An AI approach to automate transcription alignment for first nations languages

点击翻译按钮获取中文摘要

A major task in linguistics involves the transcription of audio language recordings for posterity and analysis. This process constitutes a significant amount of manual labour on the part of linguists. While this transcription step has been the subject of research for the major language families, e.g., English or Mandarin, there has not been any significant research effort for First Nations languages. In this collaborative project with the industry partner ISM Canada, an artificial intelligence (AI) approach is proposed to automate the alignment of transcription for the First Nations languages. The initial focus will be on the Upper Tannana Dene Language, spoken in Alaska and the Yukon Territory, with potential future extensions to other indigenous languages. Using a transcribed and time-aligned corpus as a training data set, a novel deep learning method is envisioned to facilitate the automation of the transcription process, matching the linguistic alignment to the audio recordings. This project will leverage audio signal processing and machine learning (ML) methods to design and test a novel AI system to automate the manual work in transcription endeavours by linguists.The outcome of this project would facilitate ISM in maintaining its competitive edge by fostering novel ML technologies, offering applicability of AI/ML services to new problem domains and cementing ISM's brand as an industry leader in Cognitive Analytics. Furthermore, ISM is a supporter of the Tech-For-Good and Data-For-Good initiatives. To this end, this collaborative project serves as a clear demonstration of ISM's commitment to contribute to the community and to protect Aboriginal culture. The technology to be developed in this proposed project has the potential to both benefit the native communities, as well as give linguists innovative automated tools in studying indigenous languages, thus supporting their important work in preserving Canadian cultural heritage.