Universal translation

• US 9,734,142 B2
• Filed: 09/22/2015
• Issued: 08/15/2017
• Est. Priority Date: 09/22/2015
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Huang, Fei
• Primary Examiner(s): Desir, Pierre-Louis
• Assistant Examiner(s): KOVACEK, DAVID M
• Application Number:
  US14/861,747
• Publication Number:
  US 20170083504A1
• Time in Patent Office: 693 Days
• CPC Class Codes: G06F 40/00 Handling natural language data speech analysis or synthesis, speech recognition G10L
• CPC Class Codes: G06F 40/10 Text processing natural language analysis G06F40/20; semantic analysis G06F40/30; processing or translation of natural language G06F40/40
• CPC Class Codes: G06F 40/117 Tagging; Marking up details of markup languages G06F40/143; Designating a block; Setting of attributes style sheets, e.g. eXtensible Stylesheet Language Transformation [XSLT], G06F40/154
• CPC Class Codes: G06F 40/166 Editing, e.g. inserting or deleting
• CPC Class Codes: G06F 40/20 Natural language analysis semantic analysis of natural language G06F40/30
• CPC Class Codes: G06F 40/263 Language identification
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30
• 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

A likely source language of a media item can be identified by attempting an initial language identification of the media item based on intrinsic or extrinsic factors, such as words in the media item and languages known by the media item author. This initial identification can generate a list of most likely source languages with corresponding likelihood factors. Translations can then be performed presuming each of the most likely source languages. The translations can be performed for multiple output languages. Each resulting translation can receive a corresponding score based on a number of factors. The scores can be combined where they have a common source language. These combined scores can be used to weight the previously identified likelihood factors for the source languages of the media item.

Claims

1. A method for identifying a most likely source language of a snippet, comprising:

receiving an indication of the snippet, wherein the snippet is a digital representation of words or character groups;

determining two or more possible source languages for the snippet;

generating, by one or more machine translation engines, two or more translations of the snippet each having a specified translation source language,

wherein at least one of the two or more translations of the snippet is generated having a first of the two or more possible source languages for the snippet set as the specified translation source language, and

wherein at least another of the two or more translations of the snippet is generated having a second of the two or more possible source languages for the snippet other than the first of the two or more possible source languages for the snippet set as the specified translation source language;

computing, by one or more translation score models trained using or more neural networks, accuracy scores for at least two of the generated two or more translations of the snippet;

producing a confidence factor for each of at least two selected possible source languages for the snippet, wherein the confidence factor for each selected possible source language is produced based on one or more of the computed accuracy scores that has a source language corresponding to the selected possible source language; and

selecting, as the most likely source language, the possible source language for the snippet that is associated with a highest confidence factor.

2. The method of claim 1,

wherein the least one of the two or more translations of the snippet comprises multiple translations, each with a common specified translation source language;

wherein computing accuracy scores includes computing a combined accuracy score for the multiple translations, each with a common specified translation source language, wherein computing each combined accuracy score is performed by combining individual accuracy scores corresponding to each of the multiple translations that have a common specified translation source language; and

wherein producing the confidence factor for the common specified translation source language is based on the combined accuracy score for the multiple translations each with that common specified translation source language.

3. The method of claim 2 further comprising:

performing an initial source language identification for the snippet;

wherein the initial source language identification for the snippet identifies one or more of the possible source languages each with a corresponding initial confidence value;

wherein each initial confidence value indicates, for a corresponding possible source language, a confidence that the corresponding possible source language is a language of the snippet; and

wherein producing the confidence factor for at least a selected one of the possible source languages comprises updating the initial confidence value for the selected one of the possible source languages using the combined accuracy score corresponding to the selected one of the possible source languages.

4. The method of claim 2,

wherein the confidence factor for the common specified translation source language is the combined accuracy score.

5. The method of claim 1 further comprising:

performing an initial source language identification for the snippet;

wherein the initial source language identification for the snippet identifies one or more of the possible source languages each with a corresponding initial confidence value; and

wherein each initial confidence value indicates, for a corresponding possible source language, a confidence that the corresponding possible source language is a language of the snippet.

6. The method of claim 5,

wherein the first of the two or more possible source languages is selected based on the corresponding initial confidence value for the first of the two or more possible source languages being above a threshold value.

7. The method of claim 5,

wherein performing the initial source language identification for the snippet comprises an analysis of a context of the snippet.

8. The method of claim 7, wherein the analysis of the context of the snippet uses one or more of:

languages that an author of the snippet is known to be facile with;

languages associated with users identified as friends of the author of the snippet;

when the snippet was created;

information on the a virtual location where the snippet was posted;

or

any combination thereof.

