fMRI evidence for shared lemma representations in speech production and comprehension


On a prominent account, lemma representations map the meaning of words to their sounds and vice-versa, and are shared across production and comprehension (Levelt et al., 1999). A meta-analysis of word production studies suggests that lemmas are localised in the middle of the left middle temporal gyrus (left mMTG) (Indefrey, 2011; Indefrey & Levelt, 2000, 2004). This account is supported by other evidence from production and comprehension studies in healthy subjects and patients (Dronkers et al., 2004; Piai et al., 2014; Roelofs, 2014; Schwartz et al., 2009). An alternative account, from a series of reviews based on comprehension data, suggests that sound to meaning mapping during word comprehension is carried out bilaterally in the posterior middle temporal and posterior inferior temporal cortex (bilateral pITG & pMTG)(Hickok & Poeppel, 2000, 2004, 2007). These claims have been challenging to test empirically due to the abstract nature of lemma representations. The sole purpose of a lemma is to link together various types of representations associated with a word (conceptual, syntactic, and phonological representations). Thus, any single task on word comprehension or production would always involve another type of representation making it hard to attribute observed effects specifically to lemmas. To circumvent this problem, we conducted an fMRI study, using four tasks that targeted different modalities (production and comprehension) and different levels of representation (semantic and syntactic) to which the lemma connects. We analysed activities common to all four tasks to determine if the left mMTG or bilateral pITG and pMTG are activated across all the four tasks regardless of modality or level of representation. We used behavioural and fMRI data from 32 healthy native speakers of Dutch. Each participant performed all the four tasks. For production, participants named pictures without determiner (bare picture naming) and with determiner (gender-marked picture naming). For comprehension, participants judged whether the spoken words denoted something man-made or natural (semantic classification -conceptual level) and whether the words were associated with the Dutch determiners “de” or “het” (gender classification – syntactic level). Participants were scanned with a Siemens 3T scanner with a multi-band multi-echo EPI sequence (TR-1.5 seconds). Contrast images of each task against the implicit baseline for each participant were compared across participants. The activation maps of all the four tasks were subjected to a conjunction analysis to identify brain regions that were activated reliably across all the four tasks. We specifically looked within two regions of interests (ROIs), namely the left mMTG and bilateral pITG and pMTG combined to see whether one or the other model was more likely. In our analysis, significant clusters were found in the left mMTG and in the pITG and pMTG, albeit only in the left hemisphere, suggesting a role of both these ROIs in lemma access and, thus, that the two models are not mutually exclusive. Our results support the claim that lemma representations underlie both production and comprehension and reveals the likelihood of a wider, more distributed neural basis of the lemma than has been suggested so far.

Poster presentation at SNL 2020
Arushi Garg
Arushi Garg
PhD candidate