Review TRENDS in Cognitive Sciences Vol.11 No.

10

Anatomy of deductive reasoning

Vinod Goel
Department of Psychology, York University, Toronto, M3J 1P3, Canada

Much of cognitive research on deductive reasoning has explanatory framework to organize and contextualize
been preoccupied with advocating for or against visuos- the data within a cognitive neuroscience framework is
patial (mental model theory) or linguistic/syntactic provided.
(mental logic theory) models of logical reasoning. Neu-
roimaging studies bear on this issue by pointing to both Overview of reasoning studies
language-based and visuospatial systems being engaged Categorical syllogisms
during logical reasoning, and by raising additional issues Syllogisms involve reasoning about quantification and
not anticipated by these cognitive theories. Here, the negation. Because they come in several logical forms
literature on the neural basis of deductive reasoning from and so are available for multiple presentations without
the past decade is reviewed. Although these results might repetition, they have constituted the stimuli in a half-
seem chaotic and inconsistent, we identify several inter- dozen neuroimaging studies. Early positron emission tom-
esting patterns and articulate their implications for cog- ography PET (Oxygen 15 [O15]) studies of reasoning by
nitive theories of reasoning. Cognitive neuroscience data Goel et al. [8,9] used syllogisms (of the form ‘some officers
point away from a unitary system for logical reasoning are generals; no privates are generals; some officers are not
and towards a fractionated system dynamically reconfi- privates’) and reported activation in the left hemisphere
gured in response to specific task and environmental (LH) frontal–temporal system. Another early PET (O15)
cues. study by Osherson et al. [10] used similar stimuli (of the
form ‘none of the bakers play chess; some of the chess
Introduction players listen to opera; some of the opera listeners are not
Two major theories have dominated the cognitive bakers’) and reported activation in right occipital lobe,
literature on deductive reasoning. The major issue of con- right basal ganglia and left prefrontal cortex (PFC). More
tention between the two theories is whether deduction is recent event-related functional magnetic resonance ima-
underwritten by a system of (linguistic) rules sensitive to ging (fMRI) studies by Goel et al. [11,12] have identified a
the logical form of the argument (mental logic theory [1]), large neural network including bilateral PFC (left to right
or whether a visuospatial representation of the argument [L > R]), left temporal lobe and bilateral parietal lobule
is constructed and evaluated (mental model theory [2]). (L > R) as underlying logical reasoning. More interest-
Until recently, this hotly debated issue relied largely upon ingly, they have demonstrated that manipulation of the
behavioral data. During the past decade findings from content of arguments leads to systematic reorganization of
neuroimaging studies and lesion studies with neurological this neural system. This issue is addressed further below.
patients have generated interesting results that need to be
considered in developing a theory of logical reasoning. Simple conditional reasoning
There is, however, some disagreement within the scientific Conditional arguments involve the sentential connective
community as to the relevance and interpretation of these ‘if-then’ and constitute an important component of human
data. Some have argued that the results are not particu- reasoning. Noveck et al. [13] used maximally simple forms
larly helpful or even relevant to evaluating cognitive of if-then arguments (‘if there is a black square than there
theories [3]. Some have been perplexed by the apparent is a yellow circle; there is a black square; there is a red
lack of consistency in neural activation across different circle’) to examine the functional neuroanatomy of modus
studies of reasoning [4,5]. Still others see compelling evi- ponens (if p then q, p; therefore q) and modus tollens (if p
dence for their favorite theory of reasoning [6,7]. then q, not q; therefore not p) reasoning. They reported
This review begins by organizing the various studies by activation in the left lateralized parietal–frontal network
the type of logical arguments tested (categorical syllo- for both inference forms, with increased amounts of right
gisms, conditionals, transitive arguments) and summar- prefrontal activity during modus tollens. Similar results
izing the findings associated with the main effect of have been reported by Prado and Noveck [14]. More com-
reasoning (i.e. the neural activation associated with plex forms of conditional reasoning have also been used in
reasoning over and above the baseline task; Box 1). The the literature.
data are indeed overwhelming and perplexing at first
glance. The similarities in the results across different Complex conditional reasoning
studies are identified and reasons are proposed for The Wason card selection task [15] is perhaps the most
the differences in results across the studies. Finally, an widely used task in the cognitive literature to assess con-
Corresponding author: Goel, V. ([email protected]).
tent effects in reasoning [16–20]. In the original version of
Available online 2 October 2007. the task, four cards are placed on a table, each having a
www.sciencedirect.com 1364-6613/$ – see front matter . Crown Copyright ß 2007 Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2007.09.003
436 Review TRENDS in Cognitive Sciences Vol.11 No.10

