Amharic Part-of-Speech Tagger for Factored Language Modeling

International Conference RANLP 2009 - Borovets, Bulgaria, pages 428–433 Amharic Part-of-Speech Tagger for Factored Language Modeling Martha Yifiru Tachbelie and Wolfgang Menzel Department of Informatics, University of Hamburg Vogt-K¨olln Srt. 30, D-22527 Hamburg, Germany tachbeli, menzel@informatik.uni-hamburg.de Abstract This paper presents Amharic part of speech tag- gers developed for factored language modeling. Hidden Markov Model (HMM) and Support Vec- tor Machine (SVM) based taggers have been trained using the TnT and SVMTool. The over- all accuracy of the best performing TnT- and SVM-based taggers is 82.99% and 85.50%, re- spectively. Generally, with respect to accuracy SVM-based taggers perform better than TnT- based taggers although TnT-based taggers are more efficient with regard to speed and memory requirement. We have developed factored lan- guage models (with two and four parents) for which the estimation of the probability for each word depends on the previous one or two words and their POS. These language models have been used in an Amharic speech recognition task in a lattice rescoring framework and a significant im- provement in word recognition accuracy has been observed. Keywords POS tagging, Amharic, factored language model 1 Introduction Language models are fundamental to many natural language applications such as automatic speech recog- nition (ASR). The most widely used class of language models, namely statistical ones, provide an estimate of the probability of a word sequence W for a given task. However, the probability distribution depends on the available training data — large amounts of training data are required so as to ensure statistical significance. Even if a large training corpus is available, there may be still many possible word sequences which will not be encountered at all, or which appear with a sta- tistically insignificant frequency (data sparseness prob- lem) [21]. In morphologically rich languages, there are even individual words that might not be encountered in the training data irrespective of its size (Out-Of- Vocabulary words problem). The data sparseness problem in statistical language modeling is more serious for languages with a rich morphology. These languages have a high vocabulary growth rate which results in a high perplexity and a large number of out of vocabulary words [19]. There- fore, sub-words (morphemes), instead of words, have been and are being used as modeling units in language modeling so as to build more robust language models even if only insufficient training data is available. 1.1 The morphology of Amharic Amharic is one of the morphologically rich languages. It is a major language spoken mainly in Ethiopia and belongs to the Semitic branch of the Afro-Asiatic su- per family. Amharic is related to Hebrew, Arabic and Syrian. Like other Semitic languages such as Arabic, Amharic exhibits a root-pattern morphological phe- nomenon. A root is a set of consonants (called radi- cals) which has a basic ’lexical’ meaning. A pattern consists of a set of vowels which are inserted (inter- calated) among the consonants of a root to form a stem. The pattern is combined with a particular pre- fix or suffix to create a …

Content truncated to 500 words