NLP Corpus Observatory – Looking for Constellations in Parallel Corpora to Improve Learners’ Collocational Skills

Abstract

The use of corpora in language learning, both in classroom and self-study situations, has proven useful. Investigations into technology use show a benefit for learners that are able to work with corpus data using easily accessible technology. But relatively little work has been done on exploring the possibilities of parallel corpora for language learning applications. Our work described in this paper explores the applicability of a parallel corpus enhanced with several layers generated by NLP techniques for extracting collocations that are non-compositional and thus indispensable to learn. We identify constellations, i.e. combinations of intra- and interlingual relations, calculate association scores on each relation and, based thereon, a joint score for each constellation. This way, we are able to find relevant collocations for different types of constellations. We evaluate our approach and discuss scenarios in which language learners can playfully explore collocations. Our explorative web tool is freely accessible, generates collocation dictionaries on the fly, and links them to example sentences to ensure context embedding.