and reported bilateral occipital, parietal and frontal lobe

Box 1. Basic research paradigm and issues for fMRI studies
activation for the main effect of reasoning. Houde et al. [22]
of logical reasoning
trained participants to find falsifying cases of a complex
The basic cognitive paradigm for studying reasoning adapts conditional rule with a negative antecedent like: ‘if there is
reasonably well to the fMRI environment, although compromise is
not a blue triangle then there is a red circle’. The training
invariably required between ideal cognitive parameters and fMRI
technical constraints. Subjects lie in the scanner and are presented involved teaching subjects to overcome the perceptual
with visual or auditory stimuli, consisting of premises and a matching bias inherent in the task. Thus, this task
conclusion. They exhibit knowledge of what logically follows from requires subjects to get beyond perceptual ‘traps’ and know
the premises by pressing one of two buttons to confirm or deny the how to falsify conditionals. They found a shift in neural
given conclusion.
activity from bilateral posterior regions in the ‘perceptual’
The technology requires 20 presentations of trials per condition
(for event-related designs). If repetition is required to generate an condition to a left frontal network after training on the
adequate number of trials per condition, memory recognition might conditional falsification task (see also [23,24]).
replace reasoning. This has been the rationale for using argument Parsons and Osherson [4] used complex conditional
forms, such as syllogisms, and avoiding tasks like the Wason arguments containing conjunctions and disjunctions like
selection task in imaging studies. The line length (for visual
presentation) or duration (for auditory presentation) of stimuli must
the following: ‘if his home is mortgaged then he is married;
be matched across conditions. Timing of trials is another issue. it is not true that he is both married and has children; his
From a scanning perspective, it is desirable to minimize trial times, home is not mortgaged.’ Furthermore, they used only
although from a cognitive perspective it needs to be sufficient to invalid (specifically indeterminate) arguments during
enable task performance. the scan, and asked subjects to make deductive judgments
Processing the stimuli and completing the task requires much
more than the cognitive process of interest. To identify neural
of validity and inductive judgments of probability. Con-
activation associated with the cognitive activity of interest, studies trary to other studies, they reported that deduction acti-
typically (but not always) use a baseline task. Low-level baseline vates the right hemisphere (RH) homologs of the LH
tasks can also be used to determine relative direction of neural language areas in the inferior frontal and middle temporal
response in active conditions. Occasionally, concerns are raised
lobes, and speculated that these regions constitute a
about the adequacy of a particular baseline task [5]. There are
undoubtedly better and worse baselines, but there is no ideal ‘deduction module’. This issue is further discussed below.
baseline for any given task; there is always a trade-off. The point
becomes moot when the activations result from direct comparisons Transitive inference
or interaction analyses, as do many of the key results reported in the Several studies have also addressed the issue of transitivity,
literature. The crucial point is that, whenever a comparison is
the logical property of transferability such that, the relation
involved, the resulting activations are relative and should be
understood as such. A has to B and B to C, transfers from A to C (A=B; B=C; A=C).
Analyzing the results requires not only a series of technical The transitivity relation is a basic cornerstone of logic and a
decisions (in terms of various image processing parameters and crucial component of our reasoning abilities. It is available
filters) but also several conceptual decisions that make key to adult humans, and on some accounts to 4-year-old chil-
assumptions about the nature of reasoning. For example, (1) what
cognitive processes should be included in the analysis of the neural
dren, apes, monkeys and even pigeons, perhaps having its
response (i.e. what constitutes reasoning)? In the studies by Goel origins in the need for socially organized species to infer
and colleagues the reading and processing of the premises, and the dominance relations [25]. There exist both human and
actual button press constituting the response, are modeled out as animal paradigms for studying transitive inferences. The
events of no interest [11,12,29,32]. The ‘reasoning’ component is animal paradigm is an implicit inference paradigm in so far
taken to occur with the presentation of the conclusion and its
integration into the premises. It is possible to argue that the
as it involves training subjects with overlapping pairs of
mapping of the premises onto a single representation is also part of stimuli with some specified relation (A:B; B:C; C:D) and
the reasoning process. (2) Should correct and incorrect responses seeing if they can determine the relationship on an
be analyzed collectively or separately? The neural responses unlearned (nonadjacent) pair (A:C). Pigeons can learn to
associated with each do differ. (3) Should valid and invalid trials
draw this ‘inference’ after the presentation of 1600 train-
be analyzed collectively or separately? Cognitive theory would say
collectively, but again, the neural activations associated with each ing trials [25]. Human subjects require the presentation of
differ. There are no absolutely right or wrong answers to these and 800 training trials [26]. The rationale for using this meth-
other similar issues. In making these decisions and compromises it odology with human subjects is presumably to bypass
is advisable to pay as much attention to the neuroscience as to the language in the assessment of inference.
cognitive theory.
Acuna et al. [27] used the animal paradigm to
train human subjects to carry out transitive inferences
letter on one side and a number on the other side. The task on an ordered list of 11 arbitrary shapes. They found
is to determine which cards need to be turned over to verify activation in bilateral PFC, pre-supplementary motor
the rule ‘if a card has a vowel on one side, then it has an area, insula, precuneus and lateral posterior parietal cor-
even number on the other side’. The basic finding is that if tex. Heckers et al. [28] used a similar paradigm and
the task involves arbitrary material, as in the above reported activations in a bilateral frontal–parietal–
example, performance is relatively poor. However, the temporal (including hippocampus) system.
presence of meaningful content (e.g. ‘if someone is drinking Within a human (explicit inference) paradigm, Goel and
beer, then they are at least 18 years old’) dramatically Dolan [29] carried out studies involving propositions such
improves performance. Canessa et al. [21] administered a as ‘George is taller than Mike; Mike is taller than Lynn;
version of this task using arbitrary descriptive statements George is taller than Lynn’ and reported activation in
(‘if someone does. . ., then he does. . .’) and socially mean- bilateral parietal–frontal systems (LH > RH). Knauff
ingful statements (‘if you give me. . ., then I give you. . .’), et al. [30] used similar stimuli (e.g. ‘the dog is cleaner than
www.sciencedirect.com
 Review TRENDS in Cognitive Sciences Vol.11 No.10 437

