Gender

Gender is a linguistic feature by which words in a given language are assigned to a class. This has consequences for the forms taken by other elements related to the word (Comrie, Reference Comrie1999) (e.g., nouns and their determiners or adjectives). French has two genders: masculine and feminine. Nouns are most often classified into either gender on a relatively arbitrary basis. For example chaise ‘chair’ is feminine, whereas banc ‘stool’ is masculine. As in most languages with gender, French shows some correlations with natural gender (i.e., inherent, as opposed to grammatical, gender). Not unexpectedly, femme ‘woman’ is feminine and homme ‘man’ is masculine. There are also correlations between formal properties of French nouns, usually final segments (consonants or syllables) and their gender, and these vary according to the segment (Desrochers et al., Reference Desrochers, Paivio and Desrochers1989). For example, a noun ending in [p], such as moppe ‘mop,’ has little predictive value as to its gender in French, while words ending in [ɔn], such as couronne ‘crown,’ have a strong chance of being feminine. In addition, most derivational suffixes have inherent gender. For example, the suffix euse meaning ‘doer of V’ (laveuse ‘washer(-woman),’ sécheuse ‘dryer,’ repasseuse ‘(flat)-iron’) derives a feminine noun and ends in the consonant -z. It is important to note that correlation research rarely distinguishes between word final consonant metrics and the presence of a derivational suffix.

Gender is realized on determiners (le/la ‘the.m/f’; un/une ‘a.m/f’; tout/e ‘all’), third person singular and plural pronoun clitics (il/elle ‘he/she’; ils/elles ‘they.m/f.’), possessive, demonstrative, indefinite and relative pronouns (e.g., le mien/la mienne ‘mine.m/.f’; celui-là/celle-là ‘that one.m/f’; aucun/e ‘no(ne).m/f’; lequel/laquelle ‘which.m/f’, etc.), a subset of adjectives (see below) and past participle forms of verbs (elle s'est assise/il s'est assis ‘she/he CL.AUX sat.m/f’).Footnote ¹ Gender is therefore a salient grammatical property of French, and, as such, a feature that must be acquired in order to master the language.

Theories of lexical acquisition

A number of language acquisition models have been proposed to account for the acquisition and processing of morphosyntactic and lexical structures. These models can generally be classified as rule- versus item-based. Rule-based models originate in linguistic theories of grammar and are believed to instantiate innate, modular, automatic, fast and obligatory processes. Examples of this type of model are proposed in Aronoff (Reference Aronoff1976), for lexemes, but more recent examples can be found. For instance, Stockall and Marantz (Reference Stockall and Marantz2006) proposed a model for the processing of inflected English verbs based on brain imaging experiments. Item-based models, on the other hand, ascribe relatively more importance to input in the development of linguistic representations. For example, Bybee and Slobin (Reference Bybee and Slobin1982) and Marchman and Bates (Reference Marchman and Bates1991, Reference Marchman and Bates1994) propose that the acquisition of inflectional paradigms, such as the English past tense, is linked to the high token frequency or high regularity of these patterns in the input. Finally, a number of hybrid models incorporating aspects of item- and rule-based approaches have been proposed, particularly in the domain of lexical learning and processing. These models generally maintain that both rule-based and item-based learning can occur at the same time. For example, Burani and Caramazza (Reference Burani and Caramazza1987) propose that known derived words are accessed whole by adult Italian speakers, while lower-frequency and newly learned words are decomposed into their morphemes during processing. Acquisition models generally hold that regular (sometimes, default) and semantically transparent structures can be morphologically analysed, but not when the structure is irregular (Clahsen and Rothweiler, Reference Clahsen and Rothweiler1992; Marcus et al., Reference Marcus, Pinker, Ullman, Hollander, Rosen and Xu1992, Reference Marcus, Brinkmann, Clahsen, Wiese and Pinker1995; Kim et al., Reference Kim, Marcus, Pinker, Hollander and Coppola1994; Pinker and Prince, Reference Pinker, Prince, Lima, Corrigan and Iverson1994; Royle, Reference Royle2007) or semantically opaque (Bertram et al., Reference Bertram, Schreuder and Baayen2000a; Reference Bertram, Laine and Virkkala2000b).

