MACHINE TRANSLATION-DRIVEN AUTHORING SYSTEM AND METHOD

• US 20140207439A1
• Filed: 01/21/2013
• Published: 07/24/2014
• Est. Priority Date: 01/21/2013
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Venkatapathy, Sriram, Mirkin, Shachar
• Granted Patent: US 9,047,274 B2
• Time in Patent Office: Days
• US Class Current: 704/4
• CPC Class Codes: G06F 40/274 Converting codes to words; Guess-ahead of partial word inputs
• CPC Class Codes: G06F 40/42 Data-driven translation
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/47 Machine-assisted translation, e.g. using translation memory

Abstract

An authoring method includes generating an authoring interface configured for assisting a user to author a text string in a source language for translation to a target string in a target language. Initial source text entered by the user is received through the authoring interface. Source phrases are selected that each include at least one token of the initial source text as a prefix and at least one other token as a suffix. The source phrase selection is based on a translatability score and optionally on fluency and semantic relatedness scores. A set of candidate phrases is proposed for display on the authoring interface, each of the candidate phases being the suffix of a respective one of the selected source phrases. The user may select one of the candidate phrases, which is appended to the source text following its corresponding prefix, or may enter alternative text. The process may be repeated until the user is satisfied with the source text and the SMT model can then be used for its translation.

Claims

1. An authoring method comprising:

generating an authoring interface configured for assisting a user to author a text string in a source language for translation to a target string in a target language;

receiving source text comprising initial text entered by a user through the authoring interface in the source language;

with a processor, selecting a set of source phrases from a stored collection of source phrases, each of the set of source phrases including at least one token of the initial source text as a prefix and at least one other token as a suffix, the selection of the set of source phrases being based on a translatability score, wherein the translatability score, for each of a stored set of source phrases, is a function of statistics of at least one biphrase in which the source phrase occurs in combination with a corresponding target phrase in the target language, each biphrase being one of a collection of biphrases;

proposing a set of candidate phrases for display on the interface, the candidate phases each comprising the suffix of a respective one of the source phrases in the set of source phrases;

providing for receiving a user'"'"'s selection of one of the candidate phrases in the set of candidate phrases and, where one of the candidate phrases in the set of candidate phrases is selected by the user, appending the selected one of the candidate phrases to the source text; and

optionally, repeating the receiving, selecting, proposing and providing to generate the text string in the source language, wherein in the repeating, the received text comprises the initial source text and previously appended text.

2. The method of claim 1, further comprising translating the text string with a machine translation system using a collection of biphrases.

3. The method of claim 2, wherein the stored collection of source phrases is derived from the same collection of biphrases used in translation of the text string.

4. The method of claim 1, wherein for a source phrase which occurs in a plurality of biphrases in the collection, the translatability score is a maximum one of translation scores over all the biphrases in which the source phrase occurs.

5. The method of claim 1, wherein the biphrase statistics are derived from a parallel corpus of sentences in the source and target languages.

6. The method of claim 1, wherein the biphrase statistics are selected from phrasal probabilities, lexical probabilities, a phrase penalty, and combinations thereof.

7. The method of claim 1, wherein the translatability score is computed as a weighted function of a plurality of the statistics.

8. The method of claim 1, wherein the translatability score of a source phrase s is computed a function of H(T|s) and p(s), where H(T|s) is the conditional entropy of a set T of target phrases t stored in biphrases as translations of the source phrase s, the conditional entropy being computed as an aggregate, over all target phrases in the set T, of a function of p(t|s), where p(t|s) is the probability of finding target phase t given s in a parallel corpus, and p(s) is computed as the relative frequency of s in a source side of the parallel corpus.

9. The method of claim 1, further comprising providing for receiving a user'"'"'s selection of text other than one of the candidate phrases in the set of candidate phrases and appending the other text to the source text.

10. The method of claim 1, wherein the selection of the set of source phrases is also based on at least one feature other than the translatability score, the at least one other feature being computed based on the input text.

11. The method of claim 10, wherein the selection of the set of source phrases includes computing a function of the translatability score and the at least one other feature.

12. The method of claim 10, wherein the at least one other feature is selected from the group consisting of a fluency score and a semantic relatedness score.

13. The method of claim 1, further comprising:

providing for the user to select at least one of the source and target languages; and

providing a plurality of the source phrase collections, the plurality of source phrase collections comprising one source phrase collection for each of a set of source and target language pairs.

14. A computer program product comprising a non-transitory medium which stores instructions, which when executed by a processor, perform the method of claim 1.

15. An authoring system comprising memory which stores instructions for performing the method of claim 1 and a processor in communication with the memory for executing the instructions.

16. An authoring system comprising:

an interface generator for generating an authoring interface on an associated display device for assisting a user to author a text string in a source language for translation to a target string in a target language;

a suggestion system for receiving source text comprising initial text entered by a user through the authoring interface in the source language and for selecting a set of source phrases from a stored collection of source phrases, each of the set of source phrases including at least one token of the initial source text as a prefix and at least one other token as a suffix, the selection of the set being based on a translatability score which, for each of a stored set of source phrases, is a function of statistics of at least one biphrase of a collection of biphrases, in which biphrase the source phrase occurs in combination with a corresponding target phrase in the target language, the suggestion system proposing a set of candidate phrases for display on the generated interface, the proposed candidate phases each comprising the suffix of a respective one of the source phrases in the set of source phrases;

the authoring interface being configured for receiving a user'"'"'s selection of one of the candidate phrases in the set and, where one of the candidate phrases in the set is selected by the user, for appending the selected one of the candidate phrases to the source text;

the suggestion system being configured generating the text string in the source language from the initial source text and appended text; and

a processor which implements the interface generator and suggestion system.

17. The system of claim 16, wherein the suggestion system comprises a scoring function which scores candidate phrases based on the translatability score of the corresponding source phrase in the set of source phrases and a fluency score and a semantic relatedness score.

18. A method for training an authoring system comprising:

acquiring a collection of source phrases in a source language from a set of biphrases derived from a parallel corpus of sentences in the source language and a target language;

with a processor, for each of the source phrases, computing a translatability score as a function of biphrase statistics derived from the parallel corpus for at least one of the biphrases in the set of biphrases for which the source phrase is a source phrase of the biphrase and storing the translatability scores in memory; and

storing a scoring function for scoring each of a set of candidate phrases to be presented to an author for appending to initial source text during authoring of a source string in the source language, each of the candidate phrases, in combination with a portion of the source text, forming one of the source phrases in the collection, the scoring function scoring candidate phrases based on the translatability score, enabling the author to be presented with candidate phrases ranked according to scores output by the scoring function through an authoring interface.

19. The method of claim 18, wherein the computing of the translatability score comprises applying weights for each of a set of features of a translation model, the features in the set being based on the statistics derived from the parallel corpus.

20. The method of claim 18, wherein the biphrase statistics are selected from phrasal probabilities, lexical probabilities, a phrase penalty, and combinations thereof.

21. The method of claim 18, wherein the scoring function score is also based on at least one of a fluency feature for the source language and a semantic relatedness model.

22. A computer program product comprising a non-transitory medium which stores instructions, which when executed by a processor, perform the method of claim 18.

23. A training system comprising memory which stores instructions for performing the method of claim 18 and a processor in communication with the memory for executing the instructions.