Developing tools for the linguistic description of referential continuity in a corpus of French and Italian school writing: a didactic perspective

What distinguishes a coherent text from a less coherent or incoherent one? How can these characteristics be described objectively? Text comprehension and production rely on several linguistic levels. Among the phenomena contributing to textual cohesion and coherence, reference chains play a central role. A reference chain consists of at least three linguistic expressions that denote the same referent within a given text. While much research has focused on coreference and anaphora in corpora of standard written texts, few resources have been developed to analyse these phenomena in learners’ writing. Furthermore, the study of school writing using computational linguistics remains challenging. This is due both to such texts deviating from the standard norm, which complicates annotation, and to the constraints involved in compiling such corpora. The development of such corpora is time-consuming and costly, which explains their scarcity. Nevertheless, these resources are essential for linguistic description and research in writing didactics. In this context, the following research questions arise: How can school writings be annotated to account for referential continuity? Can this description be achieved using tools based on existing literature on coreference? And what impact do language and educational level have on the construction of reference chains? To address these questions, this thesis makes use of the Scolinter corpus, a longitudinal, comparable collection of school writing from France, Italy and Spain. It comprises around 4,000 pieces of writing by primary school pupils, collected using the same instructions, then digitised and processed according to shared protocols. The study involved annotating a French and Italian sub-corpus from Scolinter comprising 148 French and 150 Italian texts produced by pupils in Years 2 and 3 of primary school. The annotation, based on a guide inspired by previous works on coreference, underwent a validation procedure, resulting in a reliable gold-standard corpus. The analyses combine quantitative and qualitative approaches from a comparative and didactic perspective. The results highlight recurring patterns in reference chains’ structure. The texts are structured around a limited number of referents, whose maintenance relies on strategies situated along a continuum ranging from pronominal to nominal expressions. Pupils tend to favour explicit strategies to reduce ambiguity when multiple referents coexist. Referents are primarily indicated using noun phrases, whereas less informative forms (pronouns and zero anaphora) are used to maintain those referents. Short mentions predominate, and their grammatical type varies according to their position in the chain. While cross-linguistic differences emerge in the distribution of grammatical types, variation across school levels remains limited. These differences are partly explained by text length and density, with Italian texts being more elaborated, as well as by language-specific morphosyntactic properties. This research also highlights the relevance of reference chains for understanding cohesion in pupils’ writing by identifying both regular patterns and difficulties, such as referential ambiguities and breaks in continuity, that could affect interpretation, especially in cases of competition between salient referents. These findings open new possibilities for teaching writing by integrating referential continuity from the earliest school years, as well as for developing larger-scale annotated resources.