Contextual semantic effects in compound processing

This study investigates how semantic transparency influences the phonetic realization of triconstituent nominal compounds (NNN) in English. Building on earlier work that employed averaged (static) embeddings, we use contextualized BERT representations derived from the exact carrier and context sentences presented to participants in a production experiment. This method aligns computational measures of semantic similarity with the experimental context guiding speakers’ interpretations. Our analyses reveal a branching-sensitive asymmetry, indicating that contextualized semantic information affects phonetic realization selectively, depending on structural configuration. More broadly, the results highlight how computational embeddings can enrich experimental data by capturing semantic dimensions that are not directly accessible through traditional methods.