Abstract
(Englisch)
|
The FreeText project aims at developping a hypermedia CALL system (available both on CD-ROM and via Intra/Internet) for intermediate to advanced learners of French, that relies on communicative approaches to SLA, uses authentic documents, and handles (relatively) free text productions by using NLP tools and error typologies based on statistical learner corpus analysis.Work focused during the project's second year on (1) a large L2 French corpus (collecting and keying-in ca. 250,000 words, error tagging ca. 200,000, in-depth error analyses of specific error categories), (2) didactic content development and CALL pre-formatting (2 tutorials), (3) revisions of the reference grammar, (4) NLP tools, and (5) a second CALL prototype for evaluation and testing purposes.The Dept. of Linguistics at the University of Geneva is in charge of the scientific coordination, the reference grammar, and the NLP tools, and contributes to evaluation and software realisation activities.Scientific coordination We prepared the 2001 Annual Project Review report, attended the related review meeting in Luxembourg, and prepared the follow-up on the experts' recommendations, including a short response of the consortium to the external project evaluation (originally requested by the EC and submitted as D7, Quality Assessment Report), an update of the revised version of the Dissemination and Use Plan (D3), and a first, non-binding version of the Technological Implementation Plan detailing the project's main and secondary outputs. We presented the project at the September 2001 IST Concertation meeting in Luxembourg, attended a restricted meeting called by the Project Officer and involving Softissimo, and prepared the 2nd annual report (D10, due mid-April 2002). Further activities involved the preparation of three PCC meetings (including chairing and follow-up) and three four-way conference calls.Reference grammar The linguistically-motivated reference grammar was essentially finished and its author's contract terminated as scheduled end of April 2001. Revisions had to be performed based on Softissimo's validation comments and were sub-contracted to the grammar's author.NLP tools Work focused mainly on the syntactic and 'semantic' components of the error diagnosis system. Constraint relaxation (phonetic reinterpretation in progress) was used to identify and mark errors designated by the partners as especially relevant for CALL (e.g. agreement, word order, complementation). Currently, 12 error categories are treated and 11 have been tested on learners corpora. The 'semantic' checker compares the pseudo-semantic structures extracted from a model answer and from the learner's input and identifies diverging features in various constructions (involving e.g. passivisation, cleft, raising, etc.). In order to allow Prototype II (November 2001) to integrate more NLP tools, some work planned for the project's 3rd year was anticipated upon: UGEN's existing sentence structure viewer and speech synthesiser were (minimally) adapted to FreeText and a comprehensive XML output was designed to account for all NLP tools. Contribution to software realisation activities We provided Softissimo with bug reports on Prototype II and with support for the integration of our DLLs into the software under development.Contribution to validation activities We identified potential testers for Prototype II and transmitted the relevant information to the partner in charge. We also prepared a short report on our activities within this workpackage.
|
