Tagger for Polish Computer Mediated Communication Texts

Proceedings of Recent Advances in Natural Language Processing, pages 1295–1303, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_148 1295 Tagger for Polish Computer Mediated Communication Texts Wiktor Walentynowicz Wrocław University of Science and Technology Maciej Piasecki Wrocław University of Science and Technology {wiktor.walentynowicz,maciej.piasecki,marcin.oleksy}@pwr.edu.pl Marcin Oleksy Wrocław University of Science and Technology Abstract In this paper we present a morpho- syntactic tagger dedicated to Computer- mediated Communication texts in Polish. Its construction is based on an expanded RNN-based neural network adapted to the work on noisy texts. Among several tech- niques, the tagger utilises fastText embed- ding vectors, sequential character embed- ding vectors, and Brown clustering for the coarse-grained representation of sentence structures. In addition a set of manu- ally written rules was proposed for post- processing. The system was trained to disambiguate descriptions of words in re- lation to Parts of Speech tags together with the full morphological information in terms of values for the different grammat- ical categories. We present also evaluation of several model variants on the gold stan- dard annotated CMC data, comparison to the state-of-the-art taggers for Polish and error analysis. The proposed tagger shows significantly better results in this domain and demonstrates the viability of adapta- tion. 1 Introduction Morpho-syntactic disambiguation (called also morpho-syntactic tagging) is an important pre- processing step in many text processing pipelines (e.g. terminology and information extraction), es- pecially in the case of highly inflected languages where tagging is tightly correlated with lemma- tisation. Work on the development of programs for morpho-syntactic disambiguaters, henceforth taggers, has been concentrated on texts written in the standard language, i.e. containing only small percentage free of jargon, extra-linguistic elements like codes or symbols, etc. Computer- mediated communication (CMC) texts, especially texts from different kinds of social media, are writ- ten in language that often is significantly different from the standard language, see Sec. 3. While in the case of texts in a standard language, a state-of- the-art tagger reaches near 95% of accuracy in dis- ambiguation, taggers for CMC texts express much lower accuracy. Moreover, much fewer works are devoted to tagging CMC texts than text of the stan- dard language, while the problem of processing CMC texts is continuously growing in importance and numbers of applications. Thus, the fast growth of social media requires appropriate adaptation, or even expansion, of the tagging methods, to serve growing interest and requirements in the process- ing of texts of such kinds. According to our best knowledge, a morpho-syntactic tagger dedicated to CMC texts in the Polish language has not been yet developed or at least made publicly available. The goal of this work is a far going adaptation of a tagger for the standard Polish to the demands of CMC texts in Polish. In the rest of the paper, first solutions for tag- ging the standard Polish language and CMC texts in general are discussed. Next, we analyse charac- teristic features of CMC texts on the basis of a col- lected CMC corpus. A tagger model is proposed. Finally, …

Content truncated to 500 words