The declarative/procedural model ...
37 0090-6905/01/0100-0037$19.50/0 �� 2001 Plenum Publishing Corporation Journal of Psycholinguistic Research, Vol. 30, No. 1, 2001 The Declarative/Procedural Model of Lexicon and Grammar Michael T. Ullman1 Our use of language depends upon two capacities: a mental lexicon of memorized words and a mental grammar of rules that underlie the sequential and hierarchical composition of lexical forms into predictably structured larger words, phrases, and sentences. The declarative/procedural model posits that the lexicon/grammar distinction in language is tied to the distinction between two well- studied brain memory systems. On this view, the memorization and use of at least simple words (those with noncompositional, that is, arbitrary form-meaning pairings) depends upon an associa- tive memory of distributed representations that is subserved by temporal-lobe circuits previously implicated in the learning and use of fact and event knowledge. This ���declarative memory��� system appears to be specialized for learning arbitrarily related information (i.e., for associative binding). In contrast, the acquisition and use of grammatical rules that underlie symbol manipulation is sub- served by frontal/basal-ganglia circuits previously implicated in the implicit (nonconscious) learn- ing and expression of motor and cognitive ���skills��� and ���habits��� (e.g., from simple motor acts to skilled game playing). This ���procedural��� system may be specialized for computing sequences. This novel view of lexicon and grammar offers an alternative to the two main competing theoretical frameworks. It shares the perspective of traditional dual-mechanism theories in positing that the mental lexicon and a symbol-manipulating mental grammar are subserved by distinct computa- tional components that may be linked to distinct brain structures. However, it diverges from these theories where they assume components dedicated to each of the two language capacities (that is, domain-specific) and in their common assumption that lexical memory is a rote list of items. Conversely, while it shares with single-mechanism theories the perspective that the two capacities are subserved by domain-independent computational mechanisms, it diverges from them where they link both capacities to a single associative memory system with broad anatomic distribution. The declarative/procedural model, but neither traditional dual- nor single-mechanism models, predicts double dissociations between lexicon and grammar, with associations among associative memory properties, memorized words and facts, and temporal-lobe structures, and among symbol-manipu- lation properties, grammatical rule products, motor skills, and frontal/basal-ganglia structures. In order to contrast lexicon and grammar while holding other factors constant, we have focused our investigations of the declarative/procedural model on morphologically complex word forms. I thank Roumyana Izvorski for valuable discussions and insights. Support was provided by a McDonnell���Pew grant in Cognitive Neuroscience, NSF SBR-9905273, NIH MH58189, and Army DAMD17-93-V-3018/3019/3020 and DAMD 17-99-2-9007. 1 Departments of Neuroscience and Linguistics, Georgetown University, Research Building, 3900 Reservoir Road, NW, Washington, DC 20007. email: email@example.com.
38 Ullman Morphological transformations that are (largely) unproductive (e.g., in go���went, solemn���solem- nity) are hypothesized to depend upon declarative memory. These have been contrasted with mor- phological transformations that are fully productive (e.g., in walk���walked, happy���happiness), whose computation is posited to be solely dependent upon grammatical rules subserved by the pro- cedural system. Here evidence is presented from studies that use a range of psycholinguistic and neurolinguistic approaches with children and adults. It is argued that converging evidence from these studies supports the declarative/procedural model of lexicon and grammar. KEY WORDS: language lexicon grammar declarative memory procedural memory mor- phology regular irregular frontal lobe temporal lobe basal ganglia. Language may be defined as a relation between form and meaning. The pre- dictability of this relation varies across linguistic forms. The relations of many forms, and of word ���roots��� in particular, are entirely idiosyncratic. That is, the meaning of these words cannot be predicted, or derived, from their forms, or vice versa. Thus the relation between the form cat and the mean- ing ���cat��� is entirely arbitrary, and so the pairing must be memorized. The set of memorized form���meaning pairings may be defined as the mental lexicon. In contrast, the meaning of other linguistic forms is entirely predictable (derivable), given both the memorized knowledge of the lexical forms from which they are composed, and the grammatical rules (including constraints), which specify their composition. The knowledge of these rules may be defined as the mental grammar. In addition, there are form-meaning pairings that are partially