Accounts of task-set control generally assume that the current task's stimulus-response (S-R) rules must be elevated to a privileged state of activation. How are they represented in this state? In 3 task-cuing experiments, we tested the hypothesis that phonological working memory is used to represent S-R rules for task-set control by getting participants to switch between 2 sets of arbitrary S-R rules and manipulating the articulatory duration (Experiment 1) or phonological similarity (Experiments 2 and 3) of the names of the stimulus terms. The task cue specified which of 2 objects (Experiment 1) or consonants (Experiment 2) in a display to identify with a key press. In Experiment 3, participants switched between identifying an object/consonant and its color/visual texture. After practice, neither the duration nor the similarity of the stimulus terms had detectable effects on overall performance, task-switch cost, or its reduction with preparation. Only in the initial single-task training blocks was phonological similarity a significant handicap. Hence, beyond a very transient role, there is no evidence that (declarative) phonological working memory makes a functional contribution to representing S-R rules for task-set control, arguably because once learned, they are represented in nonlinguistic procedural working memory.

When stimuli afford multiple tasks, switching among them involves promoting one of several task-sets in play into a most-active state. This process, often conceptualized as retrieving task parameters and stimulus-response (S-R) rules into procedural working memory, is a likely source of the reaction time (RT) cost of a task-switch, especially when no time is available for task preparation before the stimulus. We report 2 task-cuing experiments that asked whether the time consumed by task-set retrieval increases with the number of task-sets in play, while unconfounding the number of tasks with their frequency and recency of use. Participants were required to switch among 3 or 5 orthogonal classifications of perceptual attributes of an object (Experiment 1) or of phonological/semantic attributes of a word (Experiment 2), with a 100 or 1,300 ms cue-stimulus interval. For 2 tasks for which recency and frequency were matched in the 3- and 5-task conditions, there was no effect of number of tasks on the switch cost. For the other tasks, there was a greater switch cost in the 5-task condition with little time for preparation, attributable to effects of frequency/recency. Thus, retrieval time for active task-sets is not influenced by the number of alternatives per se (unlike several other kinds of memory retrieval) but is influenced by recency or frequency of use.

Hick’s law describes the increase in choice reaction time (RT) with the number of stimulus-response (S-R) mappings. However, in choice RT experiments, set-size is typically confounded with stimulus recency and frequency: With a smaller set-size, each stimulus occurs on average more frequently and more recently than with a larger set-size. To determine to what extent stimulus recency and frequency contribute to the set-size effect, stimulus set-size was manipulated independently of stimulus recency and frequency, by keeping recency and frequency constant for a subset of the stimuli. Although this substantially reduced the set-size effect (by approximately two-thirds for these stimuli), it did not eliminate it. Thus, the time required to retrieve an S-R mapping from memory is (at least in part) determined by the number of alternatives. In contrast, a recent task switching study (Van ‘t Wout et al. in Journal of Experimental Psychology: Learning, Memory & Cognition., 41, 363–376, 2015) using the same manipulation found that the time required to retrieve a task-set from memory is not influenced by the number of alternatives per se. Hence, this experiment further supports a distinction between two levels of representation in task-set control: The level of task-sets, and the level of S-R mappings.

Theories of instruction following assume that language contributes to our ability to understand and implement instructions. The two experiments reported here investigated that assumption. Participants (total N = 96) were required to learn a series of novel tasks, with each task consisting of six arbitrary stimulus-response rules. All tasks were preceded by an instruction phase (a visual depiction of the correct stimulus-response rules for each task), during which participants performed a verbal distractor task (articulatory suppression), a non-verbal distractor task (foot tapping) or no distractor task. Additionally, the duration of the instruction phase was varied so that it was either long (60 s) or short (30 s in Experiment 1, or 10 s in Experiment 2). In both experiments participants made more errors when they had performed articulatory suppression during the instruction interval, compared to the foot tapping and no distractor task conditions. Furthermore, Experiment 2 found that this detrimental effect of articulatory suppression was especially pronounced with a very short instruction duration. These findings demonstrate that language plays a crucial role in the encoding of novel task instructions, especially when instructions are encoded under time pressure.