--- title: "Chinese Part-of-Speech Tagging — Ng 2004" description: "Chinese text has no explicit word boundaries, so POS tagging depends on segmentation." authors: ["Hwee Tou Ng", "Jin Kiat Low"] year: 2004 venue: "Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing" source: "https://aclanthology.org/W04-3236/" licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/ng-2004-chinese-part-speech-tagging-one/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ng-2004-chinese-part-speech-tagging-one/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2004](https://md.research.elamur.ai/free/en/papers/2004/) / Chinese Part-of-Speech Tagging: One-at-a-Time or All-at-Once? Word-Based or Character-Based? # Chinese Part-of-Speech Tagging: One-at-a-Time or All-at-Once? Word-Based or Character-Based? **Authors:** [Hwee Tou Ng](https://md.research.elamur.ai/free/en/authors/hwee-tou-ng/), Jin Kiat Low · **Year:** 2004 · **Venue:** Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/) **Links:** [Publisher's page](https://aclanthology.org/W04-3236/) · [PDF](https://aclanthology.org/W04-3236.pdf) · [OpenAlex](https://openalex.org/W1575907248) **Citations:** 158 citations (openalex-snapshot · 2026-10-14) · [85](#cited-by) in our library, of the 140884 papers whose bibliographies we read ## Summary Chinese text has no explicit word boundaries, so POS tagging depends on segmentation. Ng and Low ask two design questions. First, should segmentation and tagging be done in sequence ("one-at-a-time") or jointly ("all-at-once")? Second, should the tagger rely on word-level or character-level features? They compare the options within a single maximum entropy framework, using 10-fold cross-validation on the 250K-word Penn Chinese Treebank (CTB) 3.0. To make the pipeline comparison fair, they first build a strong character-level segmenter. It labels each character with a boundary tag and decodes with dynamic programming. This segmenter beats the best SIGHAN 2003 closed-track systems on three of four corpora. Porting Ratnaparkhi's English word-based tagger directly to Chinese performs poorly. Character-based tagging works much better. The joint character-based model is the most accurate, but its advantage over the pipelined character-based variant is small, and it costs far more time to train and run. The practical lesson is that character information matters most for Chinese, and joint modelling buys only a marginal gain at a high computational price. ## Key points - **Segmenter design**: - Each character gets a b/m/e/s boundary tag from a maximum entropy model (Java opennlp maxent, feature cutoff 2, 100 iterations). - On CTB 10-fold CV it averages 95.1% F-measure, versus Luo (2003) at 94.6% (and roughly 94.9% with a 58K-entry word list). - **SIGHAN 2003 bakeoff**: - The segmenter beats the best closed-track score on AS, HK and PK. - It ranks third on CTB, whose test set has an 18.1% OOV rate. - With extra AS training data (open track), it reaches 92.2% on CTB, above the best open-track result. - **Ablation**: The new features add about 0.4% F-measure on average, and the consistency post-processing adds another 0.3%. - **POS accuracy on CTB**: | Approach | POS accuracy | Segmentation F-measure | Testing time (10% of CTB) | |---|---|---|---| | One-at-a-time, word-based | 84.1% | 95.1% | about 1 min 20 s | | One-at-a-time, character-based | 91.7% | 95.1% | about 1 min 50 s | | All-at-once, character-based | 91.9% | 95.2% | about 20 min | - **Cost and significance**: - The joint model needs about 3 hours of training, compared with 12 minutes for the segmenter. - In a paired t-test (p \< 0.01), the joint model's POS gain is significant, but its segmentation gain is not. - The all-at-once character-based setup is essentially Luo's (2003) approach. ## Interesting ideas - **Constrained decoding over word candidates**: Instead of picking the best label for each character, the decoder enumerates candidate last words up to 20 characters long. This guarantees only valid b/m/e/s sequences, and it lets word-level features such as W0C0 be used at test time. The trick could carry over to any BIO-style labelling task where invalid label sequences are a problem. - **Coarse character-type n-gram feature**: Characters are mapped to four classes (digits, date characters, Latin letters, other), and the class pattern over a five-character window is used as a feature. This lets the model generalise to unseen numbers and dates. The same idea suits other languages or noisy text with many rare tokens. - **Document-level consistency post-processing**: If a sequence of 2–6 adjacent output words, joined together, matches a 3+ character word segmented elsewhere in the same document, the sequence is merged into that word. This cheap step improves consistency for names and terms. - **Character-level evidence for unknown words**: Chinese characters carry meaning, and unseen words are far more frequent than unseen characters. Tagging from character features therefore helps on OOV words. The same logic could apply to subword or morpheme features in morphologically rich languages such as Ukrainian. - **Weighing accuracy against cost**: The comparison reports training and testing time and significance tests alongside accuracy. This framing showed that a cheaper pipeline can be a sensible choice when the joint model's gain is small. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/) Topics: [Part-of-speech tagging](https://md.research.elamur.ai/free/en/topics/part-of-speech-tagging/) ## Cited by Cited by 85 papers of our library cite it (of the 140884 whose bibliographies we read); 4 of them have a page here, 81 not yet. - [A Conditional Random Field Word Segmenter for Sighan Bakeoff 2005](https://md.research.elamur.ai/free/en/papers/tseng-2005-conditional-random-field-word-segmenter/) - [Optimizing Chinese Word Segmentation for Machine Translation Performance](https://md.research.elamur.ai/free/en/papers/chang-2008-optimizing-chinese-word-segmentation-machine/) - [Pointwise Prediction for Robust, Adaptable Japanese Morphological Analysis](https://md.research.elamur.ai/free/en/papers/neubig-2011-pointwise-prediction-robust-adaptable-japanese/) - [Automatic Cross-Lingual Similarization of Dependency Grammars for Tree-based Machine Translation](https://md.research.elamur.ai/free/en/papers/jiang-2016-automatic-cross-lingual-similarization-dependency/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{ng2004chinese, title = {{Chinese Part-of-Speech Tagging: One-at-a-Time or All-at-Once? Word-Based or Character-Based?}}, author = {Ng, Hwee Tou and Low, Jin Kiat}, booktitle = {Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing}, year = {2004}, url = {https://aclanthology.org/W04-3236/} } ``` - [← previous SMOR: A German Computational Morphology Covering Derivation, Composition and Inflection](https://md.research.elamur.ai/free/en/papers/schmid-2004-smor-german-computational-morphology-covering/) - [next → The Importance of Supertagging for Wide-Coverage CCG Parsing](https://md.research.elamur.ai/free/en/papers/aclbulk-2004-clark-importance-supertagging-wide-coverage-ccg/) - [↑ back to papers of 2004](https://md.research.elamur.ai/free/en/papers/2004/) ## References 11 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. H. Jing, R. Florian, X. Luo, T. Zhang, and A. Ittycheriah. 