Text Readability and Quality Score Feedback System for Improved Natural Language Machine Translations
Publication Date: 2018-May-16
The IP.com Prior Art Database
Text Readability and Quality Score Feedback System for Improved Natural Language Machine Translations Existing translation quality metrics such as BLEU focus on program input-output pairs. When users are not able to understand the target language, this makes correcting mis- translations impossible. For example, a user translates a document using a machine translation system (such as Google translate) from English to Japanese language. Without an understanding of Japanese characters, it is very difficult for someone to identify mistakes made by the automated system. We propose a new user interface that combines text readability scoring with language translation quality assessment. The user can remove complexity from the input language before submitting to the machine translation, thus improving the translation quality. By combining input readability and quality analysis with translation, we have found experimentally that the BLEU translation quality score improves when texts are translated with higher readability Flesch Kincaid Reading Ease score. The BLEU score and Flesch Kincaid Reading Ease score have a directly correlating relationship. When the Flesch Kincaid Reading Ease score increases, the BLEU score also increases. A.1) The Old Man and the Sea, by Ernest Hemmingway (original english) Flesch Kincaid Reading Ease: 92.2 B.1) The Old Man and the Sea, by Ernest Hemmingway (spanish human translation) B.2) The Old Man and the Sea, by Ernest Hemmingway (spanish machine trans...