the cat; the ape is dirtier than the cat; the dog is cleaner indicates that logical reasoning involves large networks
than the ape’) and reported similar activation in a bilateral encompassing much of the brain. (The overall network
parietal–frontal system (LH > RH) plus bilateral temporal activated in one of these studies [11] is presented in
systems. Fangmeier et al. [31] used visually presented Figure 1a.) It is also apparent that the activations from
nonlinguistic stimuli and reported activation in bilateral no two studies are identical. This has caused some concern
frontal–parietal systems for the reasoning phase of the and confusion in the literature. For example, if one is
task. Goel et al. [32] used transitive arguments involving searching for a ‘logic module’ these results are bound to
familiar geographic relations (e.g. ‘Paris is north of Cairo’) seem chaotic and unsatisfactory, and lead to worries about
and found activation in the bilateral hippocampus, in poor design and experimental controls [4,5]. Alternatively,
addition to the frontal–parietal system. if one is searching for evidence for a favorite theory (for
example, mental models theory [7]) with a selective eye,
Interpreting results there is much to choose from. It is suggested here that
Table 1 summarizes the neural activations of the main both of these approaches are overly constrained by,
effect of reasoning reported by the above studies. It and self-serving to, cognitive criteria at the expense of

Table 1. Summary of particulars of 19 neuroimaging studies of deductive reasoning and reported regions of activation
corresponding most closely to the main effect of reasoning
Studies Scanning Stimuli Occipital lobes Parietal lobes Temporal Basal Cingulate Frontal lobes
(organized by method modality lobes ganglia
tasks)
RH LH RH LH RH LH RH LH RH LH RH LH
Transitivity (explicit)
Goel et al. PET Visual, 19 37 Yes 24, 32 45, 46
(1998) linguistic
Goel & Dolan fMRI Visual, 17, 18, 19 7, 40 7, 40 Yes Yes 6 6, 9
(2001) linguistic 19
Knauff et al. fMRI Auditory, 7 7 21 21, 38 6 46, 47
(2003) linguistic
Goel et al. fMRI Visual, 18, 19 18, 19 7, 40 7 21, 22, 21, 22, 11, 47 6, 9, 46,
(2004) linguistic Hi Hi 11
Fangmeier et fMRI Visual, 7 40 32 6 6, 9
al. (2006) nonlinguistic
Transitivity (implicit)
Acuna et al. fMRI Visual, 7, 39, 40 39, 40 Yes 6, 8, 9, 46 6, 8, 9, 46
(2002) nonlinguistic
Heckers et al. fMRI Visual, 40 40 37, Hi 37, 21 Yes 24 PSMA, 6 6, 47
(2004) nonlinguistic
Categorical syllogisms
Goel et al. PET Visual, 21, 22 24, 32 45, 46, 47
(1998) linguistic
Osherson et al. PET Visual, 18 Yes 6
(1998) linguistic
Goel et al. fMRI Visual, 18, 19 18 7 21/22 Yes Yes 45 44, 45
(2000) linguistic
Goel & Dolan fMRI Visual, 17, 18 17, 18 21, 22, Yes 6 6, 44
(2003) linguistic 38
Goel & Dolan fMRI Visual, 18 18, 19 7 37 39 Yes Yes 6 6, 44, 45
(2004) linguistic
Conditionals (Simple)
Noveck et al. fMRI Visual, 19 7 37 32 6, 47
(2004) linguistic
Prado & fMRI Visual, 18 17 39, 40 40 6, 45, 46 9, 46
Noveck (2006) linguistic
Conditionals (complex)
Houde et al. PET Visual,
(2000) * nonlinguistic
Parsons et al. PET Visual, 18 18 21, 37, Yes Yes 24 31 10, 44, 9
(2002) linguistic 39
Canessa et al. fMRI Visual, 19 19 7, 39, 40 7, 39, Yes 32 32 6, 8, 9, 6, 8, 9, 46
(2005) linguistic 40 10, 46
Mixed stimuli
Goel et al. PET Visual,
(1997) linguistic
Knauff et al. fMRI Auditory, 19 19 7, 40 7, 14 21, 22 21, 22 32 32 6, 9 6, 9
(2001) linguistic
Numbers denote Brodmann Areas; RH = right hemisphere; LH = left hemisphere; Hi = hippocampus; PSMA = pre-sensory-motor area.
Blank cells indicate absence of activation in region. ‘‘Stimuli modality’’ refers to the form and manner of presentation of the stimuli.
Cerebellum activations are not noted in the table.
*
Brodmann Areas not provided by authors.

www.sciencedirect.com
438 Review TRENDS in Cognitive Sciences Vol.11 No.10

(e.g. comparing maximally simple if-then inferences [13]

with more complex multistep syllogisms [10]) or exper-
imental designs asking different research questions.
There are also bound to be differences in areas of
activation, as a function of interactions between exper-
imental design and imaging techniques, involving issues of
unfolding cognitive processes and measurement and
sampling techniques. For example, block versus single-
event designs, point of data sampling within an event,
shape and temporal duration of hemodynamic response
function (HRF) functions. The full understanding of such
differences will take some time to work out and are beyond
the scope of this review.
Goel and colleagues have undertaken a series of studies
in which they have held scanners, scanning parameters,
pre-processing and statistical analyses, and experimental
design (including logical form of stimuli) constant. They
have systematically varied content of arguments, presence
of conflict, and completeness of information and searched
for causal joints in the reasoning machinery by identifying
double dissociations rather than focusing on location (Box
2) [11,12,29,34]. An examination of these data, in the

Box 2. Role of neuroimaging data in informing cognitive

theories of reasoning
Figure 1. (a) The main effect of reasoning activates a large bilateral network Although few cognitive psychologists today question the value of
including occipital, parietal, temporal and frontal lobes, basal ganglia, and neuroimaging data, there is still a lack of consensus as to its role in
cerebellar regions (reproduced from Goel [62]). Further investigation reveals that informing cognitive theory. It has at least two immediate roles,
this is a composite image. (b) Reasoning about familiar material (all apples are red
localization of functions and dissociation of functions. Arguably the
fruit; all red fruit are nutritious; all apples are nutritious) activates a left frontal (BA
latter is much more important than the former.
47) temporal (BA 21/22) system. (c) Reasoning about unfamiliar material (all A are
B; all B are C; all A are C) activates bilateral parietal lobes (BA 7, 40) and dorsal PFC Localization of brain functions: it is now generally accepted that there
(BA 6). (Reproduced from Goel et al. [11].) is a degree of modularity in aspects of brain organization. Over the
years, neuropsychologists and neuroscientists have accumulated
some knowledge of this organization. For example, we know some
neuropsychological considerations. To make sense of these brain regions are involved in processing language whereas other
data we need to seriously consider the possibility that the regions process visual spatial information. Finding selective involve-
ment of these regions in complex cognitive tasks – like reasoning –
brain is organized in a different way than envisioned by can help us differentiate between competing cognitive theories that
cognitive psychology. We need to loosen (but not abandon) make different claims about linguistic and visuospatial processes in
the top-down cognitive constraints and consider the neu- the complex task (as do mental logic and mental model theories of
roimaging data in the context of 100 years of research in reasoning). However, we also know that for much of the brain there is
neuropsychology. at least a one-to-many mapping from brain structures to cognitive
processes (and probably a many-to-many mapping), which undercuts
If we look at these imaging data, allowing some distance much of the utility of localization. Despite this caveat, localization
between them and our favorite cognitive theory of reason- seems to loom large in the literature (e.g. it underlies the concerns
ing, we notice several things. First, the results of the about a unitary reasoning system and discontent about variability in
studies are not arbitrarily different. There are family patterns of neural activation across studies).
Dissociation of brain functions: brain lesions result in selective
resemblances across them. They all implicate some com-
impairment of behavior. Such selective impairments are called
bination of occipital, parietal, temporal and frontal lobes, dissociations. A single dissociation occurs when we find a case of a
basal ganglia, and cerebellar regions in logical reasoning, lesion in region x resulting in a deficit of function a but not function
and several implicate all of these regions (Table 1). b. If we find another case, in which a lesion in region y results in a
Furthermore, the results are largely consistent with deficit in function b but not in function a, then we have a double
dissociation. Recurrent patterns of double dissociation provide
patient data, as discussed below.
indication of causal joints in the cognitive system invisible in
Where there are differences in the involvement of uninterrupted normal behavioral measures [33]. Double dissocia-
specific regions, these can be explained in terms of details tions manifest themselves as crossover interactions in neuroima-
of task, experimental design, stimuli property, and ging studies. Cognitive theories will have to respect these joints.
measurement and sampling techniques (Box 1). For Thus, if in the case of reasoning, we find double dissociations along
the lines of familiarity/unfamiliarity and certainty/uncertainty, cog-
example, with respect to task, one should not be surprised nitive theories of reasoning need to be reformulated to take these
to find some differences in transitive inference studies dissociations into consideration. Indeed, some neuropsychologists
using an animal paradigm involving hundreds of learning have argued that it really does not matter where the lesions are in
trials [27,28] and those using an explicit inference para- patients (or where the activations are in neuroimaging studies), but
digm [7,29,32]. We should also not be surprised to find only that there are dissociations. Although an extreme position, it is
not without merit.
differences in studies involving different argument forms
www.sciencedirect.com
 Review TRENDS in Cognitive Sciences Vol.11 No.10 439

