The acquisition of past tense in preschool children with specific language impairment and unaffected controls: regular and irregular forms *.

Serratrice, Ludovica ; Joseph, Kate L. ; Conti-Ramsden, Gina 等

Abstract

The main aim of this study was to provide an analysis of the acquisition of past tense in preschool children with specific language impairment (SLI) and unaffected controls. Data from three children with SLI, two boys and one girl, aged 3;1, 3;5, and 4;0 at the beginning of the study, formed the basis of the investigation. These children were audiorecorded for an hour in free play sessions with their mothers at fortnightly intervals for a period of approximately ten months. For comparison purposes eleven MLU-matched children were selected from the Manchester corpus (Theakston et al. 2001) available on CHILDES (Mac Whinney 2000) to form an unaffected control group. The findings of the present study indicated that, contrary to previously reported investigations, these younger children's attempts at marking finiteness in past tense contexts were not significantly better for irregular than for regular verbs. This held true for both the control children and the children with SLI. No significant differences were observed either in the period prior to overregularization or after overregularization. A positive correlation was, however, found in both the SLI and the MLU groups between the most frequent past tense forms used by the children and those used by the eleven mothers of the MLU control children. The implications of these results for models of past tense acquisition such as the surface account, the extended optional infinitive account, and the single-mechanism account are discussed.

Introduction

The acquisition of past tense morphology in children with specific language impairment (SLI) as compared with typically developing children has recently been the subject of considerable interest. A body of findings is beginning to emerge that characterizes the performance of children with SLI as delayed with respect to language-matched typically developing children on a number of measures (Leonard et al. 1992; Rice et al. 1995; Oetting and Horohov 1997; Montgomery and Leonard 1998; Rice et al. 2000; Hansson and Leonard, this issue). All of the studies available to date include children with SLI aged five and older and the methodology is typically cross-sectional involving the elicitation of target past tense forms. The data gathered using older children are invaluable for the understanding of how children with SLI develop over time; however, information is now needed on the earlier stages of past tense acquisition. Although significant differences have been reported for older children with SLI when compared to MLU controls, there is no clear sense of how children with SLI start out in the acquisition process.

It is important to consider whether younger children with SLI may be more similar to their MLU controls than their older peers are. By age five and above it is possible that the gap between children with SLI and unaffected children has widened to a significant degree, while this may not necessarily be the case at earlier stages of development; that is, younger children with SLI may not be very different from MLU controls. Time is a variable that must be incorporated to a greater extent in the comparison of children with SLI and typically developing younger controls. The fact that children with SLI not only start producing language later than it is typically observed but also develop at a slower rate must be examined in more detail (Rice et al. 2000).

Moreover, because of the nature of the experimental tasks and of the spontaneous data available, typically consisting of thirty-minute samples with an experimenter, there is little information available on how the input that children with SLI are exposed to on a daily basis may interact with their acquisition process.

Previous research on the acquisition of past tense in children with SLI and MLU controls

A number of studies report that children with SLI have considerable and protracted difficulties in the marking of regular past tense with respect to both chronological age and MLU controls (Bishop 1994; Eyer and Leonard 1994; Leonard et al. 1992; Rice et al. 1995, 1998, 2000). When the context requires an inflected form containing an -ed affix, children with SLI are more likely than their normal MLU and chronological age controls to produce an uninflected bare form, such as walk instead of walk-ed. The difference between children with SLI and unaffected children is thus a quantitative one; this type of omission error is also found in typically developing children, but children with SLI are more likely to produce a proportionally larger number of bare forms instead of inflected forms when compared to typically developing children.

A number of recent studies have gone beyond the observation that children with SLI have difficulties with the production of regular past tense forms and have investigated the production of finite past tense forms by comparing children's performance on regular and irregular verb forms.

Three models of the acquisition of regular and irregular past tense

To date three different models have been proposed to account for the acquisition of past tense in children with SLI: the surface account (Leonard et al. 1992), the extended optional infinitive account (Rice et al. 1995; Rice and Wexler 1996; Rice et al. 2000), and the single-mechanism account (Oetting and Horohov 1997; Marchman et al. 1999).

Surface account. According to Leonard and colleagues' surface account (Leonard et al. 1992, 1997; Montgomery and Leonard 1998; Hsieh et al. 2001), children with SLI have a processing-capacity limitation that leads to difficulties with morphemes reduced in phonetic content. Although children with SLI are able to perceive word-final consonants and weak syllables of short duration, when such phonemes play a morphological role, as is the case of past tense affix -ed, overload due to their limited capacities may result in incomplete processing of the morphemes in question (Montgomery and Leonard 1998). This being the case, children with SLI are thought to need an increased amount of exposure to this morpheme (and other such phonetically reduced morphemes such as third person singular -s) before they can perceive them, categorize them according to their morphosyntactic function, and place them in the appropriate cell of the corresponding morphological paradigm (i.e. by learning the appropriate inflection for the morphological context).

The surface account makes the following predictions for children's performance on regular and irregular verbs: because regular past tense forms are associated with endings of reduced duration, which are considerably difficult to process and categorize for children with SLI, the expectation is that affected children's performance will be worse on regular past tense forms than on irregular past tense forms. In terms of a direct comparison with unaffected MLU controls, the prediction is that the performance of children with SLI will be significantly worse than the MLU controls on regular past tense forms but not on irregular forms.

Although frequency effects are not explicitly incorporated into the surface account, they are also likely to play a nontrivial role in the building of morphological paradigms, as is the possibility of children's learning of word-specific paradigms (Leonard and Eyer 1996). It is reasonable to assume that the high frequency of a given regular past tense form might override the low phonetic substance of its ending and thus reduce the length of time it will take for the child to acquire it. Oetting and Horohov (1997) in fact report a frequency effect for the suffixation of regular types for both children with SLI and normal controls and note that sensitivity to frequency is greater in the former group.

Extended optional infinitive account. The extended optional infinitive (EOI) account as originally formulated by Rice et al. (1995) and Rice and Wexler (1996) proposes that, similarly to unaffected children, children with SLI go through a period in which they do not know that tense marking is obligatory in main clauses. Unlike typically developing children, however, the period of optionality is considerably protracted for children with SLI, which results in the production of a number of nonfinite uninflected forms in contexts where finite forms are required (e.g. she walk instead of she walks, I go instead of I went).

An important claim of the EOI account is that regardless of the surface form, the finiteness principle applies in all main clauses, that is, marking of finiteness in main clauses is obligatory regardless of what form actually spells it out. By this rationale, at a time when children mark tense optionally in main clauses there is no a priori reason to expect that optionality should be different across morphemes and verb classes. In other words, one would expect that proportion of optional marking should be constant regardless of whether the tense marker required is -s for a simple present tense context (e.g. she needs), or a past tense form in a past tense context (e.g. I played). Similarly, in the case of tense marking in past tense contexts children's performance should not be influenced by whether the form required is regular or irregular.

