Abstract: A system and method for utilizing distance measures to perform text classification includes text classification categories that each have reference models of reference N-grams. Input text that includes input N-grams is accessed for performing the text classification. A text classifier calculates distance measures between the input N-grams and the reference N-grams. The text classifier then utilizes the distance measures to identify a matching category for the input text. In certain embodiments, a verification module performs a verification procedure to determine whether the initially-selected matching category is a valid classification result for the text classification.

Abstract: A system and method for effectively implementing an optimized language model for speech recognition includes initial language models each created by combining source models according to selectable interpolation coefficients that define proportional relationships for combining the source models. A rescoring module iteratively utilizes the initial language models to process input development data for calculating word-error rates that each correspond to a different one of the initial language models. An optimized language model is then selected from the initial language models by identifying an optimal word-error rate from among the foregoing word-error rates. The speech recognizer may then utilize the optimized language model for effectively performing various speech recognition procedures.

Abstract: A system and method for automatically cataloguing data by utilizing speech recognition procedures includes an electronic device that captures audio/video data and corresponding verbal narration. A speech recognition engine coupled to the electronic device automatically performs a speech recognition process upon the audio/video data and verbal narration to generate labels that correspond to respective subject matter locations in the audio/video data. A label manager of the electronic device manages a label mode for generating and storing the foregoing labels. The label manager also controls a label search mode during which a system user utilizes the labels to automatically locate corresponding subject matter locations in the captured audio/video data.

Abstract: The present invention comprises a system and method for effectively implementing a Mandarin Chinese speech recognition dictionary, and may include a recognizer configured to compare input speech data to phone strings from a vocabulary dictionary that is implemented according to an optimized Mandarin Chinese phone set. The optimized Mandarin Chinese phone set may efficiently be implemented by utilizing an allophone and phonemic variation technique. In addition, the foregoing vocabulary dictionary may be implemented by utilizing unified dictionary optimization techniques to provide robust and accurate speech recognition. Furthermore, the vocabulary dictionary may be implemented as an optimized dictionary to accurately recognize either Northern Mandarin Chinese speech or Southern Mandarin Chinese speech during the speech recognition procedure.

Abstract: A method and system for reducing lexical ambiguity in an input stream are described. In one embodiment, the input stream is broken into tokens. The tokens are used to create a connection graph comprising a number of paths. Each of the paths is assigned a cost. At least one best path is defined based upon a corresponding cost to generate an output graph. The generated output graph is provided to reduce lexical ambiguity.

Abstract: A natural language translation system contains language-neutral modules for syntactic analysis, transfer, and morphological and syntactical generation of feature structures for an input expression in a source and a target language. The language-neutral modules are driven by language-specific grammars to translate between the specified languages so that no knowledge about the languages need be incorporated into the modules themselves. The modules interface with the grammar rules in the form of compiled grammar programming language statements that perform the required manipulation of the feature structures. Because the modules are language-neutral, the system is readily adaptable to new languages simply by providing a grammar for the new language. Multiple copies of each module, each interfacing with a different natural language grammar, enables simultaneous translation of multiple languages in the same system.

Abstract: A method and system for reducing lexical ambiguity in an input stream are described. In one embodiment, the input stream is broken into tokens. The tokens are used to create a connection graph comprising a number of paths. Each of the paths is assigned a cost. At least one best path is defined based upon a corresponding cost to generate an output graph. The generated output graph is provided to reduce lexical ambiguity.

Abstract: The present invention comprises a system and method for implementing a Mandarin Chinese speech recognizer with an optimized phone set, and may include a recognizer configured to compare input speech data to phone strings from a vocabulary dictionary that is implemented according to an optimized Mandarin Chinese phone set. The optimized Mandarin Chinese phone set may be implemented with a phonetic technique to separately include consonantal phones and vocalic phones. For reasons of system efficiency, the optimized Mandarin Chinese phone set may preferably be implemented in a compact manner to include only a minimum required number of consonantal phones and vocalic phones to accurately represent Mandarin Chinese speech during the speech recognition procedure.

Abstract: A grammar programming language (“GPL”) compiler compiles each rule in a natural language grammar into a separate function that can be invoked by a translation system to apply the rule to a representation of a natural language expression. The GPL compiler can output the functions for the rules as source code for a standard computer programming language to be further compiled into object code that can be directly executed by a computer processor. The GPL compiler can also generate special functions for each rule to enable multi-layered operations on the representations and to handle the processing of representations of ambiguous expressions.

Abstract: A natural language translation system contains language-neutral modules for syntactic analysis, transfer, and morphological and syntactical generation of feature structures for an input expression in a source and a target language. The language-neutral modules are driven by language-specific grammars to translate between the specified languages so that no knowledge about the languages need be incorporated into the modules themselves. The modules interface with the grammar rules in the form of compiled grammar programming language statements that perform the required manipulation of the feature structures. Because the modules are language-neutral, the system is readily adaptable to new languages simply by providing a grammar for the new language. Multiple copies of each module, each interfacing with a different natural language grammar, enables simultaneous translation of multiple languages in the same system.

Abstract: A method and an apparatus for performing spoken language translation are provided, wherein a speech input is received comprising at least one source language. The speech input comprises words, sentences, and phrases in a natural spoken language. Source expressions are recognized in the source language. Misrecognitions of the source expressions resulting from factors comprising noise and speaker variation are minimized by the generation of intermediate data structures that encode at least one recognition hypothesis. Furthermore, misrecognitions are minimized by the generation of candidate recognized source expressions by processing the intermediate data structures using models comprising a general language model and a domain model. A recognized source expression is selected and confirmed by a user through a user interface. The recognized source expressions are translated from the source language to a target language, and a speech output is synthesized from the translated target language source expressions.

Abstract: At least one speech input is received and at least one token is generated from speech input. Morphemes of the tokens are reduced to at least one feature. Furthermore, an inflection type of the token is identified. At least one dictionary is searched for entries comprising features that match the features reduced from the morphemes. At least one lexical feature structure is generated for the token by inserting at least one morphological feature associated with the inflection type into the entry feature. An output is provided comprising at least one lexical feature structure.

Abstract: A method and an apparatus for performing spoken language translation are provided, wherein a speech input is received comprising at least one source language. The speech input comprises words, sentences, and phrases in a natural spoken language. Source expressions are recognized in the source language. Misrecognitions of the source expressions resulting from factors comprising noise and speaker variation are minimized by the generation of intermediate data structures that encode at least one recognition hypothesis. Furthermore, misrecognitions are minimized by the generation of candidate recognized source expressions by processing the intermediate data structures using models comprising a general language model and a domain model. A recognized source expression is selected and confirmed by a user through a user interface. The recognized source expressions are translated from the source language to a target language, and a speech output is synthesized from the translated target language source expressions.

Abstract: A method and apparatus for parsing in a spoken language translation system are provided, wherein an input is received comprising at least one input sentence or expression. A parsing table is accessed and consulted for a next action, wherein the parser looks up in the next action in the parsing table. During parsing operations, the parser may perform shift actions and reduce actions. In performing a shift action, a next item of the input string is shifted onto a stack or intermediate data structure of the parser. A new parse node is generated, and a feature structure or lexical feature structure of the shifted input item is obtained from a morphological analyzer and associated with the new parse node. The new node is placed on the stack or intermediate data structure. In performing a reduce action, a grammar rule and an associated compiled feature structure manipulation are applied.