REVIEWS

A NEUROCOGNITIVE PERSPECTIVE
ON LANGUAGE: THE DECLARATIVE/
PROCEDURAL MODEL
Michael T. Ullman
What are the psychological, computational and neural underpinnings of language? Are these
neurocognitive correlates dedicated to language? Do different parts of language depend on
distinct neurocognitive systems? Here I address these and other issues that are crucial for our
understanding of two fundamental language capacities: the memorization of words in the mental
lexicon, and the rule-governed combination of words by the mental grammar. According to the
declarative/procedural model, the mental lexicon depends on declarative memory and is rooted
in the temporal lobe, whereas the mental grammar involves procedural memory and is rooted in
the frontal cortex and basal ganglia. I argue that the declarative/procedural model provides a new
framework for the study of lexicon and grammar.

Language depends on two mental capacities: a memo- order of lexical items, but also their hierarchical rela-
rized ‘mental lexicon’ and a computational ‘mental tions. In this example, an abstract representation for the
grammar’1,2. The mental lexicon is a repository of verb phrase ‘glicked the plag’ contains a representation
stored information, including all idiosyncratic, word- for the noun phrase ‘the plag’. This grammatical ability
specific information. It includes those words with arbi- to derive meaning from any well-formed complex
trary sound–meaning pairings, such as the non-com- structure underlies the incredible productivity and cre-
positional word ‘cat’. It is also thought to contain other ativity of human language. Such rule-governed behav-
irregular word-specific information, such as any argu- iour is found at various levels in language; for example,
ments that must accompany a verb (‘devour’ must be in phrases and sentences (syntax), and in complex
accompanied by a direct object), and any unpredictable words such as ‘walked’ or ‘glicked’ (morphology).
forms that a word takes (‘teach’ takes ‘taught’ as its past Importantly, the rules are a form of mental knowledge,
tense). The mental lexicon might also comprise com- in that they underlie our individual capacity to produce
plex linguistic structures, such as phrases and sentences, and comprehend complex forms. Moreover, the rules
the meanings of which cannot be derived transparently underlie mental operations that manipulate words and
from their parts (for example, idiomatic phrases such as abstract representations in the composition of complex
‘kick the bucket’). structures. The learning and use of the rules and opera-
But language also consists of regularities, which can tions of grammar are generally implicit (subconscious),
be captured by the rules of grammar. The rules con- and it has been argued that such grammatical knowl-
Departments of
Neuroscience, Linguistics, strain how lexical forms can combine to make complex edge is not available to other cognitive operations — it
Psychology and Neurology, representations, and allow us to interpret the meanings is ‘informationally encapsulated’3. Last, although com-
Georgetown University, of complex forms even if we have not heard or seen plex representations (‘walked’) could be computed
Research Building, 3900 them before. For example, in the sentence ‘Clementina anew each time they are used (‘walk’ + ‘-ed’), and cer-
Reservoir Road North West,
Washington DC 20007, USA.
glicked the plag’, we know that Clementina did some- tainly must be if they have not been previously encoun-
e-mail: thing in the past to some entity. The meaning can be tered (‘glicked’), they could, in principle, also be stored
[email protected] derived from rules that underlie not only the sequential in the mental lexicon after being constructed.

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | OCTOBER 2001 | 7 1 7

REVIEWS

The declarative/procedural model cortex (including Broca’s area and the supplementary
The neurocognitive bases of the mental lexicon and the motor area), the basal ganglia, parietal cortex and the
mental grammar have been the focus of many stud- dentate nucleus of the cerebellum10–12,19–22. This system
ies2,4–7, which have concentrated on several issues might be related to the dorsal visual stream13 and is
including separability (do lexicon and grammar important for learning or processing skills that involve
depend on distinct or shared neurocognitive corre- action sequences23. The execution of these skills seems
lates?), computation (what computational mechanisms to be guided in real time by the posterior parietal cor-
underlie the learning, representation and processing of tex, which is densely connected to frontal regions13.
the two language capacities, and how is linguistic Inferior parietal regions might serve as a repository for
knowledge represented?), domain specificity (are the knowledge of skills, including information about
neurocognitive correlates of lexicon and grammar ded- stored sequences20. Similarly, the basal ganglia are also
icated only to language, or do they subserve other func- densely connected to the frontal cortex24. Basal ganglia
tions?), and the identification of their neural correlates circuits seem to be arranged in parallel and are func-
(can we localize their neural circuitry to particular tionally segregated; each of them projects through the
brain structures; what is the temporal order in which thalamus to a particular cortical region, largely in the
these structures participate during language processing, frontal cortex24.
and how do they interact?). According to the declarative/procedural model, the
Several models have attempted to address these declarative memory system underlies the mental lexi-
issues. Here I focus on one model — the declarative/pro- con, whereas the procedural system subserves aspects
cedural model — and compare its claims and predic- of mental grammar. So, declarative memory is an asso-
tions with those of competing models. The basic premise ciative memory that stores not only facts and events,
of the declarative/procedural model is that aspects of the but also lexical knowledge, including the sounds and
lexicon/grammar distinction are tied to the distinction meanings of words. Learning new words relies largely
between two well-studied brain memory systems — on medial temporal lobe structures. Eventually, the
declarative and procedural memory — that have been knowledge of words becomes independent of the
implicated in non-language functions in humans and medial temporal lobe and dependent on other neocor-
other animals8,9. tical areas, particularly those in temporal and tem-
The declarative memory system10–12 has been impli- poroparietal regions. The temporal lobe might be par-
cated in the learning, representation and use of knowl- ticularly important for storing word meanings,
edge about facts (semantic knowledge) and events whereas temporoparietal regions might be more
(episodic knowledge). This memory system seems to important in storing word sounds. Lexical memory is
be closely related to the ventral visual stream13. It not informationally encapsulated, but is accessible to
might be particularly important for learning arbitrari- multiple mental systems.
ly related items — that is, for the associative/contextual On the other hand, procedural memory subserves
binding of information. The knowledge might be the implicit learning and use of a symbol-manipulat-
explicitly (consciously) recollected, and might not be ing grammar across subdomains that include syntax,
informationally encapsulated, but accessible to multi- morphology and possibly phonology (how sounds are
ple mental systems11. Declarative memory is subserved combined). The system might be especially important
by regions of the medial temporal lobe — in particu- in grammatical-structure building — that is, in the
lar, the hippocampus — which are largely connected sequential and hierarchical combination of stored
with temporal and temporoparietal neocortical forms (‘walk’ + ‘-ed’) and abstract representations into
regions14. The medial temporal lobe is required to con- complex structures. The learning of rules should
solidate (and possibly to retrieve) new memories, depend on parts of the system that are involved in pro-
although they eventually become independent of the cedural learning. One or more circuits between the
medial temporal lobe and depend on neocortical basal ganglia and particular frontal regions might sub-
regions, particularly those in the temporal lobe15,16. serve grammatical processing and perhaps even finer
Other brain structures are also part of this system. distinctions, such as morphology versus syntax. From
Anterior prefrontal cortex might underlie the selection this point of view, the frontal cortex and basal ganglia
or retrieval17,18 of declarative memories, whereas por- are ‘domain general’, in that they subserve non-linguis-
tions of the right cerebellum might be involved in tic and linguistic processes, but contain parallel,
searching for this knowledge18. ‘domain-specific’ circuits.
The procedural memory system10–12 has been impli- It is important to note that the model does not
cated in learning new, and controlling well-established, assume that all parts of the two memory systems sub-
motor and cognitive skills. Learning and remembering serve language. At least in the procedural system, and
these procedures is largely implicit. It has been argued probably also in the declarative system, parallel circuits
that the procedural system is informationally encapsu- are posited to have analogous computational functions
lated, having relatively little access to other mental sys- in language and in other domains. Similarly, the model
tems11. (Note that I use the term ‘procedural memory’ does not assume that these two memory systems are the
to refer to only one particular brain memory system11 only systems that underlie lexicon and grammar. Other
and not to all non-declarative or implicit memory sys- neural structures and other cognitive or computational
tems.) The system is rooted in portions of the frontal components might be important for both capacities.

