Learning new sounds of speech: reallocation of neural substrates

Abstract

Functional magnetic resonance imaging (fMRI) was used to investigate changes in brain activity related to phonetic learning. Ten monolingual English-speaking subjects were scanned while performing an identification task both before and after five sessions of training with a Hindi dental–retroflex nonnative contrast. Behaviorally, training resulted in an improvement in the ability to identify the nonnative contrast. Imaging results suggest that the successful learning of a nonnative phonetic contrast results in the recruitment of the same areas that are involved during the processing of native contrasts, including the left superior temporal gyrus, insula–frontal operculum, and inferior frontal gyrus. Additionally, results of correlational analyses between behavioral improvement and the blood-oxygenation-level-dependent (BOLD) signal obtained during the posttraining Hindi task suggest that the degree of success in learning is accompanied by more efficient neural processing in classical frontal speech regions, and by a reduction of deactivation relative to a noise baseline condition in left parietotemporal speech regions.

Introduction

Infants aged 6 months or younger are able to discriminate speech sounds, including many that are not used to distinguish words in their native language. However, during development and starting as early as at 6 months of age, lack of experience with certain nonnative speech sounds results in a developmental shift from a language-general to a language-specific pattern of phonetic perception Best et al., 1988, Jusczyk, 1995, Kuhl, 2000, Kuhl et al., 1992, Polka and Werker, 1994, Werker and Lalonde, 1988, Werker and Tees, 1984a. Most adults can better distinguish two speech sounds belonging to different phonetic categories than ones belonging to the same category, even when the physical differences separating the stimuli have been equated Flege, 1984, Liberman, 1957, Liberman et al., 1957, Liberman et al., 1967, Pisoni et al., 1982, Werker and Tees, 1984b. Despite native-language phonetic perception, adults are capable of learning new languages, and thereby of learning to distinguish nonnative phonetic contrasts. Interestingly, even amongst adults with very similar language backgrounds, considerable individual differences exist in their ability to improve following phonetic training Polka, 1991, Priutt et al., 1990, Strange and Dittman, 1984, Strange et al., 1989, Werker et al., 1981. This finding leads to important questions regarding the functional neural substrates underlying the perception of native versus newly learned, nonnative speech sounds, and more specifically, regarding possible differences in functional anatomy between individuals who successfully learn new speech sounds and those who do not benefit from training.

The neural correlates of phonetic perception have been studied using functional brain imaging techniques such as PET and fMRI. These experiments have involved auditory presentation of stimuli including words, speech syllables, and meaningless speech sounds, and tasks used have included passive listening, phoneme monitoring, discrimination, or identification, and rhyming judgments. Generally, the results have shown the involvement of regions in and around what is classically known as “Wernicke's area”, including left-sided activations in perisylvian temporoparietal areas including the supramarginal and angular gyri Binder et al., 1996, Binder et al., 1997, Démonet et al., 1994a, Paulesu et al., 1993, Petersen et al., 1988, Zatorre et al., 1992, Zatorre et al., 1996. Consistent with functional imaging work, there is also evidence from lesion studies that deficits in phonological processing may arise from damage to perisylvian regions in and around Wernicke's area, including the left superior temporal gyrus and the supramarginal gyrus Benson, 1967, Benson et al., 1973, Geschwind, 1970, Geschwind, 1971. Results of functional imaging work specifically examining phonetic perception have also typically shown activity in the superior temporal gyrus (STG) bilaterally Binder et al., 1994, Hickok and Poeppel, 2000, Jäncke et al., 1998, Mazoyer et al., 1993, Mummery et al., 1999.

The involvement of regions in and around the frontal speech area classically known as Broca's area in phonological processing has been the subject of controversy. Results of some studies involving receptive speech-related tasks have not shown activation in this region Petersen et al., 1989, Rumsey et al., 1992. In contrast, a larger number of studies have shown its involvement in purely receptive language tasks that make certain specific demands Burton et al., 2000, Démonet et al., 1992, Démonet et al., 1994b, Fiez et al., 1995, Zatorre et al., 1992, Zatorre et al., 1996. The frontal regions identified differ across studies, making the interpretation of the roles of such regions more difficult. Although speech perception has not been investigated extensively in aphasic patients with lesions in and around Broca's area, existing studies have shown deficits in phonetic discrimination Blumstein et al., 1977, Tallal and Newcombe, 1978, and in temporal perception (Tallal and Newcombe, 1978) in such patients.

Plasticity of auditory function resulting from training and experience has been shown using techniques such as single cell recordings in animals Kraus and Disterhoft, 1982, Recanzone et al., 1993, magnetoencephalography (Pantev et al., 1999), and event-related potentials (ERP) Kraus et al., 1995, Tremblay et al., 1998 in humans. For example, behavioral training of two slightly different native speech stimuli in adults results in a significant change in the duration and magnitude of the mismatch negativity (MMN) (Kraus et al., 1995), an auditory cortical response to acoustic change that is introduced in a repetitive stimulus sequence Näätänen et al., 1978, Näätänen et al., 1993. This physiological change precedes behavioral discrimination improvements (Tremblay et al., 1998), suggesting that the MMN is a measure of preattentive learning (see Kraus and Cheour, 2000). A number of studies show hemispheric asymmetries in the MMN Alho et al., 1998, Csépe, 1995, Tervaniemi et al., 2000. Tremblay et al. (1997) showed that MMNs elicited by nonnative speech syllables were initially symmetrical, but that they became enhanced over the left hemisphere following training.

The aim of the present study was to determine how the pattern of brain activity may change as a result of training with speech sounds from a nonnative language. Subjects were scanned using fMRI before and after a 2-week period of phonetic training with a Hindi dental–retroflex contrast. During scanning, a native phonetic contrast was used as a control. A noise control condition was also used to subtract out lower level acoustic processing, and to make the results more comparable to those of previous studies on phonetic processing Binder et al., 2000, Zatorre et al., 1992. We wanted to address the following questions. First, does the identification of newly learned speech sounds recruit the same neural substrates as does the identification of a known, native phonetic contrast, or are new areas recruited? The second question relates to whether we can differentiate “learners” from “nonlearners” on the basis of their pattern of activation while they classify the new speech sounds. We predicted firstly that the native identification task would reveal the bilateral involvement of superior temporal regions, stronger in the left than in the right hemisphere, of the left temporoparietal region, and of the left inferior frontal gyrus (IFG) in and adjacent to Broca's area. Second, based on the above reported lateralization of the MMN response to nonnative speech sounds following training, we predicted that before training, the neural response to nonnative speech sounds would be bilateral, but that it would be more left lateralized after training. We also predicted that after training, the pattern of activation outside of the auditory regions (i.e., in the left temporoparietal and inferior frontal regions) would be similar to that found in the native condition. This prediction is also based on results of neuroimaging studies of language function in healthy bilinguals, some of which show that at the single word level, brain regions subserving the native language (L1) and the second language (L2) in fluent bilinguals appear to overlap Chee et al., 1999, Illes et al., 1999, Klein et al., 1994, Klein et al., 1995. Last, based on the assumption that more successful task performance recruits underlying neural substrates more actively, we predict that correlations between a behavioral learning measure and the blood-oxygenation-level-dependent (BOLD) signal during the posttraining nonnative task would reveal a positive relationship between learning and signal in left prefrontal and left temporoparietal speech areas.