predictable. Although these form���meaning relations may be described by rules, the application of those rules is not entirely predictable, and, therefore, the contexts in which they apply must be memorized. Here a mental model of lexicon and grammar is proposed. The model addresses issues of modularity, computation, domain specificity, and neural localization. It is posited that two systems, with distinct computational, psy- chological, and neural bases, play parallel roles in the computation of form���meaning relations. One system is an associative memory of distrib- uted (but structured) representations, over which phonological and concep- tual-semantic mappings are learned, stored, and computed. The other system follows mental rules of grammar in building the sequential and hier- archical structure of complex forms. Although it is posited that this gram- matical system subserves syntactic as well as morphological (and possibly also phonological) computations, here we focus on morphology. It is hypothesized that the rule system subserves at least those mor- phological (morphophonological) transformations that are fully productive and that involve only sequencing operations (i.e., in affixation and com- pounding e.g., look -ed ��� looked, shirt sleeve ��� shirt-sleeve), whereas the memory system underlies at least those morphophonological transfor- mations that do not involve any such sequencing operations (e.g.,
sing���sang). For a given morphosyntactic configuration, both systems attempt to compute an appropriate morphologically complex form. If a form is found in memory (sang), the rule-based computation is inhibited other- wise a rule-product (looked) is successfully computed. Neither system���s role is limited to morphology or even to language. The memory system underlies the learning, representation, and processing of the entire mental lexicon as well as nonlinguistic knowledge about facts and events, such as what a bicycle is and what you ate for lunch yesterday. The rule-subserving system underlies the implicit (nonconscious) learning and expression not only of morphological rules, but also of syntactic (and possibly phonological) rules and of motor and cognitive skills and habits, such as how to ride a bicycle. The two systems have distinct neural correlates. The memory system is rooted in temporal lobe brain regions, whereas the rule system is rooted in left frontal/basal-ganglia structures. Thus the claim is that morphology itself is modular, and that each of the two principal components underlies the computation of (at least partially) different types of morphological trans- formations, but that each relies on distinct computational operations and dis- tinct brain structures, and that each component is domain-independent, both within language and, more generally, with respect to other cognitive functions. This paper is organized as follows. First, some background to different types of morphological transformations is provided. Second, the proposed mental model is described in detail. Third, the two main competing theo- retical frameworks are discussed. Fourth, all three competing perspectives are summarized, and competing predictions are made, in terms of the rele- vant theoretical issues: modularity, computation, domain-specificity/gener- ality, and neuroanatomical localization. Fifth, evidence is presented from psycholinguistic and neurolinguistic studies aimed at probing these theoretical issues and at teasing apart the competing theories. MORPHOLOGICAL TRANSFORMATIONS Within inflectional morphology, words in a given syntactic configura- tion are associated with one or more morphophonological transformations. Here a morphophonological transformation refers to a phonological mapping between word stems or roots (i.e., bases) and inflected forms for a given (mor- pho)syntactic configuration. Thus English verbs in the past tense can undergo -ed suffixation as well as a variety of other morphophonological transfor- mations (in bring���brought, sing���sang, keep���kept, etc.). Similarly, for derivational morphology, words in a given syntactic and semantic configu- ration may be associated with one or more morphophonological transfor- mations. Thus (at least) -ity and -ness suffixation underlie the creation of Declarative/Procedural Model 39
abstract nouns from adjectives in English. Morphophonological transforma- tions (which I will also refer to as morphological transformations or map- pings) can differ along a number of dimensions, across both inflectional and derivational morphology. Here I focus on three dimensions, each of which is relevant to the claims of the proposed model. First, overt morphophonological mappings are of (at least) two types. Whereas affixation involves the sequencing of morphophonological seg- ments, sequencing may not be evident in other types of mappings. Thus suppletion involves a completely arbitrary relation between base and mor- phologically complex form (e.g., go���went), whereas phonological stem changes involve modification of only particular segments (e.g., sing���sang). A given