Collaborative Research: Contributions of Endangered Language Data for Advances in Technology-enhanced Speech Annotation

语言学家们更加努力地从濒危和很少研究的语言中收集真实的言语材料，以发现语言的多样性。然而，将这些演讲转录成书面形式以便于分析的挑战令人望而生畏。这既是因为需要转录的数字收集的语音的绝对数量，也是因为打开语音的声音的难度。SRI国际公司的语言学家安德烈亚斯·卡索尔和计算机科学家维克拉姆吉特·米特拉以及葛底斯堡学院的语言学家乔纳森·D·艾史密斯将联手开发能够大幅减少语言转录瓶颈的软件。该团队将使用来自墨西哥格雷罗州的濒危语言Yoloxochitl Mixtec作为测试案例，开发一种软件工具，该工具将使用之前转录的Yoloxochitl Mixtec语音数据来培训新一代母语人士使用实用的拼音法，并开发自动语音识别软件。识别软件的输出将被用作初步转录，母语人士将在必要时进行更正，以创建额外的高质量训练数据。这种递归方法将创建足够大的转录语音语料库，以便软件能够完成新收集的语音材料的自动转录。该项目将包括对本科生和研究生进行软件开发培训和分析Yoloxochitl Mixtec音响系统。该项目还将以互动的方式将母语人士培训为纪录员，系统地向他们介绍其语言的转录惯例。这一软件工具将有助于在更广泛的发言者群体中建立Yoloxochitl Mixtec语言的识字能力。该项目的成果将在拉丁美洲土著语言档案馆(德克萨斯大学奥斯汀分校)、Kaipuleohone(夏威夷大学数字语言档案馆)和语言数据联合会(宾夕法尼亚大学)查阅。

Linguists have increased efforts to collect authentic speech materials from endangered and little-studied languages to discover linguistic diversity. However, the challenge of transcribing these speech into written form to facilitate analysis is daunting. This is because of both the sheer quantity of digitally collected speech that needs to be transcribed and the difficulty of unpacking the sounds of spoken speech. Linguist Andreas Kathol and computer scientist Vikramjit Mitra of SRI international and linguist Jonathan D. Amith of Gettysburg College will team up to create software that can substantially reduce the language transcription bottleneck. Using as a test case Yoloxochitl Mixtec, an endangered language from the state of Guerrero, Mexico, the team will develop a software tool that will use previously transcribed Yoloxochitl Mixtec speech data to both train a new generation of native speakers in practical orthography and to develop automatic speech recognition software. The output of the recognition software will be used as preliminary transcription that native speakers will correct, as necessary, to create additional high-quality training data. This recursive method will create corpus of transcribed speech large enough so that software will be able to complete automatic transcription of newly collected speech materials. The project will include the training of undergraduate and graduate students in software development and the analysis of the Yoloxochitl Mixtec sound system. The project will also train native speakers as documenters in an interactive fashion that systematically introduces them to the transcription conventions of their language. This software tool will help in establishing literacy in Yoloxochitl Mixtec among a broader base of speakers. The results of this project will be available at the Archive of Indigenous Languages of Latin America (University of Texas, Austin), Kaipuleohone (University of Hawai'i Digital Language Archive), and at the Linguistic Data Consortium (University of Pennsylvania).