When a trigram backoff language model is created from a large body of text, trigrams and bigrams that occur few times in the training text are often excluded from the model in order to decrease the model size. Generally, the elimination of n-grams with very low counts is believed to not significantly affect model performance. This project investigates the degradation of a trigram backoff model's perplexity and word error rates as bigram and trigram cutoffs are increased. The advantage of reduction in model size is compared to the increase in word error rate and perplexity scores. More importantly, this project also investigates alternative ways of excluding bigrams and trigrams from a backoff language model, using criteria other than the number of times an n-gram occurs in the training text. Specifically, a difference method has been investigated where the difference in the logs of the original and backed off trigram and bigram probabilities is used as a basis for n-gram exclusion from...

Speech recognition performance is severely affected when the lexical, syntactic, or semantic characteristics of the discourse in the training and recognition tasks differ. The aim of language model adaptation is to exploit specific, albeit limited, knowledge about the recognition task to compensate for this mismatch. More generally, an adaptive language model seeks to maintain an adequate representation of the current task domain under changing conditions involving potential variations in vocabulary, syntax, content, and style. This paper presents an overview of the major approaches proposed to address this issue, and offers some perspectives regarding their comparative merits and associated tradeoffs. Ó 2003 Elsevier B.V. All rights reserved. 1.

In this paper, we present novel techniques for performing topic adaptation on an¢-gram language model. Given training text labeled with topic information, we automatically identify the most relevant topics for new text. We adapt our language model toward these topics using an exponential model, by adjusting probabilities in our model to agree with those found in the topical subset of the training data. For efficiency, we do not normalize the model; that is, we do not require that the “probabilities” in the language model sum to 1. With these techniques, we were able to achieve a modest reduction in speech recognition word-error rate in the Broadcast News domain.

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 significant amount of out-of-vocabulary words even when the vocabulary size is very large. In this paper we present a new approach for the integration of out-of-vocabulary words into statistical language models. We use category information for all words in the training corpus to define a function that gives an approximation of the out-of-vocabulary word emission probability for each word category. This information is integrated into the language models. Although we use a simple acoustic model for out-of-vocabulary words, we achieve a 6% reduction of word error rate on spontaneous speech data with about 5% out-of-vocabulary rate.

The vocabulary of a continuous speech recognition (CSR) system is a significant factor in determining its performance. In this paper, we present three principled approaches to select the target vocabulary for a particular domain by trading off between the target out-of-vocabulary (OOV) rate and vocabulary size. We evaluate these approaches against an ad-hoc baseline strategy. Results are presented in the form of OOV rate graphs plotted against increasing vocabulary size for each technique. 1.

Randomised techniques allow very big language models to be represented succinctly. However, being batch-based they are unsuitable for modelling an unbounded stream of language whilst maintaining a constant error rate. We present a novel randomised language model which uses an online perfect hash function to efficiently deal with unbounded text streams. Translation experiments over a text stream show that our online randomised model matches the performance of batch-based LMs without incurring the computational overhead associated with full retraining. This opens up the possibility of randomised language models which continuously adapt to the massive volumes of texts published on the Web each day. 1

One of the most important causes of failure in spoken dialogue systems is usually neglected: the problem of words that are not covered by the system's vocabulary (out-of-vocabulary or OOV words). In this paper a methodology is described for the detection, classification and processing of OOV words in an automatic train timetable information system [2]. The various extensions that had to be effected on the different modules of the system are reported, resulting in the design of appropriate dialogue strategies, as are encouraging evaluation results on the new versions of the word recogniser and the linguistic processor. 1. INTRODUCTION The majority of speech understanding systems have to face the problem of words that are not covered by their current lexicon, i.e. OOV words. In such a case the word recogniser usually recognises one or more different words with a similar acoustic profile to the unknown. These misrecognitions often result in possibly irreparable misunderstandings between ...

Generic speech recognition systems typically use language models that are trained to cope with a b variety of input. However, many recognition applications are more constrained, often to a specific or domain. In cases such as these, a knowledge of the particular topic can be used to advantage. report describes the development of a number of techniques for augmenting domain-specific lang models with dam from a more general source.

Standard statistical language models, or n-gram models, which represent the probability of word sequences, suffer from sparse-data problems in tasks where large amounts of domain-specific text are not available. This thesis focuses on improving the estimation of domain-dependent n-gram models by using out-of-domain text data. Previous approaches for estimating language models from multi-domain data have not accounted for the characteristic variations of style and content across domains. In contrast, this thesis introduces two approaches that compensate for multi-domain differences, both representing "style" by part-of-speech (POS) sequences and "content" by the particular choice of words. First, data from multiple domains is combined using similarity weighting schemes that discriminate for content and style relevance prior to pooling multi-domain text. Second, n-gram distributions from multiple domains are combined, via a POS-dependent n-gram framework that separately compensate for word and POS usage differences. Two variations are explored: explicitly transforming the out-of-domain distribution before combining with an in-domain model, and vi separately estimating components of the POS-dependent n-gram model using multidomain data. Finally, measures to analyze and predict recognition performance of language models are also investigated, resulting in an algorithm for predicting performance differences associated with localized changes in language models given a recognition system.

Collecting sufficient language model training data for good speech recognition performance in a new domain is often difficult. However, there may be other sources of data that are matched in terms of topic or style, if not both. This paper looks at the use of text normalization tools to make these data more suitable for language model training, in conjunction with mixture models to combine data from different sources. We specifically address the task of recognizing meeting speech, showing a small reduction in word error rate over a baseline language model trained from conversational speech data. 1.