In this study, we were able to delineate the cortical areas involved in the phonemic-to-articulatory translation that is necessary for the generation of articulatory codes. By directly contrasting targets with varying length, we manipulated the load on the system of postlexical articulatory-motor production and were able to identify a number of key regions underlying articulation and the overall process of transforming phonological word forms to articulatory codes. In summary, these regions included bilateral (although strongly left lateralized) mid and posterior superior temporal and frontal regions, the premotor cortex, and the SMA. These results are in agreement with current models on word production that describe a left-lateralized, perisylvian network (Indefrey and Levelt 2000
; Hickok and Poeppel 2004
To further identify the roles of the different components of the network and in particular to resolve the conflict on the role of the LIFG, we probed the network by manipulating sublexical frequency. Our hypothesis was that only regions that are directly involved in phonemic-to-articulatory translation would show an effect for frequency manipulation. Targets with components of different sublexical frequency (high vs. low) are processed differently (Guenther et al. 2006
). High-frequency clusters are precompiled and their articulatory codes are retrieved, as suggested by the fact that they are processed faster than the ones with less-frequent components (Vitevitch and Luce 1998
). The latter are thought to be compiled online on a segment-to-segment basis (Guenther et al. 2006
In our experiment, we identified 4 regions that showed an effect related to sublexical frequency (higher activation for low vs. high frequency): the LSMA, the left hemisphere PrCG, and the IFG bilaterally. From previous studies on motor planning and production, it is known that the SMA has a role in motor planning and the preparation of movements. Even though its function is not specifically associated with linguistic processes, it is also part of linguistic motor planning (Riecker et al. 2005
). In a recent fMRI study, the pre-SMA was shown to be sensitive to sequence complexity effects both within and beyond the syllable boundaries (Bohland and Guenther 2006
). The present findings are in agreement with the current theories on the function of the SMA. The observed frequency effect could simply represent the increased load that is associated with producing new and unfamiliar motor plans (low–sublexical frequency pseudowords) compared with familiar, more rehearsed, and precompiled ones (high–sublexical frequency pseudowords).
The significant activation difference for low– versus high–sublexical frequency pseudowords in the left PrCG is also in agreement with current models on word production (Hickok and Poeppel 2004
; Indefrey and Levelt 2004
; Guenther et al. 2006
). It is worth highlighting that only a small area in the dorsal PrCG was significantly active and that this area has been previously involved in studies examining sensory–motor mapping (Hickok and Poeppel 2004
). Hickok and Poeppel propose the existence of a “dorsal stream” in speech processing, which is involved in mapping sound onto articulatory-based representations. The regions that are part of this stream include a posterior inferior frontal area (including Broca's area), a dorsal premotor site, and area SPT (Hickok et al. 2003
). The latter region, which lies within the boundaries of the planum temporale, is traditionally associated with acoustic and phonological processing, as well as speech production as the interface for the sound-to-gesture transformation.
In our study, we found that the STG bilaterally shows a greater effect for target length, though the results are strongly left lateralized, and in the left hemisphere, particularly, the effect extends further in the posterior direction to area SPT (). Bilateral STG activation has been observed during both speech perception and production and reflects the processing of the acoustic and phonological properties of the target stimulus (Hickok and Poeppel 2004
). This is in contrast to area SPT, which is thought to be involved in translating between acoustic and motor representations. However, in the current study, both STG and area SPT show a similar behavior and a significant main effect for length only and not for sublexical frequency. Therefore, these findings raise doubts on the role of SPT as an auditory–motor interface and suggest that its role is not that different from the rest of the STG, that is, it could also be involved in phonological processes, such as syllabification and segmentation. This claim would be in agreement with initial claims made by Indefrey and Levelt (2000)
, whereby a portion of the superior temporal lobe was considered as a possible candidate region for syllabification. Another candidate was the LIFG.
In the current study, we found significant bilateral activation in the IFG. The presence of a sublexical frequency effect in the right IFG was surprising because this region has not been included in any of the neuroanatomical models of speech production previously discussed (Hickok and Poeppel
; Indefrey and Levelt 2000
). Activation in this region has been previously found during pitch processing and specifically for the integration of accent patterns (Geiser et al. 2008
). In the current study, the stress pattern between the 2 categories was controlled, and there were no systematic differences. However, it is possible that the increased processing demands for low–sublexical frequency pseudowords also affected the processing of metrical structure. Further research would be needed to identify the exact nature of the differences.
