The organization of perception and action. A theory for language and other cognitive skills

Abstract

, 1987, 233 pages Donald G. MacKay's monograph --in line with the current Zeitgeist--is a theoretical contribution to the effort to overcome the increasing difficulty of explaining behavior in terms of"boxological" models. Box-models, in fact, fall to accommodate flexible changes and adaptations of every kind of processing device. It is this variability that produces answer-inconsistency and post-lesional features of neuropsychological performance. The hundred-year-old use of "boxes" (ie rigid segmentations of essentially sequential processings similar to biochemical steps in a metabolic pathway) harks back to the first steps of Neuropsychology, that is to those of cognitive positivism that inspired the Wemicke (1874) --Lichrheim (1885) anatomo-psychological aphasia model and the more complex psychoanalytical theorizations. Modern cognitive psychology resumed this approach with the aim of fiUing out the black box of behaviorism with sequences of processing steps that act as inter-related working hypotheses for subsequent goal-directed experiments. Popular instances are the psychological box models for memory developed by Waugh and Norman (1965), Atkinson and Shiffrin (1972) and many others, particularly in the linguistic domain. MacKay models psychological processing in connectionistic terms, with perception as input and action/language as output taking a probabilistic path through interconnected "nodes". In MacKay's and in other models of this kind nodes and connections are loaded by "weights", whose variability controls and directs the quality of processing from input to output. The weights change autonomously according --inter alia --to quantitative features of the incoming information over long periods of time. Again the metaphor with the inihibitory/excitatory features of synapses should be avoided. Nodes and connections gradually master more and more complex cognitive performances, the network learning to produce highly sophisticated and flexible answers. This is nowadays much more than armchair speculation. Actually, artificial devices with the above properties are already in operation, providing an opportunity for routh simulation of the degraded performances of a brain-damage patient. Such simulations are studied on ad hoc"damaged" artificial devices (see Karalyn Patterns, Cambridge, research on the Seidenberg-McClelland, Montreal, model of spelling-to-sound transformation in reading). They are even capable of producing the phenomen response-inconsistency of current neuropsychological evidence. All this and much more belongs to every kind of connectionistic modelling of processing, not only to MacKay's (see "Connectionistic models and neuropsychological phenomena", chaired by Tun Shallice, Cambridge, 6th European Workshop on Cognitive Neuropsychology, Brixen, Italy, 24-29 January 1988). Unfortunately, many of these experimental aspects are not considered in MacKay's book.MacKay's "Node Structure Theroy" (NST) is a general theory of skilled behavior in speech, perception and action that, like many other ...

Time Course of Frequency Effects in Spoken-Word Recognition: Evidence from Eye Movements

“…Another way to incorporate frequency in interactive activation models, advocated by MacKay (1982MacKay ( , 1987, is to make the weights (or connection strengths) associated with lexical units proportional to their frequency. Although implementations of frequency in resting levels or in connection weights are very similar to each other (e.g., Dell, 1990), our simulations show that the two implementations differ in one crucial respect.…”

“…For example, in models with discrete lexical representations such as the ''Cohort'' model (Marslen-Wilson, 1987) or the TRACE model (McClelland & Elman, 1986), high-frequency words would be processed faster than low-frequency words because frequency determines either the baseline activation level of each lexical unit (McClelland & Rumelhart, 1981;MarslenWilson, 1990) or the strength of the connections from sublexical to lexical units (MacKay, 1982(MacKay, , 1987. In distributed learning models, the representations of high-frequency words would be activated more rapidly because highfrequency mappings are better learned, resulting in stronger connection weights (Gaskell & Marslen-Wilson, 1997;Plaut, McClelland, Seidenberg, & Patterson, 1996).…”

Online order control in the psychological refractory period paradigm.

“…One is to select the first subtask and rely on two associations that are held simultaneously: "IF [the color subtask is first] THEN [start executing the letter subtask as second]" and "IF [the letter subtask is first] THEN [start executing the color subtask as second]." Another possibility is based on a variety of inhibition mechanisms such as self-inhibition, namely inhibiting the subtask just performed (MacKay, 1987), or on lateral inhibition (Norman & Shallice, 1986). In the General Discussion section, we argue that the results support our conjecture concerning an explicit order representation and are incompatible with the just-mentioned alternative strategies.…”

Investigation of phonological encoding through speech error analyses: Achievements, limitations, and alternatives

“…One should, however, keep in mind that much of this evidence stems from analyses of errors involving word onsets, and that much less is known about the movement patterns and coherence of segments outside the word onset. In the next section, it will be shown how these generalizations are fleshed out in two models of phonological encoding, namely in the models proposed by Shattuck-ufnagel (1979Shattuck-ufnagel ( , 1983Shattuck-ufnagel ( , 1986Shattuck-ufnagel ( , 1987 and Dell (1986Dell ( , 1988; for related models see Berg, 1988;Harley, 1984;MacKay, 1982MacKay, , 1987Stemberger, 1985b). e 2.1.…”