Predicting future translations

• US 10,289,681 B2
• Filed: 07/19/2017
• Issued: 05/14/2019
• 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): Spooner, Lamont M
• Application Number:
  US15/654,668
• Publication Number:
  US 20170315991A1
• Time in Patent Office: 664 Days
• Field of Search: 704 2- 8
• CPC Class Codes: G06F 40/263 Language identification
• CPC Class Codes: G06F 40/42 Data-driven translation
• CPC Class Codes: G06F 40/47 Machine-assisted translation, e.g. using translation memory
• 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 selecting a snippet for pre-translation comprising:

receiving a potential snippet;

computing a pre-translation score for the potential snippet by;

selecting translation likelihood factors for the received potential snippet, wherein the selected translation likelihood factors are associated with one or more values for a computational cost of performing a translation of the potential snippet, the computational cost representing an amount of processing required for machine translation logic to perform a machine translation of the potential snippet, or one or more values of a comparison between a time when the translation is expected to be completed and an expected time when a translation of the snippet is needed; and

computing the pre-translation score based on the one or more values associated with the selected translation likelihood factors;

determining that the pre-translation score matches a condition; and

in response to determining that the pre-translation score matches the condition, obtaining a pre-translation of the potential snippet.

2. The method of claim 1, wherein the one or more values associated with the selected translation likelihood factors are based on one or more of:

characteristics of an author of the potential snippet;

characteristics of content of the potential snippet;

characteristics of an expected audience of the potential snippet;

a measure of user engagement with the potential snippet thus far;

an identification of when the potential snippet was created;

an amount of time before a translation of the potential snippet is expected to be needed;

a virtual location where the potential snippet is posted;

or

any combination thereof.

3. The method of claim 2, wherein the one or more values associated with the selected translation likelihood factors are based on the characteristics of the author of the potential snippet, wherein the characteristics of the author of the potential snippet include one or more of:

languages the author of the potential snippet is identified as being facile with;

an age identified for the author of the potential snippet;

a gender identified for the author of the potential snippet;

one or more locations associated with the author of the potential snippet;

an occupation identified for the author of the potential snippet;

an education level identified for the author of the potential snippet;

a number of friends of the author of the potential snippet;

a number of friends of the author of the potential snippet who speak a language other than a language spoken by the author of the potential snippet;

or

any combination thereof.

4. The method of claim 2, wherein the one or more values associated with the selected translation likelihood factors are based on the characteristics of content of the potential snippet, wherein the characteristics of content of the potential snippet include one or more of:

an identified topic or area of interest of the potential snippet;

whether an identified topic or area of interest of the potential snippet is trending;

a length of the potential snippet;

a language identified as a source language of the potential snippet;

or

any combination thereof.

5. The method of claim 2, wherein the one or more values associated with the selected translation likelihood factors are based on the expected audience of the potential snippet, wherein the expected audience of the potential snippet is identified by determining historical viewership for one or more of:

snippets with similar topics or areas of interest as the potential snippet;

snippets with similar authors as the potential snippet;

snippets posted to similar virtual locations as the potential snippet;

or

any combination thereof.

6. The method of claim 2, wherein the one or more values associated with the selected translation likelihood factors are based on the expected audience of the potential snippet by analyzing on one or more of:

gender makeup of the expected audience;

age makeup of the expected audience;

one or more locations associated with the expected audience;

languages associated with the expected audience;

a size of the expected audience;

one or more jobs associated with the expected audience;

friends of members of the expected audience;

education level makeup of members of the expected audience;

or

any combination thereof.

7. The method of claim 2, wherein the one or more values associated with the selected translation likelihood factors are based on the measure of user engagement with the potential snippet thus far, wherein the measure of user engagement with the potential snippet thus far is based on one or more of:

a total number of viewers of the potential snippet;

a total number of requests to translate the potential snippet;

a number of users who have actuated a control displayed in association with the potential snippet;

or

any combination thereof.

8. The method of claim 1, wherein the one or more values associated with the selected translation likelihood factors are based on the amount of time before the translation of the potential snippet is expected to be needed, wherein the amount of time before the translation of the potential snippet is expected to be needed is determined based on a computation of when users, who are determined likely to request a translation of the potential snippet, are also determined likely to interact with a system that provides translations of the potential snippet.

9. The method of claim 1,

wherein the receiving the potential snippet is a second time the potential snippet is received;

wherein the pre-translation score is a second pre-translation score; and

wherein the method further comprises, prior to receiving the potential snippet the second time;

receiving the potential snippet a first time;

computing a first pre-translation score for the potential snippet;

determining that the first pre-translation score does not match the condition; and

in response to determining that the first pre-translation score does not match the condition, identifying the potential snippet as a snippet that will not be pre-translated.

