Machine translation system employing classifier

• US 10,268,686 B2
• Filed: 06/24/2016
• Issued: 04/23/2019
• Est. Priority Date: 06/24/2016
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Eck, Matthias Gerhard, Goyal, Priya
• Primary Examiner(s): Thomas-Homescu, Anne L
• Application Number:
  US15/192,170
• Publication Number:
  US 20170371870A1
• Time in Patent Office: 1,033 Days
• CPC Class Codes: G06F 40/47 Machine-assisted translation, e.g. using translation memory
• CPC Class Codes: G06F 40/51 Translation evaluation

Abstract

Exemplary embodiments relate to detecting, removing, and/or replacing objectionable words and phrases in a machine-generated translation. A classifier identifies translations containing target words or phrases. The classifier may be applied to the output translation to remove target words and phrases from the translation, or to prevent target words and phrases from being automatically presented. Further, the classifier may be applied to a translation model to prevent the target words and phrases from appearing in the output translation. Still further, the classifier may be applied to training data so that the translation model is not trained using the target words of phrases. The classifier may remove target words or phrases only when the target words or phrases appear in the output translation but not the source language input data. The classifier may be provided as a standalone service, or may be employed in the context of a machine translation system.

Claims

1. A method comprising:

translating, by a machine translation system, an input in a source language to an output in a destination language;

analyzing, by a classifier, information associated with at least one of the machine translation system or the output to determine that the output of the machine translation system comprises one or more target words or phrases, the classifier mathematically dividing target words or phrases and non-target words or phrases; and

automatically modifying the machine translation system so that the one or more target words or phrases are not automatically presented to an output device, wherein the modifying comprises modifying, by the classifier, to filter the bilingual training data that trains the machine translation system, the training data comprising pairs of words or phrases in the source language and the destination language, the training data filtered to remove pairs in which the destination word or phrase is a target word or phrase.

2. The method of claim 1, wherein analyzing the information comprises determining whether the machine translation system is capable of producing a translation that includes the one or more target words or phrases from source information that lacks one or more corresponding target words or phrases.

3. The method of claim 1, wherein the modifying comprises filtering a trained phrase table that the machine translation system uses to translate the source language into the destination language.

4. The method of claim 1, wherein the modifying further comprises at least one of removing the target words or phrases from the output, replacing the target words or phrases in the output with non-target words or phrases, or obscuring the target words or phrases in the output.

5. The method of claim 1, wherein the modifying further comprises adjusting training logic that trains the machine translation system to translate the source language into the destination language.

6. The method of claim 1, wherein the modifying comprises:

preventing the machine translation system from automatically displaying the output on a display;

presenting a prompt that allows the output to be displayed; and

in response to receiving an indication that the output should be displayed from the prompt, displaying the output on the display.

16. The method of claim 1 wherein the classifier mathematically divides target words or phrases and non-target words or phrases using a maximum-margin hyperplane.

7. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to:

translating, by a machine translation system, an input in a source language to an output in a destination language;

analyze, by a classifier, information associated with at least one of the machine translation system or the output to determine that the output of the machine translation system comprises one or more target words or phrases, the classifier mathematically dividing target words or phrases and non-target words or phrases; and

automatically modify the machine translation system so that the one or more target words or phrases are not automatically presented to an output device, wherein the modifying comprises modifying, by the classifier, to filter a trained phrase table that the machine translation system uses to translate the source language into the destination language, the trained phrase table comprising pairs of words or phrases in the source language and the destination language, the trained phrase table filtered to remove pairs in which the destination word or phrase is a target word or phrase.

8. The medium of claim 7, wherein analyzing the information comprises determining whether the machine translation system is capable of producing a translation that includes the one or more target words or phrases from source information that lacks one or more corresponding target words or phrases.

9. The medium of claim 7, wherein the modifying further comprises at least one of removing the target words or phrases from the output, replacing the target words or phrases in the output with non-target words or phrases, or obscuring the target words or phrases in the output.

10. The medium of claim 7, wherein the modifying further comprises adjusting training logic that trains the machine translation system to translate the source language into the destination language.

11. The medium of claim 7, wherein the modifying comprises:

preventing the machine translation system from automatically displaying the output on a display;

presenting a prompt that allows the output to be displayed; and

in response to receiving an indication that the output should be displayed from the prompt, displaying the output on the display.

17. The computer-readable storage medium of claim 7 wherein the classifier mathematically divides target words or phrases and non-target words or phrases using a maximum-margin hyperplane.

12. An apparatus comprising:

a non-transitory computer readable medium storing logic for a machine translation system configured to translate an input in a source language to an output in a destination language;

a classifier configured to analyze information associated with at least one of the machine translation system or the output to determine that the output of the machine translation system comprises one or more target words or phrases, the classifier mathematically dividing target words or phrases and non-target words or phrases; and

a processor configured to automatically modify, by classifier, the machine translation system so that the one or more target words or phrases are not automatically presented to an output device, wherein the modifying comprises;

filtering, by classifier, bilingual training data that trains the machine translation system or filtering, by classifier, a trained phrase table that the machine translation system uses to translate the source language into the destination language, the filtering removing references to target words or phrases.

13. The apparatus of claim 12, wherein analyzing the information comprises determining whether the machine translation system is capable of producing a translation that includes the one or more target words or phrases from source information that lacks one or more corresponding target words or phrases.

14. The apparatus of claim 12, wherein the modifying further comprises at least one of removing the target words or phrases from the output, replacing the target words or phrases in the output with non-target words or phrases, or obscuring the target words or phrases in the output.

15. The apparatus of claim 12, wherein the modifying further comprises:

preventing the machine translation system from automatically displaying the output on a display;

presenting a prompt that allows the output to be displayed; and

in response to receiving an indication that the output should be displayed from the prompt, displaying the output on the display.

18. The apparatus of claim 12 wherein the classifier mathematically divides target words or phrases and non-target words or phrases using a maximum-margin hyperplane.