Psychonomic Bulletin & Review 2001, 8 (1), 44-57 The case for sensorimotor coding in working memory MARGARET WILSON North Dakota State University, Fargo, North Dakota The highly influential Baddeley and Hitch model of working memory (Baddeley & Hitch, 1974; see also Baddeley, 1986) posited analogical forms of representation that can be broadly characterized as sensorimotor, both for verbal and for visuospatialmaterial.However, difficultieswith the model of ver- bal working memory in particular have led investigators to develop alternative models that avoid ap- pealing either to sensory coding or to motoric coding, or to both. This paper examines the evidence for sensorimotor coding in working memory, including evidence from neuropsychology and from sign lan- guage research, as well as from standard working memory paradigms, and concludes that only a sen- sorimotor model can accommodate the broad range of effects that characterize verbal working mem- ory. In addition, severalfindings that have been considered to speak against sensorimotor involvement are reexamined and are argued to be in fact compatible with sensorimotor coding. These conclusions have broad implications, in that they support the emerging theoreticalviewpoint of embodied cognition. The past two decades have seen an explosion of inter- However, in addition to the considerableempirical and est in the topic of working memory. In the 1970s and theoretical successes created by the Baddeley model, a 1980s, Baddeley and Hitch (1974; Baddeley, 1986) pro- number of difficulties have emerged: In particular, several posed that previous conceptions of short-term memory of the signature effects that underlie the original model should be replaced by a multicomponent model of tem- have been found to behave in ways not originallypredicted. porary information storage and manipulation, including Some researchers have sought to adapt the model to ac- a verbal component and a visuospatial component.Since commodate the new data, whereas others have pursued then, the model—and in particular the “phonological alternative approaches, and the past 5 years have seen a loop” account of verbal rehearsal—has created a fertile proliferation of competing models of working memory. domain of research, encompassingnot only empirical con- Although the Baddeley model posits fairly peripheral firmation and expansion of the model (see Baddeley & forms of representation (visuospatial working memory Hitch, 1994, for review) but also applicationsto processes, preserves depictivepropertiesof the stimulus, verbal work- such as language acquisition and reading (e.g., Ellis & ing memory preserves phonological and articulatory Sinclair, 1996; Gathercole & Baddeley, 1993), and com- properties), more recent models have shown a consistent putational modeling of the proposed underlying mecha- trend away from sensorimotor representations. In this arti- nisms (e.g., Burgess & Hitch, 1992, 1999; Henson, 1998; cle, the viability of such a theoreticalmove is considered. Page & Norris, 1998). This questionis of considerabletheoreticalimportance (An additional recent trend has been the expansion of because it bears on the concept of embodied cognition the concept of working memory to include the capacity that is gaining currency within cognitive science (e.g., for on-line processing [Carpenter, Miyake, & Just, 1994] Clark, 1997; Glenberg, 1997; Johnson, 1987; Kutas & and semantic comprehension [Martin & Lesch, 1996] Federmeier, 1998; Lakoff & Johnson, 1999).The core in- and to explain the interactionsbetween long-term knowl- sight of this viewpoint is that many of our “central cog- edge and working memory capacity [Cantor & Engle, nitive” abilities may in fact be parasitic upon perceptual 1993; Gathercole& Martin,1996; Nairne & Kelley,1999; and motoric processes. One line of theorizing,also known Neely & LeCompte,1999; for general review, see Richard- as situatedcognition, emphasizes the on-line use of overt son et al., 1996]. For the purpose of this paper, which sensorimotoractivityto assist with cognitivetasks that in- focuses on the storage and maintenance of information, teract with the world. Of equal interest, though,is off-line the narrower usage of the term working memory will be embodied cognition, in which sensorimotor processes retained.) are run covertly to assist with the representation and ma- nipulation of information, in the temporary absence of task-relevantinput or output.Such an arrangement would This work was partially supported by NSF Grant OSR 9452892.The make sense, given our evolutionary heritage from crea- author thanks Nelson Cowan, Karen Emmorey, Arthur Glenberg, and John Richardson for their valuable comments. Correspondence should tures whose neural resources were