Toro, J. M. 1 , Pons, F. 2 , Bion, R. 3 & Sebastián-Gallés, N. 1
1 Brain and Cognition Unit, Universitat Pompeu Fabra, Barcelona, Spain
2 Departament de Psicología Básica, Universitat de Barcelona, Barcelona, Spain
3 Departament of Psychology, Stanford University, Stanford, United States
The computation of statistical regularities among syllables has been claimed to be important for various linguistic processes, including the segmentation of speech. Importantly, several aspects of the linguistic signal, including prosodic and phonotactic cues, have been shown to modulate the extraction of such regularities. In order to investigate at what level are those cues affecting the extraction of statistical dependencies, we run a series of experiments in which background linguistic knowledge conflicted with distributional information. We presented participants with an auditory speech stream composed by nonsense words violating a word-forming constraint in their native language (vowel reduction in Catalan). Results showed participants did not prefer the words from foils with smaller statistical coherence in an auditory recognition task (Experiment 1). Nevertheless, participants recognized the words when they were pitted against foils that had not appear in the stream (Experiment 2), or when the test was performed in the visual modality (Experiment 3). These results suggest background linguistic knowledge is not preventing altogether the computation of statistical regularities from the speech stream. Rather, they suggest such knowledge is filtering the recognition of statistically-coherent items during subsequent recognition.