Archchana Sindhujan
About
My research project
Quality Estimation for Neural Machine TranslationEstimating the quality of the translation is considered a highly important task to identify the reliability of the translation and for the improvement of the translation system. Predicting the quality of the translation depends on certain features such as no critical errors, number of named entities, number of prepositional phrases or nouns in source and target etc. Considering these features Quality Estimation(QE) models to obtain the quality score can be implemented using machine learning.
Evaluating the translation is mainly categorized into two parts, automated evaluation and human evaluation. Qualified automated translation evaluation systems should consists the following state of affairs: excessive relationship with human evaluations quantified in the correlation of translation quality, reliability, consistency in the translation result, responsiveness to nuances in quality among systems, massive range of fields, usability and speed. Quality Estimation for translation can be approached in word-level, sentence-level, error detection etc. Since quality estimation is still a growing research area, this study mainly focusses on how we can use linguistically grounded methods to improve the predictions obtained from MT systems by estimating the quality of the translation.
Supervisors
Estimating the quality of the translation is considered a highly important task to identify the reliability of the translation and for the improvement of the translation system. Predicting the quality of the translation depends on certain features such as no critical errors, number of named entities, number of prepositional phrases or nouns in source and target etc. Considering these features Quality Estimation(QE) models to obtain the quality score can be implemented using machine learning.
Evaluating the translation is mainly categorized into two parts, automated evaluation and human evaluation. Qualified automated translation evaluation systems should consists the following state of affairs: excessive relationship with human evaluations quantified in the correlation of translation quality, reliability, consistency in the translation result, responsiveness to nuances in quality among systems, massive range of fields, usability and speed. Quality Estimation for translation can be approached in word-level, sentence-level, error detection etc. Since quality estimation is still a growing research area, this study mainly focusses on how we can use linguistically grounded methods to improve the predictions obtained from MT systems by estimating the quality of the translation.