An attribute frequency model for the abstraction of prototypes

Abstract

An attribute frequency model for the abstraction of prototypes is proposed as an alternative to the prototype-plus-transforrnation model. A specific model is tested in a Franks and Bransford visual pattern paradigm under conditions in which the two models generate contrasting predictions. The results support the attribute frequency model. Application of the model to reported data obtained in other paradigms is illustrated and discussed.Studies of concept formation have traditionally used concepts defmed by a listing of relevant attributes (Bourne, 1966). Recently, Franks and Bransford (1971) have proposed that concepts might more generally be defined in terms of a base pattern, or prototype, plus an allowable set of transformations. In their experiments, the base pattern consisted of a particular arrangement of four discrete, discriminable visual attributes. A set of formal transformation rules was defined, and distortions of the prototype were produced by varying the prototypical pattern in accordance with various combinations of these rules. An acquisition set was assembled with the restriction that each type of distortion defined by the transformation rules was represented and that the prototype remain the "central tendency" of the acquisition set. The central tendency of the prototype is ensured by determining that the total number of individual transformations required to produce representations of all transformations defined by the set of rules is least when the prototype is used as the base pattern.Following exposure to the acquisition set, Ss were exposed to a recognition set and told to decide whether they had seen each pattern in the acquisition set and to rate their confidence on a 5-point scale. The results indicate that the prototype (generally not included in the acquisition set) received the highest positive (recognized) rating, followed by patterns of increasing transformational distance. Patterns that could not be produced by the formal transformation rules tended to receive negative (not recognized) ratings. These results were interpreted as supportive of a schematic memory structure for concepts consisting of a prototype and a set of transformation rules that produce instances of the concept. The nature of the experimental paradigm used by Franks and Bransford would seem to make these results equally (or more) amenable to the predictions of an attribute frequency model. The general model proposed here associates each attribute of the stimulus with a set of frequency counts, one for each discriminable state of each dimension upon which the attribute varies. In addition, when specific cues are present which define relationships among attributes, an additional frequency count is associated with each discriminable relationship between attributes consistent with these cues.Frequency counts are compiled from the acquisition stimuli to be shown Ss. The number of appearances of each value of each dimension upon which the attribute varies is counted. For example, if the attribute triangZe varies on t...

The effects of information order and learning mode on schema abstraction

“…There has been little investigation of factors affecting the acquisition of the original training items. In general, whether a schemaabstraction theory assumes that classification of novel items is based on similarity to stored instances (Medin & Schaffer, 1978), a central tendency or prototype (Franks & Bransford, 1971;Posner & Keele, 1968), or the frequency of feature and feature combinations during learning (Elio & Anderson, 1981;Hayes-Roth & Hayes-Roth, 1977;Neumann, 1974;Reitman & Bower, 1973), a default prediction seems to be that every study item encountered will have an equal impact on the final schema and this impact will be independent of exemplar presentation order.…”

Frequency effects in the abstraction of linguistic ideas

“…The author wishes to thank Cheryl Berryman, Lisa Gemma, and Robert Ciulla for assistance in collecting the data. Requests for reprints should be sent to the author at: Department of Psychology, University of Maryland Baltimore County, Baltimore, Maryland 21228. of nonsense syllables (Katz & Gruenwald, 1974), letter and digit sequences (Reitman & Bower, 1973), and visual patterns Neumann, 1974). Katz (1973) has shown that the effect disappears when recognition instructions are modified and subjects are only required to judge whether recognition sentences have the same meaning as acquisition sentences.…”

“…Katz, Atkeson, and Lee (1974) have shown that sentences consisting of a single idea are rated much higher when the acquisition list consists entirely of single-idea sentences, and they suggest, therefore, that the linear effect is an artifact of the method of presenting sentences of varying complexity during acquisition. These studies, taken together, suggest that the linear "linguistic" effect is a function of decision processes occurring during recognition rather than a function of semantic integration occurring during acquisition.Both Neumann (1974) and Reitman and Bower (1973) have pointed out that the Bransford-Franks paradigm is essentially a concept-formation task in which subjects are tested for recognition of the specific instances of a concept which were presented during acquisition. Although neither of these studies used linguistic stimuli, both demonstrated that a linear recognition function can be accounted for by a recognition model in which the recognition rating for a given item is a function of the frequency of presentation of its elements during acquisition, provided the recognition item contains no new elements.…”

“…Both Neumann (1974) and Reitman and Bower (1973) have pointed out that the Bransford-Franks paradigm is essentially a concept-formation task in which subjects are tested for recognition of the specific instances of a concept which were presented during acquisition. Although neither of these studies used linguistic stimuli, both demonstrated that a linear recognition function can be accounted for by a recognition model in which the recognition rating for a given item is a function of the frequency of presentation of its elements during acquisition, provided the recognition item contains no new elements.…”

Failure of visual prototype learning in the pigeon

“…Posner and Keele (1968), who showed that after learning a set of distorted patterns, human subjects could easily classify the prototype of the distorted patterns, proposed a prototype distance model to explain these interesting results. Other models, such as the attribute-frequency model (Neumann, 1974(Neumann, , 1977 and the category-density model (Fried & Holyoak, 1984), also predict abstraction of the prototype from its distorted derivatives.…”