Slide Slam C12 Sandbox Series
Homophonic speech sequences in French: The role of acoustic and contextual cues for disambiguation
Maria del Mar Cordero Rull1, Damien Vistoli2, Stéphane Pota3, Elsa Spinelli3, Fanny Meunier1; 1Université Côte d’Azur, CNRS, BCL, France, 2Université Côte d’Azur, LAPCOS, France, 3Université Grenoble Alpes, CNRS, LPNC, Grenoble, France
Due to the lack of clear word boundaries in spoken language, listeners use tacit knowledge to segment speech. Speech segmentation is proven to be affected by the listeners’ sensitivity to acoustic cues, subphonemic properties (Davis et al. 2002), and contextual information (Norris,1994). Fine acoustic details can influence word boundaries’ perception (Friederici & Wessels,1993). In French, Welby (2003) and Spinelli et al. (2007) showed that listeners used fundamental frequency rises to detect content word beginnings. The use of low-level features to segment speech is particularly salient when contextual information is insufficient (Mattys et al. 2005). Yet, little is known about the cognitive load of exposure to acoustic information not congruent with sentential information. In our study, we investigated the cognitive cost of segmenting ambiguous speech units when context is provided and explored whether and how fine-grained acoustic details can affect semantic processing. We recorded French sentences containing homophonic sequences of article+noun allowing for different segmentations, such as "l’affiche" (“the poster”) and "la fiche" (“the sheet”), both pronounced /lɑfiʃ/. Homophonic sequences were presented in 3 different context conditions: congruent, incongruent, and control. Context conditions were generated by two acoustic manipulations to avoid acoustic differences: cross-splicing yielded contexts that were not congruent with the homophonic sequence included ("La secrétaire médicale a perdu l’affiche du patient", “The medical secretary lost the patient's poster”); congruent sentences were generated after an identity-splicing including the other homophone candidate, here “la fiche”. Control sentences contained a non-congruent, non-homophonic sequence ("Le cordonnier répare la tendresse du directeur", “The shoemaker repairs the tenderness of the director”). In a first experiment, participants were presented with a semantic judgment task focusing on the meaning of sentences, mainly serving as attentional control. To assess semantic processing differences across context conditions, analyses were focused on the N400 Event-Related Potential (ERP) component. Topographic analyses revealed the presence of an N400-like component in central to parietal sides. Control contexts were significantly more negative than both congruent and incongruent ones, while these two conditions did not differ. Fine-grained acoustic information does not seem to modulate semantic integration when exposed to informative-enough contexts. In a second experiment, we assessed the role of attention to acoustic properties during speech processing. Responses and reaction times were collected during a recognition task, in addition to ERPs. Each auditory sentence was matched with three target conditions: i) a word that appeared in the sentence, ii) a word that was absent, iii) for sentences with a homophonic sequence, the word corresponding to the sequence presented. Behavioral analyses using linear and generalized mixed-effects models revealed significant differences between context and target conditions. Participants responded significantly faster to homophonic targets after congruent contexts than after incongruent contexts and more errors were made to homophone targets compared to other target conditions. Preliminary ERP and permutation analyses showed a modulation of the N400 amplitude by context condition, supporting behavioral results. Top-down processes seem to influence speech segmentation when no attention is paid to linguistic properties, whereas bottom-up processes might influence perception when attention is directed to words.