9. The method of claim 1 further comprising:

selecting from the generated two or more translations of the snippet, as translations of the snippet, the translations where the specified translation source language is the language selected as the most likely source language of the snippet.

10. The method of claim 1, further comprising:

receiving an indication of a viewing user of the snippet; and

determining an output language associated with the viewing user of the snippet;

wherein the generated two or more translations of the snippet are each in an output language matching the output language associated with the viewing user of the snippet.

11. The method of claim 1,

wherein computing the accuracy scores is performed by a translation scoring model that is trained to generate translation scores with training data comprising data points each including an input snippet, an output snippet, and a score.

12. The method of claim 1,

wherein the method is performed in response to a request that requires a translation of the snippet.

13. The method of claim 1,

wherein the method is performed in response to creation of the snippet or posting of the snippet to a social media website.

14. A non-transitory computer-readable storage medium storing instructions that, when executed by a computing system, cause the computing system to perform operations for identifying confidence factors for snippet source languages, the operations comprising:

receiving an indication of a snippet, wherein the snippet is a digital representation of words or character groups;

receiving an indication of a viewer of the snippet;

determining an output language associated with the viewer of the snippet;

generating, by one or more machine translation engines, two or more translations of the snippet of the two or more translations each having a specified translation source language and each of the two or more translations being in an output language matching the output language associated with the viewer of the snippet,

wherein at least one of the two or more translations of the snippet is generated having a first of two or more possible source languages for the snippet set as the specified translation source language, and

wherein at least another of the two or more translations of the snippet is generated having a second of the two or more possible source languages for the snippet other than the first the two or more possible source languages for the snippet set as the specified translation source language;

computing, by one or more translation score models trained using or more neural networks, accuracy scores for at least two of the generated two or more translations of the snippet; and

producing a confidence factor for each of at least two selected possible source languages for the snippet, wherein the confidence factor for each selected possible source language is produced based on one or more of the computed accuracy scores that has a source language corresponding to the selected possible source language.

15. The non-transitory computer-readable storage medium of claim 14, wherein the operations further comprise:

performing an initial source language identification for the snippet;

wherein the initial source language identification for the snippet identifies one or more of the possible source languages each with a corresponding initial confidence value; and

wherein each initial confidence value indicates, for a corresponding possible source language, a confidence that the corresponding possible source language is a language of the snippet.

16. The non-transitory computer-readable storage medium of claim 15,

wherein the first of the two or more possible source languages is selected based on the corresponding initial confidence value for the first of the two or more possible source languages being above a threshold value.

17. The non-transitory computer-readable storage medium of claim 14, wherein the operations further comprise:

selecting from the generated two or more translations of the snippet, as translations of the snippet, translations where the specified translation source language is a language with a highest confidence factor.

18. The non-transitory computer-readable storage medium of claim 14,

wherein computing the accuracy scores is performed by a translation scoring model that is trained to generate translation scores with training data comprising data points each including an input snippet, an output snippet, and a score.

19. A system for generating a translation of a snippet, comprising:

a memory;

one or more processors;

an interface configured to receive an indication of the snippet, wherein the snippet is a digital representation of words or character groups;

a pre-translation language identifier configured to determine two or more possible source languages for the snippet;

a machine translation engine configured to generate two or more translations of the snippet, each of the two or more translations of the snippet having a specified translation source language,

wherein at least one of the two or more translations of the snippet is generated having a first of the two or more possible source languages for the snippet set as the specified translation source language, and

wherein at least another of the two or more translations of the snippet is generated having a second of the two or more possible source languages for the snippet other than the first the two or more possible source languages for the snippet as the specified translation source language;

a translation scoring model trained using one or more neural networks and configured to compute accuracy scores for at least two of the generated two or more translations of the snippet; and

a confidence score generator configured to produce a confidence factor for each of at least two selected possible source languages for the snippet, wherein the confidence factor for each selected possible source language is produced based on one or more of the computed accuracy scores that has a source language corresponding to the selected possible source language;

wherein the interface is further configured to provide from the generated two or more translations of the snippet, as translations of the snippet, translations where the specified translation source language is the possible source language with the highest confidence factor.

20. The system of claim 19, wherein the pre-translation language identifier is configured to determine the two or more possible source languages for the snippet by:

performing an initial source language identification for the snippet;

wherein the initial source language identification for the snippet identifies one or more of the possible source languages each with a corresponding initial confidence value;

wherein each initial confidence value indicates, for a corresponding possible source language, a confidence that the corresponding possible source language is a language of the snippet; and

wherein the first of the two or more possible source languages is selected based on the corresponding initial confidence value for the first of the two or more possible source languages being above a threshold value.