The emergence of Information Structure in child speech: the acquisition of c’est-clefts in French

2.1 C’est-clefts in child and adult French

2.2 Usage-based theory of language acquisition

The usage-based constructionist literature posits that early child syntax is not as abstract as adult-speech, and that linguistic categories are built and not innate (Diessel 2013; Tomasello 2000, Tomasello 2003, 2015). A large body of research shows that children’s first multi-word utterances are not built from scratch, i.e., by combining words based on abstract representations of word orders present in their language as previously claimed (Pinker 1984; Wexler 1998). Rather, many early multi-word utterances seem to correspond to chunks. These chunks can be frozen, which means there is no variation in the child’s speech in the elements composing the chunks (Lieven et al. 1997). Examples of frozen chunks could include Let me see, I don’t know, What’s that?

Some multiword utterances, called frames-with-slots (Dąbrowska and Lieven 2005), may exhibit variation. Dąbrowska (2014: 621) defines frames-with-slots as ‘recurrent form-meaning combinations containing a “slot” into which novel material can be inserted’. Examples can include I wanna X in which the slot represented by the X can be filled with a paradigm of possible linguistic items.

Numerous studies support the claim that early multi-word utterances are either chunks or frames-with-slots, such as corpus analyses of children’s early production of recycled chunks of language which were previously uttered or experienced through parental input (see Dąbrowska 2014). For example, Lieven et al. (2003) investigate the speech production of one child, between age 2;0. and 2;11. The authors use the traceback method, which consists in comparing the utterances produced in a given recording with previous recordings. In this precise study, the authors compared 295 multi-word utterances produced by the child with those produced within a time span of six weeks. They report that 63% of her utterances correspond exactly to an utterance she uttered previously. Among the 37% other utterances, 74% require only one change, such as the addition or switch of a word, to match an utterance produced previously. In a study on four two-year-old English-speaking children, Lieven et al. (2009) document, again using the traceback method, that between 78 and 92% of children’s utterances match an utterance produced in a previous recording within a period of six weeks.

Research using experimental settings also demonstrate that young children store frequently occurring chunks. Bannard and Matthews (2008) show that two- and three-year-old children are more likely to produce multi-word sequences correctly if those sequences are more frequent in their parental input. Arnon and Clark (2011) also demonstrate that English four-year-old children produce fewer irregular plural mistakes (e.g., feet, teeth) if they are produced within a frequent word sequence.

Frames-with-slots are already the result of some initial degree of abstraction and categorization, for which children require two cognitive skills: (i) intention-reading and (ii) pattern-finding (Tomasello 2000, 2003, 2015). The abstraction process is assumed to unfold as follows: as children store several fixed chunks in their memory, they might notice similarities in the meaning or context of use of some chunks through their intention reading skills. Then, they use their pattern-finding skill to identify a common element between several fixed chunks. This common element between chunks of similar meaning becomes the fixed part of the frame-with-slot, and the section which differs across chunks becomes the slot (Tomasello 2003). For example, a child might store the fixed chunks I wanna eat and I wanna drink, and identify that they carry a similarity in meaning, which is to request to do something. He can then notice the similar section I wanna in both chunks, creating the frame-with-slot I wanna X that he might use to ask for something.

To this date, studies investigating the development of categories from chunks or frames-with-slots typically focus on argument categories. Other types of categories, such as information structure categories, remain unexplored. In Jourdain et al. (2020), we provided a first analysis of the emergence of the topic category in child French, through the study of frames-with-slots in dislocations. In that study, we showed that French children first use different frames-with-slots prior to developing a more abstract and general construction for dislocation. For example, we showed that some children produce the frame X ça ‘X that’, with the function of expressing the property of a referent (7). Other frequent frames-with-slots are à X ça ‘to X ça’, used to express the possessor of a referent (8) or où l’est X ‘where is X’ to ask for the location of an entity (9).

Interestingly, most functions associated with the frames-with-slots identified for early dislocations involve conveying a sort of property to an object (between 60 and 70% of dislocations, depending on the child). Considering that topics are defined as what the sentence is about (Lambrecht 1994; Reinhart 1981), producing frames to provide a property about an entity can be considered as a cognitive precursor for the topic-comment articulation. Agents were very rare, accounting for only between 7 and 15% of dislocated constituents, and they occurred mostly in later recordings, while agents have been shown in the literature to be quite frequent in child speech early on (Bambini and Torregrossa 2010). Nevertheless, as far as I am aware, no such study has been carried out to investigate categorizations preceding the development of the focus category in child speech.

