THE PUZZLE

Some child English speakers produce nonnominative pronouns in subject position, as has been noticed at least since Gruber (Reference Gruber1967), illustrated by the following “b” subentries examples, from the speech of Mackie, between the ages of 2 years, 2 months (2;2) and 2;5 (Gruber, p. 53).

1. (1) a. I wanna.

2. b. Me wanna truck.

3. (2) a. He take the wheel, fire engine.

4. b. Me take the wheel.

5. (3) a. I show you?

6. b. Me show you?

However, the optionality of the nonnominative case marking on these pronouns, illustrated by the “a” subentry examples taken from the same recording sessions, does not appear to be random but rather appears to follow a pattern. Such a pattern is illustrated in Table 1, in data from seven of the eight typically developing children reported on in Loeb and Leonard's Reference Loeb and Leonard1991 study, compiled by Schütze (Reference Schutze1997, their table 6, p. 222).Footnote ¹

Table 1.

Finiteness versus case for seven of Loeb and Leonard's (Reference Loeb and Leonard1991) normal subjects

As the second column of the table shows, nonnominative subject pronouns him and her occur rarely with “inflected” or finite verbs in the speech of typically developing English-speaking children, while nominative case pronouns he and she may occur with either inflected or uninflected verbs. A range of empiricist and nativist theoretical proposals have been put forth to explain these patterns. The main goal of this paper is provide a different kind of empirical evidence for evaluating these theories.

CASE THEORETIC APPROACHES AND THE AGREEMENT TENSE OMISSION MODEL (ATOM)

For those familiar with generative syntactic theory, the first explanation for this pattern that likely leaps to mind is the connection between verbal finiteness marking and subject (nominative) case, proposed in Rouveret and Vergnaud (Reference Rouveret and Vergnaud1980), formalized by Chomsky (Reference Chomsky1981) as the case filter, and further developed as case theory by, for example, Marantz (Reference Marantz1991), Chomsky and Lasnik (Reference Chomsky, Lasnik, Halbband and Vennemann1993), Harley (Reference Harley1995), and Jonas (Reference Jonas1995), among many others. This conclusion has been pursued by multiple researchers working in the generative developmental linguistics tradition (e.g., Aldridge, Reference Aldridge1989; Guilfoyle & Noonan, Reference Guilfoyle and Noonan1992; Radford, Reference Radford1990; Vainikka, Reference Vainikka1993–1994). The fundamental idea is that a grammatical property of all noun phrases requires that they must receive abstract case in order to occur. In this theory, the way that subject noun phrases, including pronouns, receive abstract case is by virtue of their relationship with the finite properties of the verb (tense and/or agreement) with which they co-occur. In this way, case theory attempts to account for patterns of subject occurrence such as the following:

1. (4) He left early.

2. (5) * Him left early.

3. (6) For him to leave early would be a good idea.

4. (7) * For he to leave early would be a good idea.

By case theory, (4) is grammatical because the nominative case subject co-occurs with the finiteness features of the finite verb “left.” In contrast, (5) is ungrammatical because the nonnominative case pronoun him does not need to co-occur with a finite verb for case checking, but rather must co-occur with a preposition, which can check its oblique case features, as in (6). Notice that in (6), the nonfinite verb to leave would appear to lack the finiteness features necessary to make a nominative subject, such as he, grammatical. As a consequence, the sentence is only grammatical because the preposition for checks the oblique case features of the oblique case pronoun him. In (7), the nominative case pronoun he cannot get its nominative case features checked, because it is not in a structural relationship with a finite verb, resulting in the ungrammaticality of the sentence.

In summary, case theory was designed to explain, among other things, the co-occurrence patterns of nominative case subjects with finite verbs, as in (4), and the occurrence of nonnominative subjects with nonfinite verbs, as in (6). Because child English speakers pass through a stage known as the optional infinitive stage (cf., Wexler, Reference Wexler, Lightfoot and Hornstein1994, Reference Wexler1998), in which children optionally mark finiteness on verbs, generative syntactic theory predicts that children's grammars should require nominative case pronouns when they produce finite verbs, but not require them when they produce nonfinite verbs.

