Systems and methods to identify most suitable grammar suggestions among suggestions from a machine translation model

Abstract

A semantic dependency relationship and/or a syntactic dependency relationship is determined between at least one word of a sentence and any of a set of surrounding words in the sentence. The determined semantic dependency relationship and/or the determined syntactic dependency relationship are provided as input to a machine learning model. One or more outputs of the machine learning model are obtained, the one or more outputs indicating, for a set of candidate edits for the at least one word of the sentence, a grammatical accuracy of each candidate edit of the set of candidate edits in view of the syntactic dependency relationship and/or the syntactic dependency relationship. A candidate edit is selected from the set of candidate edits based on the indicated grammatical accuracy of each candidate edit from the set of candidate edits.

What is claimed is: 
     
         1 . A method comprising:
 determining one or more of a semantic dependency relationship or a syntactic dependency relationship between at least one word of a sentence and any of a set of surrounding words in the sentence;   providing the one or more of the determined semantic dependency relationship or the determined syntactic dependency relationship between the at least one word and any of the set of surrounding words as input to a machine learning model;   obtaining one or more outputs of the machine learning model, wherein the one or more outputs indicate, for a set of candidate edits for the at least one word of the sentence, a grammatical accuracy of each candidate edit of the set of candidate edits in the sentence in view of one or more of the semantic dependency relationship or the syntactic dependency relationship between the at least one word and any of the set of surrounding words; and   selecting a candidate edit from the set of candidate edits based on the indicated grammatical accuracy of each candidate edit from the set of candidate edits.   
     
     
         2 . The method of  claim 1 , wherein determining the one or more of the semantic dependency relationship or the syntactic dependency relationship between the at least one word and any of the set of surrounding words comprises one or more of:
 determining whether the at least one word and any of the set of surrounding words have a shared semantic property, or   determining whether the at least one word and any of the set of surrounding words have a shared syntactic property.   
     
     
         3 . The method of  claim 1 , further comprising:
 identifying at least one of a part of speech identifier for each word of the sentence or a context relationship between the at least one word and one or more additional words of the sentence, wherein each of the one or more additional words is identified based on an immediate proximity of a corresponding additional word to the at least one word in the sentence; and   providing, as additional input to the machine learning model, at least one of the part of speech identifier for each word of the sentence or the context relationship between the at least one word and the one or more additional words.   
     
     
         4 . The method of  claim 1 , further comprising:
 providing, as an input to a language frequency model, each of the set of candidate edits for the at least one word of the sentence; and   obtaining one or more additional outputs from the language frequency model, wherein the one or more additional outputs indicate a frequency that each of the set of candidate edits being used in a particular set of files.   
     
     
         5 . The method of  claim 4 , wherein selecting the candidate edit from the set of candidate edits is further based on the frequency of each of the set of candidate edits being used in the particular set of files. 
     
     
         6 . The method of  claim 1 , further comprising:
 determining whether the indicated grammatical accuracy of each candidate edit from the set of candidate edits satisfies a grammatical accuracy criterion; and   responsive to determining a corresponding grammatical accuracy for one or more particular candidate edits do not satisfy the grammatical accuracy criterion, removing the one or more particular candidate edits from the set of candidate edits.   
     
     
         7 . The method of  claim 1 , further comprising:
 providing, as input to an edit generation model, the at least one word of the sentence, wherein the edit generation model is configured to identify, for a given set of words, one or more potential corrected forms of the given set of words; and   obtaining one or more outputs of the edit generation model, the one or more outputs of the edit generation model comprising the set of candidate edits for the at least one word, wherein each of the set of candidate edits corresponds to a potential corrected form of the at least one word, and wherein the set of candidate edits is provided with the one or more of the determined semantic dependency relationship or the determined syntactic dependency relationship as input to the machine learning model.   
     
     
         8 . The method of  claim 1 , wherein the machine learning model is trained to determine an accuracy of suggested edits to contents of a document based on training data comprising:
 a training input indicating one or more of a prior semantic dependency relationship or a prior syntactic dependency relationship between a prior set of words of a prior sentence that was previously corrected by replacing the prior set of words with a prior set of edit words in the corrected prior sentence, and   a target output indicating whether the prior set of edits in the corrected prior sentence is grammatically accurate in view of the one or more of the prior semantic dependency relationship or the prior syntactic dependency relationship.   
     
     
         9 . A system comprising:
 a memory; and   a processing device coupled to the memory, the processing device to perform operations comprising:
 determining one or more of a semantic dependency relationship or a syntactic dependency relationship between at least one word of a sentence and any of a set of surrounding words in the sentence; 
 providing the one or more of the determined semantic dependency relationship or the determined syntactic dependency relationship between the at least one word and any of the set of surrounding words as input to a machine learning model; 
 obtaining one or more outputs of the machine learning model, wherein the one or more outputs indicate, for a set of candidate edits for the at least one word of the sentence, a grammatical accuracy of each candidate edit of the set of candidate edits in the sentence in view of one or more of the semantic dependency relationship or the syntactic dependency relationship between the at least one word and each of the set of surrounding words; and 
 selecting a candidate edit from the set of candidate edits based on the indicated grammatical accuracy of each candidate edit from the set of candidate edits. 
   
