Memory and Working-with-Memory: A Component Process Model Based on Modules and Central Systems Morris Moscovitch Department of Psychology Erindale College, University of Toronto and The Rotman Research Institute Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/4/3/257/1755001/jocn.1992.4.3.257.pdf by guest on 18 May 2021 North York, Ontario Abstract A neuropsychological model of memory is proposed that ory that are associativekue dependent, (3) a central system, incorporates Fodor’s (1983) idea of modules and central sys- frontal-lobe component that mediates performance on explicit tems. The model has four essential components: (1) a non- tests that are strategic and on procedural tests that are rule- frontal neocortical component that consists of perceptual (and bound, and (4) a basal ganglia component that mediates per- perhaps interpretative semantic) modules that mediate perfor- formance on sensorimotor, procedural tests of memory. The mance on item-specific, implicit tests of memory, (2)a modular usefulness of the modularkentral system construct is explored medial temporallhippocampal component that mediates en- and evidence from studies of normal, amnesic, agnosic, and coding, storage, and retrieval on explicit, episodic tests of mem- demented people is provided to support the model. LNTRODUCTION “works with memory” and mediates performance on ex- plicit tests that are strategic. Memory is not unitary but depends on the operation of MODULES AND CENTRAL SYSTEMS potentially independent, but typically interactive, com- ponents. One of the jobs of a cognitive neuropsychologist A distinguishing feature of this memory model is that its is to identify these components and indicate how they basic principles are derived from a consideration of Fo- interact with each other. Although we are far from being dor’s (1983) ideas that modules and central systems are able to specify these components with the precision we the constituents of mind (and brain). Carlo Umilta and would like, either at a functional or structural level, there I, however, proposed a modified version of Fodor’s ideas is sufficient information to tempt many of us to sketch that retained its core assumptions and suggested how the outlines of what we think a complete model might Fodor’s criteria of modularity can be translated to the be like. I offer my version of such a model. neuropsychological level (Moscovitch & Umilta, 1990, The model I sketch has four essential components, 1991). each of which mediates processes that dominate perfor- Modules are computational devices that have propo- mance on four different types of memory tests: (1) a sitional content and that satisfy all of the following three nonfrontal neocortical component that consists of var- criteria: domain specijicity, informational encapsulation ious perceptual and “semantic” modules that mediate or cognitive impenetrability, and shallow output. Do- performance on item-specific, implicit (indirect) tests of main specijicity entails that the type of information mod- memory, (2) a basal ganglia component that mediates ules accept for processing is restricted or circumscribed. performance on sensorimotor procedural tests of mem- At the neuropsychological level, it must be shown that ory, (3) a medial temporaVhippocampa1 component, damage to a particular region or system, the structural which also is modular, and that mediates encoding, stor- embodiment of the module, leads to deficient processing age, and retrieval on explicit (direct) episodic memory in the purported domain with relative sparing of function tests that are associativekue dependent, and (4)a frontal- in other domains. This condition is not sufficient by itself lobe component, which is a central system structure, that since central system structures can also be localized to 0 1992 Mmacbusetts Institute of Technology Journal of Cognitive Neuroscience Volume 4, Number 3 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1992.4.3.257 by guest on 28 September 2021 circumscribed regions. Informational encapsulation im- classes of memory tests, explicit and implicit (Graf & plies that modules are resistant to the effects of higher Schacter, 1985). Explicit tests require conscious recollec- order knowledge on processing and are cognitively im- tin of past events whereas on implicit tests, memory for penetrable to probes of their content or operation. Only the past is inferred from changes in performance with the module’s shallow output is available for conscious experience or practice. To be implicit, a test must also inspection. Neuropsychologically, this criterion is satis- be highly structured so that the goal of the task and the fied if the processes mediated by a module are unaffected means to achieve it should be apparent and available to by gross intellectual decline that is caused by degener- the subject (Moscovitch, 1984). When an implicit test ation or focal damage to structures other than the mod- does not meet the latter criteria, the amnesic patient ule itself. For example, patients with generalized shows no evidence of savings even though conscious dementia caused by Alzheimer’s disease fail to under- recollection may not be involved (Nissen, Willingham, & stand even simple words or appreciate the function of Hartman, 1989). objects but they can still read relatively well (Schwartz, Implicit and explicit tests can each be subdivided fur- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/4/3/257/1755001/jocn.1992.4.3.257.pdf by guest on 18 May 2021 Saffran, & Marin, 1980) and have a good three-dimen- ther into at least two subtypes. For implicit tests, the two sional (3D) representation of objects (Chertkow & Bub, are procedural and item speczfic. Procedural tests are 1990; Moscovitch & Umilta, 1990; Warrington & Taylor, those that assess learning and retention of general sen- 1978). The converse also implies informational encap- sorimotor skills, procedures, or rules. Item-specific tests, sulation, namely, a domain-specific deficit with preserved on the other hand, assess memory for a particular item, intellectual functions and semantic knowledge about ma- such as a certain word, face, or object by seeing the effect terial in the affected domain. Patients with associative that initial presentation of the item has on accuracy or agnosia may not recognize an object visually but can speed of identification of the item when it is repeated. provide detailed semantic information about the object The benefit gained is known as the repetition priming when given its name. Yet the patient may not be able to effect. use this knowledge to identify the object visually (Mos- The two subtypes of explicit tests are mociativelcue- covitch & Umilta, 1990; Riddoch & Humphreys, 1987). dependent and strategic. Associative episodic memory Shallow ouput is output that has no meaning beyond tests are those in which the cue is sufficient for retrieval. the value assigned to it by the module; interlevel rep- When given the cue “Have you read War and Peace?” or resentations that led to the shallow output are not avail- “Have you seen Gone with the Wind?”the answer auto- able for conscious inspection. The neuropsychological matically pops into mind as surely as the word “night” correlate of this criterion is evidence of normal, domain- pops into mind to the semantic cue “day.” For strategic specific performance without any ability to interpret se- tests, the cue does not elicit the target memory auto- mantically the information pertaining to that domain. matically but provides only the starting point of a mem- Here, too, patients with associative agnosia or dementia ory search that has elements in common with problem- are the paradigmatic cases. Though such patients retain solving. Such strategic processes can be initiated by ques- the ability to process objects, faces, and words at a struc- tions that require the reinstatement of a particular spatial tural, presemantic level, they cannot assign any meaning and temporal context such as “What did you do two to the structural information they have computed (Bauer, weekends ago?” 1984; Chertkow & Bub, 1990; Warrington & Taylor, 1978; There are few, if any tests, that are comprised of only Moscovitch & Umilta, 1990, 1991). a single component. The classificatory scheme suggests Thus, a module, no matter how complex its inner ideal prototypes against which impure tests can be com- workings, is essentially a “stupid,” closed computational pared and thus provides a crude framework for fraction- device that delivers its shallow output to interpretative ating a test into its component parts. central systems where meaning and relevance are as- signed. None of the criteria of modularity applies to A NEUROPSYCHOLOGICAL MODEL central systems (but see Moscovitch & Umilta, 1990, for OF MEMORY some provisos). Unlike modules, central systems inte- Input Modules and Item-Specific Implicit grate information from superficially dissimilar domains Tests: Reactivation of Perceptual and and are open to top-down influences. The output of Semantic Records central systems is deep or meaningful and the interlevel representations that give rise to the final output may be Memory begins with the registration of information in available to consciousness. cortical modules that pick up and transform stimulus events into structural, presemantic representations (Fig. 1). The output of these modules is delivered to central CLASSIFICATION OF MEMORY TESTS system structures for early, semantic interpretation.