Argument Structure Constructions versus Lexical

Rules or Derivational Verb Templates

ADELE E. GOLDBERG

Abstract: The idea that correspondences relating grammatical relations and semantics
(argument structure constructions) are needed to account for simple sentence types is reviewed,
clarified, updated and compared with two lexicalist alternatives. Traditional lexical rules
take one verb as ‘input’ and create (or relate) a different verb as ‘output’. More recently,
invisible derivational verb templates have been proposed, which treat argument structure
patterns as zero derivational affixes that combine with a root verb to yield a new
verb. While the derivational template perspective can address several problems that
arise for traditional lexical rules, it still faces problems in accounting for idioms, which
often contain specifications that are not appropriately assigned to individual verbs or
derivational affixes (regarding adjuncts, modification, and inflection). At the same time,
it is clear that verbs play a central role in determining their distribution. The balance
between verbs and phrasal argument structure constructions is addressed via the Principles
of Semantic Coherence and Correspondence together with a usage-based hierarchy of
constructions that contains entries which can include particular verbs and other lexical
material.

1. Introduction

In order to interpret language, speakers need to assign semantic interpretations

to the overtly expressed formal patterns that they witness; conversely, in order
to produce language, speakers need to choose formal patterns to express the
meanings they want to convey. In this way, correlations between surface form
and interpretation constitute the basis of our knowledge of language. Learned
correspondences between form and function, at varying levels of complexity and
abstraction, are constructions (e.g. Fillmore Kay and O’Connor, 1988; Goldberg,
1995, 2006). Examples of various types of constructions are provided in Table 1.
For any linguistic theory to be descriptively adequate, subtle facts about semantics
and use of particular constructions need to be accounted for. For example, restrictions
on register, genre and intonation can be part of our knowledge of individual words,
idioms and productive phrasal patterns as the examples in Table 2 illustrate.
Frequency information is also assigned to constructions, since it is well known to
play a role in acquisition, processing and historical change.

I thank Sam Guttenplan, Ray Jackendoff, Paul Kay, and Stefan Müller for comments on an earlier
draft. They are of course not responsible for the perspective described here. I am grateful for
funding from the Einstein Foundation in Berlin and Freie Universität for supporting this work.
Address for correspondence: Council of the Humanities, 2-C-17 Green Hall, Princeton
University, Princeton, NJ 08540, USA.
Email: [email protected]
Mind & Language, Vol. 28, No. 4 September 2013, pp. 435–465.
© 2013 John Wiley & Sons Ltd
436 A. E. Goldberg

Word e.g. jacuzzi, tattoo, behoove

Word (partially filled) 1 e.g. anti-N, V-ing
Idiom (filled) e.g. long story short, give the Devil his due
Idiom (partially filled) e.g. Jog <someone’s> memory, <someone’s>
for the asking
(minimally filled) e.g. The longer you think about it, the less you
Correlative construction understand
The Xer the Yer
(unfilled)
Ditransitive construction: e.g. He gave her a life-saver;
Subj, V, Obj1, Obj2 He baked her a three-layer cake.

Table 1 Example constructions at varying levels of complexity and abstraction.

Register Genre
restrictions restriction Intonation
Word: Hey Amen Uh-oh
Phrasal idiom: How do you do? Once upon a time On guard!
Phrasal is-to construction limericks (e.g. there ‘Mad Magazine’
construction (e.g. He is to report once was a construction (e.g.
with open slots: at midnight)+ <person> from Him, a doctor?!)
++
<place>)

Table 2 Constructions at varying levels of complexity that specify characteristic register, genre, or
intonation (+ Goldberg and van der Auwera, 2012; ++ Lambrecht, 1990).

This article takes the constructionist viewpoint as its starting point and explains
what this view entails about the relationship between words and larger phrasal
patterns, with a specific focus on the domain of argument structure. In particular,
the idea that argument structure constructions are needed to account for simple
sentence types is reviewed, clarified, updated and compared with two lexicalist
alternatives: traditional lexical rules and the more recently proposed derivational
lexical templates. Purely syntactic alternatives are not discussed here, but see
Goldberg (2006, chapter 10) for a critique.

1 Since Saussure (1916), linguists have acknowledged that roots (or lemmas) and affixes are
conventional pairings of form and function. The constructionist perspective supports a growing
consensus among morphologists that morphemes are emergent generalizations over existing
words in the form of partially filled templates (Ackerman and Nikolaeva, 2004; Blevins, 2001;
Aronoff, 1983; Booij, 2010). In fact, phonologists are coming to a parallel conclusion as well,
namely that phonetic, phonological, and phonotactic constraints arise as generalizations over
word tokens together with constraints imposed by our auditory and production capacities (e.g.
Pierrehumbert, 2001).

© 2013 John Wiley & Sons Ltd

 Argument Structure Constructions versus Lexical Rules 437
2. Argument Structure Constructions
Much of our knowledge of language is quite general and this knowledge is
captured within a network of constructions. That is, constructions such as relative
clauses, various exclamatives, questions, topicalization and so on, form a network
of interrelated knowledge, with more specific constructions inheriting from those
that are more general (Fillmore, Kay and O’Connor, 1988; Goldberg, 1995; Lakoff,
1987; Michaelis and Lambrecht, 1996).
Argument structure constructions are form-function pairings that relate abstract
meanings with arrays of grammatical relations (Goldberg, 1995, 2002, 2006; Jack-
endoff, 2002). As an example of an argument structure construction, consider the
ditransitive construction, which has two object complements (without preposition).
The example in (1) instantiates this construction:
1. ‘She gave him the apple’
www.facebook.com/pages/Adam
The ditransitive construction is reliably associated with the meaning of actual or
potential transfer (Goldberg, 1992; Goldberg et al., 2005; Green, 1974; Levin and
Rappaport Hovav, 2005; Oehrle, 1975; Pinker, 1989). Evidence comes from the
interpretation of nonsense verbs. When people are asked what the nonsense verb
moop means in (2), a full 60% of people respond with ‘give’, and the rest offer
meanings that preserve the meaning of literal or metaphorical transfer (e.g. ‘tell’)
(Ahrens, 1995; Goldberg, 1995):
2. She mooped him something.
Additional evidence comes from the interpretation of certain familiar verbs in the
ditransitive construction. Examples 3a and 3b differ slightly in meaning in that
only 3a, which involves the ditransitive construction, requires that Sam intends
to give Pat the cake. Example 3b can be used if Pat is, for example, a baker-
in-training and Sam bakes the cake for a third party on Pat’s behalf (Green,
1974).
3. a.
. Sam baked Pat a cake.
b. Sam baked a cake for Pat.
We can account for these facts by recognizing an abstract construction, defined
by a set of grammatical relations that includes a verb ({Subjx , V, Obj1y , Obj2z }),
corresponding semantics (‘x intends to cause y to receive z’) and information
structure constraints, if relevant (in this case, Obj1 must be more topical than
Obj2). This construction licenses ditransitive expressions such as those in (1), (2)
and (3a).2

2
The example in (3b) is licensed by a transitive construction together with a benefactive adjunct
(see Goldberg, 2002).

© 2013 John Wiley & Sons Ltd

438 A. E. Goldberg
Recognizing a different, caused-motion construction allows us to account for the
fact that a verb such as sneeze, that does not itself entail caused-motion, can play a
role in a clause that does entail caused-motion as in (4):
4. ‘While waiting for doctors, he sneezed the bullet out of his right nostril’
(www.msnbc.msn.com/id/41026427/ns/world_news-weird_news/).
The caused-motion construction is defined by a different set of grammatical relations
({Subjx , V, Objy , Oblique-pathz }) and corresponding semantics (‘x causes y to move
(to/from) z’). In this way, we can identify more general correspondences between
form and function that are not naturally captured at the level of individual verbs.
These are argument structure constructions (Goldberg, 1995).
There is also ample experimental evidence that argument structure constructions
are associated with meanings, independently of the verbs used in them. For
example, researchers have demonstrated that people rely on constructional meaning
when they encounter nouns used as verbs in novel ways (e.g. to crutch) (Kaschak
and Glenberg, 2000; Goldwater and Markman, 2009). Kako (2006) has shown that
speakers also use constructional meaning when asked to interpret Jabberwocky-type
sentences containing no meaningful open-class words.
A recent experiment in our lab demonstrates in fact that these types of Jabber-
wocky sentences automatically prime words that are semantically related to the
argument structure construction of the sentence (Johnson and Goldberg, 2012). In
particular, in a speeded lexical decision task (participants had to decide whether
individual words were real or nonsense words), words were preceded by one
of four abstract skeletal constructions containing only nonsense open-class words.
For instance, an example of the ditransitive construction was He daxed her the
norp. Target words were either congruent with the hypothesized meaning of the
construction (i.e. a congruent word for the English ditransitive would be give,
handed, or transferred) or were not (e.g. put). Results showed significant priming
for congruent over incongruent target words, both for associated primes (which
occur regularly within the construction) and to a lesser extent, for primes that
are semantically related to the construction but which rarely occur in the con-
struction (e.g. transfer for the ditransitive). Thus Jabberwocky-type instances of
argument structure patterns prime verbs related to their abstract meanings or
functions.
In another recent experiment using Multi Voxel Pattern Analysis, we have
found neural evidence that the ditransitive and caused motion constructions can be
distinguished in areas of the brain known to be involved in semantic combination,
particularly BA 47 and anterior BA 22, even when propositional content, open class
words, complexity and frequency are controlled for (Allen et al., 2012). A separate
sorting study has also found that people judge argument structure constructions to
be just as important to overall sentence meaning as the morphological form of the
main verb (Bencini and Goldberg, 2000).
These studies demonstrate that argument structure is associated with meaning
even when a) the main verb involved is not stored as a verb but as a noun (Kaschak
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 439
and Glenberg, 2000; Goldwater and Markman, 2009); b) the main verb involved
is a nonsense word and thus not stored at all (Kako, 2006; Johnson and Goldberg,
2012); or c) the morphological form of the main verbs involved are controlled
for (Bencini and Goldberg, 2000; Allen et al., 2012). A partial list of argument
structure constructions and their corresponding prototypical semantics is provided
in Table 3:

Form
Label Examples Meaning
Ditransitive Subj, V, Obj, Obj2 X causes Y to
construction She gave him something. receive Z
She daxed him something.
Way construction Subji , V, <possi > way, Obliquepath X creates a path (Z)
She made her way into the room. and moves through
‘Heather handstands her way out of the it.
bathroom’.
Intransitive Subj, V, Obliquepath X moves (to/from)
motion She went down the street ‘skiers Y
construction whooshed down the pristine slopes’
Caused motion Subj, V, Obj, Obliquepath X causes Y to move
construction She put the ball in the box. (to/from) Z
‘he sneezed the bullet right out of his
right nostril’.
Resultative Subj, V, Obj, PredicateAP X causes Y to
construction He made her crazy. become Z
She kissed him unconscious.

