Predicting future translations

• US 9,805,029 B2
• Filed: 12/28/2015
• Issued: 10/31/2017
• Est. Priority Date: 12/28/2015
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Rottmann, Kay, Huang, Fei, Zhang, Ying
• Primary Examiner(s): GUERRA-ERAZO, EDGAR X
• Application Number:
  US14/981,794
• Publication Number:
  US 20170185586A1
• Time in Patent Office: 673 Days
• Field of Search: 704 2, 704 3, 704 5, 704235, 704246, 704270, 704277
• CPC Class Codes: G06F 40/51 Translation evaluation
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation

Abstract

Technology is disclosed for snippet pre-translation and dynamic selection of translation systems. Pre-translation uses snippet attributes such as characteristics of a snippet author, snippet topics, snippet context, expected snippet viewers, etc., to predict how many translation requests for the snippet are likely to be received. An appropriate translator can be dynamically selected to produce a translation of a snippet either as a result of the snippet being selected for pre-translation or from another trigger, such as a user requesting a translation of the snippet. Different translators can generate high quality translations after a period of time or other translators can generate lower quality translations earlier. Dynamic selection of translators involves dynamically selecting machine or human translation, e.g., based on a quality of translation that is desired. Translations can be improved over time by employing better machine or human translators, such as when a snippet is identified as being more popular.

Claims

1. A method for automatically improving a speed of a statistical machine translation engine:

receiving a snippet to translate, the snippet comprising a digital representation including one or more words or character groups;

determining one or more output languages for the translation of the snippet;

computing an expected amount of time before the translation of the snippet in the one or more output languages is needed;

automatically improving the speed of the statistical machine translation engine by selecting, from among multiple possible translators, a translator to perform the translation of the snippet,

wherein the selected translator is associated with a time value, the time value indicating a speed which the selected translator performs translations; and

wherein selecting the translator is performed by;

computing, based on the time value indicating the speed which the selected translator performs translations, a measure of time for the selected translator to perform the translation of the snippet; and

determining that the measure of time is less than or equal to the expected time; and

using the selected translator to perform the translation of the snippet into the one or more output languages.

2. The method of claim 1, wherein determining the one or more output languages for the translation is based on a language associated with an expected audience for the translation of the snippet.

3. The method of claim 1, wherein determining the one or more output languages for the translation is based on content of a request for the translation of the snippet.

4. The method of claim 1, wherein computing the expected amount of time before the translation of the snippet in one or more output languages is needed is based on a determination of an expected audience for the received snippet.

5. The method of claim 4, wherein computing the expected amount of time before the translation of the snippet in one or more output languages is needed is further based on:

one or more locations associated with the expected audience; and

a determination of historic times users associated with the one or more locations use a social media system.

6. The method of claim 1, wherein the expected amount of time indicates an amount of time before a first translation request is expected.

7. The method of claim 1, wherein the expected amount of time indicates an amount of time before a threshold amount of translation requests are expected.

8. The method of claim 1, wherein the multiple possible translators include at least one machine translation engine.

9. The method of claim 1, wherein the multiple possible translators include an interface to at least one human translator.

10. The method of claim 1, wherein computing the measure of time for the selected translator to perform the translation of the snippet is based on a length of the snippet.

11. The method of claim 1,

wherein selecting the translator to perform the translation of the snippet comprises selecting at least two translators from among the multiple possible translators;

wherein a first of the at least two translators is associated with a first time value and is associated with a first quality value;

wherein a second of the at least two translators is associated with a second time value and is associated with a second quality value;

wherein the first time value and the second time value indicate the first of the at least two translators is projected to translate the received snippet faster than the second of the at least two translators;

wherein the first quality value and the second quality value indicate the first of the at least two translators is projected to generate a lower quality translation than the second of the at least two translators;

wherein using the selected translator to perform the translation of the snippet comprises using the first of the at least two translators to perform a first translation of the received snippet and using the second of the at least two translators to perform a second translation of the received snippet;

wherein results of the first translation are used to respond to requests to translate the received snippet until results of the second translation are obtained; and

wherein, in response to the results of the second translation being obtained, results of the second translation are used to respond to requests to translate the received snippet.

