International Business Machines Corporation

Michael Campbell McCord

Arendse Bernth

Sue Ann Medeiros

Andrew Tsuneo Tanabe

David A Walters

Kara C Warburton

Claudia Maria Gdaniec

An improved solution for extracting and identifying key terms in content is provided. In the solution, one or more documents can be obtained and one or more candidate terms can be obtained from the content of each document. Subsequently, one or more exclusion conditions can be used to filter the candidate terms thereby generating a set of key terms. In particular, all common terms and/or all near duplicate candidate terms can be excluded from the set of key terms. The set of key terms can be used in translating the content, generating a glossary/index for the content, detecting and/or correcting incorrect terms and usages of terms in the content, building a terminology repository, and/or the like. In one embodiment, the key term extraction is performed to facilitate the translation of product documentation into a second language in order to release the product in one or more nations. In this case, candidate terms that have already been translated into the second language can also be excluded from the key terms.

Key term extraction

A computer system and method for natural language translation uses a translation process to translate a source natural language segment (e.g. English) of one or more source words/elements into a target natural language (e.g. German) segment of one or more target words/elements. An evaluation module determines a confidence measure of the natural language translation. Typically, the confidence measure indicates less confidence as the complexity of the translation increases. Various novel features for determining complexity and confidence measure at different steps in the translation are used. The translation process can be terminated if the confidence measure fails to meet a threshold criterion.

System and method for estimating accuracy of an automatic natural language translation

A computer-implemented method and system are provided for English text analysis. A file of a plurality of English sentences is received and submitted to a parser. A unified parse structure is generated by the parser for predefined ones of the English sentences. A plurality of predetermined critiquing rules are applied to each of the unified parse structures for identifying exceptions to recommended English. Identified exceptions together with related information are stored in an exception file. Then an interactive session with a user is performed utilizing the stored exception file.

Growing community of inventors

Ossining, NY, United States of America

Michael Campbell McCord