Methodological considerations in the METAL project

What it is about

An essay on how the METAL rule-based MT project ran as both a theory project and a commercial development effort, with work spread over four sites in four countries. The paper names three methodological questions: how much end-user (translator) input should steer research, how much effort should go into tools for linguistic researchers, and how a multinational team keeps one framework while each site owns language pairs. It reports no benchmark numbers. Its point is that each decision has to be justified both theoretically and by a commercial need, and that tooling for researchers pays off only if it saves more time than it costs.

Ideas

• I1. Log the origin of every lexical override: when a translator or a corpus asks for a change, record who asked, on which verses, and what it fixed. Tie each author DELTA in tools/translate to that log so the overrides stay auditable.
• I2. Justify each research tool by time saved: time one development cycle of tools/translate (the residual run, the measure run) before and after a tool is added, and keep only tools whose saving shows in the logs.
• I3. Split ownership the way the paper splits sites: one shared standard in profile.rs owned centrally, and per-author delta files owned by whoever maintains that author. Merges go through the central owner, so no author file silently changes the standard.

For Mova

The paper is about organisation, not algorithms, so its only direct use is process. It supports keeping the standard encoder single and the author deltas separate (open problem 6 and the delta rule in tools/translate), and it argues for measuring tooling cost. No measure in the paper can be reused as is.

Topics: Machine translation

Cite

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

@inproceedings{bennett1988methodological,
  title = {{Methodological considerations in the METAL project}},
  author = {Bennett, Winfield S.},
  booktitle = {Proceedings of the Second Conference on Theoretical and Methodological Issues in Machine Translation of Natural Languages},
  year = {1988},
  url = {https://aclanthology.org/1988.tmi-1.20/}
}

References

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

1. Bennett, Winfield S. 1987. The proper place of end-users in MT research. Proceedings of the 28th Annual Conference of the American Translators Association, 425-427. find
2. Gebruers, Rudi. 1988. Valency and MT: Recent developments in the METAL system. Proceedings of the Second Conference on Applied Natural Language Processing, 168-175. find
3. Margaret King (ed.). 1987. Machine translation today: The state of the art. Proceedings of the Third Lugano Tutorial. EDITS 2. Edinburgh: Edinburgh University Press. Meya, Montserrat, and J. Vidal. forthcoming. An integrated model for the treatment of time in MT systems. Proceedings of COLING 88. find
4. Nirenburg, Sergei (ed.) 1987. Machine translation: Theoretical and methodological issues. Cambridge: Cambridge University Press. find
5. Slocum, Jonathan, et al. 1987. METAL: the LRC Machine Translation System. In King (1987), 319-350. find
6. Whiffin, