A neural dissociation within language: evidence that the mental dictionary is part of declarative memory, and that grammatical rules are processed by the procedural system.

INTRODUCTION

Two capacities give human language its vast expressive power. One is a "mental lexicon" containing thousands of words, each a memorized, arbitrary sound-meaning pairing. The other is a "mental grammar" of generative rules that combine words into an infinite number of larger words, phrases, and sentences (Chomsky, 1965; De Saussure, 1959; Pinker, 1994). Previous evidence has suggested links between grammatical processing and left frontal cortex, including Broca's area, and links between lexical memory and left temporal and parietal cortex (Geschwind, 1965; Goodglass, 1993; Wernicke, 1874). The speech of aphasics with frontal (anterior) lesions is often "agrammatic," with a breakdown of sentence structure and the omission or misuse of grammatical morphemes, while access to content words such as verbs and nouns is less dramatically impaired (Damasio, 1992; Goodglass, 1993). Electrophysiological and functional neuroimaging studies have also implicated frontal regions in grammatical processing (Kluender & Kutas, 1993; Neville et al., 1991; Stromswold et al., 1996). Aphasics with temporal or parietal (posterior) lesions often speak in relatively intact sentence structures, but have severely impaired access to content words (Damasio, 1992; Goodglass, 1993). The dissociation, however, is imperfect and still controversial (Bates & Wulfeck, 1989; Goodglass, 1993). Most groups of aphasics tested have complex lesion patterns, making symptom-lesion correlations imperfect. Moreover, testing the dissociation has been problematic because tasks probing for grammar and for memory have differed in ways other than their use of the two capacities. In this study we confirm the dissociation using a simple language task in which the use of the two linguistic capacities is contrasted while other factors, such as complexity, meaning and task demands, are held constant.

We also relate the dissociation to larger principles of neural organization. Cutting across the division of the brain into systems such as language and visual perception, there is an orthogonal division into two major kinds of memory systems (Cohen & Squire, 1980; Mishkin et al., 1984; Damasio & Damasio, 1992; Squire et al., 1993). One is a declarative memory system underlying the learning and storage of information about facts and events. It is subserved by a medial temporal circuit connected largely with neocortical areas in the temporal and parietal lobes, with the medial temporal components consolidating memories that are eventually stored in neocortex (Cohen & Squire, 1980; Mishkin et al., 1984; Squire et al., 1993; Suzuki & Amaral, 1994). The other is a procedural memory system for the learning and processing of motor, perceptual, and cognitive skills. It is subserved by basal ganglia circuits connected largely with frontal cortex (Cohen & Squire, 1980; Gabrieli et al., 1993; Heindel et al., 1989; Mishkin et al., 1984; Saint-Cyr et al., 1988; Squire et al., 1993). These parallel basal ganglia circuits are functionally segregated; each receives projections from particular ipsilateral cortical and sub-cortical areas, and projects via the thalamus to a particular ipsilateral frontal lobe area. Thus a "motor circuit" projects to frontal motor areas, while other circuits project to other frontal areas. The different circuits have similar synaptic organizations within the basal ganglia (Middleton & Strick, 1994; Alexander et al., 1990; Young & Penney, 1993).

Given that word forms are like facts in being arbitrary, and in possibly having storage sites in temporal and parietal regions, the temporal-parietal/medial-temporal declarative memory system may subserve words as well as facts and events. Given that rules are like skills in requiring the coordination of procedures in real time, and in possibly having neural loci in frontal regions, the frontal/basal-ganglia procedural system may process grammatical rules as well as motor and perceptual skills. Basal ganglia circuitry may project to Broca's area (Hoover & Strick, 1993; Preuss, 1995), raising the possibility that portions of the basal ganglia subserve grammatical processing (Lieberman et al., 1992), performing operations comparable to those done for motor programming. Damasio and Damasio (1992), noting the anatomical interconnections between cortical language areas and the basal ganglia, predict that "the basal ganglia serve to assemble the components of complex motions into a smooth whole, and it seems reasonable that they might perform an analogous function in assembling word-forms into sentences."

We tested this memory/rule dissociation by devising a task based on a simple linguistic system in which reliance on grammar and lexicon differs, while other factors are held constant. Regular (look-looked) and irregular (dig-dug) past tense forms of verbs are well-matched in complexity (one word), syntax (tensed), and meaning (past). But regular verbs are predictable in form (verb stem + -ed), and new ones are constantly being added (faxed, moshed), whereas irregular verbs are unpredictable (compare sing-sang, fling-flung, bring-brought), and constitute a fixed list. A simple theory is that irregular forms are memorized, and regular forms are generated by a rule. The rule comprises two operations: copying the stem, and adding a suffix. Regulars and irregulars interact as follows: Retrieval of an irregular blocks the rule (dug preempts digged); when an irregular is not successfully retrieved, the rule may be applied, resulting in "overregularization" errors such as digged.

Alternative theories have been proposed. In one, regulars and irregulars are both computed by rules (Chomsky & Halle, 1968; Halle & Mohanon, 1985), with memory compressed to the minimum information necessary. In another, regulars and irregulars are both computed by a connectionist associative memory, with all rules eliminated (Rumelhart & McClelland, 1986; MacWhinney & Leinbach, 1991). Recent evidence from the structure, processing, and acquisition of several languages has bolstered the memory/rule theory (Marcus et al., 1992; Marcus et al., 1995; Pinker, 1991; Pinker & Prince, 1988; Prasada & Pinker, …