Does impaired grammatical comprehension provide evidencefor an innate grammar module

Children with specific language impairment (SLI) have distinctive impairments in the comprehension of sentences that involve long-distance syntactic relationships. This has been interpreted as evidence for impairment in an innate grammatical module. An alternative theory attributes such difficulties to lower level problems with speech perception or deficits in phonological working memory. These theoretical accounts were contrasted using comprehension data from three subgroups: 20 children with SLI, 19 children with mild–moderate hearing loss, and normally developing children matched on age and/or language level. There were close similarities between the hearing-impaired and SLI groups on a measure of phoneme perception. Children with SLI did poorly on tests assessing knowledge of Binding principles and in assigning thematic roles in passive sentences whereas hearing-impaired children performed close to control levels, indicating that poor speech perception cannot account for this pattern of deficit. However, the pattern of errors on syntactic tasks and the relatively weak correlation between different indicators of syntactic deficit seemed incompatible with a modular hypothesis. We propose that limited processing capacity is the principal determinant of deficient syntactic comprehension in SLI. The study of children with specific language impairment (SLI) in the context of otherwise normal development provides an opportunity to investigate the mechanisms and processes that children bring to the task of language learning and where these processes might break down in the case of SLI. Research over the years has provided essentially two contrasting causative theories of SLI. One view, which we refer to as a modular account, posits that grammatical knowledge is innate and that aspects of this knowledge are either missing in SLI (e.g., Gopnik & Crago, 1991) or mature very late (e.g., Rice & Wexler, 1996). The modular account put forward by Van der Lely and colleagues posits a “representational deficit for dependent relationships” (RDDR) in the syntactic system, which accounts for grammatical errors in both production and comprehension.  2002 Cambridge University Press 0142-7164/02 $9.50 Applied Psycholinguistics 23:2 248 Norbury et al.: Impaired grammatical comprehension This theory is the focus of the research reported here. Although modular theories vary in details, they share the crucial notion that grammatical impairment in SLI is selective and independent of other nonsyntactic linguistic domains such as lexical knowledge or phonological processing. Nevertheless, there is mounting evidence that most children with SLI experience a range of deficits outside the syntactic system. These include auditory perceptual deficits, memory deficits, and general processing capacity limitations (see Leonard, 1998, for a thorough review). A crucial question is whether these additional difficulties are simply associated symptoms of SLI or whether they are causally linked to the grammatical deficits. A nonmodular account of SLI was recently summarized by Joanisse and Seidenberg (1998): “there is good evidence that SLI is associated with impairments in the processing of speech; that these impairments affect the development of phonological representations; and that degraded phonological representations are the proximal cause of deviant acquisition of morphology and syntax, by virtue of their roles in learning and working memory” (p. 241). According to this view, grammatical deficits are seen as a secondary, downstream consequence of lower level perceptual deficits. Modular versus nonmodular accounts: Evidence from children with hearing impairments One way to tease apart the different causal theories is to investigate the syntactic abilities of children who have known deficits in areas thought to be important for language learning. Briscoe, Bishop, and Norbury (2001) noted that mild– moderate sensorineural hearing (SNH) impairment provides an informative comparison with SLI. As Moore (1998) pointed out, amplification does not normalize the hearing of people with cochlear hearing loss. Sounds may be perceived as unclear or distorted, especially when background noise is present. Of particular interest here is the fact that cochlear damage leads to poor temporal resolution at low sound levels and difficulty in following the temporal structure of everyday sounds: this suggests some similarities with the auditory perceptual difficulties described in SLI (e.g., Tallal, Miller, & Fitch, 1993). Briscoe et al. (2001) studied children with hearing losses (SNH) in the range of 20–70 dB and found that their performance on tests of phonological discrimination, short-term memory, and awareness was very similar to that of a group of children with SLI and well below control levels. However, the SNH group, unlike the SLI group, was largely unimpaired on tests of verbal short-term memory and literacy, suggesting that problems in speech perception cannot explain the pervasive verbal impairments seen in SLI. Norbury, Bishop, and Briscoe (2001) went on to compare these same groups of children on the production of finite verb morphology. The rationale was as follows: if intact speech perception is necessary for morphosyntactic development to proceed normally, then one would expect those children who experienced degraded auditory perception in the language learning years (the SNH group) to have deficits in verb morphology