This does not explain, however, why nonnominative pronouns should be able to occur with nonfinite verbs, in the absence of a preposition. To answer this question, for situations in which pronouns occur without an obvious case assigner in adult English, Akmajian (Reference Akmajian1984) suggests that languages have a default case that is assigned when no case assigner is present. Researchers have taken this to mean that in the absence of a case assigner, such as verb finiteness or a preposition, adults will produce pronouns in this default case. Examples are usually given from pronouns that occur in isolation or with no potential adjacent case assigner.

Examples such as (8a) are sometimes referred to as “Mad Magazine Sentences.” These contain no finiteness marker on the verb and are most similar to child English nonfinite sentences. There are other contexts as well, however, that lack any obvious case assigner, such as those illustrated in the sentences in (9) and (10). Notice that default case can vary across languages, and while the default appears to be the accusative in English, it appears to be the nominative in, for example, Spanish,Footnote ² as illustrated in the “b” subexamples in the following sentences.

1. (8) a. What, me worry?

2. b. ¿Qué? ¿ Yo preocuparme?

3. (9) a. Who took the cookies from the cookie jar? Me!

4. b. ¿Quién se llevó las galletas? ¡Yo!

5. (10) a. Her in a karaoke bar is what we have to avoid.

6. b. Ella en un bar de karaoke es lo que tenemos que evitar.

Part of Schütze and Wexler's (Reference Schütze and Wexler1996) ATOM hypothesis, then, is that default case steps in to assign default accusative case to subjects, when finiteness is not sufficient (i.e., does not contain agreement, for ATOM) to get the job done.

This, of course, raises the question of what a finite and a nonfinite verb look like in child language. Many constructivist studies, for example, do not assume that verbs marked for finiteness are necessarily finite in the adult sense (e.g., Pine, Lieven, & Rowland Reference Pine, Lieven and Rowland1998; Rojas-Nieto Reference Rojas-Nieto and Grinstead2009), but rather may be “frozen forms” or holophrases. Child language researchers working in the generative paradigm have made similar suggestions (e.g., Sano & Hyams, Reference Sano and Hyams1994). ATOM makes a particular set of claims about what a finite verb consists of by virtue of its assumptions regarding the correspondence of syntactic case-related finiteness features to morphological features, illustrated in Table 2.

Table 2.

Possible correspondences between case-related syntactic features and morphological representations of pronouns

Note: Compiled from Schütze and Wexler (Reference Schütze and Wexler1996, pp. 678–679).

The highlights of the ATOM theory could be summarized as follows: it assumes that finiteness and case are linked, following case theory, that some finiteness markers represent agreement while others represent tense and that the presence of third-person singular –s marking on a verb makes co-occurrence with a nonnominative subject impossible. In response to Schütze and Wexler's (Reference Schütze and Wexler1996) proposal, a series of constructivist studies have attempted to empirically falsify it, fundamentally by attempting to show counterevidence to the third premise of ATOM, that third singular –s may not occur with nonnominative pronominal subjects.

SPONTANEOUS PRODUCTION DATA, THIRD-PERSON SINGULAR –S AND NONNOMINATIVE PRONOMINAL SUBJECTS

According to most of the constructivists who have argued against ATOM (Ambridge & Pine, Reference Ambridge and pine2006; Pine, Joseph, & Conti-Ramsden, Reference Pine, Joseph and Conti-Ramsden2004; Pine, Rowland, Lieven & Theakston, Reference Pine, Rowland, Lieven and Theakston2005), the main obstacle to showing counterevidence to ATOM's claim, that nonnominative subject pronouns should not occur with finite verbs, seems to be that not all children produce enough nonnominative subject pronouns in the spontaneous data they consider to be able to test the claim. Pine et al. (Reference Pine, Joseph and Conti-Ramsden2004), for example, examine the transcripts of four English-speaking children with specific language impairment, from the Manchester SLI Corpus (Joseph, Serratrice, & Conti-Ramsden, Reference Joseph, Serratrice and Conti-Ramsden2002), to determine whether any of the children use nonnominative subject pronouns with third-person singular verbs significantly less than chance. One child (Bonnie) uses no nonnominatives at all and is excluded. Another (Harry) uses fewer than chance, but he uses very few overall tokens of nonnominative pronouns, even with nonagreeing verbs (n = 3), so his data is also not informative. The two children who do use some number of nonnominative pronouns with third-person singular verbs, Dan (n = 2) and Nathan (n = 10), actually use them significantly less than chance, as ATOM would predict. Looking in greater detail, however, Pine et al. show that the nonnominative pronouns that occur with finite verbs in the speech of Nathan and Dan both occur with uncontracted verb forms.

