Factores de riesgo de la anem ia ferropén ica

repeated the procedure with a chimpanzee subject, Fersen, Wynne, Delius and Staddon (1991) used pigeons as did Higa and Staddon (1993), and Davis (1992) demonstrated transitive inference in rats. All these studies demonstrated, to a greater or lesser degree, good evidence of transitive inference performance with non-human subjects.

Several of these studies also manipulated the transitive inference training paradigm to investigate further the precise conditions under which subjects would

demonstrate transitive inference. This involved removing the end points from the transitive inference series and seeing what effect this had on transitive inference responding.

Gillan (1981)

This manipulation was first carried out by Gillan (1981). Using a chimpanzee subject, Sadie, Gillan trained the overlapping pairs A-B+, B-C+, C-D+, D-E+ (on each trial the subject received reinforcement for selecting the stimulus indicated + , and did not receive reinforcement for selecting the stimulus indicated -)to establish the series A<B<C<D<E (Experiment 1 A). Following this training the subject demonstrated transitive inference, consistently selecting D rather than B on the critical B<D test pair. Having demonstrated transitive inference responding on B<D, Gillan extended the series by training a further pair E-F+, and again tested for transitive inference formation using two new non-adjacent test pairs BE and CE (Experiment IB). Sadie responded correctly on these new test items choosing E over B and E over C.

In both these transitive inference tests the training stimuli formed a linear series. Gillan's next manipulation was to introduce a training pair which disrupted the order of this series by effectively removing the end points. This time Sadie received training with the adjacent pairs, A-B+, B-C+, C-D+, D-E+, E-F+ as had been trained in the previous manipulations, and also with a new pair F-A+ which removed the higli and low end points of the series, effectively making it circular. Sadie then received testing on the non adjacent pairs BD, BE and CE which had been tested in the previous two manipulations (Experiment 2A). When the F-A+ pair was introduced, Sadie was much slower to reach criterion than on any of the previous pairs. The introduction of tliis pair also affected performance on the A-B+ and E-F+ pairs. However Sadie did eventually reach criterion on all six adjacent pairs. When tested, Sadie maintained accurate performance on a high percentage of the directly trained adjacent pairs and this performance was comparable to the

Chapter 3: Cognition in Non-Humans 46

performance shown in the two previous manipulations. However, performance on the non-adjacent test pairs was veiy different to that shown previously and was close to chance level.

To ensure that disruption of performance was not just due to the addition of a sixth training pair, but was in fact due to that training pair violating the established order of the series, Gillan carried out a further manipulation. Gillan now trained the pairs A-B+, B-C+, C-D+, D-E+, E-F+ and A-F+. This meant that once again six adjacent pairs were trained, but where previously F-A+ had violated the order of the series, this time A-F+ preserved the series order (Experiment 2B). Sadie reached criterion on this task which restored the series order more quickly than she had on the previous task which disrupted the order. When tested Sadie again maintained accurate performance on the adjacent pairs. When tested on the non-adjacent pairs BD, BE, CE, she consistently selected D and E over B and C. This non-adjacent test pair performance was significantly different fi*om performance on the previous manipulation which removed the end points of the series but was not significantly different to performance on the two previous manipulations which maintained the series order. This suggests that Sadie's chance level performance on the non- adjacent test pairs in experiment 2A was not caused simply by the addition of a sixth training pair but was specifically related to the way that pair violated the established linear order of the sequence. This seems particularly likely as, when the training pair which violated the linear order was replaced by one which restored that order, Sadie's performance improved to the same standard as she had shown on the four and five item series.

Bryant and Trabasso (1971) had suggested that accurate performance on the adjacent training pairs was necessary in order for the subject to demonstrate transitive inference. Gillan (1981) now suggested that there were in fact two necessary conditions if the subject was going to demonstrate accurate transitive inference performance; firstly, accurate performance on the adjacent training pairs, and secondly, the stimuli from the series must be ordered on some unidimensional scale. As a result of these experiments, Gillan suggested that transitive inference is a fundamental reasoning process in primates, and this was supported by McGonigle and Chalmer's (1977) findings with squirrel monkeys.

