This paper presents FipsOrtho, a spell checker targeted at learners of French, and a corpus of learners’ errors which has been gathered to test the system and to get a sample of specific language learners’ errors. Spell checkers are a standard feature of many software products, however they are not designed for specific language learners’ errors. After a brief review of the state of the art, we describe the system’s architecture and interfaces. Then we describe our error typology and detail the techniques used to retrieve words and to order proposals appropriately: alphacode, phoneticization, ad-hoc, capitalization, apostrophe, and word separation error methods. Proposals are sorted by a score depending on the method(s) used to retrieve them, on the expected lexical category, gender, number and person, and on the string proximity with the unknown word. Then the test results are presented: a list of individual words containing errors was submitted to the alphacode and phoneticization methods; a corpus of authentic learners’ errors was gathered and analyzed. Finally we conclude the paper with some limitations of the system and ideas for future research.

Cet article présente une recherche, menée dans le cadre du projet FreeText, qui avait pour but de développer un système de diagnostic automatique d'erreurs pour apprenants du FLE. Après un bref résumé des caractéristiques principales du projet ainsi que du corpus d'apprenants récolté et exploité dans ce cadre, l'article traite du système de diagnostic d'erreurs, en particulier de deux de ses composantes, (1) un vérificateur syntaxique qui utilise deux techniques différentes pour détecter les erreurs de nature purement grammaticale et (2) un outil de comparaison de phrases qui compare la réponse de l'apprenant avec celle correcte stockée dans le système, dans le but de détecter des différences sémantiques possibles, comme les référents ou l'usage d'un mot. Puis nous comparons les résultats obtenus par notre système avec ceux d'un système commercial. Enfin nous discutons des avantages d'un système automatique de ce type et des pistes pour une recherche future.

This paper presents an overview of the research conducted within the FreeText project to build an automatic error diagnosis system for learners of French as a foreign language. After a brief review of the main features of the project and of the learner corpus collected and used within the project, the paper focuses on the error diagnosis system itself and, more specifically, on two of its components: (a) a syntactic checker making use of two different diagnosis techniques to detect errors of purely grammatical nature and (b) a sentence comparison tool which compares learners' answers with those stored in the system to detect possible semantic discrepancies such as referents or word usage. Advantages of such an automatic system and ideas for further research are then discussed.

No abstract

No abstract