Analyzing Stemming Approaches for Turkish Multi-Document Summarization

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 702–706, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Analyzing Stemming Approaches for Turkish Multi-Document Summarization Muhammed Yavuz Nuzumlalı Arzucan ¨Ozg¨ur Department of Computer Engineering Bo˘gazic¸i University TR-34342, Bebek, ˙Istanbul, Turkey {yavuz.nuzumlali,arzucan.ozgur}@boun.edu.tr Abstract In this study, we analyzed the effects of ap- plying different levels of stemming approaches such as fixed-length word truncation and mor- phological analysis for multi-document sum- marization (MDS) on Turkish, which is an ag- glutinative and morphologically rich language. We constructed a manually annotated MDS data set, and to our best knowledge, reported the first results on Turkish MDS. Our results show that a simple fixed-length word trun- cation approach performs slightly better than no stemming, whereas applying complex mor- phological analysis does not improve Turkish MDS. 1 Introduction Automatic text summarization has gained more impor- tance with the enormous growth and easy availability of the Internet. It is now possible to reach extensive and continuously growing amount of resources. However, this situation brings its own challenges such as finding the relevant documents, and absorbing a large quan- tity of relevant information (Gupta and Lehal, 2010). The goal of multi-document summarization (MDS) is to automatically create a summary of a set of docu- ments about the same topic without losing the impor- tant information. Several approaches for MDS have been proposed in the last decade. However, most of them have only been applied to a relatively small set of languages, mostly English, and recently also to lan- guages like Chinese, Romanian, Arabic, and Spanish (Giannakopoulos, 2013). Previous studies have shown that methods proposed for languages like English do not generally work well for morphologically rich languages like Finnish, Turk- ish, and Czech, and additional methods considering the morphological structures of these languages are needed (Eryi˘git et al., 2008). For instance, Turkish is an ag- glutinative language where root words can take many derivational and inflectional affixes. This feature re- sults in a very high number of different word surface forms, and eventually leads to the data sparseness prob- lem. Hakkani-T¨ur et al. (2000) analyzed the number of unique terms for Turkish and English and showed that the term count for Turkish is three times more than En- glish for a corpus of 1M words. There are only a few studies for text summariza- tion on Turkish, all of which are about single-document summarization (Altan, 2004; C¸ ı˘gır et al., 2009; ¨Ozsoy et al., 2010; G¨uran et al., 2010; G¨uran et al., 2011). Some of these studies applied morphological analysis methods, but none of them analyzed their effects in de- tail. To our best knowledge, this paper reports the first multi-document summarization study for Turkish. We used LexRank as the main summarization algorithm (Erkan and Radev, 2004), applied and analyzed differ- ent levels of stemming methods such as complex mor- phological analysis and fixed-length word truncation. We also created the first manually annotated MDS data set for Turkish, which has been …

Content truncated to 500 words