Despite the lack of invariance problem (the many‐to‐many mapping between acoustics and percepts), human listeners experience phonetic constancy and typically perceive what a speaker intends. Most models of human speech recognition (HSR) have side‐stepped this problem, working with abstract, idealized inputs and deferring the challenge of working with real speech. In contrast, carefully engineered deep learning networks allow robust, real‐world automatic speech recognition (ASR). However, the complexities of deep learning architectures and training regimens make it difficult to use them to (...) provide direct insights into mechanisms that may support HSR. In this brief article, we report preliminary results from a two‐layer network that borrows one element from ASR, long short‐term memory nodes, which provide dynamic memory for a range of temporal spans. This allows the model to learn to map real speech from multiple talkers to semantic targets with high accuracy, with human‐like timecourse of lexical access and phonological competition. Internal representations emerge that resemble phonetically organized responses in human superior temporal gyrus, suggesting that the model develops a distributed phonological code despite no explicit training on phonetic or phonemic targets. The ability to work with real speech is a major advance for cognitive models of HSR. (shrink)

A classic problem in cognitive science concerns how listeners perceive and understand speech as comprised of discrete words. We propose a Syllable Inference account of spoken word recognition and segmentation, under which alternative hierarchical models of syllables, words, and phonemes are dynamically posited from cues that include current and past speech rate, with a goal of maximal prediction of sensory input. Three experiments using the Visual World eye‐tracking paradigm provide evidence supporting our proposal.

In three experiments, we examined priming effects where primes were formed by transposing the first and last phoneme of tri‐phonemic target words (e.g., /byt/ as a prime for /tyb/). Auditory lexical decisions were found not to be sensitive to this transposed‐phoneme priming manipulation in long‐term priming (Experiment 1), with primes and targets presented in two separated blocks of stimuli and with unrelated primes used as control condition (/mul/‐/tyb/), while a long‐term repetition priming effect was observed (/tyb/‐/tyb/). However, a clear transposed‐phoneme (...) priming effect was found in two short‐term priming experiments (Experiments 2 and 3), with primes and targets presented in close temporal succession. The transposed‐phoneme priming effect was found when unrelated prime‐target pairs (/mul/‐/tyb/) were used as control and more important when prime‐target pairs sharing the medial vowel (/pys/‐/tyb/) served as control condition, thus indicating that the effect is not due to vocalic overlap. Finally, in Experiment 3, a transposed‐phoneme priming effect was found when primes sharing the medial vowel plus one consonant in an incorrect position with the targets (/byl/‐/tyb/) served as control condition, and this condition did not differ significantly from the vowel‐only condition. Altogether, these results provide further evidence for a role for position‐independent phonemes in spoken word recognition, such that a phoneme at a given position in a word also provides evidence for the presence of words that contain that phoneme at a different position. (shrink)