devotedlargely to per- be addressed to M. Wilson, Department of Psychology, North Dakota ceptual and motor processes. Indeed, given that we have University, Fargo, ND 58105(e-mail: [email protected]). such resources, it would be odd if we did not exploitthem Copyright 2001 Psychonomic Society, Inc. 44 SENSORIMOTOR CODING IN WORKING MEMORY 45 Table 1 sions; however, translating materials into the appropriate Presence of Effects in Verbal Working Memory phonologicalcode requires some form of mental articula- Under Various Presentation Conditions tion. When this articulation is blocked, the phonological Effect code is never achieved, and so phonological confusions Presentation Phonological Irrelevant Word Articulatory will not take place. In addition, articulatory suppression Conditions Similarity Speech Length Suppression eliminates the word length effect regardless of how mate- No Additional Manipulations rials are presented. This suggests that, unlike the similar- Auditory stimuli + + + + ity effect, the lengtheffect is a direct consequenceof some Visual stimuli + + + + articulatory process: When articulation is blocked, the Articulatory Suppression length effect cannot occur. Auditory stimuli + + – (It is worth notingthat neitherarticulatorysuppression Visual stimuli – – – nor irrelevant speech reduces performance to zero. This Irrelevant Speech requires explanation—particularly since, as described Auditory stimuli + – ? + above, articulatory suppression is capable of completely Visual stimuli –? – ? + eliminating other signature effects. An immediately ob- vious interpretation is that other resources are available for maintaininginformation. A few candidatesare visual whenever possible to assist and enhance off-line cogni- representation of the physical form of the word, visual tive processing. The issue of sensorimotorinvolvementin representationof the meaning of the word [particularly for working memory, then, goes beyond the question of how concrete words], and abstract semantic representation. we perform immediate serial recall, to make contact with None of these appears to be a particularlyrobust strategy basic issues about the nature of our cognitivearchitecture. for maintaining multiple items in order, which accounts for the ubiquitous use of phonological representation in The Classic Data and the Rehearsal Loop Model these tasks.) Although visuospatialworking memory has not yielded Thus, these data have been taken as indicating a two- strong evidence regarding its internal structure, beyond its part system, consistingof a buffer that stores information analogical nature, verbal working memory from the be- in phonologicalform and an articulatoryrehearsal process ginning showed effects and interactions that demanded that can be used to load or refresh the buffer (Baddeley, explanation.The pattern of data that led to the concept of 1986). As Table 1 shows, the effects associated with the a phonologically based rehearsal loop in working mem- buffer—the phonologicalsimilarity and irrelevant speech ory has been frequently reviewed (e.g., Baddeley, 1990; effects—are disruptable with visual presentation but not Gupta & MacWhinney, 1995; Neath, Surprenant, & with auditory presentation,whereas the effects associated LeCompte, 1998; Wilson & Emmorey, 1997a) and is pre- with articulation—the word length and articulatory sup- sented here only in brief. That data pattern involves four pression effects—stand or fall regardless of modality(for effects on immediate serial recall: the phonological sim- further evidenceof two distinctcomponents,see Longoni, ilarity effect (poor recall of similar-sounding items); the Richardson, & Aiello, 1993, and Schweickert, Guentert, word length effect (poor recall of long items); the artic- & Hersberger, 1990).This two-part system has been called ulatory suppression effect (poor recall when there is the articulatoryloop, or, more recently, the phonological competingactivity of the articulators); and the irrelevant loop. speech effect (poor recall when there is competing audi- Although this basic model does not involvestrong com- tory input). The immediate suggestion these data seem to mitments on the exact nature of these two components, it make is that verbal materials are coded in working mem- is worth noting their immediate resemblance to speech ory in something like their “surface form”—either what perception and speech production,respectively.Of inter- they sound like or how we would pronounce them. est here is a general class of explanation that appeals to In addition to these effects that suggest coding in a some form of quasi-sensory coding,on the one hand, and speech-like form, interactions among the effects suggest quasi-motoric coding,on the other hand. Either or both of