Referential continuity in learner’s manuscripts

Résumé En

This presentation introduces an analysis of the co-reference chains in the ResolCo Corpus, a corpus of French student manuscripts produced at school. The ResolCo Corpus is made up with schoolwork written in reply to a dedicated instruction which has been designed for causing strategies in terms of discourse coherence. The ResolCo instruction consists in asking the students to write a narrative by inserting in it the three following predefined sentences:1. She was living in this house for a long time2. He turned around when he heard this great noise3. Since this adventure, the children no longer go out at night.Each sentence contains, amongst others, the mention of a brand-new referent (she, he, the children), co-reference via pronominal anaphora or demonstrative NP and a specific tense. Face to this instruction, the student is confront to the necessity to resolve cohesion problems such as co-referential continuities, trigger and encapsulation, sequence of tenses, generic vs. specific mood (Garcia-Debanc et al. 2017). The first step in studying co-reference chains through all the education levels consists in annotating these discourse structures in the ResolCo corpus. Our annotation process draws upon two French projects that have resulted in the publication of two large annotated standard written French corpora: ANNODIS (Asher et al. 2017) and DemoCRAT (Lattice et al. 2019). As for the ANNODIS and DemoCRAT, the annotation process is computer-assisted and aims at detecting all the mentions signalling the co- referential chains.Because texts are not written by learners and may present substantial problems in terms of cohesion and interpretability, annotation guidelines used for ANNODIS and DemoCRAT must be adapted. As a consequence, we decided to focus the annotation on the three human referents (e.g. she, he, the children). This enables to lead comparative analyses through all the education levels due to the fact that these three chains must occur in all the texts, be they very short and simple texts as those collected in elementary school or long and more complex as those collected at high school and University. In addition, such restriction facilitates the annotation process by asking the coder to focus only on mentions about “she”, “he” or “the children”. We also added a feature for annotating fuzzy mentions for letting the coder free to annotate problematic cases as in the following example providing the beginning of a 5th grade text (our approximate translation)."This house was the home of elderly people. She was living in this house for a long time. In this house they heard a great noise. This came from the locked door behind which an horrible and terrifying monster was. He turned around when he heard this great noise. They no longer go out because they saw bad guys outside..."The presentation will present the collected annotations and give results of a first analysis of the co-referential chains through education levels.Asher, N., Muller, P., Bras, M., Ho-Dac, L.-M., Benamara, F. (2017). ANNODIS and related projects: case studies on the annotation of discourse structure. In Nancy Ide and James Pustejovsky (eds). Handbookof Linguistic Annotation, Springer Netherlands, pp.1241-1264Garcia-Debanc C., Ho-Dac L.-M., Bras M. et Rebeyrolle J. (2017). Vers l'annotation discursive de textes d'élèves. Corpus, (16)LATTICE, LILPA, ICAR & IHRIM (2019). Democrat. ORTOLANG (Open Resources and TOols for LANGuage) – www.ortolang.fr.