ATLAS (Automated Tone Level Annotation System): A tonologist's and documentarian's toolkit

Abstract

This paper describes a novel computational toolkit for tonal analysis: ATLAS (Automated Tone Level Annotation System). Tone remains a challenge in many language documentation projects, and far too often still, one comes across descriptive and theoretical treatments of tone languages in which tone marking is entirely absent or of questionable accuracy. ATLAS takes as its input a WAV file and TextGrid delimiting tonebearing segments and outputs normalized pitch level annotations intermediate between raw f0 and phonemic categories. These "tone level" annotations represent a discrete numerical version of the dashes often used as a broad phonetic transcription of tone. The number of levels can be set by the researcher, and a number of raw phonetic measures are also outputted by the tool. ATLAS is designed to be used by anyone regardless of experience with tone or computational methods, thus promoting the inclusion of objective, replicable pitch data in documentary, descriptive, or theoretical materials on tone languages. We also show the utility of ATLAS's broad phonetic annotations in understanding the surface realization of already determined phonemic categories and in making hypotheses about unanalyzed tone systems.