Syllable Inference as a Mechanism for Spoken Language Understanding

A classic problem in spoken language comprehension is how listeners perceive speech as being composed of discrete words, given the variable time-course information continuous signals. We propose a syllable inference account word recognition and segmentation, according to which alternative hierarchical models syllables, phonemes are dynamically posited, expected maximally predict incoming sensory input. Generative combined with current estimates context rate drawn from neural oscillatory dynamics, sensitive amplitude rises. Over time, result local minima error between predicted recently experienced signals give rise perceptions hearing words. Three experiments using visual world eye-tracking paradigm picture-selection task tested hypotheses motivated by this framework. Materials were sentences that acoustically ambiguous numbers they contained (cf. English plural constructions, such “saw (a) raccoon(s) swimming,” have two loci grammatical information). Time-compressing, or expanding, materials permitted determination temporal at, of, each locus affected looks to, selection pictures singular referent (e.g., one more than raccoon). Supporting our account, probabilistically interpreted identical chunks consistent degree was based on chunk's gradient relation its context. interpret these results evidence arriving information, judged model predictions generated evaluated scale, informs inferences about thereby giving perceptual experiences understanding words separated time.