Automated Phraseology Extraction and Cultural Factors: An Experiment

This paper reports the results of an experiment with the Parseme 1.1. dataset for English. While the Parseme initiative represented a breakthrough in computational phraseology, it also raised a number of theoretical and practical issues. In this experiment, an attempt is made to improve the results obtained for English, by having recourse to external resources, in the form of a large web corpus. At the same time, attention is paid to the subtle interaction between linguistic tradition, culture and the manipulation of linguistic data in a supervised model for the automatic extraction of verbal multiword expressions. The results show that our algorithm, relying on an open track with external linguistic data, scores better in terms of recall, while deep learning systems yield a better precision. At various stages of the supervised model, the experiment shows that cultural factors play a crucial role