In earlier EOI accounts of SLI (Rice et al. 1995; Rice and Wexler 1996) only the use of regular past tense was taken into account and no predictions were made with respect to the use of irregular past tense forms either for children with SLI or for unaffected controls. In a recent paper Rice et al. (2000) propose an EOI account of the use of past tense marking in children with SLI and unaffected controls that reconciles the observed differences in children's performance on regular and irregular past tense. The proposal distinguishes two separate components involved in tense marking: a morphophonological component, for which the child must learn the correct realization of a past tense form according to whether the verb root is regular or irregular, and a morphosyntactic component, which requires the child to mark finiteness regardless of the morphophonological form selected. For Rice et al. (2000) the production of overregularizations shows that children know that a finite form is required in a given context, an indication that their knowledge is intact from the morphosyntactic point of view. At the same time the morphophonological knowledge that allows them to select the correct form is deficient. What children lack is specific lexical knowledge of which morphophonological form must be selected; however, they know that finiteness must be realized: failing the retrieval of an existing form, they resort to a novel overregularization to fill the lexical gap. (1)

The prediction made by the EOI acount in its strongest form would have to be that no difference should be observed between regular and irregular verbs in the marking of finiteness in past tense contexts. Rice et al. (2000: 1129; emphasis added) state that "[I]f children do not regard past tense marking as obligatory, they will not try to mark it on regular or irregular past tense forms." Presumably the converse is also true, that if children regard past tense as obligatory they will try to mark it on both regular and irregular past tense forms. By contrast, a more recent version of the EOI account, which distinguishes between a morphophonological and a morphosyntactic level of tense marking, acknowledges a distinction between regular and irregular verb types. The corollary prediction is that there will be an advantage for irregular verbs because they are more frequent in terms of both types and tokens in child-directed speech. It is, however, important to note that this prediction is based on information outside the scope of the EOI itself.

Single-mechanism account. Research on the acquisition of regular and irregular past tense in children with SLI and unaffected controls has focused to date almost exclusively on the phenomenon of overregularization (Oetting and Horohov 1997; Marchman et al. 1999; Rice et al. 2000), and virtually no information is available on children's acquisition of past tense forms prior to the stage at which they start to overregularize.

The number of studies investigating the role of frequency and phonological item-level features in connection with the onset of overregularization and the acquisition of regular and irregular past tense in children with SLI is still limited to date, but the findings so far confirm the importance of such variables in predicting how and when overregularization is likely to take place. Oetting and Horohov (1997) and Marchman et al. (1999) both provide evidence to support the hypothesis that overregularization can be predicted by the vulnerability of verb types according to their frequency and the phonological neighborhood to which they belong.

A rich literature exists on the use of regular and irregular past tense forms in typically developing English-speaking children, and on the phenomenon of overgeneralization whereby the regular affix -ed is incorrectly used with an irregular verb type to produce a novel past tense form, such as breaked instead of broke (Berko 1958; Cazden 1968; Bybee and Slobin 1982; Marcus et al. 1992; Prasada and Pinker 1993; Stemberger 1993; Marchman and Bates 1994). Children's overregularizations have been taken by some researchers as evidence for the emergence of a symbolic rule. A qualitative shift takes place when children realize that a productive rule exists to produce past tense forms; overapplication of the rule leads to erroneous irregular forms alongside correctly inflected regular forms. By the time the children have acquired this rule the past tense formation process is then regulated by a dual mechanism: a symbolic rule to compute regular -ed affixation, and a rote-learning associative mechanism that allows the learning and storage of irregular forms (Marcus et al. 1992; Prasada and Pinker 1993; Pinker and Prince 1988).

An alternative account of children's overregularization errors has been proposed by researchers working with usage-based models (Bybee and Slobin 1982; Bybee 1995) and connectionist models (Plunkett and Marchman 1991, 1993; Marchman 1997) where a single mechanism is thought to be responsible for the past tense formation process. The emergence of rule-like behavior is not determined by the acquisition of a symbolic rule, but by the gradual and incremental exposure to regular verb types. Plunkett and Marchman (1993) show that overregularizations start appearing in their network when the proportion and total number of regular verbs in the training set has reached a critical mass. Marchman and Bates (1994) report similar findings in a study of the relationship between verb vocabulary size and overgeneralizations in English-speaking children. It seems clear that the child must have accumulated an adequately large number of types before any generalization tendency becomes sufficiently strong to overrule previously established irregular past tense forms. Further evidence supporting the single-mechanism model comes from findings by Oetting and Horohov (1997), where both children with SLI and unaffected controls show sensitivity to frequency effects, not only for irregular verbs but also for regular verbs.

A key difference between the dual- and the single-mechanism models lies in the predictions they make with respect to the extent of the overgeneralization errors. For the dual-mechanism model the -ed affixation rule should in principle apply to any verb unless it is blocked by retrieval of a known irregular form. Presumably successful retrieval of an irregular form depends to some extent on how entrenched the form is, that is, how strongly the lexical representation has been established. For the single-mechanism approach two key factors predict the likelihood of an irregular verb being erroneously suffixed: token frequency of its past tense forms and the type composition of its phonological neighborhood. (2) As far as frequency is concerned the higher the frequency of a past tense form the higher the chances that the form will be learned. As for phonological neighborhood, the larger the number of friends and the smaller the number of enemies the less likely it will be that the form will be overregularized.

The predictions made by the single-mechanism account with respect to the use of regular and irregular past tense forms are stated as follows: regular and irregular forms will be learned as a measure of frequency; the more frequent a form the more quickly it will be learned. However, because the majority of high-frequency past tense forms tend to be irregular ones the obvious consequence is that irregular forms should be learned earlier. As far as the comparison between children with SLI and unaffected MLU controls is concerned, the expectation is that the former should not perform significantly differently from the latter if frequency is the determining variable affecting the acquisition sequence.

The effect of input on vocabulary and morphosyntactic acquisition

The relevance of frequency effects in the acquisition and use of lexical items in child language is by no means an uncontroversial question. However, evidence exists indicating that children's uptake is significantly affected by statistical regularities in the input such as type and token frequency. Studies of vocabulary acquisition investigating whether certain classes of words might be learned before others have typically reported a bias for noun learning in languages such as English, Italian, and Hebrew where nouns are highly frequent and tend to occur in sentence-final position (Dromi 1987; Goldfield 1993; Caseli et al. 1995). By contrast, similar research carried out in Korean and Mandarin Chinese has shown that children acquiring these languages tend to learn verbs at least as early as they learn nouns (Choi and Gopnik 1995; Tardif 1996). Unlike English and Italian child-directed speech, in Korean child-directed speech verb types outnumber noun types, and verbs are typically found in salient utterance-final position. Cross-linguistic evidence thus shows that, regardless of the cognitive demands imposed by the task, frequency and sentence position are good predictors of children's order of acquisition.

