Twitter Part-of-Speech Tagging for All: Overcoming Sparse and Noisy Data

Proceedings of Recent Advances in Natural Language Processing, pages 198–206, Hissar, Bulgaria, 7-13 September 2013. Twitter Part-of-Speech Tagging for All: Overcoming Sparse and Noisy Data Leon Derczynski University of Sheffield leon@dcs.shef.ac.uk Alan Ritter University of Washington aritter@cs.washington.edu Sam Clark University of Washington ssclark@cs.washington.edu Kalina Bontcheva University of Sheffield kalina@dcs.shef.ac.uk Abstract Part-of-speech information is a pre-requisite in many NLP algorithms. However, Twitter text is difficult to part-of-speech tag: it is noisy, with linguistic errors and idiosyncratic style. We present a detailed error analysis of exist- ing taggers, motivating a series of tagger aug- mentations which are demonstrated to improve performance. We identify and evaluate tech- niques for improving English part-of-speech tagging performance in this genre. Further, we present a novel approach to sys- tem combination for the case where available taggers use different tagsets, based on vote- constrained bootstrapping with unlabeled data. Coupled with assigning prior probabilities to some tokens and handling of unknown words and slang, we reach 88.7% tagging accuracy (90.5% on development data). This is a new high in PTB-compatible tweet part-of-speech tagging, reducing token error by 26.8% and sentence error by 12.2%. The model, training data and tools are made available. 1 Introduction Twitter provides a wealth of uncurated text. The site has over 200 million users active each month (O’Carroll, 2012) generating messages at a peak rate over 230 000 per minute (Ashtari, 2013). Information found on Twitter has already been shown to be useful for a variety of applications (e.g. moni- toring earthquakes (Sakaki et al., 2010) and predicting flu (Culotta, 2010)). However, the lack of quality part-of-speech taggers tailored specifically to this emerging genre impairs the accuracy of key down- stream NLP techniques (e.g. named entity recognition, term extraction), and by extension, overall application results. Microblog text (from e.g. Twitter) is characterised by: short messages; inclusion of URIs; username men- tions; topic markers; and threaded conversations. It often presents colloquial content containing abbrevi- ations and errors. Some of these phenomena com- prise linguistic noise, which when coupled with mes- sage brevity (140 characters for “tweets”) and the lack of labeled corpora, make microblog part-of-speech tag- ging very challenging. Alongside the genre’s informal nature, such limits encourage “compressed” utterances, with authors omitting not only needless words but also those with grammatical or contextualising function. Part-of-speech tagging is a central problem in natu- ral language processing, and a key step early in manly NLP pipelines. Machine learning-based part-of-speech (PoS) taggers can exploit labeled training data to adapt to new genres or even languages, through supervised learning. Algorithm sophistication apart, the perfor- mance of these taggers is reliant upon the quantity and quality of available training data. Consequently, lacking large PoS-annotated resources and faced with prevalent noise, state-of-the-art PoS taggers perform poorly on microblog text (Derczynski et al., 2013), with error rates up to ten times higher than on newswire (see Section 3). To address these issues, we propose a data-intensive approach to microblog part-of-speech tagging for En- glish, which overcomes data sparsity by using the thousands …

Content truncated to 500 words