Apertium

Open-source rule-based translation platform you can run and extend yourself

AI Translation ToolsFree planOverseasβ˜…β˜…β˜…β˜†β˜† 3.0

What is Apertium?

Apertium is a free and open-source translation platform built on linguistic rules rather than neural networks. Each language pair ships with dictionaries, morphological analysers and transfer rules, so behaviour is inspectable and correctable: if a term is wrong you edit the dictionary rather than retrain a model. That transparency, plus the ability to run entirely offline, is why it is used for minority and low-resource languages and for research where reproducibility matters. Coverage is uneven because every pair is a separate effort, and the output reads more literally than a neural engine, but little else offers the same level of control.

Last updated: 2026-09-20. This site only provides an index; for exact features, pricing, and licensing, see the official website.

Key features

  • Rule-based translation with dictionaries, analysers and transfer rules
  • Free and open-source software that runs offline on your own hardware
  • Coverage of closely related pairs and several minority languages
  • Editable dictionaries, so terminology fixes are direct and inspectable
  • Command-line tools plus web and library interfaces for integration
  • Format handling for documents, subtitles and web pages

Pros & cons

Strengths

  • Fully open and self-hostable, with no per-character cost
  • Behaviour can be corrected by editing rules and dictionaries
  • Handles languages that commercial engines support poorly

Watch out for

  • Output reads literally and lacks the fluency of neural engines
  • Language pair coverage is uneven and depends on community work
  • Adding a new pair requires genuine linguistic expertise

Best for & use cases

low-resource languages, research, offline translation and rule-based pipelines

If you're comparing similar products, check the alternatives below, or browse all tools in the AI Translation Tools category.

FAQ

How does rule-based translation differ from neural translation?

A neural engine learns from large parallel corpora and produces fluent but opaque output. Apertium applies explicit grammar and dictionary rules, so results are more literal, but you can see why a translation came out as it did and fix it directly.

Is the quality good enough for real use?

For closely related language pairs and for getting the gist, yes. For expressive or marketing text it is not competitive with modern neural services, and its value lies in control, offline operation and language coverage.

Can I add a new language pair?

Yes, the platform is designed for it, but it is a linguistic project rather than a data exercise: you need morphology rules and dictionaries. That is why coverage grows mainly through academic and community work.