📄 Proceedings of the Joint Workshop on Multiword Expressions and Universal Dependencies (MWE-UD) @ LREC-COLING 2024

Summary

These are the proceedings of the joint MWE-UD workshop, held on 25 May 2024 in Turin as part of the LREC-COLING 2024 conference. For the first time, two communities held a joint event: researchers of multiword expressions (MWEs: idioms, compounds, phrasal verbs, light-verb constructions, collocations) and developers of Universal Dependencies. They share common problems. UD has dedicated syntactic relations for MWEs, but there is no agreement on the annotation criterion: syntactic irregularity or semantic non-compositionality. Annotation practice also differs across treebanks and across languages. In the preface, the editors explain why the workshops were merged, report on paper selection and name the main themes. These include extending language coverage through annotated data in a standard format, and whether large language models understand idiomatic meaning. The volume also covers the usefulness of UD parsing for downstream tasks, the relationship between syntax and morphology, and automatic MWE identification. For Ukrainian NLP, the volume is useful as an overview of current approaches to consistent MWE annotation in UD-compatible corpora, in particular within the PARSEME and UniDive initiatives.

Key points

• 53 submissions were received: 29 long, 15 short and 9 non-archival. 19 long, 7 short and 9 non-archival papers were accepted, giving an acceptance rate of 59% for archival papers. One long paper was withdrawn by its authors after acceptance.
• Distribution by track: UD — 8 of 12, MWE — 9 of 16, MWE+UD — 9 of 16.
• The MWE workshop series has existed since 2003, so this event marks its 20th anniversary. For the UD series, launched in 2017, this is the seventh workshop. UD annotation has already been applied to more than 100 languages.
• The PARSEME corpora contain verbal MWE annotation for 26 languages, aligned with UD. The COST Action UniDive (CA21167) aims to harmonize PARSEME and UD practices and to extend annotation to non-verbal MWEs.
• Languages covered by the papers: Arabic, Hindi, Ancient Egyptian, Vedic, Northern Kurdish, Slovenian, Dutch, Bavarian, Gujarati, Saraiki, Swedish, Turkic and South Asian languages. Organizers and sponsors: ACL-SIGLEX (funded the participation of 2 attendees) and UniDive (covered travel and accommodation for 29 attendees).