absence of a commitment to a specific cognitive theory of inhibit the prepotent response associated with the
reasoning suggests that human reasoning is underwritten belief-bias. These belief-biased responses activate ventro-
by a fractionated system that is dynamically configured in medial PFC (VMPFC) (BA 11, 32), highlighting its role in
response to specific task and environmental cues. Three of non-logical, belief-based responses. The correct response
these lines of fractionation are reviewed below. indicates that subjects detected the conflict between their
beliefs and the logical inference, inhibited the prepotent
Systems for dealing with familiar and unfamiliar response associated with the belief-bias, and engaged the
material formal reasoning mechanism. The detection of this conflict
As noted above, cognitive theories of reasoning disagree requires engagement of right lateral/dorsal lateral PFC
whether logical reasoning is underwritten by the linguistic (BA 45, 46) [11,12,14] (Figure 2). This conflict detection role
system or a visuospatial system. According to the results of of right lateral/dorsal PFC is a generalized phenomenon
imaging studies, both systems are involved and their that has been documented in a wide range of paradigms in
engagement can be systematically manipulated by manip- the cognitive neuroscience literature [43–45]. The robust
ulating the content and context of the stimuli [11,12,32]. activation of right PFC in Acuna et al. [27] might have
More specifically, a left lateralized frontal–temporal con- resulted from the conflict introduced by the switching of
ceptual/language system (Figure 1b) processes familiar, spatial order in stimulus pairs (corresponding to the inver-
conceptually coherent material whereas a bilateral parie- sion of relations in linguistic arguments, e.g. Mary is taller
tal visuospatial system (Figure 1c) processes unfamiliar, than John versus John is shorter than Mary).
nonconceptual or incoherent material. It has been pro-
posed [34] that the frontal–temporal pathway (Broadman Systems for dealing with certain and uncertain
area [BA] 47, 21/22) corresponds to a heuristic system information
whereas the parietal pathway (BA 7, 40) corresponds to Cognitive theories of reasoning do not typically postulate
a formal/universal system, along the lines developed by different mechanisms for reasoning with complete and
Newell and Simon [35] or dual mechanism theory [6,36,37]. incomplete information. For example, the major cognitive
A proper treatment of the relationship between these theories of reasoning assume that the same cognitive system
various theoretical accounts and these data are beyond would deal with the following inferences [46,47]:
the scope of this review. But, the general idea is that Mary is smarter than John; John is smarter than
reasoning about familiar situations automatically utilizes George; Mary is smarter than George (determinate,
situation-specific heuristics, which are based on back- valid trial)
ground knowledge and experience. Where no such heur- Mary is smarter than John; John is smarter than
istics are available (as in reasoning about unfamiliar George; George is smarter than Mary (determinate,
situations), universal/formal methods must be used to invalid trial)
solve the problem. This latter system might rely on decon- Mary is smarter than John; Mary is smarter than
textualized or even visuospatial representations given the George; John is smarter than George (indeterminate,
involvement of the parietal system. invalid trial)
There is even some evidence to suggest that the However, patient and neuroimaging data suggest that
response of the frontal–temporal system to familiar situ- different neural systems underwrite these inferences. Goel
ations is content specific to some degree (in keeping with et al. [48] tested neurological patients with focal unilateral
some content specificity in the organization of temporal frontal lobe lesions (Figure 3) on a transitive inference
lobes) [38]. For example, although middle temporal lobe task while systematically manipulating completeness of
regions are activated when reasoning about categorical information regarding the status of the conclusion (i.e.
statements such as ‘all dogs or pets’, in the case of making determinate and indeterminate trials). The results demon-
transitive inferences about familiar spatial environments, strated a double dissociation such that patients with left
reasoning is mediated by posterior hippocampus and para-
hippocampal gyrus, the same structures that underwrite
spatial memory and navigation tasks [32]. Perhaps the
most robust example of content specificity in the organiz-
ation of the heuristic system is the ‘theory of mind’ reason-
ing system identified by a number of studies [39,40].

