_C_O_N_S_T_R_A_I_N_T_S _O_N _C_H_A_N_G_E _I_N _C_H_I_L_D_R_E_N'_S _P_R_O_B_L_E_M-_S_O_L_V_I_N_G: _T_H_E _R_O_L_E _O_F _S_P_E_C_I_F_I_C _S_T_R_A_T_E_G_I_E_S Stephanie Thornton CSRP 302, November 1993 Cognitive Science Research Paper Serial No. CSRP 302 The University of Sussex School of Cognitive and Computing Sciences Falmer BRIGHTON BN1 9QH England, U.K. 2 _C_O_N_S_T_R_A_I_N_T_S _O_N _C_H_A_N_G_E _I_N _C_H_I_L_D_R_E_N'_S _P_R_O_B_L_E_M-_S_O_L_V_I_N_G: _T_H_E _R_O_L_E _O_F _S_P_E_C_I_F_I_C _S_T_R_A_T_E_G_I_E_S Stephanie Thornton School of Cognitive and Computing Sciences University of Sussex Brighton BN1 9QH, England, U.K. E-mail: thorn@cogs.susx.ac.uk November 1993 _A_b_s_t_r_a_c_t This paper reports a study of the changes which children aged four to ten years made to their problem-solving whilst playing the "Twenty Questions" game. Factors specific to particular strategies are shown to play a central role in triggering episodes of change, whilst developmental factors (prior skill levels) determined the direction of the change. Analyses of these effects are discussed in relation to the debate as to the relative contributions of domain general and domain specific mechanisms in the incidence of change in children's problem-solving. 1 Children change their strategies during the course of problem-solving. Some of these changes are made in search of a solution to the problem. Others have no clear connection to the demands of the task, and occur in procedures already successfully solving the problem. Both types of effect, and particularly the latter, have been interpreted as reflecting the workings of developmental processes (e.g., Inhelder, Sinclair & Bovet, 1974; Karmiloff-Smith, 1979, 1984; Klahr, 1984). A key problem for developmental psychology is to identify the processes which generate these phenomena of change. A number of theoretical models have been put forward (e.g., Anderson, 1983; Klahr, 1982, 1984; Karmiloff- Smith, 1984, 1992) However, there is no consensual view. Major issues remain unresolved. For example, there is the problem of the relative contributions of domain-neutral and domain-specific processes. A domain-neutral process operates in all domains. It neither relies on, nor directly uses the content of the domain for its functioning. Processes of this kind are inevitably _s_y_n_t_a_c_t_i_c, in that they must act on abstract structures or forms, without regard to the concrete content or context of those structures. An example of such a syntactic domain-neutral rule, simplified from an illustration given by Hunt (1989) would be: "If the goal is to produce a structure of type Z, and the current knowledge state is a structure of type Y, and if there is an operation that changes structures of type Y into type Z, then attempt to transform the current knowledge structure into a structure of type Z". This same rule applies across any and all domains - so long as its conditions are met. In contrast, a domain-specific process is particular to a given domain, and irrelevant in all others. Processes of this kind are _s_e_m_a_n_t_i_c, in that they operate directly through manipulations of representations of the particular environment or domain in which they will be used. An example (again, modified from Hunt 1989) of such a rule would be: "Always protect 2 the king first if you want to win at chess". Rules of this type will apply only in particular contexts, just as the illustrative rule here applies only to chess. The majority of studies of mechanisms of cognitive change and development have focussed on the role of domain-neutral syntactic factors. For example, syntactic principles underly virtually all computational models (cf. Anderson, 1983, Klahr, 1984). Syntactic mechanisms are also the basic element in Karmiloff-Smith's (1992) representational redescriptive or RR processes (where the same RR mechanism operates domain-neutrally, although its products may be domain specific). These theories emphasize the role of meta-processes in producing change to problem-solving procedures, by monitoring ongoing problem-solving, identifying opportunities for the execution of some syntactic rule of the kind illustrated above, and executing it - the goals and criteria for such rules being predetermined, and common across all domains. However, it is almost certain that cognitive change also involves semantic, domain-specific processes (cf. Carey, 1985; Keil, 1984, 1986). Both domain-general and domain-neutral mechanisms have been shown to contribute importantly to problem-solving at large (see Hunt, 1989 for a review). Few developmentalists would want to argue that developmental processes differ from cognition at large in this respect. But there has been little research into the part played by domain-specific processes in development or cognitive change. Domain-specific factors have been explored almost exclusively in the context of problem-solving in the strict sense - i.e., processes searching for the solution to a problem. There is almost no work exploring how domain-specific processes might generate change in procedures already successful in completing the task. Notable exceptions to this are the work of Siegler and Jenkins (1989) on the generation of new strategies, and by Wales and Thornton (1991) on conceptual change in 3 problem-solving. Both of these lines of work emphasise the role of the semantic content of the initial strategy in mediating transitions to new strategies and representations. Both studies of domain-neutral mechanisms, and studies of domain- specific ones have yielded impressive evidence in support of the activity of each kind of mechanism in generating within-session change in problem- solving. However, research in one vein is largely independent of that in the other. There are no explorations of the relative contribution of each kind of process, or of the interaction between the two. The aim of this paper is to present a study addressing these issues. In order to examine the relative role of domain-neutral and domain- specific processes, it is essential to identify an empirical arena in which each type of process generates predictions, and where these predictions differ. This poses some problems. The data normally used in exploring domain-neutral processes (e.g., demonstrations of very similar effects across a range of situations) are very different from those used to explore domain-specific ones (detailed analyzes of individual episodes of change). Thus the normal types of prediction from each type of model are at very different levels of analysis. A worse problem is that, for most episodes of change, it is possible to generate either a domain-neutral or a domain- specific explanation: neither approach has been specified in sufficient detail to allow the identification of any critical differentiating characteristic of change contributed by one type of mechanism or the other. The approach taken here is to move away from examining the form change takes, and to look in more detail at the incidence of change. A consideration of the generic form of domain-neutral and domain-specific models reveals that both predict that the incidence of change will be differentially associated with different strategies - but for different reasons. 