Morpheme-based dictionaries and recognition units provide clear advantages for speech recognition systems, particularly by reducing out-of-vocabulary rates and word error rates in highly inflected and agglutinative languages.
The retrieved papers provide consistent empirical support for the utility of morpheme-based and subword units in speech recognition systems, specifically highlighting performance improvements in agglutinative languages.