Table 3 Incomplete list of English Argument Structure constructions. Each is defined by a set of
grammatical relations, semantics and associated information structure properties (not shown). Quoted
examples were cited from the Internet.

Abstract, phrasal argument structure constructions have been proposed and refined by
many researchers as a way to capture relationships between the form and function
of simple sentences that are not naturally attributed exclusively to the main verb
(e.g. Bergen and Chang, 2005; Booij, 2002; Croft, 2003; Goldberg, 1995, 1999,
2002, 2006; Jackendoff, 1997, 2002; Stefanowitsch and Gries, 2009; Tomasello,
2003).

2.1 Verb Specificity

At the same time that argument structure constructions increase language’s creative
potential, verbs clearly restrict which argument structure patterns they occur in.
© 2013 John Wiley & Sons Ltd
440 A. E. Goldberg
In fact verbs can be quite particular, resisting some patterns that other verbs with
similar general semantics and phonology readily appear in. For example, each of
the expressions in (5)-(7) sounds odd in the given argument structure pattern, even
though other verbs are fully acceptable in the same argument structure pattern
(examples in parentheses):
5. ??The magician vanished the rabbit. (cf. The king banished the rabbit.)
6. ??She explained him the story. (cf. She told/guaranteed him the story.)
7. ?? She considered to tell him. (cf. She wanted to tell him.)
(cf. also e.g. Boas, 2000; Faulhaber, 2011; Herbst, 2011). That is, while many
causative verbs can appear both intransitively and transitively (e.g. melt, break,
redden, etc.), vanish can only appear intransitively (and banish can only appear
transitively). Explain is semantically related to tell and has Latinate phonology
like guarantee and yet it cannot appear in the ditransitive construction while both
tell and guarantee can. Finally, consider does not allow an infinitival verb phrase
complement, while want and the majority of English verbs do. These facts do not
follow automatically from general principles; they need to be accounted for in ways
that make reference to the specific verbs involved.
Moreover, verbs that freely occur in more than one argument structure pattern
are often statistically biased to favor one over the other(s). For example give is
much more likely to occur in the ditransitive (as in 8a) than the ‘caused motion’
construction (as in 8b), whereas sell is more likely to occur in the caused-motion
than the ditransitive (Wasow, 2002, p. 87).
8. a.
. She gave him the apple.
b. She gave the apple to him.
Speakers make use of these biases in on-line comprehension (e.g. Garnsey et al.,
1997; MacDonald et al., 1994). These types of facts are verb-specific and need to
be accounted for by any theory. We return to this issue in Section 6.
But before we further explore the constructionist approach to argument structure
in Sections 4-6, let us first examine two alternative approaches: lexical rules (Section
3.1) and derivational verb templates (Section 3.2). These two types of analyses are
often viewed interchangeably (as in Müller, 2002, 2006; Briscoe and Copestake,
1999 and other HPSG work), but they are distinguished below on the basis of
whether a verb is changed into or related to a distinct verb (by lexical rule), on the
one hand, or whether it is instead combined with a derivational lexical template
to form a distinct verb. It is argued that the former perspective leads to several
problems (also discussed in Goldberg, 1995, 2002), and the latter perspective also
shares one critical problem, discussed in Section 3.2.

3. Alternatives to the Constructionist Approach to Argument Structure

As the head of the verb phrase, the main verb of the sentence is widely assumed to
play the pivotal role in interpretation by specifying the way that overt arguments
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 441
are related to one another. Almost any traditional grammar book, or introductory
logic or linguistic class will likely begin with a discussion of sentence types with a
classification of verbs according to how many arguments they ‘take’. It is generally
assumed, for example that sneeze is intransitive, kick is transitive and put requires
an agent, a theme and location arguments. In this way, basic sentence patterns of
a language are assumed to be determined by syntactic and semantic information of
the main verb alone. For example, the sentence pattern in (9) does seem to be due
to the lexical specifications of put:

9. Pat put the ball on the table.

Thus the traditional alternative to positing argument structure constructions such

as those in Table 3 is to posit lexical valences (Tesnière, 1959) or subcategorization
frames (Chomsky, 1965), which allow each verb to specify which complements it
co-occurs with.

3.1 Lexical Rules

Despite the fact that verbs often appear to determine their argument structure, they
typically can occur in more than one argument structure pattern. Traditionally,
researchers have focused on pairwise alternations including the ‘dative alternation’
(10a-b) and the ‘locative alternation’ (11a-b):

10. a.
. She gave him a book.
b. She gave a book to him.

11. a.
. She loaded the truck with hay.
b. She loaded hay onto the truck.

In order to account for such alternations, researchers have posited lexical rules,
which either take as input a verb with a particular valence and yield a verb with a
distinct valence as output, or which simply serve to relate two static lexical entries
(Jackendoff, 1975). An individual clause, then, involves either one verb or the
other but not both. In particular, neither verb is contained in the other; each is
independent, although the relationship is captured by the ‘rule’. Various approaches
have adopted some version of lexical rules in order to account for systematic
alternations (Bresnan, 1982; Fillmore, 1990; Foley and Van Valin, 1984; Grimshaw,
1990; Pollard and Sag, 1994; Pinker, 1989; Levin and Rappaport Hovav, 1994).
Although it does not involve a traditional alternation, in order to account for a
novel example like (4), lexical rule approaches posit a rule that transforms intransitive
sneeze into a verb that requires a direct object and an oblique path phrase. That is,
instead of combining sneeze with an argument structure construction as discussed
above, the lexical rule approach would posit a new verb, sneeze2 , that itself means
roughly ‘x causes y to move (to) z by sneezing’). This is in fact what Müller (2006,
p. 23) seems to suggests: ’Lexical rule-based approaches assume a lexical rule that
© 2013 John Wiley & Sons Ltd
442 A. E. Goldberg
takes an intransitive (version of a) verb as input and licenses a special lexical item
that selects for an additional object and a secondary predicate.’
At the same time, changes in argument structure have long been recognized to
involve subtle, or not so subtle, changes in meaning. For example, while example
(11b) only requires that some hay is caused to move onto the truck, (11a) requires
an interpretation in which the truck is loaded full, or is affected in some way by
the hay being loaded onto it (Anderson, 1971). In order to capture the differences
in semantics, lexical rules are widely regarded as changing verbs’ meaning (Pinker,
1989). Overt complements are then assumed to be ‘projected’ from the main verb’s
meaning by way of general linking rules (Bresnan and Kanerva, 1989; Dowty, 1991;
Grimshaw, 1990; Gleitman, 1994; Jackendoff, 1983; Pinker, 1989). In this way, for
example, the with-variant of load can be derived by lexical rule from the onto-variant
(see the rule in (12)) or vice versa. Likewise the caused-motion use of sneeze in
(4) is accounted for by a semantic lexical rule that changes the meaning of sneeze
from a sense that takes a single agent argument, to a new sense that essentially
means ‘someone causes something to move along a path by sneezing’ as represented
schematically in (13):
12. load < agent, location, instrument-theme> → load <agent, theme,
location>
13. sneeze <agent> → sneeze < agent, theme, path>
The general linking rules are then claimed to project the agent to subject, the theme
to object and the path phrase to an oblique complement.
Unfortunately, the lexical rule approach suffers several drawbacks. These include
the following, explained below. First, the lexical rule approach requires implausible,
ad hoc verb senses. Secondly, it obscures broader surface generalizations. Thirdly,
since it does not distinguish the verb from the argument structure in the output of
the rule, the lexical rule approach does not account for constraints that hold only
of the verb or only of the construction. I have touched on many of these points
and a few others previously, although several of the examples used here are new
(Goldberg, 1995, 2006).
In addition, there is another drawback that I have not discussed in previous
work. This final drawback holds equally of more recent derivational verb template
approaches to argument structure as well as lexical rule approaches (Rappaport
Hovav and Levin, 1998; Koenig and Davis, 2001; Sag, 2012). Both derivational
verb templates and lexical rules require verbs with implausibly specific arguments,
particularly in the case of VP idioms. This last point is discussed in Section 3.2.

