Enhancing Parsing of European Portuguese Verbal Idioms

Abstract

Verbal idioms (or verbal idiomatic expressions) are multiword expressions in which the main verb is distributionally frozen with one or more of its arguments (subject or complements). For the most part, they convey a non-compositional meaning that cannot be inferred from the individual meanings of their constituents when used separately.
The primary goal of this project is the creation of a system capable of processing verbal idioms from European Portuguese, seamlessly integrated into a natural language processing pipeline.
To this end, two key components were developed: (i) the creation of a corpus annotated for instances of verbal idiomatic expressions in European Portuguese, and (ii) the development of a system that generates dependency parsing rules to identify proper instances of specific verbal idioms based on the linguistic restrictions outlined in a lexicon-grammar matrix.
The system is evaluated using manually produced sentences, artificially generated ones, and real sentences from two selected corpora annotated for verbal idioms. The overall results indicate that the Precision of the system is very satisfactory; whilst the Recall is less favorable, highlighting the need for further efforts to enhance the automatic identification of verbal idioms in texts.
A portion of the annotated corpora is made available to the scientific community.