718 | OCTOBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

 REVIEWS

INFLECTIONAL MORPHOLOGY Comparison with other models representing lexical forms and abstract representations,
The modification of a word to fit The declarative/procedural model is similar in certain combining them to construct complex linguistic struc-
its grammatical role. For respects to other ‘dual-system’ models1,2,25. These models tures. These structures are often suggested to be com-
example,‘sang’ and ‘walked’ are
hold that lexicon and grammar are separable and sub- posed from their parts every time they are used2,28,29.
inflected in the past tense.
served by distinct cognitive systems1,2,26. The learning, Grammar has been claimed to depend on the left frontal
DERIVATIONAL MORPHOLOGY representation and processing of words and other arbi- cortex, particularly on Broca’s area and adjacent anterior
The creation of new words. For trary information in a rote or associative memory is sub- regions30,31. So, the declarative/procedural model shares
example, the nouns ‘solemnity’ served by one or more systems that might be specialized several features with other dual-system models. Their dif-
and ‘toughness’ are derived from
the adjectives solemn and tough,
for and dedicated to these functions1,3,27. It has been ferences will become clearer when I discuss the specific
respectively. claimed that the use of stored words might depend on left predictions made by each of them.
temporal and temporoparietal structures7. The learning, ‘Single-system’ theories posit that the learning and
knowledge and processing of grammar are also sub- use of the words and rules of language depend on a sin-
served by one or more systems that are dedicated to their gle computational system that has a broad anatomical
linguistic functions. The grammar manipulates symbols distribution32,33. According to this view, there is no cate-
gorical distinction between non-compositional and
compositional forms. Instead, rules are only descriptive
Box 1 | Dual-system models and regular/irregular morphology entities, and the system gradually learns the entire statis-
All dual-system models assume that regular (default) forms are computed by rules that tical structure of language, from the arbitrary mappings
manipulate symbols representing their parts, whereas the use of irregular (non-default) in non-compositional forms to the rule-like mappings
forms involves form-specific stored representations. However, the models differ in the of compositional forms. Modern connectionist theory
specific aspects of these claims. has offered a computational framework for the single-
system view. It has been argued that the learning, repre-
Regular forms
‘Piece-based’ theories, such as the declarative/procedural model, assume that affixes are
sentation and processing of grammatical rules and lexi-
stored lexical items that are combined with stems25,29,134. Some piece-based theories cal items take place over many interconnected, simple
assume that pieces are put together anew each time they are used9,25,29,134. For example, processing units. Learning occurs by adjustments to the
‘walked’ is the real-time product of a function combining two arguments — ‘walk’ and weights of connections on the basis of statistical contin-
‘-ed’. Other piece-based theories posit that forms are combined from pieces and then gencies in the environment4,34.
stored as whole words135, with or without more-specific representation of part–whole Deciding between these competing perspectives has
structure. So,‘walked’ is listed as a whole word but could also contain information been problematic, partly because tasks that probe for
specifying its constituent parts ‘walk’ and ‘-ed’. lexicon or grammar usually differ in ways other than
Other dual-system theories deny the piece-based computation of complex words. These their use of the two capacities. For example, it is difficult
posit that, unlike phrases and sentences, the parts of complex words do not exist as to match measures of grammatical processing in sen-
separate pieces, but are specified by relations that capture regularities among words. For tence comprehension with measures of lexical memory.
example,‘walked’ is related to ‘walk’ by an affixation function that takes ‘walk’ as its only For this reason, much recent research has focused on the
argument. The controversy between real-time computation and memorized distinction between regular and irregular morphology,
representations also exists among these theories. Although it is sometimes assumed that especially in English past tense9,25,34,35. This offers a com-
regular complex words are computed anew whenever they are used136, it is alternatively parison between two otherwise well-matched types of
claimed that the representations of existing forms are stored137,138. linguistic form. The application and construction of
Irregular forms irregular past-tense forms is not entirely predictable
Because the application of an irregular transformation is, by definition, arbitrary, all dual- (compare, for example, bring–brought, sing–sang and
system models claim that each word is associated with some type of stored information come–came), and must therefore depend on memo-
regarding any irregular transformations. Theories differ as to how the information is rized representations9,25,29. Regular past tenses follow a
represented, what type of information is stored, and with which irregular items it is stored. simple rule, the affixation of ‘-ed’, which is the default
Some models suggest that any memorized irregulars are stored as symbols in a rote transformation for the past tense. Regular past tenses
memory29,48. All such models admit that ‘suppletive’ forms (utterly idiosyncratic, as in could therefore be rule products. This distinction
go–went) must be stored. However, it has been argued that subregularities that are found between regular and irregular forms is found across lan-
among many irregular transformations (for example, the shared pattern in sing–sang, guages, in both INFLECTIONAL MORPHOLOGY and DERIVATIONAL
spring–sprang and ring–rang) can be captured by rules of grammar29,139. Although MORPHOLOGY. So, the irregular/regular distinction offers a
memorized representations that link the stems of irregulars to their individual rules must
relatively simple and well-studied cross-linguistic
exist, the irregular morphological forms themselves are suggested to be rule products that
approach for examining the neurocognitive correlates
are computed anew each time they are used. These rule products can be computed by
of lexicon and grammar.
morphophonological ‘stem-readjustment rules’ for forms that undergo stem changes
(dig–dug or sing–sang) and/or affixation rules. Last, several dual-system models, including
Predictions of the models
the declarative/procedural model, claim that irregulars are represented and processed in a
distributed associative memory that is at least partially productive, therefore allowing the Dual-system models predict that representations of
generation of new irregulars (for example, spling–splang). So, the system learns the irregulars are stored in lexical memory, whereas regu-
mappings of individual morphologically complex forms (sing–sang), learns patterns lars are grammatical rule products (BOX 1). Single-
common to the mappings of different forms (sing–sang, spring–sprang, ring–rang), and system models argue that all forms are learned, repre-
can then generalize these patterns to new forms (spling–splang). Moreover, some of these sented and processed in an associative memory, which
models25, including the declarative/procedural model, assume that the representations of can be modelled by a connectionist network. Whereas
forms can be structured, reflecting the morphophonological and phonological part–whole early connectionist models focused on the phonological
structures of words. mappings between stem and past tense34,36, a recent

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | OCTOBER 2001 | 7 1 9

REVIEWS

of a form by the procedural system should inhibit the

Box 2 | A single-system model of regular/irregular morphology memorization of that form in declarative memory,
Double dissociations between therefore decreasing the likelihood of memorizing reg-
regular and irregular forms Speech output units ular forms. However, any regular form can, in principle,
have posed a problem for be memorized. The likelihood of memorization
single-system language should increase with factors such as the frequency with
models, but a recent model35 which the item is encountered or individual variation
has tried to go beyond this in learning abilities of the declarative memory system.
limitation. The model Semantics Hidden units The different theoretical perspectives make specific
contains distinct predictions about the issues that I have discussed above,
representations for semantics, allowing the theories to be distinguished empirically.
and for input and output
phonology, each being Separability. The declarative/procedural model and
subserved by a separate set of Speech input units other dual-system models posit that lexicon and gram-
units. These units (ellipses),
mar are subserved by separable cognitive systems, with
and the pathways between
at least partially distinct neural correlates. So, these
them (arrows), are assumed to be neuroanatomically distinct, and can therefore be
models predict DOUBLE DISSOCIATIONS between the two lan-
lesioned independently. Although the model claims distinct representations and
pathways, it is a single-system model in that it assumes a uniformity of processing guage capacities. Single-system models do not invoke
mechanisms. All representations and pathways underlie the computation of both separate underpinnings for lexicon and grammar, and
regular and irregular morphological forms. It is suggested that the inconsistent therefore do not predict such double dissociations.
phonological patterns of irregulars result in their computation relying more on
semantics than on phonology. Regulars, by contrast, do not show this bias, and novel Computation. The declarative/procedural model
verbs actually show the opposite pattern, relying for their computation on assumes that language involves an associative memory
phonology but not on semantics. system and a symbol-manipulation system. This
Simulations of damage to the semantic representation led to worse performance in assumption is consistent with other dual-system models,
producing irregular rather than regular or novel past tenses. Simulations of damage although many such models adopt the distinct perspec-
to output phonology led to worse performance in producing novel rather than tive that lexical memory is a rote list of words (BOX 1).
regular and irregular past tenses, but no difference between regulars and irregulars. According to the declarative/procedural model, psycho-
So, the model revealed double dissociations between irregular and novel verbs but, logical markers of associative memory, such as FREQUENCY
crucially, not between irregular and regular forms, even from lesions to output EFFECTS and phonological-similarity effects
40–43
, should be
phonology. The results from this simulation do not fit the empirical data from found with memorized lexical items including irregular
patients. For example, several reports8,68–72,74 have revealed a consistent pattern of forms, but not with regular and other complex linguistic
worse performance by patients with anterior aphasia in processing regular than forms that are rule-computed in real time. By contrast,
irregular past tenses over five classes of task: production, reading, judgement, writing single-system models predict associative memory effects
and repetition.
for all linguistic forms.

single-system model attempted to integrate semantic Domain generality. According to the declarative/proce-
and phonological knowledge, which are assumed to be dural model, but not to other dual- or single-system
linked to temporal lobe and frontal lobe structures, models, lexicon and grammar are subserved by distinct
respectively35 (BOX 2). systems, each of which underlies a specific set of non-
The declarative/procedural model predicts that irreg- language functions. Only the declarative/procedural
ular forms are stored in declarative memory. This is model predicts associations in learning, representation
an associative memory of distributed representations, and processing, among irregular forms, non-composi-
over which the phonological and semantic mappings of tional lexical items, facts and events. Similarly, only this
the transformations are learned, stored and computed model predicts associations between regular forms,
DOUBLE DISSOCIATION
A double dissociation is (BOX 1). The procedural system, by contrast, is suggested aspects of syntax and other domains of grammar, and
observed when two different to subserve the composition of regular forms from their motor and cognitive skills.
tasks lead to complementary parts in real time (‘walk’ + ‘-ed’). The computation of a
patterns in behaviour or brain morphologically complex form involves the parallel acti- Localization. The declarative/procedural model makes
activation. Task X is normal in
patient A but not patient B,
vation of the two systems; the declarative system tries to specific claims about links between the two language
whereas task Y is normal in compute a form in associative memory, while the proce- capacities and sets of specific brain structures on the
patient B but not in A. Similarly, dural system attempts to compute a rule product in real basis of the roles of these structures in the two memory
in scanning healthy subjects, task time37. As the memory-based computation proceeds, a systems. Certain dual-system models predict similar
X leads to activation in one brain
continuous signal is sent to the rule-processing system, links, but they do not make the particular neuro-
area but not another, whereas
task Y shows the opposite indicating the probability of the successful retrieval of a anatomical claims of the declarative/procedural model.
pattern. form from declarative memory. This signal prevents the Single-system models do not predict the same func-
procedural system from carrying out its computation. So tion–structure associations.
FREQUENCY EFFECTS the computation of ‘dug’ blocks the computation of
Words stored in memory are
remembered better and faster if
‘digged’. If a memorized form is not retrieved, then the As I discuss next, these predictions are supported by
they have been more frequently rule can apply, resulting in over-regularization errors such evidence from several languages, obtained using a range
encountered. as ‘digged’8,37–40. In addition, the successful computation of methodological approaches in children and adults.

720 | OCTOBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

 REVIEWS

Psycholinguistic evidence heterogeneous disorder, homogeneous groups of people

Frequency effects5 are expected for representations with SLI have been identified54, including groups with a
stored in the lexicon, but not for representations that are hereditary form of the disorder that is accompanied by
constructed by mental rules in real time. Several studies syntactic deficits55,56. Processing of past tense was exam-
have found frequency effects for irregular but not for ined in two such groups45,57,58, both of which failed to pro-
regular past-tense forms40,41,43–45. A similar contrast has duce novel regular forms (for example, plam–plammed)
been found between irregular and regular plurals in and over-regularizations (dig–digged), indicating that
German46,47. These data indicate that the representations they were unable to apply the ‘-ed’ suffix productively.
of irregular but not regular past tenses are retrieved Both groups showed frequency effects for regular as well
from memory9,43. However, frequency effects for some as irregular past-tense forms45,57,58. These data indicate
regular past tenses, such as those in which stems rhyme that subjects with SLI had difficulty in learning grammat-
with the stems of irregulars (for example, glide–glided; ical rules and were therefore forced to memorize regular
compare with hide–hid and ride–rode), indicate that at as well as irregular forms45,57,58. Motor-skill performance
least some regular forms can be stored9,40. and brain abnormalities were probed in one of the
If multiple stored representations share distributed groups. Consistent with an underlying deficit of the pro-
memory traces, then strengthening one representation cedural memory system, they showed impairments in
will strengthen all of them (BOX 1). Such distributed- performing motor sequences59 and abnormalities in
frequency (phonological-similarity) effects have been frontal cortical regions, including the left supplemen-
found for real and novel irregular past tenses (for tary motor area and Broca’s area, and in the caudate
example, spring–sprang and spling–splang), but not nucleus of the basal ganglia60. These findings link the
for real or novel regular past tenses (walk–walked and rule of ‘-ed’ affixation for regular verbs to syntax and
grock–grocked)40,42,43,48. Analogous contrasts between procedural memory.
regular and irregular forms have been found for adjec-
tival past-tense inflection in Japanese49. These contrast- Williams’ syndrome. People with WILLIAMS’ SYNDROME
ing phonological-similarity effects support the declar- might have spared syntactic abilities but abnormal lex-
ative/procedural and other dual-system models in ical retrieval61,62. Children and adults with the disorder
which irregulars, but not regulars, are represented in a have more difficulty producing irregular than regular
distributed associative memory. past tenses (dig–dug versus look–looked) and plurals
These effects indicate that representations of irregu- (mouse–mice versus rat–rats). Most of their errors are
lar, but not regular, forms are generally memorized. over-regularizations (digged, mouses)62,63. These
However, real-time rule processing can also be exam- results help to dissociate irregular from regular forms,
ined directly. One widely used method is PRIMING5. and link irregulars to lexical memory and regulars to
Studies of English past tense, and German participles syntactic abilities.
and plurals, have consistently shown that a target word
PRIMING stem (‘walk’) is consistently primed as much by its regu- Neurological evidence
A word is recognized faster if it
lar inflected form (‘walked’) as by itself (‘walk’); this is Aphasia. There are at least two fundamental classes of
has been primed by an earlier
presentation of the same word. not the case for irregulars50–52. This indicates that regu- APHASIA — anterior and posterior
6,64–66
. Anterior aphasia
lars but not irregulars are decomposed into their stems, is associated with damage to left frontal regions — in
WILLIAMS’ SYNDROME as predicted by dual-system models. particular, to Broca’s area and nearby cortex — the basal
A hereditary developmental The real-time composition of regulars has also been ganglia and portions of inferior parietal cortex64,65.
disorder characterized by
cognitive impairment (usually
tested by examining the limited storage capacity of work- People with anterior aphasia typically show AGRAMMATISM,
mild mental retardation), ing memory. The number of items that it is possible to but are relatively spared in their use of non-composi-
distinctive facial features and hold actively in mind is relatively small. If regular com- tional words. Anterior aphasia is also linked to IDEOMOTOR
cardiovascular disease. 6,20
plex words (‘walked’) are composed in real time from APRAXIA . Posterior aphasia, in turn, is associated with
multiple independent pieces (‘walk’and ‘-ed’), then main- damage to left temporal and temporoparietal regions.
APHASIA
Language impairments acquired taining them in working memory would involve main- People with posterior aphasia show impairments in the
as a result of stroke or other taining each of their constituent pieces. By contrast, for production, reading and recognition of word sounds
brain injury. forms that are associated with a single memorized repre- and meanings. These patients tend to produce syntacti-
sentation (irregulars), one should need to maintain only cally well-structured sentences and do not omit mor-
AGRAMMATISM
Syntactic and morphological
one element. It should therefore be possible to retain fewer phological affixes such as ‘-ed’6. Posterior aphasia is also
impairments in production and forms that are composed from two or more independent linked with semantic impairments in non-language
comprehension, including those pieces than forms that are not. Indeed, performance in a domains, but not with motor deficits64,67.
in the use of free and bound working memory task was worse for regular than for People with anterior aphasia are worse at produc-
grammatical morphemes
irregular past tenses53. The results support the real-time ing8,68, reading out loud8,68–73, writing to dictation74,
(auxiliaries, determiners, and
affixes such as ‘-ed’). composition of regulars from their stems and affixes. repeating71 and judging68 regular versus irregular past-
tense forms. These patients also have more difficulty in
IDEOMOTOR APRAXIA Evidence from developmental disorders reading69,70 and writing74 regular compared with irregular
An impairment in the Specific language impairment. This term is often assigned plurals. Patients with posterior aphasia show the opposite
expression of motor skills.
Patients with ideomotor apraxia
to developmental language disorders that do not have any pattern: worse production8,68, reading68 and judgement68
have problems with imitation, other apparent social, psychological or neurological of irregular past tenses. Similar double dissociations
pantomime and tool use. cause54. Although specific language impairment (SLI) is a have been found with a priming task75,76. In Japanese, the

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | OCTOBER 2001 | 7 2 1

REVIEWS

tense have found that error rates in object naming and in

fact retrieval correlate with error rates in producing
irregular but not regular forms8,87. Patients with severe
deficits in object naming or fact retrieval make more
errors in producing irregular than regular past tenses.
Similarly, Italian patients with Alzheimer’s disease have
greater difficulty in producing irregular than regular
present tense and past participle forms in Italian88.
Semantic dementia is associated with severe degener-
ation of inferior and lateral regions of the temporal lobe.
The disorder results in the loss of lexical and non-lin-
guistic conceptual knowledge89, with spared motor, syn-
tactic and phonological abilities90. Patients with seman-
tic dementia yield a pattern like that of patients with
Alzheimer’s disease. They have more trouble producing
and recognizing irregular than regular past tenses, and
the degree of their impairment on irregular forms corre-
lates with their performance on an independent lexical
memory task91. The data link irregular forms to stored
words and conceptual knowledge, and to inferior and
lateral temporal lobe regions.
Parkinson’s disease is associated with the degenera-
tion of dopamine neurons, especially in the substantia
nigra of the basal ganglia. Loss of dopamine leads to the
suppression of motor activity (hypokinesia) and difficul-
ty in expressing motor sequences19,92,93. It might also
account for the impairments of patients with Parkinson’s
disease in acquiring motor and cognitive skills83,94, and in
grammatical processing95–97. By contrast, the temporal
lobe remains relatively undamaged in these patients, and
the use of words and facts remains relatively intact, if
dementia is not present80,83,93. In fact, non-demented
Figure 1 | Haemodynamic responses to syntactic and lexical/semantic violations patients with Parkinson’s disease that suffer from severe
detected by fMRI. Haemodynamic responses averaged over 14 subjects in a functional hypokinesia show a pattern opposite to that found
magnetic resonance imaging (fMRI) study. a | Syntactic violations elicited greater blood-oxygen-
among patients with Alzheimer’s disease, making more
level-dependent (BOLD) activations than semantic violations, primarily in bilateral superior frontal
gyrus, corresponding to Brodmann areas (BA) 6 and 8, including the supplementary motor area. errors when producing regular versus irregular past tens-
Additional activations were observed in the left insula and right anterior superior temporal sulcus. es. The level of right-side hypokinesia, which reflects
b | Semantic anomalies yielded a different pattern of activation, with substantially more temporal degeneration of the left basal ganglia, correlates with
and temporoparietal involvement than syntactic anomalies, in the angular gyri bilaterally (BA 39), error rates in the production of regular but not irregular
the right middle temporal gyrus (BA 21), and left hippocampus and parahippocampal gyrus. forms. Intriguingly, left-side hypokinesia is not accompa-
Additional activations were found in dorsolateral prefrontal cortex and medial foci. Reproduced
nied by the analogous correlations with error rates in the
with permission from REF. 103 © 2001 Plenum Publishing Corporation.
production of any type of past tense, underscoring the
role of left structures in the use of grammatical rules8,87.
double dissociation between people with anterior and Although Huntington’s disease is also associated with
posterior aphasia has been found in a judgement task of degeneration of the basal ganglia, it involves different
regular and irregular forms in derivational morphology77. structures than those affected in Parkinson’s disease; in
The findings link irregular forms to lexical and non-lin- particular, regions of the caudate nucleus. This degenera-
guistic semantic memory, and to temporal/temporopari- tion leads to unsuppressible movements (hyperkinesia)
etal cortex, and link regular forms to syntax, motor skills, instead of the hypokinesia that characterizes people with
and left frontal cortex and the basal ganglia. Parkinson’s disease92. In the language domain, patients
with Huntington’s disease also show the opposite pattern
Neurodegenerative disease. Alzheimer’s disease largely of abnormalities to that found in Parkinson’s disease8,87.
affects structures in the temporal lobe, leaving frontal cor- Patients with Huntington’s produce forms like ‘walked-
tex (particularly Broca’s area and motor regions) and the ed’ and ‘dugged’, but not analogous errors on irregular
basal ganglia relatively spared78. Temporal-lobe dysfunc- verbs like ‘dugug’ or ‘keptet’, indicating that these errors
tion might explain the impairments of patients with are not attributable to articulatory or motor deficits.
Alzheimer’s disease in learning new and using established Instead, the data point to unsuppressed ‘-ed’ suffixation.
lexical and conceptual knowledge79–81. These patients are The finding that the production rate of these over-suf-
relatively spared at acquiring and expressing motor and fixed forms correlates with the degree of chorea, across
cognitive skills79,82–84, and aspects of syntactic process- patients, strengthens this conclusion. The findings in
ing85,86. In the morphology domain, investigations of past Parkinson’s and Huntington’s diseases strongly implicate

722 | OCTOBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

 REVIEWS

frontal cortex and the basal ganglia in ‘-ed’ suffixation. about 400 ms post stimulus (N400s), and depend on
More generally, they support the hypothesis that these bilateral temporal lobe structures115–117. Difficulties in
structures underlie the expression of grammatical rule-governed syntactic processing can yield early
rules, as well as movement, and indicate that they have (150–500 ms) left anterior negativities (LANs)118,119,
a similar function in the two domains. which have been linked to rule-based automatic com-
putations120 and left frontal structures121. Intriguingly,
Amnesia. Bilateral damage to medial temporal lobe difficulties in processing word-specific syntactic knowl-
structures leads to an inability to learn new information edge can elicit an N400 rather than a LAN122. Syntactic
about facts, events and words12. Importantly, neither processing difficulties also tend to elicit late (600 ms)
phonological nor semantic lexical knowledge is centroparietal positivities (P600s)123. However, these
acquired98,99, supporting the hypothesis that these struc- positivities are associated with controlled processing120
tures underlie the learning of word forms, as well as and posterior brain regions, and are not suggested to
meanings. The ANTEROGRADE AMNESIA seen after damage to depend on the procedural system.
the temporal lobe is accompanied by variable degrees of Several ERP studies have examined regular and
RETROGRADE AMNESIA. However, knowledge acquired a long irregular inflectional morphology in German124,125,
time before the lesion tends to be spared12. So, although Italian126 and English127–129. All of these studies have
medial temporal lobe structures underlie the learning of found distinct ERP patterns for regular and irregular
new lexical information, knowledge of words learned morphology. Although the specific results have varied, a
during childhood should be intact in adult-onset amne- trend has emerged. Whereas inappropriate regular affix-
sia. As expected, our examination of the well-studied ation (anomalous addition124,125 or omission128,129 of the
global amnesic H.M.100 revealed that he did not differ from affix) can lead to a LAN, modification of irregular
normal age- and education-matched subjects in syntac- inflection tends to elicit a more central, N400-like nega-
ANTEROGRADE AMNESIA
tic processing tasks, or in the production of regular and tivity124,128,129. Moreover, this LAN does not seem to dif-
The inability to store new irregular forms in past-tense, plural and derivational fer in topography from the LAN that is elicited by syn-
information in long-term morphology. tactic anomalies128,129, underscoring common neural
memory. mechanisms for regular morphology and syntax.
Neuroimaging evidence Whereas most ERP studies examine language during
RETROGRADE AMNESIA
Loss of or inability to recall Haemodynamics. Several studies using positron emis- comprehension, a recent experiment probed regular
information that was previously sion tomography (PET) and functional magnetic reso- and irregular past-tense production, and examined cor-
stored in long-term memory. nance imaging (fMRI) have investigated the pattern of tical localization of the scalp-recorded potentials130.
brain activation during language processing. Lexical and Regular past tenses elicited more frontal activation than
H.M.
Arguably the best-studied
semantic processing is strongly associated with activa- irregular verbs, but irregular forms yielded more activi-
patient in the literature on tion in temporal/temporoparietal regions, including the ty in left temporal lobe regions, strengthening the tem-
memory, H.M. became amnesic medial temporal lobe (FIG. 1)16,101–104. In addition, selec- poral lobe/frontal lobe dichotomy that is predicted by
after the bilateral resection of tion or retrieval of lexical and semantic knowledge leads the declarative/procedural model.
large parts of the temporal lobe
to activation in anterior prefrontal cortex17. By contrast,
in an attempt to treat epilepsy
episodes. The analysis of H.M.’s several tasks that are designed to probe syntactic process- Magnetoencephalography. In a MAGNETOENCEPHALOGRAPHIC
amnesia provided a clear ing preferentially elicit activation of Broca’s area, the sup- investigation of regular and irregular past-tense produc-
dissociation between declarative plementary motor area (FIG. 1) and the left basal ganglia tion, DIPOLE MODELLING was used to localize sources of
and procedural memory. (caudate nucleus)102,105–109. Interestingly, the processing of brain activity131. Dipoles were localized to a single left
MAGNETOENCEPHALOGRAPHY
lexically stored syntactic knowledge (for example, word- temporal/parietal region for both regular and irregular
A non-invasive technique that specific knowledge about what arguments a verb takes) verbs, 250–310 ms after verb-stem presentation. Dipoles
allows the detection of the is accompanied by activation of the temporal lobe104. in left frontal regions were found only for regular verbs
changing magnetic fields that are Imaging studies have examined the production of and only for times immediately after the left temporal/
associated with brain activity. As
regular and irregular forms in the English past tense110–112, parietal dipoles (310–330 ms). No dipoles were found
the magnetic fields of the brain
are very weak, extremely and in the German past tense and past participle113. These in the right hemisphere. These results are consistent
sensitive magnetic detectors studies have found differential activation in frontal and with a dual-system model in which temporal/parietal-
known as superconducting temporal regions for the two forms, although the specific based memory is searched for an irregular form, the
quantum interference devices, regions have varied across the studies9. A study of successful retrieval of which blocks the application of a
which work at very low,
superconducting temperatures
Finnish, a morphologically very rich and productive frontal-based suffixation rule8.
(–269 °C), are used to pick up language, reported greater activation in Broca’s area for
the signal. regular morphologically complex words than for non- Conclusion
compositional words114, strengthening the view that this In summary, studies using different methodologies have
DIPOLE MODELLING
region underlies rule-based morphological processing. examined the acquisition, computation, processing and
A method to determine the
location of the sources that neural bases of lexicon and grammar, focusing on irreg-
underlie the responses measured Electrophysiology. Event-related potentials (ERPs) ular and regular morphology in several languages.
in a magnetoencephalographic reflect the real-time electrophysiological activity of the These studies have tested the predictions of different
experiment. It provides an brain elicited by cognitive processes that are time-locked single- and dual-system language models. The data
estimate of the location,
orientation and strength of the
to the presentation of target stimuli. Difficulties in largely conform to the dissociations and associations
source as a function of time after semantic processing with lexical or non-linguistic stim- that are predicted by the declarative/procedural model,
the stimulus was presented. uli elicit central/posterior bilateral negativities that peak supporting its validity (BOX 3).