Pine et al. (Reference Pine, Rowland, Lieven and Theakston2005) examine the data of the 12 children from the Manchester Corpus (Theakston, Lieven, Pine, & Rowland, Reference Theakston, Lieven, Pine and Rowland2001), to test a similar question and find that only 3 of them produce enough nonnominative case subject pronouns to be considered and that they use them at rates lower than 10%, as ATOM would predict. They then show, however, that while masculine nonnominatives with third-singular present verbs are not used significantly above 10%, feminine nonnominative pronouns are, for Gail (9/23 = 39.1%), Becky (13/39 = 33.3%), and Anne (4/12 = 33.3%), which they take to be counterevidence to ATOM. As in Pine et al. (Reference Pine, Joseph and Conti-Ramsden2004), they also find that two of the children use nonnominative pronouns with uncontracted verbs significantly more than 10%, for Anne (3/4 = 75%) and Becky (6/19 = 33.3%).

Ambridge and Pine (Reference Ambridge and pine2006) use an elicited imitation task to test children's accuracy at repeating nominative case pronouns with agreeing verbs to determine if they will produce more than 10% with nonnominative case pronouns and agreeing verbs. Of the 24 children they initially test, 5 produce nonnominative case pronouns with agreeing verbs. The remaining 19 produce no nonnominative case pronouns with agreeing verbs, which would appear to constitute support for ATOM. Ambridge and Pine, however, continue to test the 5 children who did produce nonnominatives with agreeing verbs with the rationale that only those children who produce nonnominatives can be further tested for nonnominative subject pronoun use with third-singular verbs. On this variant of the same test used for screening, they find that three of the five children use more nonnominative pronouns with agreeing verbs than predicted by chance, which they take to be evidence against ATOM.

It is interesting to note that the choice to work only with the children who produced nonnominative pronoun subjects in the screening procedure would seem to imply that the other 19 children (79%) did not have a grammatical option to use nonnominatives with agreeing verbs. A similar tacit assumption would appear to be at work in Pine et al. (Reference Pine, Rowland, Lieven and Theakston2005), where the longitudinal transcripts of 9 of the 12 children (75%) could not be studied because their spontaneous data did not contain sufficient (or any) tokens of nonnominative subject pronouns. This assumption fundamentally seems to imply that because a child does not produce a form in a set of spontaneous production recording sessions, that child cannot produce the form because he or she lacks the grammatical competence that underlies it. We will argue that this assumption and its implication are unwarranted.

Further, we note that a consequence of choosing the spontaneous production methodology is that the inferences that can be drawn from their conclusions are limited by the small number of subjects whose data is considered and by the small numbers of relevant tokens collected. Kirjavainen, Theakston, and Lieven (Reference Kirjavainen, Theakston and Lieven2009), for example, argue that children's production of finite verbs + nonnominative pronouns is a lexically dependent phenomenon that results from children hearing adults say things like “Let me do it,” which the children transform into “Me do it” in their production. Kirjavainen et al. demonstrate a degree of input dependence in the spontaneous production data of 17 children, studying their production as well as that of their adult interlocutors. In this study, 17 children are studied, but 4 of them produce no nonnominative subjects whatsoever. However, it was even more problematic that 14 of the 17 use 5 nonnominative subjects or less, leaving only 3 who use relatively larger numbers of the construction of interest (18, 30, and 211 nonnominative subjects). Thus, only 3 children in the sample provide substantial amounts of data. We will attempt to circumvent this limitation of spontaneous production data for studying this question by asking a larger number of children (N = 49) for their grammaticality judgments of nonfinite versus finite verbs and of nominative versus nonnominative subject pronouns. Finally, a common conclusion to the constructivist studies is that nonnominative forms appear to occur with uncontracted verbs but not with contracted forms. Because the contracted forms seem more likely to be lexicalized constructions, with uncertain finiteness properties, we will limit our experimental sentences to uncontracted verbs.

