A Combined approach for automatic identification of multi-word expressions for Latvian and Lithuanian

Abstract

We discuss an experiment on automatic identification of bi-gram multiword expressions (MWE) in parallel Latvian and Lithuanian corpora. Raw corpora, lexical association measures (LAMs) and supervised machine learning (ML) are used due to the scarceness and quality of lexical resources (e.g., POS-tagger, parser) and tools. Combining LAMs with ML works well for other languages, our experiments show that it perform well for Lithuanian and Latvian as well. We analyse and discuss frequency thresholds in terms of potential MWE and LAMs values. Finally, combining LAMs with ML we have achieved 98,8% precision and 57,5% recall for Latvian and 96,9% precision and 61,8% recall for Lithuanian.