Procedural Instructions 1 Running head : INTERLEAVING READING AND ACTING Interleaving

Memory for an interactive procedure learnt from written instructions is improved if the procedure can be carried out while the instructions are being read. The size of the read-act cycle was manipulated by comparing “chunked” instruction-following, in which 3 or 4 steps are read then enacted, with single-step conditions. In two experiments, enforced chunking improved subsequent unaided performance of the procedure. In Experiment 3 participants were allowed to manage the interleaving of reading and acting. The imposition of a small behavioural cost (a single mouse pointand-click operation) on the switch between instructions and device encouraged more chunking, and better subsequent test performance. We conclude that the interleaving of reading and acting is an important practical concern in the design of interactive procedures, and that more effective chunk-based strategies can quite readily be encouraged. Procedural Instructions 3 Interleaving reading and acting while following procedural instructions Good procedural instructions for interactive devices must satisfy two criteria. First, they must support performance. Like all procedural instructions they should effectively communicate the procedure they describe, so as to allow users who don’t know the procedure to enact it successfully and efficiently. Second, they must support learning. In common with instructions for all procedures that will be used repeatedly, they should facilitate subsequent memory for the procedure, so that it might later be performed without consulting the instructions. Unfortunately, the nature of human learning is such that these two criteria can sometimes conflict. Schmidt and Bjork (1992) have pointed to a general phenomenon in the relationship between performance during training episodes and later retention of the trained skill. Better performance during training does not always lead to better retention. Training should introduce difficulties for the learner, so as to better model the post-training task. (As Schmidt and Bjork discuss, this general principle of training is strongly related to the transfer-appropriate processing account of several phenomena in the literature on human memory.) How might procedural instructions be designed so as to follow the Schmidt and Bjork paradigm and provide transfer-appropriate practice opportunities for the learner? Of course, not all manipulations that introduce difficulties during learning are beneficial for the learner. Simply making the instructions unclear is unlikely to be effective, however much this idea may have informed the design of some commercial user manuals. The criterion that quality instructions must communicate the procedure that they describe cannot be ignored. Procedural Instructions 4 Some guidance can be obtained from the theoretical literature on text comprehension. This literature introduces a useful distinction between two components of the memory representation derived from reading a text (see Zwaan & Radvansky, 1998; Kintsch, 1998). The textbase represents the propositional information contained in the text itself, and will be the primary contributor to tests like recall of the text. The situation model represents the situation described by the text, it integrates text propositions with inferences and information derived from the reader’s background knowledge. The situation model is the primary contributor to tests that go beyond memory for the propositional content of the text. These tests include problem solving and inference making using knowledge derived from the text. Informed by this distinction, work by McNamara, Kintsch, Songer, and Kintsch (1996), has shown how expository text can be designed to introduce difficulties for readers in exactly the productive manner advocated by the Schmidt and Bjork conception of training. These authors created two versions of target texts, one more coherent than the other (one experiment used a text about traits of mammals, a second used a text about heart disease). Coherence cues were provided by linking clauses with appropriate connectives and by inserting topic headings. The level of readers’ background knowledge on the topic of the text was also assessed with a pretest. After reading a text participants were given tests of the textbase (free recall of the text propositions and specific factual questions about the contents of the text) and tests of the situation model (problem solving based questions, questions requiring inferences from the text, and a concept-sorting task). McNamara et al (1996) reported that for measures that tested the textbase, the high coherence texts produced better performance. However, for situation model measures, test performance for high knowledge readers was better when they read the Procedural Instructions 5 low coherence text. McNamara et al. argued that limiting the coherence of a text forced readers to engage in compensatory processing to infer unstated relations in the text. This compensatory processing supported a deeper understanding of the text, in that the information in the text became more integrated with background knowledge. Thus, for high knowledge readers the texts that were more difficult to read improved the situation model by encouraging more transfer-appropriate processing. Lowknowledge readers were, presumably, unable to achieve the compensatory inferences, and therefore did better with more coherent texts. Because the textbase does not incorporate background knowledge it was not enhanced by any compensatory processing. The work of Diehl and Mills (1995) further illustrates the relevance of the theory of text comprehension to the design of instruction for interactive procedures. They argue that in the case of procedural instructions the distinction between situation model and textbase maps directly onto a distinction between memory for the procedure (as tested by later task performance) and memory for the instructions themselves. Texts describing how to complete a task using a device (setting an alarm clock, or constructing a child’s toy) were provided. While reading a text participants were required to either perform the task (read and do), or do nothing (read only). (In addition, Diehl and Mills studied some intermediate conditions, such as read and watch experimenter do. These conditions produced intermediate results and are not relevant to the current argument.) The effect of these training methods was then examined by asking participants to recall the text, and then complete the task. Diehl and Mills reported that the increased exposure to the device in the read and do condition resulted in improved task performance times relative to the read only Procedural Instructions 6 condition. However, text recall was better in the read only condition, supporting the conceptual separation of textbase and situation model. One successful practical approach to the design of instructions for interactive devices is perhaps quite strongly related to this more theoretically oriented work. The concept of a ‘Minimal Manual’ was outlined by Carroll (1990). It sought to minimize the extent to which instructional materials obstruct learning. Crucially, a welldesigned Minimal Manual does not necessarily optimize the speed at which users can perform procedures as they read. Carroll’s manuals avoided explicit descriptions that encouraged rapid but mindless rote performance. Instead the emphasis was on active learning whereby learners were encouraged to generate their own solutions to meaningful tasks. This process was facilitated in part by reducing the amount of text provided and including information about error recovery. Like Carroll, our goal in this article is primarily practical. However, rather than developing a general heuristic framework for instruction, we focus on a particular technique that exploits the idea of transfer-appropriate practice, following the principle of Schmidt and Bjork and the methods of McNamara and colleagues. Like the manipulations of Diehl and Mills, our central interest is not the design of the instructions per se, but rather the way the instructions are read and used. Diehl and Mills’ reported advantage for reading-and-doing over reading alone has no real practical implication, as it is difficult to imagine anyone advocating isolated reading as a preferred method. However, we suggest that the way learners manage the interleaving of reading and doing will affect their later retention, and thus offers an important lever for improving instruction. Many procedural instructions have a natural step-wise structure, and in these cases it is possible to execute the procedure while reading with minimal load on Procedural Instructions 7 memory. Learners can read a single step, then execute it before reading the next step. Such an approach is low on effort (and therefore attractive to the learner), but also low on transfer-appropriate practice and therefore, we predict, poor at encouraging retention. If learners could instead be prompted to read several procedural steps before enacting them, performance would be made more effortful, but learning might benefit. Readers would be encouraged to integrate the information across the chunk of procedural steps, and the increased memory load would provide transfer-appropriate practice. Our strategy for developing and testing this idea is as follows. First, we report two experiments in which participants are forced into either a step-wise or a chunkbased strategy for interleaving reading and acting. These experiments test our prediction that reading-by-chunks will tax performance during training, but improve learning, in particular retention of the procedure. Next, we report a third experiment which develops a more subtle, indirect manipulation of chunking which we believe holds greater promise of practical application.