Evaluating word representation for hypernymy relation: with focus on Arabic

Randah Alharbi, Husni A. Al-Muhtaseb

Keywords: Word representation, hypernymy relation, hypernymy specific embedding, hypernymy detection.

TL;DR: Evaluating different types of word embedding for modeling hypernymy relation

Abstract: Hypernymy relation is one of the fundamental relations for many natural language processing and information extraction tasks. A key component of the performance of any hypernymy-related task is word representation. Traditional word embeddings capture word similarity but fall short of representing more complex lexical-semantic relationships between terms, such as hypernymy. To overcome this, recent studies have proposed hypernymy-specific representations. In this study, we conduct an evaluation of several types of word representations to determine the most effective approach for modeling hypernymy relationships in Arabic. We use an Arabic training corpus and several datasets to assess traditional embedding, hypernymy-specific embedding, and contextual embedding across several hypernymy-related tasks, including hypernymy detection. The results indicate that different embeddings have different effects on the performance. Moreover, the performance is affected by the selected datasets. This highlights that there is a need for further research to develop more robust word representation and benchmark datasets.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 14112