Amharic Idiomatic and Literal Phrase Dataset for Transformer-Based Idiom Identification

Dataset Description This dataset supports the study entitled “Transformer-Based Identification of Amharic Idiomatic Expressions: A Comparative Evaluation with Explainable AI.” It consists of Amharic idiomatic expressions and literal phrases collected and prepared for the development and evaluation of transformer-based text classification models. Each record contains three fields: label_id, representing the numerical class label; text, containing the Amharic phrase; and category, indicating whether the phrase is idiomatic or literal. The dataset is intended to support comparative evaluation of transformer models for distinguishing idiomatic expressions from literal language in Amharic. It also facilitates research on low-resource language processing, idiom identification, model interpretability, and explainable artificial intelligence (XAI). Researchers may use the dataset to investigate classification performance, compare alternative modeling approaches, and explore linguistic patterns associated with idiomatic and literal expressions, subject to the applicable data-use permissions and licensing conditions.