Extraction-empowered machine translation
View Patent ↗The invention relates to systems and methods for automatically translating documents from a first language to a second language. To carry out the translation of a document, elements of information are extracted from the document and are translated using one or more specialized translation processes. The remainder of the document is separately translated by a statistical translation process. The translated elements of information and the translated remainder are then merged into a final translated document.
1. A method for automatically translating a document from a first language to a second language comprising:
receiving the document in the first language;
processing the document to extract elements of information;
determining, using a processor, a plurality of potential translations for each of the extracted elements of information using a first translation process and a likelihood value for each of the potential translations of the elements of information;
determining a plurality of potential translations of a remainder of the document using a second, different translation process and a likelihood value for each potential remainder translation;
generating a plurality of combinations by combining a plurality of the potential translations of the elements of information with a plurality of the potential remainder translations;
determining a likelihood value for respective ones of a plurality of the combinations based on a model of the second language and corresponding likelihood values of each of the potential element of information translations and remainder translations included in the respective combinations; and
forming a translated version of the document based on the likelihood values of the combinations.
2. The method of claim 1 , wherein the elements of information comprise names of named entities, and the first translation process comprises a transliteration process.
3. The method of claim 1 , wherein the first translation process further comprises at least one of a statistical translation process, a transfer rule-based translation process, and an application of a bilingual name lexicon depending on a semantic type of the named entities.
4. The method of claim 1 , wherein the elements of information comprise at least of one of dates, times, and amounts, and the first translation process comprises applying semantic rules for the second language associated with a corresponding class of information.
5. The method of claim 1 , wherein the elements of information comprise core noun phrase descriptions and the first translation process comprises employing a mixture model to translate the extracted elements of information.
6. The method of claim 5 , wherein employing the mixture model includes employing a statistical translation process combined with a dictionary translation process.
7. The method of claim 1 , wherein the first translation process includes applying syntactic and semantic transfer rules.
8. The method of claim 1 , wherein the second translation process includes employing a statistical translation process.
9. The method of claim 1 , further comprising
identifying a span of at least two words that are related, and
translating the words of the span as a cluster.
10. The method of claim 1 , wherein the remainder of the document, prior to translation, includes placeholders for original locations of the extracted elements of information.
11. The method of claim 10 , wherein the translated remainder of the document includes placeholders for insertion of the translated elements of information.
12. A method for automatically translating a document from a first language to a second language comprising:
receiving the document in the first language;
determining, using a processor, a plurality of potential translations of a first portion of the document to the second language using a first translation process and a likelihood value for each potential first portion translation;
determining a plurality of potential translations of a second portion of the document to the second language using a second, different translation process and a likelihood value for each potential second portion translation;
generating a plurality of combinations by combining a plurality of the potential translations of the first portion with a plurality of the potential translations of the second portion;
determining a likelihood value for respective ones of a plurality of the combinations based on a model of the second language and corresponding likelihood values of each of the potential translations of the first portion and the potential translations of the second portion; and
forming a translated version of the document based on the likelihood values of the combinations.
13. The method of claim 12 , comprising:
extracting the first portion from the document; and
replacing the first portion in the document with placeholders to form the second portion.
14. The method of claim 12 , wherein the first translation process is at least partially rule driven and the second translation process is statistics-based.
15. A system for automatically translating a document from a first language to a second language comprising:
a first processor for receiving the document in the first language and processing the document to extract elements of information;
a second processor for determining a plurality of potential translations for each of the extracted elements of information using a first translation process and a likelihood value for each of the potential translations of the elements of information;
a third processor for determining a plurality of potential translations of a remainder of the document using a second, different translation process and a likelihood value for each potential remainder translation;
a fourth processor for generating a plurality of combinations by combining a plurality of the potential translations of the elements of information with a plurality of the potential remainder translations;
a fifth processor for determining a likelihood value for respective ones of a plurality of the combinations based on a model of the second language and corresponding likelihood values of each of the potential element of information translations and remainder translations included in the respective combinations; and
a sixth processor for forming a translated version of the document based on the likelihood values of the combinations.
16. A non-transitory computer-readable medium containing computer-executable instructions, which when executed by a computer carry out a method for translating a document from a first language to a second language, comprising:
receiving the document in the first language;
processing the document to extract elements of information;
determining, using a processor, a plurality of potential translations for each of the extracted elements of information using a first translation process and a likelihood value for each of the potential translations of the elements of information;
determining a plurality of potential translations of a remainder of the document using a second, different translation process and a likelihood value for each potential remainder translation;
generating a plurality of combinations by combining a plurality of the potential translations of the elements of information with a plurality of the potential remainder translations;
determining a likelihood value for respective ones of a plurality of the combinations based on a model of the second language and corresponding likelihood values of each of the potential element of information translations and remainder translations included in the respective combinations; and
forming a translated version of the document based on the likelihood values of the combinations.