The 'Chinese' Theory: For and Against

[Jorge_Stolfi]

AI models often hallucinate when dealing with ancient Chinese texts because they don't truly understand the historical printing formats like "白底黑字" (black characters on a white background) and "黑底白字" (white characters on a black background).

Gemini does seem to have that information -- not from analyzing scanned images of the ZHB drectly, but from thousands of articles that discuss the contents of the SBJ and ZHB and the printing history of the latter, in extreme detail.  Its claims about what was printed white-on-black rather than black-on-white in ZHB copies printed before 1400 often coincide with the CTP file, so it does have that knowledge.  But it is often wrong (and contradicts itself) on certain phrases or individual diseases and fields. 

Don't rely on AI for the text structure. Instead of asking Gemini "which parts are 白底黑字", you should look for scanned images of the original book. The book you are looking for is likely the "重修政和经史证类备用本草" (the famous 晦明轩 edition of the ZHB).

That would solve the problem, but I suspect that it would take me even more time than battling with Gemini, and I will end up with more errors.  Keep in mind that I know only a few hanzi, and only the simplified versions.  

I already have one PDF file with scanned images of one volume (#5) of a ZHB print with the white-on-black text.   Presumably I can get the other volumes too from the same site.

But a typical "raw" SBJ entry that I can extract from my HTML ZHB file can have 5-10 segments that were white-on-black, interspersed with black-on-white segments.  I suspect that it would be hard for me to identify the former by looking at the scanned images.  I would have to visually compare the shapes of traditional hanzis one at a time (instead of "reading" them, as you would do).

One difficulty that may seem trivial to you is that the digital version is displayed horizontally on my screen, while the scanned images use the vertical layout.

But thank you very much for your offer to help and for the tips about possible sources for the files.  I will first try to find a digital file (text, not images) of the ZHB with the white-on-black markup; that would solve my problem with minimum work.

All the best, --stolfi