Natural gender,phonological cues and the default grammatical gender in French children

Abstract

According to a dominant thesis, nominal endings are the privileged cues French children use to determine new nouns’ gender subclass. Children will rely on phonology even in cases of discordance with natural gender. Two elicited production studies involving more than 250 4- to 17-year-olds showed that while French children did not base their gender attribution choices on natural gender, they did not base them on phonology either: the masculine was the dominant choice. These results thus provide additional support to the ‘masculine as default’ view of French nouns’ gender acquisition proposed by the authors in an earlier study. The present article considers how the developmental conditions of children’s initial computations might bias their tallies towards a higher type frequency for masculine nouns, which could contribute to launching this gender as the default.

Keywords

Default French grammatical gender morphosyntactic development natural gender phonological cues

Introduction

Grammatical gender is a widespread property of many languages. In those morphosyntactic systems, the noun category is subdivided into two or more subclasses such that noun subclass membership has formal consequences for members of other syntactic categories in the sentence (Hockett, 1958). Because not all languages are gendered languages, a logically prior task for the child is to discover whether the ambient language has a gender system, and, if so, to determine the number and defining criteria of its gender subclasses (Braine, 1987; Carroll, 1989, 1995). On this basis, the gender subclass of each incoming new noun will have to be determined reliably as lexical development proceeds. According to a well-known thesis, French children would handle this by paying attention to a set of relationships between phonological shapes of nominal endings and gender subclasses (Karmiloff-Smith, 1979; Tucker, Lambert, & Rigault, 1977). However, recent production data suggest that this phonological tactic might not be the one French children favour. Boloh and Ibernon (2010) and Boloh, Escudier, Royer, and Ibernon (2012) found little evidence for a viable phonological strategy in the case of nonce nouns bearing feminine endings. Their results further suggested that the masculine might act as the default gender and not merely as the dominant one. Because those results were only based on inanimate nonce nouns, the present article reports additional production data using animate nonce nouns. We then consider the possible reasons why the masculine might be set as the default.

Which new noun belongs to which gender subclass – the gender attribution problem – is a problem that could readily be solved by children learning languages whose gender subclasses are semantically defined. However, this can not be so in others, such as French. French is a two-gender language whose gender system is regarded as a mainly morphological and/or phonological system (Corbett, 1991). Although far from perfect, regularities between nominal endings and gender subclasses do exist (Tucker et al., 1977) while correspondences between natural gender (i.e. male vs female) and grammatical gender (i.e. masculine vs feminine) only hold for about 10.5% of nouns (Séguin, 1969). Yet, two different possible routes for children’s initial steps in grammatical gender attribution have been considered for cases such as French (Levy, 1983, 1988, 1997).

A semantically driven route would have the child initially seek form–function mappings: the masculine vs feminine distinction in grammatical gender would be taken to correspond to the male vs female one that is found in animates denoted by animate nouns. As a consequence, gender attribution and gender agreement for animate nouns would be acquired before that for inanimate nouns (where no such correspondence might hold). By contrast, a formally driven route would have the child to attend to morphophonological regularities between nouns and gender subclasses from the start. Therefore, no differences between animate nouns and inanimate nouns should be found.

The first route would not actually have a chance. Its apparent plausibility rests entirely on the hidden assumptions that children might know that the morphosyntactic properties at stake are called grammatical gender, that the two subclasses are given the labels masculine and feminine, and that those labels are semantically close to the ones labelling natural gender, i.e. male vs female. Without those untenable assumptions, no learning scenario could make the child initially map those concepts onto the masculine vs feminine grammatical gender markers: the first handful of determiner–noun pairs in the input would make a sampling and hypothesis testing procedure reject those concepts as those that are grammaticalized by gender morphology. In addition, as far as French is concerned, it is simply not true that all animate nouns are either masculine nouns referring to males or feminine nouns referring to females. To mention only a few examples, témoin (witness) is masculine whether it refers to males or females, and victime (victim) and personne (person) are feminine whether they refer to males or females.

The formally driven route appeared to have much more empirical support. While metalinguistic tasks (e.g. Tucker et al., 1977) are not particularly informative with respect to how grammatical gender might be acquired, elicited production tasks showed that, as early as age 3, French children could rely on the phonological form of nominal endings to attribute gender to bare nonce nouns (Karmiloff-Smith, 1979). The phonological procedure was found to be strong enough to overcome alternative sources of information, such as the form of the article (up to age 6) or the natural gender (up to age 9). According to Karmiloff-Smith (1979, pp. 167–169), ‘as early as three to four years, i.e., as soon as articles were used consistently, the child constructed a very powerful, implicit system of phonological rules, based on the consistency, but not necessarily on the frequency, of phonological changes in word endings’. This was found because ‘it is through phonology that gender presents the most consistent patterns’.

However, diverging evidence has recently been reported. Boloh and Ibernon (2010) gave 4- to 10-year-old children and adults an elicited production task where they had to tacitly select a gender-marked article for inanimate nonce nouns. Nominal endings were either masculine or feminine, with high and equated predictive gender values. In each age group, the dominant gender attribution choice was towards the masculine article: suffix-congruent¹ choices were always at ceiling for masculine suffixes, without differences between individual suffixes, while suffix-congruent choices were either at chance or below chance for feminine suffixes. In this latter case, differences between scores for individual suffixes were found that could be related to differences in the frequencies of suffixes in the input. Critically, scores for feminine suffixes were never at a level high enough to support a viable version of the phonological scenario.

Boloh et al. (2012) used a similar elicited production task except that 4- to 10-year-old French children were presented with discordant determiner–noun pairs: the gender form of the article and the gender value of nominal endings were opposite. All nonce nouns were inanimate nouns. Results showed that the phonological procedure never won over the article gender form: suffix-congruent choices were far below chance in the case of feminine suffixes and either below or at chance in the case of masculine suffixes.

Those results suggest that French children might solve the attribution problem through a simple masculine-as-default tactic. The masculine gender would be set as the default and this assumption would be relaxed where explicitly specified otherwise: the gender of feminine nouns would be learnt on the basis of their co-occurrence with a (feminine) determiner. Under this view, sensitivity to the gender value of nominal endings would be what children might eventually induce upon acquiring grammatical gender rather than being the way they learn it.