Systems for dealing with conflict and belief-bias

One of the oldest findings in the reasoning literature is that
subjects perform better on reasoning tasks when the logical
conclusion is consistent with their beliefs about the world
than when it is inconsistent with their beliefs [41,42]. In
the former case, subjects’ beliefs facilitate the task whereas
in the latter case they inhibit it. Within inhibitory belief
trials the prepotent response is the incorrect response Figure 2. The right lateral/dorsal lateral PFC (BA 45, 46) is activated during conflict
detection. For example, in the following argument ‘all apples are red fruit; all red
associated with belief-bias. Incorrect responses in such
fruit are poisonous; all apples are poisonous’ the correct logical answer is ‘valid’/
trials indicate that subjects failed to detect the conflict ‘true’ but the conclusion is inconsistent with our world knowledge, resulting in a
between their beliefs and the logical inference and/or belief–logic conflict. (Reproduced from Goel and Dolan [12].)

www.sciencedirect.com
440 Review TRENDS in Cognitive Sciences Vol.11 No.10

unilateral focal lesions to PFC using the Wason card

selection task [59]. The results indicate that patients with
lesions to left PFC fail to benefit from the presentation of
familiar content. Patients with lesions to right PFC per-
form as well as normal controls. The only exception to this
finding is an electroconvulsive (ECT) hemispheric suppres-
sion study by Deglin and Kinsbourne [61], in which they
reported that the LH is engaged by analytic/formal reason-
ing, whereas the RH is engaged by reasoning about mean-
ingful situations. However, some technical limitations of
the study (e.g. they used only two electrodes and did not
provide details of the extent of hemispheric suppression)
leave their results open to other interpretations (e.g. rather
than a left/right-hemispheric account, a left dorsal lateral
and ventral medial PFC account would be consistent with
their placement of electrodes and the neuroimaging data).
The imaging results regarding right PFC activation for
conflict detection are also supported by lesion data [45], as
is the role of right PFC in dealing with indeterminate
information [48].

Summary and conclusion

Considerable progress has been made over the past decade
in our understanding of the neural basis of logical reason-
ing. The imaging results are overlapping with each other,
and perhaps most importantly, consistent with patient
data. Where there are discrepancies in the imaging data,
they can usually be explained in terms of task components,
experimental design and measurement technique factors.
In broad terms, these data are telling us that the brain is
organized in ways not anticipated by cognitive theory. We
Figure 3. Lesion overlay maps showing location of lesions in patients tested on should be receptive to this possibility. In particular, we
reasoning with determinate and indeterminate argument forms. Patients with
lesions to right PFC were selectively impaired in reasoning about indeterminate need to confront the possibility that there might be no
forms (Mary is taller than Mike; Mary is taller than George; Mike is taller than unitary reasoning system in the brain. Rather, the evi-
George). Patients with lesions to left PFC showed an overall impairment of
dence points to a fractionated system that is dynamically
reasoning. (Reproduced from Goel et al. [48].)
configured in response to certain task and environmental
cues. The three lines of demarcation reviewed above in-
PFC lesions were selectively impaired in trials with clude (i) systems for heuristic and formal processes (with
complete information (i.e. determinate trials), whereas evidence for some degree of content specificity in the
patients with right PFC lesions were selectively impaired heuristic system); (ii) conflict detection/resolution systems;
in trials with incomplete information (i.e. indeterminate and (iii) systems for dealing with certain and uncertain
trials). Parsons and Osherson [4] inadvertently demon- inferences. There are undoubtedly others. This is not the
strated the same thing in their imaging study described only possible interpretation of the imaging data, but it is
above. All of their deductive trials were indeterminate and one reasonable interpretation consistent with the neurop-
resulted in right PFC activation (rather than left or bilat- sychological literature and the presentation of neurological
eral PFC, as reported by other studies). patients with focal brain lesions. Sensitivity to these data

Consistency with patient data

The most consistent finding in the studies reported in Box 3. Questions for future research
Table 1 is the presence of activation in left PFC. All studies
 Will the results regarding dual mechanisms generalize to other
(except Parsons and Osherson [4]) report activation in left or
reasoning paradigms, in particular, judgment and decision-
bilateral PFC for logical reasoning. There is a long tradition making tasks?
in neuropsychology, backed by robust lesion data, sugg-  All current studies involve the recognition of valid conclusions.
esting LH dominance for reasoning processes [49–53]. With Does the generation of conclusions involve similar or different
respect to logical reasoning specifically, nine patient studies neural systems?
 Are there sex differences in logical reasoning, in particular, spatial
report impairment in reasoning accuracy as a function of reasoning?
focal lesions in left temporal lobes [45,54–56], left frontal  Real-world reasoning occurs against a backdrop of beliefs and
lobes [48,57–59] and generalized LH lesions [60]. emotions. Although several studies have addressed the issue of
The involvement of the left frontal–temporal system in how beliefs interact with the reasoning process, the question
reasoning about familiar content, reported by imaging remains, ‘How does the interaction with emotions affect the
neural basis of reasoning?’
studies [11,12,32], was tested in neurological patients with
www.sciencedirect.com
 Review TRENDS in Cognitive Sciences Vol.11 No.10 441