The relationship between frequency and the acquisition of morphosyntax has also been the subject of investigation in a number of recent studies. Hsieh et al. (1999) propose that the earlier acquisition of the morpheme -s as a noun plural marker, as opposed to -s as a third person singular verb marker, is determined by the frequency with which plural nouns outweigh third-person-singular verbs, and by the sentence position in which nouns and verbs occur.

Maternal input effects on verb acquisition have also been investigated in terms of the finite/nonfinite distinction by Pine et al. (1998a) and Wijnen et al. (2001). Pine et al. show that the correlation between the average frequency of the tensed and untensed forms in the maternal speech of twelve English-speaking mothers and the average age of acquisition of the forms in the children's speech is highly significant. Their conclusion is that frequency of occurrence in the input is a powerful predictor of the order of acquisition of tensed and untensed forms. Wijnen et al. (2001) provide support for the importance of the frequency distribution of maternal input in shaping the acquisition of finite and nonfinite verbs in Dutch.

In sum, although the effect of frequency on children's acquisition of lexical items and constructions may interact with other variables such as sentence position, semantic transparency, and phonological shape, there is sound reason to believe that frequency itself does play an important role. In fact, in some cases frequency alone can predict pattern of acquisition (Pine et al. 1998a; Rowland and Pine 2000).

Evidence for the important role played by input has been used by lexicalist-constructivist approaches to argue for a limited-scope account of language acquisition (Tomasello 1992; Lieven et al. 1997; Pine and Lieven 1997; Pine et al. 1998b). According to this view children's grammatical acquisition is a gradual phenomenon, which, at least in the initial stages, tends to be word-specific rather than category-general, and it is very much modeled on the input the child receives.

Aims of the study

The focus of the present study is on the early stages of the acquisition of regular and irregular past tense forms in preschool children with SLI and MLU-matched controls. The aims of the present study are the following:

1. To provide an account of the appearance and use of regular and irregular past tense forms in the speech of preschool children with SLI (3; 1-4; 8) and typically developing controls (1; 10-3;0).

2. To investigate children's use of regular and irregular past tense forms before and after the point of overregularization.

3. To examine the relationship between the frequency distribution of past tense forms used by children and the frequency distribution in the maternal input.

In line with the predictions of the surface account we expect that both children with SLI and unaffected controls should perform better on irregular verbs than on regular verbs. In addition we predict that input frequency will determine children's use of past tense forms, and therefore frequency could potentially override low phonetic substance effects in the acquisition of very frequent regular verbs.

As for the effect of overregularization, we consider it to be a lexically specific, gradual, and protracted phenomenon, and we do not expect an immediate across-the-board effect with a significant improvement of regulars alongside high rates of overregularized irregulars. The view taken here is of children with SLI as conservative learners who remain in the lexically specific period of language development for longer than their typically developing peers. Thus, because of the young age of our children with SLI, we would not rule out the possibility that their language might be more similar to that of the MLU controls than has been the case in previous studies of older children with SLI. This is because both groups of children are likely to be at a lexically specific stage of development.

Method

Participants

Data in the current study were collected from three children with SLI, one girl and two boys, over a period of approximately ten months. The children with SLI ranged in age from 3;1.9 to 4;0.9 at the beginning of the study, and from 3;10.22 to 4;8.30 at the end of the study. Speech problems were usually first identified by mothers, who then sought professional advice. The children were recruited through speech and language therapists in the northwest of England. All children with SLI had an IQ above 85, as measured by the Leiter (1979) performance scale. Children were not included if they presented with behavioral, hearing, or severe speech problems. Administration of the autistic screening questionnaire showed that none of the children had autistic tendencies (Berument et al. 1999). Furthermore children with SLI were included on the basis of a severe language impairment (receptive score below 16th centile), as measured by the Reynell developmental language scales (Edwards et al. 1997). The children were reported by speech therapists as being at the early stages of multiword speech.

The data for the eleven MLU-matched control children come from the Manchester corpus, collected by Theakston et al. (2001). The MLU controls ranged in age from 1;8.22 to 2;4.21 at the beginning of the study and from 2;5.8 to 3;0.10 at the end of the study. (3) All of the one-hour recording sessions took place in an informal play situation at the child's home with child's mother. The investigator was sometimes present as a participant observer. The children with SLI were recorded fortnightly and the MLU controls were recorded twice every three weeks.

Speech corpora

The three children with SLI in this study were audiotaped at fortnightly intervals for approximately an hour while playing with their mothers with toys provided by the investigator. All audio recordings were transcribed orthographically in CHAT format as described in the CHILDES manual (MacWhinney 2000), and a complete morphological tagging was automatically created using the MOR and POST programs also available on CHILDES. The criteria for inclusion of children's utterances in the corpora were that utterances (a) were fully intelligible; (b) used spontaneously (i.e. were neither self-repetitions nor imitations); and (c) were not strings from songs or nursery rhymes. The criteria for the identification of self-repetitions and imitations were the following: the utterance was counted as an imitation or a self-repetition if it was an exact repetition of a multiword utterance in the immediately preceding line.

For the mother, criteria for inclusion were that the utterances were (a) fully intelligible; and (b) not strings from songs or nursery rhymes or text read directly from books. The corpora were searched for past tense verb forms produced by the children and mothers and obligatory contexts for past tense forms that contained bare stems for the children. All no-stem-change irregular verbs for which present and past tense forms are the same (e.g. hit/hit, cut/cut) were excluded from the analyses. Obligatory contexts for past tense forms were defined as utterances containing a bare stem where there was an unambiguous interpretation from either the mother's recast or the general dialogue context that the child was talking about events in the past.

It was decided that data for individual mothers should be pooled together in order to obtain a large database of child-directed speech. Previous research into whether the input to children with SLI differs from the input to typically developing children has concentrated on the pragmatic aspects of the speech corpora. Although mothers of children with SLI have been shown to use fewer recasts (Conti-Ramsden 1990), a number of studies have shown similarities between mothers of children with SLI and mothers of MLU controls, for example in their responsiveness to topic changes and communicative acts (Messick and Prelock 1981) and in their use of requests, directives, assertions, and regulating devices (Conti-Ramsden and Friel-Patti 1983). To our knowledge studies to date have not separately considered the lexical, morphological, and syntactic properties of the input to children with SLI. Due to the lack of previous research into differences between the input produced by mothers of affected and unaffected children, we decided to analyze seven hours of speech data for the eleven mothers of the MLU controls only. This combined mother corpus contained over 60,000 utterances and was considered to be a representative sample of input language to use for both groups of children.

In order to assess the impact of the onset of overregularization it was necessary to define stages prior to and following the onset of these productions. A number of researchers have dealt with the frequency of overregularizations (c.f. Marcus et al. 1992; Maratsos 2000); however, to the best of our knowledge the issue of how to define onset of overregularization has not been dealt with. In most cases the first appearance of an overregularized form was taken as the onset of the overregularization phenomenon. In order to avoid counting one-off instances of overregularizations that could have been overheard and to ensure that the phenomenon was as robust as possible, a point of overregularization was defined here as the first of two consecutive sessions containing at least one overregularized verb.

