Publication
Entropy2Vec: Crosslingual Language Modeling Entropy as End-to-End Learnable Language Representations
MRL 2025, an EMNLP 2025 workshop ยท November 2025
Abstract
Entropy2Vec derives cross-lingual language representations from the predictive entropy of monolingual language models. The paper studies whether that uncertainty reflects structural similarity between languages, producing dense language embeddings that can be evaluated against established typological categories and multilingual NLP tasks.
Citation (ACL style)
Patrick Amadeus Irawan, Ryandito Diandaru, Belati Jagad Bintang Syuhada, Randy Zakya Suchrady, Alham Fikri Aji, Genta Indra Winata, Fajri Koto, and Samuel Cahyawijaya. (2025). "Entropy2Vec: Crosslingual Language Modeling Entropy as End-to-End Learnable Language Representations." Proceedings of the 5th Workshop on Multilingual Representation Learning (MRL 2025), 426-437. Association for Computational Linguistics.