From blueprints to brain maps: the status of the Lemma Model in cognitive neuroscience

ABSTRACT

When Levelt’s pioneering book about the Lemma Model of speech production appeared in 1989, cognitive neuroscience was just starting to take off. During the 30 years since then, this influential framework has undergone many refinements, and extensive efforts have been made to relate it to the brain. This paper provides a broad overview of these developments. The first section briefly describes the main claims of the theory. Then the next section presents multiple forms of evidence that, in keeping with these claims, spoken word generation depends on a predominantly left-hemisphere circuit in which the different levels of representation and computation postulated by the theory are subserved by mostly non-overlapping cortical regions, and the flow of information between them is largely sequential. Finally, the last section focuses on a number of other findings from cognitive neuroscience that appear to challenge the theory and hence must be addressed in future research.

KEYWORDS:

• Speech production
• lemmas
• phonology
• phonetics
• concepts

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 The morphemic phonological code of the English plural suffix is actually even more abstract than this, but the technicalities are not pertinent here. For a clear discussion, see Pinker (1999).

2 As shown in Figure 1a, syllabification takes place together with prosodification, which involves determining the stress pattern of the target word (if it is multisyllabic) and computing an intonation contour that reflects both the conventions of the given language and the emotional state of the speaker. However, prosodification and its neural substrates are not discussed in this paper (for a review see Chapter 7 of Kemmerer, 2015).

3 For some recent neuroscientific insights, though, see Bouchard, Mesgarani, Johnson, and Chang (2013), Simonyan (2014), Simonyan, Ackermann, Chang, and Greenlee (2016), Tremblay, Deschamps, and Gracco (2016), Chartier, Anumanchipalli, Johnson, and Chang (2018), and Conant, Bouchard, Leonard, and Chang (2018).

4 In this subsection and throughout the rest of the paper, the terms semantic and conceptual are used interchangeably, consistent with current practice in cognitive neuroscience.

5 A number of other studies have investigated the time-course of ATL engagement during tasks that did not involve picture naming. These studies used electrophysiology (Chan et al., 2011; Naci, Taylor, Cusack, & Tyler, 2012; see also Quiroga, 2012), MEG (Marinkovic et al., 2003; van Ackeren, Schneider, Müsch, & Rueschemeyer, 2014; Clarke, Devereux, Randall, & Tyler, 2015), and TMS (Jackson et al., 2015).

6 Besides the left IFG, the selection process also recruits the left dorsal anterior cingulate cortex, which has been implicated in more general regulatory operations that are not unique to word production (Roelofs, 2008; Price, 2012; Piai, Roelofs, Acheson, & Takashima, 2013). In addition, according to a recent intracranial recording study that used the blocked cyclic object naming paradigm (Riès et al., 2017), the selection process engages not only the left IFG and left anterior cingulate, but also some of the same cortical regions that underlie the activation of word meanings—a finding that, as the authors point out, reconciles computational models which assume that the selection process occurs outside the semantic system (e.g., Oppenheim, Dell, & Schwartz, 2010) with models which assume that it occurs inside that system (Howard, Nickels, Coltheart, & Cole-Virtue, 2006).

7 Patients with Wernicke’s aphasia likewise produce abundant phonemic paraphasias, but their speech comprehension is typically if not always impaired too. And while these patients’ lesions often include the left posterior STG, they are usually centred in the left posterior MTG, a region that has been implicated in the understanding of single-word and multi-word utterances (Dronkers, Wilkins, Van Valin, Redfern, & Jaeger, 2004; Dronkers & Baldo, 2009; Pillay, Binder, Humphries, Gross, & Book, 2017; see also Binder, 2017).

8 Likewise, Basilakos, Smith, Fillmore, Fridriksson, and Federenko (2018) found some degree of selectivity for speech production in subject-specific anatomical parcels of the left opercular IFG.

9 GODIVA stands for “gradient order DIVA,” and DIVA stands for “directions into velocities of articulators.”

10 A caveat: The 355–455 ms time window for left posterior IFG engagement derives from studies that focused exclusively on a few Indo-European languages. A recent study that focused instead on Mandarin Chinese found that the same region was engaged much earlier, by around 225 ms (Zhang, Yu, Zhang, Jin, & Li, 2018).