Reproduce “MWE as WSD: Solving Multiword Expression Identification with Word Sense Disambiguation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Joshua Tanner, Jacob Hoffman
A system developed by Joshua Tanner and Jacob Hoffman unifies multiword expression (MWE) identification and word sense disambiguation (WSD) by reframing MWE identification as a gloss-informed WSD task. This approach, utilizing a novel Distinct Codes Attention (DCA) Poly-encoder, achieves an F1 score of 61.3 on the DiMSUM dataset for MWE identification and largely retains WSD performance.
Key insights
- Unifying MWE identification and WSD into a single, semantically-aware framework by leveraging gloss information improves MWE identification, particularly precision.
- The DCA Poly-encoder architecture effectively differentiates between target MWE components and surrounding context, contributing to improved MWE identification performance.
- Pre-training on a large WSD dataset (SemCor) is crucial for the model to effectively encode context and glosses, significantly boosting downstream MWE identification.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.