Machine translation using global lexical selection and sentence reconstruction

Patent metadata

Field	Value
Publication number	US-9323745-B2
Application number	US-201414336297-A
Country	US
Kind code	B2
Filing date	Jul 21, 2014
Priority date	Mar 15, 2007
Publication date	Apr 26, 2016
Grant date	Apr 26, 2016

Abstract

Disclosed are systems, methods, and computer-readable media for performing translations from a source language to a target language. The method comprises receiving a source phrase, generating a target bag of words based on a global lexical selection of words that loosely couples the source words/phrases and target words/phrases, and reconstructing a target phrase or sentence by considering all permutations of words with a conditional probability greater than a threshold.

First claim

We claim:

1. A method comprising:
classifying, via a processor, a source phrase in a source language into a phrase meaning;
matching, via the processor, the phrase meaning to a target phrase automaton in a target language wherein:
the target phrase automaton comprises a plurality of states and a plurality of arcs interconnecting the plurality of states to define a sentence path, each of the plurality of states defining a word position slot for inserting a word from a bag of words, and each of the plurality of arcs associated with a pre-defined insertion cost for inserting the word from the bag of words based on a previous state in the sentence path; and
the pre-defined insertion cost is associated with one of penalizing and rewarding the sentence path based on how many words are in the sentence path, in order to produce more words in a target sentence relative to the source phrase;

determining, via the processor, a target sentence probability for the sentence path based on one of a lexical translation of words in the source phrase and a phrase-to-phrase mapping; and
upon determining that the sentence path has a probability above a threshold, constructing, via the processor, the target sentence using the sentence path.

2. The method of claim 1 , wherein the target sentence probability is further based on a target word possibility for each word position, the target word possibility for each word position weighted by a target language model.

3. The method of claim 2 , wherein the target word possibility for each word position is detected independently.

4. The method of claim 2 , wherein the target word possibility for each word does not use information about previous words and subsequent words.

5. The method of claim 1 , further comprising:
adjusting a length of the target sentence by adding optional deletions when constructing the target sentence.

6. The method of claim 1 , wherein function words in the target sentence possibility serve as attributes on contentful lexical items.

7. The method of claim 6 , wherein the attributes are one of definiteness, tenses and case.

8. A system comprising:
a processor; and
a computer-readable storage medium having instructions stored which, when executed by the processor, cause the processor to perform operations comprising:
classifying a source phrase in a source language into a phrase meaning;
matching the phrase meaning to a target phrase automaton in a target language wherein:
the target phrase automaton comprising a plurality of states and a plurality of arcs interconnecting the plurality of states to define a sentence path, each of the plurality of states defining a word position slot for inserting a word from a bag of words, and each of the plurality of arcs associated with a pre-defined insertion cost for inserting the word from the bag of words based on a previous state in the sentence path; and
the pre-defined insertion cost is associated with one of penalizing and rewarding the sentence path based on how many words are in the sentence path, in order to produce more words in a target sentence relative to the source phrase;

determining a target sentence probability for the sentence path based on one of a lexical translation of words in the source phrase and a phrase-to-phrase mapping; and
upon determining that the sentence path has a probability above a threshold, constructing the target sentence using the sentence path.

9. The system of claim 8 , wherein the target sentence probability is further based on a target word possibility for each word position, the target word possibility for each word position weighted by a target language model.

10. The system of claim 9 , wherein the target word possibility for each word position is detected independently.

11. The system of claim 9 , wherein the target word possibility for each word does not use information about previous words and subsequent words.

12. The system of claim 8 , the computer-readable storage medium having additional instructions stored which result in operations comprising:
adjusting a length of the target sentence by adding optional deletions when constructing the target sentence.

13. The system of claim 8 , wherein function words in the target sentence possibility serve as attributes on contentful lexical items.

14. The system of claim 13 , wherein the attributes are one of definiteness, tenses and case.

15. A computer-readable storage device having instructions stored which, when executed by a computing device, cause the computing device to perform operations comprising:
classifying a source phrase in a source language into a phrase meaning;
matching the phrase meaning to a target phrase automaton in a target language wherein:
the target phrase automaton comprising a plurality of states and a plurality of arcs interconnecting the plurality of states to define a sentence path, each of the plurality of states defining a word position slot for inserting a word from a bag of words, and each of the plurality of arcs associated with a pre-defined insertion cost for inserting the word from the bag of words based on a previous state in the sentence path; and
the pre-defined insertion cost is associated with one of penalizing and rewarding the sentence path based on how many words are in the sentence path, in order to produce more words in a target sentence relative to the source phrase;

determining a target sentence probability for the sentence path based on one of a lexical translation of words in the source phrase and a phrase-to-phrase mapping; and
upon determining that the sentence path has a probability above a threshold, constructing the target sentence using the sentence path.

16. The computer-readable storage device of claim 8 , wherein the target sentence probability is further based on a target word possibility for each word position, the target word possibility for each word position weighted by a target language model.

17. The computer-readable storage device of claim 9 , wherein the target word possibility for each word position is detected independently.

18. The computer-readable storage device of claim 9 , wherein the target word possibility for each word does not use information about previous words and subsequent words.

19. The computer-readable storage device of claim 8 , having additional instructions stored which result in operations comprising:
adjusting a length of the target sentence by adding optional deletions when constructing the target sentence.

20. The computer-readable storage device of claim 8 , wherein function words in the target sentence possibility serve as attributes on contentful lexical items.

Assignees

• At & T Ip Ii Lp

Inventors

• Bangalore Srinivas
• Haffner Patrick
• Kanthak Stephan

Classifications

• G06F40/44 Primary

  Statistical methods, e.g. probability models · CPC title

• H04N21/2662

  Controlling the complexity of the video stream, e.g. by scaling the resolution or bitrate of the video stream based on the client capabilities · CPC title

• H04N21/235

  Processing of additional data, e.g. scrambling of additional data or processing content descriptors · CPC title

• H04N21/25825

  involving client display capabilities, e.g. screen resolution of a mobile phone (optimising the visualisation of content during browsing in the Internet G06F16/9577; processing of terminal status or physical abilities in wireless networks H04W8/22; authentication in wireless network security H04W12/06) · CPC title

• G06F40/45

  Example-based machine translation; Alignment · CPC title