3 ms·
I think there's more to it than that. It's a combination of extracting the text from the PDFs, resolving persistently tricky issues (especially numbers), and fl
by us-merul 2mo ago
I think there's more to it than that. It's a combination of extracting the text from the PDFs, resolving persistently tricky issues (especially numbers), and flagging issues for human review at scale. You can view lessons from the author on the GitHub repo: https://github.com/sweisman/translation-pipeline https://github.com/sweisman/translation-pipeline
- lordgrenville 2mo agoGreat, then rather post that! Although that repo looks like 100% AI artifact, not human intervention. (What does "flagging issues for human review at scale" mean in this context?)
- us-merul 2mo agoI was referring to this bullet under Lessons Learned: "Manuscript translation is feasible but rough. A 14th-century Sephardic manuscript hand is readable at ~70–80% accuracy. The output is a useful rough draft, not a polished translation. Plan for a longer discrepancy report and more editorial review." Some of the samples I looked at had question marks or original Hebrew text throughout, which I took to mean marked uncertainties.
- sweisman 2mo agoManuscripts are doable but tougher than printed works. Hebrew likewise is tougher than Latin.
- sweisman 2mo agoThis is not what I use to translate. It was the first version to create usable output, and intentionally implemented in a code-free manner (today I have an extensive body of code in addition to the Claude Code harness instructions). Read my origin story, linked to in another comment, for more background.
- lordgrenville 2mo agoReiterating what I wrote above, sharing these would have been more interesting imo than sharing the pipeline output.
- sweisman 2mo ago[flagged]