I will attempt to tackle this gap in the remaining sections of this paper. I will investigate whether early clefts can be described in terms of chunks, exhibiting lexical and functional patterns with a low level of abstraction. This will help determine whether French children have a focus category early.

3.1 Corpus

I investigated the development of clefts in child French through a corpus study. I selected the Lyon corpus collected between 2002 and 2005 (Demuth and Tremblay 2008), available on the CHILDES platform (MacWhinney and Snow 1990). I extracted data from three children from that corpus, Anaïs, Marie and Nathan. These three children were selected because their files included a synchronisation between the text and video files. The presence of the video was important to determine the functions of children’s utterances, necessary to identify potential precursors of clefts.

The children were recorded every other week on average for 1 h at their home, interacting with a caregiver, and in rare cases for Anaïs, with a caregiver and her older sister. The interactions consist in spontaneous everyday life situations, such as playing with toys or games, or shared book-reading.

3.2 Data extraction

It is not possible to claim with certainty which early utterances of the child belong to the category of clefts, simply because such an abstract category cannot be present from the onset (Tomasello 2000, 2003). This category emerges from more concrete utterances, schemas, or patterns, which may also support the development of other constructions. Therefore, to understand the early development of clefts in child speech, I need to extract not utterances that I believe are already instantiation of the cleft construction, but rather utterances whose syntactic and functional properties can be interpreted as supporting the development of clefts.^[4]

As for syntax, the research of De Cat (2007) documents that early clefts are produced as a c’est-clause, sometimes followed by a right-dislocated constituent (10).

Hence, I considered all utterances with the third person singular copula est as potentially supporting the development of clefts, either produced with the subject pronoun c’ or without.

After having extracted all the c’est-clauses, I considered which of these utterances were produced in a context or with a function that have the potential to support the emergence of an abstract cleft category. The literature on adult French shows that c’est-clefts cannot have an identificational function (Lambrecht 2001b). Identificational copular clauses are produced to provide the name or label of an entity (Higgins 1979). Therefore, c’est-clauses that children use to label an entity do not have functional properties that can support the future development of c’est-clefts. Such functions of c’est-clauses are especially frequent when the adult is asking the child to say what or who she is currently seeing, as in (11).

Not all c’est-clauses used to provide the label of an entity answer a direct question of the adult as in (11). Hints that suggest that a c’est + LABEL combination is not used with the function to simply label an entity include looking, pointing or gesturing at a different entity or towards an action than the entity or event being labelled. For example, in (12) below, the child labels herself, but is keeping her eye gaze towards the stickers. The child is neither pointing towards herself nor looking at a picture of herself. The moi ‘me’ is therefore not simply used to label herself.

Alternatively, if the child is looking, pointing, or gesturing at the same entity as the one being labelled, I considered such instances as cases of labelling, and excluded them from the rest of the analysis. For example, in (13), the entity being labelled by the child is the sole attention of focus while being labelled.

I also excluded all predicational uses of est or c’est. Such clauses are characterized by a referential subject and a property or categorization of the referent expressed by the subject (Declerck 1988; Heycock 2012). Because the subject of predicative c’est-clauses is necessarily referential, it cannot be a case of cleft (Doetjes et al. 2004; Karssenberg 2018; Lambrecht 2001b). I considered a c’est-clause to be predicational if the element following c’est was providing the property of an entity. For example, in (14), the child is looking at an entity, and saying that it is damaged. This is a clear association between an entity and a property, which entails that the pronoun c’ is referential and therefore cannot be a precursor of clefts. All c’est + PROPERTY clauses were discarded from the final dataset.

An overview of the extracted data is provided in Table 1. Anaïs produces her first c’est-clause at age 1;10.27, Marie at age 1;8.19 and Nathan at age 2;3.6.

In addition to these data from child speech, I extracted c’est-clefts from the input of Anaïs following a similar extraction procedure as for children, excluding labelling c’est-clauses. Even though Anaïs does not produce c’est-clauses from the first recordings, I extracted c’est-clefts from all recordings for the adult data, for a total of 343 c’est-clefts.

3.3 Data analysis

A large proportion of early clefts in child French are produced only with a c’est-clause, without a relative clause (Lahousse and Jourdain 2020):

Considering that the copula in clefts does not typically undergo inflection (Blanche-Benveniste 2002), the c’est part of clefts can be considered as fixed, even in adult speech. If the relative clause is not produced, and the c’est expression is fixed, the only part of clefts that might exhibit variation is the item following directly the c’est expression, which I will refer to as the clefted constituent.