will move us beyond the sterility of mental models versus 30 Knauff, M. et al. (2003) Reasoning, models, and images: behavioral
measures and cortical activity. J. Cogn. Neurosci. 15, 559–573
mental logic debate and further the development of cogni-
31 Fangmeier, T. et al. (2006) FMRI evidence for a three-stage model of
tive theories of reasoning (See Box 3). deductive reasoning. J. Cogn. Neurosci. 18, 320–334
32 Goel, V. et al. (2004) The hippocampal system mediates logical
References reasoning about familiar spatial environments. J. Cogn. Neurosci.
1 Rips, L.J. (1994) The Psychology of Proof: Deductive Reasoning 16, 654–664
in Human Thinking, MIT Press 33 Shallice, T. (1988) From Neuropsychology to Mental Structure,
2 Johnson-Laird, P.N. (1994) Mental models, deductive reasoning, Cambridge University Press
and the brain. In The Cognitive Neurosciences (Gazzaniga, M.S., 34 Goel, V. (2005) Cognitive neuroscience of deductive reasoning. In
ed.), pp. 999–1008, MIT Press Cambridge Handbook of Thinking and Reasoning (Holyoak, K. and
3 Fodor, J. (1999) Let your brain alone. London Review of Books 21, http:// Morrison, R., eds), pp. 475–492, Cambridge University Press
www.lrb.co.uk/v21/n19/print/fodo01_.html, 30 September (1999) 35 Newell, A. and Simon, H.A. (1972) Human Problem Solving, Prentice-
4 Parsons, L.M. and Osherson, D. (2001) New evidence for distinct right Hall
and left brain systems for deductive versus probabilistic reasoning. 36 Stanovich, K.E. and West, R.F. (2000) Individual differences in
Cereb. Cortex 11, 954–965 reasoning: implications for the rationality debate. Behav. Brain Sci.
5 Monti, M.M. et al. (2007) Functional neuroanatomy of deductive 22, 645–665
inference: a language-independent distributed network. Neuroimage 37 Sloman, S.A. (1996) The empirical case for two systems of reasoning.
37, 1005–1016 Psychol. Bull. 119, 3–22
6 Evans, J.S. (2003) In two minds: dual-process accounts of reasoning. 38 McCarthy, R.A. and Warrington, E.K. (1990) Cognitive
Trends Cogn. Sci. 7, 454–459 Neuropsychology: A Clinical Introduction, Academic Press
7 Knauff, M. et al. (2002) Spatial imagery in deductive reasoning: a 39 Fletcher, P.C. et al. (1995) Other minds in the brain: a functional
functional MRI study. Brain Res. Cogn. Brain Res. 13, 203–212 imaging study of ‘‘theory of mind’’ in story comprehension. Cognition
8 Goel, V. et al. (1997) The seats of reason: a localization study of 57, 109–128
deductive & inductive reasoning using PET (O15) blood flow 40 Goel, V. et al. (1995) Modelling other minds. Neuroreport 6, 1741–1746
technique. Neuroreport 8, 1305–1310 41 Wilkins, M.C. (1928) The effect of changed material on the ability to do
9 Goel, V. et al. (1998) Neuroanatomical correlates of human reasoning. formal syllogistic reasoning. Arch. Psychol. 16, 5–83
J. Cogn. Neurosci. 10, 293–302 42 Evans, J.S.B.T. et al. (1983) On the conflict between logic and belief in
10 Osherson, D. et al. (1998) Distinct brain loci in deductive versus syllogistic reasoning. Mem. Cognit. 11, 295–306
probabilistic reasoning. Neuropsychologia 36, 369–376 43 Fink, G.R. et al. (1999) The neural consequences of conflict between
11 Goel, V. et al. (2000) Dissociation of mechanisms underlying syllogistic intention and the senses. Brain 122, 497–512
reasoning. Neuroimage 12, 504–514 44 Stavy, R. et al. (2006) Intuitive interference in quantitative reasoning.
12 Goel, V. and Dolan, R.J. (2003) Explaining modulation of reasoning by Brain Res. 1073, 1074, 383–388
belief. Cognition 87, B11–B22 45 Caramazza, A. et al. (1976) Right-hemispheric damage and verbal
13 Noveck, I.A. et al. (2004) The neural basis of conditional reasoning with problem solving behavior. Brain Lang. 3, 41–46
arbitrary content. Cortex 40, 613–622 46 Johnson-Laird, P.N. (1993) Human and Machine Thinking, Erlbaum
14 Prado, J. and Noveck, I.A. (2007) Overcoming perceptual features in Associates
logical reasoning: a parametric functional magnetic resonance imaging 47 Osherson, D. (1975) Logic and models of logical thinking. In Reasoning:
study. J. Cogn. Neurosci. 19, 642–657 Representation and Process in Children and Adults (Falmagne, R.J.,
15 Wason, P.C. (1966) Reasoning. In New Horizons in Psychology (Foss, ed.), pp. 81–92, Lawrence Erlbaum Associates
B., ed.), pp. 135–151, Penguin 48 Goel, V. et al. (2007) Hemispheric specialization in human prefrontal
16 Cosmides, L. (1989) The logic of social exchange: has natural selection cortex for resolving certain and uncertain inferences. Cereb Cortex 17,
shaped how humans reason? Studies with the Wason selection task. 2245–2250
Cognition 31, 187–276 49 Wolford, G. et al. (2000) The left hemisphere’s role in hypothesis
17 Fodor, J. (2000) Why we are so good at catching cheaters. Cognition 75, formation. J. Neurosci. 20 (6), 1–4
29–32 50 Corballis, P.M. (2003) Visuospatial processing and the right-
18 Gigerenzer, G. and Hug, K. (1992) Domain-specific reasoning: social hemisphere interpreter. Brain Cogn. 53, 171–176
contracts, cheating, and perspective change. Cognition 43, 127–171 51 Gazzaniga, M.S. (1985) The Social Brain, BasicBooks
19 Griggs, R.A. and Cox, J.R. (1982) The elusive thematic-materials effect 52 Gazzaniga, M.S. (1998) The Mind’s Past, University of California Press
in Wason’s selection task. Br. J. Psychol. 73, 407–420 53 Gazzaniga, M.S. (2000) Cerebral specialization and interhemispheric
20 Stenning, K. and Lambalgen, M. (2001) Semantics and psychology: communication: does the corpus callosum enable the human condition?
Wason’s selection task as a case study. J. Log. Lang. Inf. 10, 273–317 Brain 123, 1293–1326
21 Canessa, N. et al. (2005) The effect of social content on deductive 54 Read, D.E. (1981) Solving deductive-reasoning problems after
reasoning: an fMRI study. Hum. Brain Mapp. 26, 30–43 unilateral temporal lobectomy. Brain Lang. 12, 116–127
22 Houde, O. et al. (2000) Shifting from the perceptual brain to the logical 55 Langdon, D. and Warrington, E.K. (2000) The role of the left
brain: the neural impact of cognitive inhibition training. J. Cogn. hemisphere in verbal and spatial reasoning tasks. Cortex 36, 691–702
Neurosci. 12, 721–728 56 Whitaker, H. et al. (1991) Inference deficits after brain damage. J. Clin.
23 Houde, O. and Tzourio-Mazoyer, N. (2003) Neural foundations of Exp. Neuropsychol. 13, 38
logical and mathematical cognition. Nat. Rev. Neurosci. 4, 507–514 57 Varley, R. and Siegal, M. (2000) Evidence for cognition without
24 Houde, O. et al. (2001) Access to deductive logic depends on a right grammar from causal reasoning and ‘theory of mind’ in an
ventromedial prefrontal area devoted to emotion and feeling: evidence agrammatic aphasic patient. Curr. Biol. 10, 723–726
from a training paradigm. Neuroimage 14, 1486–1492 58 Adolphs, R. et al. (1996) Neuropsychological approaches to reasoning
25 Delius, J.D. and Siemann, M. (1998) Transitive responding in and decision-making. In Neurobiology of Decision-Making (Damasio,
animals and humans: exaption rather than adaption? Behav. A.R., ed.), pp. 157–178, Springer-Verlag
Processes 42, 107–137 59 Goel, V. et al. (2004) Asymmetrical involvement of frontal lobes in
26 Acuna, B.D. et al. (2002) Cognitive mechanisms of transitive inference. social reasoning. Brain 127, 783–790
Exp. Brain Res. 146, 1–10 60 Golding, E. (1981) The effect of unilateral brain lesion on reasoning.
27 Acuna, B.D. et al. (2002) Frontal and parietal lobe activation during Cortex 17, 31–40
transitive inference in humans. Cereb. Cortex 12, 1312–1321 61 Deglin, V.L. and Kinsbourne, M. (1996) Divergent thinking styles of the
28 Heckers, S. et al. (2004) Hippocampal activation during transitive hemispheres: how syllogisms are solved during transitory hemisphere
inference in humans. Hippocampus 14, 153–162 suppression. Brain Cogn. 31, 285–307
29 Goel, V. and Dolan, R.J. (2001) Functional neuroanatomy of three-term 62 Goel, V. (2003) Evidence for dual neural pathways for syllogistic
relational reasoning. Neuropsychologia 39, 901–909 reasoning. Psychologica 32, 301–309

www.sciencedirect.com