2003. HowtogetaChineseName (Entity): segmentation and combination issues. In Proc. of EMNLP . [find](https://scholar.google.com/scholar?q=H%2E+Jing%2C+R%2E+Florian%2C+X%2E+Luo%2C+T%2E+Zhang%2C+and+A%2E+Ittycheriah%2E+2003%2E+HowtogetaChineseName+%28Entity%29%3A+segmentation+and+combination+issues%2E+In+Proc%2E+of+EMNLP+%2E) 2. O. Y. Kwong and B. K. Tsou. 2003. Categorial fluidity in Chinese and its implications for part-of-speech tagging. In Proc. of EACL . [find](https://scholar.google.com/scholar?q=O%2E+Y%2E+Kwong+and+B%2E+K%2E+Tsou%2E+2003%2E+Categorial+fluidity+in+Chinese+and+its+implications+for+part%2Dof%2Dspeech+tagging%2E+In+Proc%2E+of+EACL+%2E) 3. X. Luo. 2003. A maximum entropy Chinese character-based parser. In Proc. of EMNLP . [find](https://scholar.google.com/scholar?q=X%2E+Luo%2E+2003%2E+A+maximum+entropy+Chinese+character%2Dbased+parser%2E+In+Proc%2E+of+EMNLP+%2E) 4. A. Ratnaparkhi. 1996. A maximum entropy model for part-of-speech tagging. In Proc. of EMNLP . [find](https://scholar.google.com/scholar?q=A%2E+Ratnaparkhi%2E+1996%2E+A+maximum+entropy+model+for+part%2Dof%2Dspeech+tagging%2E+In+Proc%2E+of+EMNLP+%2E) [our card](https://md.research.elamur.ai/free/en/papers/ratnaparkhi-1996-maximum-entropy-model-part-speech/) 5. R. Sproat, C. Shih, W. Gale, and N. Chang. 1996. A stochastic finite-state word-segmentation algorithm for Chinese. Computational Linguistics , 22(3):377-404. [find](https://scholar.google.com/scholar?q=R%2E+Sproat%2C+C%2E+Shih%2C+W%2E+Gale%2C+and+N%2E+Chang%2E+1996%2E+A+stochastic+finite%2Dstate+word%2Dsegmentation+algorithm+for+Chinese%2E+Computational+Linguistics+%2C+22%283%29%3A377%2D404%2E) [our card](https://md.research.elamur.ai/free/en/papers/sproat-1996-stochastic-finite-state-word-segmentation/) 6. R. Sproat and T. Emerson. 2003. The first international Chinese word segmentation bakeoff. In Proc. of SIGHAN Workshop . [find](https://scholar.google.com/scholar?q=R%2E+Sproat+and+T%2E+Emerson%2E+2003%2E+The+first+international+Chinese+word+segmentation+bakeoff%2E+In+Proc%2E+of+SIGHAN+Workshop+%2E) 7. W. J. Teahan, Y. Wen, R. McNab, and I. H. Witten. 2000. A compression-based algorithm for Chinese word segmentation. Computational Linguistics , 26(3): 375-393. [find](https://scholar.google.com/scholar?q=W%2E+J%2E+Teahan%2C+Y%2E+Wen%2C+R%2E+McNab%2C+and+I%2E+H%2E+Witten%2E+2000%2E+A+compression%2Dbased+algorithm+for+Chinese+word+segmentation%2E+Computational+Linguistics+%2C+26%283%29%3A+375%2D393%2E) 8. F. Xia, M. Palmer, N. Xue, M. E. Okurowski, J. Kovarik, F-D Chiou, S. Huang, T. Kroch, and M. Marcus. 2000. Developing guidelines and ensuring consistency for Chinese text annotation. In Proc. of LREC . [find](https://scholar.google.com/scholar?q=F%2E+Xia%2C+M%2E+Palmer%2C+N%2E+Xue%2C+M%2E+E%2E+Okurowski%2C+J%2E+Kovarik%2C+F%2DD+Chiou%2C+S%2E+Huang%2C+T%2E+Kroch%2C+and+M%2E+Marcus%2E+2000%2E+Developing+guidelines+and+ensuring+consistency+for+Chinese+text+annotation%2E+In+Proc%2E+of+LREC) 9. N. Xue and L. Shen. 2003. Chinese word segmentation as LMR tagging. In Proc. of SIGHAN Workshop . [find](https://scholar.google.com/scholar?q=N%2E+Xue+and+L%2E+Shen%2E+2003%2E+Chinese+word+segmentation+as+LMR+tagging%2E+In+Proc%2E+of+SIGHAN+Workshop+%2E) 10. H-P Zhang, H-K Yu, D-Y Xiong, and Q. Liu. 2003. HHMM-based Chinese lexical analyzer ICTCLAS. In Proc. of SIGHAN Workshop . [find](https://scholar.google.com/scholar?q=H%2DP+Zhang%2C+H%2DK+Yu%2C+D%2DY+Xiong%2C+and+Q%2E+Liu%2E+2003%2E+HHMM%2Dbased+Chinese+lexical+analyzer+ICTCLAS%2E+In+Proc%2E+of+SIGHAN+Workshop+%2E) 11. G. Zhou and J. Su, 2003. A Chinese efficient analyser integrating word segmentation, part-of-speech tagging, partial parsing and full parsing. In Proc. of SIGHAN Workshop . [find](https://scholar.google.com/scholar?q=G%2E+Zhou+and+J%2E+Su%2C+2003%2E+A+Chinese+efficient+analyser+integrating+word+segmentation%2C+part%2Dof%2Dspeech+tagging%2C+partial+parsing+and+full+parsing%2E+In+Proc%2E+of+SIGHAN+Workshop+%2E)