Semantic versus perceptual interactions in neural processing of speech-in-noise

Abstract

Native listeners make use of higher-level, context-driven semantic and linguistic information during the perception of speech-in-noise. In a recent behavioral study, using a new paradigm that isolated the semantic level of speech by using words, we showed that this native-language benefit is at least partly driven by semantic context (Golestani et al., 2009). Here, we used the same paradigm in a functional magnetic resonance imaging (fMRI) experiment to study the neural bases of speech intelligibility, as well as to study the neural bases of this semantic context effect in the native language. A forced-choice recognition task on the first of two auditorily presented semantically related or unrelated words was employed, where the first, 'target' word was embedded in different noise levels. Results showed that activation in components of the brain language network, including Broca's area and the left posterior superior temporal sulcus, as well as brain regions known to be functionally related to attention and task difficulty, was modulated by stimulus intelligibility. In line with several previous studies examining the role of linguistic context in the intelligibility of degraded speech at the sentence level, we found that activation in the angular gyrus of the left inferior parietal cortex was modulated by the presence of semantic context, and further, that this modulation depended on the intelligibility of the speech stimuli. Our findings help to further elucidate neural mechanisms underlying the interaction of context-driven and signal-driven factors during the perception of degraded speech, and this specifically at the semantic level.© 2013 Elsevier Inc. All rights reserved. IntroductionIn studying the neural implementation of spoken language processing, it is important to consider the complexity of linguistic processes. For example, one can ask how higher-order, semantic versus lower-order, perceptual processes interact during the processing of noisy speech -a phenomenon that is ubiquitous in our daily lives, and how the brain supports the interaction of these complementary cognitive and perceptual dimensions (Mattys et al., 2009). It is known that in one's native language, speech comprehension is often successful even when hearing noisy, or degraded speech (Nabelek and Donahue, 1984;Takata and Nabelek, 1990;van Wijngaarden et al., 2002). Further, using the Speech Perception in Noise (SPIN) paradigm (Bilger et al., 1984;Kalikow et al., 1977), in which the predictability of the final word in sentences is manipulated, it has been shown that this native language advantage can arise from the use of higher-level linguistic, contextual information (Florentine, 1985a;Mayo et al., 1997). These original studies further showed that lower SNRs (or higher noise levels) are associated with a greater context benefit (Mayo et al., 1997). We are thus capable of making use of linguistic context to compensate for poor signal quality. Linguistic context includes semantic and syntactic information, as well as pragmatic and proso...

Bilingual speech-in-noise: Neural bases of semantic context use in the native language

“…This native-language benefit is thought to arise from greater use of top-down linguistic information to assist degraded speech comprehension. Using functional magnetic resonance imaging, we recently showed that left angular gyrus activation is modulated when semantic context is used to assist native language speech-in-noise comprehension (Golestani, Hervais-Adelman, Obleser, & Scott, 2013). Here, we extend the previous work, by reanalyzing the previous data alongside the results obtained in the non-native language of the same late bilingual participants.…”

“…during semantically unrelated trials, in which no supporting semantic information is available, c.f. Golestani et al, 2013) in both languages of participants.…”

“…In the present experiment, we extend the fMRI study described above (Golestani et al, 2013) to the non-native language of late bilinguals, with the goals of replicating the behavioral native-language context benefit (Golestani et al, 2009), and of establishing, for the first time, neural evidence for the native-language specificity of this context effect. The neural basis of speech-in-noise processing has so far not been examined in both languages of bilinguals.…”

“…A third fMRI study addressed the question of prior information on the perception of degraded sentences, and found greater activation of the left angular and middle temporal gyri when degraded sentences were preceded by their nondegraded versions, thereby enabling extraction of semantic information (Clos et al, 2012). Fourth, in a recent fMRI study by our group, semantically related or unrelated word pairs were presented, in which the first word was embedded in noise (Golestani, Hervais-Adelman, Obleser, & Scott, 2013). We found relatively greater left angular gyrus activation in the presence of semantic contextual information, and this context effect was greater at high than low SNRs, consistent with the findings of Obleser and Kotz (2010).…”

Predictive processing increases intelligibility of acoustically distorted speech: Behavioral and neural correlates

“…Applying this framework to our experimental paradigm, one would expect that the responses in the auditory (i.e., lower‐level) areas decrease at the second compared to the first presentation of the distorted sentences since the prediction error (i.e., the mismatch between the internal model and the auditory input) diminishes compared to the prediction error at the initial exposure of the distorted sentence when no predictive information is available. However, most of the previous fMRI studies that have used related experimental paradigms have not found the activity within the primary auditory cortex and surrounding areas to be modulated (Golestani, Hervais‐adelman, Obleser, & Scott, 2013; Hervais‐Adelman et al., 2012; Tuennerhoff & Noppeney, 2016; Zekveld et al., 2012; see also Wild et al., 2012a) when distorted speech of low initial intelligibility is rendered more intelligible by a prior or coincident presentation of a disambiguating stimulus. To explain these unexpected results, Wild et al.…”