Elsevier

Cognition

Volume 92, Issues 1–2, May–June 2004, Pages 231-270
Cognition

Contributions of memory circuits to language: the declarative/procedural model

https://doi.org/10.1016/j.cognition.2003.10.008Get rights and content

Abstract

The structure of the brain and the nature of evolution suggest that, despite its uniqueness, language likely depends on brain systems that also subserve other functions. The declarative/procedural (DP) model claims that the mental lexicon of memorized word-specific knowledge depends on the largely temporal-lobe substrates of declarative memory, which underlies the storage and use of knowledge of facts and events. The mental grammar, which subserves the rule-governed combination of lexical items into complex representations, depends on a distinct neural system. This system, which is composed of a network of specific frontal, basal-ganglia, parietal and cerebellar structures, underlies procedural memory, which supports the learning and execution of motor and cognitive skills, especially those involving sequences. The functions of the two brain systems, together with their anatomical, physiological and biochemical substrates, lead to specific claims and predictions regarding their roles in language. These predictions are compared with those of other neurocognitive models of language. Empirical evidence is presented from neuroimaging studies of normal language processing, and from developmental and adult-onset disorders. It is argued that this evidence supports the DP model. It is additionally proposed that “language” disorders, such as specific language impairment and non-fluent and fluent aphasia, may be profitably viewed as impairments primarily affecting one or the other brain system. Overall, the data suggest a new neurocognitive framework for the study of lexicon and grammar.

Introduction

The study of language has focused largely on language itself. That is, in order to understand the representation, processing, development, neural correlates and other aspects of language, most theories and investigations have directed their attention to language. This is unsurprising, and not only because of the obvious point that directly investigating a domain generally elucidates it. Additionally, the apparent uniqueness of human language has drawn attention away from evidence suggesting the existence of biological and computational substrates that are shared between language on the one hand, and non-language domains in humans and animals on the other.

Brain organization and evolutionary principles both lead to an expectation of commonalities between language and non-language domains. First, a number of brain structures seem to be organized topographically, with sub-regions performing analogous computations on different domains of information, as a function of each sub-region's particular set of inputs and outputs. This type of brain organization has been claimed for the cerebellum, for various sub-cortical structures, including the basal ganglia, and for certain cortical regions, in particular in frontal cortex (Alexander et al., 1986, Middleton and Strick, 2000a, Shimamura, 1995). This suggests that analogous computations may underlie a range of cognitive domains, including language. Second, commonalities between language and non-language domains are not surprising from an evolutionary perspective, given the well-established pattern that biological structures tend to evolve from already-existing structures (Maynard Smith, 1975/1993, Mayr, 1963).

So, whether or not there are aspects of the neurocognition of language that are unique to this faculty and to our species, much and perhaps most aspects of language are likely to not be unique. Importantly, other cognitive domains are much better understood than language in a number of respects, including their neuroanatomy, physiology, biochemistry, evolution, development, and neural computation. This follows from the fact that most other domains have benefited from the development of animal models which allow for invasive and highly informative techniques that are not permissible to perform on humans.

A reasonable research program would thus be to identify domains that share commonalities with language: their underlying neural and computational systems will be promising candidates for those subserving language. Importantly, if the systems underlying the target domains are well understood, they should yield clear predictions about language, based solely on non-language theories and data. This should provide far greater predictive power about language than research restricted to language, whose theories and predictions are generally if not always derived from evidence solely related to language itself. Since the research tools we have at our disposal to understand language are quite impoverished compared to those available for the investigation of other domains, a research program limited to language necessarily restricts language theories and their predictions. In contrast, theories that are motivated by non-language domains as well as language have a much wider potential predictive range for language, and thus are likely to lead to important advances in our understanding of this faculty. This should be particularly likely for areas of research that have been given greater attention in non-language than language domains, such as functional neuroanatomy, physiology, biochemistry, neuroendocrinology, and pharmacology. Importantly, the converse holds as well. That is, our understanding of many aspects of the representation, computation, and processing of language has progressed far beyond that of many other cognitive domains. So, the demonstration of neurocognitive links between language and other domains should also improve our understanding of the latter. Note that I am not arguing that research programs directed solely at language should be replaced by those examining language in the context of other cognitive domains. Rather I am maintaining that the latter type of research program must crucially complement the former.

It is in this general spirit that my colleagues and I have proposed and explored the declarative/procedural (DP) model of language (Ullman, 2001a, Ullman, 2001c, Ullman et al., 1997). The basic premise of the DP model is that important aspects of the distinction between the mental lexicon and the mental grammar in language are tied to the distinction between declarative and procedural memory – two memory systems which have been implicated in non-language functions in humans and other animals (Eichenbaum and Cohen, 2001, Mishkin et al., 1984, Schacter and Tulving, 1994, Squire and Knowlton, 2000). That is, lexical memory depends largely on the declarative memory system, whereas aspects of grammar depend on the procedural memory system. Importantly, multiple characteristics of these two systems, including their computational, neuroanatomical, physiological and biochemical substrates, have been quite well studied, and thus should lead to important predictions about language.

Language depends upon a memorized “mental lexicon” and a computational “mental grammar” (Chomsky, 1965, Chomsky, 1995, de Saussure, 1959, Pinker, 1994).1 The mental lexicon is a repository of all idiosyncratic word-specific information. Thus, it includes all words whose phonological forms and meanings cannot be derived from each other (i.e. their sound–meaning pairings are arbitrary), such as the non-compositional (“simple”) word cat. It also contains other irregular—i.e. not entirely derivable—word-specific information, such as the particular arguments that must accompany a given verb (e.g. hit takes a direct object), and any unpredictable forms that a word takes (e.g. teach takes the irregular past-tense taught). The mental lexicon may comprise other distinctive information as well, smaller or larger than words: bound morphemes (e.g. the -ed or -ness suffixes, as in walked or happiness), and representations of complex linguistic structures whose meanings cannot be transparently derived from their parts (e.g. idiomatic phrases, such as kick the bucket).

Many regularities can also be found in language. These regularities can be captured by rules of grammar. The rules constrain how lexical forms and abstract symbols or features (e.g. walk, -ed, Verb, Past-Tense) can combine to make complex representations. The rules crucially allow us to interpret the meanings of complex forms even if we have not heard or seen them before. Thus, in the sentence “Clementina glicked the plag”, we know that Clementina did something in the past to some entity. The rules specify not only the sequential order (precedence) of lexical items, but also their hierarchical relations, e.g. that a verb phrase (glicked the plag) can contain a noun phrase (the plag). Such rule-governed behavior is found at various levels in language, including in the structure of phrases and sentences (syntax), and of complex words such as walked or glicked (morphology). Importantly, the rules and constraints are a type of mental knowledge in that they underlie our individual mental capacity to produce and comprehend complex forms. It is often argued that aspects of the ability to learn, represent and compute the rules and constraints that underlie grammar depend on innately-specified mental constructs (Chomsky, 1995). The learning of grammatical knowledge, and the knowledge itself, are generally not available to conscious access (Fodor, 1983); that is, they are implicit. It has been argued that grammatical processing is not influenced by other cognitive domains; that is, the underlying system is “straight-through”, or “informationally encapsulated” (Fodor, 1983, Frazier and Fodor, 1978). Moreover, at least certain aspects of grammatical processing are fast as well as automatic, in that they are not under conscious control but are rather triggered by the linguistic stimulus (Fodor, 1983, Friederici, 2002).

The two language capacities interact in a number of ways. First, the grammar combines lexical items into complex structures. Second, even though certain representations of complex linguistic structures that have idiosyncratic meanings (e.g. idioms) may be stored in the lexicon, their structures still generally follow the rules of grammar. Third, although “regular” (i.e. transparent; derivable) complex representations (e.g. walked; the cat) could be computed anew each time they are used (e.g. walk+-ed), and must be if they are new (e.g. glicked), they could in principle also be stored in the mental lexicon after being encountered. Finally, a general pattern observed in languages is that idiosyncratic, exceptional forms and meanings are selected preferentially over general, derivable ones (the “Elsewhere” principle; Halle and Marantz, 1993, Kiparsky, 1982, Pinker, 1984), suggesting that stored lexical items take precedence over those composed by the mental grammar.

The declarative and procedural memory systems have been intensively studied in humans and in several animal models, including monkeys and rodents. The demonstration of numerous double dissociations has shown that the two systems are largely independent from each other, though they interact in a number of ways (Eichenbaum and Cohen, 2001, Mishkin et al., 1984, Poldrack and Packard, 2003, Schacter and Tulving, 1994, Squire and Knowlton, 2000). As will be seen, the two memory systems share a number of characteristics with the two language capacities. Importantly, research has begun to elucidate the specific computational, developmental, anatomical, cellular, molecular and other aspects of these two systems across species. These findings lead to highly specific predictions about language.

The “declarative” memory system (Eichenbaum and Cohen, 2001, Mishkin et al., 1984, Schacter and Tulving, 1994, Squire and Knowlton, 2000) has been implicated in the learning, representation, and use of knowledge about facts (“semantic knowledge”) and events (“episodic knowledge”). It is important for the very rapid learning (e.g. based on a single stimulus presentation) of arbitrarily-related information – that is, for the associative binding of information (Cohen et al., 1997, Eichenbaum and Cohen, 2001, Squire and Knowlton, 2000). It has been argued that the information learned by this system is not informationally encapsulated, being accessible to multiple mental systems (Squire & Zola, 1996). Moreover, at least part of this knowledge can be consciously (“explicitly”) recollected.

Declarative memory depends, first of all, on medial temporal lobe structures: the hippocampal region (the dentate gyrus, the subicular complex, and the hippocampus itself), entorhinal cortex, perirhinal cortex, and parahippocampal cortex (Squire and Knowlton, 2000, Suzuki and Eichenbaum, 2000). The hippocampus projects to midline diencephalic nuclei, in particular the mammillary bodies and portions of the thalamus. These structures also play an important role in declarative memory, though they are less well studied than the medial-temporal lobe. The medial temporal structures are hierarchically organized: evidence from non-human primates indicates that the hippocampal region is heavily connected with entorhinal cortex, which is strongly connected with both the perirhinal and parahippocampal cortices, which are in turn connected extensively with temporal and parietal neocortical regions (Suzuki & Amaral, 1994).

The medial-temporal complex appears to subserve several related memory functions, including the encoding, consolidation and retrieval of new memories (Buckner and Wheeler, 2001, Eichenbaum and Cohen, 2001, Squire and Knowlton, 2000). Memories eventually (in humans, over months to years) become largely independent of the medial temporal lobe structures, and dependent upon neocortical regions, particularly in the temporal lobes (Hodges and Patterson, 1997, Squire et al., 2001). Different regions of the temporal lobes may be specialized for different types of knowledge (Damasio et al., 1996, Martin et al., 2000). It has been posited that medial temporal lobe structures associate or “bind” inputs from cortical regions, which together store an entire memory (Alvarez and Squire, 1994, McClelland et al., 1995).

The term “declarative memory system” is used here to refer to the entire system involved in the learning, representation and use of the relevant information (Eichenbaum, 2000), not just to those brain structures underlying the learning of new memories. Indeed, other brain structures also play a role in this system, although the precise regions and functions are still not entirely clear. First of all, prefrontal regions have been implicated in numerous studies (Buckner and Wheeler, 2001, Tulving et al., 1994). Ventro-lateral prefrontal cortex (VL-PFC), which corresponds to the inferior frontal gyrus and Brodmann's areas (BA) 44, 45 and 47 (Damasio, 1995), plays a role in the encoding of new memories and the selection or retrieval of declarative knowledge (Buckner and Wheeler, 2001, Thompson-Schill et al., 1997, Wagner et al., 1998). Two functionally and anatomically distinct sub-regions have been implicated: posterior/dorsal inferior frontal cortex (BA 6/44) is strongly implicated in aspects of phonology, whereas anterior/ventral inferior frontal cortex (BA 45/47) is more important for semantics (Fiez, 1997, Poldrack et al., 1999). Their precise roles may be closely related to working memory (Buckner and Wheeler, 2001, Moscovitch, 1992). Indeed, neuroimaging studies show that VL-PFC is consistently activated in working memory tasks (Smith & Jonides, 1999), and, within the same subjects, in both retrieval and working memory tasks (Braver et al., 2001). Additionally, anterior frontal-polar cortex (BA 10) is implicated in the retrieval of memories, or in the monitoring of that retrieval (Buckner & Wheeler, 2001). This area is also associated with working memory (Braver et al., 2001, McLeod et al., 1998). Finally, evidence suggests that portions of the cerebellum are involved in searching, retrieving or otherwise processing declarative memories (Desmond and Fiez, 1998, Ivry and Fiez, 2000).

The declarative memory system is closely related to the “ventral” stream system (Goodale and Milner, 1992, Ungerleider and Mishkin, 1982). This system is rooted in inferior and lateral temporal-lobe structures. It underlies the formation of perceptual representations of objects and their relations. These representations underlie the recognition and identification of objects and the long-term storage of knowledge about objects (Goodale, 2000). The ventral system is thus a memory-based system, feeding representations into long-term (declarative) memory, and comparing those representations with new ones. It has also been argued that humans are conscious of aspects of ventral stream functioning (Norman, 2002).

The declarative memory system has been intensively studied not only at functional and neuroanatomical levels, but also at cellular and molecular levels (Curran, 2000, Lynch, 2002). Acetylcholine in particular plays an important role in declarative memory and hippocampal function (Freo et al., 2002, Packard, 1998). Thus, levels of choline acetyl transferase, the synthesizing enzyme for acetylcholine, correlate with declarative memory abilities (Baskin et al., 1999). Pharmacological manipulations of the cholinergic system in normal, healthy adults have also implicated acetylcholine in declarative memory (Nissen et al., 1987, Rammsayer et al., 2000). For example, acetylcholine esterase inhibitors, which prolong the activity of acetylcholine at the synapse, improve declarative memory (Ballard, 2002, Hammond et al., 1987).

Evidence also suggests that the declarative memory system is affected by estrogen (Phillips and Sherwin, 1992, Sherwin, 1988), perhaps via the modulation of acetylcholine (Packard, 1998, Shughrue et al., 2000). Declarative memory abilities and medial temporal-lobe function are linked to estrogen, via organizational effects in utero, and/or activational effects later on. Estrogen improves declarative memory in women (Maki and Resnick, 2000, Sherwin, 1998) and men (Kampen and Sherwin, 1996, Miles et al., 1998), and strengthens the cellular and molecular correlates of long-term hippocampal learning (McEwen et al., 1998, Woolley and Schwartzkroin, 1998). Testosterone, which is the main source of estrogen in men, also improves their memory (Cherrier et al., 2001). Women with Turner's syndrome, who do not produce estrogen, have worse declarative memory (which improves with estrogen therapy; Ross, Roeltgen, Feuillan, Kushner, & Cutler, 2000) and smaller hippocampi than control subjects (Murphy et al., 1993). During declarative memory tasks, increased estrogen (i.e. hormone replacement therapy) in healthy post-menopausal women leads to greater blood flow activation changes in medial temporal lobe regions, including the hippocampus (Maki and Resnick, 2000, Resnick et al., 1998).

The “procedural memory” system (Eichenbaum and Cohen, 2001, Mishkin et al., 1984, Schacter and Tulving, 1994, Squire and Knowlton, 2000) subserves the learning of new, and the control of established, sensori-motor and cognitive “habits”, “skills”, and other procedures, such as riding a bicycle and skilled game playing. The system is commonly referred to as an “implicit memory system” because both the learning of the knowledge, and the knowledge itself, are generally not available to conscious access. Note that I use the term “procedural memory” to refer only to one type of implicit, non-declarative, memory system (Squire & Knowlton, 2000), not all non-declarative or implicit memory systems. Moreover, and analogously to how I use the term “declarative memory”, the term “procedural memory” is used here to refer to the entire system involved in the learning, representation and use of the relevant knowledge, not just to those parts of the system underlying the learning of new memories.

Although procedural memory is less well understood than declarative memory, its functional characteristics and neural bases are beginning to be revealed. Functionally, the system may be characterized as subserving aspects of the learning and processing of context-dependent stimulus-response rule-like relations (Knowlton et al., 1996, Packard and Knowlton, 2002, Poldrack et al., 1999, White, 1997, Wise et al., 1996). The system seems to be especially important for learning and processing these relations in the context of real-time sequences – whether the sequences are serial or abstract, or sensori-motor or cognitive (Aldridge and Berridge, 1998, Boecker et al., 2002, Doyon et al., 1997, Graybiel, 1995, Howard and Howard, 1997, Saint-Cyr et al., 1988, Willingham, 1998). Learning in the system is gradual, in that it occurs on an ongoing basis during multiple presentations of stimuli and responses – unlike the fast learning subserved by the declarative memory system. The relations are rule-like in that they are rigid, inflexible, and not influenced by other mental systems (Mishkin et al., 1984, Squire and Zola, 1996). Thus, this system, unlike declarative memory, appears to be informationally encapsulated (Squire & Zola, 1996). The rules apply quickly and automatically, in that the response is triggered by the stimulus rather than being under conscious control. The procedural system plays a role not only in learning and processing new sequences but also in the coordination of innate ones, such as “grooming sequences” in rodents, which follow a stereotyped sequence of “syntactic chains” that combine up to dozens of actions into a predictable order (Aldridge & Berridge, 1998). Intriguingly, although fixed linear sequences and possibly probabilistic sequences can be learned by monkeys, apes and humans, hierarchical structure is apparently commonly used and easily learned only by humans, though it has been observed in apes (Conway & Christiansen, 2001).

The procedural memory system is composed of a network of brain structures. The system is rooted in frontal/basal-ganglia circuits, with a likely role for portions of parietal cortex, superior temporal cortex and the cerebellum (De Renzi, 1989, Heilman et al., 1997, Hikosaka et al., 2000, Mishkin et al., 1984, Rizzolatti et al., 2000, Schacter and Tulving, 1994, Squire and Zola, 1996).

The basal ganglia are a set of sub-cortical structures, including the neostriatum, globus pallidus, sub-thalamic nucleus, and substantia nigra (Wise et al., 1996). In primates the neostriatum is composed of two structures: the putamen and the caudate nucleus. The putamen is particularly important for motor functions, whereas the caudate appears to underlie aspects of cognition (Alexander et al., 1986, Middleton and Strick, 2000a). Dorsal aspects of these structures (the dorsal striatum) play an important role in procedural memory, whereas ventral aspects (the ventral striatum) may be more important in affective (emotional) memory (Packard & Knowlton, 2002). The basal ganglia have been implicated in a number of functions, including implicit procedural learning in general (Eichenbaum and Cohen, 2001, Mishkin et al., 1984, Schacter and Tulving, 1994, Squire and Knowlton, 2000); stimulus-response learning (Packard and Knowlton, 2002, White, 1997), in particular of egocentric (body-centered) sensori-motor relations (White, 1997); probabilistic rule learning (Knowlton et al., 1996, Poldrack et al., 1999); sequence learning (Aldridge and Berridge, 1998, Boecker et al., 1998, Doyon et al., 1997, Graybiel, 1995, Peigneux et al., 2000, Willingham, 1998); reinforcement learning (in the dorsal striatum; cf. reward-based, in the ventral striatum) (Doya, 2000, Packard and Knowlton, 2002, White, 1997); real-time motor planning and control (Wise et al., 1996), particularly that which involves precise timing (Penhune, Zattore, & Evans, 1998) and the selection or switching among multiple motor programs (Haaland, Harrington, O'Brien, & Hermanowicz, 1997); mental rotation (Podzebenko, Egan, & Watson, 2002); interval timing and rhythm (Meck and Benson, 2002, Schubotz and von Cramon, 2001); and the context-dependent rule-based selection (Peigneux et al., 2000, Wise et al., 1996) and maintenance in working memory (Menon, Anagnoson, Glover, & Pfefferbaum, 2000) of and the real-time shifting (of attention, or focus) between sets, functions or programs. Importantly, these apparently disparate functions appear to be quite intimately related (Meck and Benson, 2002, Wise et al., 1996), although the precise nature of their relations and interactions are not yet understood.

The basal ganglia, in particular the neostriatum, receive input projections from multiple cortical areas, especially in frontal cortex, but also from other structures, including the medial temporal lobe (Alexander and Crutcher, 1990, Middleton and Strick, 2000b, Wise et al., 1996). The basal ganglia send outputs via the thalamus to neocortex, largely in frontal regions (Middleton & Strick, 2000b). The basal ganglia structures themselves are highly interconnected. Perhaps most importantly, the neostriatum projects to both the “direct” and the “indirect” pathways within the basal ganglia. The two pathways have opposing effects on the basal ganglia's outputs to frontal cortex via the thalamus. The direct pathway inhibits, whereas the indirect pathway disinhibits, the inhibitory projections from the basal ganglia to the thalamus. Thus, the direct pathway ultimately results in the disinhibition, and the indirect pathway in the inhibition, of the excitatory projections from thalamus to frontal cortex. So, frontal cortical activity is disinhibited by the direct pathway, and inhibited by the indirect pathway. The posited selection and set shifting functions of the basal ganglia may be attributed to the interaction between the direct and indirect pathways: a given cortical “set” or “program” can be disinhibited, while the rest are inhibited (Young & Penney, 1993). Imbalances between the two pathways can lead to the excessive inhibition or disinhibition of the functions that depend on the frontal cortical regions to which the basal ganglia project. This is thought to explain the inhibited/suppressed and disinhibited/unsuppressed motor and other behaviors found in Parkinson's, Huntington's and other diseases affecting the basal ganglia (Young & Penney, 1993). The two pathways are modulated by dopaminergic projections from the substantia nigra to the neostriatum. Indeed, it is this bundle of neurons which degenerates in Parkinson's disease. These nigrostriatal dopaminergic neurons appear to have a role not only in the real-time modulation of the direct and indirect pathways, but also in the reinforcement learning functions of the basal ganglia, specifically in memory consolidation (Doya, 2000, Packard and Knowlton, 2002, White, 1997).

Importantly, the various connections within the basal ganglia contain parallel and largely functionally segregated “circuits” (i.e. channels) (Alexander and Crutcher, 1990, Alexander et al., 1986, Middleton and Strick, 2000a, Middleton and Strick, 2000b). Each circuit receives projections at the neostriatum – some circuits primarily at the caudate, others at the putamen – from a particular set of cortical and sub-cortical structures. Each circuit then follows the split between the direct and indirect pathways, and projects via the thalamus to a particular cortical region, largely in frontal cortex. This cortical output area in turn projects back to the portion of the neostriatum that receives inputs for that circuit. Thus, there is at least in part a closed loop with feedback. For example, a basal ganglia “motor circuit” projects to frontal motor areas, a “prefrontal” circuit projects to prefrontal regions, and other circuits project to other frontal areas. The different basal ganglia circuits have similar synaptic organizations, suggesting that similar neuronal operations might be performed at comparable stages of each circuit (Alexander et al., 1990, Middleton and Strick, 2000b). Thus, the various parallel circuits of the basal ganglia seem to perform analogous computations; these are applied to different sets of information from different domains, depending on the particular set of input regions and frontal cortical output destinations of a given circuit (Middleton & Strick, 2000b). So, if the basal ganglia play a role in grammar, that role should be computationally analogous to that which the structures play in other domains, with the grammar-subserving circuits possibly projecting to somewhat different frontal cortical regions (e.g. in Broca's area) than other circuits.

Certain frontal cortical regions are also critical for procedural memory, in a manner which seems to be closely related to the functions of the basal ganglia. In the macaque, the basal ganglia project via the thalamus to pre-motor regions, including the supplementary motor area (SMA) and the general region of area F5 (Middleton and Strick, 2000a, Middleton and Strick, 2000b). F5 is a well-studied ventral pre-motor region that is a likely homologue of human BA 44 in Broca's area (Rizzolatti, Fadiga, Gallese, & Fogassi, 1996) (Broca's area is defined here as a portion of human inferior frontal cortex, including and perhaps limited to cortex corresponding to BA 44 – pars opercularis – and BA 45 – pars triangularis; Amunts et al., 1999). Each of these frontal regions plays an important role in the procedural memory system.

First of all, pre-motor regions (Harrington et al., 2000, Jenkins et al., 1994), including SMA (Jenkins et al., 1994) and pre-SMA (rostral SMA) (Boecker et al., 1998, Hikosaka et al., 1996), are implicated in motor sequence learning in humans. Motor sequence learning in macaques also depends on SMA and pre-SMA (Hikosaka et al., 2000). Lateral pre-motor cortex and SMA are also involved in mental rotation (Jordan et al., 2001, Kosslyn et al., 1998, Podzebenko et al., 2002), and lateral pre-motor and pre-SMA regions are implicated in aspects of timing or rhythm (Schubotz & von Cramon, 2001).

Broca's area is another critical component of the procedural memory system. Evidence suggests that motor sequence learning depends on left inferior frontal cortex (Peigneux et al., 1999), including Broca's area (Conway and Christiansen, 2001, Dominey et al., 2003), and its right homologue (Doyon, Owen, Petrides, Sziklas, & Evans, 1996). Broca's area in humans seems particularly important for learning sequences which contain abstract and potentially hierarchical structure, as opposed to fixed linear sequences (Conway and Christiansen, 2001, Dominey et al., 2003, Goschke et al., 2001). Broca's area has also been implicated in mental rotation tasks (Jordan et al., 2001, Podzebenko et al., 2002), the processing of non-motor sequences, including musical sequences (Maess, Koelsch, Gunter, & Friederici, 2001), and sequences of phonological material in working memory (Smith & Jonides, 1999). A recent study suggests that the manipulation of sequential information engages posterior Broca's area, independent of the type of information that is manipulated (Gelfand & Bookheimer, 2003). More generally, evidence suggests a close link between working memory and sequence learning and processing. These two functions share much of their circuitry: like sequence learning and processing, working memory involves Broca's area, SMA, and other pre-motor regions (Ivry and Fiez, 2000, Smith and Jonides, 1999), as well as regions in other structures (see below). It has been suggested that Broca's area, and perhaps VL-PFC more generally, may subserve particular functions that are important for working memory, including the selection and comparison of maintained information (Petrides, 1996, Petrides et al., 1995), or the maintenance of information over a delay (D'Esposito et al., 1998, Smith and Jonides, 1997). Similarly, it has been argued that the role of this region in working memory is to recall or select and maintain information that is actually stored in temporal and temporo-parietal regions (Cowan, 1999, Ruchkin et al., in press). These functions appear to be closely related to the role of this region in the selection of declarative knowledge (see above). The functions also appear to be related to the view that frontal cortex underlies the inhibition of, excitation of, and switching between (and possibly learning of) sets, programs and rules (Knight and Grabowecky, 2000, Shimamura, 1995, Wise et al., 1996), and the inhibitory and excitatory control of posterior brain regions (Knight & Grabowecky, 2000; also see Passingham, 1993). It has additionally been suggested that these attentional set shifting and sequence coordination roles may depend upon the timing functions of frontal cortex (Knight and Grabowecky, 2000, Meck and Benson, 2002, Wise et al., 1996). Indeed, Broca's area also appears to play an important role in timing and rhythm (Fiez et al., 1995, Schubotz and von Cramon, 2001, Szelag et al., 1997).

Within Broca's area, BA 44 plays an especially important role in a number of the functions described above. This region is also implicated in the observation of motor skills, and in the mental imagery of motion (Binkofski et al., 2000, Rizzolatti and Arbib, 1998, Rizzolatti et al., 2000). Intriguingly, F5 – the monkey homologue of BA 44 – contains “mirror neurons”, which fire not only at the execution of particular learned sequential motor skills (i.e. each neuron discharges with specific goal-oriented movement sequences), but also during the observation of (and even the sound of) the same action sequences (Kohler et al., 2002, Rizzolatti et al., 2001). Evidence suggests that human BA 44 also has this mirror function (Rizzolatti et al., 2001). It has been suggested that in humans such neurons may play a role in language, in particular in language comprehension (Rizzolatti & Arbib, 1998).

Portions of parietal cortex also play an important role in the procedural system. Anatomical studies of macaques show that parietal cortex projects heavily to, and reciprocally receives projections from, frontal cortex, with specific parietal regions connecting to specific frontal regions (Petrides and Pandya, 1984, Wise et al., 1997). Macaque area F5, which corresponds to the BA 44 area in humans (see above), receives strong input from a number of parietal regions (Matelli et al., 1986, Petrides and Pandya, 1984), especially macaque area AIP (Gallese et al., 2001, Matelli et al., 1994) – which is a probable homologue of human anterior intraparietal sulcus (Culham & Kanwisher, 2001) – and macaque area PF (also referred to as area 7b) (Matelli et al., 1986, Petrides and Pandya, 1984, Rizzolatti et al., 1998) – which is likely homologous to the human anterior inferior parietal lobule (supramarginal gyrus; BA 40) (Nishitani, Uutela, Shibasaki, & Hari, 1999), or possibly to part of the superior parietal lobule (BA 7) (Culham and Kanwisher, 2001, Milner, 1996).

In monkeys, AIP neurons discharge during hand movements, with the majority of neurons preferring specific types of hand grips (Sakata, Taira, Mine, & Murata, 1992). AIP neurons also contains neurons that are tuned to the specific shapes to be grasped (Sakata & Taira, 1994). Similarly, AIP's human homologue, the anterior intraparietal sulcus, is activated during visually-guided grasping (Binkofski et al., 1998, Faillenot et al., 1997) and reaching (Culham & Kanwisher, 2001), by the physical manipulation of objects (Binkofski et al., 1999), by mental rotation (Harris et al., 2000, Podzebenko et al., 2002), by the observation of hand movements made by others (Iacoboni et al., 1999), and by passively looking at manipulable objects, namely tools (Chao & Martin, 2000).

Monkey area PF is connected not only with frontal area F5, but also with parietal area AIP (Nishitani et al., 1999). Area PF is related to hand manipulation and eye movements, and may code the orientation of body parts (Nishitani et al., 1999). Like frontal area F5, area PF contains mirror neurons (Gallese et al., 2001, Rizzolatti et al., 2001). Human superior parietal lobule (BA 7), one possible homologue of PF, is strongly related to attention (Perry & Zeki, 2000). Human inferior parietal lobule, and the supramarginal gyrus (BA 40) in particular, another likely homologue of PF, has been implicated in a number of functions, including attention (Perry & Zeki, 2000), mental rotation (Harris et al., 2000, Podzebenko et al., 2002), and the execution and recognition of motor skills (Heilman et al., 1997). According to one view, inferior parietal regions may serve as a repository of stored knowledge of motor skills, including information of stored sequences (Heilman et al., 1997). This region has also been strongly implicated in phonological processing, including in working memory tasks (Ivry & Fiez, 2000).

Intriguingly, circumscribed portions of the temporal lobes also appear to play a role in the procedural memory system. In macaques neurons that respond to the observation of movement, though not to movement itself (that is, they are not mirror neurons), are found in the anterior superior temporal sulcus – a region which is connected to both frontal area F5 and parietal area PF (Rizzolatti et al., 2001). In humans, superior temporal regions in more posterior areas, including the superior temporal sulcus, have been implicated in the storage of information about motion, in contrast to more ventral temporal regions, which appear to underlie the storage of information about visual form (Martin et al., 2000).

The cerebellum has traditionally been implicated in the coordination of skilled movement and in the control of balance, as well as in motor learning (Ivry & Fiez, 2000). More recent evidence suggests that portions of the cerebellum subserve procedural memory, in particular in motor sequencing (Desmond and Fiez, 1998, Eichenbaum and Cohen, 2001, Hikosaka et al., 2000, Ivry and Fiez, 2000, Mostofsky et al., 2000, Squire and Knowlton, 2000). Some evidence suggests that the cerebellum may be involved in the modification of performance of learned sequences, rather than in the learning of those sequences (Seidler et al., 2002). Within the cerebellum, the dentate nucleus (Hikosaka et al., 2000) as well as portions of the cerebellar hemispheres and the vermis (Desmond & Fiez, 1998) play important roles in learning procedures, especially of motor sequences. The cerebellar hemispheres and vermis, especially regions at least partly overlapping those that also underlie sequence learning, have also been implicated in verbal working memory and in the retrieval or search of information from declarative memory (Desmond & Fiez, 1998). The cerebellum has additionally been implicated in imaged hand movements and in mental rotation (Ivry and Fiez, 2000, Podzebenko et al., 2002). The cerebellum has important timing functions, and seems to be involved in mental coordination and the control of attention, and in error detection and error-based learning (Doya, 2000, Ivry and Fiez, 2000). Studies of macaques have shown that, analogously to the basal ganglia, the cerebellum projects via the thalamus to frontal cortex, with each cerebellar region projecting (via the thalamus) to particular frontal regions. Intriguingly, in macaques the dentate nucleus projects via the thalamus to ventral pre-motor cortex (Middleton & Strick, 2000a), suggesting that in humans it might project to BA 44 in Broca's area.

The procedural system, and parietal cortex in particular, is closely related to the “dorsal” stream system (Goodale and Milner, 1992, Ungerleider and Mishkin, 1982). This system is rooted in posterior parietal structures, and the frontal pre-motor regions to which they are heavily connected. The system underlies the transformation of visual information into an egocentric framework that enables the execution of motor programs, such as grasping and otherwise manipulating an object. It has been argued that the main function of this system is the analysis of visual input for visually-guided motor behavior.

The declarative and procedural memory systems interact in a number of ways. In sum, together the systems form a dynamically interacting network which yields both cooperative and competitive learning and processing, such that memory function may be optimized (Poldrack & Packard, 2003).

First, brain structures which underlie procedural memory also perform context-dependent selection and maintenance (in working memory) of knowledge stored in declarative memory. Note that it is only a terminological issue as to whether we consider these structures to be part of the procedural system which plays a role in declarative memory, or vice versa, or simply (and most reasonably) brain structures that play particular roles in both systems.

Second, although there appear to be striking separations of function among the different brain areas involved in the two brain systems, it does not appear to be the case that all parts of each lobe subserve only one or the other system. In particular, we have seen that superior aspects of the temporal lobe may play some function in the procedural system, perhaps as a storage repository of procedural knowledge, and that the same or nearby areas of VL-PFC play related roles in declarative and procedural memory.

Third, the two systems can complement each other in acquiring the same or analogous knowledge, including knowledge of sequences. As was initially shown in patient H.M., the declarative memory system need not be intact for the procedural memory system to learn (Corkin, 1984, Eichenbaum and Cohen, 2001, Squire and Knowlton, 2000). However, when both systems are undamaged they can complement each other. Thus, in motor sequence learning in humans, both systems can be used cooperatively to learn the task, optimizing learning in some cases (Willingham, 1998). When the declarative memory system is able to acquire knowledge, it may do so initially, thanks to its rapid learning abilities, while the procedural system gradually learns the same or analogous knowledge (Packard and McGaugh, 1996, Poldrack and Packard, 2003). Note that if a given sequence that is normally learned and processed by the procedural system is memorized in declarative memory, its structure will likely be constrained by the rules governing the sequence in procedural memory. Interestingly, the time-course of this shift from declarative to procedural memory can be modulated pharmacologically (Packard, 1999).

Fourth, animal and human studies suggest that the two systems can also interact competitively (for reviews, see Packard and Knowlton, 2002, Poldrack and Packard, 2003). This leads to what one might call a “see-saw effect”, such that a dysfunction of one system leads to enhanced learning in the other, or that learning in one system depresses functionality of the other. Animal studies show that damage to medial-temporal lobe structures, including the hippocampus, can enhance basal-ganglia-based procedural learning (McDonald and White, 1993, Packard et al., 1989, Schroeder et al., 2002). Conversely, damage to the neostriatum in the basal ganglia can facilitate learning in declarative memory (Mitchell & Hall, 1988). A similar pattern has been found in human lesion (Halbig et al., 2002) and neuroimaging studies (Dagher et al., 2001, Jenkins et al., 1994, Poldrack and Packard, 2003, Poldrack et al., 2001, Poldrack et al., 1999).

The see-saw effect may be explained by a number of factors. In rodents there are direct anatomical projections from the medial temporal lobe (entorhinal cortex) to the dorsal striatum (Sorensen & Witter, 1983). Stimulation of both entorhinal and hippocampal neurons leads to mainly inhibitory responses in both the dorsal and ventral striatum (Finch, 1996, Finch et al., 1995). Conversely, stimulation of the caudate (in cats) reduces the occurrence of hippocampal spikes (La Grutta and Sabatino, 1988, Sabatino et al., 1985). Several studies have suggested that lesions to the hippocampus can result in increased dopamine transmission in the portion of the striatum to which the hippocampus projects (Jaskiw et al., 1990, Lipska et al., 1992). Neurochemically, acetylcholine appears to play a role in the interaction (Packard & Knowlton, 2002), perhaps by enhancing function of the hippocampus (see above), which could in turn inhibit function of the striatum via the projections described above. Moreover, acetylcholine can also directly inhibit the function of the neostriatum (Calabresi, Centonze, Gubellini, Pisani, & Bernardi, 2000). Because acetylcholine function can be enhanced by estrogen, particularly in the hippocampus (see above), it is plausible that estrogen may also contribute to the see-saw effect.

A recent and quite elegant series of neuroimaging experiments of healthy adults nicely demonstrates interactions between the two brain memory systems (Poldrack et al., 2001, Poldrack et al., 1999). Procedural learning – probabilistic rule learning – was shown to yield not only activation in the caudate nucleus, but also deactivation in the medial temporal lobe. Moreover, across subjects, the degree of activity in the caudate nucleus correlated negatively with the degree of activity in the medial temporal lobe. That is, subjects with higher caudate activity had lower medial-temporal activity, and vice versa. This suggests that individuals vary with respect to their relative dependence on the two systems. Moreover, this relationship changed over the course of learning. During early training the medial temporal lobe structures were activated while the caudate was not, whereas as learning progressed, the medial temporal structures became deactivated, while caudate activation increased. These experiments suggest some sort of competitive interaction between the two systems. Moreover, they strengthen the view that early in learning declarative memory can play a particularly important role compared to procedural learning, and that over time this balance shifts to the opposite direction. Thus, with increased dependence on procedural memory for a given function, there may be a decreased dependence on declarative memory, even if that system played a role initially in the same function. As we will see below, evidence suggests a similar pattern in language learning.

We have seen above that there are a number of striking commonalities between the functional characteristics of grammar/lexicon on the one hand, and of declarative/ procedural (DP) memory and their underlying brain structures on the other. These commonalities lead to the basic claim of the DP model: the brain systems which subserve declarative and procedural memory play analogous roles in language as in their non-language functions. So, the DP model predicts common or related computational, processing, anatomic, physiological and biochemical substrates for the language and non-language functions.

According to the DP model, the brain system underlying declarative memory also underlies the mental lexicon. This system subserves the acquisition, representation and use not only of knowledge about facts and events, but also about words. It stores all arbitrary, idiosyncratic word-specific knowledge, including word meanings, word sounds, and abstract representations such as word category. It includes, among other things, representations of simple (non-derivable) words such as cat, bound morphemes such as the past-tense suffixed -ed, “irregular” morphological forms, verb complements, and idioms. It can also contain stored complex forms and abstract structures that are “regular” in that they can also be composed or derived by the grammatical/procedural system. As with idiosyncratic knowledge, the likelihood of these “regular” representations being memorized should increase with item-related factors such as their frequency, and subject-related factors such as the individual's lexical/declarative memory abilities. The system supports a superpositional associative memory, which allows for generalizations across representations. For example, the memorization of phonologically similar stem-irregular past tense pairs (e.g. spring–sprang, sing–sang) may allow for memory-based generalization to new irregularizations, either from real words (bring–brang) or from novel ones (spling–splang). This ability to generalize could underlie some degree of productivity within the memory system.

The brain structures that subserve declarative memory play analogous roles in lexical memory. Thus, medial temporal lobe structures underlie the encoding, consolidation and access or retrieval of new memories, which eventually rely instead on neocortical regions, especially in temporal and temporo-parietal areas. Inferior and ventral temporal regions are particularly important for representing non-linguistic conceptual knowledge and word meanings. They may also contain abstract lexical representations (Damasio et al., 1996). Superior temporal cortex may be particularly important for storing phonological representations, and perhaps other grammatical (syntactic, morphological) representations. Thus, this region may be related to both the procedural and declarative memory systems. Other brain structures, particularly those related to the procedural memory system, also play roles in declarative memory. These are described below. Acetylcholine and estrogen have important functions in lexical/declarative memory, likely in the learning and/or retrieval of new lexical knowledge.

The brain system underlying procedural memory subserves the mental grammar. This system underlies the learning of new, and the computation of already-learned, rule-based procedures that govern the regularities of language-particularly those procedures related to combining items into complex structures that have precedence (sequential) and hierarchical relations. Thus, the system is hypothesized to have an important role in rule-governed structure building; that is, in the sequential and hierarchical combination – “merging” (Chomsky, 1995), or concatenation – of stored forms and abstract representations into complex structures. Procedural memory is assumed to play a role in all sub-domains of grammar which depend on these functions, including syntax; inflectional and derivational morphology – at least for default “regulars” (Pinker, 1999, Ullman, 2001a, Ullman, 2001c), but also for irregulars that appear to be affixed (Ullman, Hartshorne, Estabrooke, Brovetto, & Walenski, submitted); aspects of phonology (the combination of sounds); and possibly non-lexical (compositional) semantics (the interpretive, i.e. semantic, aspects of the composition of words into complex structures). Moreover, the computational nature of the system is likely to be similar across grammatical sub-domains – although this does not preclude the possibility that these sub-domains maintain a certain degree of independence (see discussion below).

The system is a network composed of several brain structures. These are functionally related, highly inter-connected, and dynamically interactive: the basal ganglia, especially the caudate nucleus; frontal cortex, in particular Broca's area and pre-motor regions (including SMA and pre-SMA); parietal cortex, particularly the supramarginal gyrus (BA 40) and possibly the superior parietal lobule (BA 7); superior temporal cortex, probably in close relation with the declarative memory system; and the cerebellum, including the cerebellar hemispheres, the vermis, and the dentate nucleus.

