Apparatus for providing feedback of translation quality using concept-based back translation
First Claim
Patent Images
1. A concept-based back translation system, comprising:
- a target language semantic parser module, which uses statistical techniques including decision trees to train a parser based on annotated corpus in the target language, parses a translated target sentence with the parser, once it has been trained, to extract semantic concepts, and based on results of parsing the translated target sentence, obtains a backward translation of the translated sentence into a source language;
a source language semantic parser module, which parses a source language sentence to extract semantic concepts and uses results of parsing the source language sentence to determine a significance and relevance of the extract concepts generated in the target language sentence;
a bi-directional machine translation module, which receives a translated target language sentence and converts the translated target language sentence back into the source language;
a relevancy judging module, which measures a significance and a relevance of the back translation by semantically comparing the back translation with an original source language sentence; and
a back translation display module, which displays the back translation dynamically based on the relevance and the significance of the back translationwherein the relevancy judging module measures the relevancy using a semantically weighted score based on modified n-gram precision where the original source language sentence is treated as a reference, and the back translation is treated as a translation hypothesis,wherein the relevancy between the hypothesis and the reference are compared based on a modified n-gram precision, which automatically evaluates translation performance, andwherein a higher weight is assigned to n-grams that contain significant semantic annotations, and a lower weight is assigned to n-grams that contain no significant semantic role, the significance of the n-grams is judged based on semantic parsing results, and on a pre-defined list of key concepts for an application domain.
1 Assignment
0 Petitions
Accused Products
Abstract
A concept-based back translation system includes a target language semantic parser module, a source language semantic parser module, a bi-directional machine translation module, a relevancy judging module, and a back translation display module.
31 Citations
2 Claims
-
1. A concept-based back translation system, comprising:
-
a target language semantic parser module, which uses statistical techniques including decision trees to train a parser based on annotated corpus in the target language, parses a translated target sentence with the parser, once it has been trained, to extract semantic concepts, and based on results of parsing the translated target sentence, obtains a backward translation of the translated sentence into a source language; a source language semantic parser module, which parses a source language sentence to extract semantic concepts and uses results of parsing the source language sentence to determine a significance and relevance of the extract concepts generated in the target language sentence; a bi-directional machine translation module, which receives a translated target language sentence and converts the translated target language sentence back into the source language; a relevancy judging module, which measures a significance and a relevance of the back translation by semantically comparing the back translation with an original source language sentence; and a back translation display module, which displays the back translation dynamically based on the relevance and the significance of the back translation wherein the relevancy judging module measures the relevancy using a semantically weighted score based on modified n-gram precision where the original source language sentence is treated as a reference, and the back translation is treated as a translation hypothesis, wherein the relevancy between the hypothesis and the reference are compared based on a modified n-gram precision, which automatically evaluates translation performance, and wherein a higher weight is assigned to n-grams that contain significant semantic annotations, and a lower weight is assigned to n-grams that contain no significant semantic role, the significance of the n-grams is judged based on semantic parsing results, and on a pre-defined list of key concepts for an application domain. - View Dependent Claims (2)
-
Specification