Language Weaver
From Wikipedia, the free encyclopedia
Language Weaver is a Los Angeles, California based company that was founded by USC's Kevin Knight and Daniel Marcu, marketing a new software product that learns a language by itself by reading a lot of text (extracting probabilistic translation dictionaries, patterns, and rules), and therefore is able to translate it.
It is hoped that one day such software will be able to find grammatical rules in languages that humans have not been able to pick up on. The program uses probability and cryptographic approach to translate a language. One limitation to the program is its requirement of computing power. Language Weaver products currently can translate fifteen languages, mostly 'large' languages such as English, Arabic, Chinese, French, and Spanish, among others. It may be argued that such programs would be better for translating little-known languages or highly technical documents. The statistical approach employed by Language Weaver requires large amounts of previously translated text in order to develop the system for a new language. This makes developing systems for lesser-known languages more of a challenge. Their customization server product allows users to modify the translation product's behavior to better handle highly technical and other specialized domains.
Machine translation based on cryptography was the idea of Warren Weaver, who sent a memo to 250 scientists working in the cryptography field in 1947, where he suggested solving translation as a decoding problem.