4 In the case of domain-neutral models, the content of the strategy in which these processes operate is irrelevant to their action. Nonetheless, specific strategies will still have an effect on the occurrence of change: syntactic processes will only act to generate change if their enabling conditions (e.g., knowledge structure of type X) are met. Strategies are likely to differ in the extent to which they generate the appropriate enabling conditions for syntactic rules to operate. For instance, a syntactic process for cutting out redundant steps in a procedure can only operate in a strategy which involves redundant steps. Strategies more often generating the enabling conditions for syntactic processes will be more associated with change than those less often yielding these enabling conditions. In the case of domain-specific models, the content of a strategy is more straightforwardly and obviously associated with the incidence of change. For processes of this type, the incidence of change is directly predicated on the interaction between semantic rules and the content of ongoing processing: where ongoing processing meets the rule, change will not be triggered. But where ongoing processing violates the semantic rule, change will be triggered. Where strategies differ in the extent to which they satisfy semantic rules pertinent to the domain in question (as must often be the case), they will differ in their tendency to trigger change. The relationship between specific strategies and change has been explored qualitatively (Siegler & Jenkins, 1989; Wales & Thornton, 1991). But there is no quantitative test of the general prediction that change is more associated with some strategies than with others. Confirmation of the occurrence of "strategy" effects in the incidence of change would open up a new field of empirical exploration of the constraints on cognitive change. For the purposes of the present paper, analyzes of the origin of such "strategy" effects would provide an arena in which to study the relative 5 role of domain-neutral and domain-specific processes. The aim of this study was therefore first to test the general prediction that change in children's problem-solving is associated with some strategies rather than with others; and second, to explore the contribution of either semantic or syntactic factors to this effect. Inevitably, a study of the contribution of specific strategies to change in problem-solving requires very detailed analysis of concrete situations. The analysis presented here relates to one task, the Twenty Questions task. This task has several advantages. First, it has been reported to generate a variety of strategies, and a great deal of within- session change between strategies (Thornton, 1982). In the task as used by Thornton all observed strategies successfully solve the problem, and the task does not constrain the child to become more efficient, quicker, etc. Thus the changes to problem-solving observed in this task involve change in adequate procedures as well as problem-solving in search of a solution. Furthermore, a percentage of the observed within-session change reported by Thornton parallels the classically reported age effect in this task: the transition from asking about items one by one to asking about a group of items with a single question (cf. Mosher & Hornsby, 1966). The task therefore allows scope for a comparison of a given effect as it occurs within session and with increasing age. The experiment below replicated Thornton's (1982) methodology, but with a wider age range of subjects. Experiment 1 _M_e_t_h_o_d _S_u_b_j_e_c_t_s. 49 four year olds (mean 4:11); 34 five year olds (mean 5:9); 23 seven year olds (mean 7:5); 27 nine year olds (mean 9:7). These children were drawn from schools serving a predominantly middle class area in southern England. Within each age group there were approximately equal numbers of each sex. _6 _M_a_t_e_r_i_a_l_s. 16 cards, each bearing a coloured line drawing of one of the following objects: a car, a boat, a hat or a pair of trousers. Each type of object was represented on four cards such that whilst the line drawing remained the same on all four cards, two drawings were to a large scale, and two to a small scale; one of the large scale drawings of each object was coloured blue, the other red. One of the small scale drawings of each object was yellow, the other green. _P_r_o_c_e_d_u_r_e. Children were tested individually. The materials were spread out, in random positions, on the table in front of the child. The experimenter said: "You see these little cards here? In the game we're going to play, I'm going to choose one. I'm going to choose it secretly, in my head, and I'm not going to tell you which one I chose. The game is that you have to find out which one I chose, by asking me questions. But here's the tricky bit! I'm not allowed to say anything but 'yes' or 'no' to answer your questions! Only 'yes' or 'no'. So you have to think of some good questions to ask me, to find out which card I've chosen" Since it is sometimes difficult to know exactly what a child's question means from the form of the words alone, the children were asked to turn face down, after each question, any cards which could now be eliminated. This allows the experimenter to identify the information which the child actually took from the reply to the question, rather than simply guessing from the form of the words alone. Although the child was led to believe that the target item had been chosen in advance, this was not in fact the case. The experimenter always responded so as to maximize the remaining search, but with the constraint that each response be compatible with all previous responses. Each child was asked to play this game three times, in a session lasting between five and ten minutes. This procedure is substantially the same as that reported by Thornton (1982). 