Deep model statistics method for machine translation
AssigneeABBYY InfoPoisk LLC
InventorsKonstantin Anisimovich, Diar Tuganbaev, Vladimir Selegey, Konstantin Zuev
Life of this patent
Estimating the term…
Abstract
In one embodiment, the invention provides a method for machine translation of a source document in an input language to a target document in an output language, comprising generating translation options corresponding to at least portions of each sentence in the input language; and selecting a translation option for the sentence based on statistics associated with the translation options.
Claims33
A computer method of representing in a computer memory a natural language sentence in a language-independent format, the method comprising:
- receiving one or more ratings generated based on analysis of sentences from a corpus of texts comprising parallel texts, wherein the parallel texts comprising texts in two or more languages in which each fragment in one language having a corresponding fragment in another language;
- performing by a computer processor a syntactic analysis of a natural language sentence including selecting a syntactic structure for the natural language sentence based on the one or more ratings;
- performing a semantic analysis on the natural language sentence;
- storing in the computer memory the selected syntactic structure in a language-independent semantic structure;
- and
- selecting a semantic structure based on the one or more ratings.
A computer system for representing a natural language sentence in a language-independent format, the computer system comprising:
- a memory;
- a processor;
- and
- a storage medium electrically coupled to the processor and the memory, the storage medium storing instructions which when executed by the processor cause the computer system to:
- receiving one or more ratings generated based on analysis of sentences from a corpus of texts comprising parallel texts, wherein the parallel texts comprising texts in two or more languages in which each fragment in one language having a corresponding fragment in another language;
- perform a syntactic analysis of a natural language sentence;
- select a syntactic structure based on the one or more ratings;
- and
- perform a semantic analysis so as to transition the selected syntactic structure into a language-independent semantic structure, including selecting a semantic structure based on the one or more ratings.
A non-transitory computer storage medium having stored thereon a sequence of instructions which, when executed by a computer system, cause said computer system to perform a method for representing in a computer memory a natural language sentence in a language-independent format, the instructions comprising:
- receiving one or more ratings generated based on analysis of sentences from a corpus of texts comprising parallel texts, wherein the parallel texts comprising texts in two or more languages in which each fragment in one language having a corresponding fragment in another language;
- performing by the computer system a syntactic analysis of a natural language sentence including selecting a syntactic structure for the natural language sentence based on the one or more ratings;
- performing a semantic analysis on the natural language sentence;
- storing in a computer memory the selected syntactic structure in a language-independent semantic structure;
- and
- selecting a semantic structure based on the one or more ratings.
Dependencies are inferred from the claim text.