PRONOMINAL PARADIGMS, FINITENESS, AND CASE

Matthew Rispoli offers a different proposal for why children produce nonnominative pronouns and what this might have to do with finiteness marking, in a series of studies (Rispoli Reference Rispoli1994, Reference Rispoli1998a, Reference Rispoli1998b, Reference Rispoli1999, Reference Rispoli2005). In early work (e.g., Rispoli, Reference Rispoli1994), he proposes the paradigm building hypothesis, which hypothesizes that children fail to produce nominative case pronouns in subject position because they have not yet developed a complete morphological representation of the pronoun paradigm. The basic idea is that because the pronoun her occurs twice in the pronoun paradigm, as a genitive and as an accusative, it will be more likely to occur when children make errors. On its face, the proposal has merit in that there is evidence from adult language processing that shows that adults treat, for example, nominative versus nonnominative case pronouns differently (Feldman & Fowler, Reference Feldman and Fowler1987; Kostic, Reference Kostic and Feldman1995; Lukatela, Carello, & Turvey, Reference Lukatela, Carello and Turvey1987), which suggests that different “cells” in the pronoun paradigm may have different properties. In particular, nominatives are processed more quickly than nonnominatives. That pronouns of different cases have been shown to behave differently in adult English (and German), together with the traditional assumption in theoretical morphology that morphological entries are most likely structured in paradigms (e.g., Aronoff, Reference Aronoff1994; Stump, Reference Stump1993; Zwicky, Reference Zwicky1985), lends plausibility to this proposal.

Rispoli (Reference Rispoli1994) focuses on the internal dynamics of how the pronoun paradigm is learned and says little about the role of finiteness in the explanation of pronoun case phenomena. In later work, however, he addresses this possibility directly. In Rispoli (Reference Rispoli1999), it is demonstrated that subject pronominal case marking proficiency and verb agreement proficiency correlate in spontaneous production. It is interesting that failures to mark third-singular present agreement marking (regular third-singular -(e)s, irregular has, does, says, and auxiliary and copular is) are shown to correlate with masculine errors (him for he), but not with feminine errors (her for she). Although these findings seem to corroborate those of Loeb and Leonard (Reference Loeb and Leonard1991) and Schütze and Wexler (Reference Schütze and Wexler1996), the agreement measure included contracted auxiliaries, which could have impacted the results. In particular, since Pine and colleagues have shown that nonnominative errors occur mostly with uncontracted auxiliaries (though this was not known at the time of publication of Rispoli, Reference Rispoli1999), it seems likely that an actual grammar is not consistently reflected in contracted auxiliary constructions, but rather that they are unitary, grammaticalized lexical units, as suggested by Tomasello (Reference Tomasello2003), Culicover and Nowak (Reference Culicover and Nowak2003), and others. The effect of including these items in the measure of agreement could have been to inflate this variable. In short, while Rispoli (Reference Rispoli1999) appears to give evidence of a correlation between finiteness and case, the inclusion of possibly holophrastic contracted constructions raises doubts about the nature of this evidence.

In Rispoli (Reference Rispoli2005), however, likely contractions are not included in the measure of finiteness. In this study, it is shown that finiteness as well as SDpro, a measure of the dispersion of attempts at producing pronominal forms within the paradigm, correlate with case errors. He hypothesizes that both the development of finiteness and the development of pronominal case paradigms determine children's case error rates, and that they do so interactively. Specifically, he shows that at lower levels of finiteness proficiency, below 60% correct, the children who make the most errors are those who have the lowest SDpro ratings, which means that they were the ones who were attempting to use the widest variety of pronouns. Children with scores above 60% correct for finiteness marking made levels of case errors that were unrelated to the dispersion of their variety of pronoun attempts (i.e., their SDpro score). What this seems to imply is that above a certain level of finiteness proficiency, only finiteness matters, a conclusion consistent with case theory, while below it, both finiteness and the lexical learning involved with constructing a pronominal paradigm in the lexicon matter.

