Apertium: a free/open source platform for machine translation and basic language technology

What it is about

A one-page project description of Apertium, a free/open-source shallow-transfer rule-based MT platform started in 2005. The engine is a Unix-style pipeline of small modules: format filter, finite-state morphological analyser and generator, HMM and Constraint Grammar tagging, lexical transfer, chunking, and interchunk and postchunk rules. Linguistic data is declarative XML, compiled into finite-state forms, and the same engine serves more than 40 language pairs. The paper gives no new experiment; its claim is architectural: engine and data are separate, modules talk through text streams, and shallow transfer is enough for related languages. It also notes uses beyond MT, such as the Wikimedia Content Translation tool.

Ideas

I1. Count, for tools/translate, how many translator behaviours live in data files and how many in Rust code. Each author delta should load from a data file with no rebuild; a test that adds a new delta without touching code is the Apertium-style gate for this.
I2. Keep every stage of tools/roundtrip and tools/translate dumpable as a plain text stream (meta tokens, chunk choices, frames). Then a residual per layer can be diffed between two runs, as the pipeline of Apertium allows for each module.
I3. Report a word-for-word floor per author next to the layered output: lookup of the most frequent chunk for each meta token, with no reordering. The paper argues this baseline is the intelligible part of the output; in Mova it gives the gain of each layer above nothing.

For Mova

The code is tools/translate (profile.rs for the encoder, meta.rs for the lexicon, measure.rs for the residual) and tools/roundtrip. The idea of separating engine and data matches the author-delta design (open problem 6 and the style deltas of problem 8). The shallow-transfer tradition itself is not for Mova: Mova builds trees and frames, not chunk rules. Ideas I1 and I3 are measures to run before any claim about a layer.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{forcada2016apertium,
  title = {{Apertium: a free/open source platform for machine translation and basic language technology}},
  author = {Forcada, Mikel L. and Tyers, Francis M.},
  booktitle = {Proceedings of the 19th Annual Conference of the European Association for Machine Translation: Projects/Products},
  year = {2016},
  url = {https://aclanthology.org/2016.eamt-2.4/}
}

References

1 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

Forcada, M. L., Ginestí-Rosell, M., Nordfalk, J., O’Regan, J., Ortiz-Rojas, S., Pérez-Ortiz, J. A., Sánchez-Martínez, F., Ramírez-Sánchez, G. and Tyers, F. M. (2011) Apertium: a free/open-source platform for rule-based machine translation. Machine Translation 24(1) 1–18 find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/forcada-2016-apertium-free-open-source-platform/