3.1.1 Lexical Rules Require Implausible Verb Senses. As just discussed,

lexical rules generally take a basic sense of a verb and derive an extended, distinct
sense and in this way attribute differences in clausal semantics directly to the
main verb involved. While capturing differences in semantics is clearly essential,
attributing the semantics to the main verb often requires quite implausible and ad
hoc verb senses. If argument structure were projected exclusively from a verb’s
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 443
semantics, we would need special verb senses for each of the verbs in the expressions
in (14) (e.g. Goldberg, 1992, 1995, 2006; Jackendoff, 1990, 1997, 2002):
14. a.
. ‘he was drinking the heart right out of a fine spring afternoon’ (James
Crumley, The Last Good Kiss [1978]).
b. ‘The people of this small town [ . . . ] have been unable to pray Mrs. Smith’s
two little boys home again’ (Mark Turner, personal communication).
c. ‘his thousands of travelling fans [ . . . ] had roared him into the Thomas
and Mack Center ring’ (www.topix.net/wire/world-soccer/manchester-
united).
d. ‘She tried to avoid blinking the tears onto her cheeks’ (Anne Tyler, Dinner
at the Homesick Restaurant [1992]).
e. ‘Demi Moore thinks this will Halle Berry her back to the B List’ (personal
communication 2007).
g. ‘I actually had a moth go up my nose once. I [ . . . ] coughed him out of
my mouth’ (bikeforums.net/archive/index.php/t-292132).

That is, we would need a sense of drink that meant roughly ‘to spend time by
drinking’; a special sense of pray ‘to cause to move by praying,’ a special sense of roar
that entails motion and so on. These senses are implausible in that one doesn’t find
languages that devote unique stems to these meanings. For example, it is unlikely
that one would find a word kamo, meaning ‘to cause to move by coughing’ (cf.
14g), because this is not a situation that is likely to occur regularly enough to
warrant a lexical meaning (Goldberg, 2010).
We would, in fact, require not only one additional verb sense, but also commonly
half a dozen senses. Taking cough, in (15) as an example, we find it occurs with at
least seven different argument structure constructions:

15. a. Pat coughed. Intransitive

b. She coughed the milk out of her nose. Caused motion
c. She coughed a very deep cough. Cognate object (Transitive)
d. She coughed herself hoarse. Resultative
e. She coughed onto the table. Implicit object construction
f. She coughed her way to the emergency room. Way construction
g. She coughed her head off. ‘head’ off construction

Positing additional verb senses implies that it should be possible to find languages
that have unique verb stems, perhaps cloz, kbitz, craz, ziff and burr for the requisite
meanings, but all such senses are quite unlikely. And cough is not unusual, as most
verbs, particularly frequent verbs, appear in several distinct argument structure
patterns.
Positing additional senses for each new argument structure pattern is an ad hoc
way to defend the idea that verbs determine their complement configurations.
When faced with a new complement configuration, one is always free to posit an
additional verb sense.
© 2013 John Wiley & Sons Ltd
 444 A. E. Goldberg
3.1.2 Lexical Rules Obscure Broader Surface Generalizations. Lexical
rules, which relate ‘input’ and ‘output,’ are process-oriented generalizations in
the terminology of Bybee (1985). However, in both phonology and morphol-
ogy, it has become clear that surface or product-oriented generalizations (which
emphasize the ‘output’) are typically broader, and they are obscured by an
over-emphasis on the ‘input’ (Bybee, 1985; Bybee and Slobin, 1982; cf. also
Hayes, 2009). As Kapatsinski (2008) emphasizes, citing Nesset (2008), it is not
that generalizations about phonetic and morphological alternations do not exist,
they clearly do, but they are second-order generalizations over product-oriented
constructions.
The same is true in the domain of syntax: product-oriented generalizations are
broader and more robust (Goldberg, 2002; Michaelis and Ruppenhofer, 2001).
That is, it is profitable to look beyond alternations and consider each argument
structure construction on its own terms. For example, although only (16a-b) can
be paraphrased by a ditransitive expression (given in parentheses), they pattern
together with (16c-e) both syntactically and semantically; the examples in (16)
all involve subjects, objects and oblique path phrases and they all express caused-
motion. As Baker notes, ‘it seems artificial to say that the PP in [examples like
16a] is not a locational path as well’ (Baker, 1997; cf. also Marantz, 1997). The
caused-motion construction captures the generalization across all of the examples
in 16(a-e) (Goldberg, 1995; Pinker, 1989):

f. 16. a.
. Aliza gave a book to Zach. (≈ Aliza gave Zach a book.)
b. Aliza threw a book to Zach. (≈ Aliza threw Zach a book.)
c. Aliza threw a book to the ground.
d. Aliza threw a book toward the front of the room.
e. Aliza threw a book over the hedge.
Moreover, the following example is also strikingly like those in 16 and can be
accounted for by the same caused-motion construction in combination with the
verb load:
17. Aliza loaded the hay onto the wagon.
This generalization is missed when there is too much focus on alternations and the
supposed ‘input’, as there is on lexical rule accounts. A focus on lexical rules could
easily lead one to posit at least three different rules: one for give type verbs, one
for motion type verbs such as throw and one for load type verbs (see Sag, 2012 for
such an analysis). Instead, differences in meaning and choice of prepositions that
are required are more naturally attributable to the verbs and prepositions involved.
Only a single caused-motion construction is needed.
Turning our attention to the ditransitive construction (as in (10a)), we see
that it too allows for a broad generalization if we attend to surface structure
instead of attending to possible paraphrases. Although many linguists continue to
treat (regular) ditransitives and ‘benefactive’ ditransitives (such as Aliza baked Zach
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 445
cupcakes) as distinct constructions because of their different paraphrases (Aliza sent
a book to Zach/Aliza baked cupcakes for Zach), both types of ditransitive examples
pattern alike both semantically and syntactically (Goldberg, 2002).
The constructionist approach captures these surface generalizations very naturally.
When different verbs appear in the same construction, similarities are most naturally
attributed to the construction. When faced with pairs of sentences that share words
and involve different constructions, similarities are most naturally attributed to the
words and differences to the constructions. Paraphrase relationships are captured
by explicit reference to words and/or argument structure constructions that have
related meanings.

3.1.3 Lexical Rules Incorrectly Predict Verb Meaning ↔ Constructional

Meaning. The lexical rule approach predicts that a verb meaning and the
meaning of the construction are isomorphic because they are not distinguished.
The ‘input’ is not carried within the ‘output’ on the traditional understanding
of lexical rules. Rather, the output of a lexical rule is a verb meaning and
that meaning is understood to determine the overall event interpretation of the
clause. There is no separate constructional meaning that is combined with a verb’s
meaning.
In point of fact, when a verb lexically codes a particular meaning, it generally
carries that meaning with it when it appears in other constructions. For example,
the verbs inch and worm lexically encode the meaning of real or metaphori-
cal motion despite difficulties, meanings that are very close to the meaning of
the way-construction (Goldberg, 1995). As expected, they readily occur in this
construction:
18. a.
. Snowy inched her way along the roof of the train.
b. He can’t worm his way out of this one.
Yet because inch and worm lexically encode this meaning, they retain the meaning
even when used in the intransitive construction:
19. a.
. ‘Snowy inched along the roof of the train’ (fanfiction.net/s/8069910/1/
Tintin_the_Tunnel_and_Too_Much_TNT).
b. ‘Rodman can’t worm out of this one’ (twitter.com/terrymeiners/statuses/
185042376262299648).
Similarly, give means ‘give’ when it appears in the ditransitive and in the ‘caused-
motion’ construction. In the latter case, give’s recipient participant role unifies
with the path role of the caused-motion construction. This is permitted because
recipients can be metaphorically construed as a type of metaphorical goal via a
general Transfer as Goal-directed motion metaphor (e.g. ‘He grabbed the house away
from her. She handed it over to him. He passed it on to/down to his daughter.’) The
fact that give lexically constrains its argument to be an animate recipient explains
why this argument cannot be questioned with where (Rappaport Hovav and Levin,
2008):
© 2013 John Wiley & Sons Ltd
446 A. E. Goldberg
20. ??Where did she give the flowers to?
In this way, a verbs’ meaning and the meaning of the argument structure construc-
tion it occurs in are not necessarily in a biunique correspondence. In particular,
the same verb meaning can appear in more than one argument structure con-
struction, even though each argument structure construction has its own distinct
semantics. A Semantic Coherence Principle only requires that in order for a
verbs’ roles to unify with the constructions’ roles, the former be construable as
instances of the latter (see Section 4). Additional roles may be contributed by the
construction.
A closely related problem for lexical rule approaches, which conflate the meaning
of the verb and the meaning of the construction, is that some constraints hold only
of the lexical verb; others hold only of the construction. An example of a constraint
that holds of the construction, not the lexical verb, comes from Zaenen (1991;
cited in Goldberg, 1995). The Dutch impersonal passive construction requires an
atelic situation; see Zaenen, 1991:
21. ??Er werd opgestegen.
There was taken off.
22. Van Schiphol wordt er de hele dag opgestegen.
From Schiphol, there is taking off the whole day.

In (22) the adjunct phrase insures that event is construed iteratively and is therefore
atelic. The simple act of ‘taking off’ designated by the verb alone, remains telic.
One might argue that there are two verbs opgestegen, one telic, one atelic. However,
evidence against this idea is that the verb always takes the auxiliary zijn, which
requires telic verbs:
23. Hij is (dagelijks) opgestegen.
It has taken off (daily).