Results and discussion

Analyses of naturalistic speech data for children with SLI and MLU controls were carried out in order to describe in detail the distribution of past tense forms used by the two groups and to find possible explanations for the patterns observed. The distribution of past tense forms was defined with respect to two main variables: verb class (whether verbs were regular or irregular) and finiteness (whether finite past tense forms were supplied in obligatory contexts). The effects of the onset of overregularization in child speech and of the input frequency in maternal speech were investigated as possible explanatory variables. Summary statistics for the fourteen participants are provided in Table 1.

Past tense verb forms have been traditionally classified in terms of whether they belong to the regular or irregular verb class. Table 2 presents information on the children's verb use according to the regular/irregular dichotomy in terms of verb types and tokens.

Although the children in both groups show a symmetrical distribution in the number of regular and irregular past tense verb forms overall, over 75% of all high-frequency types and over 85% of high-frequency tokens are irregular. Thus, although the children may know and produce approximately the same number of different regular and irregular types, irregular types feature far more often (i.e. they have a higher token frequency) in the past tense use of both children with SLI and MLU-matched controls. That is, around 75% of all past tense tokens produced by children are irregular.

Having established that irregular past tense forms (in terms of both types and tokens) are used more frequently than regular past tense forms, it is important to ascertain whether there is a difference in the percentage use of regular and irregular finite forms in obligatory contexts. That is, are irregular forms supplied more often than regular forms in past tense contexts? In other words, is there a relationship between word class and finiteness?

The relationship between verb class and finiteness

Table 3 shows the number of regular and irregular verb types used in finite and nonfinite contexts. The results are shown for types and not tokens to partial out the effect of cases in which a very small number of verb types would account for a large number of tokens, as in the case of one child who has a very high proportion of nonfinite tokens represented by one single verb type.

In both groups irregular finite forms were supplied either to the same degree or more often in obligatory contexts than regular finite forms. However, if the actual frequencies are considered, irregular finite past tense forms occur most frequently out of the four possible finite/nonfinite forms. The children's frequent use of irregular forms is a likely cause of the observed difference between percentage finiteness for regular and irregular types. This is because irregular past tense forms are far more frequent than regular past tense forms, constituting three-quarters of all past tense tokens produced by both groups of children. These results are inconsistent with the prediction for regular past tense made by Rice et al. (2000). It was predicted that children with SLI would be similar to MLU controls on irregular past tense but that they would perform worse than MLU controls on regular past tense. As can be seen from Table 3, there is support for the first prediction, as the two groups perform at similar levels in their use of irregular past tense. However, the children with SLI in the current study perform as well as the MLU controls, if not slightly better, on regular past tense, as shown by the percentage figures.

An analysis of variance was conducted on the data for the MLU controls shown in Table 3. No main effects were found for verb class (F(1, 10) = 2.90, p = 0.12) or finiteness (F(1, 10) = 4.40, p = 0.06); however, there was a significant interaction between verb class and finiteness in these children's use of past tense forms (F(1, 10) = 7.16, p < 0.05). This result is due to the fact that the average number of irregular past tense forms used by the children (141) is almost twice as large as the average number of regular past tense forms (72), bare stems in regular past tense contexts (74), and bare stems in irregular past tense contexts (74). That is, the children have a tendency to use finite irregular verb forms more often than they use regular finite forms and regular and irregular nonfinite forms. It was not possible to conduct an analysis of variance on the children with SLI as there were only three participants; however, visual inspection of the distributions of past tense use for the three children with SLI show a similar pattern of past tense marking for irregular verbs. Specifically, the average number of irregular past tense forms produced by the children with SLI (69) was larger than the average number of regular past tense forms (42), as well as the average numbers of unfulfilled obligatory contexts for regular and irregular past tense forms (25 and 41 respectively).

In sum, irregular finite forms constitute the most frequently used form in all past tense contexts. This result is consistent with the finding that the vast majority of the high-frequency types in the past are irregular.

The effects of overregularization

The production of overregularizations such as comed has been documented as an important creative process in language acquisition. It has been suggested that the addition of the regular -ed inflection to irregular stems represents the emergence of a symbolic rule (Marcus et al. 1992). Given that irregular finite forms are used more frequently in absolute terms, and, in the case of the MLU controls in this study, they are supplied more often in obligatory contexts, it is interesting to consider whether the onset of overregularizations affects the degree to which regular finite forms are supplied in obligatory contexts. In other words, does the onset of overregularization mark a qualitative shift in children's representation of tense and finiteness?

One child in the SLI group (Nathan) and four children in the MLU group (Anne, Becky, Carl, and Warren) met the criterion for a point of overregularization defined as the first of two consecutive sessions containing at least one overregularized verb. (4)

The data presented in Table 4 were analyzed using hierarchical log linear models for each of the five children separately. There were no third-order interactions, suggesting that there were no consistent relationships between verb class, finiteness, and stage. Significant interactions were found for two of the five children. For Warren, there was a borderline interaction between finiteness and stage ([chi square] = 3.70, p = 0.055), indicating that the child increased his use of finite forms from stage 1 (before overregularization) to stage 2 (after overregularization). For Becky there was a significant interaction between regularity and finiteness ([chi square] = 5.11, p < 0.05), indicating that the child used more regular finite forms than irregular finite forms overall.

These results imply that there was no significant improvement in finiteness marking on regular past tense forms once the children had begun to overregularize. None of the children showed an increase in their use of regular past tense forms in obligatory contexts up to five months after the point of overregularization. It is possible that the predicted differences between children with SLI and MLU controls become apparent later on in development when the gap widens between the two groups, whereas the two groups are not that different during the initial stages of past tense use.

The lack of improvement in children's production of finite regular forms is inconsistent with accounts interpreting overregularizations as the emergence of a symbolic rule applying across the board, that is, dual-mechanism accounts of past tense learning (Marcus et al. 1992; Prasada and Pinker 1993). Even invoking a blocking mechanism whereby overregularization errors are preempted by the retrieval of a stored irregular form does not solve the problem in the case of these five children. By this rationale it would be logical to expect that any irregular verb type without a known past tense form would be equally likely to be overregularized, as variables such as frequency and phonological salience play no role in the dual-route account. However, this is not the case in the present study.

First, in nine cases the verb types that are overregularized do have a corresponding past tense form; in this case blocking may not have worked if the known form is not sufficiently entrenched. Second, a large proportion of irregular verb types for which no corresponding past tense form is known by the child are used in the bare stem form rather than being overregularized. Across the five children there is a total of 36 bare forms, of which 17 do not have a corresponding past tense form.

