Dr Alfredo Maldonado, a researcher at the ADAPT Centre, recently participated in the Closed Track at the verbal MWE (Multi-Word Expression) 2017 workshop in Valencia, Spain. The competition saw Dr Maldonado take the first-place prize in three languages on token-based evaluation along with a number of second place prizes in various categories. The ADAPT Centre led team presented a system to identify verbal MWEs across 15 of the 18 official European languages.
The language families investigated comprise of a MWE containing a verb used in common language. An example of this type of expression is “have a conversation.” Common phrasal verbs such as “to look up,” “to put up,” or “shut up,” were also included as well as reflexive verb forms casually used in French or Spanish like “se trouver” (to be located), or “se dérouler (to unfold).
Speaking about the importance of his research, Dr Maldonado said: “This work is important in that it is the first time a CRF (conditional random field) sequence model is applied to the identification of verbal MWEs in a large collection of distant languages.” Of the 18 languages studied, Bulgarian, Hebrew, and Lithuanian languages were not investigated due to their lacking morpho-syntactic information—these are words with morphological and syntactic properties.
ADAPT’s team also presented the methodology behind their work. Their verbal MWE identification task was given as a name-entity recognition problem. With use of a state-of-the-art CRF sequencing label algorithm applied, testing proved successful in recognising names entities.
Oftentimes verbal MWEs don’t have a literal meaning. This type of a pattern is evident in the idiomatic expressions like, “it’s raining cats and dogs,” “the cat’s pyjamas,” “kick the bucket,” or “spill the beans.”
Expect to see future studies from this research as it develops through the ADAPT Centre to investigate MWEs focused on language-specific features.
Share this article: