Translation for research papers for both reading and publishing purposes have mostly relied on agencies' service. How can we make it cheaper and faster by utilizing existing strong machine learning tools with crowdsourcing? Literal.Press is a thought prototype to investigate this possibility.
Challenges faced and things learned: A lot of API providers for text extraction are very...subpar compared to their open source relatives (tesseract, textract, pdfttotext) with IBM Document Conversion being the only exception with intelligent sections categorization. I overestimated some companies API's language support too that led to a total change in providers used and I learned how complex Medium editing tools are (and how amazing they are in accomplishing the truest WYSIWYG editor).