Disentangling sources of difficulty associated with the acquisition of accusative clitics in French

Highlights

• 3rd person accusative clitics (ACC3) are acquired later than 1st and 2nd person.

• Three properties have been hypothesized to make ACC3 especially complex to acquire.

• Morphological marking for gender has the strongest influence on ACC3 acquisition.

• Optionality of ACC3 has a small, but detectible, role in its acquisition at age 4–6.

• Discourse-participant independence of ACC3 only impacts its acquisition at age 4.

Abstract

Accusative clitic pronouns are acquired later than both nominative and reflexive pronouns in typically developing French-speaking children. However recent research suggests that not all these clitics are equally difficult, 3rd person accusative clitics (ACC3) being more problematic than 1st and 2nd person. In this study, we explored three properties which could make ACC3 complex: (1) gender marking, (2) discourse-participant independent reference, and (3) optionality in spoken French. We have created specific experiments assessing the role of each of these properties in clitic acquisition in 41 French-speaking children aged 4–8. Results show that all three properties play a role in making ACC3 complex, with the strongest influence coming from gender marking. Implications of these results are discussed in light of previous related studies.

Introduction

Accusative clitic pronouns (ACC) have been widely investigated in studies of the acquisition of Romance languages. They have attracted particular interest because of their late acquisition as compared to both nominative and reflexive pronouns in typically developing French-speaking children (Delage, 2008, Hamann et al., 1996, Jakubowicz et al., 1998, Jakubowicz and Rigaut, 2000, Van der Velde, 2003, Zesiger et al., 2010). Their complexity has positioned them as a clinical marker of specific language impairment or ‘SLI’ (Paradis et al., 2003, Parisse and Maillart, 2004) and more generally of atypical language development (Delage, 2008, Tuller et al., 2011). More recently, it has been suggested for Greek, Romanian and Catalan that not all ACC are equally difficult (Avram et al., 2015, Coene and Avram, 2011, Gavarró and Fortón, 2014, Tsimpli and Mastropavlou, 2007). Instead what appears to be problematic is specifically 3rd person clitics, while 1st and 2nd person ACC prove to be relatively simpler. This asymmetry has now also been confirmed for French by a recent study (Tuller et al., 2011). The aim of the present work is to further explore the 3rd versus 1st person distinction for ACC in French to elucidate why 3rd person clitics are associated with delayed development. More specifically, while all ACC share certain properties (listed in a–c in Table 1), those of the 3rd person present additional characteristics which would render them particularly difficult (listed in d–f in Table 1) and which are the focus of this research.

This study considers the impact of each of these properties specific to 3rd person ACC, namely (i) the morphological marking for gender, (ii) the discourse-participant-independent reference (i.e., where reference is not tied to speaker-addressee roles) and (iii) the legitimate omission or ‘optionality’ in specific contexts of spoken French. These properties have been hypothesized to play a role in delaying the acquisition of 3rd person ACC in previous research, which we review below.

Research on the acquisition of French ACC (1) has shown that these emerge more slowly than both (2) reflexive and (3) nominative clitics in typical language development:

Hamann et al. (1996) analyzed the spontaneous language of a French-speaking child from age 2 to 2;10 and showed that while the nominative clitic was present from the first recording (i.e., at age 2;0) and was consistently used in more than half of the child productions, ACC were virtually absent until age 2;6. Moreover at age 2;9, the accusative was found only in 14% of the contexts where it should have been obligatorily realized,¹ contrary to nominative clitics which occurred in 63% of such contexts. A similar delay was observed by Jakubowicz et al. (1998) in an elicited production study of 20 French children aged 5;6 to 5;11. In addition, these authors noted a delay between the accusative as compared to the reflexive clitic (both encoding 3rd person). Thus a hierarchy emerges, with the nominative being acquired before the reflexive, which in turn is acquired before the accusative. Jakubowicz and Rigaut (2000) confirmed this hierarchy with an analysis of natural language corpora as well as with an elicited production task conducted with 12 young children aged 2 to 2;7. Subsequently, the hierarchy was replicated in a study of 99 children aged 3;5 to 6;5 tested specifically with elicited production (Zesiger et al., 2010). This study clearly illustrated the developmental trend of these elements, namely that while nominatives were at ceiling at age 3, reflexives approached this level at age 4 (85%) while ACC only did so at age 6 (90%).

