Sign detection theory and its applications

摘要

A characterisation of sign-transmission is offered which is more general than conventional signal-transmission theory. Whereas the latter relies on continuous random variables mapping from two sets of objects (the signalling set and the noise set) to portray receiver performance, sign-detection theory does so using pairs of discrete probabilities attached to elementary logical conjuncts of query terms, these conjuncts partitioning the two sets of objects. Further mappings are more general than those that map to the real numbers. The modified formalism recognises a transmitter, an intermediary, and a percipient, and a “metasignal” is defined to explain information-seeking behaviour by the percipient. This approach allows the effect of choosing all possible logical search expressions to be portrayed, for example as a bivariate distribution of these expressions over the Recall-Precision outcome space, as Recall and Precision distributions marginal to the latter, or as a distribution of a metric measuring distance between the “signed set” and the “perceived as signed” set. Sets of “successful”, i.e. more-effective, search expressions are defined by choosing intervals of Precision, Recall, etc. outcome-spaces to which such expressions map. The problem of modelling such distributions differs from that of modelling the random variables of conventional signal-detection theory. Empirical data on instances of medical information communicated through the MEDLINE data base are described for abstractly-defined percipients. A prototype psychological experiment in the area of cognitive perception is also described, in which a percipient is required to identify, using public record attributes, an attribute used privately by a transmitter. This process differs from conventional experiments in perception that describe sensory perception only and which are expressed in the language of signal-detection theory. The diffusion of signed objects through a (changing) data base with time can be simply incorporated in the theory as variation in signal-to-noise ratio, and on this basis a conjecture is offered as to sign-creation by data base producers.