Recognition of Out-of-vocabulary Words and Their Semantic Category

In almost all applications of automatic speech recognition, especially in spontaneous speech tasks, the recognizer vocabulary cannot cover all occurring words. There is always a signiicant amount of out-of-vocabulary (OOV) words even when the vocabulary size is very large. In this paper we present a new approach for the integration of OOV words into statistical language models. It is based on the fact that the context of an OOV word contains information on its semantic category. This allows us to predict the semantic category of the OOV word during the word recognition process. This information is useful for improved recognition of the word neighborhood, and can also be used by postpro-cessing modules, e.g. in spoken dialog systems. Although we use a simple acoustic model for OOV words, we achieve a 6% reduction of word error rate on spontaneous speech data with about 5% out-of-vocabulary rate.