In sum, ACC are produced substantially less than other clitics in early childhood, yielding either frequent omissions of the object altogether, or the use of full lexical DPs (Determiner Phrases) instead. Elicited production experiments for clitics in French show that frequent omissions are indeed coupled with a substantial use of lexical DPs: At mean age 3;2, children omitted ACC 55.9% of the time and produced lexical DPs instead in 20.5% of their utterances (Van der Velde, 2003). This rate tapered off with time, falling at age 4;2 (21.5% omission and 9.3% lexical DPs) and dropping even more sharply by age 6;7 (8.3% omission and 7.3% lexical DPs).

When children do attempt to produce the clitic, it has been shown that they often make mistakes in gender. Zesiger et al. (2010) reported that children aged 3;5 to 6;5 frequently made gender errors on object clitics. The most frequent pattern was the feminine la replaced by the masculine le (26.8%), the reverse being produced significantly less often (10.4%). Tuller et al. (2011) also noted that, in an elicited production task, 17% of accusative clitics produced by children at age 6 contained an error of gender. Most errors occurred when the clitic sequences (nominative + accusative) involved a mismatch in gender. In these contexts, the authors propose that the child's working memory resources are particularly taxed.

Interestingly, while the global comprehension of accusative clitics appears to fare better than their production (Grüter, 2005, Grüter, 2006, Tullerrr et al., 2004, Van der Velde, 2003, Zesiger et al., 2010), difficulty with the gender feature reported for production nevertheless carries over to comprehension. Comprehension of clitics was assessed by Zesiger et al. (2010) via a truth-value judgment task involving a puppet who produced a sentence when describing a picture. This sentence contained an object clitic which either corresponded to the sex of the referent in the picture or not. In this forced two-choice task, children at age 5 showed chance-level performance for detecting when the gender of the accusative clitic in the test stimuli mismatched that of the referent in the picture. Participants only reached 75% error detection at age 6. Once again, a difference between masculine and feminine ACC surfaced, with worse performance for detecting a gender error with feminine as compared to masculine clitics. Grüter et al. (2012) also reported that 4 year-old Spanish-speaking children encounter difficulties in making use of the gender cue on object clitics when identifying their referent in an eye-tracking study. Pirvulescu and Strik (2014) further tested the comprehension of both gender and number on clitic and strong pronouns in children aged 3 to 5 years old, by means of a picture choice task. Their results revealed that children had difficulties with these featural cues when determining antecedents for both clitic and strong pronouns, although these difficulties were more pronounced with clitics. They also showed that the youngest (3 year old) group tended to use the most recently mentioned noun to determine the clitic's reference, suggesting that this group applied another processing strategy than that of relying on the clitic's phi-features. They proposed that the younger children attempted to keep anaphoric dependencies as short as possible due to working memory limitations. Furthermore, these authors also showed that 3–4 year old children perform better with plural accusative clitics, i.e., les, than with singular ones, i.e., le/la. Note that these items differ with respect to gender marking, in that the former but not the latter morphologically encode this feature. Grüter and Crago (2012) investigated the relationship between the production of clitics and working memory in the context of L2 French. These authors showed a negative correlation between working memory (i.e., the digit span task) and the production of clitics by native Chinese-speaking children. The work by Grüter and Crago (2012) is part of a large body of work highlighting that the difficulty with 3rd person ACC extends beyond typical monolinguals to L2 learners (Grondin and White, 1996, Grüter, 2005, Paradis et al., 2003, Prévost, 2006), internationally adopted children (Delcenserie and Genesee, 2013, Delcenserie et al., 2013, Gauthier et al., 2012) and simultaneous bilinguals (Müller and Hulk, 2001, Pirvulescu et al., 2014). Protracted development of this clitic also clearly emerges in various situations of atypical language development, such as Specific Language Impairment (Hamann et al., 2003, Jakubowicz and Tuller, 2008, Paradis et al., 2003, Tuller et al., 2011), hearing loss (Delage, 2008, Delage and Tuller, 2007, Jakubowicz et al., 2000, Tuller and Delage, 2014), as well as epilepsy (Monjauze, 2007). The relationship between the development of clitic production and working memory has often been suggested to exist in a variety of these contexts (see e.g. Pirvulescu and Strik, 2014, Prévost, 2006, Tuller et al., 2011), however, Grüter and Crago (2012) and Mateu (2015) are the only empirical investigations of this link. Mateu has thus demonstrated that omission of accusative clitics correlated with results obtained in a non-word repetition task in Spanish-speaking children aged 2–3. How can we account for this phenomenon? Producing an accusative clitic arguably implies that the processing system retains and links information about two positions: the preverbal position where the clitic is pronounced (or ‘spelled-out’) after syntactic movement, and the post-verbal, canonical, ‘pre-syntactic movement’ position (or the ‘gap’). This sort of cognitive manipulation arguably poses problems for immature cognitive systems and only become accessible to systems with higher computational resources (Booth et al., 2000).

