Using a Self-Organizing Map (SOM) and the Hyperspace Analog to Language (HAL) Model to Identify Patterns of Syntax and Structure in the Songs of Humpback Whales

Abstract

Two different fully automated models were used to examine syntax and structure in humpback whale song. Songs were initially classified via a Self-Organizing Map (SOM), and then examined, via the Hyperspace Analog to Language (HAL) model, for evidence of a type of higher level organization -- global co-occurrence -- found in human language. HAL was able to identify particular "classes" of song units which were used interchangeably to form patterns in the song, not unlike the use of noun-verb-direct object in human language, where the noun, verb, or direct object can be any one of many possibilities from that particular class. Further, HAL identified specific patterns unique to the songs and their respective geographical areas. These patterns provide support for the idea that humpback whale songs are unique to specific region and may be transmitted culturally.