Human pattern recognition: Evidence for a switching between strategies in analyzing complex stimuli

Abstract

In a preceding multidimensional scaling experiment, with "size" and "brightness" as parameters, subjects were found to use individually different strategies in processing compound stimuli: Most subjects adhered to either the Euclidean or the City-block metric (Ronacher and Bautz, 1985). In the experiment reported here, participants of the previous study were induced - by a manipulative instruction - to modify their strategy. With 5 out of 10 subjects a switching to another strategy occurred, which manifested itself in a drastic shift of the respective best metric (e.g. from Euclidean to City-block or vice versa). The extent and speed of changes as well as - in some instances - the stability of estimation accuracy show that subjects were not forced by the instruction to develop a new strategy. Results rather suggest that adult subjects have easily available two, or perhaps even more, alternative processing modes, the decision for one of them being a matter of a subject's preferences rather than of individually different abilities.

Integrality/separability of stimulus dimensions and multidimensional generalization in pigeons.

“…As Shepard (1991, p. 61) concluded from his review of the human literature: “pure integrality is more or less approximated, if perhaps never strictly achieved. Some directions through the stimulus space […], although not fully separable, may always be somewhat more psychologically salient […] [I]ntegrality and separability may define a continuum rather than a dichotomy.” Furthermore, several other factors, such as individual differences or task demands, could also affect the way in which stimulus dimensions interact (e.g., Foard & Kemler Nelson, 1984; Ronacher, 1984). …”

How do bees learn and recognize visual patterns?

“…3) and in humans (due to previous experience, prejudices or in¯uencing instructions; e.g. Ronacher 1984).…”

“…In a series of multidimensional scaling experiments we obtained data for size and contrast discrimination that are directly comparable with the data of bees (Ronacher 1984;Ronacher and Bautz 1985). Apparently, human subjects are able to apply dierent strategies for assessing two-dimensional dissimilarities: for 13 out of 27 subjects the best ®tting exponent of the Minkowski metric was close to 1 (cityblock metric); for 11 subjects it was close to 2 (Euclidean metric); only 3 observers showed an intermediate exponent between 1.3 and 1.5 (see Fig.…”

“…In this respect, increasing levels of feature extraction could be regarded as a sequential process, either in phylogenetic terms or also in terms of processing stages (cf. the discussion of holistic versus analytic processing in human pattern recognition: Ronacher 1984;Ronacher and Bautz 1985; see also Treisman 1986). An open question, however, is how the bee's nervous system might accomplish the formation of, and the matching with, a template in the free-¯ight condition, in which one often observes a thorough`s canning'' of training and test patterns (see Anderson 1979;Lehrer et al 1985;Heisenberg 1995).…”

“…This interpretation is supported by the following observation: in response to an in¯uencing instruction some subjects``switched'' from an Euclidean metric to a city-block metric and vice versa. [Depending on their best exponent, in a second experiment subjects were asked either``to assess the overall dierence between the discs, as spontaneously as possible'', or``to evaluate size and brightness dierences separately''; for details see Ronacher (1984); see also Ronacher and Bautz (1985); Ronacher and SuÈ ss (1990).] In the multidimensional scaling literature the two metrics, city-block and Euclidean, are taken as indicators for two types of processing.…”