One important aspect of Gillan's (1981) experiment is that the pair comparisons were based on relative amounts of food. Bryant and Trabasso (1971) presented their comparisons in terms of relative lengths of rods, and McGonigle and

Chapter 3: Cognition in Non-Humans 47

Chalmers (1977) used relative weights. Both these presentations used some concrete dimension which may have made it easier to order the stimuli. It was possible that Gillan's presentation using relative amounts of food would be more abstract and would be treated as a series of overlapping discrimination problems rather than forming a transitive series. However, the consistent choice of D and E on the non-adjacent pair tests and the fact that Sadie's performance was susceptible to manipulation of the series order, suggests that the adjacent pairs were in fact ordered mto a transitive series. "The results of the present experiments suggest that she had acquired a mental representation of the order of the stimuli in the series" (Gillan, 1981,p.l61).

One interesting feature that Gillan (1981) pointed out was that at the time of his and McGonigle and Chalmers' (1977) transitive inference experiments with non humans, two prominent theories of transitive inference performance with humans proposed that language was central to this type of reasoning (Clark 1969; Sternberg

1980). Neither of these language orientated theories explained the data obtained with non-human subjects, and the data from these subjects suggested that language is not a necessaiy condition for the derivation of transitive inferences. Part of the problem of investigating transitive inference with non-humans had been finding an appropriate way to present the premises. This was made possible by adapting Biyant and Trabasso's (1971) procedure. It has also been difficult to demonstrate stimulus equivalence with non-humans. Possibly this is not because the ability to derive stimulus equivalence is linguistically based but because it proved more difficult than expected to develop ways of presenting the stimulus equivalence task non-linguistically.

Two further studies have repeated Gillan's (1981) procedure of disrupting the established transitive inference sequence. These studies have extended the analysis of transitive inference to non-primate subjects; Fersen, Wynne, Delius and Staddon (1991) - pigeons, Davis (1992) - rats.

Fersen, Wynne. Delius and Staddon (1991)

Fersen, Wynne, Delius and Staddon (1991) suggested that transitive inference testing might be assessing an ability with practical relevance; "the ability to rank objects on a hedonic scale and make judgements about the desirability of items that have never been encountered together must often have adaptive value" (p.334). Fersen et al aimed to show that the ability to form transitive inferences was not limited to primates by demonstrating transitive inferences with pigeon subjects. In

Chapter 3: Cognition in Non-Humans 48

their first experiment they established the series A>B>C>D>E and tested for the emergence of B>D. Four out of six subjects completed training and all four subjects showed correct selection of B over D at significantly above chance level. In experiment 2, the five term series was extended to a seven term series by the addition of two new pairs of stimuli, one at either end of the series, X+A- and E+F- . This gave six possible non-adjacent test pairs that did not include end items, AC, AD, AE, BD, BE and CE, all of which were tested. The same four subjects that completed experiment 1 were used in experiment 2. Performance for each subject on each test pair was consistent with the derivation of transitive inferences.

In experiment 3 this seven term linear series was closed into a loop by training a new pair F+X-. With Gillan's (1981) chimpanzee subject this manipulation had disrupted the transitive inference series and produced random responding on the non-adjacent test pairs. A very similar performance was shown with the pigeon subjects. Following the introduction of the inconsistent item, one subject ceased to respond. Of the tlnee remaining subjects, only two discriminated above chance level on the training pairs and were given the test items. Tests were given on seven non-adjacent pairs, XB, XC, XD, XE, BD, BE and BF. Of the two subjects tested, neither showed stimulus preferences above chance level on any pair. These experiments showed that pigeons are capable of behaving according to transitive inference mles on five and seven term Imear series but not on a circular series, performances very similar to that shown by a chimpanzee subject by Gillan (1981). Davis (1992)

Davis (1992) also investigated transitive inference performances by non-primates in a similar manner, using rats as subjects. Davis established a series A<B<C<D<E and tested for preference for D over B on the non-adjacent BD test pair. Three out of four subjects reached criterion on the training pairs and were tested on BD. Each of these subjects showed highly significant transitive inference performance. A second experiment, training the reversed series A>B>C>D>E and testing for preference for B over D established that this transitive inference performance was not an artifact of how recently the correct premise was reinforced during training. A third experiment then extended the original series by training E<F, which was consistent with the relations already trained, and then made the series circular by training F<A, which was inconsistent with the series established. This procedure replicated that carried out by Gillan (1981). Davis (1992) stated that while it was clear that dismpting the internal transitive relations should affect the way in which

Chapter 3: Cognition in Non-Humans 49

inferences were made on the test pair BD, it was not clear how this disruption would manifest itself. Should it produce random performance on BD, or would it cause the series to reverse, giving a preference for B over D ? Also, would the introduction of this inconsistent relation cause immediate disruption of the internal structure or would the disruption develop gradually ? Davis counter-balanced for the order in which the disruptive training occurred. Three rats who had already completed experiment 1 and had received traming on the series A<B<C<D<E now received training on the relations E<F and F<A. Three experimentally naive rats were first taught E<F and F<A before receiving training on the series

