Phraseology Extraction: From Corpora to Deep Learning

Colson, Jean-Pierre [UCL]

AI (Artificial Intelligence) is now present almost everywhere, from weather forecast to medicine and from financial investments to linguistics. Although some have criticized it as a hype whose cycle will soon be reaching its limits, there is no denying that AI has had major implications for applied linguistics in recent years, in particular for MT (Machine Translation). In several fields of applied linguistics, there is now a growing competition between corpus linguistics on the one hand, with a recourse to traditional statistics and huge corpora, and machine learning on the other, in particular deep learning based on neural networks. Will corpus linguistics gradually give way to deep learning? My point of view is that the automatic extraction of phraseology or multiword expressions may provide a partial answer to this question, as I will try to illustrate by a number of recent experiments. In addition, exploring phraseology from the point of view of machine learning, i.e. by drawing conclusions from its algorithmic representation, may provide fresh insights into the interaction between idiomaticity, fixedness and culture.