Relation matching-to-sample

The relational matching task is the most crucial one in testing animal knowledge about relationships between stimuli because it requires the most abstract reasoning abilities. If a non-human animal were able to learn this so-called relational matching-to-sample task, then this learning would constitute what some theorists consider to be the strongest evidence of abstract conceptual behaviour (Thompson, 1995). Until recently, the relational matching task has only been mastered by children of 5 years of age, or by chimpanzees first trained to use symbols to communicate (Thompson et al. 1997;).

In experiments of Fagot et al. (2001) baboons (Papio papio) successfully learned relational matching-to-sample: They picked the choice display that involved the same relation among 16 pictures (same or different) as the sample display, although the sample display shared no pictures with the choice displays. The baboons generalised relational matching behaviour to sample displays created from novel pictures.

Let us consider this experimental procedure as an illustrative example. The baboons were studied inside an experimental enclosure facing an analogue joystick, a metal touch pad, and a colour monitor that was driven by a computer. On the front of the enclosure were a view port, a hand port, and a food dispenser that delivered food pellets into the enclosure in accordance with the prevailing contingencies of reinforcement. As stimuli, 72 computer icons were chosen and sorted into 3 sets. Two types of stimulus arrays were created from each of these icon sets: “same” arrays and “different” arrays. The stimulus set that was used to create the sample stimuli differed from the stimulus set that was used to create the choice stimuli. This difference meant that correct choice responding could not be based on the identity of the individual icons in the sample and the choice arrays; such identity was impossible. Correct responding could only be based on the relation of the icons in the sample and the choice arrays.

The baboons were individually placed into the test apparatus. The experiment involved multiple sessions per day of many trials that involved a two-alternative forced-choice matching-to-sample procedure. By manipulating a joystick, the baboons were required to place the cursor in such a way as to initiate the presentation. Then a same or a different sample array appeared for 500 ms in either the left or the right half of the monitor. Immediately after sample stimulus offset, two choice arrays appeared on the vertical axis of the screen. One of these two choice stimuli was a same array; it involved a single icon that was repeated 16 times. The other choice stimulus was a different array; it involved 16 different icons. The location of the same and different choice arrays, either at the top or bottom of the screen, randomly varied across trials.

The choice stimuli remained on the screen for a maximum of 10 s, during which the baboons could make a single choice response by moving the cursor into contact with one or the other choice array. Failures to make a choice response in the allotted time were rare during training and never occurred during testing. Correct responses were followed by a high tone and one food pellet; incorrect responses were followed by a low tone and a short time-out. After food delivery or time-out, the next trial could be initiated by contact of the touch pad.

The acquisition of relational matching-to-sample proceeded gradually in baboons. The results were clear: animals successfully acquired relational matching-to-sample to levels in excess of 80% correct. The task was not easy for them; it took one animal 4 992 trials to respond consistently in excess of 80% correct, and it took second one 7 104 trials to do so. Earlier, the animals had learned to report 16-icon same versus different displays by making one choice response (e.g., " up") for same displays and a different choice response (e.g., " down") for different displays (Wasserman et al., 2001); it took only 600 and 700 trials to learn for them. Relational matching-to-sample is obviously much more difficult for baboons to acquire than is same-different discrimination learning.

The experimenters were also interested in whether the baboons could generalise their relational matching-to-sample behaviour to novel sample displays. Here, too, the results were clear. On the very first session of testing, the 2 baboons discriminated the displays of novel sample items at a mean level of 70% correct, whereas they discriminated the displays of familiar sample items at a mean level of 83% correct. Two subsequent sessions of testing yielded very similar levels of discriminative performance. Such robust generalisation to the novel testing displays supports the hypothesis that these monkeys had indeed learned an abstract and general concept.

Recent studies have demonstrated that baboons surpass three-year old children in their ability to solve the relational matching task Bovet et al., 2005). Whereas the baboons were able, after considerable training, to master this task, the 3-year-old children could not solve the relational matching task without any verbal guidance; they were only able to perform this task when the rationale of all of the steps was explained to them. Besides the children’s linguistic capabilities, this disparity can be explained by analogical reasoning abilities that seem to be more developed in adult baboons than in young children.

CONCLUDING COMMENTS

Both animals and researchers have achieved remarkable results in displaying cognitive skills from one end and measuring them from the other end. Basing mainly on variants of discrimination tests, students of animal cognition have revealed abilities for categorisation, abstraction, mental representation, serial learning, transitive inference, and relational matching in non-human animals such as apes, monkeys, horses, pigeons, crows, rats, and some others.

Some uncertainty still surrounds the interpretations of extraordinary capacities for categorisation, abstraction, and imagery in pigeons and bees. It is still an open question of whether these animals possess advanced cognitive abilities or they rather exploit much more simple discrimination basing on very subtle perceptual variation in properties of natural objects. However it is indicative that highest forms of categorisation are currently intensively investigated in many species. We should further examine limitations in animals thought processes in parallel with high levels of their intelligence. As we will see in Part VII, members of different species are highly selective in their responding for stimuli and in forming associations. Members of some species display very fast and advanced learning within specific domains, closely connected with their ecological traits and evolutionary history. This enables us to claim that at least some species possess high cognitive skills. The most advanced of them will be considered in Part VI.