What this is
A hundred papers that built computing, read off the published PDFs and set as text: 102 of them, 1936 to 2020. Each one keeps its formulas, its figures, its listings and its bibliography, and each one is translated into Vietnamese, Chinese and Japanese. The point is to be able to read the original work in the language you think in, with the mathematics still readable.
What is machine written
All of it, except the papers themselves. The text was got off each PDF by a tool: the text layer where the PDF had a usable one, a layout model or a vision model over the page images where it did not. The translations were written by a language model against a glossary of terms that is fixed for the whole corpus, so that one term is one term in every paper. Nothing here was translated by a person and nothing here was checked by one.
Every page says at the foot of it how it was made and which model made it. Where a page or a section was written by a model small enough that the result is doubtful, or went through a gateway rather than the model it was meant for, the page says that too, where you are reading rather than only at the top. Read a translation against the English if it matters.
- English: 98 full, 3 abstract only
- Tiếng Việt: 13 full, 80 abstract only
- 中文: 1 full, 29 abstract only
- 日本語: 1 full, 30 abstract only
A draft language is one whose glossary does not yet cover enough of the terms in the corpus. It is still published, with the page saying what it is, because hiding it would be the same corpus with less of it visible and no more of it true.
Licences
The papers belong to their authors and their publishers. Nothing here republishes a PDF: where a paper is linked, it is linked to the page the publisher put it on. What is published here is the text and the figures, and only where the licence allows it.
A paper in the public domain, or under a licence that allows redistribution, is here in full. 90 papers are under a licence that does not, and are here as front matter and a short abstract with a link to the publisher, which is what quotation allows and no more. A paper whose licence nobody has established is not here at all, in any form, until somebody establishes it.
How it is made
Two repositories. tamnd/papersis the corpus: Markdown, figures, manifests and reports, and never a PDF.tamnd/papers-reader is the toolchain that reads the PDFs, translates them, audits the result against ninety-two numbered rules and builds this site. Both are public and both take corrections.
A paper that should be here and is not can be proposed as an issue on the toolchain. So can a translation that is wrong, which is the more useful of the two: a wrong term in the glossary is wrong in a hundred papers at once, and one report fixes all of them.