This lack of across-the-board overregularization also has implications for Rice et al.'s (2000) finiteness measure on irregulars including both correct forms and overregularizations. In Rice et al.'s most recent version of the EOI model, a distinction between a morphophonological component and a morphosyntactic component allows a new measure of finiteness that includes any attempt to mark past tense, regardless of whether the child produces a correct form or an overregularized one. Both types of form mark finiteness and as such testify to children's morphosyntactic knowledge of the requirement to mark tense, although their morphophonological knowledge is still incomplete and retrieval of a correct form is not always possible. If the stage at which children start to overregularize coincides with the insight that tense must be obligatorily marked, one would expect this new piece of knowledge to be applied to any verb. However, this is not the case: as shown by proponents of the single-mechanism account (Plunkett and Marchman 1993; Marchman 1997), overregularization is a selective process governed by frequency and phonological-neighborhood effects. Rather than a qualitative shift in children's knowledge of the obligatoriness of finiteness, overregularization is better defined as a phonologically constrained mechanism that becomes operational when correct irregular past tense forms are not particularly well entrenched.

The effects of input frequency

Given that the corpora of child speech data consisted of mother-child interaction, it was possible to investigate the influence of maternal input frequency of past tense forms on the forms used by the children. In order to obtain a representative picture of the lexical statistics of English child-directed speech, frequencies of past tense forms for the eleven mothers of the MLU controls were pooled to form a mother group. Recall that only data for the mothers of the controls were used due to the lack of information regarding similarities and differences between the language of mothers of children with SLI and mothers of typically developing children. Type and token frequencies of the regular and irregular past tense forms used by the eleven mothers of the control group children in a seven-hour sample are presented in Table 5. Although the mothers use a slightly greater mean number of regular past tense types than irregular past tense types, the mean number of irregular past tense tokens is considerably larger than regular past tense tokens. Whereas there was some variation in the frequency with which mothers used regular and irregular types, ten out of the eleven mothers used over twice as many irregular tokens as regular tokens. Thus, irregular past tense forms featured far more frequently than regular past tense forms in the speech of mothers as well as of both groups of children.

The data for children in each group were pooled to form an SLI group and an MLU group, and the data for the children in each group were correlated with data for the mother group. In order to avoid spurious relationships resulting from a large number of past tense forms used rarely by both mothers and children, only high-frequency forms in the child-group data were included. These forms are the high-frequency forms that occur at least five times in the SLI-group and MLU-group data. Past tense forms were not divided into regular and irregular classes--the high-frequency lists contained both classes of forms. The lists of absolute frequencies and their corresponding ranks are shown in Table 6.

As can be seen from Table 6, the frequency distribution of past tense verb forms was highly skewed, which made it necessary to perform a log transformation on the data. The relationships between verb forms used by the SLI group and mother group and the MLU group and mother group were examined.

A Pearson's product--moment pairwise correlation between the log percentage frequencies for the SLI group and mother group use of the thirteen most frequent past tense forms in the SLI-group data was r = 0.86 (df = 12, p < 0.001) (see Figure 1). The correlation between the log percentage frequencies for the MLU-group and the mother-group use of the seventeen most frequent past tense forms in the MLU group yielded a result of r = 0.69 (df = 16, p < 0.01) (see Figure 2).

[FIGURES 1-2 OMITTED]

In addition, the [R.sup.2] values for these correlations indicate how much variance in the children's past tense use is accounted for by the distribution of past tense forms in the mother-group input. In the case of the SLI group, the distribution of past tense forms in the maternal input accounts for 74.0% of the variance in the distribution of past tense forms in the children's speech, and for the MLU group, the corresponding figure is 47.6%. Therefore, for both groups, the lexical statistics of maternal input can account for between approximately one-half and three-quarters of the variance seen in child productions of past tense forms.

These results are consistent with a limited-scope account of language learning (Tomasello 1992; Lieven et al. 1997; Tomasello and Brooks 1999; Pine et al. 1998b). According to this view, children are conservative learners who derive much of their knowledge from the input to which they are exposed. As shown by a number of other studies investigating the interaction between children's verb use and maternal input (Naigles and Hoff-Ginsberg 1998; Pine et al. 1998a; Hsieh et al. 1999; Rowland and Pine 2000; Wijnen et al. 2001), input frequency is a powerful predictor of which forms will be used by the children. In addition, by inspecting the verbs included in the list of most frequent forms used by children with SLI and the MLU controls, it is apparent that phonological effects are also a factor, given that over 80% of verb types in both groups is represented by irregulars.

Further indirect support for the surface account and the single-mechanism account is also provided by these findings. First, for both the children with SLI and the unaffected controls, irregular past tense forms account for over 80% of the most frequently used types. As predicted by Leonard and colleagues, vowel-change irregular past tense forms are more salient for both groups of children and as such are acquired earlier. Second, although to a much lesser extent, regular past tense forms are also found among the most frequently used past tense forms, thus showing that phonetically reduced forms can be learned relatively early provided they are sufficiently frequent in the input. Although this is not a prediction that is explicitly made by the single-mechanism account, it is a logical consequence of the role played by frequency in the model: the more frequent the verb form, the quicker it will be learned. Although high-frequency verbs in child-directed speech tend to be irregular, a number of regular verbs are also used with considerable frequency by mothers when speaking to their young children; exposure is such that difficulties in the perception of the unstressed -ed suffix can be overridden.

Although there are few differences in the results for the children with SLI and the MLU controls in this study, it is necessary to account for the differences found in previous work with older groups of children (Leonard et al. 1997; Montgomery and Leonard 1998; Rice et al. 2000). The limited-scope account can explain why the two groups of children in the current study performed at similar levels while older children with SLI perform significantly worse than MLU controls. In the current study, both groups of children are at the early stages of multiword speech, when there is a substantial degree of lexical specificity in verb use (Pine et al. 1998b). This is highlighted by the correlations found between the mother and child distributions of past tense forms. However, over time, typically developing children become less dependent on the input, so that their knowledge becomes less lexically specific and more category-general. It is suggested that children with SLI are particularly conservative learners who take longer to create category-general knowledge from the lexical statistics of the input language (Conti-Ramsden and Jones 1997). Thus at later stages when MLU controls have developed an "add -ed" pattern based on a prolonged period of exposure to regular past tense forms, the children with SLI have yet to form such a pattern and are only able to use correctly those forms that are most frequent in the input.

The link between lexical and grammatical acquisition has been recognized by a number of researchers (Marchman and Bates 1994; Jones and Conti-Ramsden 1997; Conti Ramsden and Jones 1997; Bates and Goodman 1999). This tie is particularly relevant in the case of past tense acquisition on the assumption that exposure to a sufficiently large number of verb types is necessary before the child can form any sort of generalization. If children with SLI are overall poorer vocabulary learners, this implies that they will not be particularly good at learning low-frequency words, that is, in this case, regular past tense forms, which are also phonetically reduced. This being the case, it will take children with SLI a considerable amount of time before they are exposed to a sufficiently large number of regular verb types and a pattern is detected. In other words, they will reach the "critical mass" point considerably later than children who have better vocabulary-learning skills. The reverse is also true; superior vocabulary learners are more likely to notice low-frequency items. With respect to past tense acquisition this would give superior vocabulary learners a rather larger group of regular verb types that would serve as a basis for pattern generalization relatively early. Maratsos (2000) reports such a case for Abe, a very gifted vocabulary learner who acquired both obligatory regular and irregular past tensing very early (2;9) and who also had a very high overgeneralization error rate.