In this context, the study of French can address a number of aspects of language acquisition that are relevant to theories of lexical representation and language acquisition. In particular, French is a morphologically richer language than English in some respects, while presenting idiosyncrasies that make structures difficult to acquire. Royle (Reference Royle2007) showed that French-speaking children from Québec are sensitive to the morphological structure of verbs, distinguishing between regular and irregular verbs. French-speaking children produce more target past tense forms under prompting when presented with regular versus irregular verbs. This occurs on early-acquired as well as low-frequency verbs. Children also tend to overregularize irregular verbs, especially if they are low-frequency. In addition, some regularization occurs into the second conjugation (-ir), which has lower type and token frequency than the default –er conjugation in French. Royle concludes that French-speaking children are sensitive to verb conjugation groups at early ages and that paradigm frequency in the corpus does not affect French-speaking children's ability to recognize these patterns. St-Pierre (Reference St-Pierre2006) studied the processing of irregular verbs and adjectives with latent consonants in school-aged French-speaking children. St-Pierre (Reference St-Pierre2006) found that the production of these forms is highly mediated by their frequencies: final consonants are more correctly produced on high-frequency than low-frequency forms. This finding suggests that the acquisition of these forms should be described as item-based rather than rule-based, and that a ‘pattern’ for the production of consonant final alternations emerges from the input (this is postulated within a connectionist framework that disallows rules in the sense of Marcus's ‘operations over variables’, Reference Marcus2001). The data from French provided by Royle and St-Pierre seem to indicate that irregular verbs and adjectives are processed using mnesic abilities, whereas regular verbs (e.g., -er and -ir verbs) are processed using rules. We still do not know how children learn the masculine-feminine opposition governing French variable adjectives. Before we present data from French child corpora, we describe intra-nominal agreement and review proposals accounting for variable adjectives in French.

French DP and Intra-Nominal Agreement

The French DP has a number of interesting features, making it useful for the study of adjective acquisition. One is the presence of agreement operations (or concord) between nouns and adjectives (1).

1. 1)

Variable French adjectives have word-final consonants that appear on the feminine forms only (see examples in Table 1). These consonants are somewhat arbitrary and cannot generally be adduced based on the masculine forms, except for some forms ending in –n in the feminine, which have nasalized vowels in the masculine. However, not all masculine forms ending in nasalized vowels have n-final feminine forms, e.g., blanc [blɑ˜] – blanche [blɑ˜ʃ] ‘white.m/f’. These variable adjectives, and other French words involving word-final consonant alternations, have been described as having ‘floating’ or ‘latent’ consonants that are realized under specific conditions (see Paradis and El Fenne, Reference Paradis and El Fenne1995, for a review of different approaches).

Table 1.

Examples of variable adjectives in French

Within the framework of constraints and repairs, Paradis and El Fenne (Reference Paradis and El Fenne1995) propose that the best way to account for different word forms in French is to posit a consonant deletion process in the absence of a timing slot (for example, syllabification can occur before an empty onset, or in a feminine form). However, this analysis is disputed in Picard (Reference Picard1996) due to its post lexical nature. An illustration of this is presented in 2b, where the masculine petit /pət^si/ becomes [pət^sit] before an onsetless syllable in the following word (this is termed liaison).

1. 2)

More recent optimality-theoretic approaches have accounted for consonant/Ø alternations in languages such as French. Tranel (Reference Tranel2000) suggests an explanation of this type for liaison; however, he leaves open the question of whether these consonants are ‘borrowed’ from the feminine form or whether they are of a completely different nature (some adjectives, for example grand ‘big.m’, have a different consonant for the feminine grande [ɡʀɑ˜d] and for masculine liaison grand [ɡʀɑ˜t]).Footnote ²

The fact remains that a morphological description of the transition from masculine (the defaultFootnote ³ in French, Fink, Reference Fink1985; Nelson, Reference Nelson2005) to feminine form is difficult and somewhat arbitrary. We could posit that there is actually no productive morphological rule for adjective gender marking in French, and there is some evidence to support this. First, in terms of linguistic productivity, no overregularizations into this pattern have been observed (ex. étanche →*étanc ‘water-/air-tight’, by analogy with blanche → blanc ‘white’). Furthermore neologisms and borrowings are systematically integrated into the invariable pattern (e.g., hot, laïc ‘secular’) and not into the consonant/Ø pattern (Herschensohn, Reference Herschensohn1993). Finally, experiments on neologism processing show that French speakers do not insert or delete final consonants, but instead tend to preserve the presented forms (e.g., bravais/e) (89% of responses) (Fink, 1986). The first two facts are linguistic indicators of a default rule (Wurzel, 1984, Marcus et al., Reference Marcus, Brinkmann, Clahsen, Wiese and Pinker1995, Dressler, 1997), and Fink's experimental data support that this default rule implies no change. Based on this interpretation, we could argue that feminine forms are not constructed via morphologically-based inflectional processes, but instead are stored as such in the lexicon in the same way as irregular and suppletive verb forms.

