Speech perception

Friday, September 30th,   2011 [17:20 - 19:20]

PS_1.086 - Is phonological knowledge on linguistic restrictions universal? A French-Japanese cross-linguistic approach

Maïonchi-Pino, N. ¹ , Takahashi, K. ¹ , Yokoyama, S. ¹ , Écalle, J. ² , Magnan, A. ² & Kawashima, R. ¹

1 Institute of Development, Aging and Cancer - Smart Ageing International Research Center - Functional Brain Imaging - Tohoku University - Japan
2 Laboratory of Cognitive Mechanism Studies - Institute of Psychology - Lyon University - France

We present results from a cross-linguistic comparison between native French- and Japanese-speaking adults which aimed at examining whether phonological knowledge on linguistic restrictions in speech perception is universal. We used two syllable counting tasks within pseudowords. In Experiment 1, we manipulated onset cluster sonority profiles to compare 15 French adults to 15 Japanese adults. Our results evidence that listeners from both languages systematically misperceive universal phonotactically-illegal (marked) onsets as phonotactically-legal ones (unmarked; /rpal/ misperceived as /rəpal/). Phonological repairs decreased as onset phonotactic legality increased (/rpal/ > /klal/) in both languages. In Experiment 2, we manipulated intervocalic cluster sonority profiles within syllable boundaries to compare 15 French listeners to 15 Japanese listeners. Of interest is that we highlight a reversed pattern following the universal markedness within syllable boundaries. French and Japanese adults misperceive intervocalic clusters disrespecting optimal syllable contact (/afmal/ misperceived as /afəmal/). Here, phonological repairs decreased as intervocalic clusters came to respect the optimal syllable contact (/aklal/ > /arpal/). Ours is a significant contribution demonstrating that listeners exhibit universal phonological knowledge on phonotactic restrictions both in languages that have (French) or do not have (Japanese) clusters. In both experiments native acoustic-phonetic properties have no straightforward influence on phonological repairs.

PS_1.087 - Phonological variation affects lexicalization of newly learnt words

Spinelli, E. ^{1, 3} , Sumner, M. . ² & Johnson , K. . ³

1 Laboratoire de Psychologie et NeuroCognition, Université Pierre Mendès France, Grenoble, France
2 Department of linguistics, Stanford University, USA
3 Department of linguistics, University of Berkeley, USA

Upon hearing two variable forms for the concept of “20”, twenty (citation form) vs. twenny (nasal-flap form), do listeners form independent phonological representations for each word, associated with the same concept? Using the novel word-learning paradigm showing that newly-learnt non-words (e.g., lantobe) are lexicalized after a night of sleep and compete with similar words (e.g., lantern; Gaskell & Dumay, 2003), we test predictions made by a multiple-variants storage account of phonological variation. Specifically, English native speakers learned novel-words (e.g., advantape) that were either non-variable (all items were presented in the citation form only) or variable (all items were presented in both the citation and nasal-flap forms). One day after the learning phase, training on non-variable stimuli inhibited the recognition of words with shared overlap (e.g., advantage; replicating Gaskell & Dumay, 2003), but training on variable stimuli did not inhibit the recognition of these words. Because each novel-word received the same amount of exposure in both conditions, our results suggest that the nasal-flap and citation forms associated with the new lexical entries were not stored and not as one unique citation form. Rather, surface forms of novel-words seemed to be stored independently, each receiving not enough exposure for subsequent inhibition effect.

PS_1.088 - Cheese and socks on audio-visual pizzas: Modality dependence of N400 effects

Feldker, K. ¹ , Hirschfeld, G. ^{1, 2} & Zwitserlood, P. ^{1, 2}

1 University of Muenster
2 Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience

This event-related potential (ERP) study examines the influence of presentation modality on semantic integration, measured by the N400. Highly constraining sentences were presented audio-visually except for the final word, a noun that was either a highly predictable or an anomalous continuation. This noun was either presented audio-visually (speech and the speakers mouth), auditorily only or visually only. The time course and strength of the context effect (predictable vs. anomalous) differed depending on the modality of presentation. A context effect was present between 200 and 350ms for the audio-visual and the auditory-only conditions, but not for the visual-only condition. The N400 time window (350 to 600ms post-onset) discriminated between predictable and anomalous continuations in all three presentation modalities. The 600 to 800ms time window revealed a context effect for auditory-only and visual-only conditions, but no longer for the audio-visual modality. In contrast to what might be extrapolated from earlier studies on phonological processing, the context effect was stronger in the auditory-only than the audio-visual condition. Our results thus show a modality dependence of the context effect, and, most interestingly, a modulation of the ERPs by context even when the critical word could only be identified by lip reading.

PS_1.089 - Temporal changes in conversational interactions induced by the presence of a simultaneous conversation

Villegas, J. ¹ , Aubanel, V. ¹ & Cooke, M. ^{1, 2}

1 Ikerbasque (Basque Science Foundation), Spain
2 Language and Speech Laboratory, Universidad del Pais Vasco, Spain

This study aims to better understand the changes in foreground conversations induced by background conversations, particularly modifications in the temporal domain including overlaps between foreground and background speech. Understanding the strategies that humans adopt to orally communicate with a peer in the presence of competing dialogs could give some useful insights for developing improved human--computer interfaces, delivering aural information more effectively, etc. In comparison to the acoustic effects of a background dialog in a conversation, our knowledge on background conversation interactional effects is rather limited. In experiments involving simultaneous conversations, we have found intensity and fundamental frequency increments, speech rate decrements, and other changes associated with the Lombard effect in speech produced in the presence of competing talkers. Interactional effects such as greater number of interruptions and dysfluencies, and less accurate turn taking were also seen. Unlike previous studies, we observed no reduction in overlap between foreground and background speech. We hypothesise that this unexpected result could be explained by visual cues used by the subjects during the conversation, methodological differences (i.e., as opposed to free conversations, previous reports focused on task-oriented experiments), stimuli differences (a single competing talker instead of a spontaneous talking pair).

PS_1.090 - Semantic priming at the cocktail party: behavioral and EEG studies

Dekerle, M. ¹ , Boulenger, V. ² & Meunier, F. ¹

1 Centre de Recherche en Neuroscience de Lyon INSERM U1028 / CNRS UMR5292
2 Laboratoire Dynamique Du Langage CNRS UMR 5596

Our studies addressed the issue of auditory masked semantic priming using cocktail party situations. Three behavioral studies showed that semantic relatedness between a multi-talker babble and a target elicited semantic priming. This effect however only appeared when the number of background voices sharing semantic features with the target was higher than the number of voices which did not. In an EEG study, we investigated to what extent the babble is semantically processed by testing the hypothesis that brain indexes of babble semantic processing could be observed even when no priming effect emerged behaviorally. Participants repeated target words embedded in multi-talker babble in which the ratio of semantically related/unrelated voices was 1/2 (no behavioral priming effect) or 2/1. Importantly, within one of the related voices, a semantically unrelated word (deviant) could be inserted. Our prediction was that semantic processing of the babble should be reflected by larger N400 effect for deviants than for non-deviants. Results showed a larger N400 for both unrelated targets and deviants compared to related targets and non-deviants respectively. There was no effect of the number of related voices. Overall these findings suggest that semantic features can be processed in low-intelligibility listening conditions, however intelligibility is necessary.