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | OCTOBER 2001 | 7 2 3

REVIEWS

The declarative/procedural model has several

Box 3 | Some open questions
implications. First, the numerous studies of the declar-
Although I argue that most of the available evidence supports the declarative/ ative and procedural memory systems in animals and
procedural model, the data are not uniform in this regard. Even the relatively simple humans are expected to help to elucidate the computa-
approach of studying regular and irregular morphology has failed to yield entirely tional and neural bases of learning, representation and
consistent results, particularly in other languages, in which the contrast is not as well processing of lexicon and grammar. For example, the
understood as in English25. neuropharmacology of declarative memory and its
Moreover, several issues remain to be further specified by the declarative/procedural underlying neural substrates132 should also pertain to
model, including the following. language. Second, because language is a relatively well-
• What is the exact computational role of the procedural system and its parts? What understood cognitive domain, linguistic theory and
particular functions do the different brain structures in the system underlie? Is the related language disciplines will probably shed light on
system involved in both structure building and learning rules? the workings of declarative and procedural memory.
• What specific regions in frontal and parietal cortex subserve these functions? Are Third, the model has direct clinical implications.
inferior parietal structures involved in lexical as well as grammatical functions? People with developmental or adult-onset disorders of
• The declarative/procedural model does not claim that these two brain systems are the the grammatical/procedural system should recover
only circuits underlying lexicon and grammar, let alone other aspects of language. What through the memorization of complex forms using
other circuits underlie these two language capacities, and how do they relate to the lexical/declarative memory. Indeed, this is what sub-
declarative and procedural systems? jects with SLI seem to do (see above)45,57,58. In addition,
preliminary evidence indicates that people with anteri-
• What types of regular and other complex structures are stored? What form do their
or (but not posterior) aphasia memorize regular forms
representations take?
after the onset of their lesion133. Such recovery could be
• Are there individual differences (for example, corresponding to sex differences) in the stimulated with pharmacological and other therapeu-
storage of complex forms? tic approaches derived from our current knowledge of
• How similar are the linguistic and non-linguistic computations in each of the two the two memory systems. Last, the existence of brain
systems? Do the similarities extend to the non-linguistic computations in other animals, systems that subserve language in humans, but are
such as non-human primates? Would this enable the development of an animal model homologous to systems that are present in other animals,
of aspects of language? has implications for the evolution of language.

1. Chomsky, N. The Minimalist Program (MIT Press, 12. Schacter, D. L. & Tulving, E. (eds) Memory Systems (MIT 25. Pinker, S. Words and Rules: the Ingredients of Language
Cambridge, Massachusetts, 1995). Press, Cambridge, Massachusetts, 1994). (Basic Books, New York, 1999).
Extends the influential Government and Binding 13. Goodale, M. A. in The New Cognitive Neurosciences (ed. Focusing on evidence from regular and irregular
framework in theoretical linguistics. Gazzaniga, M. S.) 365–378 (MIT Press, Cambridge, morphology, especially in English past tense, it is
2. Pinker, S. The Language Instinct (William Morrow, New York, Massachusetts, 2000). argued that the mental lexicon and the mental
1994). A proposal that ‘dorsal’ and ‘ventral’ visual streams grammar are distinct systems in the mind and brain.
A highly readable overview of language, arguing that it underlie distinct types of visual processing. 26. Bresnan, J. Lexical–Functional Syntax (Blackwell, Oxford,
is dependent on innate mental constructs. 14. Suzuki, W. A. & Amaral, D. G. Perirhinal and 2001).
3. Fodor, J. A. The Modularity of Mind: an Essay on Faculty parahippocampal cortices of the macaque monkey: cortical 27. Levelt, W. J. M. Speaking: from Intention to Articulation (MIT
Psychology (MIT Press, Cambridge, Massachusetts, 1983). afferents. J. Comp. Neurol. 350, 497–533 (1994). Press, Cambridge, Massachusetts, 1989).
A clearly reasoned claim that there are distinct mental 15. Hodges, J. R. & Patterson, K. Semantic memory disorders. 28. Frazier, L. in Attention and Performance Vol. 12 (ed.
‘modules’ dedicated to different mental capacities. Trends Cogn. Sci. 1, 68–72 (1997). Coltheart, M.) 559–586 (Lawrence Erlbaum, Hillsdale, New
4. Elman, J. et al. Rethinking Innateness: a Connectionist 16. Martin, A., Ungerleider, L. G. & Haxby, J. V. in Jersey, 1987).
Perspective on Development (MIT Press, Cambridge, The New Cognitive Neurosciences (ed. Gazzaniga, M. S.) 29. Halle, M. & Marantz, A. in The View from Building 20 (MIT
Massachusetts, 1996). 1023–1036 (MIT Press, Cambridge, Massachusetts, Press, Cambridge, Massachusetts, 1993).
An overview of modern connectionism, arguing that 2000). 30. Caramazza, A., Berndt, R. S., Basili, A. G. & Koller, J. J.
traditional views of cognitive innateness are flawed. 17. Buckner, R. L. in The New Cognitive Neurosciences (ed. Syntactic processing deficits in aphasia. Cortex 17,
5. Gleason, J. B. & Ratner, N. B. (eds) Psycholinguistics Gazzaniga, M. S.) 817–828 (MIT Press, Cambridge, 333–348 (1981).
(Harcourt Brace College Publishers, Fort Worth, 1998). Massachusetts, 2000). 31. Grodzinsky, Y. The neurology of syntax: language
6. Goodglass, H. Understanding Aphasia (Academic, San 18. Desmond, J. E., Gabrieli, J. D. E. & Glover, G. H. use without Broca’s area. Behav. Brain Sci. 23, 1–71
Diego, California, 1993). Dissociation of frontal and cerebellar activity in a cognitive (2000).
7. Damasio, A. R. & Damasio, H. Brain and language. Sci. Am. task: evidence for a distinction between selection and 32. MacDonald, M. C., Pearlmutter, N. J. & Seidenberg, M. S.
267, 88–95 (1992). search. Neuroimage 7, 368–376 (1998). Lexical nature of syntactic ambiguity resolution. Psychol.
An important perspective on the relation between 19. De Renzi, E. in Handbook of Neuropsychology Vol. 2 Rev. 101, 676–703 (1994).
brain and language. (eds Boller, F. & Grafman, J.) 245–263 (Elsevier, New York, 33. Bates, E. & MacWhinney, B. in The Crosslinguistic Study of
8. Ullman, M. T. et al. A neural dissociation within 1989). Sentence Processing (eds MacWhinney, B. & Bates, E.)
language: evidence that the mental dictionary is part of 20. Heilman, K. M., Watson, R. T. & Rothi, L. G. in 3–73 (Cambridge Univ. Press, Cambridge, UK, 1989).
declarative memory, and that grammatical rules are Behavioral Neurology and Neuropsychology (eds Feinberg, 34. Rumelhart, D.E. & McClelland, J.L. in Parallel Distributed
processed by the procedural system. J. Cogn. Neurosci. 9, T. E. & Farah, M. J.) 227–235 (McGraw-Hill, New York, Processing: Explorations in the Microstructures of Cognition
266–276 (1997). 1997). Vol. 2 (eds McClelland, J.L., Rumelhart, D.E. & PDP
Based on evidence from anterior and posterior 21. Hikosaka, O. et al. in The New Cognitive Neurosciences (ed. Research Group) 216–271 (Bradford/MIT Press,
aphasia, and from Alzheimer’s, Parkinson’s and Gazzaniga, M. S.) 553–572 (MIT Press, Cambridge, Cambridge, Massachusetts, 1986).
Huntington’s diseases, it is argued that grammatical Massachusetts, 2000). A seminal proposal that regular and irregular past
rules depend on a frontal/basal ganglia procedural 22. Rizzolatti, G., Fogassi, L. & Gallese, V. in The New Cognitive tense forms are learned, represented and processed
system, whereas lexical items are stored in a Neurosciences (ed. Gazzaniga, M. S.) 539–552 (MIT Press, in a single connectionist network.
temporal lobe declarative memory system. Cambridge, Massachusetts, 2000). 35. Joanisse, M. F. & Seidenberg, M. S. Impairments in verb
9. Ullman, M. T. The declarative/procedural model of 23. Willingham, D. B. A neuropsychological theory of motor skill morphology after brain injury: a connectionist model. Proc.
lexicon and grammar. J. Psycholinguist. Res. 30, 37–69 learning. Psychol. Rev. 105, 558–584 (1998). Natl Acad. Sci. USA 96, 7592–7597 (1999).
(2001). 24. Alexander, G. E., Crutcher, M. D. & DeLong, M. R. in A connectionist model that attempts to simulate
10. Mishkin, M., Malamut, B. & Bachevalier, J. in Neurobiology Progress in Brain Research Vol. 85 (eds Uylings, H. B. M., previous empirically-demonstrated neurological
of Learning and Memory (eds Lynch, G., McGaugh, J. L. & Van Eden, C. G., DeBruin, J. P. C., Corner, M. A. & Feenstra, double dissociations between regular and irregular
Weinburger, N. W.) 65–77 (Guilford, New York, 1984). M. G. P.) 119–146 (Elsevier, New York, 1990). past tense forms.
A proposal that there are distinct memory systems. A discussion of evidence that the basal ganglia are 36. Plunkett, K. & Marchman, V. U-shaped learning
11. Squire, L. R. & Zola, S. M. Structure and function of composed of parallel circuits that subserve distinct and frequency effects in a multi-layered perceptron:
declarative and nondeclarative memory systems. Proc. Natl domains, and project to distinct frontal cortical implications for child language acquisition. Cognition 38,
Acad. Sci. USA 93, 13515–13522 (1996). regions. 43–102 (1991).