However, because those results were only based on inanimate nonce nouns, it makes sense to consider Karmiloff-Smith’s (1979) experiment where natural gender was pitted against phonological cues to grammatical gender. Three- to 11-year-old children were shown pictures of Martian-like persons who were obviously either males or females. In one condition, males were introduced with nonce nouns bearing a feminine suffix while females were introduced with nonce nouns bearing a masculine suffix. In a control condition, males as well as females were introduced with nonce nouns with gender-neutral nominal endings. In the first condition, children up to age 9 supplied suffix-congruent article forms: they used a feminine article to refer to males and a masculine article to refer to females (the mean percentage of suffix-congruent scores was 73% in both cases). In the control condition, children at all ages referred to males with a masculine article (the mean percentage of masculine responses across all age groups was 88%). However, they also supplied masculine articles to refer to females. Except at age 9, choices were not based on natural gender: the mean percentage of masculine responses in 3- to 8-year-olds was 65% and it was 100% in 10-year-olds. However, according to Karmiloff-Smith (1979, p. 162), ‘whilst children can take natural gender into account in cases of no discord, they actually tend to base themselves almost entirely on phonological procedures’.

This study attracted a great deal of attention at the time (e.g. Maratsos, 1983; Pinker, 1984; Slobin, 1985) and virtually all reports on the acquisition of grammatical gender since then have made reference to it (e.g. Demuth & Weschler, 2012). However, closer inspection of the data shows that the conclusions might have been too strong. Scores were actually at chance in a number of cases and there were incredibly huge variations in scores for females in the control condition. More importantly, however, it appears that children actually tended to use the masculine article everywhere except when referring to males introduced with nonce nouns with feminine suffixes (up to age 9). This latter case is the only one where a phonological attribution procedure could be deemed to be operating.

To begin with, no phonological procedure could in principle be invoked to explain that children referred to females with masculine articles in the control condition (as claimed by Karmiloff-Smith in the quote above): endings in this condition were gender-neutral by construction. Considering now reference to females in the main condition and in the control one, we see that children actually tended to use masculine articles in both, that is, whether or not nonce nouns were bearing a masculine suffix. This pattern of data of course undermines the claim that a phonological procedure was used in the main condition for reference to females introduced with masculine nonce nouns. Rather, this suggests that the masculine might have been used here as a default just the way it was used in the corresponding control condition.

The only evidence now remaining for the claim that a phonological procedure was so consistently put to use as to override natural gender (i.e. referring with feminine articles to males introduced with nonce nouns with feminine suffixes) also happens to be at odds with Boloh and Ibernon’s (2010) results: they found that feminine suffixes never elicited suffix-based article choices at levels above chance. At first glance, it could seem paradoxical that children could successfully rely on feminine suffixes in a noisy condition involving a contradiction between sex and suffixes while they would fail to do so in a simpler condition involving inanimate nouns.

However, Karmiloff-Smith’s (1979) youngest participants were 3-year-olds. Children of that age might have sliced up the lexicon up to a point where they found out that nouns referring to males tend to belong to one gender subclass (i.e. the so-called masculine class) while nouns referring to females generally fall in the other one (i.e. the so-called feminine class). A tacit registration of the probabilistic gender value of nominal endings might also have started (whatever children actually do with it for gender attribution). Those two bits of linguistic knowledge might have prompted children to pay attention to alternating pairs such as -ier vs -ière (as in écolier vs écolière/schoolboy vs schoolgirl), or -(a)in vs -ine (as in copain vs copine/pal_masc vs pal _fem) that appear on animates. This, in turn, might trigger a phonological procedure for gender attribution that could generalize to animate nouns as long as the gender value of nominal endings is high enough, i.e. whether or not a given ending happens to be one of those suffixes that are part of alternating pairs.

To examine the hypothesis that the phonological procedure might at least generalize to animate nouns we first report results from an elicited production study similar to Karmiloff-Smith’s, although restricted to the two discordant conditions (i.e. males introduced with nonce nouns bearing feminine suffixes and conversely). Although nominal endings were chosen primarily on the basis of their gender-predictive values we also included two alternating pairs amongst those that typically appear on animate nouns. If the phonological procedure generalizes to animate nouns, one should find suffix-congruent choices, and not sex-congruent choices, in both conditions, at least within the 4–9 age range. Under the masculine as a default account, one should expect suffix-congruent choices in the case of nonce nouns with masculine suffixes referring to females. Critically, however, one should also observe masculine choices in the case of nonce nouns with feminine suffixes referring to males. In this latter case, feminine suffixes should never elicit feminine articles at a level significantly above chance, except, perhaps, under a restricted version of the phonological procedure, in cases of feminine members of alternating pairs.

Experiment 1

Participants

The experiment included 228 children, adolescents and young adults: 47 children aged 3;8–5;0 (M = 4.5, SD = .4), 52 children aged 6;7–7;11 (M = 7.2, SD = .43), 59 children aged 9;6–10;11 (M = 10.2, SD = .43), 24 children aged 11;1–12;6 (M = 12.0, SD = .48), 24 adolescents aged 14;4–15;8 (M = 15.1, SD = .48) and 22 young adults aged 17;4–18;5 (M = 17.11, SD = .35), with similar numbers of males and females in each age group. The two older age groups acted as controls. Participants were recruited from schools in the Montpellier area and were from varied socioeconomic backgrounds. All were monolingual French speakers and of Caucasian origin.

Non-linguistic material and linguistic stimuli

Participants were individually tested and introduced to nonce nouns referring to imaginary persons. They were asked to describe actions performed by an experimenter on those persons. We used 3D puppets to ensure that participants referred to the puppet rather than to the image depicting it: the French word image (image_fem) is feminine, while dessin (drawing_masc) is masculine and this could spuriously produce a bias in answers. Twelve trios of imaginary persons were prepared. In each, one exemplar was green, the other one was white and the third was purple. Puppets were approximately 12 cm high and differed from human-like puppets along dimensions such as the number of arms, legs or eyes, the presence of antenna, the relative size of the trunk and so on. Six trios comprised males while the six others comprised females. The male vs female distinction was introduced through characteristics such as wearing a moustache and a tie vs having long hair and make-up.

All nonce nouns were bi-syllabic. To guard against possible gender bias the first syllables of nonce nouns were syllables that were either rare or non-existent as initial syllables in the French lexicon (from the Petit Robert dictionary; Rey-Debove & Rey, 1993): Kib- (1 entry), fit- (no entry), baj- (2 entries), tid- (no entry), vam- (no entry) and mav- (no entry). None violated French phonotactic rules.

The second syllable was to be informative regarding the gender subclass membership of the nonce noun. We therefore only selected endings with mean informative values above 90%. All endings were suffixes, although of course not all French nouns sharing those endings are suffixed nouns. The six masculine endings were -o, -on, -ier, -a, -an and -(a)in. The six feminine endings were -ette, -ise, -ière, -ade, -ure and -ine. The two alternating pairs were -ier vs -ière (as in écolier vs écolière/schoolboy vs schoolgirl, infirmier vs infirmière/nurse_masc vs nurse_fem) and -(a)in vs -ine (as in copain vs copine/pal _mascvs pal_fem, cousin vs cousine/cousin_masc vs cousin_fem).