Corpora studies

An analysis of the CHILDES corpus for a French-speaking child from Montreal (De Cat and Plunkett, Reference De Cat, Plunkett, Pusch and Raible2002) reveals the use of adjectives in complex DPs with feminine forms (e.g., petite ‘small.f’) starting around age 2 years and 3 months. A study of noun-drop production in DPs in spontaneous speech samples of French-speaking children aged 2 to 3 years (N = 14) reveals that French-speaking children produce feminine forms of variable attributive adjectives as young as 26 months (2 years and two months) (Valois et al., Reference Valois, Royle, Bourdua-Roy and Sutton2009). Of the 103 instances of attributive adjectives produced in the corpus, 62 are variable. Of these, 41 are in feminine DPs, and only one is erroneously produced in its masculine form. This is not to say that French-speaking children completely master the French gender agreement system at these ages.Footnote ⁴ A verification of the Valois et al. corpus reveals a number of difficulties. Errors can be seen in the production of predicative adjectives (e.g., in the constructions elle est jaune, belle et *brun ‘it.f is yellow, beautiful.f and brown.m.’, 2:2), and singular determinersFootnote ⁵ (e.g., *une café ‘a/one.f coffee’, 2:1, or *le maman ‘the.m mother’, 2:0). Errors on attributive adjectives also occur (in 5 of 11 feminine variable forms, or 45.5%, and 2 of 26 masculine variable forms, or 7.7%). However, these structures are not very common in the language samples, as only 69 attributive adjectives were produced overall (48 masculine and 21 feminine, including invariable forms). Some children do not produce these structures at all, and of those who do, not all produce forms demanding the feminine (a total of 1,904 utterances are analyzed in the corpus). A subsequent study of N-Drop taken from the longitudinal Pauline corpus (Bassano and Maillochon, Reference Bassano and Maillochon1994) in CHILDES (MacWhinney and Snow, 1990) reveals similar behaviour. Valois and Royle (Reference Valois and Royle2009) show that, although Pauline produces feminine variable adjectives, she does not produce many types, and she makes mistakes on these same types. Table 2 presents an overview of different variable adjectives produced by the children in the two studies.

Table 2.

First observations of variable adjective production in the Sutton and Pauline corpora and number of different children (out of a possible 14) using them in the Sutton corpus

As can be seen, petit/e, gros/se, grand/e and vert/e are the most commonly produced forms. However, except for petit, no form is produced by more than half of the children in the Sutton corpus. Therefore, there is not much variety in terms of variable adjectives within the corpus, and even less so for each given child (excluding Pauline's longitudinal corpus, only one child, aged 2;11, produced four different variable adjectives in the feminine). Moreover, Pauline makes errors on the earliest acquired forms (petite and blanche), which she subsequently produces correctly in the same recordings. Thus, the data in the spontaneous speech corpus do not clearly indicate productive agreement abilities. In addition, the age of mastery, and whether adjective concord is truly mastered, remain in dispute. It is arguable that the children are producing lexicalized chunks à la Tomasello (Reference Tomasello, Givon and Malle2002), or collocations, without checking agreement.

In order to address this issue, we designed an experiment to elicit size and colour adjectives. To date, no elicitation study of the acquisition course of this structure in pre-school-aged French has been undertaken. The study of DPs in French allowed us to add cross-linguistic data to the research on acquisition of concord. More importantly, we could verify whether the acquisition of concord is sensitive to features such as morphological regularity, as French is manifestly not regular in this respect. We expected this lack of regularity to negatively affect French-children's ability to produce appropriate forms of variable adjectives in elicitation experiments.

Elicitation studies

DPs are relatively easy to elicit (e.g., as opposed to VPs or clauses) in very young children. A pilot project to elicit variable adjectives revealed that very young children (aged 2;0 – 4;2, N = 16) can perform adjective production tasks for DPs in French, reliably naming colours at around age 2;4, and producing size and colour adjectives around the age of 2;6 (Royle et al., accepted). However, these children had not completely mastered all adjectives, especially feminine forms. In addition, because only variable adjectives were elicited by Royle et al., information on the effect of variability versus the presence of a word-final consonant on target production is lacking. Bedore and Leonard (Reference Bedore and Leonard2001) studied grammatical morphology in Spanish-speaking children (two groups aged 2;4–3;10, 4–5;6, N = 30). They elicited the production of target structures using picture description tasks (singular and plural items). Results show that Spanish-speaking children score between 88% and 94% correct on adjective agreement. The highest number of errors occurred on feminine plural forms, which tended to be replaced by masculine plural or feminine singular forms, e.g., rojas ‘red.f.pl’ → rojos ‘red.m.pl’ and roja ‘red.f.sg’. Errors on masculine adjectives were rare. A similar study was run on Swedish-speaking children (two groups, aged 2;1–3;7 and 4;3–5;7, N = 28) (Leonard et al., Reference Leonard, Salameh and Hansson2001). Children were asked to finish contrasting sentences of the type ‘This is a green house and this is . . . Target: a yellow car. Few gender errors were observed, with 14 gender errors on adjectives in the younger group and 11 in the older group. The most common errors involved definiteness (e.g., production of ett stort tåget for det stora tåget ‘the big.indef train’ for ‘the big.def train’), which must be marked on the determiner and adjective. However, the data is difficult to interpret, as total numbers of target and correct (non-target) production of adjectives are not presented.