7 _R_e_s_u_l_t_s _A_n_d _D_i_s_c_u_s_s_i_o_n. As predicted, the incidence of within-session change in this task was strongly associated with a subset of the strategies which children used in addressing the problem: Overall, five different strategies for addressing the Twenty Questions problem were observed in this study. (All observed responses fell into one of these five categories. Scorings were made by two independent judges, yielding initial agreement of 93%. Discrepancies were the result of carelessness and were quickly resolved). The five strategies observed here were: (1) Touching an individual object, with or without a demonstrative expression such as: "That one?", and eliminating only that object after the experimenter's response. (2) Naming an individual object: i.e., identifying it by its basic category name (e.g., "car?" or "the car?"). The child normally also touched or pointed at a single card. They removed only one of the four cars on receiving the experimenter's reply. (3) Using an adjective alone to identify an individual object, for example asking: "Is it red ?" or "Red ?". The child using this strategy normally also touched or pointed at a single card. Despite its linguistic peculiarity, children scored as using this last strategy clearly intended only one object as their referent - in addition to touching or pointing at one card, they would remove only one of several possible red items, after asking "red ?", and would often ask the same question again later in the game. (4) Describing an individual object more fully, using both name and adjective: for example asking: "the red car ?", and removing the red car after the experimenter's response. (5) Referring to a group of objects with a single question, and 8 removing the whole group in response to the experimenter's reply. Patterns of usage of these strategies across different age groups are shown in Table 1. ------------------------- Insert Table 1 about here ------------------------- None of the children who began the session asking about a group of objects ever changed strategy in this study. By contrast, 58% of the children who had begun by using one or other strategy for identifying single objects subsequently changed their strategy. (This percentage was remarkably similar at all three ages: 59% for four year olds, 55% for five year olds, and 58% for seven year olds. Too few nine year olds asked about individual objects to be considered here). This difference is relatively uninteresting, in that the absence of change away from asking about groups of objects probably reflects no more than that this strategy was the most sophisticated; moves to a less sophisticated strategy would not be expected. The analyzes of the relationship between change and specific strategies below therefore concentrates on those children who began by identifying individual objects, and omits those who started by asking about a group of objects. This effectively omits the nine year old group from the analysis, as there were too few to provide an interpretable sample. Many children who changed their strategy during the experiment did so more than once. The mean number of strategy changes observed per child (among those showing a change) was 3.4, and the median 2. Whilst some children (32%) who changed their strategy more than once used three or more different strategies during the session, the majority used only two strategies, moving from one to the other and then back again, etc. Analyses below focus on initial strategy changes. Subsequent changes are hard to interpret, since they are almost certainly not independent of the initial 9 change. Among children asking about each object individually, there are striking patterns of association between specific strategies and the incidence of change. The data are summarized in Table 2. Collapsing data across age shows that children naming objects more often changed strategy than those touching them (chi square = 26.45, df = 1, N = 59, p<0.001). Children naming objects also changed their strategy more often than those describing objects more fully (chi square = 26.65, df = 1, N = 32, p<0.001). Equally, children using adjectives alone more often changed strategy than those describing the object more fully (Fisher exact test, p<0.01) or touching it (chi square = 4.55, df = 1, N = 56, p<0.05). These strategy effects are very strong. 100% of children who began the task asking about individual objects by either by name or by adjective alone changed their strategy at some point in the session. Only 26% of those touching individual objects, and 17% of those describing objects more fully ever changed strategy. These effects hold up at each age: seven year olds using the "touch" strategy are less likely to change strategy than those using the "adjective alone" strategy (Fisher exact, p<0.005). Five year olds using the "naming" strategy are more likely to change strategy than those using the "touch" strategy (Fisher exact p<0.005) or the "describe more fully" strategy (Fisher exact, p<0.01). Four year olds using the "naming" strategy are more likely to change strategy than those using 'touch" (chi square = 16.12, df = 1, N = 38, p<0.001). ------------------------- Insert Table 2 about here ------------------------- This differential association between certain strategies and within- session change also holds up where the only transition considered is that 10 from asking about single objects to asking about groups of objects. 20% of children who began by asking about individual objects switched to asking about groups of objects at some point during the session. 15% of four year olds, 20% of five year olds and 30% of seven year olds initially asking about single objects showed this effect. The age trend fails to reach statistical significance. This transition occurred in all three games. For some children it was permanent. For others, the effect was more transient, the child moving back and forth between questions about individual objects and questions about groups. Among five and seven year olds beginning the task either touching individual objects or describing them fully, _n_o_n_e ever switched to asking about a group of items. By contrast, fully 50% of five year olds, and 57% of seven year olds who began by either naming individual objects or using an adjective alone made the transition to eliminating groups of objects with each question at some point during the experimental session. This "strategies" effect was not observed for the four year old group. Among 4 year old children, 12% of those initially either touching individual objects or describing them fully switched to asking about a group of items. Only 14% of children initially naming individual objects made this transition. Statistical analyzes of these effects were made, combining the two strategies highly associated with change (naming and using an adjective alone) and the two less associated with change (touching, and describing more fully). The interaction between the linear component of age and initial strategy type was significant (chi Square = 5.3, df = 1, p<0.05). Examination of the data suggests partitioning the 4 year olds from the older children. An analysis of the remaining four cells, relating to the five year olds and the seven year olds, produces a non-significant age effect (chi square = 0.1) but a significant effect for initial strategy (chi square 11 = 14.4, df = 1, p<0.001). There is no significant interaction between these two terms. Thus for each of