Applying natural language processing technology to CALL

Background:  If students are to be allowed to produce their own sentences and errors in response to CALL exercises, the CALL program must be smart enough to detect the full range of errors the student is likely to produce.  Canned (multiple choice or fill-in-the-blank) exercises cannot do this.  Natural language processing technology is required
 

Research question:  Is it possible to apply natural language processing technology to CALL  in a practical manner that yields demonstrable pedagogical benefits in the classroom?

Answer:  yes.

Suggested methodology/comments: Write a parser-driven CALL program using natural language processing technology.  Systematically vary the user interface and perform empirical experiments to determine what sort of exercise yields the greatest pedagogical benefits, both in terms of comprehension and production. 

Contact:  Noriko Nagata   nagatan@usfca.edu

Submitted August 2002

Reader Comments: --

August 6, 2003: Stephen Nightingale   nakidorioka@yahoo.co.uk   Parser driven language checking sounds good, but so far there is no deterministic parser which has been developed which can recognize enough of the difficult cases to be anything other than a toy. Indeed, parser based Machine Translation is giving way to Statistical and Example based methods, recognizing that the best examples of language are human developed.  Similarly for monolingual Language processing, corpus based methods are coming to dominate the field. I propose therefore frequency based analyses of complex sentential structures, to make sure that these be favoured over simple SVO, or Subject-Predicate-Adjunct examples.