Machine translation using information retrieval

1. A computer-implemented method comprising:

providing a received input segment of text in a source language to be translated into a particular target language as a query to a search engine, the search engine searching an index of one or more collections of documents;

receiving one or more candidate segments in the source language in response to the query;

determining a similarity of each candidate segment to the received input segment;

for one or more candidate segments in the source language having a determined similarity that exceeds a threshold similarity, providing a translated target segment corresponding to the respective candidate segment as a target language translation of the input segment; and

for a candidate segment in the source language having a determined similarity that does not exceed the threshold similarity:

identifying one or more subsegments of the input segment,

providing each of the one or more subsegments of the input segment as a respective query to the search engine, and

for each provided subsegment:

receiving one or more candidate subsegments in the source language in response to the respective query, and

determining a similarity of each candidate subsegment to the provided subsegment;

using the similarity of the one or more candidate subsegments to determine one or more non-matching subsegments of the one or more subsegments;

replacing the one or more non-matching subsegments with respective substitute subsegments in the source language to form a substitute candidate segment; and

in response to determining that the substitute candidate segment has a similarity to the input segment exceeding the threshold similarity, providing a translated target segment in the target language corresponding to the substitute candidate segment in the source language as a target language translation of the input segment.

2. The method of claim 1 , further comprising:

for candidate subsegments having a similarity that does not exceed the threshold similarity, using a model-based statistical machine translation to identify target subsegments corresponding to the input subsegments.

3. The method of claim 2 , further comprising:

receiving an input adjusting the similarity threshold, where adjusting the similarity threshold results in different sized candidate subsegments.

4. The method of claim 1 , further comprising:

providing target subsegments for each candidate subsegments having a determined similarity to the corresponding subsegment of the input segment that exceeds a threshold similarity.

5. The method of claim 4 , where providing target subsegments includes using alignment data for the candidate subsegment to identify a location of the target subsegment in a repository.

6. The method of claim 1 , where providing the target segment corresponding to the candidate segment further includes:

using alignment data for the candidate segment to identify a location of the corresponding target segment.

7. A system comprising:

one or more computers operable to perform operations including:

providing a received input segment of text in a source language to be translated into a particular target language as a query to a search engine, the search engine searching an index of one or more collections of documents;

receiving one or more candidate segments in the source language in response to the query;

determining a similarity of each candidate segment to the received input segment;

for one or more candidate segments in the source language having a determined similarity that exceeds a threshold similarity, providing a translated target segment corresponding to the respective candidate segment as a target language translation of the input segment; and

for a candidate segment in the source language having a determined similarity that does not exceed the threshold similarity:

identifying one or more subsegments of the input segment,

providing each of the one or more subsegments of the input segment as a respective query to the search engine, and

for each provided subsegment:

receiving one or more candidate subsegments in the source language in response to the respective query, and

determining a similarity of each candidate subsegment to the provided subsegment;

using the similarity of the one or more candidate subsegments to determine one or more non-matching subsegments of the one or more subsegments;

replacing the one or more non-matching subsegments with respective substitute subsegments in the source language to form a substitute candidate segment; and

in response to determining that the substitute candidate segment has a similarity to the input segment exceeding the threshold similarity, providing a translated target segment in the target language corresponding to the substitute candidate segment in the source language as a target language translation of the input segment.

8. The system of claim 7 , the system further operable to perform operations comprising:

for candidate subsegments having a similarity that does not exceed the threshold similarity, using a model-based statistical machine translation to identify target subsegments corresponding to the input subsegments.

9. The system of claim 8 , the system further operable to perform operations comprising:

receiving an input adjusting the similarity threshold, where adjusting the similarity threshold results in different sized candidate subsegments.

10. The system of claim 7 , the system further operable to perform operations comprising:

providing target subsegments for each candidate subsegments having a determined similarity to the corresponding subsegment of the input segment that exceeds a threshold similarity.

11. The system of claim 10 , where providing target subsegments includes using alignment data for the candidate subsegment to identify a location of the target subsegment in a repository.

12. The system of claim 7 , where providing the target segment corresponding to the candidate segment further includes:

using alignment data for the candidate segment to identify a location of the corresponding target segment.

13. A computer program product, encoded on a non-transitory computer-readable medium, operable to cause data processing apparatus to perform operations comprising:

providing a received input segment of text in a source language to be translated into a particular target language as a query to a search engine, the search engine searching an index of one or more collections of documents;

receiving one or more candidate segments in the source language in response to the query;

determining a similarity of each candidate segment to the received input segment;

for one or more candidate segments in the source language having a determined similarity that exceeds a threshold similarity, providing a translated target segment corresponding to the respective candidate segment as a target language translation of the input segment; and

for a candidate segment in the source language having a determined similarity that does not exceed the threshold similarity:

identifying one or more subsegments of the input segment,

providing each of the one or more subsegments of the input segment as a respective query to the search engine, and

for each provided subsegment:

receiving one or more candidate subsegments in the source language in response to the respective query, and

determining a similarity of each candidate subsegment to the provided subsegment;

using the similarity of the one or more candidate subsegments to determine one or more non-matching subsegments of the one or more subsegments;

replacing the one or more non-matching subsegments with respective substitute subsegments in the source language to form a substitute candidate segment; and

in response to determining that the substitute candidate segment has a similarity to the input segment exceeding the threshold similarity, providing a translated target segment in the target language corresponding to the substitute candidate segment in the source language as a target language translation of the input segment.

14. The computer program product of claim 13 , further operable to perform operations comprising:

for candidate subsegments having a similarity that does not exceed the threshold similarity, using a model-based statistical machine translation to identify target subsegments corresponding to the input subsegments.

15. The computer program product of claim 14 , further operable to perform operations comprising:

receiving an input adjusting the similarity threshold, where adjusting the similarity threshold results in different sized candidate subsegments.

16. The computer program product of claim 13 , further operable to perform operations comprising:

providing target subsegments for each candidate subsegments having a determined similarity to the corresponding subsegment of the input segment that exceeds a threshold similarity.

17. The computer program product of claim 16 , where providing target subsegments includes using alignment data for the candidate subsegment to identify a location of the target subsegment in a repository.

18. The computer program product of claim 13 , where providing the target segment corresponding to the candidate segment further includes:

using alignment data for the candidate segment to identify a location of the corresponding target segment.

19. A system comprising:

one or more computers configured to perform operations comprising:

providing a received input segment of text in a source language to be translated into a particular target language as a query to a search engine, the search engine searching an index of one or more collections of documents;

receiving one or more candidate segments in the source language in response to the query;

determining a similarity of each candidate segment to the received input segment;

for one or more candidate segments in the source language having a determined similarity that exceeds a threshold similarity, providing a translated target segment corresponding to the respective candidate segment as a target language translation of the input segment; and

for a candidate segment in the source language having a determined similarity that does not exceed the threshold similarity, means for:

identifying one or more subsegments of the input segment,

providing each of the one or more subsegments of the input segment as a respective query to the search engine, and

for each provided subsegment:

receiving one or more candidate subsegments in the source language in response to the respective query, and

determining a similarity of each candidate subsegment to the provided subsegment;

using the similarity of the one or more candidate subsegments to determine one or more non-matching subsegments of the one or more subsegments;

replacing the one or more non-matching subsegments with respective substitute subsegments in the source language to form a substitute candidate segment; and

in response to determining that the substitute candidate segment has a similarity to the input segment exceeding the threshold similarity, providing a translated target segment in the target language corresponding to the substitute candidate segment in the source language as a target language translation of the input segment.