Concluding remarks

It is important to note that this study differs from previous work on the development of the past tense in two main ways. First, unlike previous research that draws evidence from probe-test data, the current work is based on a rich longitudinal database of spontaneous mother-child interaction. Second, both the children with SLI and the MLU control group were at the early stages of multiword speech (MLU 1.6-3.8), which represents an earlier stage of development than has previously been considered. It is therefore necessary to view the results obtained in light of these methodological differences.

The current findings shed light both on the distribution of past tense use in young children with SLI and MLU controls and on some of the proposed explanations for patterns observed. The distribution of regular and irregular past tense forms used by the children in both the SLI and MLU groups was found to be symmetrical in terms of verb types overall, but asymmetrical in terms of high-frequency verb types, with irregular past tense forms being used more often than regular ones. For the MLU controls, there was a significant interaction between finiteness and verb class, with irregular verb types significantly more likely to be used in a tensed form in obligatory past tense contexts.

These results only partially confirm the predictions made by the surface account with respect to the asymmetry between regular and irregular types. Only the MLU controls, but not the children with SLI, showed a significant difference between the percentage of correct regular past tense forms and irregular past tense forms in obligatory contexts, with irregulars being supplied in a tensed form more often than regulars. For the children with SLI no advantage for irregular verbs in obligatory contexts was found. This unexpected result could be accounted for by the relatively poor lexical learning skills that characterize this group of children. Note also that an advantage for irregular verbs has often been reported in the literature for children with SLI who are considerably older than the children included here. It is possible that the older children included in other studies will have reached a sufficiently large critical mass for a number of irregular past tense forms to show an advantage over the phonetically reduced regular past tense forms, which take even longer to be acquired.

For one child with SLI and four MLU controls who had an identifiable point of overregularization, there was no increase in the use of regular past tense forms in obligatory contexts in the stage following this point. This could suggest that overregularizations do not constitute evidence of an across-the-board rule. Nevertheless it is possible that a longer period of time after the onset of overregularizations may be needed in order to assess the impact of overregularization more fully.

The absence of improvement in the proportion of regular past tense forms in obligatory contexts and the selective overregularization of a limited number of irregular verb types are inconsistent with claims made by proponents of the EOI model with respect to finiteness marking. If overregularization marks a qualitative shift in the child's mental representation of finiteness and obligatoriness of tense marking, the expectation would be that any regular verb would be appropriately tense-marked, and at least any irregular verb without a sufficiently entrenched correct past tense form would be a candidate for overregularization. This is, however, not the case either for Nathan or for any of the other unaffected children. Our results point to an interpretation of overregularization as a lexical phenomenon driven by analogy and schema formation, in line with predictions made by the single-mechanism account and usage-based models.

Finally, the positive correlations between frequencies of past tense form use in the mother group and both the SLI and MLU groups testify to the importance of frequency as a predictor of order of acquisition and are consistent with a limited-scope account of grammatical development. The fact that the verb forms used most frequently by both children with SLI and typically developing children correspond to those used by a group of mothers suggests that children could be drawing upon distributional information from the input during early past tense acquisition. Thus, both groups of children seem to be similarly influenced by frequency effects in their use of past tense forms. Furthermore, the MLU controls show an advantage in the use of irregular forms in obligatory contexts, unlike the children with SLI. The role played by vocabulary-learning skills is thus even more obvious in the group of affected children. For them not only does it take longer to form a regular schema for regular verb types, typically lower in frequency than irregulars, but even learning the irregular forms seems to be taking a considerable amount of time. We take this as further evidence for the close relationship between lexical and grammatical development, and for the conservative learning strategies of children with SLI.

University of Manchester

Appendix. Overregularizations for the five children who met the
criteria for point of overregularization

 Child (group) Overregularization No.
 of
 tokens

Nathan (SLI) felled 4
 comed 3
 catched 1
 swimmed 1

Anne (MLU) bited 1
 borned 1
 hiddened 1
 lied (instead of lay) 1
 runned 1
 thoughted 1

Becky (MLU) stealed 4
 broked 2
 comed 2
 brokened 1
 sawed 1

Carl (MLU) broked 8
 comed 8
 runned 3
 catched 1
 doned 1
 falled 1
 goned 1

Warren (MLU) comed 11
 stucked 4
 blowed 1
 breaked 1
 broked 1

Table 1. Summary statistics

Group
 Child Age range MLU No. of No. of Mean no. of
 name range sessions past past tense
 analyzed tense contexts
 contexts per session

SLI
 Bonnie 4;0.09-4;8.30 2.2-3.4 14 113 8.1
 Harry 3;5.0-4;5.2 1.7-3.3 24 115 4.8
 Nathan 3;1.9-3;10.22 1.6-3.6 22 163 7.4

MLU controls
 Anne 1;10.07-2;7.01 1.5-2.8 7 64 9.1
 Aran 2;0.2-2;8.12 1.7-3.4 7 77 11.0
 Becky 2;2.15-2;10.25 1.7-3.1 7 49 7.0
 Carl 1;8.22-2;58 1.9-3.1 7 57 8.1
 Dominic 2;1.11-2;9.26 1.6-2.7 7 98 14.0
 Gail 1;11.27-2;8.6 1.6-3.0 7 94 13.4
 Joel 1;11.22-2;9.13 1.5-2.8 7 102 14.6
 John 1;11.15-2;7.24 1.8-2.8 7 21 3.0
 Liz 2;0.07-2;8.14 1.6-3.5 7 56 8.0
 Nicole 2;4.21-3;0.10 1.6-3.0 7 19 2.7
 Warren 1;10.6-2;6.23 1.9-3.8 7 46 6.6

Table 2. Regular and irregular past tense in terms of total number
of verb types, number of high-frequency types, and percentage of
irregular tokens for the children with and without SLI

Group Past tense forms

 No. of No. of % irregular
 regular irregular tokens of all
 types types past tense
 forms

SLI 29 32 77.4

MLU controls 45 33 74.8

Group High-frequency past tense form (a)

 No. of No. of % irregular
 high high tokens of all
 frequency frequency past tense
 regular irregular forms
 types types

SLI 2 11 92.9

MLU controls 5 18 86.9

(a.) High-frequency past tense verb types were those that appeared
with five or more tokens in the pooled speech data for the SLI and
MLU groups.

