Example based machine translation system

Abstract

The present invention performs machine translation by matching fragments of a source language sentence to be translated to source language portions of an example in example base. When all relevant examples have been identified in the example base, the examples are subjected to phrase alignment in which fragments of the target language sentence in each example are aligned against the matched fragments of the source language sentence in the same example. A translation component then substitutes the aligned target language phrases from the matched examples for the matched fragments in the source language sentence.

1 - 14 . (canceled) 
   
   
       15 . A method of performing machine translation of a source language (SL) input to a translation output in a target language (TL), comprising:
 selecting examples, from an example base, that match fragments of the SL input;   aligning TL portions of the selected examples with SL portions that match the fragments of the SL input by, for each example:
 performing word alignment to identify anchor alignment points corresponding to words in the SL portion that are translations of words in the TL portion; 
   finding continuous alignments between the TL portion and the SL portion based on the anchor alignment points;   wherein finding continuous alignments comprises:
 obtaining SL boundary information indicative of positions of words in the SL input that define a boundary for a fragment of the SL portion to be aligned; 
 obtaining TL boundary information identifying boundary positions of words in the TL portion of the example that are aligned with the SL portion, based on the anchor alignment points, to obtain a minimum possible alignment (MinPA); 
 identifying a maximum possible alignment (MaxPA) by extending boundaries identified by the TL boundary information until an inconsistent alignment anchor point is reached; 
   finding non-continuous alignments between the TL portion and the SL portion; and   translating the SL input to the translation output from the continuous and non-continuous alignments.   
   
   
       16 . The method of  claim 15  comprises:
 generating a plurality of translation outputs based on the continuous and non-continuous alignments;   calculating a score for each translation out put; and   selecting at least one translation output.   
   
   
       17 . The method of  claim 16  and further comprising:
 calculating a confidence measure for the selected translation output; and   identifying one or more portions of the translation output that have a confidence measure below a threshold level.   
   
   
       18 . (canceled) 
   
   
       19 . (canceled) 
   
   
       20 . (canceled) 
   
   
       21 . The method of  claim 15  wherein finding continuous alignments further comprises:
 generating all alignments between MinPA and MaxPA, all of which include MinPA.   
   
   
       22 . The method of  claim 18  wherein finding all non-continuous alignments comprises:
 identifying a word set in the TL portion of the example that corresponds to the SL portion to be aligned, based on the anchor alignment points.   
   
   
       23 . The method of  claim 22  wherein finding all non-continuous alignments further comprises:
 identifying a word set in the SL portion of the example that aligns to a portion of the word set in the TL portion but is outside the SL boundary information.   
   
   
       24 . The method of  claim 23  wherein finding all non-continuous alignments further comprises:
 if the word set in the SL portion is continuous, finding all possible continuous alignments for the word set in the SL portion and the TL portion of the example.   
   
   
       25 . The method of  claim 23  wherein finding all non-continuous alignments further comprises:
 removing from the word set in the TL portion the words that align with the words in the SL portion that are outside the SL boundary information to obtain a minimum possible alignment (MinPA).   
   
   
       26 . The method of  claim 25  wherein finding all non-continuous alignments further comprises:
 extending boundaries of MinPA, until an inconsistent alignment anchor point is reached, to obtain a maximum possible alignment (MaxPA).   
   
   
       27 . The method of  claim 26  wherein finding all non-continuous alignments further comprises:
 generating continuous substrings from the TL portion between MinPA and MaxPA, all of which include MinPA.   
   
   
       28 . The method of  claim 15  wherein performing word alignment comprises:
 accessing a bilingual dictionary to obtain dictionary information indicative of word translations between the SL portion and the TL portion of the example.   
   
   
       29 . The method of  claim 28  wherein word alignment further comprises:
 if the TL portion of the example is in a non-segmented language, performing word segmentation on the example.   
   
   
       30 . (canceled) 
   
   
       31 . (canceled) 
   
   
       32 . (canceled) 
   
   
       33 . (canceled) 
   
   
       34 . (canceled)