IdiomX: A large-scale English–Arabic dataset for idiom detection, semantic understanding, and cross-lingual NLP

IdiomX is a research-oriented Natural Language Processing (NLP) project that investigates the translation of English idiomatic expressions using transformer-based neural models.

A web browser extension helps to highlight and translate texts, as well as interact with LLMs.

This project pioneers advancements in Natural Language Processing (NLP) by introducing a deep neural network approach to address challenges in interpreting multi-word sentences, detecting idioms, and classifying text. Focusing on text classification, it evaluates various models, with BERT outperforming others with an F1 score of 0.7980.

Detection Without Localization: Compositional Generalization Failures in Transformer Models for Hebrew Idiom Boundaries. M.Sc. thesis - 4,800-sentence dataset, 6 transformers, 5 LLMs.