The following measures were assessed. Type predictivity refers to the proportion of noun types of the dominant gender relative to all noun types with this suffix or ending. Token predictivity refers to the proportion of noun occurrences of the dominant gender relative to the total number of occurrences of nouns with this ending or suffix. Those measures were computed based on the following sources. First, we combined an analysis of the lexical input of Philippe and of Stéphane, children whose verbal production and verbal input were recorded from age 1;1 to 3;3 for Philippe (Suppes, Smith, & Léveillé, 1973) and from age 2;3 to 4;9 for Stéphane (Rondal, 1985; Rondal, Bachelet, & Peree, 1985) and made publicly available through the CHILDES project (MacWhinney, 2000). Combining the two inputs (henceforth, P&S input) yielded a lexical database of 1810 types (57% masculine) and 26,470 tokens (55% masculine). We categorized nouns according to gender and suffix/ending type, only removing special, highly frequent words such as maman (mommy), papa (daddy) and so on.

Second, we used the CD-ROM version of Le Robert junior illustré (Rey-Debove & Rey, 1999), a 20,000 word dictionary specifically designed for 8- to 12-year-olds (henceforth RJ). Because Lyster (2006) made a number of coding decisions specifically aimed at French L2 learners, we repeated the tallies to assess type predictivity of suffixes/endings from a database of 10,967 nouns. Finally, we used the database described by New, Brysbaert, Veronis, and Pallier (2007; New, Pallier, Ferrand, & Matos, 2001), a corpus of 52 million French words taken from film subtitles (henceforth, Freq Film).

As shown in Table 1, the mean type-predictive values were above .90 and similar for masculine and feminine suffixes. Tokens’ predictive values were higher for feminine than for masculine suffixes.

Table 1.

Informative values (type and tokens) per suffixes and per database.^a

Suffixes	Masculine							Feminine
	-o	-on	-ier	-a	-an	-(a)in	Mean	-ette	-ise	-ière	-ade	-ure	-ine	Mean
Type predictivity
P&S input	90.3	95.2	97.3	94.4	98.4	95.2	95.2	100	92.3	82.4	90.9	93.1	96.3	92.5
RJ	95.6	89.2	94.3	87.9	99.4	98.6	94.2	90.9	95.5	92.4	95	91.9	94.9	93.4
Freq Film	94.4	92.9	98.6	77.7	99.3	97	93.3	92.8	94	93.5	90.2	88.5	95.2	92.4
Mean	93.5	92.5	96.7	86.7	99	96.9	94.2	94.6	93.9	89.4	92	91.2	95.5	92.8
Token predictivity
P&S input	62.7	97.6	99	99.7	94.9	82.4	89.3	100	99.2	97.9	61.2	94.2	99.2	92
Freq Film	81.5	69.7	86.5	91.1	91	78	83	97.2	99.1	82	90	90.2	97.1	92.6
Mean	72.1	83.6	92.8	95.5	93	80.2	86.2	98.6	99.2	90	75.6	92.2	98.2	92.3

P&S corresponds to Philippe’s lexical input and Stephane’s lexical input. RJ corresponds to the CD-ROM version of Le Robert junior illustré (1999). Freq Film corresponds to lexique.org database. Type predictivity refers to the proportion of noun types of the dominant gender relative to all noun types with this suffix or ending. Token predictivity refers to the proportion of noun occurrences of the dominant gender relative to the total number of occurrences of nouns with this ending or suffix.

Procedure and design

Nonce nouns were arranged in two blocks of six nonce nouns each, with three masculine endings and three feminine endings in each block. In the first block, the endings were -o, -on, -ier, -ette, -ise and -ière. In the second block, the endings were -a, -an, -(a)in, -ade, -ure and -ine. For each block, 24 lists of six different nonce nouns were prepared such that each initial possible syllable (of six) and each possible ending (of six) appeared only once in each list. Each list was further subdivided into two sets of three nonce nouns, with two nouns with masculine endings and one noun with a feminine ending in one set, and two nouns with feminine endings and one noun with a masculine ending in the other set. Each participant in each age group was presented one of the 24 lists. Within the constraints indicated above, the particular endings appearing in each set were systematically varied across participants. The order of sets within a list and the order of blocks were systematically varied across participants.

The participant was told that he or she and the experimenter would send puppets to another person. The participant’s role would be to record a message indicating to this person in which bag (a big bag, a medium bag or a little bag) each of the three exemplars (the white one, the green one and the purple one) would be sent. Thus, while participants would focus their attention on the relation between the colours of the puppets, the size of the bags and the experimenter’s successive actions, their answers would reveal which gender forms they tacitly selected for the article. In order to attract and maintain younger participants’ attention we added a fictional context. The person to whom puppets were sent was an imprisoned fairy who needed magical puppets with special powers to escape. The participants’ task was thus to help the fairy by indicating precisely in which bag each puppet would be found.

Each trio of puppets was presented to the participants along with the noun for them. Male puppets were introduced with nouns bearing a feminine suffix while female puppets were introduced with nouns bearing a masculine suffix. Participants were asked to tell whether puppets were boys or girls. In case of failure to supply the correct identification, the participant’s data were discarded. Nouns were repeated three times. However, no repetition was required from participants. Nouns were introduced with the gender-neutral numerical adjective trois (three), as follows: Voici trois kibains (repeated three times). Maintenant je les mets dans un sac et tu dis où je les mets (Here are three kibains. Now I put them in a bag and you say where I put them). Participants were expected to produce descriptions such as, e.g. tu mets le kibain vert dans le petit sac (you put the _masc green _masc kibain in the small bag).

The first block of items was given in a morning session and the second during an afternoon session. Pauses were inserted between sets and additional pauses were given upon participants’ request. Finally, to ensure that participants understood the game and knew the colour adjectives, two pre-tests were presented with familiar objects, one referred to with a masculine noun (stylo/pen _masc), the other referred to with a feminine noun (gomme/eraser _fem). No feedback on gender was given.

Scoring

The dependent variable was the number of times a participant chose a definite or indefinite article whose gender form fitted the probabilistic value of the noun suffix (e.g. choosing un/a _masc or le/the _masc to refer to a kibon, or choosing une/a _fem or la/the _fem to refer to a bajette). The adjective form was not taken into consideration in scoring for attribution (see Boloh & Ibernon, 2010). Given that French has only two genders, the only possible responses were either a masculine article or a feminine article. Due to the pragmatics of our experimental situation, there were no alternative responses such as a plural article or a numerical adjective. For a given ending, scores could range from 0 to 3. For the six masculine endings as well as for the six feminine endings, scores could range from 0 to 18.