An example of a constraint that holds of the lexical verb and not the construction,
comes from the English prefix re-. This prefix seems to occur only with verbs that
are lexically causative. The prefix is markedly less acceptable with verbs that are not
themselves causative, even when these verbs appear in causative constructions (see
Table 4):

3 Paul Kay (personal communication 13 March 2012) suggested the following examples as possible
counterexamples to the generalization:
i. ??She repushed him into the closet.
ii. ??She rewhisked more foam off the cappuccino.
But push is not causative (She pushed the wall but it didn’t move). Whisk is causative, but due to
its meaning, once something is whisked, it cannot normally change state in the same way again.
When it can, rewhisk is fine as in the following attested example: ‘rewhisk the [oil and vinegar],
then pour it over the vegetables’ http://www.liposonix-info.com/

© 2013 John Wiley & Sons Ltd

 Argument Structure Constructions versus Lexical Rules 447

Lexically causative Lexically NON-causative verbs

(in causative constructions) in causative constructions
John repositioned the book on the table. ??She reate herself sick.
John refilled the tub with water. ??She recried herself to sleep.
John resprayed paint on the wall. ??The dog rebarked them awake.
John resold him the car. ??She resneezed the foam off the
cappuccino.

Table 4 Re-prefix occurs with (subset of) verbs that are lexically causative. 3

Since the lexical rule approach fails to distinguish the lexical verb from its
argument structure properties, these phenomena, which make reference to only the
argument structure pattern or only the verb, are quite difficult to account for.
The arguments provided so far against lexical rules arise directly or indirectly
from the fact that the lexical rule approach makes it impossible to distinguish the
lexical contribution of the verb from the contribution of the construction. There
is an alternative to lexical rules that can, in principle, avoid this particular problem:
the derivational verb template approach.

3.2 Derivational Verb Templates

Derivational verb templates have been proposed more recently as an alternative to
lexical rules. These templates embed a lexical verb and yield a complex verb that
specifies the formal and semantic properties of what we have been calling argument
structure constructions (Briscoe and Copestake, 1999; Koenig and Davis, 2001;
Meurers, 2001; Müller, 2002, 2006; Müller and Wechsler, forthcoming; Rappaport
Hovav and Levin, 1998; Sag, 2012). This approach essentially treats the argument
structure pattern as an invisible derivational affix that combines with the root verb
to yield a derived verb.
The approach can potentially capture many of the same insights as the con-
structionist approach since the embedded verb root and the invisible derivational
template are, at least in principle, distinguishable. To the extent that this distinc-
tion is emphasized, the derivational lexical template and the phrasal construction
approach are close cousins and I have therefore not attempted to distinguish them
previously.4 They both specify a simple verb, as well as an argument structure
defined by an array of grammatical relations, associated semantics and information
structure. Relevant facts about frequencies, genre, or register can, in principle, be
assigned to the simple verb and/or to the argument structure construction.
The problems with lexical rules discussed above need not hold of the derivational
template approach. In particular, implausible verb senses are not necessarily a

4 Müller (2007, p. 375) also notes, ‘the difference between the two approaches is rather small.’

© 2013 John Wiley & Sons Ltd

448 A. E. Goldberg
problem because the new senses are assumed to be complex combinations of verb
root plus derivational template and can be created on the fly. We do in fact find
languages with productive derivational morphology that allows quite novel verb
meanings to be created (e.g. Croft, 2003). Broader surface generalizations can
be captured by the derivational verb template approach, since the template itself
corresponds to the argument structure construction, regardless of which verbs it
combines with. Finally, the derivational verb template approach need not assume an
isomorphic relationship between verb meaning and what we have been calling the
constructional meaning; instead, constraints can in principle be assigned to either
the verb root or the derivational verb template (see in fact Müller and Wecshler,
forthcoming). Thus the derivational verb template approach avoids many of the
problems with lexical rules. The main difference between the constructionist and
the template approaches is that latter assumes that the combination of verb and
template creates a new verb: a V0. This leads to certain difficulties described below.

3.2.1 Lexical Rules and Derivational Templates Require Verbs with

Implausibly Specific Requirements. Like the lexical rule approach, the
derivational template approach assumes that a verb always determines the number
and specific types of complements of the clause. Therefore, both approaches need to
posit verbs that constrain their arguments in quite specific and ultimately implausibly
ways. That is, in order to account for the example in (24), these approaches need to
posit a verb, laugh (whether simple, on the lexical rule approach; or derived on the
derivational template approach) that can only appear with a possessive determiner
+ the particular noun, way and a path phrase.
24. She laughed her way across town.
Notice that examples are unacceptable if any of these features is missing or altered.
25. a.
. ??She laughed him.
b. ??She laughed her path across town.
c. ?? She laughed a way across town.
But we do not generally find verbs or verbs + derivational affixes that involve such
specific constraints cross linguistically. While verbs often semantically constrain
the type of object they can occur with, we do not find unique verbs or overt
derivational verbal morphemes that specify a bound direct object with one particular
lexical noun.5 This problem is magnified in the case of verb phrase idioms.

5
Moreover, the requirement of both a possessive determiner and the specific noun way appears
to violate locality, a principle that has been strongly defended by those who have argued in favor
of derivational verb templates (e.g. Sag, 2007, 2012). A constraint is local if and only it can be
stated as a constraint on a head and the head’s sisters. The head daughter of a sister node may
also be constrained insofar as the features of the head are shared by its mother node, the sister.
However, if a constraint on V is local, it can’t constrain both the determiner and head noun of
a sister, since only one or the other can be the head of the sister.

© 2013 John Wiley & Sons Ltd

 Argument Structure Constructions versus Lexical Rules 449
3.2.2 Derivational Templates Do Not Extend Naturally to Idioms. The
complex, full syntactic information associated with many VP idioms is far richer
than that associated with individual verbs (again, whether derived with an overt
morpheme or not). For example, idioms often require adjuncts, modifiers, or
conjunction (Fellbaum, 2007):
26. modifier:
look on the bright side =/= ? look on the side
27. adjunct:
taking candy from a baby =/= ? Taking candy.
28. conjunction
eat <someone> out of house and home. =/= ?eat <someone> out of house.
In order to account for (26), verb-based approaches would need a verb look that
specifies not only that it takes a PP phrase headed by on but also that this phrase
must have the modification bright in the NP within the PP. To account for (27), the
lexical rule and derivational template approaches are both forced to say that there is
a verb take that specifies that it must occur with what is normally an adjunct and
thus optional: the phrase from a baby. The simple or complex verb eat (in 28) must
specify that it requires a prepositional phrase that contains a particular conjunction.
Moreover, the lexical approach must admit inflectional properties inside of lexical
derivations, since idioms often specify inflectional properties of their complements.
For example, pull strings must involve strings in the plural (29a-b),
29. a.
. She pulled strings to get him admitted.
b. ?? She pulled a string to get him admitted.
Importantly, the distinction between argument structure constructions and
idiomatic phrases is hard to detect in many particular instances. Table 5 pro-
vides semi-idiosyncratic examples of some of the argument structure constructions
provided earlier in Table 3. It is therefore theoretically desirable to treat idioms
and argument structure constructions such as those in Table 3 alike, which means
treating either both phrasally or both lexically. The argument from implausibly
constrained verbs argues in favor of a phrasal approach.
As far as I can tell, the derivational template approach has been proposed in
response to the evidence in favor of constructional meaning, as a way to retain
the traditional assumption that the main verb ‘projects’ the argument array of
the clause (Chomsky, 1982). This allows it to account for passives and long-
distance dependency relations as has been done in decades past. For example,
passive is assumed to be a lexical rule takes a V0 as input and yields a pas-
sive V0 as output (Bresnan, 1982). The auxiliary verb is then assumed to select
for the passive participle and the adjunct by phrase. A verb’s arguments can
be realized distantly, by transformation, or by a ‘subcat’ or ‘valence’ feature
that keeps track of which arguments are accounted for (Pollard and Sag, 1994;
© 2013 John Wiley & Sons Ltd
450 A. E. Goldberg

ditransitive
give < someone> a kiss
give <someone> a piece of <one’s> mind.
way construction:
work <one’s> way through (<type of >) school.
sleep <one’s> way to the top.
caused-motion:
make <one’s> hair stand on end.
resultative:
eat <oneself> sick
make <oneself> scarce

Table 5 Semi-idiomatic instances of argument structure constructions.

Sag, 2012). But these mechanisms have clear correlates on the phrasal argu-
ment structure approach. Let us examine the constructionist approach in more
detail.

4. Constraints on the Combination of Verb and Construction

Generalizations about how verbs and argument structure constructions combine

are discussed at some length in Goldberg (1995,especially chapters 2 and 5; 2002).
Slots in the argument structure constructions are argument roles of the construction.
These often correspond roughly to traditional semantic roles such as agent, patient,
instrument, source, theme, location, etc. At the same time, because they are defined in
terms of the semantic requirements of particular constructions, argument roles in
this framework are more specific and numerous than traditional thematic roles (see
also Jackendoff, 1990, 2002).
Argument roles capture generalizations over individual verbs’ participant roles.
That is, each distinct sense of a verb is conventionally associated with rich frame
semantic meaning that in part specifies the number and type of slots that are
associated with a given sense of a verb. A subset of those roles, namely those
roles which are lexically profiled, are obligatorily expressed, or, if unexpressed, must
receive a definite interpretation.6 Lexical profiling, following the general spirit of
Langacker (1987), is designed to indicate which participant roles associated with
a verb’s meaning are obligatorily accessed, functioning as focal points within the
scene, achieving a special degree of prominence. Fillmore (1977) similarly notes

6
This generalization is true for English. In many other languages profiled arguments are omissible
as long as they are given and non-focal in the context. Typically in these languages, however,
lexically profiled roles are also expressed by a small set of core grammatical relations, when they
are expressed.

