Abstract
Construction grammar (CG) has been proposed as an adequate grammatical formalism for building intelligent language tutoring systems because it is highly compatible with the learning strategies observed in second language learning. Unfortunately, the lack of computational CG implementations has made it impossible in the past to corroborate these proposals with actual language tutoring prototypes. However, recent advances in Fluid Construction Grammar (FCG) now offer exciting new ways of operationalizing robust and open-ended language processing within a CG approach. This paper demonstrates its adequacy for ICALL applications through a case study on error diagnosis in the domain of Spanish tense, aspect and modal morphology. The performance of the FCG tutor is tested on the Spanish Learner Language Oral Corpus (SPLOCC 2). This first FCG Spanish error diagnostic prototype achieves an accuracy of 70% on a total of 500 conjugation errors in four oral tasks carried out by 20 low intermediate and 20 advanced English learners of Spanish. Follow-up experiments will test this prototype on larger learner corpora of differing proficiency levels.
Acknowledgments
This research has been made possible by a basic research grant of the Flemish Fund for Applied Science (IWT 489). It is because of the devotion of the FCG teams at the Artificial intelligence Laboratory in Brussels and the Sony Computer Science Laboratory in Paris that the framework can be applied today in a CALL setting. Finally, a special thanks to the director of these labs, Luc Steels, for his unconditional support and creative ideas in the onset of this work.
Notes
1. For recent developments and interactive web demos the interested reader is referred to www.fcg-net.org.
2. You find a list explaining all abbreviations used in the text at the end of this paper.
3. Without subscript, “errors” refers to the general notion of an error.
4. The first element of a meaning predicate.