The evolution of language models, particularly transformer-based architectures like BERT and Sentence-BERT, has significantly impacted natural language processing (NLP), introducing advancements in text analysis, translation, and content generation. These models excel in understanding the subtle nuances of language, including multi-word expressions (MWEs) like compound nouns, which are known for their complex semantic properties. Our research investigates the ability of transformer encoder models to discern and interpret the distinct meanings of compositional compound nouns, crucial for tasks such as semantic analysis and information retrieval. By conducting a systematic comparison of similarity scores among compound nouns and exploring the effects of modifier and head-noun manipulation, we uncover insights into the models’ capabilities to comprehend the semantic depth of these expressions. Our findings suggest that transformer encoders can capture the nuances of compound nouns. Our work highlights the potential of these models to advance our understanding and processing of natural language, promising significant improvements in search technologies and the broader domain of NLP.