Machine studying and different synthetic intelligence (AI) strategies have had immense success with scientific and technical duties comparable to predicting how protein molecules fold and recognising faces in a crowd. Nevertheless, the applying of those strategies to the humanities are but to be absolutely explored.
What can AI inform us about philosophy and faith, for instance? As a place to begin for such an exploration, we used deep studying AI strategies to analyse English translations of the Bhagavad Gita, an historical Hindu textual content written initially in Sanskrit.
Utilizing a deep learning-based language mannequin known as BERT, we studied sentiment (feelings) and semantics (meanings) within the translations. Regardless of large variations in vocabulary and sentence construction, we discovered that the patterns of emotion and which means had been broadly comparable in all three.
This analysis opens a path to the usage of AI-based applied sciences for evaluating translations and reviewing sentiments in a variety of texts.
An historical ebook of knowledge
The Bhagavad Gita is without doubt one of the central Hindu sacred and philosophical texts. Written greater than 2,000 years in the past, it has been translated into greater than 100 languages and has been of curiosity to western philosophers because the 18th century.
The textual content of the Bhagavad Gita relates a dialog between the Hindu deity Lord Krishna and a prince known as Arjuna. They focus on whether or not a soldier ought to go to struggle for ethics and responsibility (or “dharma”) if they’ve shut mates or household on the opposing aspect.
The textual content has been instrumental in laying the foundations of Hinduism. Amongst many different issues, it’s the place the philosophy of karma (a religious precept of trigger and impact) originates.
There have been numerous English translations of the Bhagavad Gita, however there’s not a lot work that validates their high quality. Translations of songs and poems not solely break rhythm and rhyming patterns, however may also end result within the lack of semantic data.
In our analysis, we used deep studying language fashions to analyse three chosen translations of the Bhagavad Gita (from Sanskrit to English) with semantic and sentiment analyses which assist in the analysis of translation high quality.
We used a pre-trained language mannequin referred to as BERT, developed by Google. We additional tuned the mannequin utilizing a human-labelled coaching dataset primarily based on Twitter posts, which captures 10 totally different sentiments.
These sentiments (optimistic, grateful, empathetic, pessimistic, anxious, unhappy, aggravated, denial, shock, and joking) had been adopted from our earlier analysis into social media sentiment through the onset of the COVID-19 pandemic.
Patterns of sentiment
The three translations we studied used very totally different vocabulary and syntax, however the language mannequin recognised comparable sentiments within the totally different chapters of the respective translations. In accordance with our mannequin, optimistic, aggravated and shocked sentiments are probably the most expressed.
Furthermore, the mannequin confirmed how the general sentiment polarity modifications (from unfavourable to optimistic) over the course of the dialog between Arjuna and Lord Krishna.
Arjuna is pessimistic in the direction of the start and turns into optimistic as Lord Krisha imparts data of Hindu philosophy to him. The emotions expressed by Krishna present that with philosophical data of dharma and mentorship, a troubled thoughts can get readability for making the fitting selections in instances of battle.
One limitation of our mannequin is that it was skilled on information from Twitter, so it recognises “joking” as a typical sentiment. It applies this label inappropriately to some elements of the Bhagavad Gita. Humour is sophisticated and strongly culturally constrained, and understanding it’s an excessive amount of to ask of our mannequin at this stage.
As a result of nature of the Sanskrit language, the truth that the Bhagavad Gita is a track with rhythm and rhyme, and the various dates of the translations, totally different translators used totally different vocabulary to explain the identical ideas.
The desk under reveals a number of the most semantically comparable verses from the three translations.
The makes use of of sentiment evaluation
Our analysis factors the best way to the usage of AI-based applied sciences for evaluating translations and reviewing sentiments in a variety of texts.
This expertise may also be prolonged to assessment sentiments expressed in leisure media. One other potential utility is analysing motion pictures and songs to supply insights to oldsters and authorities concerning the suitability of content material for youngsters.