Results

Preliminary inspection of the data revealed no significant difference between male and female participants’ scores. As shown in Table 2, suffix-congruent scores for masculine suffixes (where nouns referred to females) were close to ceiling from age 4 to age 12. They then abruptly declined at ages 15 and 17, where they were at chance. By contrast, suffix-congruent scores for feminine suffixes (where nouns referred to males) were never above chance: they gradually declined from age 4 to age 12 and were non-existent at ages 15 and 17.

Table 2.

Means and standard deviations of suffix-congruent article choices, per age and per condition (maximum score = 18).^a

Age	Female natural gender + masculine suffixes	Male natural gender + feminine suffixes
4;5	14.89 (4.21)	7.11 (5.82)
7;2	16.19 (3.41)	5.71 (5.91)
10;2	15.59 (3.73)	3.79 (4.83)
12;0	16.38 (4.16)	2.33 (4.79)
15;1	6.79 (7.93)^*	0 (0)
17;11	6.45 (7.58)^*	0 (0)

All scores below chance level for feminine suffixes and above chance level for masculine suffixes (p < .05), except for *.

An analysis of variance (ANOVA) with age group (6) as a between-participants factor and the gender value of suffixes (masculine vs feminine) as a within-participants factor yielded a significant main effect for age, F(5, 222) = 42.78, p < .001, η²_p = .08, a significant main effect for the gender value of suffixes, F(1, 222) = 300.05, p < .001, η²_p = .91, and a significant interaction between age and suffix gender value, F(2, 222) = 4.56, p < .001, η²_p = .01. Post hoc tests (Tukey for different Ns) on scores for masculine suffixes only showed significant differences between the 15- and 17-year-olds and any of the other age groups (p < .001 at least). Post hoc tests on scores for feminine suffixes showed that 4-year-olds’ scores were higher than those of any other age group, except 7-year-olds’ ones (p < .007 at least). Scores of 7- and 10-year-olds were higher than those of 15- and 17-year-olds (p < .05 at least).

Suffix-congruent scores of 4-year olds were close to ceiling for masculine suffixes (M = 14.89, SD = 4.21) but below chance for feminine suffixes (M = 7.11, SD = 5.82), a highly significant difference, F(1, 46) = 36.37, p < .001, η² = .97. Sign tests (Bonferroni correction, α set at p = .003) revealed no significant difference between masculine suffixes or between feminine suffixes except, in this latter case, for the -ette vs -ure comparison, p = .0023. Children’s scores were significantly higher for every masculine suffix as compared to every feminine suffix (Bonferroni correction, α set at p = .0014) except for comparisons involving -ette, -ine (except for the -ine vs -ier and the -ine vs -ain comparisons) and for the -o vs -ise and the -o vs -ière comparisons. Participants relying consistently on feminine suffixes (i.e. getting scores ranging from 13 to 18) were outnumbered by those almost never doing so (i.e. getting scores ranging from 0 to 5): N = 10 vs N = 21, although this difference was only marginally significant (χ² (1) = 3.23, p = .072).

Suffix-congruent scores of 7-year olds were close to ceiling for masculine suffixes (M = 16.19, SD = 3.41), but below chance for feminine suffixes (M = 5.71, SD = 5.91), a highly significant difference, F(1, 51) = 91.86, p < .001, η² = .99. Sign tests (α set at p = .003) revealed no significant difference between masculine suffixes or between feminine suffixes. Children’s scores were significantly higher for every masculine suffix as compared to every feminine suffix (α set at p = .0014) except for the -a vs -ière comparison. Participants relying consistently on feminine suffixes were outnumbered by those almost never doing so: N = 6 vs N = 29, χ² (1) = 13.83, p < .001.

Suffix-congruent scores of 10-year olds were close to ceiling for masculine suffixes (M = 15.59, SD = 3.73), but below chance for feminine suffixes (M = 3.79, SD = 4.83), a highly significant difference, F(1, 58) = 181.37, p < .001, η² = .99. Sign tests (α set at p = .003) revealed no significant difference between feminine suffixes or between masculine suffixes, except for the -a vs -ain comparison. Children’s scores were significantly higher for every masculine suffix as compared to every feminine suffix (α set at p = .0014). Participants relying consistently on feminine suffixes were outnumbered by those almost never doing so: N = 4 vs N = 38, χ² (1) = 25.93, p < .001.

Suffix-congruent scores of 12-year-olds were close to ceiling for masculine suffixes (M = 16.38, SD = 4.16) but below chance for feminine suffixes (M = 2.33, SD = 4.8), a highly significant difference, F(1, 23) = 107.44, p < .001, η² = .99. Sign tests (α set at p = .003) revealed no significant difference between feminine suffixes or between masculine suffixes. Children’s scores were significantly higher for every masculine suffix as compared to every feminine suffix (α set at p = .0014). Participants relying consistently on feminine suffixes were outnumbered by those almost never doing so: N = 2 vs N = 20, binomial test, p < .001.

Finally, the patterns of scores were identical in 15- and 17-year-olds. Suffix-congruent scores were at chance for masculine suffixes (respectively, M = 6.79, SD = 7.93, and M = 6.45, SD = 7.58) while they were non-existent in the case of feminine suffixes. There was no significant difference between masculine suffixes, either at age 15 or at age 17.

Discussion

Although both feminine and masculine suffixes were chosen to have high type-predictive values, and although token-predictive values were higher in the former case than in the latter, the patterns of responses were very different in the two conditions. In the case of females introduced with nouns bearing a masculine suffix, choices were clearly suffix-congruent from age 4 to age 12. There was then a very clear shift towards sex-congruent choices which became dominant, although not significantly so, in 15- and 17-year-olds. In the case of males introduced with nouns bearing a feminine suffix, suffix-congruent choices were never above chance and they gradually declined as to be non-existent in 15- and 17-year-olds. Feminine members of alternating pairs (i.e. -ière and -ine) did not elicit suffix-congruent choices above chance either, even though they sometimes, but not consistently, tended to elicit such choices in a higher proportion as compared to other suffixes (e.g. in 4- and 7-year-olds). As predicted, then, we did not replicate Karmiloff-Smith’s (1979) results in the critical condition that we pointed out in the Introduction section (i.e. males introduced with nouns bearing feminine suffixes): the mean percentage of suffix-congruent responses was 31% in our 4- to 10-year-olds while it was 73% in Karmiloff-Smith’s 3- to 9-year-olds.

Considering results in the 4–12 age range, one could envision two competing interpretations of our data in each condition. Suffix-congruent scores in reference to females might be true suffix-based (masculine) responses but they also might be (masculine) default responses (see note 1). The dominant masculine responses in reference to males might be true sex-based responses or (masculine) default responses. Note that the two conditions were exactly symmetrical and equal regarding the predictive values of suffixes and that there seems to be no a priori reason why participants would have based their responses on sex in one condition (reference to males) and on phonology in the other one (reference to females). Obviously, the masculine-as-default account would be the most parsimonious one. One has thus to consider things beyond the apparently binary nature of the possible responses: a non-suffix-congruent response is not ipso facto a sex-based one and a suffix-congruent response is not ipso facto a suffix-based response.

