The Meteo system

What it is about

A description of METEO, a machine translation system in operation since May 1977 that translates Canadian weather bulletins from English into French. It runs on a narrow sublanguage: telegraphic forecast text with a fixed bulletin structure. The software is a graph-transformation engine (the Q-system) driven by a grammar and dictionary kept apart from the algorithm. Units the system cannot analyse, because of an unknown word, an ambiguity or a broken format, are rejected whole and sent to a human translator; the system never offers partial output. Translators took part in choosing the corpus, building the dictionary and refining the system, and the paper notes that many rejected texts broke the bulletin format rules. No accuracy figure is given in the part of the text read.

Ideas

• I1. Reject instead of guessing: when a meta token has no concept, or falls back to a bare form from morph, mark the verse as rejected and count it as its own class in the residual. Report the rejection rate per author so that a low score cannot hide a large set of guessed words.
• I2. Measure the sublanguage coverage: for each book of the Bible pairs, count what share of meta tokens the standard covers, and report it as METEO reported its share of rejected units.
• I3. Format as a first stage: recognise the verse and section structure of the source (the bulletin headings in METEO) before reading, so that structure errors are not counted as lexical errors.
• I4. Keep the two modes apart as the paper does (a verbose mode for new users, a terse mode for experts): the translator's output can have a debug view with the source words of each meta token and a plain view, both from one run.

For Mova

The main use is the reject rule: tools/translate should prefer a flagged gap to a silent guess in exact mode, and count gaps by class. It supports the sublanguage view of the Bible pairs as a test domain for the reader before wider text. The paper has no statistics to compare with, so each idea needs its own measure, run twice for identical output.

Topics: Machine translation

Cite

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

@inproceedings{thouin1981meteo,
  title = {{The Meteo system}},
  author = {Thouin, Benoit},
  booktitle = {Translating and the Computer: Practical experience of machine translation},
  year = {1981},
  url = {https://aclanthology.org/1981.tc-1.4/}
}

References

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

1. Gobeil, F., La traduction automatique au Canada, vol. 14, no 5 (May 1981). find
2. Colmerauer, A., Les Systèmes-Q, ou un formalisme pour analyser et synthétiser des phrases sur ordinateur, TAUM, Université de Montréal (Sept. 1970). find
3. Thouin, B., Pour les traducteurs, par des traducteurs, Congrès mondial de la F.I.T. (Warsaw, 1981). find
4. Chandioux, J. and Guéraud, M.F., METEO, un système a l'épreuve du temps, META, vol. 26, no 1 (March 1981). L'Actualité terminologique, in: Proceedings, IX e find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/thouin-1981-meteo-system/