© 2013 John Wiley & Sons Ltd

 Argument Structure Constructions versus Lexical Rules 451
that certain participant roles are obligatorily ‘brought into perspective’ achieving
a certain degree of ‘salience’. The notion of lexical profiling is intended to be a
semantic one: it is a stable aspect of a word sense and can differentiate the meaning
difference between words—cf. buy versus sell (Fillmore, 1977) or rob versus steal
(Goldberg, 1995, p. 47). Participant roles may be highly specific and are often
unique to a particular verb’s meaning; they therefore naturally capture traditional
selectional restrictions.
Two general principles constrain the ways in which the participant roles of a verb
and the argument roles of a construction can be put into correspondence, fused,
or ‘unified’: the Semantic Coherence Principle and the Correspondence Principle
(Goldberg, 1995, chapter 2; 2002).
The Semantic Coherence Principle ensures that in order to unify, the participant
roles of the verb and the argument roles of the construction must be semantically
compatible. In particular, the more specific participant role of the verb must be
construable as an instance of the more general argument role, if the two roles are
to unify. General categorization processes are responsible for this categorization
task and it is always operative. This principle follows from the idea that argument
structure constructions are learned by generalizing over the semantics of instances
of the pattern used with particular verbs (e.g. Tomasello, 1992, 2000, 2003;
Cameron-Faulkner et al., 2003; Goldberg et al., 2004; Goldberg, 2006).
As is the case with lexical items, only certain argument roles are profiled. In par-
ticular the subject, direct object and the second object in ditransitives are considered
profiled. These are the same grammatical relations that receive a special status in
most theories as the set of ‘terms’ which correspond to ‘core,’ ‘nuclear’ or ‘direct’
arguments. These grammatical relations afford a high degree of discourse promi-
nence, being either topical or focal (see Keenan and Comrie, 1977; Fillmore, 1977;
Langacker, 1987 for arguments to this effect.). Specifically, the Correspondence
Principle states that profiled participant roles of the verb must be encoded by pro-
filed argument roles of the construction, with the exception that if a verb has three
profiled roles, one can be represented by an unprofiled argument role (and realized
as an oblique argument). The Correspondence Principle is a default principle.
The intuition behind the Correspondence Principle is that lexical semantics and
discourse pragmatics are in general aligned. That is, the participants that are highly
relevant to a verb’s meaning (the profiled participant roles) are likely to be the ones
that are relevant or important to the discourse, since this particular verb was chosen
from among other lexical alternatives. In particular, the Correspondence Principle
requires that the semantically salient profiled participant roles are encoded by
grammatical relations that provide them a sufficient degree of discourse prominence:
i.e. by profiled argument roles. The Correspondence Principle is a default principle
and is therefore overridden by particular constructions that specify that a particular
argument should be deemphasized and expressed by an oblique or not at all.
Passive, for example is a construction that overrides the Correspondence Principle
and insures that a normally profiled role (e.g. the agent) be optionally expressed in
an oblique by phrase.
© 2013 John Wiley & Sons Ltd
452 A. E. Goldberg
Constructions may add certain roles that are not lexically specified by the verb.
These roles are indicated by a smaller vertical line in the examples below (dashed
line in Goldberg, 1995).
The Semantic Coherence and Correspondence principles may be clarified with a
couple of examples. The participant roles of load are the verb-specific roles: loader,
loaded-theme and container. These roles combine with the argument roles in the
caused motion construction and causative + with constructions as follows:

30. Caused-motion (e.g. Pat loaded hay onto the truck)

CAUSE-MOVE (cause theme path/location)
|
Load (loader loaded-theme container)
31. Causative construction + with construction (e.g. Pat loaded the truck with hay)
CAUSE (cause patient) + INTERMEDIARY(instrument)
|
Load (loader container loaded-theme)

All three of load’s roles are profiled. (Profiling is indicated with boldface). This
includes the loaded-theme role even though that role is optional. When omitted,
it receives a definite interpretation as indicated by the strangeness of the following
mini-conversation (see Fillmore, 1986 for tests to distinguish definite from indefinite
omission):
32. She loaded the trucks. #I wonder what she loaded onto the trucks.
Because all three roles are profiled, one of the roles may be expressed as an
oblique argument, in accordance with the Correspondence Principle. The Semantic
Coherence Principle insures that only semantically compatible roles may be fused.
As indicated above, the loaded-theme role of load may either be construed to be a
type of theme as in (30) or an intermediary as in (31). The container role can either
be construed to be a path/location as in (30) or a patient role as in (31); the patient
role is what results in the interpretation that the container is somehow affected by
the loading.
This example illustrates how argument structure constructions address semantic
similarities and subtle differences between paraphrases. Construing the verb’s roles
as instances of distinct argument roles leads to differences in semantic construals.
The fact that the meanings are related is attributable directly to the shared verb
involved. That is, the verb evokes the same frame semantic scene and the same
profiled participant roles.7

7
On this view, there is no need to say that the with phrase itself designates a theme relation (cf.
e.g. Jackendoff, 1990). Instead, the fact that the hay is interpreted to be loaded onto the truck
even in the with variant is attributed, not to the argument structure construction, but to the
specifications of the verb load.

© 2013 John Wiley & Sons Ltd

 Argument Structure Constructions versus Lexical Rules 453
Another example can be used to illustrate a way that certain unusual uses of verbs
are licensed. Notice that the caused-motion construction is again involved in the
main clause of example (4). In this case the construction itself contributes the theme
and path arguments, while sneeze’s single participant role, the sneezer, is construed
to be a type of causer, which is possible because of our real-world understanding
of what sneezing involves. This example demonstrates how constructions can add
arguments that are not associated directly with the verb involved. Both the theme
argument and the path/location arguments are contributed by the construction.

33. Caused-motion (e.g. he sneezed the bullet out of his right nostril)
CAUSE-MOVE (cause theme path/location)
|
Sneeze (sneezer )

Thus the constructionist account is able to account naturally for paraphrase rela-
tionships and it allows for creative uses of verbs.

5. Argument Structure Constructions Interacting with Other Syntactic

Phenomena

Argument structure constructions are phrasal in the sense that they are not zero
level grammatical categories but rather consist formally of an array of grammatical
relations such as subject, object and oblique. They may, but need not, specify
lexical material as well. For example, the object grammatical relation may specify
particular words (e.g. <coreferential pronoun’s> way).
At the same time, argument structure constructions do not specify phrase
structure trees or word order directly (cf. also Osborne and Gross, 2012). Other
constructions that they combine with do. In particular, a general VP construction
specifies statistical constraints on the ordering of postverbal complements, dependent
on weight and information structure (cf. Wasow, 2002). There has been some
misunderstanding about this in the literature, as it is sometimes assumed that
‘phrasal’ means that a particular tree configuration must be specified and that linear
order is necessarily fixed in advance (e.g. Müller, 2006). But it is not necessary, nor
advantageous to make this assumption. In fact, the same caused motion construction
is involved in both (34a) and (34b) below:
34. a.
. She threw the book to Joe.
b. She threw to Joe the book she had just finished reading.
Müller (2006) even assumes that new argument structure constructions must be
posited when argument structure constructions combine with adjuncts. The obvi-
ously preferable alternative, however, is that an independent adjunct construction(s)
specifies the placement of adjuncts along with their corresponding scope properties.
For example, an adjunct placement construction such as that sketched in (35) (scope
© 2013 John Wiley & Sons Ltd
454 A. E. Goldberg
properties not shown) accounts for the fact that English does not generally allow
adverbs between the verb and direct object (cf. (36)):
35. English: [Adjunct∗ V (Obj) Adjunct∗ (PP∗ Adjunct∗ )∗ ]VP
36. ??She threw hard the book to Joe.
Topicalization and cleft constructions allow for various ‘displaced’ arguments.
These types of long distance dependency constructions generally require that
grammatical relations be satisfied non-locally (but see Sag, 2007). For example in
(37), what bears the grammatical relation Obj2 of the ditransitive.
37. What did he think she gave Pat?
In this way, actual expression typically includes the combination of at least half a
dozen different constructions. For example, the expression in (38) involves the list
of constructions given in (39a-f).
38. What did Aliza give Zach?
39. a.
. Ditransitive construction
b. Non-subject question construction
c. Subject-Auxiliary inversion construction
d. VP construction
e. NP construction
f. Aliza, give, Zach, what, do lexical constructions
Thus, the recipient argument is an Object whether or not it appears directly
after the verb or as a distantly instantiated question word. The same ditransitive
construction is involved in the active declarative form as well as in topicalized,
clefted, or questioned forms. The ‘valence’ or ‘subcat’ feature innovation in Pollard
and Sag, 1994 or Sag, 2012, is intended to keep track of how arguments are
expressed; on the constructionist approach, this feature can be associated with the
phrasal construction instead of lexical verb.
Clearly we don’t compute all possible combinations of constructions in advance.
Instead, constructions unify on the fly to form utterances as long as they don’t
conflict (Fillmore, 1990; Culicover and Jackendoff, 2005; Ambridge and Goldberg,
2008). Conflicts can arise in semantics (e.g. if a participant role cannot be construed
as an instance of an argument role), information structure (e.g. in the case of many
syntactic islands; cf. Ambridge and Goldberg, 2008), or in syntactic specifications
of particular constructions (e.g. a verb must be classified as an ‘auxiliary’ in order
to occur in subject-auxiliary inversion). In addition, domain general processes
of induction and preemption further constrain which combinations are judged
acceptable, as discussed in Section 6.
The phrasal approach allows us to avoid positing invisible derivational verb
templates, which, from a learning and comprehension point of view are problematic.
That is, since they are invisible, the learner and the comprehender more generally,
can only discern their existence by observing the phrasal array of grammatical
relations. From this point of view, the array of grammatical relations must be
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 455
primary (cf. also Fisher et al., 1994). In addition, the constructionist approach
generally tries to avoid positing invisible elements (cf. also Carlson, 2006), favoring
a what-you-see-is-what-you-get to syntax. Therefore, phrasal approach is, ceteris
paribus, preferable to the lexical approach, since the complement array is overt, as is
the verb.
The variable expression of deformable idioms can be captured by appealing to
an underspecified phrasal form (Jackendoff, 1997, section 7.3, 2009). For example
‘to pull strings’ can be captured by the combination of pull and noun strings that is
constrained to bear the semantic role of ‘patient.’
40. pull, stringsN-patient ‘to manipulate or control’ (pull) by using connections
(strings)’
This approach is distinct from the lexical approach primarily in the fact that it
does not assume that pull selects strings (or that strings selects pull). It is instead the
combination of the two words that gives rise to the idiomatic reading. In this way,
we do not need to stipulate that a word (pull) specifies the inflectional properties
(i.e. plurality) of its argument. And we only need to state the meaning of the
construction once.
Note that the form of the phrasal construction is underspecified so that it allows
strings to occur as a direct object, as a subject of a passive, or as a topicalized phrase.
That is, as is the case with argument structure constructions, idioms combine with
other constructions as long as the constraints on the constructions being combined
do not conflict. For example, Strings can be topicalized if and only if ‘strings’
(the connections) can be construed as topical. This predicts that only relevantly
‘compositional’ idioms are deformable (Nunberg, Wasow and Sag, 1994; Fellbaum,
2011).

