Permutation transformations on phrase structures in letter sequences.

Abstract

A paired-associate (PA) learning task in which the stimuli were strings of five consonants followed by a numeral and the responses were reorderings of the same five consonants was presented to college Ss. Each response string was related to its stimulus string by one of four permutation rules. Each rule was associated with one of the stimulus numerals. The four rules together segmented the consonant string into a hierarchical system of units. With standard PA instructions, there was an average of 4.06 trials to criterion in initial learning. When subsequently transferred to a list with five entirely new consonants but the same rules, 5"s performed virtually errorlessly. In responding, Ss revealed a tendency to pause in a pattern appropriate to the hierarchical units imposed by the permutation rules. However, the similarity of these experimentally induced rules with those involved in natural language behavior was rendered doubtful by the long experimental response latencies of nearly 5 sec. per rule.What is learned by a child when he acquires the ability to speak and to comprehend a natural language has been a matter of dispute. The knowledge that a speaker possesses of his language has been viewed, on the one hand, as a complex network of stimulus-response connections (Jenkins & Palermo, 1964;Mowrer, 1960;Osgood, 1953) and, on the other hand, as a system of rules which generate the well-formed utterances in his language (Chomsky, 1957(Chomsky, , 1965. According to the latter view, phrase structure rules and transformational rules are essential features of the syntax of all natural languages. Each of these types of rules are general enough that they can be applied to many different materials in addition to natural language. For example,

Remember that old theory of memory? Well, forget it.

“…With a small set of examples, subjects may remember particulars. With a liberal set of examples, however, they cannot reliably discriminate the instances on which they were trained from other acceptable instances in the system (Foss, 1968;Keeney, 1969;Segal, 1962).…”

Hierarchical organization as a determinant of sequential learning

“…The task for the subjects was to learn a set of permutations of those items that either did or did not define a hierarchical organization for the sequence that was permuted. Keeney (1969) has shown that certain sets of permutations will lead subjects to learn an implicit structuring of the elements that are permuted. The rule for selecting the permutations is that only those permutations that assign elements to unique nodes in the structure may be used.…”

“…For example, consider the sequence ABCD and the hierarchical structure «AB) (CD». According to Keeney (1969), the permutation CDAB defines AB and CD as two subgroupings of ABCD since only elements that are represented by unique nodes may be permuted. The permutation BACD defines A and B and lower order constituents of AB and the permutation ABDC defines C and D as constituents of CD.…”

“…The permutation BACD defines A and B and lower order constituents of AB and the permutation ABDC defines C and D as constituents of CD. To support the structureinducing property of the technique, Keeney (1969) showed that subjects who learned one set of systemdefining permutations and then transferred to a list of new letters that retained the old system produced fewer errors than subjects who transferred to permutations defining a new system. Furthermore, the pausal patterns of subjects learning different response systems were judged to vary in accordance with the particular system learned.…”