Constructing Uyghur Name Entity Recognition System using Neural Machine Translation Tag Projection

Constructing Uyghur Named Entity Recognition System using Neural Machine Translation Tag Projection Azmat Anwar, Xiao Li, Yating Yang, Rui Dong and Turghun Osman Xinjiang Technical Institute of Physics & Chemistry, Chinese Academy of Sciences, Urumqi, China University of Chinese Academy of Sciences, Beijing, China Xinjiang Laboratory of Minority Speech and Language Information Processing, Urumqi, China {azmat,xiaoli,yangyt,dongrui,turghun}@ms.xjb.ac.cn Abstract Although named entity recognition achieved great success by introducing the neural networks, it is challenging to apply these models to low resource languages including Uyghur while it de- pends on a large amount of annotated training data. Constructing a well-annotated named entity corpus manually is very time-consuming and labor-intensive. Most existing methods based on the parallel corpus combined with the word alignment tools. However, word alignment methods introduce alignment errors inevitably. In this paper, we address this problem by a named en- tity tag transfer method based on the common neural machine translation. The proposed method marks the entity boundaries in Chinese sentence and translates the sentences to Uyghur by neural machine translation system, hope that neural machine translation will align the source and target entity by the self-attention mechanism. The experimental results show that the Uyghur named entity recognition system trained by the constructed corpus achieve good performance on the test set, with 73.80% F1 score(3.79% improvement by baseline). 1 Introduction Named Entity Recognition (NER) is a task of identifying named entities (NEs), especially person names (PER), location names (LOC), organization names (ORG), and classifying them into some pre-defined target entity classes (Hobbs et al., 1997). NER is essential to many natural language processing (NLP) tasks such as relation extraction (Christopoulou et al., 2019), event detection (Cakır and Virtanen, 2019), knowledge graph construction (Bosselut et al., 2019) and so on. Although the NER achieves great success by the introduction of the advanced neural networks (Collobert et al., 2011; Huang et al., 2015; Chiu and Nichols, 2016; Lample et al., 2016; Ma and Hovy, 2016; Yang et al., 2016; Peters et al., 2017; Liu et al., 2018; Peters et al., 2018), these methods are highly dependent on a large amount of annotated training data, and thus challenging to apply these models to low resource languages including Uyghur. Constructing a well-annotated NE corpus manually is very time-consuming and labor-intensive. Instead, Cross-lingual transfer is an effective solution, which addresses this challenge by transferring knowledge from a high-resource source language with abundant entity labels to a low-resource target language with few or no labels. According to the resource availability of the target language, different types of NER methods are proposed, such as bilingual parallel corpus based tag projection (Yarowsky et al., 2001; Ehrmann et al., 2011; Wang et al., 2013; Fang and Cohn, 2016; Ni et al., 2017), cross- lingual word embedding (Fang and Cohn, 2017; Wang et al., 2017; Huang et al., 2018), cross-lingual Wikification (Kim et al., 2012; Nothman et al., 2013; Tsai et al., 2016; Pan et al., 2017) or multi-task learning (Yang et al., 2016; Lin et al., 2018). As a low …

Content truncated to 500 words