Towards Parsing Language Learner Utterances in Context

Intelligent computer-assisted language learning (ICALL) systems give feedback on grammatical errors based on the syntactic analysis of an utterance. However, parsing a language learner sentence in isolation can result in multiple interpretations, which even humans may not be able to resolve. For deriving a helpful error diagnosis, an interpretation that is consistent with the intended meaning is needed. To achieve this, we propose an approach that models information about the context of a sentence by means of conceptual relationships and integrates them into the decision procedure of a syntactic-semantic parser. This way, we obtain a syntactic structure which complies best with a given context model. We identify and categorize cases where the context influences both the structure and the error diagnosis derived from it.