Posted on 2025/01/12

i hate software translations

this is not about all translations nor an absolute statement, but rather a rant on how bad software translations are and how the incentives for translations in software projects are broken.

i speak nepali, which belongs to the indo-european language family. since many south asian indo-european languages form a linguistic continuum, i can often decipher certain words and sentences of related languages. this has allowed me to read translations for software in a couple of languages, and the results are not good.

while "mainstream" languages like Spanish, French, German, etc., seem to receive high-quality translations, smaller languages suffer from neglect. the issue isn’t just a lack of translations but the large volume of bad translations that do more harm than good.

the bad translation cycle:

  1. someone uses Google Translate to generate text.

  2. they submit the unverified output as a contribution to a project.

  3. the project maintainers, seeing no immediate downside, merge the translation without review.

  4. users encounter a bad interface, become frustrated, and switch back to English.

  5. the experience reinforces the belief that translations are inherently unreliable, leading people to ignore them in the future.

  6. because translations are ignored, they don’t get improved.

this isn’t just bad localization. it’s anti-localization where poor translations actively discourage people from using their native language in software.

in open-source projects, contributors are usually driven by personal fulfillment. the problem is, for some reason, people seem to find fulfillment in submitting garbage translations. either they do it for the sake of contribution statistics, or they believe they are helping when they are actually just offloading the work to Google Translate, which is notoriously unreliable.

the bigger issue is that projects do not care. internationalization (i18n) is often treated as a checklist feature. thought to quality. if someone submits translations, maintainers just think, eh, whatever, i’ll merge it. this whatever, i’ll merge it attitude enables low-effort contributions, creating a self-sustaining loop of bad localization.

some projects, like Astro, have chosen to avoid this problem by not offering translations for non-major languages at all. while this prevents bad translations (to some extent, the hindi translations for Astro docs is bad), it does nothing to solve the root issue: the lack of sustainable, high-quality localization efforts for smaller languages.

i believe governments and institutions should invest in professional translations for essential systems and we should also expect translations from commercial software that we pay for. but for open-source projects, the path forward is unclear. volunteers can’t sustain high-quality translations, and bad translations are worse than none.

if a language can’t be translated well (and kept maintained), should it be excluded? how can we improve incentives for meaningful contributions? can we establish better oversight to prevent unchecked translations? how do we ensure proper maintenance and prevent language loss in digital tools?