Large Corpora for Turkic Languages and UnsupervisedMorphological Analysis
Vít Baisa, Vít Suchomel · 2012
In this article we describe six new web corpora for Turkish, Azerbaijani, Kazakh, Turkmen, Kyrgyz and Uzbek languages. The data for these corpora was automatically crawled from the web by SpiderLing. Only minimal knowledge of these languages was required to obtain the data in raw form. Corpora are tokenized only since morphological analyzers and disambiguators for these languages are not available (except for Turkish). Subsequent experiment with unsupervised morphological segmentation was carried out on the Turkish corpus. In this experiment we achieved encouraging results. We used data provided for MorphoChallenge competition for the purpose of evaluation.