The text focuses on English translations of the Bhagavad Gita, a sacred text originally written in the Sanskrit language. Although AI has been especially helpful in advancing the sciences, the researchers note that AI-based tools have been used less extensively in the humanities. In the study at hand, the researchers attempt to showcase AI’s uses in the humanities, while also exploring the issue of translation quality.
“Recent progress of language models powered by deep learning has enabled not only translations but a better understanding of language and texts with semantic and sentiment analysis,” the team of researchers claims.
Using Google’s Bidirectional Encoder Representations from Transformers (BERT), the researchers conducted an analysis of three prominent translations of the Bhagavad Gita to compare their emotional content with each other. Specifically, the researchers conducted semantic analysis and syntactic analysis to determine ways in which the translations deviated from each other in terms of the emotions expressed in the passages of the Bhagavad Gita.
“Sentiment analysis provides an understanding of human emotions and affective states which has been prominent in understanding customer behavior, health and medicine, stock market predictions, and modeling election outcomes,” the study reads.
The researchers found that the three translations had vastly different vocabulary and syntactic structures. However, the emotional qualities of each passage remained extremely similar. “The meaning conveyed across the different translations were similar although different language and vocabulary were used,” the study concludes.
The researchers note that this AI-informed approach to assessing translation quality has its limitations, particularly when focusing on holy texts like the Bhagavad Gita, as the distinct lexicon of a given religion tends to get lost in translation. However, the research presents a novel technique to assess translation quality and may allow literary translators to better understand how the sentiment and emotions of their translations map onto the original content.