The effect of enhanced working memory on language

Abstract

AlthoughweareinalmostcompleteagreementwithMarti´n-Loeches (2006), his critique of our argument presents us withan opportunity for clarification. We commend his descriptionof the possible neural basis for enhanced working memory(EWM), especially the possible role of parallel activation. Ashe makes clear (clearer than we did), EWM need not havebeen based in a massive increase in the number of neurons.We differ from Marti´n-Loeches primarily in our treatmentof language. One of our motivations for advancing theEWM hypothesis was a desire to challenge the heavy (andoften uncritical) reliance that many anthropologists place onlanguage as the key to cognitive modernity. Language issuch a powerful ability, and our understanding of it is somuch more comprehensive than that of many other cognitiveabilities, that it tends to eclipse our understanding of all otherdevelopments. Marti´n-Loeches’reworking of our argument forthe phonological loop reflects, we believe, this ‘‘bullying’’ ef-fect of language, and syntax in particular. We suggest thatmuch of what Marti´n-Loeches attributes to a language proces-sor might be more parsimoniously explained by elements ofBaddeley’s original model of working memory. One compo-nent of WM mentioned, but not emphasized, by Marti´n-Loeches is its access to long term memory (LTM). Ericsson(Ericsson and Delaney, 1999), for example, has built a power-ful explanation of expert performance based on facilitatedaccess to LTM via WM. In essence, cues held in and manipu-lated by WM provide quick access to much longer and morecomplex encodings held in LTM. This strikes us as being pre-cisely what happens in Waters and Caplan’s (1996) ‘‘psycho-linguistic resource pool,’’ when the ‘‘more automatic’’ accessto complex syntax occurs. In other words, this cognitive abilitymay be a general feature of WM tapping into a long term storeof syntactical models, rather than anything specific to lan-guage. And the ‘‘post-interpretive processing’’ used in verbalreasoning strikes us as a function of the episodic buffer ofthe central executive, rather than a separate module.It is interesting that our differing accounts of the relation-ship between language and WM ultimately take us to thesame place, for we too would argue for Marti´n-Loches’secondevolutionary scenario e that the developments in WM thatproduced the modern mind enhanced the capabilities of analready-existing language (syntactical) processor. Recursionprovides a telling example of how this would have worked(and especially pertinent as some linguists now maintain thatrecursion is the crucial piece to syntax [Hauser et al.,2002]). Recursion is the mechanism in grammar that enablesa speaker to use an entire phrase as an object of a higher levelphrase (e.g., ‘‘He said that she said.’’). It is this feature thatsupplies native speakers of a language with the ability to pro-duce, in principle, an infinite number of meaningful sentences.In practice, the size of this ‘‘infinity’’ is constrained by severalpractical limitations, one of which is WM. The number of re-cursions must be held and processed in attention if they are tobe understood. ‘‘He said that she said that they said that wesaidthat IsaidthatGeorge W.BushisatrueTexan,’’isagram-matically correct sentence, but one that just about exhausts thecapacity of WM to analyze. Add two more levels of recursionand few native speakers could keep track. The recursive rule,held presumably in a syntactical processor, has not changed,but the sheer size of the task has. Perhaps the simplest inter-pretation of the effect EWM had on linguistic communicationis to conclude that it enlarged the recursive capacity of lan-guage. An enhancement of WM would yield immediate resultsin the length and complexity of sentences. Marti´n-Loechesmay well be correct in arguing that the phonological loop,sensu strictu, is too specialized (or too encapsulated) to ade-quately encompass this increased capacity (TW and FC them-selves disagree amicably on this point). However, we are alsoreluctant to subdivide WM into smaller and smaller subsys-tems if features of the general model can account for theexperimental results.