Constraints in Free-input Question-Answering Drills

This article describes a set of question-answer drills for language learning for a richly inflected language. The drills have been in actual use for some time. They allow for free input and make use of a constraint-grammar-based system, which anticipates a number of grammatical errors and common misspellings and gives certain response types. The interactions between student and computer are recorded, and the log reveals that the free-input approach comes at a price: students tend to avoid complex constructions. In order to force the student to answer with more complex constructions, while still keeping the free-input approach, we implemented a solution with more constraints for the input. The exercise items are generated and each template gives rise to a huge numbers of exercises. Constraint grammar makes it easy to control for both grammar errors and adherence to the constraints given in the task. The evaluation on authentic learner data shows that constraining the user’s input with the question itself, makes it possible to analyse the student’s free input, with very good precision and recall. But parsing the input is only a part of the challenge of designing real-life ICALL systems. The article discusses other design issues related to question-answering drills.