely, suggesting that they were only ever of ephemeral value.
However, the opposing viewpoint is well represented by Massimo Piattelli-
Palmarini, who, in his introduction to Juan Uriagereka’s Rhyme and Reason
(Uriagereka 1998), claimed that generative grammar

is well on the way to becoming a full-blown natural science, offering a seri-

ous promise of an advanced field of scientific inquiry whose idealizations,
abstractions, and deductions will eventually match in depth and subtlety those
of the most advanced domains of modern science. (Uriagereka 1998: xxv)

This pronouncement is unambiguous, and uttered only three years after Postal’s
contemptuous assessment. So which analysis is correct? Is generative gram-
mar a science or not? The divergence between Postal and Piattelli-Palmarini’s
views is rather alarming to say the least. Surely it should not be possible for
the opinions of informed, practising linguists to differ so greatly on this mat-
ter? Surely it should be possible for the linguistics community to agree as to
whether generative grammar has indeed produced any genuine results or not?
Surely it should at least be possible to determine whether, after over half a
century of formalisms, debates, analyses, assertions, counter-assertions, mis-
understandings, revelations, and revolutions, the researchers involved with the
generative grammar enterprise (whether TGG or EST or P&P or the MP) have
actually been responsible for some kind of scientific research or not? And if
such possibilities are not, in fact, possible, then surely it is the case that some-
thing is rather deeply wrong with syntactic theory, or at least with the particular
branch of it known as generative grammar? Unfortunately, there is not space
here fully to explore all the implications raised by these queries. However, if
200 Conclusion

the confusion and disagreement concerning the scientific status of generative

grammar is indeed as profound as has been indicated above, then it is surely
time to reconsider the various assumptions, intentions, and methods associated
with this particular subset of linguistic theory; and, in order to reflect sensibly
upon the scientific status (or otherwise) of contemporary generative grammar,
it is perhaps necessary first to brood protractedly upon the scientific aspects of
its past – a task that this book has attempted to accomplish.
 Notes

Introduction
1 For short but insightful discussions of the standard classifications (and some of the
attendant difficulties) see Suppe 2000 and Schaffer 1997.
2 For a notorious negative review of Chomsky’s book see Aarsleff 1970. For a more
dispassionate assessment, and for evidence of Chomsky’s annoyance at the reception
of Cartesian Linguistics, see the discussion in Koerner 1999, especially pp. 10–12,
p. 178, and pp. 210–214.
3 Some of the mathematical aspects of Hockett’s own work are discussed in
section 5.2.
4 Chomsky’s rejection of nominalism is discussed at length in section 5.3. For Lyons’
discussion of Chomsky’s critique of behaviourism, see 1970: 83ff., and for Chom-
sky’s own words, see Chomsky 1959b.
5 For the details see Matthews 1993: 131–134.

The consequences of analysis

1 Throughout this chapter the term ‘calculus’ is used in the standard way to refer
to the set of algorithmic procedures associated with differentiation and integration,
while the term ‘analysis’ is used to denote various extensions of the calculus-based
techniques, which include differential equations, the calculus of variations, and mul-
tivariate calculus.
2 ‘singulare . . . calculi genus’.
3 For this particular example in full, see Newton 1967[1665]: 273.
4 ‘quantitates per singula temporis infinite parva intervalla augentur’.
5 For a detailed discussion of Newton’s fluxional calculus, see Westfall 1980: 131–140.
6 For Leibniz’s full response, see Leibniz 1863b[1695]: 332.
7 A useful collection of papers (in English translation) that contributed to the
development of analysis during the eighteenth century can be found in Struik
1969, especially section 3. While often hostile towards each other, the researchers
associated with the English and Continental traditions appear to have interacted
more frequently than is sometimes suggested. For clear discussions of some
of the complexities, see Boyer 1949 (chapter 6), Guicciardini 1989, and Fraser
1997.

201
202 Notes to pages 27–38

8 See section 2.5 below for a brief summary of the development of logic in the nine-
teenth century.
9 The most authoritative overview of Cauchy’s contribution to the nineteenth-
century reformulation of analysis is Grabiner 1981. The development of Cauchy’s
ideas in relation to logic and set theory is considered in Grattan-Guinness 2000:
64–68.
10 Good surveys of various aspects of Weierstrass’s work can be found in Dugac
1973 and Manning 1975. As an aside, it should be mentioned that the consensus
amongst the mathematical community concerning the definitions of differentiation
and integration was undermined when Abraham Robinson introduced Non-Standard
Analysis in the 1960s. This theory provided a model-theoretic basis for Leibnizian
infinitesimals, therefore reviving the very conceptual foundation for the calculus that
Weierstrass had effectively eliminated. For further information about this fascinating
development, see Robinson 1996 and Dauben 1995.
11 A good explanation of ‘Dedekind cuts’ can be found in Hrbacek and Jech 1984: 100–
102.
12 An excellent introduction to Cantor’s research into number theory can be found in
Dauben 1979, especially chapter 2.
13 For more detailed accounts of Cantorian set theory, see Dauben 1979 (chapters 10
and 11), Jech 1991, and Grattan-Guinness 2000 (chapter 3). For a brief but insightful
summary, see Hallett 1984: 1–11.
14 For more information about set theory in the twentieth century, see Johnson
1972.
15 For general overviews of the development of logic, see Kneale and Kneale 1962 and
Grattan-Guinness 2000, chapter 4.
16 For more information about Frege’s logical systems see Baker 1984 and Dummett
1991. As the twentieth century progressed Frege was gradually appropriated by
analytical philosophers. For a recent attempt to reclaim him for logicians and math-
ematicians, see Grattan-Guinness 2000: 177–199.
17 By comparison with the work of other logicians concerned with foundational issues,
Peano’s work has been rather neglected. For a reasonably detailed overview, see
Grattan-Guinness 2000, chapter 5.
18 For more information concerning Russell’s intellectual development, see Clark 1975
(chapter 5), and Hylton 1990 (chapter 3).
19 In the years following the publication of Whitehead and Russell’s ideas concerning
the theory of logical types, a number of criticisms of the theory emerged. For a
general overview of the theory itself and the controversy surrounding it, see Copi
1971.
20 For a more detailed discussion, see Grattan-Guinness 2000, especially chapter 7.
21 For more information about the terminology used in PM, see Grattan-Guinness
2000: 384–400.
22 For a succinct introduction to various non-classical logics, see Priest 2001.
23 For more biographical information about Hilbert, see Reid 1996, and, for more
information about Formalism in general, see Kreisel 1958, Detlefsen 1993, and
Hintikka 1995.
 Notes to pages 39–65 203

24 As mentioned in section 2.4, Hilbert seems to have known about Cantor’s own doubts
concerning set theory as early as 1896. For more information, see Grattan-Guinness
2000: 117–119.
25 For instance, see Hilbert 1932[1918], in which he considers the utility of such a
language when constructing axiomatic-deductive arguments.
26 For timely words of caution see Ewald 1996: 1106–1107 and Mancosu 1998: 163–
164.
27 By far the best source of information concerning all aspects of Brouwer’s life and
work is van Stigt 1990.
28 It is of interest, incidentally, that Brouwer’s topological papers largely conform to
classical assumptions about mathematics and do not overtly manifest his Intuitionist
concerns. No doubt his reticence during these years was partly due to his desire to
ingratiate himself with his peers.
29 ‘Brouwer, that is the revolution!’
30 For more information concerning Intuitionistic logic see Mancosu 1998 (part 4) and
van Stigt 1990, especially chapter 5. For a discussion of the relationship between
Intuitionism and proof theory, see Kino et al. 1970.
31 For more information concerning Intuitionist set theory see van Stigt 1990
(chapter 6) and Mancosu 1998: 1–27.

Mathematical linguistics
1 These particular examples can be found in Bloomfield 1926: 154.
2 For more information concerning the influence of Weiss upon Bloomfield’s work,
see Belyi 1967. Bloomfield cites Weiss’ paper (Weiss 1925) explicitly in his own
paper (Bloomfield 1926).
3 Certain aspects of Bloomfield’s later work concerning this topic are discussed in
section 3.7.
4 These particular examples can be found in Bloch 1948: 6–7.
5 Another important example, namely Bar-Hillel’s development of Categorial Gram-
mar, is discussed at length in section 3.4.
6 It should be noted that Chomsky discussed Harwood’s paper explicitly in Syntactic
Structures and compared Harwood’s system with his own version of phrase struc-
ture grammar. For specifics, see Chomsky 1957b: 26n1. Some of the reasons why
the axiomatic-deductive character of TGG impressed contemporary linguists are
considered in various places in chapters 4 and 5.
7 Detailed discussions of the set of primitive recursive functions can be found in
Kleene 1952 (chapter 9), Crossley and Dummett 1965, and Fitting 1981. For an
approachable and insightful analysis of the various terminological problems that
have enveloped recursive function theory since the late 1930s, see Soare 1996.
8 The well-known schemata discussed here were influentially presented in Kleene
1952: 219.
9 For more details, see Gödel 1986b[1934] and Kleene 1952: 270–276.
10 Various aspects of the work of Bar-Hillel are discussed in several sections of this
chapter. This focus is necessary since his role in the development of TGG has often
204 Notes to pages 65–77

been neglected. This neglect is unjustifiable, though, since, as Chomsky himself

has acknowledged, Bar-Hillel was a ‘constructive participant’ (quoted in Kasher
1991: 6) in the development of TGG.
11 Bar-Hillel’s work was also influenced by Carnap’s 1952 paper ‘Meaning Postulates’,
which is discussed at length in section 4.3.
12 The various ellipses in the following examples are Bar-Hillel’s.
13 For more information about the Lvov-Warsaw school in general see Wolénski 1989.
A good collection of papers in English translation can be found in McCall 1967, and
for an excellent collection of papers exploring the linguistic interests of the members
of the school, see Historiographia Linguistica 25:1/2 1998. In the current study, the
work of Tarski is considered separately in section 4.5.
14 For more information concerning Leśniewski’s system, see Luschei 1962.
15 For more information concerning meaning-rules, see Ajdukiewicz 1978a[1934]: 57–
66. His interest in pragmatics eventually culminated in Ajdukiewicz 1965.
16 The best source of information concerning Harris’ ‘procedures’ is MSL itself. As an
example, though, a typical distributional (morphophonemic) procedure was the one
called ‘Interchanging Phonemes among Alternates of One Morpheme’. The purpose
and function of this procedure is summarised by Harris as follows: ‘we group together
into one morphophoneme the phonemes which replace each other in corresponding
parts of the various members of a morpheme’ (Harris 1951: 224). As a result, ‘knife’
and ‘knive’ are grouped together and associated with the single morphophonemic
sequence //nayF//. For more information concerning distributional linguistics, see
Matthews 1993, chapter 3.
17 The title of Carnap’s text poses a few problems for translators. The actual title was
Der logische Aufbau der Welt: Versuch einer Konstitutionstheorie der Begriffe, and
the word ‘Aufbau’ is commonly rendered into English in the philosophical litera-
ture as ‘Structure’. This is done because Carnap also uses the noun ‘Konstitution’
regularly, and it is usually deemed appropriate to reserve the term ‘Construction’
for this concept. However, ‘Aufbau’ is more accurately converted into English if it
is translated as ‘Construction’, and since these terminological difficulties are not a
central concern here, this translation has been adopted. For further discussion of the
complexities, see Richardson 1998: 6n3.
18 Like all paragraph-long accounts of the Vienna Circle, this summary is inevitably
inadequate in many respects. Fortunately, there is a substantial body of literature
concerning logical positivism in general and the work of the Vienna Circle in par-
ticular. For instance, Weinberg’s classic book An Examination of Logical Positivism
(1936) is still a revealing, contemporaneous introduction, as is Ayer’s influential
Language, Truth and Logic (1936). Two useful anthologies of core Vienna Circle
papers are Ayer 1959 and Hanfling 1981. In recent years the standard interpretation
of the whole movement has been destabilised by provocative revisionist accounts.
Excellent examples of this current trend are Richardson 1998 and Friedman
1999.
19 The relationship between LCW and Russell’s work has been reconsidered recently.
For instance, see the assessment in Richardson 1998: 22–30.
20 For details, see Carnap 1967[1928]: 65ff.
 Notes to pages 77–102 205

21 Quine’s own detailed account of his time in Vienna can be found in Quine
1985: 86ff. A useful recent assessment of Quine’s work is Orenstein 2002, which
includes some discussion of the complex relationship between Chomsky and
Quine.
22 The complete text of Quine’s 1934 lectures on Carnap can be found in Creath
1990: 47–103.
23 For specifics concerning Carnap’s interpretation of the connection between his work
and that of Goodman, see Carnap 1963: 19.
24 In brief, Goodman realised that ‘joints’ and ‘segments’ were more effective measures
of simplicity. An n-place predicate joins each occupant of one of its n places with
the occupants of the other places. Therefore, Goodman was able to define joints
as follows: ‘The number of joints . . . is the number of places in any minimum
set of pairs of places that ties all the places of the predicate together – i.e., that
provides between every two places in the predicate a route consisting entirely of
steps between places paired in the set’ (Goodman 1949: 36). By contrast, the notion
of a segment can be expressed in terms of place occupants that are identical, and the
guiding intuition is that an n-place predicate is more complex if fewer of its places
have the same occupant. As Goodman puts it: ‘since all the places that have the
same occupant as any given place belong to what I call a segment, clearly segments
rather than places are to be counted in determining complexity’ (Goodman 1949:
37).
25 An analytic truth is one that is true for logical reasons (e.g., p → ( p ∨ ¬ p)), while
a synthetic truth depends upon extralogical considerations (e.g., ‘Socrates is a man’
is true or false depending upon the state of the universe of discourse).
26 For example, see Fries 1954. A broad discussion of these issues can be found in
Matthews 1993: 118–122. The intensity of Bloomfield’s distaste for semantics has
been questioned from time to time. For some discussion of this position, see Murray
1994: 130–132.
27 The remaining fragments can be found in Bloomfield 1970b: 333–338.
28 This fact alone suggests some kind of familiarity with the work of Russell (perhaps
his accessible The Principles of Mathematics (1903)?), since Russell had been the
most assiduous paradox collector, and, as mentioned in section 2.5, the theory of
logical types had been designed primarily to obviate the kind of set-theoretical self-
reference that engendered the paradoxes.
29 A clear presentation of Grelling’s paradox can be found in Grelling 1936.
30 This paper is mentioned briefly in Bloomfield (1970a[1937]: 333).
31 The redefinition of the notion of a set was one common response to the work of Can-
tor, Whitehead, and Russell during the first half of the twentieth century. For instance,
as discussed in section 3.4, in the 1920s Leśniewski’s ‘mereology’ was devised in
order to avoid some of the problems of self-reference. For more information, see
Luschei 1962.
32 The task of attempting to determine which researchers unambiguously belonged
to the group now known as the post-Bloomfieldians has preoccupied a number
of linguistic historiographers. For a consideration of the various complexities, see
Hymes and Fought 1981[1975].
206 Notes to pages 109–123

Systems of syntax: 1951–1995

1 For example, Chomsky 1988, Otero 1994b, Barsky 1997, and Smith 2004.
2 For instance, in the introduction to The Minimalist Program Chomsky writes:
The shift in perspective provided by the P&P [i.e., Principles and Parame-
ters] approach also gives a different cast to the question of how simplicity
considerations enter into the theory of grammar. As discussed in the earliest
work in generative grammar, these considerations have two distinct forms: an
imprecise but not vacuous notion of simplicity that enters into rational enquiry
generally must be clearly distinguished from a theory-internal measure of sim-
plicity that selects among I-languages (see Chomsky 1975[1955], chapter 4).
The former notion of simplicity has nothing special to do with the study of
language, but the theory-internal notion is a component of UG [i.e., Universal
Grammar], part of the procedure for determining the relation between expe-
rience and I-language; its status is something like that of a physical constant.
In early work, the internal notion took the form of an evaluation procedure to
select among proposed grammars (in present terms, I-languages) consistent
with the permitted format for rule systems. The P&P approach suggests a way
to move beyond that limited though nontrivial goal and to address the problem
of explanatory adequacy. With no evaluation procedure, there is no internal
notion of simplicity in the earlier sense. (Chomsky 1995: 8)
3 For this particular reference, see Chomsky 1979b[1951]: 67n5.
4 Since it is one of the core texts involved in the development of TGG, the publication
history of LSLT deserves comment. The manuscript was complete by the spring of
1955 and revised for publication the following year. Having been rejected by the
Technology Press at MIT, it circulated widely in a mimeographed version during
the late 1950s, 1960s, and early 1970s. Chomsky finally agreed to have the book
published, in a revised and truncated form, in 1975. For more detailed information
concerning the complex genesis of this seminal text, see Chomsky’s introduction
to the 1975 edition of LSLT. However, as mentioned in the introduction, it is also
worth reading Murray 1980 and Murray 1994 as an antidote, since Murray provides
alternative reasons for the delay in publication.
5 The definition of grammatical levels in LSLT, which draws heavily upon construc-
tional system theory, is considered in section 5.4.
6 Evaluation procedures are further discussed in section 5.3.
7 The advent of Generalised Phrase Structure Grammar in the late 1970s and early
1980s prompted a reassessment of these issues. For details, see Gazdar et al. 1985.
8 Chomsky had initially submitted this paper to the Journal of Symbolic Logic in
October 1952.
9 In this context, constructional homonymity occurs whenever sentences sharing
superficially identical forms are assigned to different syntactic classes. For instance,
the two sentences
r The man stood by the woman (with the sense ‘the man championed the
woman’)
r The man stood by the tree (with the sense ‘the man stood beside the tree’)
obviously differ significantly in their formal structure, despite apparent similarities.
 207

10 Chomsky’s shift from (apparent) empiricism to rationalism is considered separately

in section 5.3.
11 Bar-Hillel’s appreciation of Carnap was considered in section 3.7.
12 Carnap’s formation and transformation rules are discussed in section 3.7, while the
difference between Chomskyan and Carnapian ‘transformations’ is considered at
length in section 5.5.
13 Bar-Hillel acknowledges that, in recent work, Harris had begun to introduce rules that
he referred to as ‘transformations’, but he points out that this is merely an attempt
to incorporate transformation rules into the formation part of syntax (Bar-Hillel
1954: 231–232).
14 For specifics, see Carnap 1942: x.
15 The paper ‘The Concept of Truth in Formalized Languages’ was presented to the
Warsaw Scientific Society on 21 March 1931. It was published in Polish in 1933
and in German in 1936. An English translation appeared in Tarski 1956: 152–
278.
16 The paper ‘The Establishment of Scientific Semantics’ appeared in Polish in 1936
and in German later the same year. An English translation appeared in Tarski
1956: 401–408.
17 Just to clarify, Carnap’s book was called Der logische Syntax der Sprache and this
title is most accurately translated as The Logical Syntax of Language. However,
the standard English translation that was published in 1937 was given the title The
Logical Structure of Language, and it is this version of Carnap’s title that Chomsky
is clearly echoing.
18 For more information about Montague’s work, see Davis and Mithun 1979, and, for
a useful collection of Montague’s own papers, see Montague 1974.

Transforming generative grammar: 1955–1957

1 For details, see the discussion in Chomsky 1957b: 72–105.
2 Shannon was responsible for the first part of the book (pp. 1–91), while Weaver
was responsible for the remainder, hence the rather specific reference below to the
individual authors of what appears to be a co-authored book.
3 Chomsky’s review of Hockett’s A Manual of Phonology appeared in 1957 and drew
heavily upon ‘TMDL’ and SS, hence the focus of the following discussion. For
Chomsky’s explicit discussion of Hockett, though, see Chomsky 1957a.
4 For some scene-setting, see Barsky 1997: 87–88.
5 For instance, see Matthews 1993: 134–142 and Newmeyer 1980: 19–36.
6 Although the introduction to LSLT is dated 1973, it was not published until 1975.
7 For a stimulating discussion of the associations between logical positivism and
behaviourism see Smith 1986. For a primarily philosophical perspective on the
type of epistemology Chomsky came to advocate in his later work, see ‘Chomsky:
Linguistics and Epistemology’ in Nagel 1995. It should be noted, just for clarity,
that while the conspicuous, mind-based philosophy that has been associated with
generative grammar since the early 1960s is usually assumed to be a crucial part of
Chomskyan syntax, it is (in theory) entirely separable from the formal procedures
and theoretical constructs that constitute the machinery of generative grammar.
208 Notes to pages 154–198

8 It is worth mentioning that in his (now usually ignored) 1954 review of Rieger’s Mod-
ern Hebrew, which had been published in 1953, Chomsky commented favourably
upon the type of data collection techniques that Rieger advocates, remarking that
they constitute ‘a method that a linguist might be tempted to use in constructing a
linguistic corpus’ (Chomsky 1954: 180). This brief observation may indicate that,
in 1954, Chomsky was still primarily concerned with the sort of post-Bloomfieldian
corpus-based discovery procedures that he would later reject, or, at least, it may
reveal that he was not yet willing to question the validity of these methods in public.
9 For example, see the discussion in Katz 1981, chapter 1. Katz’s assessment of the
topic is a good example of the sort of opportunities for explication that have been
missed in the past. His approach is philosophical rather than historiographical, and
therefore he is too swift to project some of Chomsky’s later ideas (especially his
later ‘rationalism’) onto his earlier work. The title of Katz’s chapter, ‘From Harris’s
Nominalism to Chomsky’s Conceptualism’, indicates the nature of the deficiency.
Also, inexplicably, Katz does not mention Carnap in the chapter, despite his obvious
awareness of the influence of Carnap’s work on linguistics in the 1950s.
10 See Chomsky 1975[1955]: 294–306 for full details concerning various problems
with phrase structure grammar.
11 For example, see the succinct and (comparatively) non-technical discussion in Chom-
sky 1957b: 61–84.
12 For details, see Matthews, 1993: 133ff.
13 It should be noted that, although it appeared in the 1975 publication of LSLT, the
introduction (part of which is quoted here) is dated 1973.
14 For this particular example, see Chomsky 1975[1955]: 175.

Conclusion
1 Feyerabend is here quoting Lovis Rosenfeld, and he has added the italics himself.
2 i.e., Bloomfield 1926, Sapir 1929, Lees 1957, Hockett 1967, Chomsky 2000b.
3 For similar diatribes, see Postal 2004.
 Bibliography

Aarsleff, H. (1970), ‘The History of Linguistics and Professor Chomsky’, Language

44: 570–585.
Ajdukiewicz, K. (1965), Pragmatic Logic, Holland and the United States: D. Reidel
Publishing Co.
Ajdukiewicz, K. (1978a[1934]), ‘Language and Meaning’, reprinted in English
translation in Ajdukiewicz 1978c: 35–66.
Ajdukiewicz, K. (1978b[1931]), ‘On the Meaning of Expressions’, reprinted in
English translation in Ajdukiewicz 1978c: 1–34.
Ajdukiewicz, K. (1978c), The Scientific World-Perspective and Other Essays:
1931–1963, ed. J. Giedymin, Holland and the United States: D. Reidel Publishing
Co.
Ajdukiewicz, K. (1978a[1936]), ‘Syntactic Connexion’, reprinted in English
translation in Ajdukiewicz 1978c: 118–139.
Ayer, A. J. (1936), Language, Truth and Logic, London: Gollancz.
Ayer, A. J. (1959), Logical Positivism, New York: Free Press.
Bach, E. (1964), An Introduction to Transformational Grammars, New York: Holt,
Rinehart & Winston, Inc.
Baker, G. P. (1984), Frege: Logical Excavations, New York and Oxford: Oxford
University Press.
Bar-Hillel, Y. (1953a), ‘On Recursive Definitions in Empirical Science’, 11th
International Congress of Philosophy 5: 160–165.
Bar-Hillel, Y. (1953b), ‘A Quasi-Arithmetical Notation for Syntactic Description’,
Language 29: 47–58.
Bar-Hillel, Y. (1954), ‘Logical Syntax and Semantics’, Language 30: 230–237.
Bar-Hillel, Y. (1963), ‘Remarks on Carnap’s Logical Syntax of Language’, in Schilpp
1963: 519–543.
Bar-Hillel, Y. (1964), Language and Information: Selected Essays on their Theory and
Application, Reading, Massachusetts: Addison-Wesley Publishing Co.
Barsky, R. F. (1997), Noam Chomsky: A Life of Dissent, Cambridge, Massachusetts:
MIT Press.
Belyi, V. V. (1967), ‘Some Facts about Weiss’ Influence on Bloomfield’, in Fought
1999, II: 115–118.
Berkeley, G. (1992[1734]), The Analyst, reprinted in Jesseph, D. M. (1992), De Motu
and The Analyst: A Modern Edition, with Introductions and Commentary, Boston,
and London: Kluwer Academic Publishers Dordrecht.