724 | OCTOBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

 REVIEWS

37. Pinker, S. & Prince, A. Regular and irregular morphology and 68. Ullman, M. T. et al. Neural correlates of lexicon and grammar: 98. Gabrieli, J. D. E., Cohen, N. J. & Corkin, S. The impaired
the psychological status of rules of grammar. Berkeley evidence from the production, reading, and judgement of learning of semantic knowledge following bilateral medial
Linguist. Soc. 17, 230–251 (1991). inflection in aphasia. Brain Lang. (in the press). temporal-lobe resection. Brain Cogn 7, 157–177 (1988).
38. Pinker, S. Rules of language. Science 253, 530–535 (1991). 69. Marin, O. S. M., Saffran, E. M. & Schwartz, M. F. 99. Postle, B. R. & Corkin, S. Impaired word-stem completion
39. Marcus, G. F. et al. Overregularization in language Dissociations of language in aphasia: implications for normal priming but intact perceptual identification priming with
acquisition. Monogr. Soc. Res. Child Dev. 57, 1–165 (1992). function. Ann. NY Acad. Sci. 280, 868–884 (1976). novel words: evidence from the amnesic patient H. M.
40. Ullman, M. T. The Computation of Inflectional Morphology. 70. Coslett, H. B. Dissociation between reading of derivational Neuropsychologia 15, 421–440 (1998).
Thesis, Massachusetts Inst. Technol. (1993). and inflectional suffixes in two phonological dyslexics. 100. Kensinger, E. A., Ullman, M. T. & Corkin, S. Bilateral medial
41. Prasada, S., Pinker, S. & Snyder, W. Some evidence that Academy of Aphasia (Nashville, 1986). temporal lobe damage does not affect lexical or
irregular forms are retrieved from memory but regular forms 71. Badecker, W. & Caramazza, A. The analysis and grammatical processing: evidence from the amnesic patient
are rule-generated. 31st Annu. Meet. Psychonomics Soc. morphological errors in a case of acquired dyslexia. Brain H. M. Hippocampus 11, 347–360 (2001).
(New Orleans, 1990). Lang. 32, 278–305 (1987). 101. Damasio, H., Grabowski, T. J., Tranel, D., Hichwa, R. D. &
42. Prasada, S. & Pinker, S. Generalization of regular and 72. Badecker, W. & Caramazza, A. Morphological composition Damasio, A. R. A neural basis for lexical retrieval. Nature
irregular morphological patterns. Lang. Cogn. Processes 8, in the lexical output system. Cogn. Neuropsychol. 8, 380, 499–505 (1996).
1–56 (1993). 335–367 (1991). 102. Bookheimer, S. Y., Zeffiro, T. A., Gaillaird, W. & Theodore, W.
43. Ullman, M. T. Acceptability ratings of regular and irregular 73. Ullman, M. T., Hickok, G. & Pinker, S. Irregular and regular Regional cerebral blood flow changes during the
past tense forms: evidence for a dual-system model of inflectional morphology in an aphasic. Brain Cogn 28, 88–89 comprehension of syntactically varying sentences. Soc.
language from word frequency and phonological (1995). Neurosci. Abstr. 19, 843 (1993).
neighbourhood effects. Lang. Cogn. Processes 14, 47–67 74. Coslett, H. B. A selective morphologic impairment in writing: 103. Newman, A. J., Pancheva, R., Ozawa, K., Neville, H. J. &
(1999). evidence from a phonological dysgraphic. Academy of Ullman, M. T. An event-related fMRI study of syntactic and
44. Stemberger, J. P. & MacWhinney, B. in Theoretical Aphasia (Montreal, 1988). semantic violations. J. Psycholinguist. Res. 30, 339–364
Morphology: Approaches in Modern Linguistics (eds 75. Marslen-Wilson, W. D. & Tyler, L. K. Dissociating types of (2001).
Hammond, M. & Noonan, M.) 101–116 (Academic Press, mental computation. Nature 387, 592–594 (1997). 104. Kuperberg, G. E. et al. Common and distinct neural
New York, 1988). 76. Marslen-Wilson, W. & Tyler, L. K. Rules, representations, and substrates for pragmatic, semantic, and syntactic
45. Van der Lely, H. K. J. & Ullman, M. T. Past tense morphology the English past tense. Trends Cogn. Sci. 2, 428–435 processing of spoken sentences: an fMRI study. J. Cogn.
in specifically language impaired and normally developing (1998). Neurosci. 12, 321–341 (2000).
children. Lang. Cogn. Processes 16, 177–217 (2001). 77. Hagiwara, H., Ito, T., Sugioka, Y., Kawamura, M. & Shiota, 105. Stromswold, K., Caplan, D., Alpert, N. & Rauch, S.
46. Clahsen, H., Eisenbeiss, S. & Sonnenstuhl, I. Morphological J.-I. Neurolinguistic evidence for rule-based nominal Localization of syntactic comprehension by positron
structure and the processing of inflected words. Theor. suffixation. Language 75, 739–763 (1999). emission tomography. Brain Lang. 52, 452–473 (1996).
Linguist. 23, 201–249 (1997). 78. Boller, F. & Duyckaerts, C. in Behavioral Neurology and 106. Ni, W. et al. An event-related neuroimaging study
47. Penke, M. & Krause, M. German noun plurals: a challenge Neuropsychology (eds Feinberg, T. E. & Farah, M. J.) distinguishing form and content in sentence processing.
to the dual-mechanism model. Brain Lang. (in the press). 521–544 (McGraw-Hill, New York, 1997). J. Cogn. Neurosci. 12, 120–133 (2000).
48. Bybee, J. L. & Moder, C. L. Morphological classes as natural 79. Nebes, R. D. in Behavioral Neurology and Neuropsychiatry 107. Embick, D., Marantz, A., Miyashita, Y., O’Neil, W. & Sakai,
categories. Language 59, 251–270 (1983). (eds Feinberg, T. E. & Farah, M. J.) 545–550 (McGraw-Hill, K. L. A syntactic specialization for Broca’s area. Proc. Natl
49. Fujiwara, M. & Ullman, M. T. in Proc. 12th Annu. CUNY New York, 1997). Acad. Sci. USA 97, 6150–6154 (2000).
Conf. Hum. Sentence Process. Vol. 12, 88 (CUNY Graduate 80. Sagar, H. J., Cohen, N. J., Sullivan, E. V., Corkin, S. & 108. Indefrey, P. et al. in Proc. 12th Annu. CUNY Conf. Hum.
School and University Center, CUNY, New York, 1999). Growdon, J. H. Remote memory function in Alzheimer’s and Sentence Process. Vol. 12, 93 (CUNY Graduate School and
50. Stanners, R. F., Neiser, J. J., Hernon, W. P. & Hall, R. Parkinson’s disease. Brain 111, 185–206 (1988). University Center, CUNY, New York, 1999).
Memory representation for morphologically related words. 81. Grossman, M. et al. Language comprehension and regional 109. Moro, A. et al. Syntax and the brain: disentangling grammar
J. Verb. Learn. Verb. Behav. 18, 399–412 (1979). cerebral defects in frontotemporal degeneration and by selective anomalies. Neuroimage 13, 110–118 (2001).
51. Marslen-Wilson, W., Hare, M. & Older, L. in Proc. 15th Annu. Alzheimer’s disease. Neurology 50, 157–163 (1998). 110. Jaeger, J. J. et al. A positron emission tomographic study of
Conf. Cogn. Sci. Soc. 1–6 (Hillsdale, New Jersey, 1993). 82. Gabrieli, J. D. E., Corkin, S., Mickel, S. F. & Growdon, J. H. regular and irregular verb morphology in English. Language
52. Sonnenstuhl, I., Eisenbeiss, S. & Clahsen, H. Morphological Intact acquisition and long-term retention of mirror-tracing 72, 451–497 (1996).
priming in the German mental lexicon. Cognition 72, skill in Alzheimer’s disease and in global amnesia. Behav. 111. Ullman, M. T., Bergida, R. & O’Craven, K. Distinct fMRI
203–236 (1999). Neurosci. 107, 899–910 (1993). activation patterns for regular and irregular past tense.
