Apertium is a rule-based machine translation toolchain and ecosystem, with many of our tools based on finite-state transducers.

Our language agnostic tools are native and written in C++. The various development helpers are mostly in Python.

Our language data is in various formats, including XML and other human-editable texts. Language data is split into single-language packages that can analyse and generate a given language, and translation pairs that perform transfer and transformation between two languages. The single-language packages are shared amongst many pairs.

If you wish to contribute to the language agnostic native tools you'll need to know C++.

If you wish to contribute language data to Apertium, your contributions should fit in our existing pipeline. That is, it should be rule-based and deterministic. We will happily help you learn our formats and methods, and we know from experience it is possible to learn and use Apertium in short time.

We do not currently include any statistical or neural machine translation tools or methods. We are often asked if contributions can be made with statistical or neural systems, but for now they cannot.

For more information about how to contribute, see Contributing.