the two older groups, there is a strong effect of initial strategy on the probability of their making the transition from asking about individual objects to asking about groups of objects. The size of this effect does not differ between the 5 year old and the 7 year old group. On this basis, the 5 and 7 year old data was aggregated and compared to that for the 4 year old group. Two comparisons were made, one for the children who initially identified individual objects by touch or by both name and adjective together, and the other for the children who began by identifying objects by either name alone or adjective alone. Where the first strategy identified objects by name alone or adjective alone, the 5 and 7 year olds show significantly more change to asking about groups of objects than do the 4 year olds (chi square = 5.3, df = 1, p<0.05). Where the first strategy identified single objects by touch or by both name and adjective together, there was no significant difference between the aggregated older groups and the 4 year olds (chi square = 3.0, df = 1, NS). Taking the results in relation to the incidence of within-session change together, these data suggest that the same specific strategies (naming and using an adjective alone) are more highly associated with within-session change at all ages - but that there is a developmental change, between the four and five year old groups, in the form that the change takes. Specifically, the older children are more likely to show an effect parallel to the developmental effect than are the four year olds. The lesser tendency for four year olds to switch from asking about individual objects to asking about groups of objects can be interpreted in the context of the developmental changes observed in initial strategy choices (Table 1). Age related change in patterns of strategy usage were analyzed using the statistical package GLIM (Payne, 1986), and following the 12 procedure described by Fienberg (1980). This form of Chi Square allows the decomposition of effects into main effects and interactions similar to that obtained for continuous data with the use of analysis of variance. As has been reported in other studies, the tendency to eliminate a group of objects with a single question increased linearly with age (chi square = 36.89, df = 1, p<0.001). This was the modal strategy used by the nine year old group. Analyses of the patterns of strategy choice among children asking about individual objects exclude the nine year old group - since only two nine year olds used a strategy of this type. For the remaining children, at the start of the session there was a linear decline with age in the tendency to make identifications by touch (chi square = 3.86, df = 1, p<0.05). The tendency to identify individual objects by name showed a quadratic relationship with age (chi square = 4.06, df = 1, p<0.05) - rising in frequency between the four and five year old group, and thereafter falling. There is no linear effect. Identification of an individual object by an adjective alone showed a strong linear increase with age (chi square = 24.49, df = 1, p<0.001), contributed by the fact that only the oldest group included in the analysis (the seven year olds) ever used it. There was no significant linear relationship between age and the tendency to give fuller descriptions of individual objects, but there is a quadratic effect (chi square = 4.21, df = 1, p<0.05). 20% of five year olds asking about individual objects began the task this way, in contrast to only 2.56% of four year olds, and 8% of seven year olds. These developmental changes in strategy can be largely characterized as reflecting an age related increase in the child's success in handling the variance present in a set of items (cf. Denney, Denney, & Ziobrowski, 1973; Inhelder & Piaget, 1964; Kofsky, 1966; Thornton, 1982). The most sophisticated use of the variance in the set is made in the strategy of 13 grouping items together, and eliminating a whole group with each question. Strategies for describing individual objects, either fully or partially, make less sophisticated use of the variance than this - but make more explicit reference to it than is the case for the strategy of identifying individual objects by touch alone. That the "touching" strategy is associated with poorer use of the variance than the "naming" strategy (for example) is confirmed by an analysis of successive item choices made by children using each strategy: successive item choices can be scored 1 where item N+1 is from the same category as item N, and 0 where it is not. Scoring the first five choices from each game provides a score between 0 and 12. Chance rates for these scores were approximated by 10,000 Monte Carlo runs of artificial, computer generated subjects (this procedure corrects for the non-independence of successive choices). Scores on this measure made by children using the "naming" strategy differed from chance, both for the four year old group (p<0.02) and for the five year old group (p<0.014), implying that these children were implicitly grouping the items across successive questions. By contrast, scores on this measure did not differ from chance, at either four or five years, for children using the "touching" strategy, implying that these children did not make implicit grouping across successive questions. (Seven year olds were not included in this analysis as few used either "touching" or "naming"). Thus age related patterns of initial strategy choice reflect differing levels of usage of the variance in the set of items to be considered - with four year old children showing very little use of variance. In this context, one may infer that factors associated with skill in handling variance limited the four year olds' capacity to show the transition from asking about individual objects to asking about a group of objects, relative to that of the older children. The general thesis that age related change in the capacity for handling 14 the variance present in the set of objects constrains the direction or range of change in children's procedures can be further explored by detailed analysis of the overall direction of all types of change observed here. This analysis supports the view that the four year olds are less successful than their elders in moving to strategies using the variance in the set more effectively. Table 3 summarizes the patterns of initial and final change observed for children starting out either naming individual objects (four and five year olds) or using adjectives alone (seven year olds). All of the children change strategy. Few (only 13%) reverted to their original strategy by the end of the game. 50% of four year olds had switched to strategies making more explicit use of the variance by the end of the session (i.e., describing more fully or asking about groups of objects), in contrast to 80% of five