Clitic omission has also been accounted for in terms of the Unique Checking Constraint or UCC (see e.g. Gavarró et al., 2010, Wexler et al., 2003). The UCC predicts clitic omission in languages where the past participle morphologically agrees with the accusative clitic, as is the case in French (e.g. Il l’a peinte, la fille: He her painted + feminine, the girl). This is claimed to stem from young children's systems not being mature enough to carry out such a complex operation. Indeed the DP ‘pro’ which co-occurs with clitics in the canonical object position (see (4)) must raise to spec,ClP by passing also through Spec,AgrOP, because past participle agreement is required (see also Sportiche, 1996):

In these instances, the child system opts instead to not project ClP, so that the ‘pro’ DP only has to check a single D-feature against AgrO, thus giving rise to clitic omission. It is relevant to note that this reasoning predicts that omission should not only occur with 3rd person ACC in French, but also with other clitics such as 1st/2nd person ACC, since past participle agreement occurs with these latter elements as well (Il t’a décrite, Maria: ‘He has you described + feminine Maria’).

In contrast to the bulk of studies discussed above, few investigations have been conducted of clitics in typically developing French other than those of the 3rd person (except for Hamann et al., 1996, who counted all accusative clitics and Tuller et al., 2011 described below). For other Romance languages, a few authors have observed an asymmetric development between 3rd person (5) and 1st/2nd person ACC (6), with the former emerging slower than the latter (Avram and Coene, 2008, Baauw, 2000, Coene and Avram, 2011, Gavarró and Fortón, 2014, Silva, 2010, Tsimpli and Mastropavlou, 2007):

Coene and Avram (2011) analyzed longitudinal corpora of the speech of two monolingual Romanian children (aged 1;05–3;05) and found that in obligatory clitic contexts, ACC of the third person were frequently omitted until age 3 years, while those of the first and second person were always present from the earliest utterances. This asymmetric omission pattern was also observed for very young Catalan-speaking children (Gavarró and Fortón, 2014), for Greek-speaking children with SLI aged 3;5 to 7 (Tsimpli, 2001) and for adult L2 learners (Tsimpli and Mastropavlou, 2007).

Tuller et al. (2011) studied this phenomenon in French-speaking adolescents suffering from different pathologies. These authors conducted an elicited production task with 36 typically-developing (TD) children aged 6 and 11, as well as 71 adolescents (age range 11–20 years) with atypical language development. The clinical population was divided into three groups: SLI, mild-to-moderate hearing loss (MMHL), and Rolandic Epilepsy (RE). Both 1st and 3rd person clitics were evaluated. The authors report that all clinical groups, as well as the youngest TD group (aged 6), showed difficulty specifically with 3rd person ACC. More precisely, the TD 11-year-old group performed at ceiling for all clitic pronouns, while the TD 6-year-olds were at ceiling performance only for 3rd person nominatives (97.4%) and 3rd person reflexives (96.4%) but not for 3rd person accusative clitics which yielded an accuracy of 70.3% (versus 90.6% for 1st person accusative clitics). Similarly, the group with SLI produced 3rd person nominative clitics and 3rd person reflexive clitics at mean rates of 86.6% and 93.6% respectively, while 3rd person accusative clitics were produced with a significantly lower mean rate of 49.7%. Performance by this group with 3rd person ACC also strikes a significant contrast with that of 1st person ACC, which revealed a mean rate of 85%. For MMHL, the mean rate of production of 1st person accusative clitics was high as well, attaining 88.8% and in RE it was as much as 95%, in contrast to the mean production rates for 3rd person ACC, 80.9% for MMHL and 85% for RE. These results attest to the status of 3rd person ACC as a persistent marker of atypical language development stemming from different etiologies. In addition, most crucially for the present study, 3rd person ACC were significantly more subject to avoidance than 1st person ACC in both TD-6-year-olds and in adolescents with atypical language development.

What makes these clitics particularly complex? One possibility is that the 3rd person feature itself is the source of their difficulty, independently of the accusative case. However, the person feature alone cannot account for the difficulty of 3rd person accusative clitics since nominative clitics of the 3rd person are even produced earlier than those of the 1st and 2nd person in very early typical development of French, between the ages of 1;9 and 2;9 (Pierce, 1992, Hamann et al., 1996, Van der Velde, 1999). No differences are reported between the production rates of 1st and 3rd person nominative clitics later, by age 4 (respectively 85% and 88%, Dupuy, 2009). By age 6, children reach a production rate of 98% for production of both 1st and 3rd person nominatives (Tuller et al., 2011), and commit very few errors: In an elicited production task, typically-developing children aged 6 to 11 (N = 36, M age = 8;8) only produced 1.9% of 3rd person nominative clitics with a gender error and 0.6% with a person error (Delage, 2008), contrasting with the frequent gender errors on 3rd person accusatives reported in the literature (as previously described). Given that ceiling performance is observed with 3rd person nominatives by age 6 while difficulty with 3rd person accusative clitics persists (Tuller et al., 2011), the 3rd person feature alone does not seem to be the source of complexity, leading to the conclusion that other properties specifically associated with 3rd person accusative clitics also affect its acquisition, “each of these properties adding complexity, separately and conjointly” (Tuller et al., 2011:437). It is to these specific properties proposed by Tuller et al. which we now turn.

A first potential source of difficulty evoked for the 3rd person ACC is the morphological marking for gender. Indeed these clitics are marked for both gender and number in the singular (le/la) while their counterpart object clitics of other persons only inflect for number. It is worth recalling that 1st and 2nd person ACC do not morphologically encode gender in either singular or plural forms (me/te for singular, vous/nous for plural). Indeed all forms may be used to refer both to masculine or feminine antecedents. Given previous evidence showing the high prevalence of gender errors in studies targeting production of 3rd person ACC, the factor of gender may thus constitute a valuable source of complexity of this element.

