Enchaînements verbaux - étude sur le temps et l'aspect utilisant des techniques d'apprentissage non supervisé
Fiche du document
- handle: 10670/1.9cxaem
- hal: hal-00153327
info:eu-repo/semantics/OpenAccess
Sujets proches
Hours (Time) Sheltered employment Workshops for people with disabilities Natural science Sciences Science of science CAT (Personality test) Duración Horas (Tiempo) Enfants, Test d'aperception pour CAT Bellak, Test de Test de Bellak Aperception, Test d' pour enfants Centres d'aide par le travail Centres de travail adapté CAT ÉSAT Modes opératoires Techniques Mode opératoire technique technique de fabrication techniquesCiter ce document
Catherine Recanati et al., « Enchaînements verbaux - étude sur le temps et l'aspect utilisant des techniques d'apprentissage non supervisé », HAL-SHS : linguistique, ID : 10670/1.9cxaem
Métriques
Partage / Export
Résumé
Unsupervised learning allows the discovery of initially unknown categories. Current techniques make it possible to explore sequences of phenomena whereas one tends to focus on the analysis of isolated phenomena or on the relation between two phenomena. They offer thus invaluable tools for the analysis of sequential data, and in particular, for the discovery of textual structures. We report here the results of a first attempt at using them for inspecting sequences of verbs coming from sentences of French accounts of road accidents. Verbs were encoded as pairs (cat, tense) – where cat is the aspectual category of a verb, and tense its grammatical tense. The analysis, based on an original approach, provided a classification of the links between two successive verbs into four distinct groups (clusters) allowing texts segmentation. We give here an interpretation of these clusters by using statistics on semantic annotations independent of the training process.