From phonemes to articulatory codes: an fMRI study of the role of Broca's area in speech production

Abstract

We used event-related functional magnetic resonance imaging to investigate the neuroanatomical substrates of phonetic encoding and the generation of articulatory codes from phonological representations. Our focus was on the role of the left inferior frontal gyrus (LIFG) and in particular whether the LIFG plays a role in sublexical phonological processing such as syllabification or whether it is directly involved in phonetic encoding and the generation of articulatory codes. To answer this question, we contrasted the brain activation patterns elicited by pseudowords with high- or low-sublexical frequency components, which we expected would reveal areas related to the generation of articulatory codes but not areas related to phonological encoding. We found significant activation of a premotor network consisting of the dorsal precentral gyrus, the inferior frontal gyrus bilaterally, and the supplementary motor area for low- versus high-sublexical frequency pseudowords. Based on our hypothesis, we concluded that these areas and in particular the LIFG are involved in phonetic and not phonological encoding. We further discuss our findings with respect to the mechanisms of phonetic encoding and provide evidence in support of a functional segregation of the posterior part of Broca's area, the pars opercularis.

Figure 1.

During the experiment, subjects were asked to listen to pseudowords and to repeat them either overtly or covertly after a 6-s delay. The structure of each trial is shown in (A). The stimulus is presented auditorily at 0 s and then subjects wait for the response probe. During the delay period, they are instructed to covertly rehearse the stimulus and are not aware of the type of response (overt or covert) before they hear the probe. The type of stimulus that will be presented in each trial is determined pseudorandomly by a combination of 3 m-sequences. In (B), we present an example of 3 binary sequences that resemble those used in the experiment. Each sequence is associated with an experimental factor. In the example provided, the top sequence controls the length of the stimulus (1 for 4 syllables and 0 for 2 syllables), the middle sequence controls sublexical frequency (1 for high and 0 for low), and the bottom sequence controls response type (1 for overt and 0 for covert). For example, the combination 0 1 0 would retrieve a 2-syllable, high-frequency pseudoword and the covert response probe.

Figure 2.

Surface renderings of significant activations in the whole-brain group analysis for length (A) and sublexical frequency (B). In (A), an extended perisylvian and premotor activation including the LIFG showed significantly higher activation for 4 versus 2 syllables. In (B), premotor areas including the dorsal PrCG and the IFG bilaterally showed significantly higher activation for low- versus high-frequency pseudowords. In (C), we show the main effect of length within left BA44 (significantly activated voxels appear in magenta) using a small volume correction approach (SVC). BA44 (shaded area) was defined using a cytoarchitectonic probability map of the area (Eickhoff et al. 2005). Maps are thresholded voxelwise at P <0.001 uncorrected and clusterwise at P <0.05 FWE corrected. Color grading in (A) and (B) reflects depth, with brighter voxels on the surface. The maximum depth of the projected voxels is 20 mm. L, sagittal view of the left hemisphere.

Figure 3.

Significant activations within left hemisphere BA44 as defined by a cytoarchitectonic probability map of the area (Eickhoff et al. 2005). Shown in red are voxels significantly more activated for 4 versus 2 syllables. This cluster extends from z = −2 (slice not shown) to z = 28. The largest effect for length is located dorsally, at [−60 4 20]. Shown in blue are voxels significantly more activated for low versus high sublexical frequency. The largest effect for frequency is located at [−54 12 12]. Finally, shown in green are voxels that are overlapping for both conditions (size of overlap = 9 voxels). Activations are thresholded at P <0.001 uncorrected voxelwise and P <0.05 FWE corrected clusterwise. Z coordinates are in MNI space.