Morphological Analysis Corpus Construction of Uyghur

Morphological Analysis Corpus Construction of Uyghur Gulinigeer Abudouwaili1,3, Kahaerjiang Abiderexiti1,3, Jiamila Wushouer 1,3,∗, Yunfei Shen2,3, Turenisha Maimaitimin1,3 and Tuergen Yibulayin1 1School of Information Science and Engineering, Xinjiang University, Urumqi, Xinjiang 830046, China 2School of Software, Xinjiang University, Urumqi, Xinjiang 830091, China 3Laboratory of Multi-Language Information Technology, Urumqi, Xinjiang 830046, China {107556518131, shenyunfei}@stu.xju.edu.cn, {jiamila, turgun}@xju.edu.cn, kaharjan@aliyun.com, turanisa11@163.com Abstract Morphological analysis is a fundamental task in natural language processing, and results can be applied to different downstream tasks such as named entity recognition, syntactic analysis, and machine translation. However, there are many problems in morphological analysis, such as low accuracy caused by a lack of resources. In this paper, to alleviate the lack of resources in Uyghur morphological analysis research, we construct a Uyghur morphological analysis corpus based on the analysis of grammatical features and the format of the general morphological analysis corpus. We define morphological tags from 14 dimensions and 53 features, manually annotate and correct the dataset. Finally, the corpus provided some informations such as word, lemma, part of speech, morphological analysis tags, morphological segmentation, and lemmatization. Also, this paper analyzes some basic features of the corpus, and we use the models and datasets provided by SIGMORPHON Shared Task organizers to design comparative experiments to verify the corpus’s availability. Results of the experiment are 85.56%, 88.29%, respectively. The corpus provides a reference value for morphological analysis and promotes the research of Uyghur natural language processing. 1 Introduction Morphological analysis is the process of dividing words into different morphologies or morphemes and analyzing their internal structure to obtain grammatical information. It is an essential step in lexical analysis. Therefore, Strakov´a et al. [2015] showed that the related research of morphological analysis has also attracted the attention of most researchers. In natural language processing based on deep learn- ing, researchers have found that if modeling to words directly, it is easy to ignore the relationship within the words, which will also bring limitations to the model. Thus, the model’s input has changed from a single word to a character, subword, and morpheme or morphology. When the word is split into different granularities (Zhuang et al. [2018], ¨Ust¨un et al. [2019], Zhu et al. [2019]), the performance of the model is improved. Among these words’ segmentation methods, morphology or morpheme relies on the mor- phological analysis tagger or manual annotation. In languages with mature natural language processing technologies such as English, Finnish, and Chinese, there are many manually annotated morpholog- ical analysis corpus, and the lexical analysis (including morphological analysis) technology of these languages has reached a high level. To further promote the lexical analysis of minority languages in Xin- jiang, many researchers have obtained preliminary results in lexical analysis. Ibrahim and Baoshe [2011] constructed a Uyghur lexical tagging corpus of 1.23 million words; Enwer et al. [2015] built a Uyghur stemming corpus of 10,000 sentence sets; Altenbek et al. [2014] annotated about 30,000 Kazakh sen- tences and studied a lexical analysis of Kazakh; Osman et al. [2019] constructed a character-level Uyghur morphological …

Content truncated to 500 words