Automatic Voice Recognition technologies and their incorporation into indigenous language transcription methods

Abstract

In August 2018 a group of documentary linguists, speakers of endangered languages, endangered language activists, and natural language processing specialists came together at a retreat in Quechee, Vermont. The goal of the retreat was to help resolve the current bottleneck in natural languages transcription.  Researchers at the retreat were particularly interested in ways to utilize new automatic speech recognition technologies, especially artificial neural networks, in the field. Prior work in natural languages transcription did not have access to such technology and these discussions were extremely fruitful.

   In this welcoming environment, the participants from different fields and backgrounds had the opportunity to get to know one another, exchange ideas, knowledge, and experiences. The retreat hinged on discussions about the different resources each group used in the work of linguistic documentation and voice recognition. Both camps shared their latest advances and current conditions of their respective fields, including the linguistic vitality of their respective languages, the size of their corpus, their workflows, among other equally related themes.

   Participants also expressed needs and wants related to implementation of such technologies in the field. In particular, speakers of endangered languages do not traditionally have access to advances natural language processing technology, while those with such technology are often unable to interact with the communities most in need. Bridging this gap was determined to be a key goal of all the groups in attendance.