209
210 Bibliography

Birkhoff, G. (ed.) (1973), A Source Book in Classical Analysis, Cambridge,

Massachusetts: Harvard University Press.
Bloch, B. (1948), ‘A Set of Postulates for Phonemic Analysis’, Language 24: 3–46.
Bloomfield, L. (1926), ‘A Set of Postulates for the Science of Language’, Language
2: 153–164.
Bloomfield, L. (1933), Language, New York: Henry Holt & Co.
Bloomfield, L. (1935), ‘Linguistic Aspects of Science’, Philosophy of Science
2: 499–517.
Bloomfield, L. (1936), ‘Language or Ideas?’, Language 12: 89–95.
Bloomfield, L. (1955[1939]), ‘Linguistics Aspects of Science’, in Neurath, O. Carnap,
R. and Morris, C. (eds.), reprinted in 1955, The International Encyclopedia of
Unified Science, Vol. I, Chicago: University of Chicago Press.
Bloomfield, L. (1970a[1937]), The Language of Science, unpublished manuscript
fragments collected in Bloomfield 1970b: 333–338.
Bloomfield, L. (1970b), A Leonard Bloomfield Anthology, ed. C. F. Hockett,
Bloomington: Indiana University Press.
Borgström, C. H. J. (1949), ‘The Technique of Linguistic Descriptions’, Acta
Linguistica 5: 1–14.
Boyer, C. B. (1949), The History of the Calculus and its Conceptual Development,
New York: Dover Publications Inc.
Bradley, F. H. (1883), The Principles of Logic, Oxford: Oxford University Press.
Brody, M. (2002), ‘On the Status of Representations and Derivations’, in Epstein and
Seely 2002: 19–41.
Brouwer, L. E. J. (1912), ‘Intuitionisme en formalisme: inaugurale rede’, Amsterdam:
Clausen.
Brouwer, L. E. J. (1979[1907]), Over de grondslagen derwiskunde, reprinted in
English translation in van Stigt, W. P. (1979), ‘The Rejected Parts of Brouwer’s
Dissertation on the Foundations of Mathematics’, Historia Mathematica 6: 385–404.
Cantor, G. F. L. P. (1883), Grundlagen einer allgemeinen Mannigfaltigkeitslehre,
Leipzig: Teubner.
Cantor, G. F. L. P. (1895), ‘Beiträge zur Begründung der transfiniten Mengelehre’, in
Zermelo, E. (ed.) (1932), Gesammelte Abhandlung mathematischen und
philosophischen Inhalts, Berlin: Springer, 282–311.
Cantor, G. F. L. P. (1937[1899]), letter in Noether, E., and Cavaillès, J. (1937),
Briefweschel Cantor-Dedekind, Paris: Hermann, pp. 405–411.
Carnap, R. (1928), Der logische Aufbau der Welt: Versuch einer Konstitutionstherorie
der Begriffe, Berlin: Welt-Kreis.
Carnap, R. (1937[1934]), Logische Syntax der Sprache, Vienna, reprinted in English
translation in 1937 as The Logical Structure of Language, London and New York:
Routledge.
Carnap, R. (1942), Introduction to Semantics, Cambridge, Massachusetts: Harvard
University Press.
Carnap, R. (1944), Formalization of Logic, Cambridge, Massachusetts: Harvard
University Press.
Carnap, R. (1952), ‘Meaning Postulates’, Philosophical Studies 3: 65–73.
 Bibliography 211

Carnap, R. (1963), ‘Intellectual Autobiography’, in Schilpp 1963: 1–84.

Carnap, R. (1967[1928]), The Logical Structure of the World: Pseudoproblems in
Philosophy, translated by R. A. George, London: Routledge & Kegan Paul.
Cauchy, A. L. (1821), Analyse Algébrique, Paris: Debure Frères.
Cauchy, A. L. (1823), Leçon sur le calcul infinitésimal, Paris: Debure Frères.
Cherry C. E., Halle M., and Jakobson R. (1953), ‘Toward the Logical Description of
Languages in their Phonemic Aspect’, Language 29: 34–46.
Chomsky, N. (1953), ‘Systems of Syntactic Analysis’, Journal of Symbolic Logic
18: 242–256.
Chomsky, N. (1954), review of Rieger, E., Modern Hebrew, Language 30: 180–181.
Chomsky, N. (1955a), ‘Logical Syntax and Semantics: Their Linguistic Relevance’,
Language 31: 36–45.
Chomsky, N. (1955b), ‘Semantic Considerations in Grammar’, Monograph Series in
Language and Linguistics: 141–153.
Chomsky, N. (1956), ‘Three Models for the Description of Language’, IRE
Transactions of Information Theory, IT-2: 113–124.
Chomsky, N. (1957a), review of Hockett’s A Manual of Phonology, International
Journal of American Linguistics 23: 223–234.
Chomsky, N. (1957b), Syntactic Structures, The Hague: Mouton Publishers.
Chomsky, N. (1959a), ‘On Certain Formal Properties of Grammars’, Information and
Control 2, no. 2: 137–167.
Chomsky, N. (1959b), review of Skinner’s Verbal Behaviour, Language 35: 26–58.
Chomsky, N. (1966), Cartesian Linguistics: A Chapter in the History of Rationalist
Thought, New York and London: Harper & Row.
Chomsky, N. (1975[1955]), The Logical Structure of Linguistic Theory, Cambridge,
Massachusetts: MIT Press.
Chomsky, N. (1979a), Language and Responsibility, New York: Pantheon.
Chomsky, N. (1979b[1951]), Morphophonemics of Modern Hebrew, New York:
Garland.
Chomsky, N. (1988), The Chomsky Reader, London: Serpent’s Tail.
Chomsky, N. (1995), The Minimalist Program, Cambridge, Massachusetts: MIT Press.
Chomsky, N. (2000a), ‘Minimalist Inquiries: The Framework’, in Michaels, D.,
Uriagereka, J., and Martin, R. (eds.), Step by Step: Essays on Minimalist Syntax in
Honor of Howard Lasnik, Cambridge, Massachusetts: MIT Press.
Chomsky, N. (2000b), New Horizons in the Study of Language and Mind, Cambridge:
Cambridge University Press.
Chomsky, N. (2004), ‘Beyond Explanatory Adequacy’, in Belletti, A. (ed.), Structures
and Beyond: The Cartography of Syntactic Structures, Oxford Studies in
Comparative Syntax, New York and Oxford: Oxford University Press, III: 104–131.
Chomsky, N., and Miller G. A. (1958), ‘Finite State Languages’, Information and
Control 2, no. 2: 91–112.
Church, A. (1944), Introduction to Mathematical Logic: Part 1, Annals of
Mathematics Studies, Princeton: Princeton University Press.
Church, A. (1956), Introduction to Mathematical Logic: Volume 1, Princeton:
Princeton University Press.
212 Bibliography

Clark, R. W. (1975), The Life of Bertrand Russell, London: Jonathan Cape and
Weidenfeld & Nicolson.
Collins, C. (2002), ‘Eliminating Labels’, in Epstein and Seely 2002: 42–64.
Copi, I. (1971), The Theory of Logical Types, London: Routledge & Kegan Paul.
Creath, R. (ed.), (1990), Dear Carnap, . . . , Dear Van: the Quine–Carnap
Correspondence and Related Work, Berkeley: University of California Press.
Crossley, J. N., and Dummett, M. A. E. (1965), Formal Systems and Recursive
Functions, Amsterdam: North-Holland.
Dauben, J. W. (1979), Georg Cantor: His Mathematics and Philosophy of the Infinite,
Cambridge, Massachusetts: Harvard University Press.
Dauben, J. W. (1995), Abraham Robinson: The Creation of Nonstandard Analysis: A
Personal and Mathematical Odyssey, Princeton: Princeton University Press.
Davis, S., and Mithun, M. (1979), Linguistics, Philosophy and Montague Grammar,
Austin: University of Texas Press.
Dedekind, R. J. W. (1932[1872]), Stetigkeit und irrationale Zahlen, reprinted in Fricke,
R., Noether, E., and Ore, O. (eds.), (1930–1932), Gesammelte mathematische
Werke, Vieweg: Braunschweig; III: 315–334.
Dedekind, R. J. W. (1932[1888]), ‘Was sind und was sollen die Zahlen?’, reprinted in
Fricke, R., Noether, E., and Ore, O. (eds.), (1930–1932), Gesammelte
mathematische Werke, Vieweg: Braunschweig; III: 335–391.
Detlefsen, M. (1993), ‘Hilbert’s Formalism’, Revue Internationale de Philosophie
47: 285–304.
Dugac, P. (1973), ‘Eléments d’analyse de Karl Weierstrass’, Archive for the History of
Exact Science 10: 41–176.
Dummett, M. A. A. (1977), Elements of Intuitionism, Oxford: Clarendon Press.
Dummett, M. E. (1991), Frege: Philosophy of Mathematics, London:
Duckworth/Cambridge, Massachusetts: Harvard University Press.
Embleton, S., Joseph, J. E., and Niederehe H.-J. (eds.) (1999), The Emergence of
Modern Language Science, 2 vols., Amsterdam: John Benjamins.
Epstein, S. D., and Hornstein, N. (eds.) (1999), Working Minimalism, Cambridge,
Massachusetts: MIT Press.
Epstein, S. D., and Seely, T. D. (eds.) (2002), Derivation and Explanation in the
Minimalist Program, New York and London: Blackwell.
Euler, L. (1912[1755]), Institutiones Calculi Differentialis, reprinted in Opera Omnia,
series 1, vol. 10: 69–72.
Ewald, W. (1996), From Kant to Hilbert: A Source Book in the Foundations of
Mathematics, vols. I and II, Oxford: Clarendon Press.
Feyerabend, P. (2001[1975]), Against Method, revised 3rd ed, London and New York:
Verso.
Fitting, M. (1981), Fundamentals of Generalised Recursive Function Theory,
Amsterdam: North-Holland.
Fought, J. (ed.) (1999), Leonard Bloomfield: Critical Assessments of Leading
Linguists, London and New York: Routledge.
Frampton, J., and Gutmann, S. (2002), ‘Crash-Proof Syntax’, in Epstein and Seely
2002: 90–105.
 Bibliography 213

Fraser, C. G. (1997), Calculus and Analytical Mechanics in the Age of Enlightenment,

London: Ashgate.
Frege, F. L. G. (1879), Begriffsschrift, eine der arithmetischen nachgebildete
Formelsprache des reinen Denkens, Halle, Saale: Niebert.
Frege, F. L. G. (1884), Die Grundlagen der Arithmetik. Eine logisch-mathematische
Untersuchung über den Begriff der Zahl, Breslau: Köbner.
Frege, F. L. G. (1893), Die Grundgesetze der Arithmetik, begriffsschriftlich abgeleitet
1, Jena: Pohle.
Friedman, M. (1999), Reconsidering Logical Positivism, Cambridge: Cambridge
University Press.
Fries, C. C. (1952), The Structure of English: An Introduction to the Construction of
English Sentences, New York and London: Longmans, Green.
Fries, C. C. (1954), ‘Meaning and Linguistic Analysis’, reprinted in Fought 1999,
II: 84–97.
Gazdar, G., Klein, E., Pullum G., and Sag, I. (1985), Generalised Phrase Structure
Grammar, Oxford: Basil Blackwell.
Gödel, K. (1986a), Kurt Gödel: Collected Works, 2 vols, Oxford: Oxford University
Press.
Gödel, K. (1986b[1934]), ‘On Undecidable Propositions of Formal Mathematical
Systems’, reprinted in Gödel 1986a, I: 346–371.
Gödel, K. (1986c[1931]), ‘Über formal unentscheidbare Sätze der Principia
Mathematica und verwandter Systeme’, reprinted in Gödel 1986a, I: 144–194.
Goodman, N. (1943), ‘On the Simplicity of Ideas’, Journal of Symbolic Logic
8: 107–121.
Goodman, N. (1949), ‘The Logical Simplicity of Predicates’, Journal of Symbolic
Logic 14: 32–41.
Goodman, N. (1950), ‘An Improvement in the Theory of Simplicity’, Journal of
Symbolic Logic 14: 228–229.
Goodman, N. (1951), The Structure of Appearance, Cambridge, Massachusetts:
Harvard University Press.
Goodman, N. (1952), ‘New Notes on Simplicity’, Journal of Symbolic Logic
17: 189–191.
Goodman, N. (1954), Fact, Fiction, and Forecast, London: Athlone Press.
Goodman, N. (1955), ‘Axiomatic Measurement of Simplicity’, Journal of Philosophy
52: 709–722.
Goodman, N. (1963), ‘The Significance of Die logische Aufbau der Welt’, in Schilpp
1963: 545–558.
Goodman, N. (1990[1941]), A Study of Qualities, Harvard Dissertations in Philosophy
Series, Cambridge, Massachusetts: Harvard University Press.
Goodman, N., and Leonard, H. S. (1940), ‘The Calculus of Individuals and its Uses’,
Journal of Symbolic Logic 5: 45–55.
Goodman, N., and Quine, W. V. O. (1940), ‘Elimination of Extra-Logical Postulates’,
Journal of Symbolic Logic 5: 104–109.
Goodman, N., and Quine, W. V. O. (1947), ‘Steps Towards a Constructive
Nominalism’, Journal of Symbolic Logic 12: 105–122.
214 Bibliography

