Statistical method for building a translation memory
A statistical translation memory (TMEM) may be generated by training a translation model with a naturally generated TMEM. A number of tuples may be extracted from each translation pair in the TMEM. The tuples may include a phrase in a source language and a corresponding phrase in a target language. The tuples may also include probability information relating to the phrases generated by the translation model.
1. A method comprising:
training a statistical translation model on a plurality of translation pairs, each translation pair including a text segment in a source language and a corresponding text segment in a target language;
extracting a plurality of tuples from each of a plurality of said translation pairs, each tuple comprising a pair of phrases and a corresponding word level alignment between said pair of phrases, the pair of phrases including a phrase in the source language comprising at least two words and a corresponding phrase in the target language comprising at least two words, the corresponding word level alignment based on the trained statistical translation model; and
constructing automatically a statistical translation memory from said plurality of tuples, wherein the statistical translation memory encodes a plurality of said pair of phrases and the corresponding word level alignment between each said pair.
2. The method of claim 1 , wherein the phrase in the target language corresponds to the phrase in the source language occurring most frequently in the plurality of tuples.
3. The method of claim 1 , wherein the phrase in the target language corresponds to the phrase in the source language having an alignment of highest probability.
4. The method of claim 1 , further comprising:
judging a correctness of the plurality of tuples; and
selecting a tuple as a translation equivalent in response to said judgment.
5. An automatically constructed statistical translation memory comprising:
a plurality of tuples extracted from each of a plurality of translation pairs, each tuple comprising a pair of phrases and a corresponding word level alignment between said pair of phrases, the pair of phrases including a phrase in the source language comprising at least two words and a corresponding phrase in the target language comprising at least two words, the corresponding word level alignment based on a statistical translation model trained on the translation pairs.
6. The automatically constructed statistical translation memory of claim 5 , wherein the phrase in the target language corresponds to the phrase in the source language having an alignment of highest probability.
7. The automatically constructed statistical translation memory of claim 5 , wherein the phrase in the target language corresponds to the phrase in the source language occurring most frequently in the plurality of tuples.
8. Apparatus comprising:
a statistical translation model operative to train on a plurality of translation pairs, each translation pair including a text segment in a source language and a corresponding text segment in a target language;
an extraction module operative to extract a plurality of tuples from each of a plurality of said translation pairs, each tuple comprising a pair of phrases and a corresponding word level alignment between said pair of phrases, the pair of phrases including a phrase in the source language comprising at least two words and a corresponding phrase in the target language comprising at least two words, the corresponding word level alignment based of the training of the statistical translation model; and
an automatically constructed statistical translation memory based on said plurality of tuples, wherein the statistical translation memory encodes a plurality of said pair of phrases and the corresponding word level alignment between each said pair.
9. The apparatus of claim 8 , wherein the extraction module is operative to select a phrase occurring most frequently in the plurality of tuples.
10. The apparatus of claim 8 , wherein the extraction module is operative to select a phrase having a highest probability of being a correct translation of said phrase in the source language.
11. The apparatus of claim 8 , wherein the extraction module is operative to select a phrase having an alignment of highest probability with said phrase in the source language.
12. An article comprising a machine-readable medium including instructions operative to cause a machine to perform:
training a statistical translation model on a plurality of translation pairs, each translation pair including a text segment in a source language and a corresponding text segment in a target language;
extracting a plurality of tuples from each of a plurality of said translation pairs, each tuple comprising a pair of phrases and a corresponding word level alignment between said pair of phrases, the pair of phrases including a phrase in the source language comprising at least two words and a corresponding phrase in the target language comprising at least two words, the corresponding word level alignment based on the trained statistical translation model; and
constructing automatically a statistical translation memory from said plurality of tuples, wherein the statistical translation memory encodes a plurality of said pair of phrases and the corresponding word level alignment between each said pair.
13. The article of claim 12 , wherein the phrase in the target language corresponds to the phrase in the source language occurring most frequently in the plurality of tuples.
14. The article of claim 12 , wherein said constructing comprises selecting a phrase having a highest probability of being a correct translation of said phrase in the source language.
15. The article of claim 12 , wherein constructing comprises selecting a phrase having an alignment of highest probability with said phrase in the source language.