Metaphors are ubiquitous in human communication; they serve to enhance the conveyance of ideas between individuals. By clarifying and drawing parallels between various concepts, metaphors deepen and enrich our language and expression. However, with the rise of cross-cultural communications and a globalized economy, to better understand these nuances, we rely on Large Language Models (LLMs), which are currently limited in metaphorical and analogical comprehension. With the introduction of encoder-only transformers, such as BERT, prior studies have demonstrated superior metaphor detection performance compared to earlier machine learning techniques in recent years. In this paper, we propose DisBERT, a transformer-based model for metaphor detection that introduces a novel Word Sentence Discrepancy (WSD) module. Rooted in Black’s linguistic theory of metaphor interaction, WSD reflects the semantic divergence between a word’s semantics and its sentence context. Evaluated on four standard datasets—VUA18, VUA20, MOH-X, and TroFi—DisBERT demonstrated consistent improvements over its underlying model and achieves results competitive with state-of-the-art models. Overall, our findings suggest that DisBERT generalizes well and indicate that incorporating WSD is a promising approach for word-level metaphor detection ( t(238) = −2.20, p = .028*, 95% CI [−0.0018, −0.0001], d = −0.28 ).