The first step of this analysis is to determine whether c’est-clefts in child French undergo a ‘chunk’ stage, i.e., with the element following c’est being lexically fixed, or whether they show a frame-with-slot pattern, with the element following c’est being free. I will also consider whether children are using both chunks and frames-with-slots, depending on their communicative intents or context. For this, I will first investigate the degree of productivity of clefts through the distribution of clefted constituents.

The next step is to identify whether the array of items that children produce as clefted constituents exhibit a form of concrete categorization, that can be interpreted as a precursor of the adult focus category. For this, I assessed potential semantic regularities across the different clefted constituents. In the study on the acquisition of topic in French children’s dislocations in Jourdain et al. (2020), we show that in the frames-with-slots functioning as precursors of the topic-marking dislocation construction, children produce mostly items referring to inanimate entities whose thematic function is theme. I therefore decided to check whether (i) animacy and (ii) the thematic function have an impact on the development of focus.

For animacy, I differentiated between human referents and other types of animate referents, due to the very high number of human referents in my data (see the section below). As for the thematic roles, I used the same categorization as Bambini and Torregrossa (2010), themselves based on the semantic categories identified in child speech by Brown (1973) and Slobin (1985). The list of the thematic roles that I identified for the clefted constituents of my data, along with definitions and examples, is provided in Table 2.

4.1 Distribution of clefted constituents

The frequency of each item following the copular expression c’est is provided in Table 3. The most frequent item produced in clefts is moi ‘me’ for all children and is produced in over 70% of the clefts of Anaïs and Nathan. Marie produces moi less frequently, in only 32% of her early clefts. Marie also produces the largest number of different items in her early clefts: 35 in total, against 9 for Anaïs and 5 for Nathan.

To determine whether the children of the Lyon corpus start producing clefts from a fully frozen chunk c’est moi, rather than from a frame-with-slot, I compared the production of moi in clefts with the other items produced through time. For Marie, considering the lower percentage represented by the main item, I also included the second item in the comparison, qui ‘who’. The results are provided in Figures 1 –3.

Figure 1:

Developmental distribution of the slot-filler moi versus other fillers in the clefts of Anaïs.

Figure 2:

Developmental distribution of the slot-fillers moi and qui versus other fillers in the clefts of Marie.

Figure 3:

Developmental distribution of the slot-filler moi versus the other fillers in the clefts of Nathan.

In the data of Anaïs, the item moi is frequent in all recordings. In the data of Marie, which exhibit greater variety, moi has varying frequency between recordings, with no clear developmental pattern: moi is very frequent at age 2;5.16, not documented at age 2;9 in this corpus, but is very frequent again at age 2;10.24. Therefore, despite moi being most frequent, children exhibit variation to some extent quite early on. The construction c’est X seems to function as a frame-with-slot. In the next section, I will investigate clefted constituents’ semantic properties, to determine whether the free slot exhibits some form of early categorization.

4.2 Semantic properties of the slot fillers

The first semantic dimension that I analysed is animacy. The main striking element, common to all children, is that most items in clefts refer to human entities. For Anaïs, humans represent 94,7% of items in her clefts, 79% in the data of Marie, and 94,7% in the data of Nathan (16).^[5]

Another important property is that most items referring to humans also refer to a discourse participant: 89,4% of all items in the clefts of Anaïs, 51,4% in the data of Marie, and 89,5% in the data of Nathan (17). The developmental overview of the degrees of animacy of items in clefts is provided in Figures 4 –6.

Figure 4:

Discourse participants, human, animate and inanimate items in the clefts of Anaïs.

Figure 5:

Discourse participants, human, animate, and inanimate items in the clefts of Marie.

Figure 6:

Discourse participants, human, animate and inanimate items in the clefts of Nathan.

It is not possible to claim that there is an evolution in the type of items produced in clefts, going from a first stage with only c’est moi or c’est toi to a second stage with c’est X, as far as animacy is concerned. In the data of Anaïs, in the very last recording, the child produces more human items which are not discourse participants than previously. Nevertheless, considering that there is no data for her recordings afterwards, it remains unclear whether this is due to chance or to a developmental path.

The next property that I considered is the thematic role (see Section 3.3 in the methodology section). The results are shown in Figures 7 –9. Due to the very low number of beneficiaries, experiencers and patients, I collapsed these categories in the graphs, for readability.

Figure 7:

Agent versus other thematic roles in the clefts of Anaïs.

Figure 8:

Agent versus other thematic roles in the clefts of Marie.

Figure 9:

Agent versus other thematic roles in the clefts of Nathan.

Most items in the clefts of all children can be interpreted as agents. They represent 91,7% of items in the clefts of Anaïs, 74,3% in the data of Marie, and 89,5% in the data of Nathan (18).

