Universal Dependencies for Ainu

Universal Dependencies for Ainu Hajime Senuma†‡, Akiko Aizawa‡† †University of Tokyo, ‡National Institute of Informatics {senuma, aizawa}@nii.ac.jp Abstract This paper reports an on-going effort to create a dependency tree bank for the Ainu language in the scheme of Universal Dependencies (UD). The task is crucial both language-internally (language revitalization) and language-externally (providing sources for new features and insights to UD). Since the language shows many of the representative phenomena of a type of languages called polysynthetic languages, an annotation schema to Ainu can be used as a basis to extend the current specification of UD. Our language resource comprises an annotation guideline, dependency bank based on UD, and a mini-lexicon. Although the size of the dependency bank will be small and contain only around 10,000 word tokens, it can serve as a base annotation for the next step. Our mini-lexicon is encoded under the W3C OntoLex specification with UD and UniMorph (UM) features with the system-friendly JSON-LD format and thus bearable to future extensions. We also provide a brief description of dependency relations and local features used in the bank such as pronominal cross-indexing and alienability. 1. Introduction The project of Universal Dependencies (UD) (Nivre et al., 2016) marks a milestone in the history of natural lan- guage processing (NLP), as it unifies syntactic annota- tion schemes across languages/corpora and enables cross- lingual processing. At the CoNLL Shared Task 2017 for UD, 33 teams participated in the task, proving that the com- munity is quickly thriving (Zeman et al., 2017). We are currently working on the UD annotation scheme and corpus of Ainu, a language spoken by the Ainu people, an ethnic minority in Japan. A part of the work was presented elsewhere (Senuma and Aizawa, 2017). The task aims to tackle two purposes: 1. to promote the language revitalization of Ainu, as it is a highly endangered language but has a rich amount of the records of oral literature, and 2. to serve as a basis to test if universal specifications in NLP such as UD and UniMorph (UM) (Sylak- Glassman, 2016) can encode the world’s languages. The first issue is urgent because Ethnologue classifies the languages as nearly extinct (Lewis et al., 2016), and scholars have been alerting the status of language usages (DeChicchis, 1995; Sato, 2012). To mitigate the situation, Bugaeva (2011) created a freely-accessible Ainu dictionary for daily conversation based on the old dictionary published in 1898 by Kotora Jimb¯o and Sh¯ozaburo Kanazawa. Our work will be in line of these movements. Whereas the first motivation is language-internal, the sec- ond motivation is external, as it contributes to our deep understanding of natural languages. Being a member of polysynthetic languages (Baker, 1996), the language ex- hibits many peculiar properties from the view point of the speakers of major languages. Complex verb forma- tion rules called noun incorporation (Mithun, 1984) may be the hallmark of such properties. Other properties (some of which will be discussed later) include pluractionality, possessed case, alienability, and multiple pronominal mark- ers. These …

Content truncated to 500 words