year olds and 80% of seven year olds. Combining the two older groups who initially named individual objects or used adjectives alone, there is a significant tendency for both initial changes in strategy (p = 0.006, Binomial) and final strategy choice (p = 0.018, Binomial) to be toward more explicit use of the variance. Neither effect reaches significance for the four year olds. ------------------------- Insert Table 3 about here ------------------------- Unlike the older groups, a fair number (8, i.e., 33%) of four year olds initially touching individual objects also changed their strategies during the session. Inevitably, any change away from touching must involve a greater use of the variance, and so initial changes are of relatively little interest in this context. Seven of the eight children initially switched from touch to naming; of these, three subsequently tried describing items more fully, and one tried - transiently - asking about a group of objects. However, seven of the eight children reverted to touching individual objects 15 by the end of the session. This tendency to revert to the original strategy is significantly greater for these children than for those starting out naming individual objects or using adjectives alone (Chi square = 17.5 , df = 1, p<0.001, N = 39). The remaining child who had begun by touching the items ended the session describing items more fully. These data again confirm the general view that the four year olds have some difficulty in moving to the strategies involving better use of variance. _C_o_n_c_l_u_s_i_o_n The results obtained in this experiment confirm the prediction that some strategies are more highly associated with the incidence of change than others. The effect occurs for each age group examined. The principle contribution of these strategies seems to be in triggering change: the direction which change subsequently takes is a function of other factors, - specifically here the child's capacity to handle the variance present in the set of objects to be considered. _E_x_p_e_r_i_m_e_n_t _2 The aim of this second study is to analyze the origins of the "strategy" effect obtained above, specifically exploring the contributions of semantic or syntactic factors. In analyzing the genesis of any effect, it is as well to begin from a consideration of the contribution made by the direct demands of the task. The Twenty Questions task as used here was designed to be very open ended, (as was described above). There is no task demand to be quicker, or more efficient, or more sophisticated. (Only a minority of the observed phenomena of change would meet these contingencies, in any case). Thus the observed effects are not the result of contingencies created by the task per se. The two strategies associated with change successfully satisfy the task demands as much as do the strategies not associated with change. One possible cavil to this last conclusion is immediately apparent: 16 from the point of view of an objective observer, the strategies highly associated with change include a verbal label which partially - but only partially - identifies the object the child intends to select. For example, saying "car" in the "naming" strategy provides a label which narrows the field of reference, but does not uniquely identify a single card, since there are several cars in the set. There is, therefore, a certain ambiguity about this label. Labels used in strategies which were not associated with change were not ambiguous in this way: the "full description" strategy provides a unique verbal identification of each item ("The red car", for example). The "touching" strategy does not offer any verbal identification of the object at all, but only a neutral reference in support of touch ("That one?"). One might hypothesize, therefore, that the ambiguity of reference inherent in the "naming" and "adjective alone" strategies triggered change by creating difficulties in communication, and hence generating a task demand for change. But this is not the case. It should be recalled that children using the "naming" and the "adjective alone" strategies virtually always touched each object as they asked about it. Their questions were in fact, therefore, no more ambiguous than those asked by children using the "touching" strategy. There was no communicative difficulty associated with these strategies at all. Nor did the experimenter call attention to the potential ambiguity of a verbal label: her "script" was to behave as if every question were clear in every respect. A truly ambiguous question "car?" without clarificatory pointing would have been answered straightforwardly (e.g., "no!"), and the child left to interpret that reply as referring to any individual car - or all of the cars - as he or she pleased. The experimenter always accepted the child's interpretation (as revealed by subsequent behaviour) of her responses as correct, without comment. In sum, this task does not present 17 either task demands or social cues suggesting a need for better or less ambiguous communication. Nevertheless, the inclusion of a potentially ambiguous label for identifying an object might still have been the key factor leading the "naming" and the "adjective alone" strategies to trigger change. A fundamental rule in most communicative situations is to avoid ambiguity (Grice, 1975). Rules of this kind are commonly over-extended, becoming ends in themselves, rather than remaining subservient to the direct contingencies of the task in hand (cf. Fodor, 1987; Piaget, 1977). The general rule to avoid _a_c_t_u_a_l ambiguities may have been over-extended here, to include the avoidance of even the _p_o_t_e_n_t_i_a_l ambiguities inherent in the inclusion of ambiguous elements (labels) in a question. The rule "avoid ambiguities" is a semantic rule, deriving from the domain of communication, and predicating a direct manipulation of the semantic content of a strategy in relation to its context. Thus the account offered above of the strategy effects observed here implies that the processes constraining the incidence of change in this task were primarily domain-specific, semantic ones. This hypothesis can be tested by altering the relationship between the materials and the two strategies associated with change. The argument is that "naming" and "adjective alone" strategies triggered change because the child over-extended the rule "avoid ambiguity". These two strategies trigger this rule only because there were several items sharing the labels favoured by each strategy. If the materials could be changed so that there was only one item corresponding to labels of these types, then the "naming" and "adjective alone" strategies would not generate ambiguous labels, would not trigger the rule, and would not be highly associated with the incidence of change. The aim of this second experiment is to test this hypothesis. The 