In summary, Rispoli's earlier work was chiefly concerned with illuminating the nature of the relationship within the lexicon of pronouns of different cases. In later work, he considers the relevance of finiteness for pronoun case errors and finds an interaction. This work will serve as a starting point for our study in that we will concern ourselves primarily with children who are above the 60% correct level on finiteness marking. Working with children at this stage of development should allow us to see the clearest interaction of case with finiteness, without the possible confound of lexical development of the English pronominal paradigm. If Rispoli is correct, we should find no lexically specific effects at this level of finiteness proficiency, such as the gender asymmetry in pronominal errors (i.e., more feminine than masculine) that he has consistently observed in children with lower levels of verb finiteness proficiency. Further, we take seriously the finding of Pine and colleagues to the effect that nonnominatives occur with uncontracted auxiliaries and not with contracted expressions, which are likely to be holophrastic, at least some of the time. We will address this in our experiments by only using uncontracted forms. In addition, following insights gleaned from the work of Pine and colleagues, we will not use spontaneous production data, as many children do not happen to produce nonnominative case pronouns in recording sessions. We depart from Pine and colleagues by assuming that though children's productive language may not include case errors in particular samples, their grammars may still produce them and they may still judge them to be grammatical receptively. Instead, we will use the receptive grammaticality choice task (GCT) of Pratt and Grinstead (Reference Pratt, Grinstead, Belikova, Meroni and Umeda2007), which allows a larger sample of children to give their judgments of the grammaticality of the constructions in question.

RESEARCH QUESTIONS

1. 1. The first question we ask is whether the contingency between finite verbs and nominative case pronouns observed by Loeb and Leonard (Reference Loeb and Leonard1991), Schütze and Wexler (Reference Schütze and Wexler1996), Wexler, Schütze and Rice (Reference Wexler, Schütze and Rice1998), and Rispoli (Reference Rispoli1999, Reference Rispoli2005) in spontaneous production can be cross-validated in a receptive task, with children with levels of finiteness marking over 60% correct, which, as Rispoli has shown, appears to make knowledge of the pronoun paradigm irrelevant to the determination of case error rates.

2. 2. If there is a threshold of finiteness marking above which knowledge of pronoun paradigms does not matter, then what is the relationship of masculine and feminine gender pronoun errors above this threshold? Are they asymmetrically feminine, as they are below this level, or does the “double cell” effect, observed by Rispoli (Reference Rispoli1998a, Reference Rispoli1998b, Reference Rispoli1999), Pine et al. (Reference Pine, Joseph and Conti-Ramsden2004), and Pine et al. (Reference Pine, Rowland, Lieven and Theakston2005) disappear?

3. 3. Last, we ask whether receptive tasks show evidence that children accept verb forms inflected for third-person singular –s with nonnominative pronouns, contra ATOM.

VERB FINITENESS MARKING AND PARADIGM REGULARITY

Rice, Wexler, and Hershberger's (Reference Rice, Wexler and Hershberger1998) and Rice, Wexler, and Redmond's (Reference Rice, Wexler and Redmond1999) longitudinal studies of typical and atypical child English finiteness marking demonstrated, through spontaneous production and elicited production data and receptively through traditional grammaticality judgment tasks, that multiple markers of finiteness have correlated growth curves between roughly 3 and 6 years of age. Although the growth of these markers, including third singular –s, regular past tense –ed, have, do, as well as copula and auxiliary be, is correlated, differing degrees of proficiency are found for each marker.

The finding that their growth correlates, even with varying degrees of proficiency for each marker, makes sense if one assumes that a single underlying knowledge construct, tense, is largely responsible for their growth, while other factors involved in their use, such as processing and memory demands as well as the degree of regularity of the paradigm, scatter the resulting proficiencies, though in consistent proportions, to different levels. Along these lines, Yang (Reference Yang2002) argues that the regularity of morphological paradigms should matter for how quickly children learn to use them. For example, past tense –ed is more regular than auxiliary be, in that it is simply added to all regular past tense verbs. To use auxiliary be, however, the child must memorize the five forms of the verb be (is, am, are, was, were), which are completely irregular, and additionally must add –ing as a discontinuous morpheme to the lexical verb that it co-occurs with, which may be separated from the form of be by negation, adverbs, and other intervening constituents.

