Access to informational content of texts by children – TextoKids
Methods of deep machine learning
- Use of a recent machine learning model (Transformers): the results show that the proposed method (currently based on the age groups given by the editors) obtains very good scores, both on sentences and on texts, and even surpasses the predictions of psycholinguistic experts in children's reading comprehension
- Development of a corpus annotated in emotions
Putting the processing chain online
Rashedur Rahman, Gwénolé Lecorvé,Aline Etienne, Nicolas Béchet, Jonathan Chevelu, Delphine Battistelli (2020) - «Mama/Papa, Is this Text for Me?«. in Actes COLING'20 (28th International Conference on Computational Linguistics), 8-13 décembre 2020, Barcelone, Espagne
Gwénolé Lecorvé, Alexis Blandin, Delphine Battistelli, Aline Etienne (2020) - «Age Recommendation for Texts«. in Actes LREC'20 (12th International Conference on Language Resources and Evaluation), 13-15 mai 2020, Marseille, France
Alexis Blandin, Gwénolé Lecorvé, Delphine Battistelli, Aline Etienne (2020) - «Recommandation d’âge pour des textes«. In Actes TALN’20 (Traitement automatique du langage naturel 2020).
Aline Etienne, Delphine Battistelli, Gwénolé Lecorvé (2020) - «L’expression des émotions dans les textes pour enfants : constitution d’un corpus annoté«. In Actes TALN’20 (Traitement automatique du langage naturel 2020)
Aline Etienne, Delphine Battistelli, Gwénolé Lecorvé (2020) - «Apports de la linguistique et du TAL à l'analyse des émotions dans les textes pour enfants«. In Actes de la 3ème édition du Colloque «Langage et éMOTions«, poster, 26-27 novembre 2020, Bordeaux
The TextToKids project aims to develop tools to facilitate children's access to information contained in texts. This involves work both on the production of these texts by adults and on children's access to adapted texts. The target age group is young children, i.e. 7-12 years old. The consortium, which brings together linguists, psycholinguists, computer scientists and specialised journalists, will seek to characterise the psycholinguistic and linguistic constraints (in particular of a temporal and emotional nature) to respecte and to offer support tools (automatic text analysis, information retrieval, reformulation, good practices). The experimental frameworks will be the narratives of current events (for example, the reception of migrants in France or the Oscars) and the implementation of an Internet search engine that respects the linguistic highlighted constraints. In terms of impact, the project works towards a "Children's Internet" and paves the way for other modalities (speech, images) to support the production of multimedia content for children.
Project coordination
Delphine Battistelli (Modèles, Dynamiques, Corpus)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
Partner
SYNAPSE DEVELOPPEMENT
QWANT
IRISA Institut de Recherche en Informatique et Systèmes Aléatoires
MoDyCo Modèles, Dynamiques, Corpus
Help of the ANR 649,310 euros
Beginning and duration of the scientific project:
November 2019
- 42 Months
