Automatic Symbol Processing for Language Model Building in Slavic Languages

When we want to adapt an existing automatic speech recognition system to a new language, we need a large corpus of texts to create a lexicon, a language model and a database of annotated recordings to train an acoustic model. Usually the texts in the corpus (or in annotations) contain not only words but also some other symbols, mainly strings of digits, special characters and some frequent abbreviations of units. The common feature of all these symbols is that there is not a straightforward correspondence between their printed form and the spoken one. The main goal of this work was to develop efficient tools for automatic translation of symbols or symbolic terms to words for almost all Slavic languages. In this paper we present the research of the basic elements and the production rules in Slavic languages which was used for design of our universal text preand post-processing tools.