1. (11) He walk-ed fast.

2. (12) He talk-ed fast.

3. (13) He is walk-ing fast.

4. (14) He is not walk-ing fast.

5. (15) He is almost always walk-ing fast.

6. (16) He is really never quite walk-ing fast.

These kinds of differences explain differing levels of proficiency for each morpheme type, but they do not predict that the different levels of proficiency should not correlate. As mentioned, it has been demonstrated that they do correlate, though at different levels of proficiency, a finding that we will attempt to validate in our first study.

EXPERIMENT 1: THE VERB FINITENESS TASK

In this experiment, our goal is to determine the degree to which children judge sentences with verbs ending in either past tense –ed or with auxiliary be to be grammatical. All sentences contain a subject pronoun and all subject pronouns are in the nominative case. In this way, subject pronominal case is held constant in the experiment, as verb finiteness will be in the next experiment. We note that although we could have used full noun phrases instead of pronouns (e.g., the boy, instead of he), doing this would have introduced another variable relating to frequency in the input of these nouns, which is controlled for here by using the same subjects in both the verb finiteness task and the pronoun case task.

Participants

The participants in this task were 62 monolingual child speakers of midwestern US English in Columbus, Ohio. African American English and varieties of English spoken in the southern Ohio region of Appalachia, which are spoken in Columbus, Ohio; were not spoken by the children in the daycare centers and preschools from which this sample was recruited, who were overwhelmingly European American, from upper middle-class families. Fifty-seven children passed the filler items (at least four out of five correct), described below, which ensured that they understood the format, were paying attention and were not answering according to a number of strategies we observed, detailed below. Fifty-two children passed these fillers on both experiments reported here, and their mean age was 4;7 with an age range of 3;6–5;9 (SD = 7;52 months). The rest of the study will center on the results of these 52 children. All children were given each of the two tests on different days. Their distribution is illustrated in Figure 1, which shows that they are primarily 3- and 4-year-olds.

Figure 1.

Distribution of the children in the sample, given in age in months, by frequency.

It is important to note that the age range of the children who participated in these experiments begins at 3;6 (42 months) when Rice et al. (Reference Rice, Wexler and Hershberger1998) report that the mean composite tense score of the longitudinally studied 30 typically developing children in their language control group was just above 60% correct. This means that at this age, children are likely to not be at ceiling on a measure of verb finiteness, but are also likely to be above Rispoli's 60% correct threshold for knowledge of the pronominal case paradigm. As a consequence, there should be variation in their scores, which can be analyzed in a correlation test.

Procedures

Children were introduced to three puppets: a dog and a cat (who were both identified as male) and a turtle (who was identified as female). The animals were identified to the children as very young, and consequently as needing help to learn how to talk. It was emphasized that sometimes they would “talk like grown-up, big kids, like you” and that other times they would say things that were not right. For this reason, children were urged to pay close attention so that they could help the young animals learn to speak better by telling them who had said the sentence best.

On each item, children were shown a picture in which one of the three puppets appeared performing an action (Figure 2). Then, each puppet would utter a sentence relating to the picture, and the child was encouraged to decide who had said the sentence best. For example, after being shown the following picture, one puppet would say, “He kicked a ball.” Then, the other puppet would say, “He kick a ball.” The child was then asked who said it better. If children said that they were both good, they were told that they had to choose one or the other. Whether the first or second sentence was correct was counterbalanced and showed no effect (p > .05) and the animal that presented the correct sentence was also counterbalanced and showed no effect (p > .05). We observed these two strategies at play in pilot work and actively discouraged them in our warm-up items. Finally, the order of presentation within both present and past tense was counterbalanced and also showed no effect, for either present (p > .05) or past (p > .05) tense.

Figure 2.

Verb finiteness sample picture.