The language-related functions of each of these structures is expected to be highly related to its non-language functions. Thus, the basal ganglia are posited to play a role in one or more aspects of the real-time selection and maintenance in working memory of, and switching between, sequentially and hierarchically structured elements in complex linguistic representations, and in the learning of rules over those representations. Grammar is learned and processed by one or more channels that run throughout the basal ganglia to the thalamus and thence to frontal cortex. These channels are parallel to and largely functionally segregated from other channels that undergo analogous computations but subserve other domains. The channel(s) subserving grammar might also subserve other domains, such as non-linguistic sequence learning; that is, they may be at least somewhat domain-independent. Alternatively, there may be one or more channels dedicated to grammar, and perhaps distinct (sub)channels for distinct grammatical sub-domains (e.g. syntax, morphology). Such channels and their frontal outputs may be considered domain-specific “modules” dedicated to grammar or its sub-domains. Though these hypothesized grammatical (sub)channels are domain-specific in that they underlie only grammatical processing, they are part of a domain-general procedural system, in which the same or analogous computations are performed on parallel and largely functionally segregated channels subserving other domains. This is a somewhat different view of modularity than is traditionally discussed in the study of language (Fodor, 1983).

The frontal cortical areas to which the basal ganglia project – in particular Broca's area, SMA and pre-SMA – also underlie aspects of grammar. Broca's area or portions thereof – especially BA 44 – may play an especially important role. Based on the functions of Broca's area in non-linguistic domains, this region is expected to subserve aspects of the selection and maintenance in working memory of elements in complex linguistic structures, and the learning and processing of rule-governed sequential and hierarchical patterns over those structures. These functions are, not surprisingly, quite related to the hypothesized functions of the basal ganglia in language, though Broca's area and the basal ganglia likely differ at least somewhat in their specific functions (Ullman, 2003; Ullman and Pierpont, in press).

Although the other structures that constitute the procedural system network are also expected to subserve language, their functional roles are perhaps less clear than those of the structures discussed above. Following evidence from their non-linguistic functions, the supramarginal gyrus and/or the superior parietal lobule may each play a role in attentional selection, which could be related to the selection functions described above. Parietal cortex may also play a role not only in sensori-motor transformations, but also in transforming structured representations stored in superior temporal regions to the dynamic representations created by Broca's area. The cerebellum is expected to be involved in the search of lexical items, and possibly in the error-based learning of the rules that underlie the regularities of complex structures.

The lexical/declarative memory system and the grammatical/procedural system are hypothesized to interact in several ways. First, the procedural system is hypothesized to build complex structures, and learn rule-governed patterns over those structures, by selecting lexical items from declarative memory, and maintaining and structuring those items together in working memory. Second, superior aspects of the temporal lobe may play a role in the storage of knowledge about procedural memory related relations of structured representations. Third, the same or similar types of knowledge can in at least some cases be acquired by both systems. The rapid lexical/declarative storage of sequences of lexical forms should provide a database from which grammatical rules can gradually and implicitly be abstracted by the procedural memory system. Moreover, in some cases explicit knowledge of the rules themselves may help guide processing, perhaps enhancing the procedural rule acquisition. Fourth, the two systems interact competitively in a number of ways. Access to a stored representation which has similar mappings to one which could be composed compositionally by the procedural system (e.g. an irregular vs. a regular past-tense form of the same verb) would block completion of the latter computation. Damage to the declarative system is expected to lead to enhanced learning and processing by the procedural system, and vice versa. Moreover, learning in one system may depress functionality of the other. It is possible that, at least under certain circumstances, enhancing acetylcholine or estrogen function in medial temporal lobe structures may result not only in improved lexical/declarative memory function, but also in suppressed grammatical/procedural function.

The DP model is motivated by a set of commonalities between language functions on the one hand, and the functions of the memory brain systems on the other. However, the commonalities do not suggest, and indeed it is not the claim, that there are isomorphic (one-to-one) relations between lexicon and declarative memory, or between grammar and procedural memory. That is, it is not expected that all parts of the brain system underlying procedural memory subserve all aspects of the mental grammar, and similarly for declarative memory and the mental lexicon. First, there may be parts of each system that subserve non-language functions but which play no role in language, or a minimal role, or perhaps a role only in special circumstances (e.g. after breakdown). Indeed, this seems likely. For example, the declarative memorization of visual images clearly depends in part on cortical regions which may be specialized for and perhaps dedicated to visual processing, and thus are unlikely to be involved in the memorization of phonological word forms. Second and conversely, the DP model does not claim that all aspects of language depend upon the two brain memory systems. Other neural structures and other cognitive or computational components, perhaps even dedicated to language, may play an important role in the two language capacities. Third, as we have seen above, structures with topographically segregated sub-regions (i.e. the basal ganglia, cerebellum, and possibly frontal cortex) may contain distinct sub-regions or circuits that subserve language and non-language functions (see above). On this view, the procedural memory system may be domain-independent in that it subserves many functions, but is also domain-specific in that each function is subserved by parallel and functionally segregated sub-components or modules. Fourth, as discussed above, even in the context of such topographic organization, there may be anatomical and functional specialization of sub-regions, such as in Broca's area.

The predictions follow from the model described above. Most fundamentally, language and non-language functions that are posited to depend on the same brain systems should pattern together, showing similar computational, anatomic, physiological, biochemical and other characteristics. Moreover, this should apply not only to normal functioning, but also to the breakdown of these brain systems, and to recovery from this breakdown.

The DP model is proposed in the same spirit as, and is similar in a number of respects to, several other models and proposals (Dominey, 1997, Dominey et al., 2003, Friederici, 1990, Greenfield, 1991, Lieberman et al., 1992, Rizzolatti and Arbib, 1998). These focus on the relation between grammar on the one hand, and implicit procedural memory, motor sequencing and hierarchical structure on the other. Some of these proposals are quite well-specified in certain respects. For example, Dominey and his colleagues have developed a computational model of the type of sequencing that may underlie both grammar and non-linguistic sequencing (Dominey et al., 2003). These models complement the DP model, specifying some aspects to a greater depth than the DP model, which in turn provides further specification in other dimensions, in particular the anatomical, physiological and biochemical substrates, and the functional roles in language played by those substrates.

The DP model is also largely, though not completely, compatible with many “dual-system” (“dual-mechanism”) or multiple-system linguistic and psycholinguistic models of language (Bever et al., 1998, Chomsky, 1995, Fodor, 1983, Fodor et al., 1974, Frazier and Fodor, 1978, Pinker, 1994, Pinker, 1999). On these views, language is subserved by separable components. At the very least, the mental lexicon is assumed to be distinct from a “computational” mental grammar, which moreover is often claimed to be composed of several components (Chomsky, 1995). These theories also claim or assume that at least the grammatical component or components are domain-specific. As can be seen from the discussion above, all of these claims are compatible with the DP model. Thus, the DP model can be thought of as a neurocognitive model of aspects of these linguistic and psycholinguistic proposals. The neurocognitive model provides further specification in certain respects, particularly of the underlying brain structures and their functions, whereas the linguistic and psycholinguistic models provide much greater specification at the level of representation, computation and processing.

In contrast, the DP model is at least partially inconsistent with certain claims about the domain specificity of the neural basis of grammar. Thus, the DP model is not consistent with the particular claim (Grodzinsky, 2000) that Broca's area is dedicated to language and performs a different set of linguistic computations than are claimed by the DP model. The DP model is also not consistent with the claims made by certain connectionist models, in particular connectionist models that deny grammatical composition (Bates and MacWhinney, 1989, Joanisse and Seidenberg, 1999, MacDonald et al., 1994, Rumelhart and McClelland, 1986). These models also do not predict empirical associations among grammatical domains and procedural memory, or dissociations between these and lexical and declarative memory.

