Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Earlier sensory encoding for surprising words in connected speech

Poster Session B - Sunday, March 8, 2026, 8:00 – 10:00 am PDT, Fairview/Kitsilano Ballroom

David Hernández-Gutiérrez¹ (), Shyanthony Synigal¹, Judy Thompson¹, Edmund Lalor¹; ¹University of Rochester

Understanding spoken language involves hierarchical processing, from early sensory analysis of sound to higher-level linguistic comprehension. Predictive neural models such as temporal response functions (TRFs) have shown that these stages are not independent. Yet, it remains unclear how different levels interact in neural timing. Recent work from our lab (Dou et al., 2025) found that TRFs to word surprisal—resembling an N400-like response—are faster for more predictable words. Here, we test whether this effect extends to earlier sensory processing. Specifically, we asked whether acoustic TRFs that track the speech envelope also show shorter latencies for predictable words, suggesting that higher-level predictions can influence the temporal dynamics of low-level auditory encoding. We reanalyzed EEG data from Broderick et al. (2018), in which 19 participants listened to ~60 minutes of an audiobook. Word surprisal and entropy were computed and grouped into four levels (very low, low, high, very high). Separate TRF models were built for each group within each linguistic complexity metric, focusing on acoustic TRFs while controlling for word-level features. Envelope TRFs showed shorter latencies for highly surprising words compared to very predictable ones—contrary to our initial hypothesis. This pattern replicated in a second dataset (N = 21). No latency shifts were found for word entropy, suggesting that the effect may relate specifically to prediction error rather than contextual uncertainty. Although additional analyses are needed to exclude attentional effects, these findings suggest that surprisal shapes the temporal dynamics of early auditory encoding, highlighting top-down interactions during naturalistic speech comprehension.

Topic Area: LANGUAGE: Lexicon