A<B<C<D<E. After training, all subjects were tested on two non-adjacent pairs, BD and BE. All subjects were affected by exposure to information that altered the logical structure of the transitive series. Although there were no significant differences between the two training conditions on the non-adjacent pairs, Davis believed there were some subtle effects according to the order in which premise inconsistencies were introduced. Subjects exposed to the inconsistencies before training seemed to have greater difficulty reaching criterion on the training trials. When tested, naive subjects failed to show a preference for stimulus D and in the subjects who had previously demonstrated transitive inference this preference for D was eliminated or reversed. On the test pair BE, E had formerly been the endpoint of the series and so it was expected that on the consistent series there would be a veiy strong preference for this stimulus. On the inconsistent series this preference for E was eliminated in 2 out of 3 subjects who had previously demonstrated transitive inference. The naive subjects showed a slight (non-significant) preference for E on the inconsistent series but Davis believed that this could be accounted for by reinforcement of E in tlie final stages of training with these subjects.

These results with rats as subjects were very similar to the results obtained with other non-human subjects. Both human and non-human subjects seem to be capable of making transitive inferences, as long as the stimuli form an ordered linear sequence. When the series' linear order is disrupted, transitive inference responding is also disrupted. However, Davis (1992) points out that while there may be evidence for a continuum of mental abilities, it may not be wise to assume that different species approach transitive inference problems in the same manner. As there are considerable neurological differences between humans, chimpanzees, monkeys, rats and pigeons it would hardly be surprising to find considerable cognitive differences as well.

Chapter 3: Cognition in Non-Humans 50

"If different species are truly approaching logical transitivity in the same way , then, to put the case crudely, either rats must be stretching their ability to the breaking point or humans must be barely using theirs" (Davis, 1992, p.348) So while non-humans may be able to make transitive inferences it may not be wise to assume that they do this in the same way or with the same ease as human

subjects. It is also doubtful that they can solve problems of the same complexity as those which human subjects can deal with.

Careful experimental investigation has demonstrated that non-human subjects are capable of solving quite complex cognitive tasks such as transitive inference and probably, stimulus equivalence. However, their behaviour on these tasks may not be the same as that of human subjects, so while Thorndike's view that data from non-human subjects is relevant to tracing "the origin of human faculty", it is important to be cautious when comparing human and non-human performance on similar tasks. A detailed analysis of both human and non-human responding is necessary in order to ascertain if the processes operating are similar.

51

CHAPTER 4

Examples of Human Cognition

Chapter 3 considered the demonstration of cognitive abilities in non-human subjects. This chapter looks at the demonstration of the same abilities in human subjects.

Transitive Inference

To make a transitive inference requires the co-ordination of two pieces of

information to reach a correct inferential conclusion. For instance, if stimuli A and B are related to each other by relation r (ArB), and stimuli B and C are related to each other by the same relation r (BrC), then it can be inferred that A and C must also bear relation r to each other (ArC).

Piaget (1928) frequently used transitive inference tasks as a measure of cognitive development in children. Piaget, with others, proposed that children were unable to make these inferences until they reached the stage of concrete operations at around age seven. Piaget felt that pre-operational children were unable to make transitive inferences because they were dominated by immediate perceptual input and thus unable to reorganise this input to combine the information from two relations. Transitive inferences became possible when the child progressed to a "relativistic"

conception of relations (Breslow 1981) and understood that a term can have two relations simultaneously, e.g. B can be smaller than A (B<A) and larger than C (B>C). This allows the child to co-ordinate the two independent relations B<A, B>C to infer the relation A>C.

B m n t and Trabasso (1971)

Some researchers, however, have challenged Piaget's conclusions that children can not make inferences before the age of seven. Bryant and Trabasso (1971) suggested that much younger children were able to make transitive inferences provided they could remember the comparisons they had to combine. They felt that previous failures may have been failures of memory rather than failures of inferential ability. Bryant and Trabasso controlled for this possibility by training the initial

comparisons veiy thoroughly, and also checking retention of these comparisons during testing. They tested for transitive inference by teaching four overlapping pair comparisons A>B, B>C, C>D, D>E and testing performance on the non- adjacent pair B>D, using a series of wooden rods of differing lengths and colours.