Section snippets

Empirical evidence

Here I examine neurocognitive evidence pertaining to the claims and predictions of the DP model in native (first) language. I focus on the relation between brain and cognition. For a discussion of purely behavioral (psycholinguistic) evidence from cognitively unimpaired individuals, see Pinker (1999), Ullman (2001a), or Pinker and Ullman (2002), among others. Three broad types of evidence are examined here. First, I provide brief overviews of hemodynamic (PET, fMRI) and electrophysiological

Conclusion

According to the DP model, the brain systems underlying two well-studied memory capacities, declarative and procedural memory, also subserve aspects of the mental lexicon and the mental grammar. Both brain systems play similar functional roles across language and non-language domains, which depend on common anatomical, physiological, and biochemical substrates. Evidence from neuroimaging (fMRI, PET, ERPs) and from developmental and adult-onset disorders supports this claim. Moreover, I have

Acknowledgements

Support was provided to MTU by a McDonnell-Pew grant in Cognitive Neuroscience, a research grant from the National Alliance For Autism Research, NSF SBR-9905273, NIH MH58189, and Army DAMD-17-93-V-3018/3019/3020 and DAMD-17-99-2-9007. I would like to thank all the members of the Brain and Language Laboratory, especially Claudia Bonin, John Drury, Ivy Estabrooke, Joshua Hartshorne, Matthew Moffa, Rene Pierpont, Karsten Steinhauer, and Matthew Walenski, for lively discussions and for helpful

References (318)

  • M. D'Esposito et al.

    Functional MRI studies of spatial and nonspatial working memory

    Brain Research: Cognitive Brain Research

    (1998)
  • P.F. Dominey

    An anatomically structured sensory-motor sequence learning system displays some general linguistic capacities

    Brain and Language

    (1997)
  • K. Doya

    Complementary roles of basal ganglia and cerebellum in learning and motor control

    Current Opinion in Neurology

    (2000)
  • J. Doyon et al.

    Role of the striatum, cerebellum, and frontal lobes in the learning of a visuomotor sequence

    Brain and Cognition

    (1997)
  • D. Finch et al.

    Neurophysiology and neuropharmacology of projections from entorhinal cortex to striatum in the rat

    Brain Research

    (1995)
  • L. Frazier et al.

    The sausage machine: a new two-stage parsing model

    Cognition

    (1978)
  • A. Friederici

    Towards a neural basis of auditory sentence processing

    Trends in Cognitive Sciences

    (2002)
  • A.D. Friederici et al.

    Verb-argument structure processing: the role of verb-specific and argument-specific information

    Journal of Memory and Language

    (2000)
  • A.D. Friederici et al.

    First-pass versus second-pass parsing processes in a Wernicke's and a Broca's aphasic: electrophysiological evidence for a double dissociation

    Brain and Language

    (1998)
  • A.D. Friederici et al.

    Event-related brain potentials during natural speech processing: effects of semantic, morphological and syntactic violations

    Cognitive Brain Research

    (1993)
  • J.D.E. Gabrieli et al.

    The impaired learning of semantic knowledge following bilateral medial temporal-lobe resection

    Brain and Cognition

    (1988)
  • J.R. Gelfand et al.

    Dissociating neural mechanisms of temporal sequencing and processing phonemes

    Neuron

    (2003)
  • P. Georgiewa et al.

    Phonological processing in dyslexic children: a study combining functional imaging and event related potentials

    Neuroscience Letters

    (2002)
  • J.W. Aldridge et al.

    Coding of serial order by neostriatal neurons: a “natural action” approach to movement sequence

    Journal of Neuroscience

    (1998)
  • G.E. Alexander et al.

    Basal ganglia-thalamocortical circuits: parallel substrates for motor oculomotor ‘prefrontal’ and ‘limbic’ functions

  • G.E. Alexander et al.

    Parallel organization of functionally segregated circuits linking basal ganglia and cortex

    Annual Review of Neuroscience

    (1986)
  • M.P. Alexander

    Aphasia: clinical and anatomic aspects

  • P. Alvarez et al.

    Memory consolidation and the medial temporal lobe: a simple network model

    Proceedings of the National Academy of Sciences USA

    (1994)
  • K. Amunts et al.

    Broca's region revisited: cytoarchitecture and intersubject variability

    Journal of Comparative Neurology

    (1999)
  • S.R. Anderson
    (1992)
  • S.E. Arnold et al.

    The topographical and neuroanatomical distribution of neurofibrillary tangles and neuritic plaques in the cerebral cortex of patients with Alzheimer's disease

    Cerebral Cortex

    (1991)
  • P. Bach et al.

    Conceptual and structural relations in action comprehension

  • A. Bailey et al.

    Autism: towards an integration of clinical, genetic, neuropsychological and neurobiological perspectives

    Journal of Child Psychology and Psychiatry

    (1996)
  • C. Ballard

    Advances in the treatment of Alzheimer's disease: benefits of dual cholinesterase inhibition

    European Journal of Neurology

    (2002)
  • R.A. Barkley et al.

    Sense of time in children with ADHD: effects of duration, distraction, and stimulant medication

    Journal of the International Neuropsychological Society

    (1997)
  • G. Bartolucci et al.

    Cross-sectional studies of grammatical morphemes in autistic and mentally retarded children

    Journal of Autism and Developmental Disorders

    (1980)
  • D.S. Baskin et al.

    Brain choline acetyltransferase and mental function in Alzheimer disease

    Archives of Neurology

    (1999)
  • E. Bates et al.

    Functionalism and the competition model

  • M.L. Bauman

    Motor dysfunction in autism

  • W.W. Beatty et al.

    Preserved cognitive skills in dementia of the Alzheimer type

    Archives of Neurology

    (1994)
  • L. Bennetto et al.

    Intact and impaired memory functions in autism

    Child Development

    (1996)
  • P.C. Berquin et al.

    Cerebellum in attention-deficit hyperactivity disorder: a morphometric MRI study

    Neurology

    (1998)
  • T.G. Bever et al.

    The emperor's psycholinguistics

    Journal of Psycholinguistic Research

    (1998)
  • F. Binkofski et al.

    Broca's region subserves imagery of motion: a combined cytoarchitectonic and fMRI study

    Human Brain Mapping

    (2000)
  • F. Binkofski et al.

    A parieto-premotor network for object manipulation: evidence from neuroimaging

    Experimental Brain Research

    (1999)
  • F. Binkofski et al.

    Human anterior intraparietal area subserves prehension: a combined lesion and functional MRI activation study

    Neurology

    (1998)
  • D.V. Bishop

    Motor immaturity and specific speech and language impairment: evidence for a common genetic basis

    American Journal of Medical Genetics

    (2002)
  • H. Boecker et al.

    Role of the human rostral supplementary motor area and the basal ganglia in motor sequence control: investigations with H2 15O PET

    Journal of Neurophysiology

    (1998)
  • S. Bozeat et al.

    Non-verbal impairment in semantic dementia

    Neuropsychologia

    (2000)
  • Cited by (1193)

    • Unraveling sequence learning in obsessive–compulsive disorder

      2023, Current Opinion in Behavioral Sciences
    View all citing articles on Scopus
    View full text