Unproofed texts
Ambuda's goal is to create a complete archive of Sanskrit literature. Since such an archive is massive and our team is small, we must balance the quality of a text with the speed at which we can publish it.
To better strike this balance and make Ambuda more useful, we have started publishing texts with
minimal proofreading from a human expert. These unproofed texts are marked with a red
circle that links back to this page:
Unproofed texts have a higher defect rate than our other texts and may even have severe errors. We do not recommend them for beginners or for scholarly work. But despite their shortcomings, we are publishing them anyway for three reasons.
First, unproofed texts are good enough for many use cases. Optical character recognition — the process by which a computer converts a scanned book to readable text — has historically been of middling quality when applied to Sanskrit texts. But recently, Sanskrit OCR has substantially improved, and in our experience it is now common for a processed page to have zero mistakes. We think the current defect rate is acceptable for skimming, light reading, machine translation, and careful use by expert readers.
Second, unproofed texts are clearly marked and link back to a source of truth. All of our unproofed texts are integrated with our proofing environment, which links each verse and paragraph back to the scanned page it came from. This integration means that a user who has suspicions about what they're reading can quickly verify what the text actually says.
Third, unproofed texts can improve over time. Our proofing environment lets users edit and improve a text directly. Our proofing environment also includes automatic checks and rules that are becoming more capable over time. While our automatic checks will likely never reach the level of a human expert, they can narrow the gap and even uncover errors that a human expert is likely to miss.
On this basis, we plan to massively expand our library with unproofed texts and gradually refine them over time.