18 specific design of the new set of materials to test the hypothesis will be different, in relation to the "naming" strategy and the "adjective alone" strategy. For simplicity, the study below explores the issue in relation to only one of these strategies: the "naming" strategy. In this second experiment, the set of items the child was asked to consider fell into superordinate categories (e.g., "vehicles"), and included only one item of each of the basic category labels (e.g., "car") favoured by children using the "naming" strategy. Subjects were five year old children, i.e., the age group most often using this strategy in Experiment 1. If the association between the "naming" strategy and within-session change is indeed driven by the over-extension of a rule specifying the avoidance of ambiguity, then a comparison of results from Experiments 1 and 2 should show that: (a) this association is much stronger in Experiment 1 than Experiment 2; (b) the incidence of within session change is much higher among children using the "naming" strategy in Experiment 1 than for those using it in Experiment 2. In comparing the results from these two experiments, one would also expect to see other differences, following on from factors identified by Rosch (Rosch, Mervis & Boyes-Bream, 1976; Rosch, 1978). Rosch has shown that young children will use classes at the level of basic categories much more readily than classes at the level of superordinate categories, in a range of situations. On this basis, one would expect fewer children in Experiment 2 to begin the session asking about groups of objects, and fewer to switch to this strategy during the experimental session than was observed in Experiment 1. The factors identified by Rosch are independent of those put forward here to explain the incidence of within-session change. However, Rosch's work must be taken into account in testing the predictions generated by the "ambiguity" hypothesis: the first prediction from the "ambiguity" hypothesis 19 (that the degree of association between the "naming" strategy and change will be different between Experiment 1 and 2) cannot be predicted on the basis of the factors identified by Rosch. But the second "ambiguity" hypothesis (that rates of within-session change will be greater in Experiment 1 than in Experiment 2) also follows to some extent from Rosch's work. This difficulty in generating clear predictions from the "ambiguity" hypothesis can, however, be easily overcome. Rosch's work would predict only that the rate of change from asking about individual objects to asking about groups will be lower in Experiment 2 than in Experiment 1. Unlike the "ambiguity" hypothesis, Rosch's work makes no predictions as to the relative rates of change between one strategy for asking about individual objects and another. Confounding effects from the factors identified by Rosch can therefore be avoided by comparing specifically the rates of change between one strategy for identifying individual objects and another, in testing the "ambiguity" hypothesis. Rosch's work on children's category use identifies a further area which might be explored in comparing the results of Experiments 1 and 2. >From that work, one would expect five year old children asking about individual objects to show a strong preference for using the "naming" strategy rather than other strategies for identifying individual objects - as was observed in Experiment 1. Will this bias be still stronger in Experiment 2? If so, then one may conclude that the same factors influencing children to abandon naming within-session were also influential in determining initial strategy choice. In the analyzes below, the key comparisons in testing the "ambiguity" hypothesis involve only those children who begin the task asking about individual objects. Five year olds in Experiment 1 who began by asking about a group of objects do not contribute to these analyzes. These 20 children were run (as a separate group) in Experiment 2, in addition to the sample of new, naive subjects. The inclusion of this "experienced" sample in Experiment 2 adds power to the test of the hypothesis of interest here: if these children behave as predicted with the superordinate materials (i.e., identifying individual objects by name, showing low rates of within-session change overall, and no association between "naming" and change) then one may be confident that this reflects factors governing strategy choices, and not simply an inability to use, or to think of, alternative strategies. _M_e_t_h_o_d _S_u_b_j_e_c_t_s. 37 five year olds (mean age 5 years 8 months). 19 children (the "naive" group) were new to the task. 18 children (the "experienced" group) had all done the task before, with the materials used in Experiment 1, and had all asked about groups of objects from the outset. 14 of these were subjects from Experiment 1, and 4 were children from a pilot study, added to bring up the N in this group. _M_a_t_e_r_i_a_l_s. 16 objects, each drawn on a separate card as in Experiment 1. The objects fall into four superordinate categories (transports, furniture, fruit, clothing). The drawings were coloured: 4 were yellow (boat, chair, banana, hat). 4 were red (trousers, television set, train, cherries). 4 were green (bed, plane, shirt, apple). 3 were blue (car, table, glove). The orange was orange. _P_r_o_c_e_d_u_r_e. The procedure and instructions were as in Experiment 1. Children in the "experienced" group were tested shortly after completing the task used in experiment one. Results obtained for the two new groups of children were compared to those obtained for five year olds in Experiment 1. 21 _R_e_s_u_l_t_s _a_n_d _d_i_s_c_u_s_s_i_o_n The predictions derived from the hypothesis that "naming" triggered change in Experiment 1 through the overextension of a rule to avoid ambiguities were supported in the results from this experiment. The key comparisons in testing the "ambiguity" hypothesis involved only those children who had begun the session asking about individual objects. 