Back to results
US 7,680,647 B2Granted
Association-based bilingual word alignment
AssigneeMicrosoft Corporation
InventorRobert C. Moore
Life of this patent
Estimating the term…
Abstract
A pair of bilingual, aligned text fragments are received and word type alignment is performed on the text fragments. Word token alignment is then performed.
Claims17
A computer implemented method of aligning words in pairs of bilingual, aligned text fragments using a computer with a processor, comprising:
- linking, with the processor, word types in a plurality of pairs of the aligned text fragments, based on statistical word association scores;
- estimating, with the processor, a conditional probability for each pair of word types in a selected pair of aligned text fragments that is linked during the linking step, based on whether the pair of word types is also linked in other pairs of aligned text fragments during the linking step;
- generating the word type alignment, with the processor, by linking pairs of word types based on the conditional probability estimated;
- wherein linked word types in the word type alignment comprise a cluster, and further comprising, repeating the step of generating a word type alignment for unlinked words in the selected pair of text fragments by linking the unlinked words to other unlinked words or to clusters, by preferentially linking the unlinked words to other unlinked words instead of linking the unlinked words to clusters;
- calculating, with the processor, a measure of non-monotonicity for each of a plurality of word token alignments corresponding to the word type alignment;
- identifying, with the processor, a word token alignment having a desired measure of non-monotonicity;
- and
- outputting a textual input, generated based on the word type alignment and the identified word token alignment, to a translation system.
5 more elements
A computer implemented method of aligning words in pairs of bilingual, aligned text fragments, using a computer with a processor, comprising:
- generating a word type alignment for a selected pair of bilingual, aligned text fragments, the word type alignment including a plurality of word token alignments;
- calculating, with the processor, a measure of non-monotonicity of each of the word token alignments;
- identifying, with the processor, a word token alignment having a desired measure of non-monotonicity;
- and
- outputting an indication of word alignments in the identified word token alignment for use in a machine translation system.
2 more elements
A computer implemented method of aligning words in pairs of bilingual, aligned text fragments, using a computer with a processor, comprising:
- generating, with the processor, a word type alignment for a selected pair of aligned text fragments by linking word types in the aligned text fragments based on statistical word association scores for words in the text fragments, the word type alignment including a plurality of word token alignments;
- calculating, with the processor, a non-monotonicity measure, indicative of a degree of monotonicity, for the word token alignments;
- generating additional links of instances of unlinked words in the word token alignments, if the additional links do not increase the non-monotonicity measure for the word token alignment for which they are generated;
- and
- outputting a translation output for use in a machine translation system based on links and additional links in the word token alignment.
2 more elements
Dependencies are inferred from the claim text.