This paper analyzes the impact of German compound words on speech recognition. It is well known that, due to an idiosyncrasy of German orthography, compound words make up a major fraction of German vocabulary. And most OutOf-Vocabulary (OOV) compounds are composed of frequent words already in the lexicon. This paper introduces a new method for handling the components of compounds rather than the compounds themselves. This not only reduces the vocabulary, and therefore the perplexity, but also improves word accuracy. And reduced perplexity means a more robust language model.

No abstract

This paper analyzes the impact of German compound words on speech recognition. It is well known that, due to an idiosyncrasy of German orthography, compound words make up a major fraction of German vocabulary. And most Out-Of-Vocabulary (OOV) compounds are composed of frequent words already in the lexicon. This paper introduces a new method for handling the components of compounds rather than the compounds themselves. This not only reduces the vocabulary, and therefore the perplexity, but also improves word accuracy. And reduced perplexity means a more robust language model.

This paper presents a novel approach to using condence scores for word graph rescoring. For each w ord in the system's vocabulary, w e computed the probability that the observation is correct given its acoustic score. Afterwards, we used these probabilities for rescoring word graphs outputted by the recognizer. We will present some implementation details as well as accuracy improvements obtained using this method.

This paper presents an analysis of the Out-0f-Vocabulary (OOV) word problem and results of experiments in language modeling of OOV words. In particular, we introduce the method of iterative substitution for correcting distortions caused by OOV words in the language model. We evaluate the results on two well known spontaneous speech tasks: Verbmobil and ATIS. We show that perplexity as well as error rate reductions can be achieved using iterutive substitution. Further, we present preliminary results in combining newspaper texts with the Verbmobil (spontaneous speech) corpus. We could reduce the perplexity of the Verbmobil test set by augmenting the training corpus with newspaper texts. Preliminary recognition results show only slight improvements in the Word Accuracy and detection of OOV words.