that mimic the deficits seen in SLI. This hypothesis was contrasted directly with the extended optional infinitive (EOI) account of SLI put forward by Rice and Wexler (1996). Applied Psycholinguistics 23:2 249 Norbury et al.: Impaired grammatical comprehension The results appeared to favor the EOI account in finding severe and distinctive difficulties with verb morphology in only the SLI group. The SNH group of children were superior to those with SLI on expressive tasks of regular and irregular past tense morphology and measures of third person singular agreement, even though their speech discrimination abilities resembled those of the SLI group. This clearly indicated that low-level perceptual deficits alone cannot account for the impairments of morphosyntax seen in SLI. However, there were several pieces of data that did not fit well with a modular interpretation. First, 6 out of 19 children with SNH exhibited deficits similar to those seen in SLI. Although they constituted a minority of the SNH group, the number is considerably higher than one would expect to find in a normally hearing population, suggesting that perceptual deficits could act as a risk factor for SLI. Second, for all groups, the children’s omission of verb inflections was dependent on nonsyntactic properties of the verb such as frequency and phonological complexity. Third, there was a relationship between tense marking and nonsyntactic language measures, such as vocabulary, nonword repetition, and recalling sentences, all tasks on which children with SLI do poorly. This suggested that processing limitations might play a crucial role in morphosyntactic deficits. Although the study by Norbury et al. (2001) raised important issues about the nature of underlying deficits in SLI, it dealt exclusively with expressive syntax. One might expect that, if defective speech perception affects language learning, its effects would be seen more in receptive language than expressive. Although most of the literature on SLI has focused exclusively on expressive grammatical abilities, there is a small and growing literature on the comprehension of syntax. Again, causal explanations can be divided into modular and nonmodular accounts. Grammatical comprehension: A modular account Van der Lely, Rosen, and McClelland (1998) proposed that there is a homogeneous subtype of grammatical SLI (G-SLI), which is characterized by severe deficits in grammatical production and comprehension relative to nongrammatical language domains such as lexical knowledge or phonology. These authors suggest that these deficits are attributable to a RDDR in the underlying syntactic system, which leads to disproportionate difficulties using sentence elements that mark syntactic dependencies. An example of such a relationship includes the overt marking of verb tense (e.g., “I washed my hair yesterday”), which involves a syntactic dependency between the verb and the functional category of inflection (Van der Lely & Stollwerck, 1997). Children without this syntactic knowledge may produce bare verb stems in contexts where an inflected form is obligatory, an error type that is prolific in SLI (e.g., Rice & Wexler, 1996). In a series of studies, Van der Lely and colleagues provided evidence that children with SLI have difficulty in comprehension, as well as production, of long-distance dependencies. This is evidenced in poor comprehension of reversible active and passive sentences (Van der Lely & Harris, 1990); in a bias toward adjectival, rather than verbal, interpretation of truncated passives (Van der Lely, 1996); and in problems in applying Binding principles to assign pronominal reference Applied Psycholinguistics 23:2 250 Norbury et al.: Impaired grammatical comprehension in sentences (Van der Lely & Stollwerck, 1997). Van der Lely et al. (1998) argued that children with these deficits may perform normally on auditory and cognitive tasks, and they concluded that where nonlinguistic deficits are found in SLI may not be the cause of the grammatical impairments. They further argued that their results point to “the existence of a genetically determined specialisation of a sub-system in the brain required for grammar, and, it appears, for nothing else” (p. 1257). Grammatical comprehension: Nonmodular accounts The nonmodular accounts considered two factors that might affect children’s comprehension of grammar: auditory perception and processing capacity. Bishop (1997) argued that perceptual deficits could impact the acquisition of a range of syntactic structures by increasing the ambiguity of both the inflected items and the items with which they contrast. For instance, if a child had difficulty perceiving items of low perceptual salience, such as by or the inflection -s, then statements such as “the boy is hit by the girl” and “the boy hits the girl” would both be represented as “boy hit girl,” resulting in a breakdown of comprehension. Leonard and Eyer (1996) argued that morphemes such as -ed and -s and function words such as the and a are crucial for language development because they give cues to the grammatical category of unfamiliar words. For example, one does not need to known the meaning of “the zoop” to realize that zoop is a noun. Other accounts focused on processing capacity, which are limitations in the amount of material that a child can comprehend when computing meaning from rapidly incoming input. Montgomery (1995, 2000) found that children with SLI showed deficits in both phonological memory capacity (as evidenced by poor repetition