Collaborative Research: RI: Medium: From Acoustic Signal to Morphosyntactic Analysis in One End-to-End Neural System

当今世界上大约有7,000种语言，但这一数字正在急剧下降。即使是目前有成千上万人使用的许多语言，也可能在一代人的时间内消失。对于这些语言的使用者来说，这是文化和语言遗产的悲剧性损失，而这些遗产是他们社会身份的重要支柱。每一种语言也都携带着关于语言作为人类行为现象的不可替代的数据--语言变异的限度以及语言结构和发展的模式。语言学家和语言活动家目前正致力于尽可能多的语言快速和全面的文件。如果不幸的是，语言从使用中消失了，文档可以确保它的数据可以用于未来的文化或科学分析。该项目使用自然语言处理和机器学习的工具部分自动化语言文档的过程。它与类似项目的不同之处在于使用一个集成系统来处理语音和单词的结构，而不是使用两个或多个单独的组件。在母语学者的合作下，研究人员正在将他们的方法应用于四种语言：高地普埃布拉纳瓦特尔语，Yoloxóchitl Mixtec，圣佩德罗AmuzgosAmuzgo和北坡Iñupiaq。拟议的研究将通过引入一个端到端系统，将语音作为输入，并将线间注释作为输出，极大地改变自动形态句法和形态音位分析的前景。研究小组建议建立一个端到端的系统，一个单一的神经网络，使用母语语言学家产生的少量标记数据，可以直接将记录的语音转换为分析的文本，产生四个输出：（1）表面转录，（2）表面形式的形态分割，（3）每个词素的基础或规范形式，以及（4）每个词素的注释或标准化标签。提出的单一端到端神经网络代表了将上述四项任务集成到单一神经网络中的第一次尝试，避免了困扰早期创建管道和减轻最终用户技术复杂性的错误传播问题。研究人员还提出了将语言知识融入神经网络的创新方法，包括使用可微分加权有限状态传感器，这些传感器由迭代自训练架构独立激励。这种迭代自我训练的方法本身将代表机器学习的一个进步--一种增加单词和词素权重的新算法。这项研究也对计算形态学做出了重要贡献，它包括对现有的分割和注释方案进行简单但富有表现力的修改，特别是对不连续语素的表示。此外，该建议将流行的方法扩展到形态分析（例如，UniMorph）通过系统地解决派生以及变形。这个奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估来支持。

There are approximately 7,000 languages in the world today, but this number is declining precipitously.Even many languages that currently have thousands upon thousands of speakers are likely to fall outof use within a generation. For the speakers of these languages, this represents a tragic loss of culturaland linguistic heritage, which are important anchors of their social identity. Each language also carriesirreplaceable data about language as a phenomenon of human behavior—the limits of its variation andthe patterns in its structure and development. Linguists and language activists are currently working toquickly and comprehensively document as many languages as possible. In the unfortunate event that alanguage fades from use, documentation ensures that its data will remain available for future cultural orscientific analysis. This project partially automates the process of language documentation using toolsfrom Natural Language Processing and Machine Learning. It differs from similar projects in using oneintegrated system to process the sounds of speech and the structure of words, instead of using two ormore separate components. With the collaboration of native speaker scholars, the researchers are applyingtheir methodology to four languages: Highland Puebla Nahuatl, Yoloxóchitl Mixtec, San Pedro AmuzgosAmuzgo, and North Slope Iñupiaq.The proposed research will dramatically transform the landscape of automatic morphosyntactic andmorphophonological analysis by introducing an end-to-end system that consumes speech as an input andproduces interlinear annotations as an output. The research team proposes to build an end-to-end system,a single neural net that, with small amounts of labeled data produced by native speaker linguists, candirectly convert recorded speech to analyzed text, producing four outputs: (1) surface transcription, (2)morphological segmentation of surface forms, (3) an underlying or canonical form for each morpheme,and (4) a gloss or standardized label for each morpheme. The proposed single end-to-end neural networkrepresents the first attempt to integrate the four aforementioned tasks into a single neural network, avoidingthe error-propagation problems that have plagued earlier attempts at creating a pipeline and mitigating thecomplexity of the technology for end-users. The researchers also propose innovative ways to incorporate linguisticknowledge into neural networks, including the use of differentiable weighted finite-state transducers,which are independently motivated by an iterative self-training architecture. This approach to iterative self training,in its own right, will represent an advance in machine learning — a new algorithm for upweightingwords and morphemes. The research also makes significant contributions to computational morphology.It includes a simple but expressive modification to existing schemes for segmentation and glossing, specificallyfor the representation of discontinuous morphemes. Furthermore, the proposal extends popularapproaches to morphological analysis (e.g., UniMorph) by systematically addressing derivation as well asinflection. This proposal addresses glossing of reduplication and noun-incorporation, which earlier workhas not.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.