Foundations of Language: Brain, Meaning, Grammar, Evolution

(ff) #1

interaction of components is constrained by the rules of grammar—that is, it is notentirelyfree.


7.3 Remarks on working memory


One often sees lexical access characterized simply as something like“activating a word node”in memory. This is far
too crude, and calls for several refinements.


First of all, the competence theory shows there is no single“word node.”A lexical ite mis a co mplex association of
phonological,syntactic,and semanticstructures. I stress herestructures, notjustfeatures. Onecannotjust thinkofa word
node as activating a number of other nodes that together constitute a collection of independent features. Rather, any
adequate account has to include provisionfor hierarchical phonological structure in complex morphological items, for
hierarchical syntactic structure in idioms, and for hierarchical conceptual structure in just about any word (see Chapter
11).


Next, let us return to an issue raised in section 3.5. There I argued that working memory cannot just consist of the
nodes in long-term memory that are activated in current processing. The most important reason, it will be recalled, is
the“Proble mof2”: a structureconstructed inworkingmemory may containtwoor moreinstances ofa typestored in
long-term memory. For instance,The little star's beside a big starcontains two instances ofstar. These two instances
cannot be kept distinct if each consists simply of an activation of the lexical entry forstarin long-term memory.


For thisreason, theolder view (e.g. Neisser 1967, Baddeley 1986) ofworkingmemory as a separatefacilityin thebrain
appears moreappropriate: material from long-term memory is“retrieved”intoworkingmemory whereitcan undergo
transient processing. If we don't like the idea of“copying”material into working memory, an alternative way to think
about it might be that working memory is a set of indices or pointers or transient bindings to long-term memory.
Although itis still altogether mysterioushow such a mechanism can be implementedin neural terms, thenature of the
task demands something with these functional properties.


In addition, just having an item in working memory still isn't enough for it to be part of an utterance. It must also be
integrated into the larger structures being assembled by the integrative processors—in the phonological, syntactic,and
semantic departments of the“blackboard.”


It is important to differentiate the present view of working memory from an extensive research tradition associated
with such work as Baddeley (1986), Gathercole and Baddeley (1993), and Gathercole (1999). This tradition thinks


IMPLICATIONS FOR PROCESSING 205

Free download pdf