     
     
         10 . The system of  claim 9 , wherein determining the one or more of the semantic dependency relationship or the syntactic dependency relationship between the at least one word and any of the set of surrounding words comprises one or more of:
 determining whether the at least one word and any of the set of surrounding words have a shared semantic property, or   determining whether the at least one word and any of the set of surrounding words have a shared syntactic property.   
     
     
         11 . The system of  claim 9 , wherein the operations further comprise:
 identifying at least one of a part of speech identifier for each word of the sentence or a context relationship between the at least one word and one or more additional words of the sentence, wherein each of the one or more additional words is identified based on an immediate proximity of a corresponding additional word to the at least one word in the sentence; and   providing, as additional input to the machine learning model, at least one of the part of speech identifier for each word of the sentence or the context relationship between the at least one word and the one or more additional words.   
     
     
         12 . The system of  claim 9 , wherein the operations further comprise:
 providing, as an input to a language frequency model, each of the set of candidate edits for the at least one word of the sentence; and   obtaining one or more additional outputs from the language frequency model, wherein the one or more additional outputs indicate a frequency that each of the set of candidate edits being used in a particular set of files.   
     
     
         13 . The system of  claim 12 , wherein selecting the candidate edit from the set of candidate edits is further based on the frequency of each of the set of candidate edits being used in the particular set of files. 
     
     
         14 . The system of  claim 9 , wherein the operations further comprise:
 determining whether the indicated grammatical accuracy of each candidate edit from the set of candidate edits satisfies a grammatical accuracy criterion; and   responsive to determining a corresponding grammatical accuracy for one or more particular candidate edits do not satisfy the grammatical accuracy criterion, removing the one or more particular candidate edits from the set of candidate edits.   
     
     
         15 . The system of  claim 9 , wherein the operations further comprise:
 providing, as input to an edit generation model, the at least one word of the sentence, wherein the edit generation model is configured to identify, for a given set of words, one or more potential corrected forms of the given set of words; and   obtaining one or more outputs of the edit generation model, the one or more outputs of the edit generation model comprising the set of candidate edits for the at least one word, wherein each of the set of candidate edits corresponds to a potential corrected form of the at least one word, and wherein the set of candidate edits is provided with the one or more of the determined semantic dependency relationship or the determined syntactic dependency relationship as input to the machine learning model.   
     
     
         16 . A non-transitory computer readable medium comprising instructions that, when executed by a processing device, cause the processing device to perform operations comprising:
 determining one or more of a semantic dependency relationship or a syntactic dependency relationship between at least one word of a sentence and each of a set of surrounding words in the sentence;   providing the one or more of the determined semantic dependency relationship or the determined syntactic dependency relationship between the at least one word and any of the set of surrounding words as input to a machine learning model;   obtaining one or more outputs of the machine learning model, wherein the one or more outputs indicate, for a set of candidate edits for the at least one word of the sentence, a grammatical accuracy of each candidate edit of the set of candidate edits in the sentence in view of one or more of the semantic dependency relationship or the syntactic dependency relationship between the at least one word and each of the set of surrounding words; and   selecting a candidate edit from the set of candidate edits based on the indicated grammatical accuracy of each candidate edit from the set of candidate edits.   
     
     
         17 . The non-transitory computer readable medium of  claim 16 , wherein determining the one or more of the semantic dependency relationship or the syntactic dependency relationship between the at least one word and any of the set of surrounding words comprises one or more of:
 determining whether the at least one word and any of the set of surrounding words have a shared semantic property, or   determining whether the at least one word and any of the set of surrounding words have a shared syntactic property.   
     
     
         18 . The non-transitory computer readable medium of  claim 16 , wherein the operations further comprise:
 identifying at least one of a part of speech identifier for each word of the sentence or a context relationship between the at least one word and one or more additional words of the sentence, wherein each of the one or more additional words is identified based on an immediate proximity of a corresponding additional word to the at least one word in the sentence; and   providing, as additional input to the machine learning model, at least one of the part of speech identifier for each word of the sentence or the context relationship between the at least one word and the one or more additional words.   
     
     
         19 . The non-transitory computer readable medium of  claim 16 , wherein the operations further comprise:
 providing, as an input to a language frequency model, each of the set of candidate edits for the at least one word of the sentence; and   obtaining one or more additional outputs from the language frequency model, wherein the one or more additional outputs indicate a frequency that each of the set of candidate edits being used in a particular set of files.   
     
     
         20 . The non-transitory computer readable medium of  claim 19 , wherein selecting the candidate edit from the set of candidate edits is further based on the frequency of each of the set of candidate edits being used in the particular set of files.