Generally, there does not seem to be a clear increase of other thematic roles in children’s clefts, and the proportion of agent remains high for all children throughout development. Only at age 2;9, Marie produces fewer agents than items with a different thematic role, but she then shifts to producing more agent in later recordings.

To sum up this section, in addition to children producing some specific items in their clefts, their most frequently produced items exhibit similar semantic properties. Most items in clefts refer to agents and to discourse participants themselves. In the next section, I will interpret these clefts in context, to identify possible discourse functions, precursors of the focus-background articulation.

4.3 Functions of early clefts

To understand how IS categories might develop in child speech, it is crucial to consider the more general context, to be able to identify which function the construction might serve. I was able to identify two frequent functions with lower levels of abstraction, (i) the request for an action (presented in Section 4.3.1), and (ii) identifying the agent of an action (Section 4.3.2). I will also discuss remaining cases with no clear discourse function except for the expression of the focus (Section 4.3.3) and will present these functions developmentally in Section 4.3.4.

4.3.4 Developmental overview of early clefts’ functions

A summary of the frequency of each function in the clefts of the children from the Lyon corpus is provided in Table 4.

Table 4:

Frequency of each function in the clefts of the Lyon corpus.

	Request action	Whodunnit	Other function	Unclear	Total
Anaïs	84 (64%)	38 (28%)	2 (2%)	8 (6%)	132 (100%)
Marie	36 (34%)	38 (36%)	30 (29%)	1 (1%)	105 (100%)
Nathan	15 (79%)	2 (11%)	1 (5%)	1 (5%)	19 (100%)

The distribution of the possible functions of clefts is illustrated developmentally in Figures 10 –12.

Figure 10:

Distribution of clefts’ functions in the data of Anaïs.

Figure 11:

Distribution of clefts’ functions in the data of Marie.

Figure 12:

Distribution of clefts’ functions in the data of Nathan.

The functions do not seem to undergo a clear developmental path: clefts with neither of the two functions with low levels of abstraction, requesting to do or someone else to carry out an action or identifying the agent of an action, are not more frequent in later than in earlier recordings. This could be due to the age span of the children investigated, and maybe the development into a more abstract construction occurs later than at age 3.

To conclude the results on child speech, early c’est-clefts seem to exhibit both a fixed chunk (c’est moi) and a frame-with-slot pattern (c’est X). The regularities in the semantic properties of clefted constituents suggest an early form of categorization within the slot that could be a precursor of the focus category. Finally, most instances of clefts (90% for Anaïs and Nathan, 70% for Marie) can be ascribed one of two functions: request for action and whodunnit. In the next section, I will discuss the implications of these results for the acquisition of information structure in child speech.

4.4 Input data

As mentioned in the methodology section, the input data concern the input of Anaïs only. I will compare the input with the children’s result, to identify whether there are similar restrictions in clefted constituents (Section 4.4.1) and functions (Section 4.4.2) in children’s input.

4.4.1 Distribution of clefted constituents

Using similar categories as in child speech, I determined whether in the input of Anaïs, most clefted constituents are moi, or correspond to another discourse participants, or whether they express another human, animate or inanimate entity. The results are shown in Figure 13 below. Due to the higher number of recordings from the input with clefts, I collapsed categories into the input that the child heard either between age 1 and 1,5, between age 1,5 and 2, between age 2 and 2,5 and between 2,5 and 3.

Figure 13:

Distribution of clefted constituents in Anaïs’s input.

As illustrated by this table moi ‘me’ is also frequent in the input, accounting to around 16% of all clefted constituents. Nevertheless, this number is drastically lower than in the child’s data. Toi ‘you’ is actually a more frequent clefted constituents in the input, accounting for around 27% of all clefted constituents.

Finally, inanimate constituents seem to represent the majority of clefted constituents. However, when all categories representing humans are taken together, it becomes clear that the majority of clefted constituents in all age spans refer to a human referent. This could be a potential hint of a semantic preference for clefts.

5.1 Early clefts as chunks

The clefts analysed in the present study can be said to be instances of either a fixed chunk in the case of c’est moi or the frame-with-slot c’est X.

Despite variation being present early, as Marie and Nathan produce other clefted constituents than moi from the first recording and Anaïs from the second recording, the phrase c’est moi represents over 73% of the clefts of Anaïs and Nathan, and 32% of the clefts of Marie. Considering this very high frequency, it is safe to assume that the chunk c’est moi has its own mental representation as such. Nevertheless, the early variation of clefted items, present even though limited, suggests that children have both a chunk and a frame-with-slots as early constructions for c’est-clefts. This multi-representation of a construction at different concreteness levels is documented for the acquisition of different constructions (see for example Dąbrowska and Lieven 2005 on the development of questions) and is not dissimilar to adults’ constructional network either, as adults themselves store frequent chunks along with abstract constructions (see for example Wray 2002, 2009).

