Implementation and Conversational Environment of ARIANE 78.4, An Integrated System for Automated Translation and Human Revision

What it is about

A description of ARIANE-78 version 4, an operational environment at GETA in Grenoble for building, running and revising machine translation programs. Linguists write each component in one of four specialised languages, and an interactive environment manages the text corpora, the intermediate results of each step, the grammars and dictionaries, and the test and debug cycle. The part most discussed is human revision: the rough machine output is never altered, and the reviser works on a copy with the source and the rough translation side by side in an editor. The paper gives storage figures: an intermediate result takes about four times the size of its source text, and a Russian-French corpus of about 700,000 words took about 12 MB.

Ideas

• I1. Keep the machine output immutable: the encoder output stays fixed, and every human correction goes to a separate delta file. Each correction then has a source, and the residual can be measured with and without it, which matches the author DELTA rule in tools/translate.
• I2. Provenance queries: add a command that answers which verse, which merge and which author produced a given meta token or output word, as ARIANE answers which lexical units come from which dictionary. It helps debugging the lexicon (open problem 2).
• I3. Storage per stage: record the size of each stored intermediate (reader, encoder, stack) for a full run in the test set, so a change that inflates a stage is seen early.

For Mova

The copy-and-revise design maps onto the author delta: a machine output that never changes, plus an override file that is measured. The provenance command helps with the open problem of noisy merges (be with have). The storage record is a cheap addition to the roundtrip logs. The paper gives no accuracy figure, so every idea needs its own measure.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/991813.991817

@inproceedings{boitet1982implementation,
  title = {{Implementation and Conversational Environment of ARIANE 78.4, An Integrated System for Automated Translation and Human Revision}},
  author = {Boitet, Ch. and Guillaume, P. and Quezel-Ambrunaz, M.},
  booktitle = {Coling 1982: Proceedings of the Ninth International Conference on Computational Linguistics},
  year = {1982},
  doi = {10.3115/991813.991817},
  url = {https://doi.org/10.3115/991813.991817}
}

References

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

1. N~dobejkine N. - " A p p l i c a t i o n du syst~me ATEF l ' a n a l y s e automatique des textes russes". COLING-73, Pisa, 1973. find
2. [I] [2] Qu~zel-Ambrunaz M., Guillaume P. - "Analyse automatique de textes par un syst~me d ' t a t s f i n i s " . COLING-73, Pisa, 1973. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/boitet-1982-implementation-conversational-environment-ariane-78/