Grabiner, J. V. (1981), The Origins of Cauchy’s Rigorous Calculus, Cambridge,

Massachusetts: MIT Press.
Graffi, G. (2001), 200 Years of Syntax: A Critical Survey, Amsterdam: John
Benjamins.
Grattan-Guinness, I. (2000), The Search for Mathematical Roots, 1870–1940: Logics,
Set Theories and the Foundations of Mathematics from Cantor through Russell to
Gödel, Princeton: Princeton University Press.
Grelling K. (1936), ‘The Logical Paradoxes’, Mind 45: 481–486.
Guicciardini, N. (1989), The Development of Newtonian Calculus in Britain,
1700–1800, Cambridge: Cambridge University Press.
Hallett, M. (1984), Cantorian Set Theory and Limitation of Size, Oxford: Oxford
University Press.
Hanfling, O. (ed.) (1981), Essential Reading in Logical Positivism, Oxford: Basil
Blackwell.
Harris, R. A. (1993), The Linguistic Wars, Oxford: Oxford University Press.
Harris, Z. S. (1946), ‘From Morpheme to Utterance’, Language 27: 161–183.
Harris, Z. S. (1951), Methods in Structural Linguistics, Chicago: University of
Chicago Press.
Harris, Z. S. (1952), ‘Discourse Analysis’, Language 28: 1–30.
Harris, Z. S. (1957), ‘Co-occurrence and Transformation in Linguistic Structure’,
Language 33: 283–340.
Harris, Z. S. (1991), The Theory of Language and Information, Oxford: Clarendon
Press.
Harwood, F. W. (1955), ‘Axiomatic Syntax: The Construction and Evaluation of a
Syntactic Calculus’, Language 31: 409–413.
Haugen, E. (1951), ‘Directions in Modern Linguistics’, Language 25: 211–222.
Hauser, M., Chomsky, N., and Fitch, T. (2002), ‘The Faculty of Language: What is it,
Who has it, and How did it evolve’, Science 298: 1569–1579.
Heine, E. H. (1870), ‘Über trigonometrische Reihen’, Journal für reine angewandte
Mathematische 71: 353–365.
Hilbert, D. (1899), Grundlagen der Geometrie, Leipzig: Teubner.
Hilbert, D. (1900), ‘Über den Zahlbegriff’, Jahresbericht der Deutschen
Mathematiker-Vereinigung 8: 180–184.
Hilbert, D. (1932[1918]), ‘Axiomatisches Denken’, reprinted in Gesammelte
Abhandlungen, Berlin: Springer-Verlag, vol. III, 146–156.
Hilbert, D. (1967a[1927]), ‘Die Grundlagen der Mathematik’, Abhandlungen aus dem
mathematischen Seminar der Hamburgische Universität, 6, 65–85, reprinted in
English translation in van Heijenoort 1967: 464–479.
Hilbert, D. (1967b[1904]), ‘Über die Grundlagen der Logik und der Arithmetik’,
Verhandlungen des Dritten Internationalen Mathematiker-Kongress in Heidelberg
vom 8 bis 13. August 1904, 174–185, Leipzig: Teubner, reprinted in English
translation in van Heijenoort 1967: 130–138.
Hilbert, D. (1998[1922]), ‘Neubegründung der Mathematik’, Erste Mitteilung,
Abhandlungen aus dem mathematischen Seminar der Hamburgische Universität,
reprinted in English translation in Mancosu 1998: 198–214.
 Bibliography 215

Hilbert, D., and Bernays, P. (1934–1939), Grundlagen der Mathematik, vols. I and II,
Berlin: Verlag von Julius Springer.
Hintikka, J. (1995), From Dedekind to Gödel: Essays on the Development of the
Foundations of Mathematics, London and New York: Kluwer.
Hiorth, F. (1974), Noam Chomsky: Linguistics and Philosophy, Oslo, Bergen, Tromsφ:
Universitetsforlaget.
Hjelmslev, L. (1961[1943]), Omkring sprogteoriens grundlaeggelse, reprinted in
English translation as Prolegomena to a Theory of Language by F. J. Whitfield,
Madison: University of Wisconsin Press.
Hockett, C. F. (1953), review of Shannon and Weaver’s The Mathematical Theory of
Communication, Language 29: 69–93.
Hockett, C. F. (1955), A Manual of Phonology, Chicago: University of Chicago
Press.
Hockett, C. F. (1967), Language, Mathematics and Linguistics, The Hague: Mouton de
Gruyter.
Hockett, C. F. (1968), The State of the Art, Janua Linguarum Series, Minor, 73, New
York: Cornell University Press.
Hrbacek, K., and Jech, T. (1984), Introduction to Set Theory, 2nd edn, New York and
Basel: Marcel Dekker, Inc.
Huck, G. J., and Goldsmith, J. A. (1995), Ideology and Linguistic Theory: Noam
Chomsky and the Deep Structure Debates, New York and London: Routledge.
Huybregts, R., and Riemsdijk, H. V. (1982), Noam Chomsky on the Generative
Enterprise, New Jersey: Foris Publishers.
Hylton, P. (1990), Russell, Idealism, and the Emergence of Analytic Philosophy,
Oxford: Oxford University Press.
Hymes, D., and Fought, J. (1981[1975]), American Structuralism, The Hague, Paris,
and New York: Mouton.
Jech, T. (1991), Set Theory, 2nd edn, Berlin and New York: Springer.
Johnson, P. E. (1972), A History of Set Theory, Boston: Prindle, Weber, & Schmidt.
Kasher, A. (ed.), (1991), The Chomskyan Turn, Oxford: Basil Blackwell.
Katz, J. (1981), Language and Other Abstract Objects, Oxford: Basil Blackwood.
Kino, A., Myhill, J., and Vesley, R. E. (eds.) (1970), Intuitionism and Proof-Theory:
Proceedings of the Summer Conference at Buffalo, N.Y., 1968, Amsterdam:
North-Holland.
Kleene, S. (1952), Introduction to Metamathematics, Amsterdam: North-Holland.
Kneale, W., and Kneale, M. (1962), The Development of Logic, Oxford: Clarendon
Press.
Koerner, E. K. (1999), Linguistic Historiography: Projects and Prospects, Amsterdam:
John Benjamins.
Kreisel, G. (1958), ‘Hilbert’s Programme’, Dialectica 12: 346–372.
Kuhn, T. (1962), The Structure of Scientific Revolutions, Chicago: University of
Chicago Press.
Lagrange, J. L. (1867[1760]), ‘Essai d’une nouvelle méthode pour déterminer les
maxima et les minima des formules intégrales indéfinies’, Miscellanea Taurinensia
2: 173–195, reprinted in Oeuvres, I (1867), 355–362.
216 Bibliography

Lakoff, G. (1989), ‘Philosophical Speculation and Cognitive Science’, Philosophical

Psychology 2: 55–76.
Lees, R. (1957), review of Syntactic Structures, Language 33: 375–408.
Leibniz, G. W. (1863a[1686]), ‘De geometria recondita et analysi indivisibilum atque
infinitorum’, Acta Eruditorum 5, reprinted in Gerhardt, C. I. (ed.) (1863),
Mathematische Schriften, Abth. 2, Band III, 226–235.
Leibniz, G. W. (1863b[1695]), Letter to Nieuwentijdt, in Gerhardt, C. I. (ed.) (1863),
Mathematische Schriften, Abth. 2, Band V, 332–336.
Leibniz, G. W. (1863c[1684]), ‘Nova Methodus pro maximis et minimis, itemque
tangentibus, quae nec fractas nec irrationales quantitates moratur, et singulare pro
illi calculi genus’, Acta Eruditorum 3: 467–473, reprinted in Gerhardt, C. I. (ed.)
(1863), Mathematische Schriften, Abth. 2, Band III.
Leibniz, G. W. (1978[1666]), Dissertatio de Arte Combinatoria, in Gerhardt, C. I. (ed.)
(1875–1890; reprinted 1978), Die philosophischen Schriften, vol. IV, Berlin:
Weidmannsche Buchhandlung.
Leśniewski, S. (1992a), Collected Works, 2 vols., ed. Surma, S. J., et al., Dordrecht,
Bristol, and London: Kluwer.
Leśniewski, S. (1992b[1914]), ‘Is a Class of Classes not Subordinated to Themselves,
Subordinate to Itself?’, reprinted in Leśniewski 1992a: 115–128.
Leśniewski, S. (1992c[1927]), ‘On the Foundations of Mathematics’, reprinted in
Leśniewski 1992a: 174–382.
Linsky, L. (1952), Semantics and the Philosophy of Language, Urbana: University of
Illinois Press.
Luschei, E. C. (1962), The Logical Systems of Lésniewski, Amsterdam: North-Holland.
Lyons, J. (1970), Chomsky, London: Fontana/Collins.
McCall, S. (ed.), (1967), Polish Logic: 1920–1939, Oxford: Oxford University
Press.
Mancosu, P. (1998), From Brouwer to Hilbert, Oxford: Oxford University Press.
Mandelbrot, B. (1954), ‘Structure formelle des textes et communication: deux études’,
Word 10: 1–27.
Manning, K. R. (1975), ‘The Emergence of the Weierstrassian Approach to Complex
Analysis’, Archive for History of Exact Science 14: 297–383.
Martin, R. (1999), ‘Case, the Extended Projection Principle, and Minimalism’, in
Epstein and Hornstein 1999: 1–26.
Matthews, P. H. (1993), Grammatical Theory in the United States from Bloomfield to
Chomsky, Cambridge: Cambridge University Press.
Matthews, P. H. (2001), A Short History of Structuralist Linguistics, Cambridge:
Cambridge University Press.
Miller, G. A. (1951), Language and Communication, New York: McGraw-Hill.
Montague, R. (1970), ‘English as a Formal Language’, reprinted in Montague
1974: 188–221.
Montague, R. (1974), Formal Philosophy: Selected Papers of Richard Montague, ed.
R. H. Thomason, New Haven: Yale University Press.
Murray, O. S. (1980), ‘Gatekeepers and the “Chomskian Revolution” ’, Journal of the
History of the Behavioural Sciences 16: 73–88.
 Bibliography 217

Murray, O. S. (1994), Theory Groups and the Study of Language in North America: A
Social History, Amsterdam: John Benjamins.
Nagel, T. (1995) Other Minds: Critical Essays, 1969–1994, Oxford: Oxford University
Press.
Newmeyer, F. J. (1980), Linguistic Theory in America: The First Quarter-Century of
Transformational Grammar, New York and London: Academic Press.
Newmeyer, F. J. (1986), Linguistic Theory in America, 2nd edn, Orlando, Florida:
Academic Press.
Newmeyer, F. J. (1996), Generative Linguistics: A Historical Perspective, London:
Routledge.
Newton, I. (1967[1665]), ‘A Method for Finding Theorems Concerning Quaestiones
de Maximis et Minimis’, in Whiteside, D. T. (ed.), The Mathematical Papers of
Isaac Newton 1664–1666, Cambridge: Cambridge University Press, vol. I, 272–297.
Newton, I. (1969[1670]), ‘De methodis serierum et fluxionum’, in Whiteside, D. T.
(ed.), The Mathematical Papers of Isaac Newton 1664–1666, Cambridge:
Cambridge University Press, vol. III, 32–353.
Newton, I. (1972[1687/1726]), Philosophiae Naturalis Principia Mathematica, 3rd
edn, ed. A. Koyré and B. Cohen; 2 vols., Cambridge: Cambridge University Press.
Orenstein, A. (2002), W.V. Quine, Boston, Massachusetts: Acumen Press.
Nunes, J. (1999), ‘Linearization of Chains and Phonetic Realization of Chain Links’,
in Epstein and Hornstein 1999: 217–250.
Otero, C. P. (1994a), ‘The Emergence of Transformational Generative Grammar’, in
Otero 1994b, 1–36.
Otero, C. P. (ed.), (1994b), Noam Chomsky: Critical Assessments, 4 vols., New York
and London: Routledge.
Peano, G. (1959a[1889]), Arithmetices Principia Nova Methodo Exposita, reprinted in
Unione Matematica Italiana (eds.), Opera scelte 2: 20–55.
Peano, G. (1959b[1891]), ‘Principii di Logica Matematica’, reprinted in Unione
Matematica Italiana (eds.), Opera scelte 2: 92–101.
Post, E. (1944), ‘Recursively Enumerable Sets of Positive Integers and their Decision
Problems’, Bulletin of the American Mathematical Society 50: 284–316.
Postal, P. M. (2004), Skeptical Linguistic Essays, Oxford: Oxford University Press.
Priest, Graham (2001), An Introduction to Non-Classical Logic, Cambridge:
Cambridge University Press.
Quine, W. V. O. (1940), Mathematical Logic, Cambridge, Massachusetts: Harvard
University Press.
Quine, W. V. O. (1946), ‘Concatenation as a Basis for Arithmetic’, Journal of
Symbolic Logic 11: 105–114.
Quine, W. V. O. (1951), Mathematical Logic, revised 2nd edn, Cambridge,
Massachusetts: Harvard University Press.
Quine, W. V. O. (1953), From a Logical Point of View, Cambridge, Massachusetts:
Harvard University Press.
Quine, W. V. O. (1985), The Time of My Life: An Autobiography, Cambridge,
Massachusetts: MIT Press.
Reichenbach, H. (1947), Elements of Symbolic Logic, New York: Macmillan.
218 Bibliography

Reid, C. (1996), Hilbert, New York: Springer-Verlag.

