At+t Corp.

At&t Intellectual Property I, LP

Hiyan Alshawi

Shona Douglas

Mazin G Rahim

Narendra Kumar Gupta

Robert Elias Schapire

Yoram Singer

David Arthur Berkley

Giuseppe DiFabrizzio

Giuseppe DiFabbrizio

Embodiments of the present invention relate to a method and system for augmenting a training database of an automated language-understanding system. In one embodiment, a training example in a first language is received from the training database. The first language-training example is translated to a second language output. The second language output is translated to a first variant of the first language-training example. An action pair including the first variant of the first language-training example and an action command associated with the first language-training example is stored in an augmented training database.

Language-understanding training database action pair augmentation using bidirectional translation

Embodiments of the invention relate to a system and method for providing speech-enabled application programs. The speech-enabled programs automatically execute requests input by users. One or more natural language variants may be mapped with at least one natural language exemplar. The natural language exemplar may correspond to a typical way to express a request relevant to the speech-enabled application program. The natural language variant may correspond to an alternative way of expressing the same request. A recognized input string is received and a prospective variant that most resembles the received recognized input string is selected from the natural language variants. The natural language exemplar mapped to the prospective variant is identified. An action instruction associated with the identified natural language exemplar is executed to fulfill the user's request.

Method and apparatus for creation and user-customization of speech-enabled services

Embodiments of the present invention relate to a method and system for augmenting a training database of an automated language-understanding system. In one embodiment, a training example in a first language is received from the training database. The first language-training example translated to a second language output. The second language output is translated to a first variant of the first language-training example. An action pair including the first variant of the first language-training example and an action command associated with the first language-training example is stored in an augmented training database.

A system for understanding entries, such as speech, develops a classifier by employing prior knowledge with which a given corpus of training entries is enlarged threefold. A rule is created for each of the labels employed in the classifier, and the created rules are applied to the given corpus to create a corpus of attachments by appending a weight of &#x3b7;p(x), or 1&#x2212;&#x3b7;p(x), to labels of entries that meet, or fail to meet, respectively, conditions of the labels' rules, and to also create a corpus of non-attachments by appending a weight of 1&#x2212;&#x3b7;p(x), or &#x3b7;p(x), to labels of entries that meet, or fail to meet conditions of the labels' rules.

Spoken language understanding that incorporates prior knowledge into boosting

Embodiments of the present invention relate to a method and system for augmenting a training database of an automated language-understanding system. In one embodiment, a training example in a first language may be received from the training database. The first language-training example may be translated to a second language output. The second language output may be translated to a first variant of the first language-training example. An action pair including the first variant of the first language-training example and an action command associated with the first language-training example may be stored in an augmented training database.

Language-understanding systems employing machine translation components

A system for understanding entries, such as speech, develops a classifier by employing prior knowledge with which a given corpus of training entries is enlarged threefold. A rule is created for each of the labels employed in the classifyier, and the created rules are applied to the given corpus to create a corpus of attachments by appending a weight of &#x3b7;p(x), or 1&#x2212;&#x3b7;p(x), to labels of entries that meet, or fail to meet, respectively, conditions of the labels' rules, and to also create a corpus of non-attachments by appending a weight of 1&#x2212;&#x3b7;p(x), or &#x3b7;p(x), to labels of entries that meet, or fail to meet conditions of the labels' rules.

Methods and systems for language translation are disclosed. The translator is based on finite state machines that can convert a pair of input symbol sequences to a pair of output symbol sequences. The translator includes a lexicon associating a finite state machine with a pair of head words with corresponding meanings in the source and target languages. The state machine for a source language head word w and a target language head word &ngr; reads the dependent words of w to its left and right in a source sentence and proposes corresponding dependents to the left and right of &ngr; in a target language sentence being constructed, taking account of the required word order for the target language. The state machines are used by a transduction search engine to generate a plurality of candidate translations via a recursive process wherein, a source language head word is first translated as described above, and then the heads of each of the dependent phrases are similarly translated, and then their dependents and so on. Only the state machines corresponding to the words in the source language string are activated and used by the search engine. The translator also includes a parameter table that provides costs for actions taken by each finite state machine in converting between the source language and the target language. The costs for machine transitions are indicative of the likelihood of co-occurence of pairs of words in the source language, and between corresponding pairs of words in the target language. The transduction search engine provides a total cost, using the parameter table, for each of the candidate translations. The total cost of a translation is the sum of the cost for all actions taken by each machine involved in the translation.

Method and apparatus for language translation

A method and apparatus for automatically constructing hierarchical transduction models for language translation is presented. The input to the construction process may be a database of examples each consisting of a transcribed speech utterance and its translation into another language. A translation pairing score is assigned (or computed) for translating a word in the source language into each of the possible translations it has in the target language. For each instance of the resulting training dataset, a head transducer may be constructed that translates the source string into the target string by splitting the source string into a source head word, the words preceding the source head word, and the words following the source head word. This process may be performed recursively to generate a set of transducer fragments. The transducer fragments may form a statistical head transducer model. The head transducer translation model may then be input into a transduction search module.

Method and apparatus for automatic construction of hierarchical transduction models for language translation

A method for the treatment of unrecognized words and phrases in spoken language translation systems is applicable to open-ended word or phrase categories, such as proper names, and involves the demarcation of one or more unrecognized segments from the input speech signal and splicing these segments into the appropriate position of the synthesized speech signal for the translation.

Growing community of inventors

Madison, NJ, United States of America

Hiyan Alshawi