A computational approach to modeling population differences

Abstract

Four experiments were conducted to determine whether the Hyperspace Analogue to Language (HAL) model of semantic memory could differentiate between two different populations. An analysis ofthe differences in densities (or average distances between word neighbors in semantic space) in HAL matrices-generated from text corpora derived from younger and older adults-confirmed that HAL was able to distinguish between the two age groups. This difference was again detected when structured interview data were used to build the corpora. Athird experiment, designed to test the specificity of HALin detecting differences between groups, did not detect any difference in the densities of the memory representations when older adults generated both the test corpora. The final experiment, conducted on the language of adults with Alzheimer's and normal adults, again demonstrated that HAL could discriminate between the two populations. These results suggest that HALis capable of modeling, on the basis of changes in mean density, some of the differences between populations without modifying the model itself but, rather, by changing the text corpus from which the model creates its representations in semantic space.The ability to adequately discriminate between populations is one of the foundations of psychological research. Detecting differences between groups allows researchers to isolate those variables that systematically change between the groups, permitting a better understanding of the processes and relationships between variables within the system under study. Even within the specific domain of cognitive psychology, the list of potential group contrasts is endless: younger and older adults, children and adults, adults with schizophrenia and normal adults, patients with Alzheimer's and normal controls, to name a few. Computational models have much to offer such research. First, computational models require the explicit operationalization of variables. Such operationalization itself often reveals confounds and hidden variables previously left unconsidered and allows for explicit testing of the contribution ofeach variable to the global performance of the model. Second, computational models often allow competing theories of differences between groups to be compared directly. Finally, computational approaches allow the modeling of processes that would be difficult to measure or even detect in actual human participants.Computational models to date have primarily been used in specific domains, such as the effects of aging on cognition. For instance, Hannon and Hoyer (1994) demonThis paper received the Castellan Award for best student submission at the 1999 conference ofthe Society for Computers in Psychology. The research was supported by an NSF Presidential Faculty Fellow Award SBR-9453406 to c.B. We thank Guila Glosser and Susan Kemper for providing us with the interview transcripts and subject data reported in this paper. Correspondence concerning this article should be addressed to P. Conley, Psychology Department, University of...

Age-Related Degree and Criteria Differences in Semantic Categorization

“…Evidence in favor of more elaborate semantic networks in older age comes from semantic representations built from word co-occurrences in text corpora elicited from young and older adults. The representations of the latter are characterized by denser word neighborhoods (Conley & Burgess, 2000a, 2000b). In word association studies, the interconnections between words are elicited in a direct manner, by having participants answer with the first words that come to mind in response to a cue word.…”