Cross-Phoneme Generalisation of Dimension-Based Statistical Learning for Stop Voicing: Probing Subject Design and...

Area of Science:

• Psycholinguistics
• Auditory Perception
• Speech Processing

Background:

• Adult listeners adapt their reliance on acoustic cues for spoken word recognition.
• Dimension-based statistical learning (DBSL) adjusts cue weighting based on statistical deviations.
• DBSL is known to be context-sensitive, particularly for stop voicing cues like Voice-Onset Time (VOT) and fundamental frequency (F0).

Purpose of the Study:

• To investigate the generalization of DBSL across different places of articulation for stop voicing.
• To examine DBSL under enhanced learning conditions, including an additional contrast and reduced word frame variability.
• To determine if DBSL for stop voicing is specific to the learned place of articulation.

Main Methods:

• A conceptual replication (Experiment 1) using a between-subjects design with two word-frame contexts.
• Generalization experiments (Experiments 2 and 3) probing cross-place of articulation learning using between-subjects and within-subjects designs.
• Exposure to statistical deviations in the covariance between VOT and F0 cues for stop voicing.

Main Results:

• No evidence of DBSL generalization across place of articulation in any of the four generalization sub-experiments.
• Larger learning effect estimates were observed in between-subjects experiments with enhanced learning conditions compared to previous studies with variable word frames.
• Learning effect estimates were larger for -ALE compared to -ILL word frames, indicating context-sensitivity.

Conclusions:

• Dimension-based statistical learning (DBSL) for stop voicing is highly specific to the place of articulation.
• Lexical or word-frame context significantly influences the degree of DBSL observed.
• The findings highlight the limited generalizability of perceptual learning in speech processing across different phonetic contexts.