10. The method of claim 9, wherein the first pre-translation score is computed as a different score than the second pre-translation score, the difference being based on computing the first pre-translation score for a first output language and computing the second a pre-translation score for a second output language, wherein the first output language and the second output language are different.

11. The method of claim 9, wherein the first pre-translation score is computed as a different score than the second pre-translation score, the difference being based on:

computing the first pre-translation score for a first output language and computing the second a pre-translation score for a second output language, wherein the first output language and the second output language are the same; and

(A) the one or more values associated with the selected translation likelihood factors selected for the first pre-translation score are different from (B) the one or more values associated with the selected translation likelihood factors selected for the second pre-translation score.

12. A computer-readable storage medium storing instructions that, when executed by a computing system, cause the computing system to perform operations for pre-translation of snippets, the operations comprising:

receiving a potential snippet;

computing a pre-translation score for the potential snippet based on a selection of translation likelihood factors for the potential snippet, wherein the selected translation likelihood factors are associated with one or more corresponding values, the one or more corresponding values computed as;

a computational cost of performing a translation of the potential snippet, the computational cost representing an amount of processing required for machine translation logic to perform a machine translation of the potential snippet, or

a comparison between a time when the translation is expected to be completed and an expected time when a translation of the snippet is needed;

determining that the pre-translation score matches a condition; and

in response to determining that the pre-translation score matches the condition, obtaining a translation of the potential snippet.

13. The computer-readable storage medium of claim 12, wherein the translation likelihood factors comprise characteristics of an author of the potential snippet and/or characteristics of an expected audience of the potential snippet.

14. The computer-readable storage medium of claim 13, wherein the translation likelihood factors comprise the characteristics of the expected audience of the potential snippet;

and wherein the expected audience of the potential snippet is identified by determining historical viewership for snippets with similar topics or areas of interest as the potential snippet and/or snippets with similar authors as the potential snippet.

15. The computer-readable storage medium of claim 12, wherein the translation likelihood factors comprise characteristics of content of the potential snippet and/or a measure of user engagement with the potential snippet.

16. The computer-readable storage medium of claim 15, wherein the translation likelihood factors comprise a measure of user engagement with the potential snippet;

and wherein the measure of user engagement is based on a total number of viewers of the potential snippet and/or a number of users who have actuated a control associated with the potential snippet.

17. The computer-readable storage medium of claim 12,

wherein the receiving the potential snippet is a second time the potential snippet is received;

wherein the pre-translation score is a second pre-translation score; and

wherein the operations further comprises, prior to receiving the potential snippet the second time;

receiving the potential snippet a first time;

computing a first pre-translation score for the potential snippet;

determining that the first pre-translation score does not match the condition; and

in response to determining that the first pre-translation score does not match the condition, identifying the potential snippet as a snippet that will not be pre-translated.

18. A system for pre-translating snippets, the system comprising:

one or more processors;

a memory;

an interface configured to receive a potential snippet;

pre-translator logic executable on the processor and configured to select one or more translation likelihood factors for the potential snippet, wherein selected translation likelihood factors are associated with corresponding values, the corresponding values computed as;

a computational cost of performing a translation of the potential snippet, the computational cost representing an amount of processing required for machine translation logic to perform a machine translation of the potential snippet, or

a comparison between a time when the translation is expected to be completed and an expected time when a translation of the snippet is needed;

pre-translation scorer logic executable on the processor and configured to compute a pre-translation score for the potential snippet based on the values associated with the translation likelihood factors selected by the pre-translator;

wherein the pre-translator is further configured to determine that the pre-translation score matches a threshold; and

one or more instances of logic executable on the processor and configured to, in response to the determining that the pre-translation score matches the threshold, perform a translation of the potential snippet.

19. The system of claim 18 further comprising:

translation timer logic executable on the processor and configured to compute an expected amount of time before a translation of the potential snippet is needed; and

translator selector logic executable on the processor and configured to select the one or more translators to perform the translation of the potential snippet, wherein each selected one of the one or more translators are associated with a performance metric indicating a projected amount of time required to perform a translation by the selected one of the one or more translators;

wherein the selection of the one or more translators is based on a comparison between the expected amount of time before a translation of the potential snippet is needed and the performance metrics associated with the one or more translators.

20. The system of claim 18 wherein the translation likelihood factors comprise one or more of:

characteristics of an author of the potential snippet;

characteristics of content of the potential snippet;

characteristics of an expected audience of the potential snippet;

or

any combination thereof.