20 five year olds from Experiment 1 (i.e., 59%) took part in this analysis, as did 34 five year olds (i.e., 92%) from Experiment 2. (Conflating "naive" and "experienced" children asking about each object individually in Experiment 2 in this sample does not confound within and between subject variance, in comparing this group with subjects from Experiment 1: all "experienced" subjects had asked about groups in Experiment 1, and so were excluded from the Experiment 1 sample used here). As these figures imply, (and confirming expectations derived from Rosch's work) the tendency to ask about groups of objects was significantly greater in Experiment 1 than in Experiment 2. Whereas 41% of the five year olds in Experiment 1 initially asked about groups of objects, none of the naive five year olds in Experiment 2 did so (chi square = 10.6, df = 1, N = 53 p<0.01). More dramatically, of the 18 "experienced" subjects, all of whom had begun by asking about groups of objects in Experiment 1, only 3 did so in Experiment 2. The difference between the two conditions is significant (Mcnemar test, chi square = 17, df = 1, p<0.001). This last result confirms Rosch's thesis that children's use of categories in problem-solving reflects the structure of the categories into which materials fall: the effect observed here cannot be explained as a general inability to use groups, or a general metacognitive failure to realize the utility/possibility of doing so: 15 of 18 children failing to use groups here had done so, with different material, shortly beforehand. The prediction that the incidence of change will be more highly 22 associated with the naming strategy in Experiment 1 than in Experiment 2 was supported. Whereas in Experiment 1, 10 of the 11 children showing within- session change (91%) originally used the naming strategy, none of those showing within-session change in Experiment 2 used this strategy. The difference is significant (fisher exact, p = 0.025). Thus the association between the naming strategy and a high incidence of within session change in Experiment 1 was a function of the fit between that strategy and the materials. The prediction that there would be a lower overall rate of within- session change between one strategy for identifying individual objects and another in Experiment 2 as compared with Experiment 1 was also supported. The analysis omits all children who subsequently switched to asking about groups of objects (four in Experiment 1, none in Experiment 2). Of the residual children, seven of the 16 from Experiment 1 (i.e., 44%) switched between one strategy for identifying single objects and another, whereas only three of the 34 children from Experiment 2 (i.e., 9%) did so. The difference is significant (Chi square = 8.3, df = 1, N = 50, p<0.01). Among children asking about individual objects, 80% of those in Experiment 2 began by naming items, as opposed to only 50% who did so in Experiment 1. The difference is significant (chi square = 6.31, df = 1, N = 54, p<0.02). _C_o_n_c_l_u_s_i_o_n The results of this second experiment fully support the conclusion that the "naming" strategy triggered within-session change in Experiment 1 through the overextension of a rule specifying the avoidance of ambiguities. With materials which are uniquely described by the labels favoured in this strategy, the association between naming and change disappears, and the overall incidence of change declines. Initial strategy choice is also constrained by this factor: the tendency to start the session using the 23 naming strategy is significantly greater where that strategy provides unique reference than where it provides potentially ambiguous reference. _G_e_n_e_r_a_l _D_i_s_c_u_s_s_i_o_n The studies reported here support the thesis that within-session change is more highly associated with some strategies than with others. This relationship is in fact predicted by virtually all existing models of mechanisms of cognitive change - but it is seldom explicitly considered, or taken into account in theoretical models. It is worth pointing out that differential strategy effects can easily be overlooked, unless explicitly sought out: e.g., Thornton's (1982) analysis assumed (but did not test for) equal association between strategies and change in the Twenty Questions task, and drew theoretical inferences accordingly. The results obtained here demonstrate the utility of explicit analysis of strategy effects in constraining models of change. More than this, the demonstration of so strong a strategy effect here must call into question the interpretation of any micro-analytic study of the genesis of within-session change which does not explore, and take account of possible strategy effects. In the within-session phenomena observed here, strategy effects interact with other factors in determining change in children's problem- solving. For instance, the same strategy (naming individual objects) is the principle origin of within-session change for both four and five year old children. But the five year old group are more likely to change in the direction of strategies making more explicit use of the variance in the set of items (describing fully, asking about groups) than are the four year olds. In effect, the strategy seems to trigger change, and then other factors constrain the direction which that change takes (here, these other factors are presumably connected with well documented developmental change in children's representations of classifications and variance - cf. Denney 24 et al., 1973; Kofsky, 1966; Thornton, 1982). The result endorses the view that studies of within-session change may tap only a subset of the factors contributing to developmental effects. The data support the conclusion that the within-session effects observed in Experiment 1 were due to the application of a rule specifying the avoidance of even potential ambiguities. This rule is semantic, domain-specific: it derives from the domain of communication, and is moderated by the detailed environment in which it is used. The implication is that semantic, domain-specific processes, rather than syntactic domain general ones played the critical role in the incidence of change in this task. No doubt it would be possible to adduce a syntactic model which would account for the data equally well. No empirical analysis of a mental process can ever be definitive. However, a syntactic account of these data would be post hoc. No presently published mechanisms would predict it. The proposed semantic rule provides a better hypothesis, in that it actually predicted the observed effects in a principled way. At the very least, it must be taken into account as an active factor in these phenomena, until directly discredited. Nevertheless, the conclusion that semantic, domain-specific processes played a major role in determining the incidence of change in this task does not necessarily imply that syntactic, domain general ones are not also important. Rather, what it does is to place constraints on the phenomena to be explained through syntactic models. For example, one might argue here that semantic processes are responsible for triggering, and for setting up the broad goals for change. However, it is possible, and even likely, that syntactic processes contribute to the actual generation and shape of the ensuing change. Of course, the identification of semantic factors as the principle 25 determinants of change in this task does not allow the conclusion that this will be the case in every task. That is an empirical question, requiring the analysis of many different tasks. But it is the nature of detailed research of this kind to progress one task at a time. The occurrence of such an effect here