From Word Segmentation to POS Tagging for Vietnamese

From Word Segmentation to POS Tagging for Vietnamese Dat Quoc Nguyen1, Thanh Vu2, Dai Quoc Nguyen3, Mark Dras1 and Mark Johnson1 1Department of Computing, Macquarie University, Australia {dat.nguyen, mark.dras, mark.johnson}@mq.edu.au 2NIHR Innovation Observatory, Newcastle University, United Kingdom thanh.vu@newcastle.ac.uk 3PRaDA Centre, Deakin University, Australia dai.nguyen@deakin.edu.au Abstract This paper presents an empirical com- parison of two strategies for Vietnamese Part-of-Speech (POS) tagging from unseg- mented text: (i) a pipeline strategy where we consider the output of a word seg- menter as the input of a POS tagger, and (ii) a joint strategy where we predict a combined segmentation and POS tag for each syllable. We also make a comparison between state-of-the-art (SOTA) feature- based and neural network-based models. On the benchmark Vietnamese treebank (Nguyen et al., 2009), experimental results show that the pipeline strategy produces better scores of POS tagging from unseg- mented text than the joint strategy, and the highest accuracy is obtained by using a feature-based model. 1 Introduction POS tagging is one of the most fundamental nat- ural language processing (NLP) tasks. In English where white space is a strong indicator of word boundaries, POS tagging is an important first step towards many other NLP tasks. However, white space when written in Vietnamese is also used to separate syllables that constitute words. So for Vietnamese NLP, word segmentation is referred to as the key first step (Dien et al., 2001). When applying POS tagging to real-world Viet- namese text where gold word-segmentation is not available, the pipeline strategy is to first segment the text by using a word segmenter, and then feed the word-segmented text—which is the out- put of the word segmenter—as the input to a POS tagger. For example, given a written text “thu∏ thu nh™p cá nhân” (individualcá_nhân incomethu_nh™p taxthu∏) consisting of 5 syllables, the word seg- menter returns a two-word phrase “thu∏_thu_nh™p cá_nhân.”1 Then given the input segmented text “thu∏_thu_nh™p cá_nhân”, the POS tagger returns “thu∏_thu_nh™p/N cá_nhân/N.” A class of approaches to POS tagging from un- segmented text that has been actively explored in other languages, such as in Chinese and Japanese, is joint word segmentation and POS tagging (Zhang and Clark, 2008). A possible joint strat- egy is to assign a combined segmentation and POS tag to each syllable (Kruengkrai et al., 2009). For example, given the input text “thu∏ thu nh™p cá nhân”, the joint strategy would produce “thu∏/B- N thu/I-N nh™p/I-N cá/B-N nhân/I-N”, where B refers to the beginning of a word and I refers to the inside of a word. Shao et al. (2017) showed that this joint strategy gives SOTA results for Chi- nese POS tagging by utilizing a BiLSTM-CNN- CRF model (Ma and Hovy, 2016). In this paper, we present the first empirical study comparing the joint and pipeline strate- gies for Vietnamese POS tagging from unseg- mented text. In addition, we make a comparison between SOTA feature-based and neural network- based models, which, to the best of our knowledge, has not done in any prior work on Vietnamese. …

Content truncated to 500 words