Conversational dialogue is not only shaped by acoustic features and timing but also by the evolving semantic content exchanged between interlocutors. While prior work, such as Ellag et al. (2024), demonstrated that background noise alters turn-taking dynamics, it did not consider what speakers say. Yet, semantic cues have been shown to play a critical role in speech perception, especially under adverse listening conditions, where contextual predictability can facilitate intelligibility. In this study, we analyze the recordings collected in Ellag et al. (2024) to examine the semantic structure of dialogues between native-English speakers in both quiet and noisy environments during free-form and task-based interactions. Our goal is to understand how noise and task demands influence semantic coherence, alignment, and predictability across conversational turns. Using natural language processing techniques, we analyze phrase-level embeddings and word predictability to explore how meaning shifts within and between speakers under different conditions. This approach allows us to investigate whether noise leads to more constrained or elaborative speech, and how it affects the alignment of semantic trajectories across interlocutors. Together, these findings will offer new insights into how environmental challenges influence both the structure and coherence of conversational interaction.
Stepanenko et al. (Wed,) studied this question.