53. Ullman, M. T., Walenski, M., Prado, E., Ozawa, K. & 83. Saint-Cyr, J. A., Taylor, A. E. & Lang, A. E. Procedural Neuroimage 5, S549 (1997).
Steinhauer, K. in Proc. 14th Annu. CUNY Conf. Hum. learning and neostriatal dysfunction in man. Brain 111, 112. Rhee, J. et al. Neural substrates of English past tense
Sentence Process. Vol. 14, 64 (CUNY Graduate School and 941–959 (1988). generation. Abstr. Cogn. Neurosci. Soc. 8,131 (2001).
University Center, Philadelphia, 2001). 84. Beatty, W. W. et al. Preserved cognitive skills in dementia of 113. Indefrey, P. et al. A PET study of cerebral activation patterns
54. Leonard, L. B. Children with Specific Language Impairment the Alzheimer type. Arch. Neurol. 51, 1040–1046 (1994). induced by verb inflection. Neuroimage 5, S548 (1997).
(MIT Press, Cambridge, Massachusetts, 1998). 85. Bayles, K. A. Language function in senile dementia. Brain 114. Laine, M., Rinne, J. O., Krause, B. J., Teras, M. & Sipila, H.
55. Gopnik, M. Feature-blind grammar and dysphasia. Nature Lang. 16, 265–280 (1982). Left hemisphere activation during processing of
344, 715 (1990). 86. Schwartz, M. F., Marin, O. S. M. & Saffran, E. M. morphologically complex word forms in adults. Neurosci.
56. Van der Lely, H. K. J. & Stollwerck, L. A grammatical specific Dissociations of language function in dementia: a case Lett. 271, 85–88 (1999).
language impairment in children: an autosomal dominant study. Brain Lang. 7, 277–306 (1979). 115. Barrett, S. E. & Rugg, M. D. Event-related potentials and the
inheritance? Brain Lang. 52, 484–504 (1996). 87. Ullman, M. T. Evidence that lexical memory is part of the semantic matching of pictures. Brain Cogn 14, 201–212
57. Ullman, M. T. & Gopnik, M. in The McGill Working Papers in temporal lobe declarative memory, and that grammatical (1990).
Linguistics: Linguistic Aspects of Familial Language rules are processed by the frontal/basal-ganglia procedural 116. Nobre, A. C., Allison, T. & McCarthy, G. Word recognition in
Impairment Vol. 10 (ed. Matthews, J.) 81–118 (McGill, system. Brain Lang. (in the press). the human inferior temporal lobe. Nature 372, 260–263
Montreal, 1994). 88. Cappa, S. & Ullman, M. T. A neural dissociation in Italian (1994).
58. Ullman, M. T. & Gopnik, M. Inflectional morphology in a verbal morphology. Abstr. Cogn. Neurosci. Soc. 5, 63 117. Segalowitz, S. J. & Chevalier, H. in Handbook of
family with inherited specific language impairment. Appl. (1998). Neurolinguistics (eds Stemmer, B. & Whitaker, H. A.) 95–109
Psycholinguist. 20, 51–117 (1999). 89. Bozeat, S., Lambon Ralph, M. A., Patterson, K., Garrard, P. (Academic, San Diego, California, 1998).
59. Vargha-Khadem, F., Watkins, K., Alcock, K., Fletcher, P. & & Hodges, J. R. Non-verbal impairment in semantic 118. Neville, H., Nicol, J. L., Barss, A., Forster, K. I. & Garrett, M. F.
Passingham, R. Praxic and nonverbal cognitive deficits in a dementia. Neuropsychologia 38, 1207–1214 (2000). Syntactically based sentence processing classes: evidence
large family with genetically transmitted speech and 90. Graham, K. S., Patterson, K. & Hodges, J. R. Episodic from event-related brain potentials. J. Cogn. Neurosci. 3,
language disorder. Proc. Natl Acad. Sci. USA 92, 930–933 memory: new insights from the study of semantic dementia. 151–165 (1991).
(1995). Curr. Opin. Neurobiol. 9, 245–250 (1999). 119. Friederici, A. D., Pfeifer, E. & Hahne, A. Event-related brain
60. Vargha-Khadem, F. et al. Neural basis of an inherited speech 91. Patterson, K., Lambon Ralph, M. A., Hodges, J. R. & potentials during natural speech processing: effects of
and language disorder. Proc. Natl Acad. Sci. USA 95, McClelland, J. L. Deficits in irregular past-tense verb semantic, morphological and syntactic violations. Brain Res.
12695–12700 (1998). morphology associated with degraded semantic Cogn. Brain Res. 1, 183–192 (1993).
61. Bellugi, U., Bihrle, A., Jernigan, T., Trauner, D. & Doherty, S. knowledge. Neuropsychologia 39, 709–724 (2001). 120. Friederici, A. D., Hahne, A. & Mecklinger, A. The temporal
Neuropsychological, neurological, and neuroanatomical 92. Young, A. B. & Penney, J. B. in Parkinson’s Disease and structure of syntactic parsing: early and late effects elicited
profile of Williams syndrome. Am. J. Med. Genet. 6 (Suppl.), Movement Disorders (eds Jankovic, J. & Tolosa, E.) 1–11 by syntactic anomalies. J. Exp. Psychol. Learn. Mem. Cogn
115–125 (1990). (Williams & Wilkins, Baltimore, Maryland, 1993). 22, 1219–1248 (1996).
62. Clahsen, H. & Almazan, M. Syntax and morphology in 93. Dubois, B., Boller, F., Pillon, B. & Agid, Y. in Handbook of 121. Friederici, A. D., Hahne, A. & Von Cramon, D. Y. First-pass
Williams syndrome. Cognition 68, 167–198 (1998). Neuropsychology Vol. 5 (eds Boller, F. & Grafman, J.) versus second-pass parsing processes in a Wernicke’s and
63. Bromberg, H. et al. in Proc. 18th Annu. Boston Univ. Conf. 195–240 (Elsevier, Amsterdam, 1991). a Broca’s aphasic: electrophysiological evidence for a
Lang. Dev. (Boston, Massachusetts, 1994). 94. Harrington, D. L., Haaland, K. Y., Yeo, R. A. & Marder, E. double dissociation. Brain Lang. 62, 311–341 (1998).
64. Alexander, M. P. in Behavioral Neurology and Procedural memory in Parkinson’s disease: impaired motor 122. Friederici, A. D. & Frisch, S. Verb-argument structure
Neuropsychology (eds Feinberg, T. E. & Farah, M. J.) but not visuoperceptual learning. J. Clin. Exp. Neuropsychol. processing: the role of verb-specific and argument-specific
133–150 (McGraw-Hill, New York, 1997). 12, 323–339 (1990). information. J. Mem. Lang. 43, 476–507 (2000).
65. Damasio, A. R. Aphasia. N. Engl. J. Med. 326, 531–539 95. Grossman, M., Carvell, S. & Peltzer, L. The sum and 123. Osterhout, L. & Holcomb, P. J. Event-related brain potentials
(1992). substance of it: the appreciation of mass and count qualifiers elicited by syntactic anomaly. J. Mem. Lang. 31, 785–806
66. Tyler, L. K., Ostrin, R. K., Cooke, M. & Moss, H. E. in Parkinson’s disease. Brain Lang. 44, 351–384 (1993). (1992).
Automatic access of lexical information in Broca’s aphasics: 96. Illes, J., Metter, E. J., Hanson, W. R. & Iritani, S. Language 124. Weyerts, H., Penke, M., Dohrn, U., Clahsen, H. & Münte, T. F.
against the automaticity hypothesis. Brain Lang. 48, production in Parkinson’s disease: acoustic and linguistic Brain potentials indicate differences between regular and
131–162 (1995). considerations. Brain Lang. 33, 146–160 (1988). irregular German plurals. Neuroreport 8, 957–962 (1997).
67. Farah, M. J. & Grossman, M. in Behavioral Neurology and 97. Lieberman, P. et al. Speech production, syntax 125. Penke, M. et al. How the brain processes complex words:
Neuropsychology (eds Feinberg, T. E. & Farah, M. J.) comprehension, and cognitive deficits in Parkinson’s an event-related potential study of German verb inflections.
473–477 (McGraw-Hill, New York, 1997). disease. Brain Lang. 43, 169–189 (1992). Brain Res. Cogn. Brain Res. 6, 37–52 (1997).

