Kellogg, Ronald T., Casey E. Turner, Alison P. Whiteford, and Andrew Mertens. “The Role of Working Memory in Planning and Generating Written Sentences.” The Journal of Writing Research 7.3 (2016): 397-416. Web. 04 Apr. 2016.

Kellogg et al. conduct an experiment designed to shed light on how working memory (WM) relates to the sequence of the processes that go into generating written sentences. They draw on a body of research like that of Linda Flower and John Hayes that posits a conceptual “planning” phase that is largely nonverbal and a phase of grammatical and orthographical translation that leads to writing or typing (398).

The models propose that the planning stage draws on a visual component when the information to be translated involves concrete nouns that evoke images (399). The researchers hypothesized that a visual working-memory load (that is, a non-verbal load) during the writing task would interact with the planning, or non-verbal, stage, while imposing a verbal working-memory load would impact transcription of the written sentences during the phase when grammatical processing was taking place.

To test these hypotheses, Kellogg et al. provided participants with two different kinds of prompts and asked for two different kinds of sentences using these prompts. The prompts were paired words to be used in sentences. One set of pairs involved words that were semantically related, such as “door” and “knob.” The other set provided words that were not related, such as “ice” and “jail.” All word pairs were concrete and taken from a scale tested to calculate how easily they would be recognized (401).

The rationale for these choices was that more planning was needed to link unrelated word pairs in a sentence, so that the load on visual memory would also increase (399). But because the models specified that planning takes place before “grammatical encoding,” the kind of word pair required should have no effect on this latter process (400).

To investigate the relationship of verbal memory to the sentence-generating process, the researchers imposed another manipulation by asking half of the subjects to compose active-voice sentences, while the other half composed passive-voice sentences. In this case, the rationale was research showing that passive-voice constructions are more complex than active and thus should burden “verbal working memory” but not visual (400). Subjects were screened for their ability to produce the kinds of sentences required as instructed in the prompts (403).

Thus the protocol required subjects to produce sentences, either in the passive or active voice as instructed, in three conditions: while being asked to perform a “concurrent” visual working-memory task, a concurrent verbal working-memory task, or no concurrent task (401). The visual concurrent task involved studying and learning a set of images chosen from “the SPSS Marker Set in Microsoft Word” that were “designed to be not readily named” (401). In contrast, the verbal concurrent task was composed of nine digits that “could be coded verbally” (401). In both cases, after completing the required sentence, the participants were asked to determine whether a set of symbols or digits matched the ones they had previously seen (402-03).

The level of accuracy in this ability to match the symbols or digits displayed after the writing task with those seen previously constituted the main data for the study (409). The researchers wanted to test the hypothesis that impeding visual working memory would not interact with the verbal process of grammatical encoding, but that impeding verbal working memory would do so (399).

Additional data came from measuring factors such as the length of the sentences produced, the number of words produced per second in each trial, the time involved in producing the sentences, and the time before typing began. These factors were controlled for in the interpretation of the major findings.

The authors’ predictions, therefore, were that factors that made planning harder, such as efforts to work with unrelated nouns, would show up in less accurate results for the visual working-memory task, the symbols, while factors that impeded grammatical encoding, such as writing passive-voice sentences, would manifest themselves in less accuracy in recalling the verbal working-memory task components, the digits.

Unexpectedly, they found that even though, as predicted, passive voice sentences took longer to write, required more words, and resulted in fewer words per second, the type of sentence made “no reliable difference” on the verbal concurrent task (the digits): “if anything, actives produced more interference than passives” (410). They found also that, contrary to expectations, related word pairs “most disrupted the verbal WM task” (410). Thus, the operations assumed to be simplest required the most verbal working-memory effort, and factors that were expected to affect visual working memory because they were presumably involved in planning did not produce the hypothesized interference with the visual task (409-10).

That the presumably simpler effort involved in producing an active-voice sentence using a related pair of words demanded more verbal working memory led the researchers to consult the work of M. Fayol, who proposed that the “Kellogg (1996) model” may fail to take into account “an understanding of the temporal dynamics” involved in writing sentences (411). To explain their findings in the current study, Kellogg et al. posited that the conceptual work of planning for the simpler active-voice/related-pair resulted in “a single chunk that was then immediately cascaded forward to grammatical encoding” (411). In contrast, they suggest, the more difficult planning for a sentence using the unrelated pair occurred incrementally, possibly in parallel with grammatical encoding or during pauses, for example, after typing the first of the words. Thus, the grammatical encoding process that would have shown up as a demand on verbal working memory was broken up into “piecemeal” activities by planning phases rather than implemented all at once (411-12). Such intermittent planning/encoding has been demonstrated in studies of speech and typing (412). In short, easier planning can result in more pressure on verbal working memory.

The authors conclude that “the role of WM in written sentence production is markedly more complex than previously postulated” (414).