INTELLIGENCE SOLUTIONS FOR GOVERNMENT FROM CARAHSOFT ^TM

Statistical Machine Translation Software

Language Weaver's primary product, Statistical Machine Translation Software (SMTS), now in its version number 4.1, has undergone considerable growth in capabilities and accuracy since statistical methodologies for translation were first introduced four years ago.

SMTS v4.1 saves time and money for customers while increasing translation productivity and ease-of-use. Unlike linguistic rule-based methodology that tries to correlate grammar and syntax rules of one language with that of another, Language Weaver uses statistical probabilities to assess all of the possible, and identify the most probable, translation for any given text.

Using the best standard code (Unicode UTF-8), this translation engine can translate 5,000 words per minute per CPU and is scalable to 500,000 words per minute on a high throughput network, improving users' abilities to process millions of pages of foreign language information in record time. It is easy for both users and integrators to use, having a simple graphical user interface and well-documented Web-services/SOAP API, with sample code and documents.

The translation engines are built using a parallel corpus that has been derived from several domains. The systems will perform optimally on topics and domains they have been trained on.

However, Language Weaver has the ability to easily and very quickly retrain our systems if new parallel data is available. This process can continue to improve the quality of the output if the engines are trained on specific domain material. High quality, domain specific material enhances the accuracy and fluency or readability of the translations dramatically.

Generally it is true that the more parallel corpus we have available for training and for subsequent refinement, the better our translations become, being able to incorporate style and idioms that may be particular to your industry or company. This is a claim that cannot be made by rule-based systems. The most successful customer applications are those where the customer contributes data to the learning system in order to modify it for their particular focus. In essence, you receive your own language pairs that cannot be found elsewhere and that will do the best job of translating your material.