Towards Building an Automatic Transcription System for Language Documentation: Experiences from Muyu

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2893–2900 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2893 Towards Building an Automatic Transcription System for Language Documentation: Experiences from Muyu Alexander Zahrer, Andrej ˇZgank, Barbara Schuppler Institute for Linguistics, University of M¨unster, Germany Laboratory for Digital Signal Processing, University of Maribor, Slovenia Signal Processing and Speech Communication Laboratory, Graz University of Technology, Austria a.zahrer@uni-muenster.de, andrej.zgank@um.si, b.schuppler@tugraz.at Abstract Since at least half of the world’s 6000 plus languages will vanish during the 21st century, language documentation has become a rapidly growing field in linguistics. A fundamental challenge for language documentation is the ”transcription bottleneck”. Speech technology may deliver the decisive breakthrough for overcoming the transcription bottleneck. This paper presents first experiments from the development of ASR4LD, a new automatic speech recognition (ASR) based tool for language documentation (LD). The experiments are based on recordings from an ongoing documentation project for the endangered Muyu language in New Guinea. We compare phoneme recognition experiments with American English, Austrian German and Slovenian as source language and Muyu as target language. The Slovenian acoustic models achieve the by far best performance (43.71% PER) in comparison to 57.14% PER with American English, and 89.49% PER with Austrian German. Whereas part of the errors can be explained by phonetic variation, the recording mismatch poses a major problem. On the long term, ASR4LD will not only be an integral part of the ongoing documentation project of Muyu, but will be further developed in order to facilitate also the language documentation process of other language groups. Keywords: Muyu, language documentation, phone recognition, American English, Austrian German, Slovenian 1. Introduction Language documentation is a rapidly growing field in lin- guistics due to its urgency. At least half of the world’s 6000 plus languages will vanish during the 21st century. Along with the speech communities themselves, linguists are as- siduously working to document these languages through recordings, which they translate to global languages, for example English. A fundamental challenge for language documentation is the ”transcription bottleneck”. Transcrib- ing spoken languages is a rather mechanic but very time- consuming task, with a recording-to-transcription-time- ratio of up to 1:60. Speech technology, as a fast developing field in itself, may deliver the decisive breakthrough for overcoming the tran- scription bottleneck. The existing methods, however, are not directly applicable to language documentation, as good performing automatic speech recognition (ASR) systems are based on machine learning techniques requiring large- scale corpora. For the development of a transcription sys- tem suitable for language documentation, two main re- quirements need to be considered: 1) The tool needs to be able to deal with little or no data from the target language, and 2) it needs to integrate well into the work-cycles of a typical language documentation process. This leads to a constantly growing body of primary data and opportunity to increase the algorithm’s performance in course of the language documentation process. A working transcription system can also serve …

Content truncated to 500 words