Table 3. The number of past tense verb types in terms of verb class
and finiteness used by the children in the SLI and MLU groups

 + finite - finite % finiteness

SLI (N = 3)
 regular 42 25 63
 irregular 69 41 63

MLU controls (N = 11)
 regular 72 74 49
 irregular 141 74 66

Table 4. The number of past tense types used before and after the point
overregularization for one child with SLI and four MLU controls

Child (group) Regular and irregular %
 Before/after past tense verb types occurring finiteness
 overregularization in finite and nonfinite form

 + finite - finite

Nathan (SLI)
 before
 regular 14 8 64
 irregular 21 5 81
 after
 regular 10 2 83
 irregular 13 2 87

Anne (MLU)
 before
 regular 5 1 83
 irregular 9 2 82
 after
 regular 7 7 50
 irregular 13 5 72

Becky (MLUU)
 before
 regular 7 0 100
 irregular 9 7 56
 after
 regular 11 1 92
 irregular 18 4 82

Carl (MLU)
 before
 regular 3 4 43
 irregular 1 2 33
 after
 regular 21 9 70
 irregular 8 6 57

Warren (MLU)
 before
 regular 2 7 22
 irregular 3 5 38
 after
 regular 11 9 55
 irregular 7 4 64

Table 5. Type and token frequencies for the eleven individual mothers'
use of regular and irregular past tense

Mother of Past tense types Past tense tokens
MLU control regular irregular all regular irregular all
child

Anne 37 32 69 87 181 268
Aran 105 51 156 254 425 679
Becky 32 25 57 57 165 222
Carl 33 34 67 80 220 300
Dominic 48 40 88 92 270 362
Gail 55 38 93 124 298 422
Joel 51 40 91 93 179 272
John 38 31 69 69 170 239
Liz 28 28 56 53 104 157
Nicole 47 25 72 70 169 239
Warren 34 42 76 52 154 206

Group mean 46.2 35.1 81.3 93.7 212.3 306.0

Table 6. Token frequency and ranks for past tense forms used by the SLI
and MLU groups and their corresponding frequencies and ranks in the
mother group data

High- Frequency Frequency High- Frequency Frequency
frequency in SLI in mother frequency in MLU in mother
past tense group group past tense group group
forms for (rank) (rank) forms for (rank) (rank)
SLI groups MLU
 control
 group

got 27 (1) 252 (1) had 45 (1) 229 (2)
lost 25 (2) 45 (5) got 44 (2) 252 (1)
had 23 (3) 229 (2) said 31 (3) 163 (3)
did 21 (4) 128 (4) went 27 (4) 159 (4)
said 16 (5) 163 (3) lost 22 (5) 45 (9)
broke 10 (6.5) (a) 21 (10) bumped 19 (6) 9 (16)
found 10 (6.5) 43 (6) found 18 (7) 43 (10)
finished 7 (8.5) 23 (9) did 16 (8) 128 (5)
bought 7 (8.5) 38 (7) fell 15 (9) 35 (12)
fell 6 (10.5) 35 (8) happened 14 (10) 88 (6)
forgot 6 (10.5) 8 (12) bought 13 (11) 38 (11)
crashed 5 (12.5) 10 (11) saw 10 (12) 53 (8)
won 5 (12.5) 7 (13) came 9 (13) 68 (7)
 broke 8 (15) 21 (14)
 crashed 8 (15) 10 (15)
 forgot 8 (15) 8 (17)
 gave 5 (17) 30 (13)

(a.) Tied ranks are given as mean averages.

Notes

* This research was supported by the Economic and Social Research Council, ESRC grant to Gina Conti-Ramsden (grant number R00023 7767). We would like to thank Rachel F. Hick for help with data collection and Brian Faragher for help with statistical analyses. In addition, our special thanks go to the three children with SLI and their families who participated in the study. Correspondence address: Professor Gina Conti-Ramsden, Human Communication and Deafness, School of Education, University of Manchester, Oxford Road, Manchester M13 3PL, UK.

E-mail: gina.conti-ramsden@man.ac.uk.

(1.) It must be noted that the average age of the children with SLI at the start of the Rice et al. (2000) study is 4;8 and the average age of the language-age controls is 3;0 (3N group). The evidence reported indicates that for a protracted period of time both the SLI group and the 3N group resort to overregularization with some consistency. However, because of the age range selected, no information is available on children's marking of finiteness in past tense contexts prior to the onset of overregularization.

(2.) Phonological neighborhood is defined with respect to the phonological features that are relevant for the clustering of past tense forms, i.e. stem-final vowel--consonant sequence. Neighborhoods are composed of "friends" and "enemies." For example the verb throw (past tense form threw) has a number of friends that have similar sounding past tense forms (blow/blew, grow/grew, flow/flew) but it also has a number of enemies, who despite having the same stem-final vowel in the root have a regular past tense form (mow/ mowed, show/showed, snow/snowed). If the number of enemies is higher than the number of friends the vulnerability to overregularization increases considerably.

(3.) MLU was calculated in words throughout.

(4.) See Appendix 1 for a list of overregularizations.

References

Bates, Elizabeth and Goodman, Judith (1999). On the emergence of grammar from the lexicon. In The Emergence of Language, Brian MacWhinney (ed.), 1-79. Mahwah, NJ: Erlbaum.

Berko, Jean (1958). The child's learning of English morphology. Word 14, 150-177.

Berument, Sibel Kazak; Rutter, Michael; Lord, Catherine; Pickles, Andrew; and Bailey, Anthony (1999). Autistic screening questionnaire: diagnostic validity. British Journal of Psychiatry 175, 444-451.

Bishop, Dorothy (1994). Grammatical errors in specific language impairment: competence or performance limitations? Applied Psycholinguistics 15, 507-550.

Brown, Roger (1973). A First Language: The Early Stages. Cambridge, MA: Harvard University Press.

Bybee, Joan (1995). Regular morphology and the lexicon. Language and Cognitive Processes 10, 425-455.

--; and Slobin, Dan (1982). Rules and schemas in the development and use of the English past tense. Language 58, 265-289.

Caselli, Maria Christina; Bates, Elizabeth; Casadio, Paola; Fenson, Judi; Fenson, Larry; Sanderl, Lisa; and Weir, Judy (1995). A cross-linguistic study of early lexical development. Cognitive Development 10, 159-199.

Cazden, Courtney (1968). The acquisition of noun and verb inflections. Child Development 39, 433-448.

Choi, Sonja; and Gopnik, Alison (1995). Early acquisition of verbs in Korean: a cross-linguistic study. Journal of Child Language 22, 497-530.

Conti-Ramsden, Gina (1990). Maternal recasts and other contingent replies to language-impaired children. Journal of Speech and Hearing Disorders 55, 262-274.

--; and Friel-Patti, Sandy (1983). Mothers' discourse adjustments to language-impaired and non-language-impaired children. Journal of Speech and Hearing Disorders 48, 360-367.

--; and Jones, Melanie (1997). Verb use in specific language impairment. Journal of Speech and Hearing Research 40, 1298-1313.

Dromi, Esther (1987). Early Lexical Development. Cambridge: Cambridge University Press.

