Assessing animal vocal communication using the Hyperspace Analog to Language (HAL) model.

摘要

The Hyperspace Analog to Language (HAL) model is used to measure contextual co-occurrence in human language (Lund & Burgess, 1996). In this dissertation, the HAL model was applied to three non-human animal systems; the vocalizations of an African gray parrot, the songs of humpback whales, and the courtship songs of male mice (from both a wild-type population and a genetic model for Fragile X syndrome). In all cases, HAL found evidence of contextual co-occurrence and therefore higher order structure in the communication systems. In the case of the parrot, HAL showed contextual clusters stemming from common phrases in the repertoire, showing these phrases had been arrived at via individual word learning and substitution (as opposed to memorization of each and every phrase as a specific entity). In the humpback whale songs, HAL identified Classes of units that could be combined into patterns specific to individual regions. Changes in these patterns and the usage of the Classes may be additional support for the idea of cultural or geographic clans in these marine mammals. In the mouse song, HAL analysis found different co-occurrence Classes for the wild type and knock out (Fragile X model) mice, and established that although the Fragile X mice appear to be putting together courtship songs with the correct syntax, they may not be doing this using a global co-occurrence schema. Much of this research is preliminary and required subjective judgments, in addition to the creation of new statistical techniques. The judgments made and statistical methods developed were seen as the most reasonable options, however further experimentation is necessary in the case of all three experiments.