Richardson, A. (1998), Carnap’s Construction of the World: The Aufbau and the
Emergence of Logical Positivism, Cambridge: Cambridge University Press.
Riemann, G. F. B. (1990[1867]), ‘Über die Darstellbarkeit einer Function durch eine
trigonometrische Reihe’, reprinted in Gesammelte mathematische Werke, Berlin:
Springer, 227–271.
Robinson, A. (1996), Non-Standard Analysis, revised edn, Princeton: Princeton
University Press.
Rosenbloom, P. C. (1950), Elements of Mathematical Logic, New York: Dover.
Russell, B. A. W. (1914), Our Knowledge of the External World as a Field for
Scientific Method in Philosophy, La Salle: Open Court.
Russell, B. A. W. (1938[1903]), Principles of Mathematics, New York: W. W.
Norton.
Sapir, E. (1929), ‘The Status of Linguistics as a Science’, Language 5: 207–214.
Schaffer, S. (1997), ‘What is Science?’, in J. Krige and P. Dominique (eds.), Science in
the Twentieth Century, London: Harwood Academic Publishers.
Schilpp, P. A. (ed.), (1963), The Philosophy of Rudolf Carnap, La Salle: Open Court.
Shannon, C., and Weaver, W. (1949), The Mathematical Theory of Communication,
Urbana: University of Illinois Press.
Sheynin, O. B. (1988), ‘A. A. Markov’s Work on Probability’, Archive for the History
of Exact Science 39: 337–377.
Simon, H. A. (1955), ‘On a Class of Skew Distribution Functions’, Biometrika
42: 425–440.
Smith, L. (1986), Behaviorism and Logical Positivism: A Reassessment of The
Alliance, Stanford: Stanford University Press.
Smith, N. V. (2004), Chomsky: Ideas and Ideals, 2nd edn, Cambridge: Cambridge
University Press.
Soare, R. I. (1996), ‘Computability and Recursion’, Bulletin of Symbolic Logic
2: 284–321.
Steinberg, D. (1999), ‘How the Anti-Mentalist Skeletons in Chomsky’s Closet Make
Psychological Fiction of his Grammars’, in Embleton et al. 1999: 267–282.
Struik, D. J. (ed.), (1969), A Source Book in Mathematics, 1200–1800, Cambridge,
Massachusetts: Harvard University Press.
Suppe, F. (2000), ‘Axiomatization’, in Newton-Smith, W. H. (ed.), A Companion to the
Philosophy of Science, Oxford: Blackwell, 9–12.
Tarski, A. (1944), ‘The Semantic Conception of Truth and the Foundation of
Semantics’, Philosophy and Phenomenological Research 4: 341–375.
Tarski, A. (1956a[1931]), ‘The Concept of Truth in Formalized Languages’, reprinted
in English translation in Tarski 1956c: 152–278.
Tarski, A. (1956b[1936]), ‘The Establishment of Scientific Semantics’, reprinted in
English translation in Tarski 1956c: 401–408.
Tarski, A. (1956c), Logic, Semantics and Metamathematics, English translation by J.
H. Woodger, Oxford: Clarendon Press.
Tomalin, M. (2002), ‘The Formal Origins of Syntactic Theory’, Ling̀ua
112(10): 827–848.
 Bibliography 219

Tomalin, M. (2003), ‘Goodman, Quine, and Chomsky: from a Grammatical Point of

View’, Ling̀ua 113(2): 1223–1253.
Tomalin, M. (2004), ‘Leonard Bloomfield: Linguistics and Mathematics’,
Historiographia Linguistica 31(1): 105–136.
Turing, A. (1936), ‘On Computable Numbers, With an Application to the
Entscheidungsproblem’, Proceedings of the London Mathematical Society
42: 230–265.
Uriagereka, J. (1998), Rhyme and Reason, Cambridge, Massachusetts: MIT Press.
Uriagereka, J. (1999), ‘Multiple Spell-Out’, in Epstein and Hornstein 1999: 251–282.
Van Heijenoort, J. (1967), From Frege to Gödel: A Source Book in Mathematical
Logic, 1879–1931, Cambridge, Massachusetts: Harvard University Press.
Van Stigt, W. P. (1990), Brouwer’s Intuitionism, Amsterdam: North-Holland.
Weinberg, J. R. (1936), An Examination of Logical Positivism, London: Kegan Paul.
Weiss, A. P. (1925), ‘One Set of Postulates for a Behaviourist Psychology’,
Psychology Review 32: 83–87.
Wells, R. (1947), ‘Immediate Constituents’, Language 23: 81–117.
White, M. (1952), ‘The Analytic and the Synthetic: An Untenable Divide’, in Linsky
1952: 272–286.
Whitehead, A. N. (1898), A Treatise on Universal Algebra with Applications,
Cambridge: Cambridge University Press.
Whitehead, A. N., and Russell, B. A. W. (1925[1910]), Principia Mathematica, 3
vols., Cambridge: Cambridge University Press.
Wilder, R. L. (1952), Introduction to the Foundations of Mathematics, New York: John
Wiley.
Westfall, R. S. (1980), Never at Rest: A Biography of Isaac Newton, Cambridge:
Cambridge University Press.
Wolénski, J. (1989), Logic and Philosophy in the Lvov–Warsaw School, Dordrecht:
Kluwer Academic Press.
Young, J. W. (1917), Foundational Concepts of Algebra and Geometry, New York:
Macmillan Co.
Zipf, G. (1935), The Psycho-Biology of Language: An Introduction to Dynamic
Philology, Cambridge, Massachusetts: MIT Press.
Zipf, G. (1949), Human Behavior and the Principle of Least Effort: An Introduction To
Human Ecology, New York: Hafner Publishing Co.
 Index

λ-calculus, 61 and nineteenth-century mathematics, 27,

λ-definability, 61 28
and pre-TGG linguistic theories, 19
Aarsleff, H., 201 and proof theory, 60
abstraction, 49, 76, 163 and syntactic theory, 50
active–passive constructions, 121, 134, 161, and TGG, 2, 5, 19, 60, 177–179, 186, 188,
163, 164, 166 203
Ajdukiewicz, K., 67–72, 88, 103, 105, 126, and the formal sciences, 3
129, 130, 184, 204 and the foundations of mathematics, 51
algebra, 26, 33, 34, 38, 48, 51, 56, 73, 97 axioms, 2, 3, 5, 27, 28, 31, 34–36, 39, 41–43,
algebraic topology, 46 49, 50, 55–57, 59, 68, 69, 73, 83, 101,
alphabet, 49, 147, 178 123, 124, 176–178, 188
America, 11, 33, 44, 77, 78, 101, 102 Ayer, A. J., 204
analysis, 25–28, 31, 38, 39, 201, 202
analysis of language, 106 Bach, E., 4, 5, 7
analytic philosophy, 186 Baker, G. P., 202
analytic truth, 88, 205 Bar-Hillel, Y.
anthropology, 6 and Bloomfield, 105, 106
anti-mentalism, 17 and Carnap, 8, 105, 126, 127, 135, 138, 204,
antinomies, 69 207
apparent economy, 118 and Categorial Grammar, 71, 72, 203
applied mathematics, 2 and Chomsky, 8, 18, 132, 135, 138, 140
Aristotle, 46 and discovery procedures, 71
arithmetic, 21, 28, 31–34, 38, 39, 42, 44, 45, and distributional procedures, 126
47, 55, 78, 86, 158 and formal consequence, 133
automata theory, 104 and formal languages, 127
automatic speech recognition, 187 and Harwood, 58
auxiliary predicates, 86 and logical syntax, 131
axiomatic geometry, 38, 48 and logical systems, 54, 88
axiomatic number theory, 39 and recursion, 54, 65, 66, 168, 184, 191,
axiomatic-deductive method, 125 193
and Bloch, 58 and semantics, 127, 128, 135, 136
and Bloomfield, 55–57, 93, 177, 184 and synonymy, 127
and geometry, 38, 39 and TGG, 73, 170, 184, 203, 204
and Harwood, 59, 186 and the axiomatic-deductive method, 184
and Hilbert, 203 and transformations, 127, 161–163, 167,
and linguistic theory, 54, 60, 64, 73, 106 207
and Logicism, 60 and truth conditions, 128

220
 Index 221

on the relationship between logic and Cambridge, 9, 34, 175

linguistics, 109, 125, 126, 131, 133, Cambridge (Mass.), 78
138, 184 Cantor, 28–32, 34, 46, 47, 205
Barrow, I., 22 Carnap, R., 89
Barsky, R. F., 110–112, 206, 207 and Bar-Hillel, 138
behaviourism, 154, 201, 207 and Bloomfield, 130, 131
Bernoulli, J., 24 and constructional system theory, 73–77,
biological theory, 1 79, 151, 204
biology, 6 and formal languages, 89–91, 102, 134
Birkhoff, G., 26 and formal linguistics, 102
Bloch, B., 12, 54, 57–59, 67, 102, 175, 178, and formal syntax, 54, 93, 105
184, 203 and Formalism, 176, 181
Bloomfield, L. and formation rules, 126, 161, 167, 168, 207
and Carnap, 130, 131 and logical empiricism, 9, 11, 12, 150, 151,
and distributional linguistics, 103, 104 153, 154
and formal languages, 59 and logical epistemology, 78, 79, 87
and formal syntax, 54, 93, 105 and logical syntax, 106, 126, 168, 176
and Formalism, 13, 14, 94, 95, 180 and meaning in linguistic theory, 91, 180
and mathematics, 58 and meaning postulates, 129, 130
and meaning in linguistic theory, 95, 141, and metalanguage, 89
149 and natural languages, 90, 181
and precision in linguistics, 67 and proof theory, 49
and the axiomatic-deductive method, 54–57, and semantics, 128–130
178, 184, 203 and transformation rules, 127, 133, 134,
and the foundations of mathematics, 96–101 162–164, 167, 186, 207
and the Vienna Circle, 93 and transformations, 92, 93
influence on Bar-Hillel, 105, 130 criticism of Chomsky, 138
influence on Bloch, 58 influence on Bar-Hillel, 8, 105, 106, 127,
on linguistics as a science, 57, 198, 208 128, 130, 204, 207
on the relationship between logic and influence on Bloomfield, 93, 94
linguistics, 88 influence on Chomsky, 111
on the relationship between mathematics influence on Goodman, 54
and linguistics, 93, 94, 205 influence on Goodman and Quine, 78, 84
on the relationship between mathematics influence on Harris, 104, 132
and natural language, 56, 57 influence on Hjelmslev, 102
the influence of Weiss, 56 influence on linguistics, 208
Bloomfieldians, 7 influence on Quine, 48, 77, 205
Bohr, N., 196 on formation and transformation rules,
Boole, G., 33 90–92, 162, 164
Borgström, C. H. J., 106, 107 on the relationship between formal and
Bradley, F. H., 33 natural language, 128
Brouwer, L. E. J., 44–48, 103 on the relationship between linguistics and
Burali-Forti, C., 32, 34 logic, 138
Burali-Forti Paradox, 32 on the relationship between syntax and
logic, 126
calculus, 18, 21–28, 55, 202 case, 196
calculus of individuals, 79, 122, 123, 150, 197, categorial analysis, 72
198 Categorial Grammar, 70, 72, 103, 126, 184,
calculus of relations, 37 203
calculus of variations, 26, 201 Cauchy, A. L., 26–28, 55, 177
222 Index