The task consisted of 6 practice items to introduce children to the format of the experiment, 16 experimental items, and 5 filler items. The filler and warm-up items tested constructions that have been shown to be mastered much earlier than the lower bound of our age range (cf. Rice et al., Reference Rice, Wexler and Redmond1999; McDaniel & Cairns, Reference McDaniel and Cairns1990), namely, noun plurals and –ing marking of present progressive participles, illustrated in (17) and (18).

1. (17) a. The turtle wants two cookies. versus *The turtle wants two cookie.

2. b. The dog had two books. versus *The dog had two book.

3. (18) a. The turtle is dancing. versus *The turtle is dance.

4. b. The turtle was using a pencil. versus *The turtle was use a pencil.

During the warm-up phase of the experiment, children were corrected if they chose the nonadult like noun plural or present progressive form and complimented if they chose the adult like form. No differential feedback of this kind was given after the warm-up period. Rather, they were complimented, regardless of correctness. The warm-up items were configured so that neither the first item nor the second item was always correct, nor was it the case that any particular animal was always correct. The absence of these patterns and the need to listen carefully to what each animal said was emphasized. The experimental items are described in Table 3. See Appendix A for the complete list of experimental items.

Table 3.

Example sentences from the verb finiteness task

Results

The overall mean percentage correct for the sample was 82.56% correct (“correct” will be used to indicate a choice of the adult like value). Further, the children did not appear to be guessing, as the mean correct (11.54) was statistically greater than chance (p < .001), which is 50%, or 8 items correct out of 16 possible in our design. The histogram in Figure 3 illustrates that the great majority of the children in the sample (48 of 52) scored above Rispoli's 60% correct (10 or more correct) threshold.

Figure 3.

The distribution of verb finiteness scores (100% correct = 16) by frequency.

The results of the verb finiteness task were submitted to an arcsine transformation, upon which a mixed effects analysis of variance (ANOVA) was performed with age and morpheme as fixed factors and child as a random factor. The results of this ANOVA were that there was a main effect of morpheme, f (1, 49) = 18.518, p < .001. There was no effect for age, f (2, 49) = 3.029, p = .058, and there was no interaction of age and morpheme, f (2, 49) = 0.112, p = .894. Pairwise comparisons showed that –ed was significantly better than auxiliary be (p < .001). Subscores for –ed and auxiliary be nonetheless correlated (r = .430, p = .001). The children's scores by age group are graphically represented in Figure 4.

Figure 4.

Mean correct judgments of auxiliary be, past –ed and combined verb finiteness items, across three age groups.

EXPERIMENT 2: THE PRONOUN CASE TASK

In this experiment, our goal was to determine the degree to which children judge sentences with feminine and masculine third-person, singular pronominal subjects, in nominative and accusative case, to be grammatical. It is important to note that nonnominative subject pronouns frequently do not manifest themselves in child English spontaneous production data. Our position, following case theory, is that they occur as a function of optional syntactic finiteness on verbs. Thus, nonnominative subject pronouns should be as acceptable as nonfinite verbs are, with the caveat that children, following Rispoli (Reference Rispoli2005), have learned the English pronominal paradigm. In this view, the reason that there is not a total correlation of the two constructions (nonfinite verbs and nonnominative subject pronouns) in spontaneous production data is that not all apparently finite verbs are actually finite. Some of them are holophrases, which appear to be finite, but they lack underlying syntactic finiteness features. This explains why researchers have been able to show that nonnominative subject pronouns sometimes occur with third-singular –s (Ambridge & Pine, Reference Ambridge and pine2006; Pine et al., Reference Pine, Joseph and Conti-Ramsden2004, Reference Pine, Rowland, Lieven and Theakston2005), contra the basic intuition of case theory and the specific predictions of ATOM. In sum, testing children's knowledge of pronominal subjects receptively allows us to overcome the empirical difficulty posed by the scarcity of nonnominative subjects in production data.