demonstrates that semantic processes can contribute importantly to change, and shows that the identification of such effects can constrain inferences as to the functioning of syntactic elements. It emphasizes the need for semantic processes to be explored more directly in other situations, both to understand them for themselves, and to focus studies of syntactic effects. More detailed studies of the interaction between specific strategies and cognitive changes can provide rich constraints on models of change, and an arena for the exploration of the interactions between semantic, domain-specific and syntactic, domain general ones. 26 References Anderson, J. R., (1983). _T_h_e _a_r_c_h_i_t_e_c_t_u_r_e _o_f _c_o_g_n_i_t_i_o_n. Cambridge, MA: Harvard University Press. Carey, S., (1985). _C_o_n_c_e_p_t_u_a_l _C_h_a_n_g_e _i_n _C_h_i_l_d_h_o_o_d. Cambridge, MA: MIT Press. Denney, J. R., Denney, N. W., & Ziobrowski, M. J. (1973). Alterations in the information processing strategies of young children following observation of adult models. _D_e_v_e_l_o_p_m_e_n_t_a_l _P_s_y_c_h_o_l_o_g_y, _8, 202-208. Fienberg, S. E. (1980). _T_h_e _a_n_a_l_y_s_i_s _o_f _c_r_o_s_s-_c_l_a_s_s_i_f_i_e_d _c_a_t_e_g_o_r_i_c_a_l _d_a_t_a. Cambridge, MA: MIT Press. Fodor, J. A. (1987). _P_s_y_c_h_o_s_e_m_a_n_t_i_c_s: _T_h_e _p_r_o_b_l_e_m _o_f _m_e_a_n_i_n_g _i_n _t_h_e _p_h_i_l_o_s_o_p_h_y _o_f _m_i_n_d. Cambridge, MA: MIT Press. Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), _S_p_e_e_c_h _a_c_t_s. New York: Academic Press. Hunt, E. (1989). Cognitive science: Definition, status and question. _A_n_n_u_a_l _R_e_v_i_e_w _o_f _P_s_y_c_h_o_l_o_g_y, _4_0, 603-629. Inhelder, B., & Piaget, J. (1964). _T_h_e _e_a_r_l_y _g_r_o_w_t_h _o_f _l_o_g_i_c _i_n _t_h_e _c_h_i_l_d. New York: Harper & Row. Inhelder, B., Sinclair, H., & Bovet, M. (1974). _L_e_a_r_n_i_n_g _a_n_d _t_h_e _d_e_v_e_l_o_p_m_e_n_t _o_f _c_o_g_n_i_t_i_o_n. Cambridge, MA: Harvard University Press. Karmiloff-Smith, A. (1979). Micro and macro developmental changes in language acquisition and other representational systems. _C_o_g_n_i_t_i_v_e _S_c_i_e_n_c_e, _3, 81-118. Karmiloff-Smith, A. (1984). Children's problem solving. In M. E. Lamb, A. L. Brown, & B. Rogoff (Eds.), _A_d_v_a_n_c_e_s _i_n _d_e_v_e_l_o_p_m_e_n_t_a_l _p_s_y_c_h_o_l_o_g_y Vol. 3). Hillsdale, NJ: Erlbaum. Karmiloff-Smith, A. (1992). _B_e_y_o_n_d _m_o_d_u_l_a_r_i_t_y. Cambridge, MA: MIT Press. Keil, F. (1984). Mechanisms in cognitive development and the structure of knowledge. In R. Sternberg (Ed.), _M_e_c_h_a_n_i_s_m_s _o_f _c_o_g_n_i_t_i_v_e _d_e_v_e_l_o_p_m_e_n_t. 27 New York: Freeman Keil, F. (1986). On the structure-dependent nature of stages of cognitive development. In I. Levin (Ed.), _S_t_a_g_e _a_n_d _s_t_r_u_c_t_u_r_e: _R_e_o_p_e_n_i_n_g _t_h_e _d_e_b_a_t_e. Klahr, D. (1982). Non-monotone assessment of monotone development in information processing analysis. In S. Strauss & R. Stavey (Eds.), _U- _s_h_a_p_e_d _b_e_h_a_v_i_o_r_a_l _g_r_o_w_t_h. New York: Academic Press. Klahr, D. (1984). Transition processes in quantitative development. In R. Sternberg (Ed.), _M_e_c_h_a_n_i_s_m_s _o_f _c_o_g_n_i_t_i_v_e _d_e_v_e_l_o_p_m_e_n_t. New York: Freeman. Kofsky, E. (1966). A scalogram study of classificatory development. _C_h_i_l_d _D_e_v_e_l_o_p_m_e_n_t, _3_7, 191-204. Mosher, F. A., & Hornsby, J. R. (1966). On asking questions. In J. Bruner et al. (Eds.), _S_t_u_d_i_e_s _i_n _c_o_g_n_i_t_i_v_e _g_r_o_w_t_h. New York: Wiley. Payne, C. D. (Ed.). (1986). _T_h_e _G_L_I_M _s_y_s_t_e_m. Royal Statistical Society. Piaget, J. (1976). In W. Overton & J. Gallagher (Eds.), _K_n_o_w_l_e_d_g_e _a_n_d _d_e_v_e_l_o_p_m_e_n_t: _V_o_l. _1. _A_d_v_a_n_c_e_s _i_n _r_e_s_e_a_r_c_h _a_n_d _t_h_e_o_r_y. New York: Plenum Press. Rosch, E. (1978). Principles of categorization. In E. Rosch & B. Lloyd (Eds.), _C_o_g_n_i_t_i_o_n _a_n_d _c_a_t_e_g_o_r_i_s_a_t_i_o_n. New York: Lawrence Erlbaum Associates. Rosch, E., Mervis, C. B., Gray, W. D., & Boyes-Beam, P. (1976). Basic objects in natural categories. _C_o_g_n_i_t_i_v_e _P_s_y_c_h_o_l_o_g_y, _8, 382-439. Siegler, R. S. & Jenkins, E. (1989). _H_o_w _c_h_i_l_d_r_e_n _d_i_s_c_o_v_e_r _n_e_w _s_t_r_a_t_e_g_i_e_s. Hillsdale, NJ: Erlbaum. Thornton, S. (1982). Challenging "Early Competence": a process-oriented analysis of children's classifying. _C_o_g_n_i_t_i_v_e _S_c_i_e_n_c_e, _6, 77-100. Wales, R. & Thornton, S. (1991). What's new? Psychological issues for modelling creativity. IJCAI satellite conference, Queensland. (and forthcoming). 28 _T_A_B_L_E _1 _N_u_m_b_e_r _o_f _C_h_i_l_d_r_e_n _a_t _E_a_c_h _A_g_e _U_s_i_n_g _E_a_c_h _o_f _t_h_e _5 _S_t_r_a_t_e_g_i_e_s _f_o_r _I_d_e_n_t_i_f_y_i_n_g _O_b_j_e_c_t_s _i_n _t_h_e _2_0 _Q_u_e_s_t_i_o_n_s _G_a_m_e ________________________________________________________________ Four Five Seven Nine Years Years Years Years ________________________________________________________________ Touching 24 6 4 2 Naming 14 10 1 0 Adjective alone 0 0 6 0 Describing fully 1 4 1 0 Eliminating a 10 14 11 25 Group of Objects ________________________________________________________________ 29 _T_A_B_L_E _2 _P_e_r_c_e_n_t_a_g_e_s _o_f _C_h_i_l_d_r_e_n _W_h_o _E_v_e_r _C_h_a_n_g_e_d _S_t_r_a_t_e_g_y, _a_s _a _F_u_n_c_t_i_o_n _o_f _A_g_e _a_n_d _I_n_i_t_i_a_l _S_t_r_a_t_e_g_y __________________________________________________________________ Original Four Five Seven Overall Strategy Years Years Years Mean __________________________________________________________________ Touching 33.3% 16.6% 0% 26% Naming 100% 100% 100% 100% Adjective Alone NA NA 100% 100% Describing Fully 100%* 0% 0% 17% *(N=1 child) __________________________________________________________________ 30 _T_A_B_L_E _3 _S_e_c_o_n_d _a_n_d _F_i_n_a_l _C_h_o_i_c_e _o_f _S_t_r_a_t_e_g_y _M_a_d_e _b_y _C_h_i_l_d_r_e_n _S_t_a_r_t_i_n_g _b_y _I_d_e_n_t_i_f_y_i_n_g _I_n_d_i_v_i_d_u_a_l _O_b_j_e_c_t_s _E_i_t_h_e_r _b_y _N_a_m_e _o_r _b_y _U_s_i_n_g _a_n _A_d_j_e_c_t_i_v_e _A_l_o_n_e - (_P_e_r_c_e_n_t_a_g_e_s _b_y _A_g_e) ____________________________________________________________________ Second Strategy Chosen (i.e., First Change) touching describing eliminating fully a group ____________________________________________________________________ 4 Years 43% 43% 14% (initial strategy naming) 5 Years 30% 50% 20% (initial strategy naming) 7 Years 0% 33% 66% (initial strategy adjective alone) ____________________________________________________________________ ____________________________________________________________________ Final Strategy Used touching describing eliminating reverts to fully a group 1st strategy ____________________________________________________________________ 4 Years 43% 28% 14% 14% (initial strategy naming) 5 Years 20% 50% 30% 0% (initial strategy naming) 7 Years 14% 14% 43% 28% (initial strategy adjective alone) ____________________________________________________________________ 31