The significant move towards sex-congruent responses in reference to females from age 15 onwards provides indirect evidence for the former claim. One could not account for this shift through a vanishing of the masculine-as-default strategy and even less, under the phonological account, through a loss of the predictive values of suffixes: those interpretations would just not be tenable. Thus, what seems to be at stake here is a clear move towards true sex-based responses. Now, if, for whatever reason, only older participants based their gender attribution choices on natural gender, there would be little sense in maintaining that 4- to 12-year-olds’ responses in the condition where males were introduced with nouns bearing feminine suffixes were sex-based responses. Thus, those non-suffix-congruent responses were most probably (masculine) default responses.

Evidence that a suffix-congruent response is not ipso facto a suffix-based one can be found in Karmiloff-Smith’s (1979) data. As indicated, children opted for the masculine article for reference to females whether suffixes were neutral or masculine. This pattern of data undermines the suffix-based interpretation in the main condition and favours the masculine-as-default one. As noted above, however, there were huge and unexplained variations in scores in Karmiloff-Smith’s control condition where reference should be made to females. In addition, one of the supposedly neutral suffixes used for this condition (-a) actually is a masculine suffix in French, which might have biased the results. Therefore, we ran a small control experiment, similar to Karmiloff-Smith’s one, where nonce nouns with neutral endings were introduced in reference either to males or to females.

Experiment 2

Participants

Forty-eight typically developing children from varied socioeconomic backgrounds participated in the experiment. They came from two age groups: 24 children (11 boys) aged 3;8–4;4 (M = 4.0, SD = .25) and 24 children (12 boys) aged 7;7–8;4 (M = 8.1, SD = .26). Children were recruited from French schools and all were monolingual French speakers.

Non-linguistic material and linguistic stimuli

The non-linguistic material was the same as that in experiment 1 except that we only used four trios of puppets: two trios with three males in each and two trios with three females in each. All nonce nouns were bi-syllabic. We used four of the six initial syllables used in experiment 1: kib-, fit-, baj- and tid-. We selected four endings that were rare enough to be considered neutral regarding grammatical gender: -uv, -ink, -ib and -uf. None of those endings appeared as endings of real nouns in the P&S input, thus up to age 4;9. The RJ database confirmed the scarcity of those endings amongst nouns supposedly known by 8- to 12-year-old French children: -ub (two types, one masculine, one feminine), -ink (two types, both masculine), -uf (one type, feminine) and -uv (three types, one masculine, two feminine).

Two sets of four nonce nouns were prepared and assigned each to half of the total number of participants in each age group. The first set comprised bajuv, kibib, tiduf and fitink. The second set comprised kibuf, bajib, fituv and tidink.

Procedure and design

Each trio of puppets was presented to the participants along with the noun for them. The procedure was exactly the same as in experiment 1 except that the four nonce nouns were all given in one session. The order of nonce nouns and the order of male trios vs female trios were systematically varied across participants.

Scoring

Given that endings were neutral, the dependent variable was expressed as the number of times a participant chose a definite or indefinite masculine article. For a given ending, scores could range from 0 to 3. For the two endings assigned either to trios of males or to trios of females, scores could range from 0 to 6.

Results and discussion

Preliminary inspection of the data revealed no significant difference between male and female participants’ scores. There was no difference either between individual endings. As shown in Table 3, gender attribution choices were almost exclusively towards the masculine article in both conditions, that is, whether reference should be made to males or to females. Our participants did not even supply the small number of feminine responses that were seen in the Karmiloff-Smith’s data, especially in the condition where reference should be made to females. Our results thus replicate the major trend towards the masculine that was seen in her control condition. This, in turn, confirms our view that suffix-congruent (masculine) responses in reference to females in experiment 1 were actually default (masculine) responses.

Table 3.

Means and standard deviations of masculine article choices, per age and per condition (maximum score = 6).

Age	Male natural gender + neutral suffixes	Female natural gender + neutral suffixes
4;0	6	5.75 (.85)
8;0	6	5.88 (.61)

Because the general scarcity of sex-based responses might be regarded as striking, especially in experiment 2, it is worth considering further what might have been at stake from children’s point of view. Recall that we checked whether participants were able to identify imaginary persons as males or females before providing nonce nouns. Before nonce nouns were introduced, children occasionally produced spontaneous commentaries on females puppets such as elles sont jolies (they _fem are pretty), thus using a feminine form of pronouns. From the moment the experimenter introduced nonce nouns (e.g. here are three kibufs), the child would then shift to the masculine gender to produce the requested description, e.g. tu mets le kibuf vert dans le petit sac (you put the _masc green _masc kibuf in the small bag).

While pronouns refer either to an individual or to a collection of individuals, nouns refer to categories and it has been shown that naming can prompt the formation of categories even in 12-month-olds (Waxman & Markow, 1995). Providing nonce nouns in our tasks might have led children up to age 12 to shift to a categorical level of representation. Children referred to imaginary persons that they knew were either males or females but they might have referred to them as mere members of the category that was suggested to them by the new taught label i.e. they were kibufs (experiment 2) or kibains or tidettes (experiment 1) and so on. Categorization demoted sex to a secondary information and the masculine popped out as the default response due to the lack of the full-proof cue for gender (i.e. articles).

However, one cannot rule out another possible reason why sex-based responses were rare. Puppets, and for that matter, drawings, are just objects. Although younger participants were able to identify males vs females, the lack of animacy cues (e.g. auto initiated movement) might have led them to refer to what they subsequently regarded as mere objects. Neither our experiments nor Karmiloff-Smith’s one can securely ascertain whether natural gender was part of younger children’s representation while they supplied their responses. Controlling for this would require videos depicting actors performing actions. Note, however, that in the case this produced an increase of sex-based responses (in reference to females) this would just be another rebuttal for the phonological procedure although not for the masculine-as-default proposal: the correspondence between grammatical gender and natural gender only holds for a small subset of French nouns.

General discussion

Alongside Tucker et al.’s (1977) judgement studies, Karmiloff-Smith’s (1979) production studies have been taken to provide the empirical justification for the phonological procedure as the main solution to gender attribution in French. While contrary evidence has recently been reported for inanimate nouns, her classic experiment where grammatical gender was pitted against natural gender could have shown that a phonological procedure does apply to animate nouns. We have shown that this conclusion could only be based on results from one experimental condition (choice of a feminine article to refer to males with feminine nonce nouns) and we failed to replicate this finding. The masculine article was the dominant choice everywhere except at ages 15 and 17 for reference to females. This latter developmental trend, the symmetrical nature of main conditions in experiment 1 and the corroborating results from experiment 2 suggest that the masculine acted as a default. We further suggested that the effect of naming on categorization demoted sex-related information, which paved the way for the default.

We found no evidence for a phonological procedure that would apply to animate nouns, not even for a limited version of it: feminine members of alternating pairs were unable to elicit suffix-congruent choices above chance. While it could be argued that this might be demonstrated with some other pairs (e.g. -ais vs -aise or -ien vs -ienne), such effects of a limited number of suffix pairs, within a particular semantic niche (i.e. nouns referring to beings), would be of restricted scope: this would only speak to a tiny fraction of French nouns. Both Karmiloff-Smith (1979) and Levy (1983, 1988, 1997) were right in claiming that (younger) French children do not base their grammatical gender attribution choices on natural gender. On the other hand, the phonological procedure was then dramatically overestimated as a general, viable solution for gender attribution.

When presented with inanimate nonce nouns whose endings are masculine or feminine, 4- to 10-year old children and adults overwhelmingly opt for the masculine article: suffix-congruent scores for feminine suffixes are far below those for masculine suffixes (Boloh & Ibernon, 2010). Suffix-congruent scores for feminine suffixes are also far below those for masculine suffixes when 4- to 10-year-olds are presented with gender-discordant article–noun pairs, although in both cases the dominant response is to adhere to the provided form of the article (Boloh et al., 2012). Finally, when presented with a contradiction between natural gender and the gender value of nominal endings, children up to age 12 also overwhelmingly supply masculine articles (the current study). Suffix-based feminine responses are nowhere above chance and this is found despite the fact that, in each study, the feminine and masculine endings were equated regarding gender-predictive values. Note that the feminine nominal endings used in those experiments have exceptionally high gender-predictive values. Thus things could only be worse for the many other feminine endings/suffixes with lower ones.

These results cannot be reduced to the sort of methodological artefact proposed by Seigneuric, Zagar, Meunier, and Spinelli (2007). Those authors claimed that participants would fall back on the masculine article either because the masculine form of the adjective is the simpler, canonical form or because they do not know, or are not sure of, the exact form of the feminine adjective. This would thus remove the processing load regarding agreement between the (feminine) article and the (feminine) adjective. Unfortunately, amongst the many reasons why this account is untenable is the fact that children also overwhelmingly supplied the masculine article when well-known adjectives whose forms do not vary with gender (as is the case for two-thirds of French adjectives) were used (Boloh & Ibernon, 2010).

Frequency effects could not explain these patterns of data either. It could have been argued that although gender-predictive values were equated across masculine vs feminine nominal endings the latter ones might have been less frequent in the input, thus less well known, which would explain lower scores in those cases. The most appropriate task where to assess such possible frequency effects is the one using bare nonce nouns without any other sort of information (i.e. articles or natural gender). However, frequencies of those particular endings were computed either as suffixes per se or as the last two to three phonemes and there was no evidence that these could explain the differences between scores for masculine vs feminine nonce nouns (Boloh & Ibernon, 2010; these endings are the same as those used in Boloh et al., 2012, and in the current study).

The evidence so far thus undermines the notion of a fully fledged phonological procedure that would handle gender attribution. It also shows that the masculine and the feminine genders are clearly not equally probable. The masculine gender overwhelmingly applies to nonce nouns bearing masculine suffixes, without frequency effects. It generalizes to nonce nouns with neutral endings and overgeneralizes to nonce nouns with feminine ones. This suggests an account whereby the masculine would be initially set and represented as the default gender. The definitive source of blocking for this default would be the registration of a connection between a noun and one of its fool-proof feminine determiners, which might require more than a couple of instances to be strongly memorized (Boloh et al., 2012; note that French nouns must appear with a determiner, and, except if one assumes that most occurrences are in the plural, this will be with an overtly gender-marked determiner). Phonological regularities within feminine nouns would be represented in an associative memory (as suggested by frequency effects amongst feminine suffixes), which could also block the default, although only to a very moderate extent.

That being said, further evidence is required to show that suffix-congruent responses (i.e. masculine articles) in the case of nonce nouns with masculine endings actually are default responses and not suffix-based ones. This could be done in combining the ‘ending alone’ condition from Boloh and Ibernon (2010) and the ‘discordant condition’ from Boloh et al. (2012) in a within-participants design. Correlations could thus be computed between suffix-congruent scores to feminine nouns in the former condition and suffix-congruent scores to masculine nouns in the latter. Because individual variations were found in children’s propensity to supply suffix-congruent responses and because suffix-congruent scores in the former condition cannot be but suffix-based responses one could derive the following predictions. If suffix-congruent scores for masculine nouns in the ‘discordant condition’ are true suffix-based choices there should be a significant positive correlation between those scores and suffix-based scores for feminine nouns in the ‘ending alone’ condition. Alternatively, if scores in the former condition are default responses, there should be a significant negative correlation with suffix-based responses in the latter condition.

Second, it would be necessary to show that low suffix-congruent scores for nonce nouns with feminine suffixes are really due to the default overflowing the (feminine) suffix-driven information rather than to the fact that children would not have extracted those gender probabilistic values or only weakly so. Results such as Tucker et al.’s (1977) or Seigneuric et al.’s (2007) cannot clarify this issue. Besides the fact that their tasks were clearly metalinguistic, the experimental question forced participants to a priori equate the probabilities that a given noun would be masculine or feminine (thus, setting those probabilities at .50 in both cases). However, there are now reasons to believe that acquisition and production do not abide to those hidden probabilities assumptions. It would thus be necessary to devise tacit tasks that would also be non-gender attribution tasks. For example, returning to the natural vs grammatical gender issue, we would expect children to be successful in a situation where six imaginary persons (three males and three females, of same kind) were introduced with here are six tidiens and they would then be asked to give two tidiennes. The child’s primary task would not be to decide on nouns’ gender subclass, but he or she could use their registered knowledge of phonological regularities in relation to natural gender to guess the answer. Analogous (artificial) object-selection tasks could be devised to probe children’s knowledge of nominal endings in the case of inanimate nonce nouns (see Arias-Trejo & Alva, 2012, for such evidence in Spanish-speaking 30-month-olds).

