Interactive Multilingual Word-Alignment Techniques

Abstract

Techniques for interactively presenting word-alignments of multilingual translations and automatically improving those translations based upon user feedback are described herein. With one or more implementations of the techniques described herein, a word-alignment user-interface (UI) concurrently displays a pair of bilingual sentences, where one is a translation of the other, and interactively highlights linked (i.e., “word-aligned”) words and phrases of the pair. Other implementations of the techniques described herein offer an option for a user to provide feedback about the existing word-alignments or realign the words or phrases. In still other described implementations, word-alignment is automatically improved based upon that user feedback.

1 - 20 . (canceled) 
     
     
         21 . A method of facilitating interactive collection of user feedback regarding word-alignments between a bilingual sentence pair, the method comprising:
 concurrently displaying each sentence of the bilingual sentence pair via a user-interface (UI) on a display including a control to facilitate word-alignment;   obtaining word-alignments from via the control, the obtaining including obtaining selection of a word or phrase of a first sentence of the bilingual sentence pair and selection of a word or phrase of a second sentence of the bilingual sentence pair, the selected word or phrase of the second sentence being linked to the selected word or phrase of the first sentence;   receiving user feedback regarding a quality of the word-alignments between the linked selected words or phrases of the bilingual sentence pair; and   generating a new training dataset comprising one or more bilingual sentence pairs, the generating being based at least in part on the received user feedback regarding the quality of the word-alignments.   
     
     
         22 . The method of  claim 21 , further comprising:
 creating a new word-alignment model based at least in part on the new training data set; and   applying the new word-alignment model to a new multilingual textual dataset to realign existing sentence pairs according to the new word-alignment model.   
     
     
         23 . The method of  claim 21 , further comprising storing the user feedback in a data structure, the user feedback indicating how many positive feedbacks were given to the word-alignments and how many negative feedbacks were given to the word-alignments. 
     
     
         24 . The method of  claim 21 , further comprising rating the quality of the word-alignments based at least in part on an amount of time between obtaining the word alignments and receiving the user feedback. 
     
     
         25 . The method of  claim 24 , wherein rating the quality of the word-alignments comprises:
 assigning a high confidence rating when the amount of time between obtaining the word alignments and receiving the user feedback is greater than a predetermined threshold; and   assigning a low confidence rating when the amount of time between obtaining the word alignments and receiving the feedback is less than a predetermined threshold.   
     
     
         26 . The method of  claim 21  wherein the first sentence of the bilingual sentence pair is in a first human language and the second sentence of the bilingual sentence pair is in a second human language, the second human language being different than the first human language. 
     
     
         27 . One or more computer storage media storing processor-executable instructions that, when executed, cause one or more processors to perform operations for improving word-alignments between a dataset of bilingual sentence pairs at least based in part on collected user feedback, the operations comprising:
 obtaining existing word-alignments of the bilingual sentence pairs of the dataset;   obtaining user feedback from multiple users, the user feedback comprising word-realignments of the existing word-alignments to create realigned bilingual sentence pairs of the dataset;   based upon the obtained user feedback, calculating an updated word-realignment of the bilingual sentence pairs of the dataset; and   generating a new word-alignment model based at least in part on the updated word-realignment of the bilingual sentence pairs of the dataset.   
     
     
         28 . The one or more computer storage media as recited in  claim 27 , wherein the obtaining user-feedback comprises:
 defining a quality threshold regarding the user feedback comprising word-realignments;   identifying a group of word-realignments exceeding the quality threshold; and   creating the updated word-realignment of the realigned bilingual sentence pairs of the dataset based at least in part on the group of word-realignments exceeding the quality threshold.   
     
