Learning about phraseology from corpora: A linguistically motivated approach for Multiword Expression identification

Area of Science:

• Computational Linguistics
• Natural Language Processing

Background:

• Multiword Expressions (MWEs) present significant challenges in Natural Language Processing (NLP).
• Verbal MWEs, specifically verb+noun MWEs (VNMWEs), are particularly difficult to identify due to their complex morphosyntactic flexibility.

Purpose of the Study:

• To develop and evaluate a linguistically motivated method for extracting detailed information about VNMWEs from corpora.
• To assess the adaptability of the method to different languages, with a primary focus on Spanish.

Main Methods:

• Utilized monolingual and parallel corpora as input data.
• Extracted data on the morphosyntactic variability of VNMWEs.
• Tested the extracted information in a VNMWE identification task.

Main Results:

• Achieved an F score of 0.52 in the VNMWE identification task.
• The obtained F score is considerably higher than that of related work in the field.

Conclusions:

• The proposed linguistically motivated method effectively gathers information on VNMWEs.
• The method demonstrates potential for improving the accuracy of VNMWE identification in NLP applications.