In addition to these lexical properties, semantic and functional restrictions also point towards an interpretation of early clefts as a chunk for c’est moi or as instances of the frame-with-slot c’est X. First, the chunk c’est moi is associated with one specific function, to request to do an action. This form-function pairing is both specific and frequent, exhibiting the degree of entrenchment of a fixed chunk. As for the frame-with-slot c’est X, they are mostly associated with the function of identifying the instigator of an event, and with a few cases of children requesting the adult to carry out an action. This function is less abstract than in adult French and limits the type of constituents that can be clefted, yielding the interpretation of a construction with a low abstraction.

5.2 Building a functional network

For each of the two main functions that I identified in early clefts, I found some instances that deviate from the original function while retaining other properties: for the request to do an action, I showed that children may ask someone else present in the physical context to carry out an action. As for the whodunnit function, I showed that instead of an agent, children might use a beneficiary, an experiencer or an entity.

This idea of prototypical and less prototypical functions of constructions is theorized in cognitive linguistics (Geeraerts 1988; Lakoff 1987; Langacker 1987a, 1987b). This notion of prototype stems from the research of Rosch and colleagues (Mervis and Rosch 1981; Rosch 1983), and was aimed to explain how humans form categories. The authors provide the example of the bird category: some birds seem to be more representative of the category than others. The concept of prototype can be extended to constructions. For example, Næss (2007) highlights that a prototype of the transitive construction involves an agent intentionally instigating an action, which results in a patient being affected in some way. The closer exemplars of the transitive construction are to this definition, the closer they are to the prototype. Exemplars which deviate the most from this are less prototypical instances of the transitive constructions.

These early functions of clefts can therefore emerge from a sort of prototype that children identify from their input. Even though clefts in the input seem to be more diverse, almost half of them (45%) express a function related to the association of an agent with an action. Early functions of children’s clefts therefore seem to stem directly from one part of the input, maybe because these clefts from the input have functions which are more similar than others. A more in-depth analysis of the functions of clefts in adult speech and in child-directed speech is however required to assess whether it is possible to identify a network of functions for clefts in adult French showing a strong nucleus, or even prototype towards the identification of agents.

Deviations from the main functions of clefts in child speech provide evidence for the beginning of the creation of a more complex functional network. The fuzziness of the boundaries suggests that the children from the Lyon corpus can extend the original prototypical function of a construction. Nevertheless, considering the very low density of the recordings of the Lyon corpus, it remains impossible to claim from this analysis alone whether the functions’ boundaries are blurred from the start, or whether children first built their prototype before extending the constructions’ functions to more contexts.

5.3 Functions of early clefts: precursors of IS articulations?

The present study aims at identifying the nature of information structure categories in very early child speech. Now that I have identified the form-function properties of early clefts, what do they entail for the acquisition of the abstract function of marking a focus-background articulation found in adult French? Evidence for gradual emergence of the focus category comes from the semantic properties of the clefted constituents, while evidence for the gradual emergence of the focus-background articulation comes from the functions of early clefts.

Most clefted constituents refer to human agents, while in our previous study on dislocation based on the same corpus as the present study (Jourdain et al. 2020), almost all dislocated constituents produced by children are inanimate entities, in total contrast with clefted constituents. This lack of overlap of item categories across different chunks or frames-with-slots is frequently reported in early child speech (Tomasello 1992, 2000, 2003, 2015). This dichotomy could suggest that the IS categories of topic and focus emerge from these more concrete levels of categorization the same way as other linguistic categories.

Next, both functions could be potential precursors of the focus-background articulation. The function of requesting to do an action, which is the most specific function of the two, might be also a precursor of the whodunnit function. Indeed, in the request, the child is identifying an agent to associate with an action, but within the physical situation of the discourse.

In turn, the function of identifying the agent of an action could potentially function as a stepping stone towards the focus-background articulation. In a sentence with a focus-background articulation, the background refers to a general event, situation, or property. The focus allows to identify one element associated with a specific context (event, situation, or property). The whodunnit function could be an over-specific interpretation of the focus-background articulation, by restricting (i) the element identified, an agent, and (ii) the context, an action.

More research is needed to make the bridge between the human agent category and the focus category, and between the whodunnit function and the focus-background articulation, both in adult and child French. Considering that the age span covered by these analyses is quite narrow, more research on longitudinal data on a wider age span would help understand how children develop these initial concrete categories to the same categories as in adult speech.