5.1 Allostructions for Passives and Middles

Argument structure constructions (and derivational verb templates) specify an array
of grammatical relations that can be associated with semantics. Since the array of
grammatical relations is not the same for actives, passives and middles, separate
but related allostructions are required (Bergen and Chang, 2005; cf. also Cappelle,
2006; Pike, 1962; Lambrecht, 1994 for use of this term in slightly different
contexts). That is, for example, we need to posit a passive-ditransitive as well as an
active-ditransitive.
Referring to the passive-ditransitive and the active ditransitive as allocon-
structions is intended to evoke the relationship among allophones of a given
phoneme. For example, both the partially devoiced word final /d/ and the inter-
vocalic flap are allophones of the English voiced /d/ phoneme. These allophones
are motivated by articulatory and auditory factors and recur cross-linguistically,
but their specifics differ. For example, British English does not use the flap
intervocalically and German devoices word final stops including /d/ more than
English does.
© 2013 John Wiley & Sons Ltd
456 A. E. Goldberg
Similarly, the actual form of the passive-ditransitive is not strictly predictable. The
passive-ditransitive didn’t arise until around 1375, almost two centuries after case
marking on the recipient disappeared (Allen, 2001). Moreover, in some languages,
both the recipient and patient arguments can passivize (Alsina and Mchombo,
1990), where as in English only the recipient argument can be passivized. The fact
that there is something non-predictable about the passive-ditransitive entails that a
construction must be posited.
Müller argued against the phrasal account of argument structure constructions
by citing examples from Yucatec Maya (2006, p. 21, ex 35d). He pointed out that
it is possible to causativize the passive verb ‘learn’ in (41) and suggested that the
causativized form can be passivized again as in (42):

41. K=u ka’an -s -ik le teoria-o’

Incompl=3.erg learn.PASS-cause-impf det theory-D1
‘He is teaching the theory’ (He causes that the theory is being learned)

42. K=u ka’an-s -a’al le teoria-o

Incompl=3.erg learn.PASS-cause-PASS.impf det theory-D1
‘The theory is being taught.’ (that the theory is being learned is caused by
somebody’

This would seem to imply that we need not only a causative-passive construction,
but also a passive-causative-passive construction. However, passive-causative-
passive forms are extremely rare cross-linguistically. For example, (43c) is impossible
in English:

43. a.
. He hit the dog. (active)
b. The dog was hit. (passive)
c. ??The dog was hit the cat. (causative of passive)

Note that the Yucatec Maya causative example intended to mean ‘cause to learn’
is glossed as ‘teach,’ which is a verb meaning that is very often lexicalized. In fact,
the causative of passive pattern is not actually productive in Yucatec Maya (Müller,
2007). Thus it seems there simply exist two verbs, ‘to learn’ and ‘to teach’, both
of which can be passivized. No general passive-causative-passive construction is
warranted.

5.2 Capturing Generalizations Across Argument Structure Constructions

Do we miss generalizations by appealing to constructions instead of to linking
rules? Not at all. For example, the often-cited linking generalizations of Dowty
(1991) can be summed up as follows: in simple active clauses, if there is a subject
and an object and if there is an agent and a patient, then the agent role will be
expressed by the subject and the undergoer role as direct object (cf. also Foley and
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 457
8
Van Valin, 1984). When stated thus it is clear that this generalization concerns
the transitive construction. The fact that several constructions inherit from the
transitive construction, including the caused-motion and resultative constructions,
is naturally captured via an inheritance hierarchy (cf. Goldberg, 1995, chapter 4).
Beyond the transitive construction, many rules linking semantic roles to gram-
matical relations are actually construction-specific. Examples are provided below in
Table 6:

theme → direct object (in the transitive construction)

She kicked the ball
theme →subject (in intransitive construction)
The ball rolled.
theme → second object (in ditransitive)
She kicked him the ball.
recipient → subject (in the transitive construction)
She received it.
recipient → first object (in ditransitive)
He gave her a letter.
recipient → prepositional phrase (in ‘dative’ construction)
She gave the letter to her.

Table 6 Semantic roles that predict grammatical relations only for particular constructions.

By recognizing that linking between grammatical and semantic relations is done

at the level of constructions, which are in turn related in an inheritance hierarchy,
both generalizations and construction-specific aspects of linking are naturally
accounted for.

8
This is a modest proposal that has been taken by some to express an innate linguistic
universal. In fact, the facts are even more modest: there are syntactically ergative languages
in which agents are not generally expressed as subjects, there are many languages that do
not have canonical subjects, and there are many constructions within a given language that
violate the generalizations (e.g. passives which express the agent argument as an oblique).
But again, there is something to the generalization. The facts can be restated as follows:
semantic actors and undergoers tend to be expressed in formally prominent slots. Prominent
syntactic positions can be defined as those positions that license agreement and/or lack
overt case and/or may be obligatory. Once stated this way, the generalization is much less
mysterious: actor and undergoer arguments are generally expressed in prominent slots cross-
linguistically because human beings’ attention is naturally drawn to the actors and undergoers in
events (Goldberg, 2006).

© 2013 John Wiley & Sons Ltd

458 A. E. Goldberg
6. Lexical Specificity

Returning to the important issue of lexical specificity raised in Section 2.2, we can
now see that it is an equivalent issue for lexical rule, derivational verb template,
or constructionist approaches. For example, if lexical rules are assumed to be
productive, they will overgeneralize and incorrectly produce ill-formed examples
such as those in (5)-(7). If lexical rules are assumed to be non-productive, then
each of all possible distinct verb senses will need to be listed in advance for each
verb. Neither solution is attractive because in fact argument structure patterns are
partially, but not fully productive. That is, argument structure patterns are used
productively with certain verbs while being constrained so as not to apply to all
verbs. The derivational verb template and the phrasal constructionist approach face
exactly the same issue.
The solution requires recognizing that speakers have a usage-based model of
linguistic knowledge (Langacker, 1988). We record which verbs we’ve heard used
in which constructions previously and generalize over that knowledge (Tomasello,
2003). Thus we need verb-specific argument structure constructions: that is, phrasal
constructions that specific which verbs have been used with the construction. These
verb-specific constructions need to be phrasal even though they include a lexical
verb (cf. also Croft, 2003), in order to account for the issues discussed in Section
3. Idiomatic cases must contain other lexical material in addition to a verb. The
collection of phrasal constructions that have a verb (or other material) specified
are related in an inheritance network as daughters of the more general argument
structure constructions such as those in Table 3.
This perspective interfaces naturally with language learning. Each general argu-
ment structure construction is learned inductively from witnessing various verbs
used in the construction (e.g. Abbot-Smith, Dittmar and Tomasello, 2007; Bates
and MacWhinney, 1987; Goldberg, 1999, 2006; Tomasello, 1992, 2003). It is not
an accident that the verbs that occur most frequently in each argument structure
construction tend to encode the general meaning of that construction. For example,
give is the most frequent verb in the ditransitive. If learners record various verbs used
in particular constructions and generalize over those tokens to induce more abstract
constructions, it makes sense that the ditransitive construction comes to be associated
with ‘giving’ even when give is not used in the pattern (Goldberg et al., 2004).
Several factors that are known to be related to induction more generally deter-
mine how productive a construction is. These include type frequency, variability
of attested instances, similarity of the coinage to attested tokens and statistical pre-
emption (Ambridge et al., 2012; Barðdal, 2008; Boyd and Goldberg, 2011; Suttle
and Goldberg, 2011; Wonnacott et al., 2008). Thus productivity depends on how
the witnessed instances are distributed and how related the potential coinage is to
those instances. Also relevant is whether there exists a prepackaged alternative that
statistically preempts it (Boyd and Goldberg, 2011; Brooks and Tomasello, 1999;
Goldberg, 1995, 2006, 2011).
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 459
7. Conclusion

