Specific Language Impairments in Children Phonology, Semantics, and the English Past Tense

Theories of specific language impairment (SLI) in children turn on whether this deficit stems from a grammarspecific impairment or a more general speech-processing deficit. This issue parallels a more general question in cognitive neuroscience concerning the brain bases of linguistic rules. This more general debate frequently focuses on past-tense verbs, specifically, whether regular verbs (bake-baked) are encoded as rules, and whether irregular forms (take-took) are processed differently. Children with SLI have difficulties with past tenses, so SLI could represent an impairment to rules. An alternative theory explains past-tense deficits in SLI as resulting from a phonological deficit. Evidence for this theory has been obtained from connectionist models of past-tense impairments and from behavioral studies of languageand reading-impaired children. The data suggest that SLI is not an impairment to linguistic rules, that past-tense impairments can be explained as resulting from a perceptual deficit, and that a single processing mechanism is ideally suited to account for these children’s difficulties. KEYWORDS—specific language impairment; connectionism; English past tense; speech perception A key question in cognitive neuroscience concerns the neural mechanism by which humans encode the rules of language. The English past tense represents an interesting case of rulelike processes: Although regular patterns (bake-baked, step-stepped) appear to be rulelike, English also has a number of irregular forms (take-took, sleep-slept) that conflict with the rule that the past tense is formed by adding -ed to the present tense. Irregular forms are problematic to a rule-based approach because they call into question whether rules alone are sufficient for explaining linguistic phenomena, and whether a secondary mechanism is required for encoding these irregular forms. In 1986, Rumelhart and McClelland proposed a connectionist model in which both regular past tenses and exceptions were encoded within a single type of neural mechanism. The connectionist approach to cognitive neuroscience explains cognitive processes as arising from the behavior of simple neuronlike processing units. To the extent that Rumelhart and McClelland’s approach to grammar proposed that rules are not necessary to explain systematic grammatical processes such as past-tense formation, it represented a radical departure from the accepted wisdom that language was processed within a symbolic ruleprocessing module, sparking a debate that continues to this day. Modern linguistics characterizes mental grammars as sets of symbolic rules. For example, rules are proposed to be the basis of grammatical morphology, a process by which words are created from ‘‘building blocks’’ such as stems, suffixes, and prefixes (bake-d, hand-s, talk-ing). One popular extension of this theory holds that past-tense morphology is encoded using two distinct types of neural mechanisms: a rule module that generates regular forms and an associative memory system that encodes exceptions case by case (Pinker, 1991). Recently, there has been some interest in studying this dual-mechanism theory from the perspective of developmental language disorders. Of particular interest are past-tense deficits in children with specific language impairment (SLI), an impairment marked by poor acquisition of grammar. Language problems in SLI cannot be explained by poor speech articulation, hearing loss, frank neurological deficit, or pervasive developmental disorder (Joanisse & Seidenberg, 1998; Leonard, 1998). If SLI is a grammar-specific disorder, as some linguists have indicated, it should exclusively affect regular past-tense forms while leaving irregular forms relatively intact (Pinker, 1991; van der Lely & Ullman, 2001). In this review, I discuss evidence regarding the pattern of past-tense impairment in SLI and suggest that the facts do not support the view of SLI as a rule-learning disorder. Instead, I present an alternative theory that explains SLI as an impairment in speech perception and processing. This deficit has important consequences for learning grammatical processes such as past-tense formation, and also makes interesting predictions about past-tense deficits in other populations of children. THE CONNECTIONIST THEORY The connectionist theory of morphology presented here builds on McClelland and Rumelhart’s (1986) model of the past tense. It holds that all morphological forms are processed within one type of processing mechanism (connectionist networks) distributed across multiple brain regions (Joanisse & Seidenberg, 1999). The connectionist Address correspondence to Marc F. Joanisse, The University of Western Ontario, Department of Psychology, London, Ontario, Canada N6A 5C2; e-mail: [email protected]. CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE 156 Volume 13—Number 4 Copyright r 2004 American Psychological Society approach assumes that complex cognitive abilities arise from interactions among simple processing units (i.e., neurons), and that complex behaviors such as language reflect the more basic characteristics of these neural networks. In the case of the past tense, this approach assumes that word knowledge relies on information about sound (phonology) and meaning (semantics). Although the relationship between sound and meaning is usually arbitrary, morphology is a special case because morphologically related words share both phonological and semantic relationships; thus, morphology represents a ‘‘convergence of codes.’’ For instance, walk and walked overlap both in phonology and in semantics (‘‘to move by foot,’’ ‘‘to move by foot in the past’’). This convergence of codes means that it is not necessary for speakers to use linguistic rules per se in order for them to exhibit knowledge of morphology. Instead, they may encode rules as statistical regularities—patterns of semantics-phonology relationships. An interesting prediction of this theory is that impairments should affect word forms regardless of whether they are regular, because regular and irregular forms are encoded in the same way. Thus, a deficit in past-tense formation should affect both regular and irregular forms of the past tense, because in both cases phonological and semantic information is used to encode the relationship between present and past tense. Rather than resulting from an impairment in rule learning, the morphology deficits in SLI can be explained by a phonological impairment that affects all past-tense forms, but is especially deleterious in the case of forms that the speaker has not encountered previously. IS SLI A RULE-LEARNING IMPAIRMENT? The primary source of data about morphological deficits in children with SLI comes from studies using a sentence-completion task (e.g., The girl likes to walk. She did the same thing yesterday; she ——). Theories of SLI as a rule-learning disorder predict that children with this impairment will have difficulty producing the past tenses of regular verbs on this task, but will perform better with irregular verbs. Surprisingly, however, most studies of children with SLI have found either no difference between regular and irregular verbs or numerically worse performance on irregular verbs (van der Lely & Ullman, 2001). These results seem inconsistent with a rule-specific deficit, because the children seem just as impaired on forms that are not rule governed as on those that are. However, the results are not surprising in light of the fact that English-speaking children generally find irregular verbs difficult and tend to learn them later than regular verbs. It is possible, then, that children with SLI have difficulty on irregular verbs because of an overall delay in learning past tenses. Other aspects of past-tense performance in children with SLI might be more informative, however. For instance, these children tend to make fewer overregularization errors (i.e., applying the regular pasttense form to an irregular verb, as in sleeped and taked) than normally developing children. Overregularization errors are thought to indicate the creative use of a past-tense rule, because they involve producing a form that has never been heard. The fact that children with SLI produce fewer overregularized forms than control children suggests that they have not acquired a rule for the past tense. Similarly, these children perform very poorly when asked to produce the past tenses of nonwords (for which the ‘‘correct’’ past tense is considered to be a regular ending, e.g., wug-wugged; Fig. 1). This is an ideal test of morphological knowledge because a speaker must use a rule creatively to answer correctly. (In contrast, a familiar form, like walked, might be recalled from memory.) Thus, the fact that children with SLI are poor at generating past tenses of nonwords again suggests that they have not encoded a rule for generating past tenses. According to the rule-based account, the pattern of deficits in SLI indicates that regular past tenses are processed differently in children with SLI than in other children (Pinker, 1991; van der Lely & Ullman, 2001). There is clearly room for an alternative explanation, however. For instance, a rule-based account does not explain why children with SLI have difficulty with irregular verbs. Similarly, the traditional characterization of SLI as a grammar-specific deficit fails to capture the full range of impairments that have been discovered in these children; ideally, an account of SLI should also explain these children’s delayed vocabulary development, impaired phonology, and impaired speech perception. THE PHONOLOGICAL-DEFICIT HYPOTHESIS The phonological-deficit hypothesis takes a different perspective on language impairments in SLI, proposing that a perceptual deficit leads to a phonological deficit that is the direct cause of the language deficits seen in this disorder. This hypothesis builds on the connectionist theory that past tenses normally arise through the integration of semantic and phonological information. A phonological deficit that is accompanied by relatively intact semantic representations will most severely impair the ability to generalize from known forms (bugbugged) to novel ones (wug-wugged ); intact semantic representations will help support some ability in the case of regular and irregular forms that are familiar, but do not come into play in the case of unfamiliar forms. Because connectionist theory posits that knowledge is implemented in a distributed and interactive neural system, the phonological-deficit hypothesis predicts that an impairment to any aspect of this system will tend to affect all word forms. However, because nonwords can rely only on phonology, these forms are predicted to be more severely impaired by a phonological deficit than true words are. Thus, this hypothesis explains why children with SLI do not show the same performance deficits for nonwords and familiar forms; in contrast, the rule-based theory predicts wholesale deficits on regular forms and intact performance on irregular forms. The phonological-deficit hypothesis grew out of earlier perceptionbased theories of SLI, which held that the language difficulties of children with SLI stem from a perceptual deficit that makes it difficult for them to perceive certain grammatical markers, such as the pasttense marker (Leonard, 1998; Tallal, Miller, & Fitch, 1993). The hypothesis proposes that phonological deficits are the key link between perceptual and grammatical impairments. The hypothesis further suggests that children with SLI show morphological impairments not only because they have difficulty perceiving grammatical markers that occur in difficult-to-perceive contexts, such as the endings of words, but also because they have difficulty translating the auditory forms of words into a phonological code necessary for learning word forms. Regular and Irregular Forms in Aphasia The theory that inadequate phonological representations can impair grammar has been previously tested in adults with aphasia (language impairments following brain damage). Like SLI, Broca’s aphasia Volume 13—Number 4 157 Marc F. Joanisse