More recently, Roulet-Amiot and Jakubowicz (Reference Roulet-Amiot and Jakubovicz2006) investigated the elicited production and processing of French gender, focussing on determiners and adjectives in the DP. This study included two groups, aged 4;4 – 4;11 and 6;4 – 6;11. They evaluated children's ability to produce appropriate determiners and pre- or postnominal attributive adjectives in a picture description task and verified interference effects of disagreement (e.g., un *nouvelle balai ‘a.m new.f broom’) in a semantic categorization task. Significantly more errors were observed on the overall production of feminine DPs, and masculine forms were produced as a default.Footnote ⁶ However, the older group did not show a masculine/feminine asymmetry in their errors and made few errors overall (5.2% on average). During the perception task, the experimenter gave the child some background context (e.g., ‘Now the monkey is going to the store where one can buy clothes, things we can wear. Do you think that in this store, it will be able to buy. . .’), and the child was then presented with one of three response types: grammatical, ungrammatical determiner (e.g., *une nouveau balai), and ungrammatical adjective (e.g., un *nouvelle balai). The child was asked to respond YES or NO by pushing a button. Reaction times (RTs) and error rates were recorded. Only the older group was able to perform semantic categorization. They showed an agreement effect, where children produce faster RTs on concordant stimuli than non-concordant ones. However, this effect was found only for the discordant determiner condition and did not reach significance for the adjective condition, although a trend was found. Error rates in the 6-year-olds were similar to those found in adults (5.8% vs. 4.1%). These data show that mastery of DP agreement emerges sometime between 4 and 6 years old in French-speaking children, and that definite determiners are reliably produced with appropriate gender at 4 years in French. However, the data on adjectives does not show such robust effects.

The above-presented experiments were run to investigate the nature of specific language impairment in Spanish, Swedish and French. The focus of the analyses was therefore on impaired language, with control children serving as benchmarks for normal language acquisition. Little information is given about their production abilities, overregularization patterns (if any), alternate production strategies, and so forth. In addition, the picture description and contrastive approaches used to elicit complex DPs in these children raise the possibility that the child focussed on some aspect of the drawing that was not relevant to the study, and thus produced non-target adjectives.Footnote ⁷ We therefore decided to design a controlled elicitation task comparing the use of variable and invariable adjectives in order to document typical language acquisition patterns and verify whether the ability to produce variable adjectives is acquired early in French.

Research questions

The present study aims to bring together data from spontaneous speech samples and elicited tasks in order to verify whether children show productive rule-governed behaviour when producing DP-internal adjective concord at young ages. It also aims to determine when DP concord becomes productive, and whether this morphosyntactic process is correlated with productivity in the spontaneous corpus and other morphosyntactic processes involving gender agreement, such as determiner concord.

We posit a number of hypotheses relating to the acquisition of concord, but also to theories of language acquisition. A first general hypothesis draws on previous research on agreement in language. Following Valois and collaborators’ studies (Valois et al., Reference Valois, Royle, Bourdua-Roy and Sutton2009; Valois and Royale, Reference Valois and Royle2009), we propose that concord (i.e., checking for agreement within the DP) is acquired early in French, but that different structures in the DP are productively produced at different acquisition stages. More specifically, determiners are the first to emerge, and are mastered quite rapidly with low error levels, whereas variable adjectives take longer to master due to the idiosyncratic nature of feminine variable forms. Accordingly, the rule linking feminine and masculine adjectival forms (if rule there is) should emerge only in later acquisition stages.

If the correct production of feminine adjectives is related to the acquisition of specific and idiosyncratic lexemes, then we should expect mastery of the elicited feminine structures to be strongly related to lexical acquisition of the feminine forms. That is, even if the child knows the masculine form of the adjective, we should observe some difficulty in production when the feminine counterpart is required. In theory, we would also expect occasional reversals of this pattern, i.e. comprehension or production of the feminine but not the masculine form, for children having learned only the feminine. However, because the masculine is the default in French, we would not expect to observe this pattern often, especially because we expect the children will be old enough to have acquired the semantics of the experimental lexical items.

Finally, if spontaneous speech corpora reflect the emergence and acquisition of linguistic rules, we expect to observe parallels between children's agreement productions in the spontaneous speech samples and their ability to produce the elicited structures. We expect types and tokens to be relatively common and without error in the corpora of children having no difficulty doing the tasks. Nonetheless, we might observe a relative lack of these structures in the corpora, even in children able to perform the tasks, because adjectives are not obligatory (unlike determiners), but also due to the deictic nature of play situations in which the spontaneous speech is gathered (i.e., children can point to, rather than describe, the topic).