Much of the training research investigating specific training has focused on the nine-dot problem, which is a visuo-spatial insight problem (e.g., Burnham & Davis,
1969; Chronicle et al., 2001, Experiment 3; Kershaw & Ohlsson, 2004; Weisberg & Alba, 1981, Experiment 2), with training effects ranging between 40% (Kershaw & Ohlsson, 2004, Experiment 3) and 43% (Weisberg & Alba, 1981, Experiment 2). When training entails practice in drawing non-dot turns, which is critical to solving
this problem, performance is higher (Kershaw & Ohlsson, 2004; Weisberg & Alba, 1981,), which suggests that the degree of specificity of the training to test problems is crucial for positive transfer to occur. However, the training effects are modest, and suggest that other sources of difficulties may operate that affect performance on the nine-dot problem (Kershaw & Ohlsson, 2004). Further, the context generated by this problem has a strong inhibitory effect, therefore despite the fact that training is designed to improve performance on a specific problem, the solution rate is far from
100%. Finally, specific training reduces the novelty of the test problem by giving practice in problems that are structurally similar. From an applied perspective, the generality of these studies is limited to the test problem under study.
There is a lack o f specific training studies designed to improve performance on verbal insight problems. This is most likely due to the fact that, unlike visuo- spatial problems, verbal insight problems do not have a clear goal state. Further, there is greater variability in the content and constraints associated with verbal problems. Therefore, the challenge is to devise and implement specific training that can facilitate transfer to verbal insight problems. One suggestion is to train solvers to use heuristics or rules of thumbs to solve insight problems that share certain characteristics. The remainder of this section provides a discussion of how heuristic-based training can be used to solve verbal insight problems.
Heuristic-based training is one type o f training that has been used successfully in other problem solving domains, including industrial faultfinding (Shepherd,
Marshall, Turner, & Duncan, 1977) and mathematics (Schoenfeld, 1979). General heuristics such as hill-climbing and means-end analysis have been successfully applied to solve well-defined, move problems (Chronicle et al., 2004; MacGregor et al., 2001; Newell & Simon, 1972; Ollinger et al., 2006; Ormerod et al., 2002). For
example, Kaplan and Simon (1990) suggested that, in solving the mutilated
checkerboard problem, a visuo-spatial insight problem, solvers applied heuristics to narrow the space o f possible moves to achieve solution. However, such heuristics cannot be applied to solve verbal insight problems as their goal state is ill-defined. It is also important to note that most training studies provide training with only one test problem rather than looking at a wider range o f problems, which limits the
applicability of the training to other insight problems. Thus, general problem solving heuristics that might apply across a wider range of problems is lacking in the insight problem solving domain (Chronicle et al., 2004), and further research is needed in this area, which is explored by the two experiments reported in this chapter. Although heuristic training is also narrow in some sense, it nonetheless is better than some literature that has only looked at one test problem (e.g., Weisberg & Alba, 1981).
One of the difficulties faced in designing heuristic-based training for verbal insight problems is that the nature of the stereotypical assumptions or constraints associated with the problems are so idiosyncratic (cf. Isaak & Just, 1995), such that it is difficult to envisage how knowledge of any one of these could be used to facilitate the solution of other problems. One possibility is to develop an intermediate
categorisation that identifies commonalities between particular types of insight
problem and to design heuristic-based training based on these categories. This was the rationale for Experiments 1 and 2 reported in this chapter.
One study that attempted to categorise insight problems for the purpose of training is that by Dow and Mayer (2004; reviewed in Chapter 3, Section 3.3.1). Dow and Mayer categorised problems by their overall nature i.e., verbal, visuo-spatial, mathematical or a combination of these. They trained participants in solving one or more of these types o f problems and performance was tested on the different problem
categories. It was found training in verbal or mathematical problems did not improve performance on test problems of the same category. Only visuo-spatial training improved solution to the same category of test problems. In another experiment, no difference was found between verbal and visuo-spatial training in solving verbal problems when compared to a control group. It was possible that there was too much variability between the nature of the problem constraints and the categories were too broad which affected the results. Although it is useful to categorise problems for the purpose of training, it may be more beneficial to categorise problems in terms of constraints in order to narrow the categories and thus increase the rate of transfer
Consequently, as part of the training, Experiment 1 in this chapter first identified commonalities in constraints between particular types o f insight problem and heuristics were developed. Thus, participants were first made aware of problem constraints and then trained on two heuristics that involved identifying ambiguous words and ambiguous names within a problem that lead to solution. For example, ‘guide* is an ambiguous word as it can refer to either a human/animal guide or a map. For ambiguous names, the names in the Anthony and Cleopatra refer to animals, not humans. The heuristic to consider names as ambiguous should discourage participants from making this assumption during testing. Experiment 2 aimed to improve solution of problems containing ambiguous words by adapting the training utilised in
Experiment 2 and by using a new set of test problems. It was predicted that, in each experiment, positive transfer would be restricted to trained category of test problems as suggested by theoretical formulations of transfer (e.g., Anderson, 1983; Gick & Hollyoak, 1980; Thorndike & Woodworth, 1901), and no transfer would take place on problems that were out of scope to the training. The think aloud methodology was utilised in both experiments.