ALUMCoCo : Mesuring the influence of lexical context on the sense annotation task
Résumé
Lexical sense annotation, known as the lexical disambiguation task, is not always easy. It is even considered complex for both manual and automatic annotation. Confronted with an annotation instance such as the one we implement, the annotator - whether human or automatic - is confronted with a given textual context En cette saison, le __soleil__ se lève tard. in which he must associate a given occurrence (soleil) with a sense (I.a « L’astre autour duquel tourne la Terre, considéré depuis la Terre. [Le soleil s’est levé à 8h13.] »), taken from a given inventory of senses (Réseau Lexical du français https://www.ortolang.fr/market/lexicons/lexical-system-fr/). Each of the elements (context, occurrence, sense and inventory of senses) has its own characteristics, which can be used to establish different levels of complexity. The ALUMCoCo project (https://perso.atilf.fr/sollinge/projets/alumcoco/) focuses on the issue of textual context complexity. Here we present the project's methodology: preparation of an experimental protocol; online data collection from non-expert speakers (time to select a sense from a given inventory for a given context, feeling of complexity in carrying out the task, diversity of annotations proposed in the same context); quantitative and qualitative analysis of results.
Domaines
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |