Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Individual Differences in Pattern Completion Support Lexical Bias
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
Julia Drouin1, Alexandria Swaine2; 1University of North Carolina at Chapel Hill
Introduction: Listeners rely on lexical context cues to categorize acoustically degraded speech signals, resulting in speech categorization functions that favor real-word interpretations. However, there is substantial individual variability in how lexical context and acoustic input are weighted during speech perception, such that some listeners demonstrate stronger lexical bias. For example, listeners with hearing loss who use cochlear implants show heightened lexical bias, a finding also observed in listeners with normal hearing categorizing noise-vocoded speech, suggesting that lexical information is increasingly upweighted under acoustic uncertainty. While these findings are often described in terms of cue-weighting strategies, these accounts do not specify the neurocognitive mechanisms that support perceptual reconstruction of degraded speech signals. One candidate mechanism is pattern completion, a hippocampal-dependent process that supports reconstruction of missing information by mapping partial input onto stored representations. We hypothesized that individuals who show stronger reliance on pattern completion would demonstrate greater lexical bias during categorization of degraded speech. Methods: Participants (n = 85) completed four phonetic categorization tasks and a Memory Image Completion (MIC) task which measures individual differences in visual pattern completion. In the speech task, normal-hearing listeners categorized clear speech and 8-channel noise-vocoded speech along two /b/-/g/ continua (BAP–GAP; BACK–GACK), with lexical bias quantified as the Ganong shift. In the MIC task, participants identified whether partially degraded images of familiar scenes corresponded to previously learned or novel scenes, indexing the tendency to complete incomplete visual input based on prior memory, which varied across individuals. Results: We observed a robust effect of acoustic degradation on lexical bias, with significantly greater Ganong shift for vocoded speech (M = 19.79%, SE = 2.34) compared to clear speech (M = 6.16%, SE = 1.12), indicating increased reliance on lexical context under reduced acoustic detail. In the MIC task, participants showed increased reliance on learned representations as image completeness decreased, consistent with stronger pattern completion under conditions of visual uncertainty. Critically, individual differences in pattern completion were related to speech perception behavior: participants with stronger pattern completion exhibited greater lexical bias during degraded (vocoded) speech categorization (p = 0.037), but not clear speech (p = 0.163). This condition-specific relationship suggests that pattern completion is selectively engaged when acoustic input is insufficient, aligning perceptual reliance on lexical knowledge with a memory-based reconstruction process. Discussion: Together, these findings suggest that variability in speech perception under degraded listening conditions may reflect differences in memory-based reconstruction mechanisms. Our findings suggest individuals who more readily rely on pattern completion mechanisms to resolve visual ambiguity also rely more strongly on lexical context to categorize degraded speech. These results have implications for understanding individual differences in speech perception in noise and under signal degradation, including variability in outcomes among listeners using cochlear implants and other hearing assistive technologies.
Topic Areas: Speech Perception, Multisensory or Sensorimotor Integration