Evaluation Phonemic Transcription of Low-Resource Tonal Languages for Language Documentation

Evaluating Phonemic Transcription of Low-Resource Tonal Languages for Language Documentation Oliver Adams,1 Trevor Cohn,1 Graham Neubig,2 Hilaria Cruz,3 Steven Bird,4,5 Alexis Michaud6 1Computing and Information Systems, The University of Melbourne, Australia 2Language Technologies Institute, Carnegie Mellon University, USA 3Linguistics Program, Dartmouth College, USA 4Northern Institute, Charles Darwin University, Australia 5International Computer Science Institute, University of California Berkeley, USA 6CNRS-LACITO, National Center for Scientific Research, France oliver.adams@gmail.com, trevor.cohn@unimelb.edu.au, gneubig@cs.cmu.edu, steven.bird@cdu.edu.au, hilaria.cruz@dartmouth.edu, alexis.michaud@cnrs.fr Abstract Transcribing speech is an important part of language documentation, yet speech recognition technology has not been widely harnessed to aid linguists. We explore the use of a neural network architecture with the connectionist temporal classification loss function for phonemic and tonal transcription in a language documentation setting. In this framework, we explore jointly modelling phonemes and tones versus modelling them separately, and assess the importance of pitch information versus phonemic context for tonal prediction. Experiments on two tonal languages, Yongning Na and Eastern Chatino, show the changes in recognition performance as training data is scaled from 10 minutes up to 50 minutes for Chatino, and up to 224 minutes for Na. We discuss the findings from incorporating this technology into the linguistic workflow for documenting Yongning Na, which show the method’s promise in improving efficiency, minimizing typographical errors, and maintaining the transcription’s faithfulness to the acoustic signal, while highlighting phonetic and phonemic facts for linguistic consideration. Keywords: low-resource languages; Asian languages; Mesoamerican languages; speech recognition; language documentation. 1. Introduction Language documentation involves recording the speech of native speakers. Transcribing these recordings, which are rich cultural and linguistic resources, is an integral part of the language documentation process. However, transcrip- tion is slow: it often takes a linguist between 30 minutes to 2 hours to transcribe and translate a minute of speech, depend- ing on the transcriber’s familiarity with the language and the difficulty of the content. This is a bottleneck in the docu- mentary linguistics workflow: linguists accumulate consid- erable amounts of speech, but do not transcribe and translate it all, and there is a risk that untranscribed recordings could end up as “data graveyards” (Himmelmann, 2006, 4,12-13). There is clearly a need for “devising better ways for lin- guists to do their work” (Thieberger, 2016, 92). There has been work on low-resource speech recognition (Besacier et al., 2014), with approaches using cross-lingual information for better acoustic modelling (Burget et al., 2010; Vu et al., 2014; Xu et al., 2016; Müller et al., 2017) and language modelling (Xu and Fung, 2013). However, speech recognition technology has largely been ineffective for endangered languages since architectures based on hid- den Markov models (HMMs), which generate orthographic transcriptions, require a large pronunciation lexicon and a language model trained on text. These speech recognition systems are usually trained on a variety of speakers and hun- dreds of hours of data (Hinton et al., 2012, 92), with the goal of generalisation to new speakers. Since large amounts of text are used for language model training, such systems of- ten do not incorporate pitch …

Content truncated to 500 words