Cavalieri, 22 and the autonomy of syntax, 141, 149, 180,

Cherry, C. E., 143 181
Chomsky, N. and the axiomatic-deductive method, 177,
‘LSS’, 109, 126, 133–137 178, 188, 203
‘SSA’, 121–124, 135, 136, 150, 151, 153, and the calculus of individuals, 123, 125,
206 135
‘TMDL’, 145, 147, 148, 178, 207 and the development of TGG, 3, 4, 12, 108,
and anti-mentalism, 17 140
and Bar-Hillel, 18, 138 and the formal sciences, 6, 18, 111, 198, 208
and Carnap, 137, 138 and the foundations of mathematics, 135
and constructional system theory, 12, 73, and the history of linguistics, 7
74, 125, 144, 155 and the innateness hypothesis, 185
and constructive nominalism, 8, 122, 156, and the post-Bloomfieldians, 18, 149, 175
186, 198, 201 and the Spec–Head relation, 197
and discovery procedures, 125, 132, 140, and transformations, 133, 159–168, 181
149, 151, 154, 208 as an aging dictator, 16
and empiricism, 185, 207 at Harvard, 11, 110
and epistemology, 207 at MIT, 110, 175
and evaluation procedures, 150, 154, biographical information, 109, 206
155 influence of analytic philosophy, 88
and finite-state grammars, 145–148 influence of Bar-Hillel, 12, 67, 140, 170,
and formal consequence, 133, 134 184, 204
and formal languages, 124, 136 influence of Carnap, 111
and formal linguistic theories, 8, 133, 176, influence of Goodman, 9, 14, 109, 112, 113,
177, 182 117, 118, 121, 152, 156
and formal syntax, 53 influence of Goodman and Quine, 7, 17, 18,
and Formalism, 13, 14, 174, 176, 178, 179, 121–123, 150, 153, 159, 185
188 influence of Harris, 110, 112, 114, 121, 122,
and grammaticality, 148 136
and linguistic levels, 156, 157 influence of his mother, 109
and logic, 133, 134, 137, 158, 159 influence of Quine, 9, 112, 118, 133, 135,
and logical empiricism, 11, 154, 185 153, 158
and logical semantics, 134–136, 141 influence of White, 133, 136
and logical syntax and semantics, 8 influence on contemporary culture, 15
and logical systems, 135 LSLT, 3, 111, 115, 117, 118, 124, 137, 150,
and mathematical linguistics, 15 155–158, 164–166, 170, 172, 173, 176,
and meaning, 132–136, 181 177, 181, 184, 195, 206–208
and P-markers, 164 MMH, 17, 110, 112, 114, 160, 162
and politics, 187 MP, 188–190, 194, 197, 198, 206
and predictive grammars, 169 on linguistics as a science, 198
and rationalism, 207 on the origins of TGG, 1, 5, 6, 8–10
and recursion, 137, 171–173, 189–193 on the relationship between logic and
and recursively enumerable sets, 169, 170 linguistics, 109, 126, 133, 162
and simplicity, 18, 84, 109, 112–120, 166, relationship with Bar-Hillel, 8
167, 172, 195 relationship with Goodman, 111
and ‘SSA’, 8 relationship with Harris, 110, 111
and statistical models of language, 141, 145, relationship with Quine, 110, 205
148, 149, 207 SS, 4, 149, 178, 203, 207
and syntactic theory, 206 Chomsky, W. Z., 109
and syntax and semantics, 179 Christianity, 25
 Index 223

Church, A., 15, 49–51, 55, 61, 64 Copenhagen Circle, 101, 107
classical sets, 69 copy theory of movement, 196
cognitive function, 190 corpora, 71, 114–116, 118, 120, 121, 124, 132,
competing grammars, 115, 116, 119, 120, 149, 150, 154, 155, 168, 169
154–156, 185, 194, 195 counterargument, 83
computability theory, 61, 130, 190–195 Curry, A., 96
computer science, 15, 199
Conant, J. B., 110 D’Alembert, 26, 27
concatenation, 78, 86, 117, 157–159 data, 2, 103, 154, 177
conjectures, 177 Dauben, J. W., 202
conjunction, 34, 36, 41, 42, 65, 86, 92 Davis, S., 207
consistency of axioms, 43 De Morgan, A., 33
consolidated grammar, 118, 119 declarative, 52
constant function, 64 Dedekind, R. J. W., 28, 29, 31, 46, 61
constructional system theory, 114 deduction, 3, 27, 35, 36, 55, 91, 102, 130, 132,
and Carnap, 74, 75, 79, 151, 204 152, 179, 186
and constructional levels, 204 deep-structure, 199
and discovery procedures, 124, 144 definienda, 122
and extralogical bases, 77, 80, 112, 117, definiendum, 37
118, 158 definition, 27, 41, 55, 63, 77, 91, 102, 117,
and extralogical primitives, 82 128, 129, 140, 156, 176–178
and formal languages, 14 dependence, 39
and Formalism, 122 derivation, 41, 43, 75, 83, 171, 174, 179, 189,
and Goodman and Quine, 85 196
and linguistic levels, 140, 158 derivatives, 23, 25
and logical analysis, 106 Descartes, 22
and multigrade relations, 80 descriptive adequacy, 193
and nominalism, 85, 87, 121 descriptivism, 17
and simplicity, 80–84, 113, 118, 121, 155, desideratum, 131
185, 193 differential equations, 201
and syntactic analysis, 121–123 differentiation, 22–24, 27, 28, 202
and syntactic theory, 123, 124, 158, discourse analysis, 161
159 discourse structure, 162
influence on syntactic theory, 73, 122 discovery procedures, 19, 71, 104, 124, 140,
influence on TGG, 9, 12, 14, 17, 19, 117, 149–151, 155, 169, 185, 208
125, 131, 140, 151, 154–156, 185, 197, discrete expression, 193
206 disjunction, 36, 41, 42, 76
neglect of in historiographical studies of distributional linguistics, 103, 127, 128, 176,
TGG, 14 183, 204
the development of, 19, 54 Dugac, P., 202
constructive nominalism, 7, 8, 12, 19, 85, Dummett, M. A. A., 202
87, 89, 109, 122, 156, 185, 186, 197,
198 economy, 14, 80–84, 113, 114, 116, 120, 167,
continuity, 47 195, 196, 198
continuous number line, 28 electromagnetism, 34
continuum, 46, 47 elegance, 196
contradictions, 32, 43, 69, 102 Elegant Syntax, 196
conversations, 85 elementary syntax, 49
conversions, 117–120, 160, 170–173, 181, elements, 3, 30, 31, 69, 75–77, 80, 92, 115,
196 119, 123, 127, 157, 160, 178, 190
224 Index

empiricism, 7–9, 15, 17–19, 65–67, 84, 87, and Formalism, 39, 44, 89–91, 102, 128, 203
123–125, 132, 151–153, 155, 156, 185, and Intuitionism, 45
207 and Leibniz, 32
epistemology, 74, 75, 89, 153, 207 and logic, 60, 69, 73, 78, 84, 88–90, 153
EPP feature, 197 and logical syntax, 49
EST, 188, 199 and Logicism, 33, 68
ethics, 74 and mathematics, 97
Euler, L., 24, 25, 27 and meaning, 128–129, 207
Europe, 44, 78, 102 and natural languages, 49, 52, 53, 58, 69, 70,
evaluation procedures, 12, 19, 115, 117, 125, 73, 90, 123, 127, 130, 139, 182, 184
140, 149, 150, 155, 156, 185, 195, 206 and nominalism, 69, 85, 86, 123
Ewald, W., 203 and science, 90
existential operator, 41 and simplicity, 119
explanatory adequacy, 193, 194 and syntactic theory, 56, 70, 73, 123, 124,
explanatory theory, 150 128, 131, 133, 135, 178, 187
extralogical bases, 118, 124, 158 and the axiomatic-deductive method, 56,
extralogical postulate, 81 59
extralogical primitives, 80–82, 153 formal linguistics, 102, 106, 121, 158,
175–177, 193
faculty of language, 190, 191, 194 formal notation, 160, 161
fallacy, 98, 100 formal proofs, 42
Fermat, P. de, 22 formal relations, 134
Feyerabend, P., 196 formal rules, 90
Fine, N., 110 formal sciences
finite alphabets, 178 and Formalism, 95
finite arithmetic, 30, 38 and linguistic theory, 1, 4, 7, 16, 18, 66, 71,
finite classes, 165 73, 95, 104, 109, 183
finite formulae, 100 and recursion, 64, 65, 191, 192
finite grammars, 168–171, 193 and TGG, 1, 2, 4, 5, 15, 18–19, 140, 159,
finite lexicon, 58, 59, 178 174, 187, 188
finite mathematics, 50, 60, 63, 169, 190, 193 and the axiomatic-deductive method, 56
finite proofs, 63, 168, 179 and the post-Bloomfieldians, 10, 183
finite sequences, 52, 64, 69 definition of, 2, 3
finite sets, 30, 31, 85, 142, 158, 169, 178, 190, in the twentieth century, 21, 54
193 formal symbols, 41, 43, 50
finite state grammars, 143, 145–148 formal syntax, 49, 53, 54, 86, 89, 91, 93–94,
finite state machines, 141, 144, 148 96, 105, 117, 128, 166, 180
finite strings, 123 formal systems, 14, 40, 43–45, 49, 50, 52, 58,
Fitch, T., 190–193 65, 73, 80, 89, 91, 95, 102, 128, 137, 178,
FLN hypothesis, 190–192 179, 186
fluxional calculus, 23, 201 and TGG, 5
formal consequence, 134, 162 formal theories, 5, 8, 53, 106, 107, 121, 155,
formal deduction, 89 160, 162, 176, 177, 208
formal discourse, 94 Formalism
formal features, 188, 197 and Bloomfield, 94–96, 101, 104, 183
formal grammars, 119–121, 125, 169, 179, and Carnap, 90, 91, 181
190, 193, 195 and Chomsky, 122
formal inference, 43 and Church, 49, 50
formal languages and formal languages, 52
and concatenation, 78, 158 and Gödel, 45
 Index 225

and Harris, 103 generative grammar, 1, 6, 11, 16–20, 139, 150,

and Hjelmslev, 102 154, 169, 182, 187–191, 193, 194,
and Intuitionism, 45 197–200, 206–208
and Logicism, 39, 42 generative procedures, 31, 140
and mathematics, 48 Generative Semantics, 13, 14
and meaning, 44, 50, 51, 91, 95, 130, 180 geometric intuition, 38
and Quine, 49 geometry, 34, 39, 48, 55, 56, 58, 97
and recursion, 60, 61, 63, 168 German, 77, 89, 90
and textbooks, 18 Germany, 33
and TGG, 13, 17, 123, 140, 174–177, 180, Gestalt psychology, 76
181, 186 Gödel, K., 15, 45, 54, 61, 63, 64, 74, 104,
and the foundations crisis, 18, 21, 32, 38, 191–193, 203
40, 45, 183 Gödel’s incompleteness theorem, 45, 61
and the post-Bloomfieldians, 102 Goldsmith, J. A., 14, 17, 199
and well-formedness, 70 Goodman, N.
and Wilder, 51 and analytic and synthetic truths, 87
in the twentieth century, 183 and constructional system theory, 14, 112,
formation rules, 92, 126, 163 121, 123, 124, 140, 197
formula, 40, 41, 43, 82, 100, 101 and constructive nominalism, 7, 54, 85–88,
formulae, 23, 40, 41, 43, 87, 91, 178 122, 185
Fought, 10, 11, 205 and induction, 9, 87, 152, 155
foundations crisis, 21, 29, 47, 51, 54, 94, and Leonard, 79, 80
96–100, 183 and logical empiricism, 17, 18, 84, 150,
foundations of algebra, 48 151, 153
foundations of analysis, 21, 22, 25, 28, 177 and logical epistemology, 87
foundations of arithmetic, 28, 33 and Quine, 9, 78–80, 84
foundations of geometry, 34, 38, 39, 44 and simplicity, 80–84, 116–118, 120, 193,
foundations of logic, 39 197
foundations of mathematics, 8, 15, 18, 32, and the calculus of individuals, 123, 125,
39–40, 45–48, 51, 55, 85, 99, 103, 135, 198
136, 151 influence of Carnap, 74, 78, 79, 85, 89,
foundations of number theory, 28, 31 205
foundations of science, 97 influence on Chomsky, 7, 8, 11, 12, 14, 18,
Fraenkel, 31 88, 109–112, 116–118, 121, 122, 150,
Frank, 74 154, 159, 185, 186
Frege, 33, 34, 73, 111, 202 influence on Harris, 104
French, 65, 66 neglect of in historiographical studies, 8
Friedman, 204 rejection of Chomsky, 111
Fries, 71, 126, 205 Grabiner, J. V., 26, 202
function, 37, 42, 43, 52, 61–65, 68, 147, 156, Graffi, G., 18
158, 169, 170, 175 grammar, 2, 5, 7, 52, 112, 113, 115, 117, 118,
function-theoretic grammar, 52 121, 141, 144, 145, 147, 150, 154, 156,
functional categories, 197 160, 165, 167–171, 173, 174, 178, 181,
fundamental theorem of the calculus, 22 191, 196, 206
grammar construction, 113–116, 127, 142,
Gassendi, P., 25 150, 154, 156, 159, 169, 197
general recursive functions, 64 grammar evaluation, 12, 19, 115, 117, 125,
general theory of series, 38 140, 149, 150, 154–156, 181, 185, 195,
Generalised Phrase Structure Grammar, 12, 206
206 grammatical levels, 206
226 Index

