The Use of NLP Techniques in CALL for the Diagnosis of Specific Errors Made by Learners of French

This thesis investigates the use of Natural Language Processing, (NLP) in Computer Assisted Language Learning (CALL) and its possible contribution to the learning experience of students. The specific question that it addresses is whether NLP techniques can help diagnose problematic learner errors, identified in secondary research and through teaching experience. As accurate error detection and the generation of relevant feedback are considered to be particularly important for successful language learning, projects that incorporate parsers in error detection and soft parsing in feedback are examined. This project employs a parser that provides morphological, syntactic and semantic analyses of natural language and develops a French concept dictionary. With these in place, morphological analyses of complex forms such as heavily inflected verbs are derived from base forms (lexical entries) and numerous affixes. The syntactic properties of verbs and their complements are examined in the context of the processing of clitic pronouns. The softening of the relevant constraints is explored with a view to diagnosing errors and producing appropriate feedback. By combining morpho-syntactic information from the sentence with a knowledge-base and an inference engine, it is possible to tackle temporal and pragmatic questions that require real world knowledge. There is an indication of how the parser, when harnessed in a CALL environment, could provide useful diagnosis of errors and relevant student feedback in simulated substitution or translation exercises (source and target sentences)