A central issue is of course to determine what might make French children set the masculine as the default gender. Default morphological patterns have been said to be explainable by input statistics such as type frequency (e.g. Bybee, 1995; Plunkett & Marchman, 1993; Rumelhart & McClelland, 1986), distributional asymmetry (e.g. Hare & Elman, 1995) or faithfulness to the stem (e.g. Stemberger, 1995). However, in virtually every case, there is also evidence for defaults without such input properties (e.g. Berent, Pinker, & Shimron, 1999, 2002; Marcus, Brinkmann, Clashen, Wiese, & Pinker, 1995). In what follows, we consider a possible initial, although not subsequent, role for type frequency that could stem from the means children recruit to interconnect masculine vs feminine determiners on the one hand and from the effect of the liaison rule between possessive pronouns and vowel-initial feminine nouns on the other.

As indicated, 58% of French noun types are masculine (Séguin, 1969). Our P&S input data showed that 57% of noun types and 55% of noun occurrences are masculine. Whether or not a difference of such magnitude is sufficient to launch the masculine as the default is hardly determinable as long as there is no direct evidence for how the child runs and updates his or her tallies. However, one can still consider the developmental conditions under which the child would count. Recall first, that masculine and feminine nouns are nothing but nouns that co-occur with a set of mutually predictive determiners, respectively the masculine one and the feminine one. As long as determiners in each set have not been interconnected, the masculine vs feminine subclasses simply do not exist and it makes no sense to consider counting the number of nouns in each.

A relevant question is thus to know what are masculine vs feminine nouns from the child’s point of view so that he or she may count them. Grouping determiners as a common class to then categorize novel words as nouns is something 14-month-olds are able to do (Shi & Melançon, 2010). However, 17-month-olds are unable to categorize determiners based on gender, something that 30-month-olds will do (Cyr & Shi, 2010). More generally, learning dependencies among arbitrary grammatical classes is a form of distributional learning that has been shown to be extremely difficult if not impossible (e.g. Smith, 1969) except if partial, within-subclass similarities are available to the learner (Braine, 1987; Brooks, Braine, Catalano, Brody, & Sudhalter, 1993; although see Pinker, 1984, chapter 5).

Building on Braine’s proposal, Boloh and Ibernon (2010, pp. 22–23) noted that masculine nouns might benefit from a clearer within-similarity space than feminine nouns. The number of vocalic ending types is far below the number of consonant ending types and vocalic endings showing a gender bias of .90 at least in the P&S input are always biased towards the masculine. By contrast, consonant endings with .90 predictive values are sometimes biased towards the feminine gender and sometimes towards the masculine gender. Thus, similarities in the vocalic/masculine case do not even seem to require that differences between vocalic endings types be necessarily processed, which could not be dispensed with in the consonant/feminine case. The interconnection of masculine determiners might thus be achieved earlier than that of feminine determiners or there might always be a greater number of interconnected determiners in the masculine case than in the feminine one during the early phases of gender learning.

A quantitative consequence of this is that the child’s number of masculine nouns will increase much faster that his or her number of feminine nouns. A new noun coming in with either one of the previously interconnected masculine determiners will count as one more member of this class. However, a new noun coming in with an unconnected determiner could only count as one more noun with this determiner but not as one more feminine or masculine noun. As soon as the feminine class is properly defined, the initial asymmetry in counts will progressively vanish and the child’s tallies will eventually converge to the 58% vs 42% adult type frequencies.

A further input property that could contribute to an initial increase in the number of masculine nouns is the liaison phenomenon involving possessive pronouns and vowel-initial nouns. Vowel-initial feminine nouns actually take masculine forms of possessive pronouns as in, e.g. mon oreille, ton oreille, son oreille (respectively, my _masc ear, your _masc ear, his ear). Nothing of this sort ever happens to masculine nouns. As long as feminine determiners have not been fully interconnected, any occurrence of such feminine nouns with possessive pronouns might count as one more masculine noun due to the previous interconnection of mon, ton, son with other masculine determiners. This bias in the tallies might play a role until the child has extracted a liaison rule such that masculine forms of possessive pronouns before vowel-initial feminine nouns should be converted and represented as feminine forms. Input analyses would be needed to assess the quantitative impact of this characteristic of French possessive determiners. In any case, current evidence suggests difficulties with liaison-related phenomena (e.g. Babineau & Shi, 2011; Chevrot & Fayol, 2001).

The statistics and qualitative properties of the input might thus conspire to strongly bias children’s type frequency tallies during the initial steps of grammatical gender learning and this might launch the masculine as the default. As soon as the feminine subclass has been properly built this asymmetry would disappear but the masculine would then keep its default status. With respect to this, it is noteworthy that occurrences of masculine determiners vastly outnumber those of feminine determiners in 20- to 30-month-olds’ production, while the magnitude of the difference reduces to approximately 60% vs 40% occurrences in 3-year-olds (Bassano, Maillochon, & Mottet, 2008). Whether a 16% (58% vs 42%) difference in masculine vs feminine nouns’ type frequencies would then be sufficient to sustain the default or whether this could even provide the appropriate explanation for it deserves further clarification.

Footnotes

Acknowledgements

We would like to thank the heads of public schools in the Montpellier area for their support and the many children, adolescents and young adults for their participation. We are grateful to Chloe Marshall and two anonymous reviewers for many stimulating and helpful comments. We would also like to thank Aurélia Julien, Céline Castejon, Cécile Larroque and Marie Pouplin for their help in running pilot versions of the experiment and Mélanie Cabanillas, Maëlys Rebouillat, Amandine Hernandez, Anne-Laure Bret, Charlotte Botella and Elodie Rey for their help in running the experiments.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

References

Arias-Trejo

Alva

E. A.

(2012). Early Spanish grammatical gender bootstrapping: Learning nouns through adjectives. Developmental Psychology. Advance online publication. doi:10.1037/a0029621.

Babineau

Shi

(2011). Processing of French liaisons in toddlers. In Danis

Mesh

Sung

(Eds.), Proceedings of the 35th annual Boston University conference on language development (Vol. 1, pp. 25–37). Boston, MA: Cascadilla Press.

Bassano

Maillochon

Mottet

(2008). Noun grammaticalization and determiner use in French children’s speech: A gradual development with prosodic and lexical influences. Journal of Child Language, 35, 403–438. doi:10.1017/S0305000907008586.

Berent

Pinker

Shimron

(1999). Default nominal inflection in Hebrew: Evidence for mental variables. Cognition, 72, 1–44.

Berent

Pinker

Shimron

(2002). The nature of regularity and irregularity: Evidence from Hebrew nominal inflection. Journal of Psycholinguistic Research, 31, 459–502.

Boloh

Escudier

Royer

Ibernon

(2012). Determiners and nouns’ endings in French children’s gender attribution choices. First Language, 32, 398–410. doi:10.1177/0142723711422621.

Boloh

Ibernon

(2010). Gender attribution and gender agreement in 4- to 10-year-old French children. Cognitive Development, 25, 1–25. doi:10.1016/j.cogdev.2009.09.011.

