Unlimited vocabulary speech recognition for agglutinative languages

Proceedings of the Human Language Technology Conference of the North American Chapter of the ACL, pages 487–494, New York, June 2006. c⃝2006 Association for Computational Linguistics Unlimited vocabulary speech recognition for agglutinative languages Mikko Kurimo1, Antti Puurula1, Ebru Arisoy2, Vesa Siivola1, Teemu Hirsimäki1, Janne Pylkkönen1, Tanel Alumäe3, Murat Saraclar2 1 Adaptive Informatics Research Centre, Helsinki University of Technology P.O.Box 5400, FIN-02015 HUT, Finland {Mikko.Kurimo,Antti.Puurula,Vesa.Siivola}@tkk.fi 2 Bogazici University, Electrical and Electronics Eng. Dept. 34342 Bebek, Istanbul, Turkey {arisoyeb,murat.saraclar}@boun.edu.tr 3 Laboratory of Phonetics and Speech Technology, Institute of Cybernetics, Tallinn Technical University, Estonia tanel.alumae@phon.ioc.ee Abstract It is practically impossible to build a word-based lexicon for speech recogni- tion in agglutinative languages that would cover all the relevant words. The prob- lem is that words are generally built by concatenating several prefixes and suffixes to the word roots. Together with com- pounding and inflections this leads to mil- lions of different, but still frequent word forms. Due to inflections, ambiguity and other phenomena, it is also not trivial to automatically split the words into mean- ingful parts. Rule-based morphological analyzers can perform this splitting, but due to the handcrafted rules, they also suf- fer from an out-of-vocabulary problem. In this paper we apply a recently proposed fully automatic and rather language and vocabulary independent way to build sub- word lexica for three different agglutina- tive languages. We demonstrate the lan- guage portability as well by building a successful large vocabulary speech recog- nizer for each language and show superior recognition performance compared to the corresponding word-based reference sys- tems. 1 Introduction Speech recognition for dictation or prepared radio and television broadcasts has had huge advances during the last decades. For example, broadcast news (BN) in English can now be recognized with about ten percent word error rate (WER) (NIST, 2000) which results in mostly quite understandable text. Some rare and new words may be missing but the result has proven to be sufficient for many im- portant applications, such as browsing and retrieval of recorded speech and information retrieval from the speech (Garofolo et al., 2000). However, besides the development of powerful computers and new al- gorithms, a crucial factor in this development is the vast amount of transcribed speech and suitable text data that has been collected for training the mod- els. The problem faced in porting the BN recogni- tion systems to conversational speech or to other lan- guages is that almost as much new speech and text data have to be collected again for the new task. The reason for the need for a vast amount of train- ing texts is that the state-of-the-art statistical lan- guage models contain a huge amount of parameters to be estimated in order to provide a proper probabil- ity for any possible word sequence. The main reason for the huge model size is that for an acceptable cov- erage in an English BN task, the vocabulary must be very large, at least 50,000 words, or more. For languages with a higher degree of word …

Content truncated to 500 words