ABSTRACT
Many foreign language learners, especially students at the level of secondary or tertiary education who are learning to write in the target language, want feedback on the grammatical quality of the sentences they produce. This raises the question how Intelligent Computer-Assisted Language Learning (ICALL) systems can provide feedback on the grammatical structure of their L2 sentences-for instance, in essay writing exercises. The usual Natural-Language Processing (NLP) approach to this problem is based on parsing. After the student has typed a sentence, the parser evaluates it and provides feedback on the grammatical quality. However, the more errors a sentence contains, the less accurate the feedback tends to be: A parser working with a large lexicon and a rich grammar usually fi nds many correction options but has no criteria to select the option that fi ts the message the student wishes to express. A related problem is caused by ambiguity. Hardly any sentence can be parsed unambiguously (cf. the proverbial Time fl ies like an arrow, for which Wikipedia lists no less than seven different interpretations). Hence, it is notoriously diffi cult to produce highly reliable feedback based on the parsing results.