With respect to the LIFG, the pars opercularis showed consistent effects for both length and sublexical frequency (4 vs. 2 syllables and low vs. high frequency, respectively), as well as evidence of functional segregation. The more dorsal part of the area (dPOp) was modulated by differences in stimulus length, whereas the ventral part (vPOp) was modulated by differences in both length and sublexical frequency. The idea that Broca's area is functionally segregated into its 3 anatomical parts (pars opercularis, triangularis, and orbitalis) is well known and well founded (Bokde et al. 2001
; Chein et al. 2002
; Devlin et al. 2003
; Heim et al. 2007
). Recently, however, there have also been claims concerning a functional segregation within pars opercularis (Molnar-Szakacs et al. 2005
). In a meta-analysis of imaging studies on imitation and action observation, Molnar-Szakacs et al. (2005)
identified 2 distinct foci within the pars opercularis, a dorsal and a ventral one, that serve different functions. DPOp shows mirror neuron properties and is significantly active during both action observation and imitation, whereas vPOp shows only motor properties and is only active during imitation.
In agreement with this segregation, we also identified 2 distinct clusters within the pars opercularis with one extending more dorsally than the other. The more dorsal cluster is located closer to the IFS and the premotor cortex and shows greater activation for length manipulation. The vPOp, on the other hand, shows both a main effect of length and sublexical frequency. In the current study, the dPOp is part of a wider area of activation in the left hemisphere PrCG. Therefore, based on its relation to premotor areas, as well as the fact that it is only active for the length condition, we can conclude that the dPOp is involved in phonological encoding and syllabification as proposed by Indefrey and Levelt (2000
). This role is in agreement with other proposed roles such as sequencing discrete units (Gelfand and Bookheimer 2003
) or sublexical processing requiring explicit segmentation (Zatorre et al. 1996
; Burton et al. 2000
; Chein et al. 2002
The vPOp on the other hand shows a significant effect of both length and frequency, which is in agreement with a role as the cite of the speech sound map or mental syllabary that has been proposed by Guenther et al. (2006)
. These results are also partially in agreement with the claims made by Molnar-Szakacs and colleagues, who propose that it holds a form of representation of the motor plans that is communicated to the posterior part of the STS (Molnar-Szakacs et al. 2005
). In this account, the vPOp is not the location of the speech sound map but only holds a copy of the articulatory codes. The codes themselves are generated elsewhere. The only other possible candidate in our case would be the dorsal premotor cortex, which also showed a significant effect of sublexical frequency. Based on our results, we cannot exclude either possibility.
Research into the functional segregation of the pars opercularis is still in a preliminary phase. The anatomy of the LIFG is highly variable across subjects (Amunts et al. 1999
), which makes it difficult to draw any precise conclusions about the exact anatomical borders of the hypothesized segregation of the pars opercularis based on group-averaged results. For the purposes of this study, we have also described the functional segregation of the region using gross anatomical terms such as ventral and dorsal and only in terms of the group tendency. Future research using higher spatial resolution at the single-subject level will be needed to further verify and specify the exact anatomical features of this functional segregation.
Finally, we also note that we did not find any regions showing significant effects for the inverse contrast high– versus low–sublexical frequency. Based on our hypothesis, we would expect that a significant activation for this contrast would reveal the location of the mental syllabary versus the network underlying articulatory code generation. However, based on the computational model proposed by Guenther et al. (2006)
, the speech sound map (the equivalent of the mental syllabary) does not just contain precompiled frequent syllables but also motor representations for phonemes, common words, phrases, etc. The speech sound map is therefore involved in both processes, though the online compilation of articulatory codes would be computationally more demanding than the retrieval of precompiled gestural scores. This would explain why we do not see increased activity for high- versus low-frequency stimuli because it would be the same network that is underlying both processes.
To conclude, in this fMRI study, we investigated the process of phonological-to-articulatory translation and the role of the LIFG. Based on our findings, we conclude that the LIFG, BA44 in particular, is functionally segregated into 2 subregions following a dorsal–ventral gradient. The dorsal part seems to be involved at the level of phonological encoding as suggested by Indefrey and Levelt (2000
), whereas the ventral part seems to be involved at the level of phonetic encoding and possibly in the translation between phonemic and articulatory representations as proposed by Hickok and Poeppel (
. This finding is in agreement with recent observations on the functional segregation of the pars opercularis and further clarifies the role of the LIFG in language production.