Edwards, Susan; Fletcher, Paul; Garman, Michael; Hughes, Arthur; Letts, Carolyn; and Sinka, Indra (1997). The Reynell Developmental Language Scales III: The University of Reading Edition. Windsor: NFER-NELSON.

Eyer, Julia; and Leonard, Laurence (1994). Learning past tense morphology with specific language impairment: a case study. Child Language, Teaching and Therapy 10, 127-138.

Goldfield, Beverly (1993). Noun bias in maternal speech to one-year-olds. Journal of Child Language 20, 85-99.

Hsieh, Li; Leonard, Laurence; and Swanson, Lori (1999). Some differences between English plural noun inflections and third singular verb inflections in the input: the contribution of frequency, sentence position and duration. Journal of Child Languag 26, 531-543.

Jones, Melanie; and Conti-Ramsden, Gina (1997). A comparison of verb use in children with SLI and their younger siblings. First Language 17, 165-193.

Leiter (1979). Leiter International Performance Scale. Chicago: Stoelting.

Leonard, Laurence; Bortolini, U.; Caselli, Maria Christina; McGregor, K.; and Sabbadini, L. (1992). Morphological deficits in children with specific language impairment: the status of features in the underlying grammar. Language Acquisition 2, 151-179.

--; and Eyer, Julia (1996). Surface properties of grammatical morphology and morphological deficits in children with specific language impairment. In Signal to Syntax, J. Morgan and K. Demuth (eds.), 233-247. Hillsdale, NJ: Erlbaum.

--; Eyer, Julia; Bedore, Lisa; and Grela, Bernard (1997). Three accounts of the grammatical morpheme difficulties of English-speaking children with specific language impairment. Journal of Speech, Language and Hearing Research 40, 741-753.

McGregor, K.; and Allen, G. (1992). Grammatical morphology and speech perception in children with specific language impairment. Journal of Speech and Hearing Research 35, 1076-1085.

Lieven, Elena; Pine, Julian; and Baldwin, Gillian (1997). Lexically-based learning and early grammatical development. Journal of Child Language 24, 187-219.

MacWhinney, Brian (2000). The CHILDES Project: Tools for Analysing Talk. Mahwah, NJ: Erlbaum.

Maratsos, Michael (2000). More overregularizations after all: new data and discussion on Marcus, Pinker, Ullman, Hollander, Rosen and Xu. Journal of Child Language 27, 183-212.

Marchman, Virginia (1997). Children's productivity in the English past tense: the role of frequency, phonology and neighborhood structure. Cognitive Science 21, 283-304.

--; and Bates, Elizabeth (1994). Continuity in lexical and morphological development: a test of the critical mass hypothesis. Journal of Child Language 21, 331-366.

--; Wulfeck, Beverly; and Ellis Weismer, Susan (1999). Morphogical productivity in children with normal language and SLI: a study of the English past tense. Journal of Speech, Language and Hearing Research 42, 206-219.

Marcus, Gary; Pinker, Steven; Ullman, Michael; Hollander, Michelle; Rosen, T.; and Xu, Fei (1992). Overregularization in Language Acquisition. Monographs of the Society for Research in Child Development 57(4), serial no. 228.

Messick, C.; and Prelock, P. (1981). Successful communication: mothers of language-impaired children versus mothers of language-normal children. Paper presented at the Convention for the American Speech-Language-Hearing Association, Los Angeles.

Montgomery, James; and Leonard, Laurence (1998). Real-time inflectional processing by children with specific language impairment: effects of phonetic substance. Journal of Speech, Language and Hearing Research, 41, 1432-1443.

Naigles, Letitia; and Hoff-Ginsberg, Erika (1998). Why are some verbs learned before other verbs? Effects of input frequency and structure on children's early verb use. Journal of Child Language 25, 95-120.

Oetting, Janna; and Horohov, Janice (1997). Past tense marking by children with and without specific language impairment. Journal of Speech, Language and Hearing Research 40, 62-74.

Pine, Julian; and Lieven, Elena (1997). Slot-and-frame patterns and the development of the determiner category. Applied Psycholinguistics 18, 123-138.

--; Lieven, Elena; and Rowland, Caroline (1998a). Tense optionality and children's use of English verb morphology: testing the optional infinitive hypothesis. Paper presented at the Child Language Seminar, Sheffield.

--; Lieven, Elena; and Rowland, Caroline (1998b). Comparing different models of the development of the English verb category. Linguistics 36, 807-830.

Pinker, Steven; and Prince, Alan (1988). On language and connectionism: analysis of a parallel distributed processing model of language acquisition. Cognition 28, 59-108.

Plunkett, Kim; and Marchman, Virginia (1991). U-shaped learning and frequency effects in a multi-layered perceptron: implications for child language acquisition. Cognition 38, 43-102.

--; and Marchman, Virginia (1993). From rote learning to system building: acquiring verb morphology in children and connectionist nets. Cognition 48, 21-69.

Prasada, Sandeep; and Pinker, Steven (1993). Generalizations of regular and irregular morphological patterns. Language and Cognitive Processes 8, 1-56.

Rice, Mabel; and Wexler, Kenneth (1996). Towards tense as a clinical marker of specific language impairment in English-speaking children. Journal of Speech, Language and Hearing Research 39, 1239-1257.

--; Wexler, Kenneth; and Cleave, Patricia (1995). Specific language impairment as a period of extended optional infinitive. Journal of Speech and Hearing Research 38, 850-863.

--; Wexler, Kenneth; and Hershberger, Scott (1998). Tense over time: the longitudinal course of tense acquisition in children with specific language impairment. Journal of Speech, Language and Hearing Research 41, 1412-1431.

--; Wexler, Kenneth; Marquis, Janet; and Hershberger, Scott (2000). Acquisition of irregular past tense by children with specific lnguage impairment. Journal of Speech, Language and Hearing Research 43, 1126-1145.

Rowland, Caroline; and Pine, Julian (2000). Subject-auxiliary inversion errors and wh-question acquisition: "what children do know?" Journal of Child Language 27, 157-181.

Stemberger, Joseph (1993). Vowel dominance in overregularizations. Journal of Child Language 20, 503-521.

Tardif, Twila (1996). Nouns are not always learned before verbs: evidence from Mandarin speakers' early vocabularies. Developmental Psychology 32, 492-504.

Theakston, Anna; Lieven, Elena; Pine, Julian; and Rowland, Caroline (2001). The role of performance limitations in the acquisition of verb-argument structure. Journal of Child Language 28, 127-152.

Tomasello, Michael (1992). First Verbs: A Case of Early Grammatical Development. Cambridge: Cambridge University Press.

--; and Brooks, Patricia (1999). Early syntactic development: a construction grammar approach. In The Development of Language, M. Barrett (ed.). Sussex: Psychology Press.

Wijnen, Frank; Kempen, Masja; and Gillis, Steven (2001). Root infinitives in Dutch early child language: an effect of input? Journal of Child Language 28, 629-660.

Received 20 March 2001

Revised version received 19 November 2001