Formality Sensitive Machine Translation

Magana, Gregory [Browse]
Senior thesis


Li, Xiaoyan [Browse]
Princeton University. Department of Computer Science [Browse]
Class year
Summary note
While machine translation efforts get better every day, they still fall short when it comes to understanding the complexities of language, from idiom to poesy. One of the more important nuances of language that modern translators have difficulty understanding is formality. Thispaper provides the design, development, and evaluation of a formality sensitive machine translation (FSMT) framework that attempts to build on top of existing translation progress by adding formality augmentation- making informal sentences more formal during the translationprocess. Specifically, this paper addresses the design of three major components in the translation process that allow for formality adjustment: raw translation, classification of rawtranslations as formal or informal, and formality augmentation of informal sentences. It is found that the combination of these components yields a highly modularized framework that provides modest gains in formality in many cases. This paper will also discuss the large potential for rapid improvement with future work offered by this framework as well as some of the related work already done concerning machine translation and sentence classification.

Supplementary Information