     
         29 . The one or more computer storage media as recited in  claim 27 , wherein obtaining the user-feedback comprising word-realignments comprises:
 displaying the bilingual sentence pairs of via a user interface (UI);   receiving, from at least one of the multiple users, a selection of a first word in a first sentence of at least one of the bilingual sentence pairs of the dataset, the first word being linked to a linked word contained in a second sentence of the at least one of the bilingual sentence pairs;   receiving, from at least one of the multiple users, a selection of a second word in the second sentence of the at least one of the bilingual sentence pairs, the second word being different than the linked word; and   reassigning the link between first word and the linked word to a link between the first word and the second word.   
     
     
         30 . The one or more computer storage media as recited in  claim 27 , the operations further comprising:
 applying the new word-alignment model to another dataset of bilingual sentence pairs; or   exposing, via a user-interface (UI) on an output display, bilingual sentence pairs of the another dataset of bilingual sentence pairs.   
     
     
         31 . The one or more computer storage media as recited in  claim 27 , wherein the calculating of the updated word-realignment of the existing word-alignments comprises:
 calculating a user-specific confidence value based, at least in part, upon factors associated with the user;   repeating the calculating for a plurality users;   weighing each realignment based upon the calculated user-specific confidence value for each user of the plurality users; and   selecting the updated word-realignment of the realigned sentence pairs based upon the weighted realignments.   
     
     
         32 . A system comprising:
 one or more processors;   memory communicatively coupled to the one or more processors;   one or more components stored on the memory that, when executed by the one or more processors, communicate with an end-user computing device to perform operations that facilitate interactive exposing of word-alignments between a bilingual sentence pair, the operations comprising:
 simultaneously displaying, via a user-interface (UI) on a display of the end-user computing device, each sentence of the bilingual sentence pair; and 
 presenting to a user, via the UI, a control to facilitate word-alignment between the bilingual sentence pair. 
   
     
     
         33 . The system as recited in  claim 32 , wherein the operations further comprise:
 receiving a user selection of an of-interest word or phrase of a first sentence of the bilingual sentence pair; and   in response to the receiving, identifying a linked word in a second sentence of the bilingual sentence pair that corresponds to the of-interest word; and   reassigning, via the control, at least one of the of-interest word or the linked word.   
     
     
         34 . The system as recited in  claim 33 , wherein the operations further comprise identifying an emphasis word of the first sentence, the emphasis word being associated with a query. 
     
     
         35 . The system as recited in  claim 32 , wherein the reassigning is performed while still simultaneously displaying each sentence of the bilingual sentence pair via the UI. 
     
     
         36 . The system as recited in  claim 32 , wherein the receiving a user selection of an of-interest word comprises:
 determining that a user-directable position indicator is identifying a first of-interest word or phrase when input causes a graphical cursor to hover over or near the first of-interest word or phrase in the UI; and   selecting the first of-interest word identified by the position indicator.   
     
     
         37 . The system as recited in  claim 36 , wherein the receiving a user selection of an of-interest word further comprises:
 determining that the user-directable position indicator is identifying a second of-interest word or phrase when input causes the graphical cursor to hover over or near a second of-interest word or phrase in the UI; and   selecting the second of-interest word identified by the position indicator.   
     
     
         38 . The system as recited in  claim 37 , wherein the reassigning comprises:
 in response to the selecting the first of-interest word identified by the position indicator, highlighting the first of-interest word; and   in response to the selecting the second of-interest word identified by the position indicator, removing the highlighting from the first of-interest word and highlighting the second of-interest word.   
     
     
         39 . The system as recited in  claim 32 , wherein the operations further comprise, in response to the receiving, performing a dictionary look-up of the of-interest word and displaying the results of the dictionary look-up. 
     
     
         40 . The system as recited in  claim 32 , wherein the operations further comprise:
 detecting that a user selected a first word or phrase of the first sentence of the bilingual sentence pair;   in response to detecting that the user selected the first word or phrase, highlighting the first word or phrase via the UI;   detecting that the user selected a second word or phrase of a second sentence of the bilingual sentence pair;   in response to detecting that the user selected the second word or phrase, highlighting the second word or phrase via the UI; and   storing a user-feedback word-alignment associated between the first word or phrase and the second word or phrase.