Nuance Communications Austria Gmbh

U.S. Philips Corporation

Koninklijke Philips Corporation N.v.

Nuance Communications, Inc.

Deutsches Zentrum Fur Luft-Und Raumfahrt E.v.

Universität Des Saarlandes

Dietrich Klakow

Jochen Peters

Carsten Meyer

Evgeny Matusov

Reinhard Kneser

Volker Steinbiss

Juergen Rataj

Hartmut Helmke

The invention relates to a method, a computer program product, a segmentation system and a user interface for structuring an unstructured text by making use of statistical models trained on annotated training data. The method performs text segmentation into text sections and assigns labels to text sections as section headings. The performed segmentation and assignment is provided to a user for general review. Additionally, alternative segmentations and label assignments are provided to the user being capable to select alternative segmentations and alternative labels as well as to enter a user defined segmentation and user defined label. In response to the modifications introduced by the user, a plurality of different actions are initiated incorporating the re-segmentation and re-labeling of successive parts of the document or the entire document.

Text segmentation and label assignment with user interaction by means of topic specific language models, and topic-specific label statistics

Text segmentation and label assignment with user interaction by means of topic specific language models and topic-specific label statistics

The invention relates to a method, a computer program product, a segmentation system and a user interface for structuring an unstructured text by making use of statistical models trained on annotated training data. The method performs text segmentation into text sections and assigns labels to text sections as section headings. The performed segmentation and assignment is provided to a user for general review. Additionally, alternative segmentations and label assignments are provided to the user being capable to select alternative segmentations and alternative labels as well as to enter a user defined segmentation and user defined label. In response to the modifications introduced by the user, a plurality of different actions are initiated incorporating the re-segmentation and re-labelling of successive parts of the document or the entire document. Furthermore the method comprises a learning functionality, logging and analyzing user introduced modifications for adaptation of user's preferences and for further training of the statistical models.

The present invention relates to a method, a computer system and a computer program product for speech recognition and/or text formatting by making use of topic specific statistical models. A text document which may be obtained from a first speech recognition pass is subject to segmentation and to an assignment of topic specific models for each obtained section. Each model of the set of models provides statistic information about language model probabilities, about text processing or formatting rules, as e.g. the interpretation of commands for punctuation, formatting, text highlighting or of ambiguous text portions requiring specific formatting, as well as a specific vocabulary being characteristic for each section of the recognized text. Furthermore, other properties of a speech recognition and/or formatting system (such as e.g. settings for the speaking rate) may be encoded in the statistical models. The models themselves are generated on the basis of annotated training data and/or by manual coding. Based on the assignment of models to sections of text an improved speech recognition and/or text formatting procedure is performed.

Topic specific models for text formatting and speech recognition

A small vocabulary pattern recognition system is used for recognizing a sequence of words, such as a sequence of digits (e.g. telephone number) or a sequence of commands. A representation of reference words is stored in a vocabulary . Input means are used for receiving a time-sequential input pattern representative of a spoken or written word sequence. A pattern recognizer comprises a word-level matching unit for generating a plurality of possible sequences of words by statistically comparing the input pattern to the representations of the reference words of the vocabulary . A cache is used for storing a plurality of most recently recognized words. A sequence-level matching unit selects a word sequence from the plurality of sequences of words in dependence on a statistical language model which provides a probability of a sequence of M words, M&gE;2. The probability depends on a frequency of occurrence of the sequence in the cache. In this way for many small vocabulary systems where no reliable data is available on frequency of use of word sequences, the cache is used to provide data representative of the actual use.

Language model based on the speech recognition history

The invention relates to a method of estimating probabilities of occurrence of speech vocabulary elements in a speech recognition system. By modification of the linguistic speech modeling, further alternatives for reducing the error rate and perplexity of a speech recognition system are proposed. The method according to the invention is characterized in that, in the estimation of a probability of occurrence of a speech vocabulary element, several M-gram probabilities of this element are raised to a higher power by means of an M-gram-specific optimized parameter value, and the powers thus obtained are multiplied by each other, in which the estimation of the probability of occurrence of a speech vocabulary element does not include the case where an M-gram probability with M>1 estimated by means of a first training vocabulary corpus for the speech vocabulary element is multiplied by a quotient raised to a higher power by means of an optimized parameter value, which optimized parameter value is determined by means of the GIS algorithm, and a unigram probability of the element estimated by means of a second training vocabulary corpus serves as a dividend of the quotient, and a unigram probability of the element estimated by means of the first training vocabulary corpus serves as a divisor of the quotient.

Method of estimating probabilities of occurrence of speech vocabulary elements

Language models which take into account the probabilities of word sequences are used in speech recognition, in particular in the recognition of fluently spoken language with a wide vocabulary, in order to increase the recognition reliability. These models are obtained from comparatively large quantities of text and accordingly represent values which were averaged over several texts. This means, however, that the language model is not well adapted to peculiarities of a special text. To achieve such an adaptation of a given language model to a special text on the basis of only a short text fragment, according to the invention, it is suggested that first the unigram language model is adapted with the short text and, in dependence thereon, the M-gram language model is subsequently adapted. A method is described for adapting the unigram language model values which automatically carries out a subdivision of the words into semantic classes.

Growing community of inventors

Saarbrücken, Germany

Dietrich Klakow