of polysyllabic nonwords) and working memory (as evidenced by poor recall of word lists when required to organize the recalled words into semantic category and then a size sequence). In a sentence comprehension task, children with SLI were impaired at comprehending the longer redundant sentences compared to controls and their own performance on the nonredundant sentences. These two sentence types varied only in length; they contained the same syntactic structures and semantic information. Furthermore, both memory measures were reliably correlated with the sentence comprehension task, suggesting that processing capacity was a key factor in sentence comprehension. Distinguishing between theories: Sources of evidence One kind of evidence that has been advanced to distinguish between theoretical accounts is the pattern of performance on tests of syntactic comprehension. An important point in Van der Lely’s argument for modularity is that children with SLI are not simply immature in their syntactic comprehension: when compared with younger normally developing children matched on language level, they typically do worse. Nevertheless, Bishop (1997) noted that, in the production and comprehension of syntactic structures, children with SLI do not behave as if they have no grammatical knowledge. Their performance is typically worse Applied Psycholinguistics 23:2 251 Norbury et al.: Impaired grammatical comprehension than that of other children but nevertheless well above chance, which would suggest that the problem is one of deploying grammatical knowledge under realtime processing constraints, rather than the lack of such knowledge. A second source of evidence comes from considering the relationship between different syntactic deficits. If a single modular deficit leads to problems in tense marking, thematic role assignment in reversible sentences, and the use of Binding principles, then we should be able to find children with G-SLI who show all these characteristics. Just such a case, AZ, was reported by Van der Lely and colleagues (Van der Lely, 1997; Van der Lely et al., 1998) who studied him between the ages of 6 and 13 years. Van der Lely et al. (1998) also briefly reported on six children with linguistic and cognitive profiles similar to those of AZ. Bishop, Bright, James, Bishop, and Van der Lely (2000) investigated the notion of a distinct syndrome of G-SLI. They studied the understanding of active–passive sentences and Binding principles in a sample of 141 twin pairs, aged 7–13 years, including some selected for the presence of SLI. Like Van der Lely, Bishop et al. found that in both comprehension tasks the overall performance differentiated language-impaired children from those with normal language. However, the particular pattern of performance predicted by the RDDR was not entirely evident. Most children displayed only partial deficits on the RDDR measures, and those who had significant deficits across measures had concomitant impairments in other language functions. Bishop et al. concluded that the results leave open the possibility that children with SLI make errors on measures of grammatical comprehension for reasons other than a difficulty in deriving syntactic relationships among sentence constituents. The finding of partial manifestations of G-SLI in many children raised the question of whether the cases reported by Van der Lely are a qualitatively distinct subgroup of children or rather represent the extreme on a continuum of performance. It is possible, however, that the failure by Bishop et al. (2000) to find more cases of G-SLI reflects the fact that many children in their study had relatively mild forms of SLI. In this article we consider a further source of evidence: children with hearing loss. We know that children with severe and profound hearing losses have difficulties in comprehending English syntax and may show unusual patterns of performance resembling those seen in receptive SLI (Bishop, 1983a). However, the interpretation of such findings is complicated, because these children typically learn language visually via signing, written language, and lipreading. Spoken English is, in effect, a second language for such children. A more realistic test of how far auditory deficits may influence the learning of syntax is provided by children with milder levels of permanent hearing loss, who are not part of the deaf community and are not exposed to signing. To date, only a handful of studies has been conducted on language development in such children and no study has focused specifically on syntactic comprehension. Following the Norbury et al. (2001) findings regarding expressive syntax in this population, in the current article we report data on syntactic comprehension in the same children, comparing their performance to that of a sample of children with SLI. Applied Psycholinguistics 23:2 252 Norbury et al.: Impaired grammatical comprehension Aims of the current study The study reported here replicates and extends the investigation conducted by Bishop et al. (2000) and explicitly compares children with documented mild– moderate hearing impairment to groups of normally hearing children and a group of normally hearing children with SLI. In doing so we sought to elucidate the following: 1. the extent to which children with mild–moderate hearing loss exhibit deficits similar to those seen in SLI and 2. how far the patterns of responding in syntactic comprehension are associated and can be explained by the RDDR versus alternative nonmodular accounts