12. The method of claim 1,

wherein determining the one or more output languages for the translation of the snippet comprises determining at least a first output language and a second output language different from the first output language;

wherein computing the expected amount of time before the translation of the snippet is needed comprises computing a first expected amount of time before a first translation in the first output language is needed and computing a second expected amount of time before a second translation in the second output language is needed;

wherein selecting the translator to perform the translation of the snippet comprises selecting a first translator to perform the first translation based on the first expected amount of time and selecting a second translator to perform the second translation based on the second expected amount of time; and

using the selected translator to perform the translation of the snippet comprises performing the first translation using the first translator and performing the second translation using the second translator.

13. The method of claim 1 further comprising, prior to receiving the snippet to translate, selecting the snippet for pre-translation by:

computing a pre-translation score for the snippet by;

selecting translation likelihood factors for the snippet, wherein the selected translation likelihood factors are associated with corresponding likelihood scores, the corresponding likelihood scores computed as an estimation of an effect the corresponding translation likelihood factor will have on an amount of translations or time of translation of the snippet; and

computing, as the pre-translation score, a combination of the likelihood scores corresponding to the selected translation likelihood factors;

determining that the pre-translation score is above a threshold; and

in response to determining that the pre-translation score is above the threshold, identifying the snippet as a snippet that will be pre-translated.

14. A system for automatically improving a speed of a statistical machine translation engine comprising:

one or more processors;

a memory;

an interface configured to receive a snippet to translate, the snippet comprising a digital representation including one or more words or character groups;

a translation timer configured to compute an expected amount of time before a translation of the snippet is needed;

two or more translators, wherein each particular one of the two or more translators is associated with a time value, the time value indicating a speed which the particular one of the two or more translators performs translations; and

a translator selector configured to automatically improve the speed of the statistical machine translation engine by selecting a translator to perform the translation of the snippet from among the two or more translators by;

computing, based on the time value indicating the speed which the selected translator performs translations, a measure of time for the selected translator to perform the translation of the snippet; and

determining that the measure of time is less than or equal to the expected time;

wherein the selected translator is used to perform the translation of the snippet.

15. The system of claim 14, wherein the translation timer is configured to compute the expected amount of time before the translation of the snippet is needed based on:

a determination of an expected audience for the received snippet;

one or more locations associated with the expected audience; and

a determination of historic times users associated with the one or more locations use a social media system.

16. The system of claim 14, wherein the two or more translators comprise:

an interface to at least one human translator; and

at least one machine translation engine.

17. The system of claim 14, wherein the translator selector is configured to compute the measure of time for the selected translator to perform the translation of the snippet based on a length of the snippet.

18. A non-transitory computer-readable storage medium storing instructions that, when executed by a computing system, cause the computing system to perform operations for automatically improving a speed of a statistical machine translation engine, the operations comprising:

receiving a snippet to translate, the snippet comprising a digital representation including one or more words or character groups;

computing an expected amount of time before a translation of the snippet is needed; and

automatically improving the speed of the statistical machine translation engine by selecting, from among multiple possible translators, a translator to perform translation of the snippet,

wherein the selected translator is associated with a time value, the time value indicating a speed which the selected translator performs translations; and

wherein selecting the translator is performed by;

computing, based on the time value indicating the speed which the selected translator performs translations, a measure of time for the selected translator to perform the translation of the snippet; and

determining that the measure of time is less than or equal to the expected time.

19. The non-transitory computer-readable storage medium of claim 18,

wherein computing the expected amount of time before the translation of the snippet is needed is based on:

a determination of an expected audience for the received snippet;

one or more locations associated with the expected audience; and

a determination of historic times users associated with the one or more locations use a social media system.

20. The non-transitory computer-readable storage medium of claim 18, wherein the multiple possible translators comprise:

an interface to at least one human translator; and

at least one machine translation engine.