Syllable Inference as a Mechanism for Spoken Language Understanding

Area of Science:

• Cognitive Science
• Psycholinguistics
• Auditory Perception

Background:

• Spoken word recognition faces challenges due to continuous speech signals.
• Listeners must segment continuous speech into discrete words.

Purpose of the Study:

• To propose and test a syllable inference account of spoken word recognition and segmentation.
• To investigate how temporal information and speech rate influence word perception.

Main Methods:

• Developed hierarchical generative models for syllables, words, and phonemes.
• Utilized the visual world eye-tracking paradigm with a picture-selection task.
• Manipulated speech rate (time-compression/expansion) in acoustically ambiguous sentences.

Main Results:

• Listeners dynamically posited hierarchical models to predict sensory input.
• Perception of words emerged from minimizing prediction errors.
• Speech rate influenced the probabilistic interpretation of speech segments (singular vs. plural).

Conclusions:

• The syllable inference account explains how listeners segment continuous speech.
• Temporal information, contextual speech rate, and language model predictions are crucial for word perception.
• This framework elucidates the continuous, probabilistic nature of spoken language understanding.