of working memory as a limited-capacity“phonological loop”in which phonological material is stored, maintained in
sequence, and rehearsed. The tasks of processing this information and directing itsflow are ascribed to a“central
executive.”Characteristic experimental tasks used to probe working memory in this tradition are tests of digit span
(frontward and backward), recall of lists of words, and repetition of nonsense words.
What strikes me about this tradition is that it apparently neglects what would appear to be by far the most important
function of linguistic working memory: understanding and producing spoken language. In order to construct a
meaningfora sentence oneis hearing, onemust do morethan retrievethemeanings oftheindividual words:onemust
determine the relations among the word meanings, based on the syntactic structure of the sentence—which is in turn
based in part on the linear order of the words. Consequently, the string of words must be stored in working memory,
not just in order to be able to repeat the mback, but in order to understand what is related to what. Si milarly, in
language production, phonological workingmemory is the“place”wherethe pronunciations of wordsare put in linear
order on the basis of the syntactic and semantic relations in the intended utterance, prior to the construction of a
motor program that produces the utterance.^99
206 ARCHITECTURAL FOUNDATIONS
(^99) Logically speaking, theworkingmemory system used for understanding and producingsentencesmightbe different from theonewhose capacityis beingmeasured by digit
spans and lists of unrelated words. One reason they might seem superficially different is that language processing can deal with much longer strings of words than are
observed in digit span tasks. However, the earliest literature on working memory (Miller 1956) observed that its limited capacity depends not on the sheer amount of
information, but rather on how that information is“chunked.”Phonologicalprocessing certainly involveschunking word strings into larger segments—a process that this
tradition of research on working memory simply avoids by restrictingtest stimuli to lists and nonsense words.Gathercoleand Baddeley (1993) do explicitlyargue that their
“phonological loop”does not playa central role in speechperceptionand production. Addressing production (ch. 4), theyproduceevidence that the“phonologicalloop”is
not to be identified with motor programs for speech, and conclude that it is not involved in production. But they fail to sufficiently distinguish levels of structure:
phonological structure simply is not a motor program, so phonological workingmemory should not be expected to contain motor programs. They also citeevidence from
Shallice and Burterworth (1977) concerning a patient with“conduction aphasia,”whose repetition and digit span skills were impaired but who spoke fairly normally.
Problems with repetition, however, surely involve lexical access as well as phonological working memory, so without further analysis of the task it is not clear how to
interpret this case.Turning to perception (1993: ch. 8), Gathercole and Baddeley give evidence that working memory load impairs only complex language processing
(embedded relative clauses, reversible passives, and the like). They also cite patients whose digit spans are impaired but whose language comprehension is not impaired
except with complex sentences. They therefore make a distinction between“online”comprehension based on“raw input forms”and“offline”comprehension based on
storage in the“phonological loop”theyclai monlythelatter is affectedin thecases theycite.Again, theydo notask what couldpossiblybe meant by“raw inputform.”It is
surely not acoustic: it must already be converted into phonological form before it can invoke lexical access and give rise to semantics. My intuition is that an impaired
phonological working memory in the sense proposed here (Fig. 7.2) would have just the desired effect: it would work acceptably for easy sentences thatcould be rapidly
“chunked,”butwouldbreakdownif“chunks”couldnotbedeterminedwithoutextensivesubsequentprocessing.Tobesure, itisofinteresttoknowwhichelementsofa list
are more likely to be remembered after different time intervals and under different simultaneous loads. But it would by now seem time to establish a meaningful line of
communication between this enterprise and psycholinguistic research.