NATURE REVIEWS | NEUROSCIENCE VOLUME 2 | OCTOBER 2001 | 7 2 5

REVIEWS

126. Gross, M., Say, T., Kleingers, M., Münte, T. F. & Clahsen, H. 132. Curran, H. V. in The New Cognitive Neurosciences (ed. Acknowledgements
Human brain potentials to violations in morphologically Gazzaniga, M. S.) 797–804 (MIT Press, Cambridge, Support was provided by a McDonnell-Pew grant in Cognitive
complex Italian words. Neurosci. Lett. 241, 83–86 Massachusetts, 2000). Neuroscience, and by grants from the National Institutes of Health
(1998). 133. Ullman, M. T. Neural plasticity and morphological and the, the National Science Foundation and the Department of
127. Münte, T. F., Say, T., Clahsen, H., Schiltz, K. & Kutas, M. processing in agrammatic aphasia: implications for Defence.
Decomposition of morphologically complex words in recovery. Abstr. Cogn. Neurosci. Soc. 7, 53
English: evidence from event-related brain potentials. Brain (2000).
Res. Cogn. Brain Res. 7, 241–253 (1999). 134. Dell, G. S. A spreading-activation theory of retrieval Online links
128. Newman, A., Izvorski, R., Davis, L., Neville, H. & Ullman, M. in sentence production. Psychol. Rev. 93, 283–321
T. Distinct electrophysiological patterns in the processing of (1986). DATABASE LINKS
regular and irregular verbs. Abstr. Cogn. Neurosci. Soc. 6, 135. Halle, M. Prolegomena to a theory of word formation. The following terms in this article are linked online to:
47 (1999). Linguistic Inq. 4, 3–16 (1973). MIT Encyclopedia of Cognitive Sciences:
129. Ullman, M. T., Newman, A., Izvorski, R. & Neville, H. in Proc. 136. Anderson, S. R. A-morphous morphology (Cambridge Univ. http://cognet.mit.edu/MITECS/
13th Annu. CUNY Conf. Hum. Sentence Process. Vol. 13, Press, New York, 1992). magnetic resonance imaging | positron emission tomography
87 (CUNY Graduate School and University Center, La Jolla, 137. Jackendoff, R. S. Morphological and semantic regularities in OMIM: http://www.ncbi.nlm.nih.gov/Omim/
California, 2000). the lexicon. Language 51, 639–671 (1975). Alzheimer’s disease | Huntington’s disease | Parkinson’s disease
130. Lavric, A., Pizzagalli, D., Forstmeier, S. & Rippon, G. 138. Aronoff, M. Word Formation in Generative
Mapping dissociations in verb morphology. Trends Cogn. Grammar (MIT Press, Cambridge, Massachusetts, FURTHER INFORMATION
Sci. 5, 301–308 (2001). 1976). MIT Encyclopedia of Cognitive Sciences:
131. Rhee, J., Pinker, S. & Ullman, M. T. 139. Ling, C. X. & Marinov, M. A symbolic model of the http://cognet.mit.edu/MITECS/
A magnetoencephalographic study of English past tense nonconscious acquisition of information. Cogn. Sci. 18, grammar, neural basis of | language, neural basis of | lexicon,
production. Abstr. Cogn. Neurosci. Soc. 6, 47 (1999). 595–621 (1994). neural basis of | linguistics and language | psycholinguistics

726 | OCTOBER 2001 | VOLUME 2 www.nature.com/reviews/neuro

 ONLINE

Database links At a glance

• Several models have been proposed to account for the neurocognitive
MIT Encyclopedia of Cognitive Sciences basis of the mental lexicon (a repository of stored words) and the
http://cognet.mit.edu/MITECS/ mental grammar (which captures the regularities of language). The
declarative/procedural model argues that lexicon and language depend
magnetic resonance imaging on two neural systems that are intensively studied in the context of
http://cognet.mit.edu/MITECS/Entry/ugurbil memory: declarative and procedural memory.
• The declarative/procedural model links lexicon with the declarative
positron emission tomography system and with brain structures in temporal/temporoparietal regions.
http://cognet.mit.edu/MITECS/Entry/raichle On the other hand, the model links grammar with the procedural
system, and with structures in the basal ganglia and frontal cortex.
OMIM • The declarative/procedural model makes a set of specific predictions
http://www.ncbi.nlm.nih.gov/Omim/ about the neurocognitive basis of lexicon and grammar, regarding
their separability, computation, domain generality and localization.
Alzheimer’s disease These predictions, which have been thoroughly tested in the context of
http://www.ncbi.nlm.nih.gov/htbin-post/Omim/dispmim?104300 the use of regular versus irregular word forms (walk–walked versus
go–went), have been helpful in contrasting this model with other com-
Huntington’s disease peting perspectives.
http://www.ncbi.nlm.nih.gov/htbin-post/Omim/dispmim?143100 • Several lines of evidence support the declarative/procedural model
over alternative views. This evidence has come from psycholinguistic
Parkinson’s disease studies, the analysis of developmental disorders of language, neurolog-
http://www.ncbi.nlm.nih.gov/htbin-post/Omim/dispmim?168600 ical cases, haemodynamic studies and neurophysiological observa-
tions. Collectively, the data show a double dissociation. On the one
Further information hand, there is a link between lexicon, associative-memory markers, the
knowledge of facts and events, and temporal/temporoparietal regions.
MIT Encyclopedia of Cognitive Sciences On the other, there is a link between grammar, motor and cognitive
http://cognet.mit.edu/MITECS/ skills, and structures in the frontal lobe and the basal ganglia.
• The declarative/procedural model has several implications. First, stud-
grammar, neural basis of ies of declarative and procedural memory should help to elucidate the
http://cognet.mit.edu/MITECS/Entry/caplan neural bases of lexicon and grammar, and vice versa. Second, the
model has clinical implications for people with developmental or
language, neural basis of adult-onset disorders of grammar, as they might recover through the
http://cognet.mit.edu/MITECS/Entry/dronkers memorization of complex forms using the declarative system. Last, the
existence of systems that subserve language in humans and are homol-
lexicon, neural basis of ogous to systems present in other animals has implications for the evo-
http://cognet.mit.edu/MITECS/Entry/caramazza lution of language.

linguistics and language

http://cognet.mit.edu/MITECS/Entry/ling.intro

psycholinguistics
http://cognet.mit.edu/MITECS/Entry/clark

Biography
Michael Ullman is a faculty member of the Department of
Neuroscience and at the Georgetown Institute for Cognitive and
Computational Sciences at Georgetown University, Washington DC. He
holds secondary appointments in Linguistics, Psychology and
Neurology. He received his Ph.D. from the Department of Brain and
Cognitive Sciences at the Massachusetts Institute of Technology. His
group examines the neural, psychological and computational underpin-
nings of language, focusing on the mental lexicon of memorized words
and the mental grammar, which underlies rule-governed behaviour. He
uses various methodological approaches, including neuroimaging, neu-
ropsychology and psycholinguistics, to probe the structure of words
(morphology), and phrases and sentences (syntax), in several languages,
including English, Italian, Spanish and Japanese.