This paper has argued that phrasal correspondences relating grammatical relations
and semantics (argument structure constructions) are needed to account for simple
sentence types. Traditional lexical rules fall prey to problems due to their failure to
distinguish the main verb from the argument structure it occurs in. In particular,
lexical rules create implausible verb senses, miss broader generalizations due to
an emphasis on the ‘input’ of the rule and fail to account for constraints that
hold only of the verb, or the construction, but not both. A newer suggestion
of invisible derivational verb templates addresses these drawbacks, but nonetheless
treats argument structure as a verb-level phenomenon. We saw that this leads
to difficulties in accounting for idioms, which often contain specifications about
adjuncts, modification and inflection; these specifications are not appropriately
specified by individual verbs.
At the same time, it was acknowledged that verbs do play an important role
in argument structure patterns, as verbs can be quite finicky about which patterns
they occur in and there is much evidence that we record information about which
verbs are used in which constructions. The fact that argument structure patterns
are partially but not fully productive is an issue that must be addressed by any
approach. The phrasal argument structure (constructionist) approach that is argued
for here adopts a usage-based hierarchy of phrasal instances and generalizations.
New coinages are acceptable to the extent that they are similar to an existing
similarly cluster of cases and are not preempted by an alternative formulation.
The constructionist approach to argument structure was clarified in certain
respects. It was observed that word order and adjunct placement are contributed
by independent constructions that combine with argument structure constructions.
Both argument structure constructions and idioms can underspecify aspects of
their syntactic expression, allowing them to combine with other constructions that
specify linear order and grammatical categories. It has been argued that verb-specific
properties, idioms and general argument structure constructions can all be treated as
the same type of entity: phrasal patterns whose overt syntax is often underspecified
and which contain more or less lexically specified material. Correlations between
form and function at the level of argument structure are most naturally assigned to
phrasal argument structure constructions.

Council of the Humanities

Princeton University

References

Abbot-Smith, K., Dittmar, M. and Tomasello, M. 2007: Graded representations in the

