Machine translation system

Abstract

Abstract A machine translation system which operates by reading a text from an input source, then entering that text into an analysis module, which processes the text by drawing on information from source language grammars, dictionaries and other knowledge sources provided in the system to deduce information about the usage and meaning of the words of the source language text as they occur in context. This information is collected in repositories which are associated with the individual morphemes of the input text. The information repositories which have been constructed and stored in the buffer of the analysis module are then passed to an equivalencing module, which matches them against the source language sides of a set of bilingual equivalence statements provided within the system. The matching of the source language side of the equivalence statements causes the equivalence module to produce as output a set of information repositories describing words of the target language. The target language information repositories are then passed to a generation module which will construct either a single target language text or a plurality of target language texts, such that these texts are possible translations of the source language text. The generation module has internal structure, consisting of a constraint checking module and a tree construction module. In the preferred embodiment the target language information repositories are simultaneously fed to both the constraint checker and the tree building module, which operate in parallel to produce a set of candidate translations of the original text. The translated texts are then made available to the user by conveying them to an accessible location, such as a screen, a printer or a second computer file, and optionally by initiating a structured interaction during which the user can explore the alternative translations which the system has produced.

Inventors

• 👤 Christopher Hardie Brew

Bibliographic Details