Towards the Construction of Language Resources for Greek Multiword Expressions: Extraction and Evaluation

Multiword Expressions have been posing problems for Natural Language Processing systems for many years. Their automatic identification has, as a result, been in the focus of NLP research for almost two decades now. The advances in the most widely spoken languages like English, French, German, etc. are remarkable and have been extensively documented. This paper presents our work towards the creation of a dictionary of Multiword Expressions for Greek using automatic extraction and human validation. We investigate the use of a knowledge-poor statistical approach based on four association measures. The results obtained by these measures on the Greek Europarl corpus are compared and contrasted with those obtained by the same measures using the web as a corpus. The manual evaluation of the results by Greek native speakers shows that the automatic approach performs well enough to help in the construction of a lexical resource, despite of the difficulty of the task.