grammatical transformations, 140, 162, 165, heterological paradox, 98

167, 181 heuristic procedure, 83, 120
grammaticality, 59, 60, 66, 70, 72, 106, 114, Heyting, A., 47
115, 120–122, 145, 147–149, 152, 157, hierarchies, 68, 70, 158, 189, 190, 197
160, 164, 169, 178–180, 186, 197 hierarchy of types, 122
Grassmann, H., 33 Hilbert, D., 14, 15, 32, 38–44, 47–50, 61, 63,
Grattan-Guinness, I., 44, 202, 203 89, 91, 98, 102, 160, 203
Grelling, K., 205 Hiorth, F., 7, 8, 124
Grelling’s paradox, 98, 205 historiography, 7, 16, 17, 183
Hjelmslev, L., 54, 101, 102, 105, 107, 181
Hahn, H., 73, 74 Hockett, C. F., 6, 7, 11, 13, 101, 102, 110,
Halle, 143 143–145, 148, 169, 175, 198, 207, 208
Hallett, M., 202 homonymity, 122, 123, 206
Hanfling, O., 204 Hrbacek, K., 202
Harris, R., 14 Huck, G. J., 14, 15, 17, 199
Harris, Z. S. Humboldt, 5
and discovery procedures, 71, 117, 121, Hylton, P., 202
124, 149, 204 Hymes, D., 10, 11, 205
and distributional procedures, 104, 114, hypothesis, 97, 187, 191
126–127, 146
and distributional techniques, 121 I-language, 195, 206
and formal languages, 105 illogicality, 28, 34
and formal syntax, 54, 103, 105, 107 immediate constituent analysis, 103
and Formalism, 103 immediate dependent, 64
and grammar construction, 113, 114 Immediate-Successor, 65
and grammatical transformations, 126, imperfections, 68
159–164, 166–168, 207 implication, 2, 36, 37, 41, 42, 88, 92, 113, 120,
and linguistic theory, 110 130, 132, 152, 160, 162, 181
and mathematics, 6, 104, 107, 175 incompleteness, 45
and meaning in linguistic theory, 141, 149 induction, 9, 18, 61–64, 151–153, 155, 193
as a structural linguist, 71 inference, 41, 43, 44, 63, 92, 93, 127, 134,
defended by Chomsky, 138 135, 152, 162, 163, 168
his ‘theoretical nihilism’, 110, 113 infinite classes, 96, 100, 101
influence of Carnap, 104, 105, 162 infinite regress, 21
influence on Bar-Hillel, 106 infinite sequence, 31
influence on Chomsky, 7, 9, 18, 110, 112, infinite series, 27
122, 136, 185 infinite sets, 29–31, 100, 169, 190
influence on post-Bloomfieldians, 71 infinite structures, 147, 168–171, 173, 174,
intellectual development of, 103 190, 193
negative attitude towards simplicity, 113, infinitesimals, 22–26, 202
114 infinity, 29, 31, 63
on logic and linguistics, 131, 159, 162 informal discourse, 94
relationship with Chomsky, 110, 111 information theory, 141, 145
Harvard, 8, 11, 77, 78, 84, 85, 110–112, 151 initial function, 64
Harwood, F. W., 54, 58, 59, 67, 88, 163, 164, innateness hypothesis, 187, 194
167, 184, 186 inscriptional nominalism, 86, 122
Haugen, E., 107 integrals, 25
Hauser, M., 190–193 integration, 22, 27, 28, 201, 202
Hebrew, 109 interpretable features, 196
Heine, E. H., 29 intuition, 42, 43, 45, 85, 117, 169, 176
 Index 227

Intuitionism, 18, 21, 32, 45, 46, 51 and logical empiricism, 9, 151
irrational numbers, 100 and mathematics, 101
iteration, 171 and meaning, 13, 95, 132, 136, 138, 180
and recursion, 65, 66, 170
Jakobson, R., 143 and simplicity, 113, 115, 116, 119, 196
Jech, T., 202 and stochastic processes, 141, 143, 144
Jena, 73 and syntax and semantics, 18, 179
Joos, M., 102 and TGG, 1–3, 7, 9, 11, 19, 108, 110, 156
and the axiomatic-deductive method, 2, 57,
Kasher, A., 204 178
Katz, J., 208 and the formal sciences, 1, 4–7, 54, 183, 200
kernel, 165, 173, 174 and the foundations crisis, 96
Kleene, S., 15, 50, 51, 54, 55, 61, 63, 65, 203 and the post-Bloomfieldians, 6, 10, 110, 176
Kneale, W., 202 and transformations, 127
Koerner, E., 17, 201 linguistics, 183–186
Köhler, 76 and constructional system theory, 19, 122
Kuhn, T., 16, 110 and empirical science, 67
and Formalism, 5, 60, 89, 101, 102, 107, 176
labels, 190, 196 and historiography, 13
Lagrange, J. L., 26 and logic, 5, 89, 94, 105–107, 125,
Landen, J., 24 130–134, 136–138, 159, 208
language faculty, 194 and mathematics, 5, 6, 19, 54, 57, 93, 94,
language of science, 97 96, 103, 175
Latin, 90 and nominalism, 122, 155
Lees, R., 1–3, 5, 188, 198, 208 and philosophy, 8
Leibniz, 22, 24, 32, 201 and psychology, 95
Lenneberg, E., 187 and recursion, 65
Leonard, 80 and semiotics, 102
Leśniewski, S., 67–70, 85, 103 and simplicity criteria, 115
lexical features, 188 and statistics, 143, 148
lexical items, 189, 190 and structuralism, 102
lexical properties, 189, 197 and the axiomatic-deductive method, 27,
Lexical-Functional Grammar, 12 55–57, 60, 177
lexicalist theory, 197 and the formal sciences, 16, 21, 32, 48, 73,
lexicon, 197 95, 105, 106, 198
limits, 23, 26, 28, 100, 147 and the foundations crisis, 97, 98
linear grammar, 120, 171 and the post-Bloomfieldians, 151, 176
linguistic levels, 115, 117, 140, 156, 157 at MIT, 16
linguistic theory Chomsky’s introduction to, 110
and constructional system theory, 14, 151, distributional linguistics, 104, 110, 127
155, 158 structural linguistics, 127
and discovery procedures, 150, 154, 185 Linsky, L., 133
and distribution procedures, 128 logic
and evaluation procedures, 154 and algebra, 33
and explanatory adequacy, 193 and analysis, 26, 27
and Formalism, 102, 104, 121, 177 and arithmetic, 39
and linguistic levels, 157, 158 and constructional system theory, 79
and logic, 12, 19, 51, 52, 73, 94, 109, 121, and Formalism, 39, 49–51, 89, 175, 182
124, 131, 133, 135–137, 158, 162, 176, and inference, 162
184 and Intuitionism, 46, 203
228 Index

logic (cont.) Lukasiewicz, J., 67, 103

and linguistics, 5, 11, 12, 19, 51, 52, 58, 71, Luschei, E. C., 204, 205
91, 93, 94, 105, 106, 109, 125, 126, Lvov, 67
130–132, 136–138, 148, 158, 159, 162, Lvov-Warsaw school, 54, 67, 102, 103, 128,
175, 176, 184, 199 129, 184, 204
and Logicism, 33, 35, 36, 38, 39, 42
and mathematics, 35, 48, 73, 74, 94 machine translation, 187
and nominalism, 85 Maclaurin, C., 24
and operators, 33 Mancosu, P., 203
and recursion, 10, 19 Mandelbrot, 149
and relations, 33 Markov, A., 141, 142
and set theory, 33, 122 Markov process, 142, 144, 146, 148
and TGG, 5, 6, 8, 60, 148, 151, 159 mathematics
and the formal sciences, 2, 7, 18 and algebra, 48
and the foundations crisis, 33, 103 and analysis, 21, 22, 25
and the Lvov-Warsaw school, 67, 68, 103 and Bar-Hillel, 105
Boolean logic, 33 and Bloomfield, 55, 56, 94, 96, 99, 184
classical logic, 2, 33, 46 and Chomsky, 6, 9, 110, 151, 175, 182, 199
conditional logic, 38 and deduction, 35
constructional system theory, 84 and Formalism, 40, 41, 43, 44, 48, 50, 89,
fuzzy logic, 38 91, 95, 106
in the nineteenth century, 33, 202 and inference, 43
non-classical logic, 38 and Intuitionism, 45, 46, 203
origins of, 32, 202 and linguistics, 5–7, 10, 11, 54, 56, 93, 94,
symbolic logic, 2, 32–35, 38, 48–51, 58, 67, 96, 97, 101, 102, 175, 183
73, 74, 76, 89, 95, 98, 107, 122, 123, and Logicism, 33–36, 38, 39, 68, 94
130, 131, 137, 158, 163, 175, 206 and natural language, 56–58, 94, 97, 101
logical analysis, 74, 90, 104, 106, 107, 161 and philosophy, 34
logical arguments, 89 and proofs, 27, 41, 42, 62
logical axioms, 42, 153 and set theory, 31, 40, 96, 100
logical constants, 76 and symbols, 102
logical empiricism, 2, 9, 11, 84, 88, 150, 151, and TGG, 5, 8, 18, 123, 135
153, 155, 156, 159, 185 and the axiomatic-deductive method, 5, 50,
logical epistemology, 74, 75, 78, 79, 87 55, 93
logical form, 188 and the formal sciences, 2, 74
logical implication, 134 and the foundations crisis, 18, 29, 32, 33,
logical inference, 41, 159, 162 40, 47
logical operators, 86 Matthews, P. H., 13, 14, 16–18, 201, 204, 205,
logical positivism, 74, 204, 207 207, 208
logical relations, 93 meaning, 2, 3, 13, 44, 50, 68–70, 80, 87, 88,
logical semantics, 8, 69, 126, 129, 132, 135, 91, 95, 96, 98, 102, 126–136, 138, 149,
141, 184 176, 179–181
logical syntax, 3, 8, 12, 18, 19, 48, 49, 52, 69, meaning postulates, 130
73, 77, 78, 88–90, 92, 93, 96, 98, mechanical grammars, 155
104–106, 126, 128, 130–132, 134–136, mechanical procedures, 71, 72, 83, 95, 117,
138, 162, 168, 176, 186, 207 154
logical systems, 35, 49, 52, 133, 135 mentalism, 130
Logicism, 18, 21, 32, 33, 38, 42, 45, 51, 60, mereology, 68, 79, 80
94, 186 Merge, 189, 190, 195
logistic theory, 102 metalanguage, 65, 89, 128, 129, 178
 Index 229

metamathematics, 10, 13–15, 43, 44, 48–51, and meaning, 69, 87, 132, 134, 141, 181,
65, 89, 102, 106, 175, 176, 178, 179, 181, 182
182 and Montague Grammar, 139
metaphysics, 74, 78 and pragmatics, 69
Miller, G. A., 143, 187 and recursion, 67, 169, 170, 174
minimal search, 197 and simplicity, 113, 195, 206
Minsky, M., 110 and statistical modelling, 11, 141–149
MIT, 16, 145, 175, 206 and structuralism, 143
models, 142, 145 and synonyms, 133
modern science, 199 and TGG, 1, 141, 177
Montague, R., 138, 139, 207 and the axiomatic-deductive method, 2, 5,
Montague Semantics, 138 55, 57, 58, 184
morphemes, 103, 112, 123, 127, 144, 157, and the foundations crisis, 98, 135, 136
204 and the foundations of mathematics, 103
MP, 194, 195, 197, 198 and the real world, 2
multivariate calculus, 21 and transformations, 161–164
Murray, O. S., 12, 16, 17, 205, 206 and Zipfian distributions, 141
complexity of, 134, 156
Nagel, T., 207 English, 52, 53, 139
natural languages French, 65
adequacy criteria for, 112, 130, 193 German, 90, 207
and and constructional system theory, 123 interdisciplinary approaches to the study of,
and Categorial Grammar, 70, 72 9, 10, 175
and cognition, 56 Latin, 90
and competing grammars, 154 Polish, 207
and constructional system theory, 14, 121, properties of, 194
122 natural numbers, 64
and constructive nominalism, 89 natural philosophy, 23
and Esperanto, 53 natural science, 199
and finite-state grammars, 145, 147 negation, 36, 37, 41, 42, 76
and formal consequence, 134 neologisms, 181
and formal languages, 52, 53, 58, 70, 90, 95, network, 83, 116
123, 134–136, 139, 147, 178, 182 Neurath, O., 73, 93
and Formalism, 183 Newmeyer, F. J., 11–13, 16, 207
and grammaticality, 60, 120 Newton, 22–24, 201
and induction, 152, 153 nominalism, 7, 8, 54, 68, 69, 79, 82, 85–89,
and inference, 134 121–124, 155, 158, 201, 208
and infinity, 168, 169 non-standard analysis, 202
and knowledge, 175 North America, 16
and linguistic data, 154 noun phrase, 65, 117
and linguistic levels, 157 number theory, 21, 26–32, 38, 39, 43, 48, 51,
and linguistic theory, 2 202
and linguists, 127, 133 number-theoretic functions, 61–63
and logic, 38, 51, 52, 58, 59, 73, 89, 94, 97, numeration, 189
126, 130, 131, 137, 158, 159, 175, 176,
184 object language, 86
and logical epistemology, 74 ontology, 68, 75
and logical syntax, 90, 92, 93, 104 optimality conditions, 120
and mathematics, 45, 56, 94, 97, 101, 104, ordinal numbers, 31
105, 142, 183 ordinary language, 175
230 Index

Orenstein, A., 205 principles, 5, 9, 14, 22, 33, 35, 40, 42, 46, 47,
Otero, C. P., 15, 206 151, 158, 193, 199
Oxford, 9, 175 Principles and Parameters, 12, 188, 195, 199,
206
paradigm, 150 probability, 33, 142, 144, 149
paradoxes, 21, 32, 35, 69, 85, 99, 100, 205 proof theory, 40–45, 48–51, 60, 62, 65, 89,
Pascal, 22 102, 128, 160, 168, 174, 178–180, 184
passive transformation, 166 proofs, 27, 41–43, 45, 49, 179, 183, 186
passives, 166 propositional calculus, 33, 36, 49
Peano, G., 28, 33, 34, 61, 62, 65 propositional functions, 37, 52
Peirce, C., 33 psychology, 6, 7, 11, 56, 58, 67, 76, 92, 95,
permutation, 165, 166 112, 130
philosophy, 2, 6, 8, 15, 33, 34, 45, 46, 67, 68, pure mathematics, 2
73, 74, 84, 87, 88, 105, 110, 136,
151–155, 175, 207 qualities, 78, 113, 153
philosophy of language, 133 quantification, 33, 48, 87, 118
philosophy of mathematics, 104 quantifiers, 49
phonemes, 57, 119, 127, 141, 143, 164, 204 quantum logic, 38
phonemic analysis, 57, 58, 143, 157, 171 quantum mechanics, 15
phonetic analysis, 143, 158 quasi-analysis, 76
phonetic features, 98, 99 Quine, W. V. O.
phonological analysis, 143, 194 and analytic and synthetic truths, 87, 88, 133
phonological form, 188 and concatenation, 78, 86, 158
phrasal projections, 194, 197 and constructional system theory, 121, 140,
phrase, 98 197
phrase structure grammar, 120, 145, 157, 164, and constructive nominalism, 85–88, 122
166, 167, 170, 172–174, 178, 181, 197, and Goodman, 78, 80, 84
203, 208 and logic, 48, 49, 55
phrase structure level, 172 and logical empiricism, 9, 18, 135, 151, 153
physics, 199 and logical epistemology, 87
Piattelli-Palmarini, M., 199 and logical syntax, 78
Platonism, 85, 86, 197 and meaning, 135
poetry, 141 and simplicity, 80, 81, 116–118, 197
Polish logicians, 67, 71, 102 and the Vienna Circle, 77
Post, 15, 54, 61, 64, 104, 169, 170 his autobiography, 205
post-Bloomfieldians, 6, 7, 9, 10, 12, 13, 15, 16, his Ph.D., 77
18, 71, 102–104, 107, 136, 141, 143, 146, influence of Carnap, 74, 77–79, 85, 89
149–152, 154, 168, 169, 175, 176, 183, influence on Chomsky, 7, 8, 12, 14, 18, 88,
205, 208 110, 112, 118, 150, 153, 154, 159, 185,
Postal, P., 198, 199 186
postulates, 55–58, 116, 130, 133, 184 influence on Harris, 105
Prague, 77 neglect of in historiographical studies, 8
Prall, D., 78 relationship with Chomsky, 110, 111
predicate calculus, 36, 37, 49, 50
predicates, 62, 80, 82, 83, 86, 118, 123 rational numbers, 28
predictive grammars, 169 rationalism, 7, 9, 17, 88, 154, 195, 207
primitive recursion, 63 real economy, 118
primitive recursive functions, 63, 64, 203 real numbers, 39, 47
primitives, 75, 80, 81, 83, 91, 116, 122, 133, recursion, 19, 43, 61–63, 65, 66, 96, 167, 168,
188, 190, 197 171–173, 188, 190–193
 Index 231

Recursion Convention, 192 sentence, 52, 53, 58, 60, 65, 66, 70–72, 87, 88,
recursive components, 60, 64, 169–174, 190, 91–94, 117, 128, 129, 145–148, 159, 160,
197 164, 166, 173, 178, 180, 186
recursive definitions, 19, 62–67, 106, 131, 137, sentence generation, 179
140, 168, 170, 184, 186, 191, 193 sentential calculus, 68, 103
recursive function theory, 10, 43, 50, 54, set theory, 18, 21, 29–37, 39, 40, 46–48, 50,
60–65, 67, 73, 96, 104, 168, 169, 174, 51, 60, 68–70, 86, 96, 98, 100, 101, 122,
175, 191–193 202, 203, 205
recursive rules, 12, 19, 170–174 sets, 30–32, 34, 47, 64, 69, 100, 123, 134, 157,
recursive steps, 169, 189 165, 168, 169, 197
recursively enumerable languages, 174 Shannon, C., 141–146, 148
recursively enumerable sets, 61, 64, 168–170 Sheynin, O. B., 141
reductionism, 157 similarity relations, 76, 124
Reichenbach, H., 51, 52, 105 simplicity, 185
relativity theory, 15 and constructional system theory, 9, 12,
Richardson, A., 204 80–84, 116, 118, 193
Rieger, 208 and linguistic theory, 9, 12, 113, 115–117,
Riemann, G. F. B., 29 130, 196
rigorisation, 18, 29, 55, 67 and recursion, 197
rigour, 2, 7, 26–28, 55, 64, 107, 137 and TGG, 109, 113–120, 151, 155, 156,
Rosenbloom, P. C., 52 166, 171, 193–195
Rosenfield, 208 simplicity criteria, 2, 14, 18, 19, 59, 81, 82,
Russell, B., 34–39, 42, 48, 49, 60, 67, 69, 70, 87, 106, 112–121, 125, 149, 155, 156,
76, 98, 111, 122, 205 166, 167, 171, 172, 185, 193–195, 197,
Russell’s paradox, 68, 98 198, 206
Russia, 109 theory-external, 195
theory-internal, 81, 84, 113, 121, 132,
Sapir, E., 103, 104, 198, 208 195–197, 206
Saussure, 102 Skolem, 103
Schaffer, S., 201 Smith, L., 207
schemata, 64, 203 Smith, N. V., 206
scheme, 71, 85, 153 Soare, R. I., 192
Schlick, M., 74, 77, 93 sociology, 130
Schröder, E., 33 solipsism, 46
Schnitt, 28 spatial intuition, 38
Schützenberger, M., 187 Spec–Head, 197
science, 6, 40, 57, 65, 74, 89, 90, 94, 97, 104, speech forms, 97, 100
130, 153, 185, 192, 198–200 spell-out, 196
science of language, 55 statistical modelling, 3, 11, 141, 143–145, 148,
scientific language, 97 149
scientific method, 3, 186 statisticians, 141
scientific theory, 2, 9, 16, 90, 116, 151, 158 Steinberg, D., 17
self-reference, 35, 41, 62, 96, 98, 99, 190, 193, Stevenson, C., 78
205 stochastic grammars, 140, 141, 143–145, 148,
semantic, 70 149
semantics, 3, 18, 19, 68, 70, 95, 96, 102, 106, stochastic processes, 19, 141–143, 145, 148
121, 126, 128–130, 134–136, 138, 141, structural linguistics, 11, 18, 71, 72, 95, 111,
149, 179–181, 183, 186, 194, 205 125, 127, 130, 138, 143
semiotics, 102 structuralism, 10
sense data, 76, 153 Struik, D. J., 201
232 Index

subject-predicate analysis, 33, 52 Tarski, A., 67, 78, 85, 105, 129, 130, 204,
subsets, 30, 32, 58–60, 90, 165, 176 207
substrings, 181 tautology, 36, 88, 129
successor function, 64 Tawardowski, K., 67, 68
Suppe, F., 201 Taylor, B., 24, 27
surface-structure, 199 terminal symbols, 164
syllables, 141 TGG
symbols, 45, 49, 50, 56, 72, 91, 92, 95, 103, and Bar-Hillel, 184, 203
106, 123, 142, 145, 160, 164, 165, 176, and constructional system theory, 14, 73,
179 125
synonymy, 88, 127, 133, 134, 136 and constructive nominalism, 8, 12, 197
syntactic categories, 164 and discovery procedures, 71, 124
syntactic classes, 124 and Goodman, 185
syntactic objects, 189, 190, 196 and induction, 152
syntactic theory, 197 and logical empiricism, 155, 159
and constructional system theory, 73, and logical syntax, 73, 105
112–114, 121, 123–125, 140 and meaning, 91, 181
and constructive nominalism, 122, 123, 186 and Montague Grammar, 139
and discovery procedures, 154 and previous syntactic theories, 5
and empiricism, 17, 156 and recursion, 60, 64, 67, 140, 168–170,
and explanatory adequacy, 193 172–175, 190
and Formalism, 44, 122, 136, 175, 177, and simplicity, 14, 59, 84, 117, 120, 121,
179–182, 184, 186 185, 193–195, 206
and induction, 13 and the axiomatic-deductive method, 5, 60,
and logic, 8, 69, 70, 73, 89, 91, 109, 114, 125, 177, 178, 203
184 and the formal sciences, 2, 15, 19, 96, 174,
and logical empiricism, 77 176, 186, 187
and logical epistemology, 74 and the history of ideas, 6
and mathematics, 10, 11, 14, 21, 22, 58, 105 and the post-Bloomfieldians, 7, 10–13, 16,
and nominalism, 19, 86 18, 169
and philosophy, 15 and transformations, 91, 93, 140, 159,
and recursion, 12, 60, 61, 64, 66, 67, 168, 163–168
169, 184 as a scientific theory, 1, 4, 16, 198, 199
and simplicity, 117, 132 as an explanatory theory, 150
and stochastic processes, 142 Chomsky on the origins of, 8, 10, 151
and TGG, 4, 6, 10, 11, 16, 108, 111, 114, definition of, 3
140, 149 Goodman’s knowledge of, 111
and the axiomatic-deductive method, 58, 60, Harris’ attitude towards, 111
184, 188 Hockett on the origins of, 6
and the formal sciences, 1, 4, 16, 19, 109, in the 1960s, 13, 14
199 influence of analytic philosophy, 88
and the post-Bloomfieldians, 10, 12 influence of Carnap, 104, 186
and transformations, 92, 140, 164 influence of Formalism, 50, 140, 174, 179,
categorial grammar, 70, 72, 126 180, 186, 188
generative grammar, 188 influence of logic, 8
syntacticians, 131, 141, 168 influence of Logicism, 60
syntax, 70, 90, 91, 95, 126, 129, 132, 138, 140, influence of Quine, 78, 88
141, 143–145, 149, 161, 163, 169, 179, influence of the formal sciences, 4, 5, 18,
180, 186, 191, 207, 208 19, 21
synthetic truth, 205 influence of White, 186
 Index 233

other histories of, 2, 4–8, 12–17, 186, 187 UG, 4, 195, 206
reception of, 2 ungrammaticality, 59, 60, 66, 148
the definition of, 4, 108 universal operator, 41
the development of, 1, 4, 17, 21, 54, 102,
107, 108, 140, 183, 186–188, 195, Van Stigt, W. P., 46, 203
198 verb phrase, 117
the reception of, 1 verbs, 56, 65, 143
theorems, 2, 3, 5, 26, 27, 31, 32, 45, 50, 63, 81, verifiability, 49, 180
103, 123, 124, 178 Vienna, 67, 73, 74, 77, 205
theoretical syntax, 49, 50 Vienna Circle, 67, 73, 74, 77, 89, 93, 94, 98,
theory construction, 1 204
theory of descriptions, 163
theory of functions, 25, 27 Warsaw, 68, 207
theory of logical types, 35, 36, 68, 70, 122, Warsaw-Lvov school, 130
202, 205 Weaver, 141–144, 148, 207
theory of meaning, 135, 180 Weierstrass, K., 28, 46, 202
theory of reference, 135 Weiss, A. P., 56, 203
theory of relations, 76 well-formedness, 70, 179, 180
theory of simplicity, 84 well-ordering, 100
theory-internal economy, 113, 114, 195 Wells, R., 102–105, 175
theory-internal simplicity, 19, 81, 84, 121, Wertheimer, 76
132, 193–197, 206 Weyl, H., 46
thought and language, 45 wh-questions, 166
traditional grammars, 107 White, M., 87, 88, 110, 133, 135
transfinite arithmetic, 31, 32, 38, 42 Whitehead, A. N., 15, 34–39, 42, 60, 67–69,
transfinite set theory, 29 76, 77, 111, 122, 202, 205
transformational level, 157, 164 Wilder, R. L., 51, 55
transformations, 12, 19, 90–93, 102, 119–121, Wittgenstein, 111
126, 127, 133, 140, 145, 159–168, 173, Wolénski, J., 204
174, 181, 186, 207 word formation, 53
translation, 89, 126, 137, 168, 176, 204, 207
trigonometric functions, 29 X-bar theory, 194, 197, 199
truth-conditional semantics, 2, 128, 129, 139,
207 yes-or-no questions, 166
Turing, A., 61, 104, 191, 192 Young, J. W., 48, 51, 55
Turing computable functions, 192
Turing machine, 191, 192 Zermelo, E., 31
Turing Thesis, 192 Zipf, G., 141