Another property of 3rd person accusative clitics is their reference which is characterized by discourse-participant-independence, in contrast to ACC of the 1st and 2nd person which are restricted to discourse-participants for reference and, as a result, are obligatorily animate (Tuller et al., 2011). Discourse-participant independence as such widens the source of possible referents and may thus also be a source of difficulty in the acquisition of 3rd person ACC. Gibson (1998) considers that building a structure for a new discourse referent increases the processing cost, which may lead to “decays in the activations associated with preceding lexical items” (Gibson, 1998:12). This would explain why sentences like (7a), with 1st or 2nd person nominative clitics, are simpler to process (by adults) than (7b) where the referent of the 3rd person clitic is lacking (Gibson and Warren, 1998).

A third property of 3rd person ACC which may influence their acquisition and which is specific to 3rd person ACC as compared to other ACC is that they may be felicitously deleted in spoken French when the reference is sufficiently salient in the discourse (see Fónagy, 1985, Lambrecht and Lemoine, 1996, Lemoine, 1997, Tuller, 2000). Consider the examples (8) below which illustrate this optionality in spoken French (Fónagy, 1985):

In addition to discourse-salience restrictions, object omission is also subject to lexical restrictions (see Fónagy, 1985, Tuller, 2000), as can be observed below (from Tuller, 2000). In other words, third person ACC can be omitted only if they are used with specific verbs.

In sum, the legitimate omission of 3rd person ACC is subject to discourse restrictions, but also to lexical ones, which may require time to be mastered, resulting in prolonged object omission. Children exposed to a variety of null object contexts such as seen above might thus assume that such null objects have a wider distribution than they actually do in the target grammar (Pérez-Leroux et al., 2008). Indeed the object omission stage is attested across languages (Deen, 2006, Jakubowicz et al., 1997, Schaeffer, 1997, Schaeffer, 2000, Wexler et al., 2003, and others). However as Pérez-Leroux et al. (2008) have shown empirically, children learning languages with a complex typology of null objects, such as French, are later to restrict the distribution of these null objects than children learning a language with a simpler object drop typology, such as English. This reasoning can be extended to explain the asymmetry within French between 3rd person ACC and 1st and 2nd person ACC, which are typologically simpler in that they are never legitimately dropped.

The explanations based on the above properties, initially reported by Tuller et al. (2011), suggest that 3rd ACC possess various properties which may have an impact on their acquisition. It has already been shown that a major source of complexity of object clitics is their non-canonical position resulting from internal merge (Belletti, 1999). Other authors have also evoked processing difficulties in chain crossing to explain the difficulty of accusative clitics, as opposed to reflexive ones (See Zesiger et al., 2010). These considerations however do not explain the specific difficulty of the 3rd person ACC as compared to the 1st. Based on this observation, the complexity of 3rd person object clitics is arguably the result of a combination of several properties aside from their non-canonical position. As we have shown, the properties associated with third person accusative clitics that set them apart from 1st and 2nd, include their morphologically marking for gender, their reference not being restricted to discourse participants and their being subject to legitimate omission in certain contexts. The reasoning regarding these properties being complex is as follows: (i) If morphological agreement is part of the difficulty involved in producing third person accusative clitics, because of the need to keep in working memory the gender of the referent, we expect that ACC3, which display gender agreement, would be harder than 1st and 2nd person ACC (ACC1/2), which are void of such agreement. (ii) If establishing non-discourse-participant reference adds difficulty because the larger set of possible referents for ACC3 would tax working memory resources, then we expect ACC3 to be harder to master than ACC1/2 whose set of possible referents is highly restricted, i.e., respectively the speaker or the hearer. (iii) If the legitimate omission of ACC3 in certain contexts increases its difficulty because children become more likely to overgeneralize this omission, we expect more difficulty with the mastery of those contexts that demand the obligatory presence of ACC3, a difficulty which would not exist for ACC1/2, which are invariably obligatory in the input.

In this work we thus propose to assess the empirical effect of gender marking, discourse-participant independence and legitimate omission, by using experimental materials which allow us to disentangle these properties of 3rd person ACC. We predict that each of these properties will have a significant effect on acquisition, and we spell out the nature of the effect expected in our description of each corresponding experiment.