ArXiv's latest language policy & governance

Published
Published .

The policy

A recent policy change has caused a minor uproar (is that an oxymoron) in the math community around here. In a blog post from last November, arXiv announced that non-English submissions are essentially forbidden, and that all new submissions would need to include a translation into English. They even clarified a couple of weeks ago, mid-January, that the English version would need to appear first. Despite the diplomatic framing—claiming this will “expand the reach” of papers and ensure “transparency,” this is effectively a soft ban on non-English scientific publishing.

Perhaps the most dismaying part of this announcement is the casual dismissal of translation quality. The policy states:

“We realize that many arXiv submitters may not have access to professional translation services; non-English paper versions that use automated translation are acceptable, as long as their content is faithful to the original paper.”

Mathematics requires precision. Nuance matters. The idea that we should feed serious research through Google Translate or DeepL just to satisfy a bureaucratic moderation requirement is baffling. We are being told that a machine-garbled English approximation is preferable to a precise, rigorously written paper in French, German, or Russian. If the English version is generated by AI and acknowledged to be potentially lower quality, yet it is forced to appear first, what are we actually publishing? We are prioritizing English accessibility over mathematical accuracy.

While I sympathize with the logistical load of running the repository, the mathematical community is inherently global. Are we really to believe that arXiv cannot find a single trustworthy mathematician to screen a paper in French or Mandarin?

This policy doesn’t “encourage a bilingual model.” It encourages authors to give up on their native languages entirely. It solves the “problem” of non-English papers by making them structurally irrelevant.

Geographical breakdown of arXiv’s governance

Just for fun, I cooked up a data visualization about the geographical location of members of arXiv’s various governance boards, using the data available here. There are no board members from Africa, South America, or Oceania. The colors are chosen to be (hopefully) accessible to people with color vision deficiency.

North America
Europe
Asia
Executive Leadership (4 members)
100%
Science Advisory Council (11 members)
55%
36%
9%
Editorial Advisory Council (9 members)
78%
22%
Institutions Advisory Council (8 members)
50%
37%
13%