Braine

M. D. S.

(1987). What is learned in acquiring word classes: A step toward an acquisition theory. In MacWhinney

(Ed.), Mechanisms of language acquisition (pp. 65–87). Hillsdale, NJ: Erlbaum.

Brooks

P. J.

Braine

M. D. S.

Catalano

Brody

R. E.

Sudhalter

(1993). Acquisition of gender-like noun subclasses in an artificial language: The contribution of phonological markers to learning. Journal of Memory and Language, 32, 79–95.

10.

Bybee

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

11.

Carroll

(1989). Second-language acquisition and the computational paradigm. Language Learning, 39, 535–594.

12.

Carroll

(1995). The hidden dangers of computer modelling: Remarks on Sokolik and Smith’s connectionist learning model of French gender. Second Language Research, 11, 193–205.

13.

Chevrot

J. P.

Fayol

(2001). Acquisition of French liaison and related child errors. In Almgren

Barrena

Ezeizabarrena

M. J.

Idiazabal

MacWhinney

(Eds.), Proceedings of the 8th conference of the International Association for the Study of Child Language (pp. 760–774). Somerville, MA: Cascadilla Press.

14.

Corbett

(1991). Gender. Cambridge, UK: Cambridge University Press.

15.

Cyr

Shi

(2010). Grammatical gender categorization in infants. In Franich

Iserman

K. M.

Keil

L. L.

(Eds.), Proceedings of the 34th annual Boston University conference on language development (Vol. 1, pp. 102–113). Boston, MA: Cascadilla Press.

16.

Demuth

Weschler

(2012). The acquisition of Sesotho nominal agreement. Morphology, 22, 67–88. doi:10.1007/s11525-011-9192-7.

17.

Hare

Elman

(1995). Learning and morphological change. Cognition, 56, 61–98.

18.

Hockett

C. F.

(1958). Two models of grammatical description (Journal of the Linguistic Circle of New York. 1945–1985). New York, NY: International Linguistic Association.

19.

Karmiloff-Smith

(1979). A functional approach to child language: A study of determiners and reference. Cambridge, UK: Cambridge University Press.

20.

Levy

(1983). It’s frog all the way down. Cognition, 15, 75–93.

21.

Levy

(1988). On the early learning of formal grammatical systems: Evidence from studies of the acquisition of gender and countability. Journal of Child Language, 15, 179–187.

22.

Levy

(1997). Autonomous linguistic systems in the language of young children. Journal of Child Language, 24, 651–671.

23.

Lyster

(2006). Predictability in French gender attribution: A corpus analysis. Journal of French Language Studies, 16, 69–92. doi:10.1017/S0959269506002304.

24.

MacWhinney

(2000). The childes project: Tools for analyzing talk. Mahwah, NJ: Lawrence Erlbaum.

25.

Maratsos

M. P.

(1983). Some current issues in the study of the acquisition of grammar. In Mussen

(Ed.), Carmichael’s manual of child psychology (4th ed., pp. 707–786). New York, NY: John Wiley & Sons.

26.

Marcus

Brinkmann

Clashen

Wise

Pinker

(1995). German inflection: The exception that proves the rule. Cognitive Psychology, 29, 189–256.

27.

New

Brysbaert

Veronis

Pallier

(2007). The use of film subtitles to estimate word frequencies. Applied Psycholinguistics, 28, 661–677. doi:10.1017/S014271640707035X.

28.

New

Pallier

Ferrand

Matos

(2001). Une base de données lexicales du français contemporain sur internet: LEXIQUE. L’Année Psychologique, 101, 447–462. Retrieved from http://www.lexique.org

29.

Pinker

(1984). Language learnability and language development (Reprinted with a new introduction, 1996). Cambridge, MA: Harvard University Press.

30.

Plunkett

Marchman

(1993). From rote learning to system building: Acquiring verb morphology and connectionist nets. Cognition, 48, 21–69.

31.

Rey-Debove

Rey

(Eds.). (1993). Le Nouveau Petit Robert. Paris, France: Dictionnaires Le Robert.

32.

Rey-Debove

Rey

(Eds.). (1999). Le Robert junior illustré [Edition CD-ROM]. Paris, France: Bureau Van Dijk.

33.

Rondal

J. A.

(1985). Adult–child interaction and the process of language understanding. New York, NY: Praeger.

34.

Rondal

J. A.

Bachelet

J. F.

Peree

(1985). Analyse du langage et des interactions verbales adulte-enfant. Bulletin d’Audiophonologie, 6, 507–537.

35.

Rumelhart

D. E.

McClelland

J. L.

(1986). On learning the past tense of English verbs: Implicit rules or parallel distributed processing? In McClelland

J. L.

Rumelhart

D. E.

, and the PDP Research Group (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition, Vol. 2: Psychological and biological models (pp. 216–271). Cambridge, MA: MIT Press.

36.

Séguin

(1969). Les marques du genre dans le lexique français écrit contemporain: Compilation des cas et essai de classement. Montréal and Québec, Canada: Mémoire de DES, Département de Linguistique, Université de Montréal.

37.

Seigneuric

Zagar

Meunier

Spinelli

(2007). The relation between language and cognition in 3- to 9-year-olds: The acquisition of grammatical gender in French. Journal of Experimental Child Psychology, 96, 229–246. doi:10.1016/j.jecp.2006.12.003.

38.

Shi

Melançon

(2010). Syntactic categorization in French-learning infants. Infancy, 15, 517–533. doi:10.1111/j.1532-7078.2009.00022.x.

39.

Slobin

D. I.

(1985). Crosslinguistic evidence for the language-making capacity. In Slobin

D. I.

(Ed.), The crosslinguistic study of language acquisition, volume 2: Theoretical issues (pp. 1157–1256). Hillsdale, NJ: Lawrence Erlbaum.

40.

Smith

K. H.

(1969). Learning co-occurrence restrictions: Rule learning or rote learning? Journal of Verbal Learning and Verbal Behavior, 8, 319–321.

41.

Stemberger

J. P.

(1995). Phonological and lexical constraints on morphological processing. In Feldman

L. B.

(Ed.), Morphological aspects of language processing (pp. 247–267). Hillsdale, NJ: Lawrence Erlbaum.

42.

Suppes

Smith

Léveillé

(1973). The French syntax of a child’s noun phrases. Archives de Psychologie, 42, 207–269.

43.

Tucker

G. R.

Lambert

W. E.

Rigault

A. A.

(1977). The French speaker’s skill with grammatical gender: An example of rule-governed behavior. The Hague, The Netherlands/Paris, France: Mouton.

44.

Waxman

Markow

D. B.

(1995). Words as invitations to form categories: Evidence from 12- to 13-month old infants. Cognitive Psychology, 29, 257–302.