All sentences on the pronoun case task contain an uncontracted finite verb, marked with either copula be or third person singular –s, neither of which was used in the first experiment. In this way, verb finiteness is held constant and the lack of overlap between this experiment and the verb finiteness task allows us to test the two variables independently. An additional consideration pointed out in two studies by Pine et al. (Reference Pine, Joseph and Conti-Ramsden2004, Reference Pine, Rowland, Lieven and Theakston2005) is that nonnominatives occur with uncontracted verbs. We follow their lead and assume that any relationship between nonfinite verbs and nonnominative pronouns is best measured using uncontracted verbs. Only the latter are used in our experimental sentences. To our knowledge, this is the first time that a receptive task has been used to measure children's knowledge of nonnominative pronominal case.

Procedures

As in the previous experiment, GCT was used to assess children's judgments of pronominal case in subject position. Also as in the previous experiment, it consisted of 6 practice items, 16 experimental items, and 5 filler items. As before, correction was provided in the warm-up period, but none thereafter. Half of the 16 experimental items used third singular –s, whereas half used copula be. Similarly, half of the 16 items used a masculine pronoun (him vs. he), and half used a feminine pronoun (her vs. she). We again note that these finiteness markers are distinct from those used in the verb finiteness task, in order to maintain the independence of the case variable. Table 4 summarizes the item types used. See Appendix B for the complete list of experimental items.

Table 4.

Example sentences from the pronoun case task

Results

The overall mean percent correct for the sample was 79.43%. Further, the children were not guessing, as the mean correct (12.71) was statistically greater than chance (p < .001), which is 50%, or 8 items correct out of 16 possible, in our design. The results of the pronoun case task were submitted to an arcsine transformation, upon which a mixed effects ANOVA was performed with age and pronoun gender as fixed factors and child as a random factor. The results of this ANOVA showed there was no main effect of pronoun gender, f (1, 49) = 0.242, p = .625. There was an effect for age, f (2, 49) = 4.523, p = .016, and there was no interaction of age and pronoun gender, f (2, 49) = 1.821, p = .173. Pairwise comparisons showed that the scores of the 5-year-old group were significantly higher than those of the 3-year-old group (p = .004), but that there were no significant differences either between the 3-year-olds and the 4-year-olds (p = .078) or between the 4-year-olds and the 5-year-olds (p = .081). Subscores for masculine and feminine pronoun judgments correlated (r = .648, p < .001). The children's scores by age group are graphically represented in Figure 5.

Figure 5.

Mean correct judgments of he-him, she-her, and combined pronoun case items across three age groups.

In addition, in contrast to the findings of Pine and colleagues, the majority of the children in our sample accepted at least some nonnominative pronominal subjects as grammatical, as illustrated in Figure 6.

Figure 6.

The distribution of pronoun case scores (100% correct = 16) by frequency.

There were 16 experimental items on the pronoun case task. Figure 6 shows that 12 children made all 16 case judgments in an adult like fashion. It also shows that the remaining 40 (77%), of the 52 children who took the test, chose a nonnominative case pronoun as grammatical at least some of the time. Case errors, then, are widely distributed across a great majority of the participants in the study, contrary to what has apparently been assumed in the majority of studies of this phenomenon.

Turning to the main goal of the study, we find that children's judgments on the pronoun case task correlate with those of the verb finiteness task (r = .375, p = .006, N = 52). This effect is illustrated in Figure 7.

Figure 7.

Judgments of verb finiteness and pronoun case in subject position correlate (some data points overlap).

The alert reader will note that there are a small number of outliers in Figure 7. In correlational analyses, which are maximum likelihood statistics, even small numbers of outliers can be influential and mask important tendencies. Because there are myriad nonlinguistic factors (e.g., distractedness, lack of sleep, hunger, not feeling well) that could have produced the results of this small number of outliers, I will remove the three whose results are greater than 2 SD from the mean. Removing the results of these 3 children, leaving 49 children still in the sample, increases the correlation coefficient by almost .2 (r = .549, p < .001, N = 49). Thus, with the removal of a very small number of cases, an already significant relationship becomes much more so. Consequently, we will consider the results without the influential outliers to be most representative of the relationship between case and finiteness under consideration. Contrast the difference between Figure 7 and Figure 8.

Figure 8.

Judgments of verb finiteness and pronoun case in subject position correlate, with three influential outliers removed (some data points overlap).