Reach Exceeds Grasp: Comments on Frawley's “Control and Cross‐Domain Mental Computation: Evidence from Language Breakdown”
暂无分享,去创建一个
A particular version of the modular theory of cognition goes like this. Cognition consists of modules, each with its own domain of expertise. In this picture, a module does not interact directly with another module. A module reports the results of its activity to a central working memory (WM), with its own more or less complex structure. Tasks often require involvement of more than one module, thus setting up the problem of how the different modules work together. This problem is solved by having the modules not only report their results to the WM, but also by having the modules that need information from another module simply read off the WM the results reported by the latter module. This way, the task gets accomplished without the need for the modules to directly communicate with other modules. The WM plays the role of a central blackboard in this view. Software systems in general, even those that are modular, as almost all large systems have to be, are not restricted to the architecture outlined above. Especially in parallel systems, there is in principle no reason for a single centralized WM, or a WM at all. In the modular view sketched above, the individual modules might themselves have submodules for all we know, but the activities of these submodules are not coordinated by the central WM. In cognitive science we are so used to metaphors from computer science that we are not always careful to ask where the metaphor becomes a stretch, where some artifact of the computational model inadvertently takes on a reality as part of the cognitive science explanation. For example, cognitive models are often written in Lisp, and unless one is careful, one might slip into thinking of Cons operations and garbage collection as true cognitive phenomena. (They could be, but evidence has to be adduced.) Frawley proposes ‘‘control’’ between domains—or modules—as an explanatory idea in understanding certain issues in language processing. Certainly control—or more generally, coordination between modules—has to be an important source of explanation of cognitive phenomena in the modular theory. The structure of WM, exactly what qualify as modules and how they write to and read from WM are details to be worked out in the theory, but, as I indicated above, the proposal that modules coordinate their activities by means of reading from and writing to the central WM is a promising framework for explanation. I think the author’s attempt to explain some language disorders in the framework of module coordination is valuable. Let me restate what I see as valuable using my own language, in particular completely avoiding the logic versus control language that he employs. Performance in any task that involves many modules working together can suffer because of problems in one or more of the following: processing problems within one or more of the modules, systemic damage to WM, problems in a module reporting its results to the WM, or problems in a module reading from WM. Frawley analyzes various languagerelated disorders from this perspective, and gives his accounts about which of these disorders correspond to within-module problems and which to problems in reporting to the working memory. He rules out systemic damage to WM as explanation of these disorders. As far as I can understand, he does not make a distinction between problems with reporting
[1] Richard L. Lewis. Architecture matters: What Soar has to say about modularity , 1993 .
[2] Allen Newell,et al. The Knowledge Level , 1989, Artif. Intell..
[3] Richard L. Lewis. Interference in short-term memory: The magical number two (or three) in sentence processing , 1996, Journal of psycholinguistic research.
[4] Todd R. Johnson,et al. Task-structure analysis for knowledge modeling , 1992, CACM.