word can undergo none of these types of overt changes (e.g., hit���hit), or one or more of them, such as stem changes in addition to affix- ation (e.g., break���broken). Second, certain morphophonological transformations may have prece- dence over others in the same syntactic configuration, blocking those with lower precedence. Such precedence is a function of the fact that many mor- phophonological transformations apply only or primarily under specific con- ditions. Some apply only to a fixed list of words (e.g., go���went, bring���brought, teach���taught), whereas others appear to apply in particular phonological, semantic, and/or morphological conditions. If in a given syn- tactic (and semantic) configuration two or more morphological transforma- tions have overlapping conditions, and if one of these transformations always applies instead of the others in these overlapping conditions, then it can be said to have strict precedence over the others. For example, in English past tense, the various transformations underlying irregular forms (e.g., take���took, bring���brought, sing���sang, keep���kept) take precedence over -ed suffixation, thus precluding forms like taked, bringed, and keeped. Third, morphological transformations vary widely with respect to their productivity. Here, the level of productivity of a morphological transforma- tion refers to the extent to which it applies to new forms (new words and novel forms) that meet any particular set of conditions, within a given syntactic con- figuration. Unproductive transformations apply only to a fixed list of words (e.g., go���went) that is, they apply to no new forms. Fully productive trans- formations apply to every form which meets a necessary and sufficient set of conditions, apart from those forms to which a transformation of higher prece- dence has applied. For example, the application of the -ity suffix to -able-suf- fixed adjectives appears to be fully productive. Such a transformation may be called a local default, because it is defaulted to under particular (i.e., local) conditions within a broader syntactic semantic configuration. A fully productive morphological transformation that applies under any phono- logical, semantic, or morphological conditions, within a given syntactic/ 40 Ullman
semantic configuration, may be referred to as a global default. For example, English past tense -ed suffixation is a global default. The necessary and suffi- cient conditions under which a global default may apply are simply equivalent to its associated configuration. Thus, for a given syntactic/semantic configu- ration, a local default can apply in a particular subset of the phonological, semantic, or morphological feature space, whereas a global default can apply in the entire feature space. Partially productive transformations apply to some or perhaps many or most, but crucially not to all forms���apart from those forms to which a transformation of higher precedence has applied���that meet any possible set of phonological, semantic, or morphological conditions within a given syn- tactic/semantic configuration. In other words, there exists no necessary and sufficient set of conditions under which a partially productive transforma- tion can be fully productive. Consequently, there is no strict precedence relationship between a partially productive transformation and the transfor- mation(s) that it can precede: For any set of conditions, a given partially productive transformation will only sometimes precede other transforma- tions. This variable precedence relationship is therefore a key test of partial productivity. For example, the past tense mapping in feed���fed, lead���led, breed���bred, speed���sped can apply to new forms (e.g., cleed���cled), but does not apply to heed, seed, need, weed, which are -ed suffixed. Partially pro- ductive transformations can range in productivity, and can be found both among stem change and affixal transformations (e.g., German participle -en suffixation (Marcus, Brinkmann, Clahsen, Wiese, & Pinker, 1995) and Japanese adjectival past-tense -katta suffixation (Fujiwara & Ullman, 1999)). THE MENTAL MODEL A mental model of morphology is presented here. It is similar in cer- tain respects, but crucially different in others, to models that have previ- ously been proposed by my colleagues and myself (Marcus et al., 1992, 1995 Pinker, 1991 Pinker & Prince, 1988, 1991 Prasada & Pinker, 1993 Ullman, 1993, 1999a Ullman, Corkin et al., 1997). According to the model (also see Ullman, Corkin et al., 1997 Ullman et al., in press), morphological transformations can be computed by either of two components, which have distinct cognitive, computational, and neural bases. One component is a memory system. This is not a rote memory con- taining a list of the morphologically complex forms yielded by morpho- logical transformations. Rather it is an associative memory of distributed representations, over which the transformations��� phonological and con- ceptual���semantic mappings are learned, stored, and computed. The system Declarative/Procedural Model 41