Improving Contextual Translations with ChatGPT

There's a Herculean task ahead of us here at Dubverse, as we navigate the complex universe of translations. We find that despite Google translate's efficacy in AI translation, it's unfortunately lacking in providing contextual or vernacular translations. Take for instance, when the word 'sleek' was translated to `चिकना` (Hindi) in a tech video. We pondered, can this truly convey the intended meaning?

Inspired by GPT-4's text-prompted translation and flexibility, we startd experimenting with ChatGPT. We compared translation outputs for a tech conference video and noticed that grounding the translations with Google translate's high-quality output yielded much better results. While the results aren't perfect, we found it to be a promising solution for our translation dilemma, particularly in terms of configurability and faster result generation.

However, considering the high cost of GPT-4, we investigated further with GPT-3.5-Turbo and discovered that GPT-4 remains superior in quality.

As we tread this uncharted territory, we've become aware of ChatGPT's own limitations such as its proneness to hallucinations with longer text length variables and difficulty in setting metrics for translations.

In the end, when we evaluated both ChatGPT and Google's Cloud Platform (GCP) for translations across domains like cooking and pharmacy, GCP surprisingly outperformed in some cases. Nevertheless, our work in translation, particularly the challenge of making them contextually and vernacularly sound in the realm of AI, is still ongoing and full of potential.

Stay tuned as we continue to explore possibilities and watch this space for updates on the availability of our ChatGPT translations!