Selección de unidades léxicas para reconocimiento automático del habla continua en euskera

Basque is an agglutinative language, which implies that corpusvocabulary can not be defined with words, because they growcombinationally making medium and large vocabulary tasks intractable. Pseudo-morphemes, generated with an automatic segmentation tool, could be an alternative choice for building the lexicon and the language model, for they notably reduce the vocabulary size. In Basque, there are many short and acoustically very similar morphemes. This phenomenon has to be taken into account, because the acoustic-phonetic decodification proccess can influence the CSR task, increasing the possibility of confussion and insertion of certain lexical units (very short units with high rates of acoustic confussion). A feasible way to deal with this problem is to avoid the segmentation of those units. The next step to improve the CSR system in Basque is the use of a language model in order to guide the recognition proccess.