acquisition of English and German transitive constructions. Cognitive Development,
23, 48-66.
© 2013 John Wiley & Sons Ltd
460 A. E. Goldberg
Ackerman, F. and Nikolaeva, I. 2004: Comparative Grammar and Grammatical Theory: A
Construction-Based Study of Morphosyntax. Stanford, CA: CSLI Publications.
Ahrens, K. 1995: The mental representation of verbs. UCSD Dissertation.
Allen, C. 2001: The development of a new passive in English. In M. Butt and T. H.
King (eds), Time Over Matter: Diachronic Perspectives on Morphosyntax. Stanford, CA:
CSLI Publications, 43-72.
Allen, K., Pereira, F., Botvinick, M. and Goldberg, A. E. 2012: Distinguishing
grammatical constructions with fMRI pattern analysis. Brain and Language, 123,
174-82.
Alsina, A. and Mchombo, S. A. 1990: The syntax of applicatives in Chichewa: problems
for a theta theoretic asymmetry. Natural Language and Linguistic Theory, 8, 493-506.
Ambridge, B., Pine, J. M., Rowland, C. F. and Chang, F. 2012: The roles of verb
semantics, entrenchment and morphophonology in the retreat from dative argument
structure overgeneralization errors. Language, 88, 45-81.
Ambridge, B. and Goldberg A. E. 2008: The island status of clausal complements:
evidence in favor of an information structure explanation. Cognitive Linguistics, 19,
357-89.
Anderson, S. R. 1971: On the role of deep structure in semantic interpretation.
Foundations of Language, 6, 197-219.
Aronoff, M. 1983: Potential words, actual words, productivity and frequency. Paper
presented to the Proceedings of the 13th International Congress of Linguists., Oslo.
Baker, M. 1997: Thematic roles and syntactic structure. In L. Haegeman (ed.), Elements
of Grammar. Dordrecht: Kluwer, 73-137.
Barðdal, J. 2008: Productivity: Evidence from Case and Argument Structure in Icelandic.
Amsterdam: John Benjamins.
Bates, E. and MacWhinney, B. 1987: Competition, variation and language learning. In
B. MacWhinney (ed.), Mechanisms of Language Acquisition. Hillsdale, NJ: Lawrence
Erlbaum Associates, 157-93.
Bencini, G. M. L. and Goldberg, A. E. 2000: The contribution of argument structure
constructions to sentence meaning. Journal of Memory and Language, 43, 640-51.
Bergen, B. and Chang, N. (2005). Embodied construction grammar in simulation-based
language understanding. In J-O. Ostman and M. Fried (eds), Construction Grammar(s):
Cognitive and Cross-Language Dimensions. Philadelphia, PA: John Benjamins.
Blevins, J. P. 2001: Realisation-based lexicalism. Journal of Linguistics, 37, 355-65.
Boas, H. 2000: Resultatives at the crossroads between the lexicon and syntax: where
are they formed? Proceedings of the 1999 Western Conference on Linguistics, 11, 38-52.
Booij, G. 2010: Construction Morphology. Oxford: Oxford University Press.
Boyd, J. K. and Goldberg, A. E. 2011: Learning what not to say: categorization and
statistical preemption in the production of a-adjectives. Language, 87, 1-29.
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 461
Bresnan, J. 1982: The Mental Representation of Grammatical Relations. Cambridge, MA:
MIT Press.
Bresnan, J. and Kanerva, J. 1989: Locative inversion in Chichewa: a case study in the
factorization of grammar. Linguistic Inquiry, 20, 1-50.
Briscoe, T. and Copestake, A. 1999. Lexical rules in constraint-based grammars.
Computational Linguistics, 25(4), 487–526.
Brooks, P. J. and Tomasello, M. 1999: How children constrain their argument structure
constructions. Language, 75(4), 720-38.
Bybee, J. 1985: Morphology: A Study of the Relation between Meaning and Form. Amsterdam:
John Benjamins.
Bybee, J. and Slobin, D. I. 1982: Rules and schemas in the development and use of the
English past tense. Language, 58(2), 265-89.
Cameron-Faulkner, T., Lieven, E. and Tomasello, M. 2003: A construction based
analysis of child directed speech. Cognitive Science, 27, 843-73.
Cappelle, B. 2006: Particle placement and the case for ‘allostructions’. Constructions, SV
1-7.
Carlson, G. 2006: When morphology . . . disappears. In C. O. Orgun and P. Sells
(eds), Morphology and the Web of Grammar: Essays in Memory of Steven G. Lapointe.
Stanford, CA: CSLI Publications.
Chomsky, N. 1965: Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Chomsky, N. 1982: Lectures on Government and Binding. Dordrecht: Forest.
Croft, W. 2003: Lexical rules vs. constructions: a false dichotomy. In H. Cuyckens, T.
Berg, R. Dirven and K-U. Panther (eds), Motivation in Language. Philadelphia, PA:
John Benjamins, 49-68.
Culicover, P. W. and Jackendoff, R. 2005: Simpler Syntax. Oxford: Oxford University
Press.
Dowty, D. 1991: Thematic proto-roles and argument selection. Language, 67, 547-619.
Faulhaber, S. 2011: Idiosyncrasy in verb valency patterns. ZAA, 4(4), 331-46.
Fellbaum, C. 2007: The ontological loneliness of idioms. In A. Schalley and D. Zaefferer
(eds), Ontolinguistics. Berlin: Mouton de Gruyter, 419-34.
Fellbaum, C. J. 2011: Idioms and collocations. In C. Maienborn, K. von Heusinger
and P. Portner (eds), Handbook of Semantics. Berlin: Mouton de Gruyter,
441-56.
Fillmore, C. J. 1977: Topics in lexical semantics. In R. Cole (ed.), Current Issues in
Linguistic Theory: Bloomington, IN: Indiana University Press, 76-138.
Fillmore, C. J. 1986: Pragmatically controlled zero anaphora. Berkeley Linguistics Society.
12, 95-107.
Fillmore, C. J. 1990: Construction grammar. Course reader for Linguistics 120A,
University of California, Berkeley.
© 2013 John Wiley & Sons Ltd
462 A. E. Goldberg
Fillmore, C. J., Kay, P. and O’Connor, M. C. 1988: Regularity and idiomaticity in
grammatical constructions: the case of let alone. Language, 64, 501-38.
Fisher, C., Hall, D. S., Rakowitz S. and Gleitman, L. 1994: When it is better to receive
than to give: syntactic and conceptual constraints on vocabulary growth. Lingua, 92,
333-75.
Foley, W. A. and Van Valin, R. 1984: Functional Syntax and University Grammar.
Cambridge: Cambridge University Press.
Garnsey, S. M., Pearlmutter, N. J., Myers, E. and Lotocky, M. 1997: The contributions
of verb bias and plausibility to the comprehension of temporarily ambiguous
sentences. Journal of Memory and Language, 37, 58-93.
Gleitman, L. 1994: The structural sources of verb meanings. In P. Bloom (ed.), Language
Acquisition: Core Readings. Cambridge, MA: MIT Press.
Goldberg, A. E. 1992: The inherent semantics of argument structure: the case of the
English ditransitive construction. Cognitive Linguistics, 3, 37-74.
Goldberg, A. E. 1995: Constructions: A Construction Grammar Approach to Argument
Structure. Chicago, IL: Chicago University Press.
Goldberg, A. E. 1999: The emergence of argument structure semantics. In B. MacWhin-
ney (ed.), The Emergence of Language. Hillsdale, NJ: Lawrence Erlbaum Publications.
Goldberg, A. E. 2002: Surface generalizations: an alternative to alternations. Cognitive
Linguistics, 13, 327-56.
Goldberg, A. E. 2006: Constructions at Work: The Nature of Generalization in Language.
Oxford: Oxford University Press.
Goldberg, A. E. 2010: Verbs, frames and constructions. In M. R. Hovav, E. Doron
and I. Sichel (eds), Syntax, Lexical Semantics and Event Structure. Oxford: Oxford
University Press.
Goldberg, A. E. 2011: Corpus evidence of the viability of statistical preemption.
Cognitive Linguistics, 22,131-53.
Goldberg, A. E., Casenhiser, D. M. and Sethuraman, N. 2004: Learning argument
structure generalizations. Cognitive Linguistics, 14(3), 289-316.
Goldberg, A. E., Casenhiser, D. and Sethuraman, N. 2005: The role of prediction in
construction-learning. Journal of Child Language, 32(2), 407-26.
Goldberg, A. E. and van der Auwera, J. 2012: This is to count as a construction. Folia
Linguistica, 46, 1-24.
Goldwater, M. B. and Markman, A. B. 2009: Constructional sources of implicit agents
in sentence comprehension. Cognitive Linguistics, 20, 675-702.
Green, G. M. 1974: Semantics and Syntactic Regularity. Bloomington, IN: Indiana
University Press.
Grimshaw, J. 1990: Argument Structure. Cambridge, MA: MIT Press.
Hayes, B. 2009: Introduction to Phonology. Oxford: Wiley-Blackwell.
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 463
Herbst, T. 2011: The status of generalizations: valency and argument structure con-
structions. ZAA, 4(4), 347-68.
Jackendoff, R. 1975: Morphological and semantic regularities in the lexicon. Language,
51, 639-71.
Jackendoff, R. 1983: Semantics and Cognition. Cambridge, MA: MIT Press.
Jackendoff, R. 1990: Semantic Structures. Cambridge, MA: MIT Press.
Jackendoff, R. 1997: The Architecture of the Language Faculty. Cambridge, MA: MIT
Press.
Jackendoff, R. 2002: Foundations of Language. Oxford: Oxford University Press.
Jackendoff, R. 2009: Meaning and the Lexicon: The Parallel Architecture, 1975-2010.
Oxford: Oxford University Press, 227.
Johnson, M. A. and Goldberg, A. E. 2012: Evidence that Constructional Meaning is
accessed automatically: Jabberwocky sentences prime associated verbs. Language and
Cognitive Processes, 1-14.
Kako, E. 2006: Thematic role properties of subjects and objects. Cognition, 101, 1-42.
Kapatsinski, V. 2008: When and why we (fail to) extend morphophonological alter-
nations: towards a usage-based harmonic grammar. Ms. University of Oregon,
1-46.
Kaschak, M. P. and Glenberg, A. M. 2000: Constructing meaning: the role of affordances
and grammatical constructions in sentence comprehension. Journal of Memory and
Language, 43(3), 508-29.
Kay, P. and Fillmore, C. J. 1999: Grammatical constructions and linguistic generaliza-
tions: the what’s X doing Y? construction. Language, 75, 1-34.
Keenan, E. and Comrie, B. 1977: Noun phrase accessibility and universal grammar.
Linguistic Inquiry, 8, 63-99.
Koenig, J-P. and Davis, A. 2001: Sublexical modality and the structure of lexical
semantic representations. Linguistics and Philosophy, 24, 71-124.
Lakoff, G. 1987: Women, Fire and Dangerous Things: What Categories Reveal About the
Mind. Chicago, IL: University of Chicago Press.
Lambrecht, K. 1990: ‘What, me worry?’ Mad Magazine sentences revisited. In Proceedings
of the 16th Annual Meeting of the Berkeley Linguistics Society. Berkeley, CA: BLS,
215-228.
Lambrecht, K. 1994: Information Structure and Sentence Form. Cambridge: Cambridge
University Press.
Langacker, R. W. 1987: Foundations of Cognitive Grammar Volume I. Stanford, CA:
Stanford University Press.
Langacker, R. W. 1988: A usage-based model. In B. Rudzka-Ostyn (ed.), Topics in
Cognitive Linguistics. Amsterdam: John Benjamins.
Levin, B. and Rappaport Hovav, M. 1994: A preliminary analysis of causative verbs in
English. Lingua, 92, 35-77.
© 2013 John Wiley & Sons Ltd
464 A. E. Goldberg
Levin, B. and Rappaport Hovav, M. 2005: Argument realization. Research Surveys in
Linguistics Series. Cambridge: Cambridge University Press.
MacDonald, M. C., Pearlmutter, N. J. and Seidenberg, M. S. 1994: The lexical nature
of syntactic ambiguity resolution. Psychological Review, 101(4), 676-703.
Marantz, A. 1997: No escape from syntax: don’t try morphological analysis in the
privacy of your own lexicon. In A. Dimitriadis and L. Siegel (eds), University of
Pennsylvania Working Papers in Linguistics, Vol. 4.2, Proceedings of the 21st Annual
Penn Linguistics Colloquium, 1997, 201-25.
Meurers, W. D. 2001: On expressing lexical generalizations in HPSG Vertical general-
izations. Nordic Journal of Linguistics, 24(2), 161–217.
Michaelis, L. A. and Lambrecht, K. 1996: Toward a Construction-based model of
language function: the case of nominal extraposition. Language, 72, 215-47.
Michaelis, L. A. and Ruppenhofer, J. 2001: Beyond Alternations: A Constructional Account
of the Applicative Pattern in German. Stanford, CA: CSLI Publications.
Müller, S. 2002: Syntax or morphology: German particle verbs revisited, In N. Dehé,
R. Jackendoff, A. McIntyre and S. Urban (eds), Verb Particle Explorations, Interface
Exploration, 1. Berlin: Mouton de Gruyter, 119-39.
Müller, S. 2006: Phrasal or lexical constructions? Language, 82(4), 850-83.
Müller, S. 2007: Phrasal or lexical constructions?: some comments on underspecification
of constituent order, compositionality and control. In S. Müller (ed.), Proceedings of
the HPSG07 Conference. Stanford, CA: CSLI Publications, 373-93.
Müller, S. and Wechsler, S. (forthcoming). Lexical approaches to argument structure.
Nesset, T. 2008: Abstract Phonology in a Concrete Model. Cognitive Linguistics and the
Morphology- Phonology Interface. Berlin: Mouton de Gruyter.
Nunberg, G., Wasow, T. and Sag, I. 1994: Idioms. Language, 70(3), 491-538.
Oehrle, R. 1975: The grammatical status of the English dative alternation. Doctoral
dissertation, MIT.
Osborne, T. and Gross, T. 2012: Constructions are catenae: construction grammar
meets dependency grammar. Cognitive Linguistics, 23, 165-216.
Pierrehumbert, J. 2001: Exemplar dynamics: word frequency, lenition and contrast.
Typological Studies in Language, 45, 137-58.
Pike, K. 1962: Dimensions of grammatical constructions. Language, 38(3), 221-45.
Pollard, C. J. and Sag, I. A. 1994: Head-driven phrase structure grammar. Stanford, CA:
CSLI Publications.
Pinker, S. 1989: Learnability and Cognition: The Acquisition of Argument Structure. Cam-
bridge, MA: MIT Press/Bradford Books.
Rappaport Hovav, M. and Levin, B. 1998: Building verb meanings. In M. Butt and W.
Geuder (eds), The Projection of Arguments: Lexical and Compositional Factors. Stanford,
CA: CSLI Publications, 97-134.
© 2013 John Wiley & Sons Ltd
 Argument Structure Constructions versus Lexical Rules 465
Rappaport Hovav, M. and Levin, B. 2008. The English dative alternation: the case for
verb sensitivity. Journal of Linguistics, 44, 129-67.
Sag, I. A. 2007: Remarks on locality. In S. Müller (ed.), HPSG07 Conference. Stanford,
CA: CSLI Publications.
Sag, I. A. 2012: Sign-based construction grammar: an informal synopsis. In H. Boas
and I. Sag (eds) Sign-Based Construction Grammar. Stanford, CA: CSLI Publications,
69-202.
Saussure, F. de. 1916: Course in General Linguistics. New York: Philosophical Library.
Stefanowitsch, A. and Gries, S. T. 2009. Corpora and grammar. In A. Lüdeling and M.
Kytö (eds), Corpus Linguistics. Berlin: Mouton de Gruyter, 933–52.
Suttle, L. and Goldberg, A. E. 2011: The partial productivity of constructions as
induction. Linguistics, 6, 1237-70.
Tesnière, L. 1959. Éleménts de syntaxe structurale. Paris: Klincksieck.
Tomasello, M. 1992: First Verbs: A Case Study of Early Grammatical Development.
Cambridge: Cambridge University Press.
Tomasello, M. 2000: Do young children have adult syntactic competence? Cognition,
74, 209-53.
Tomasello, M. 2003: Constructing a Language: A Usage-Based Theory of Language Acquisi-
tion. Cambridge, MA: Harvard University Press.
Wonnacott, E., Newport, E. L. and Tanenhaus, M. K. 2008: Acquiring and processing
verb argument structure: distributional learning in a miniature language. Cognitive
Psychology, 56(3), 165-209.
Wasow, T. 2002: Postverbal Behavior. Stanford, CA: CSLI Publications.
Zaenen, A. 1991: Subcategorization and pragmatics. Presentation at the Center for the
Study of Language and Information, Stanford University.

© 2013 John Wiley & Sons Ltd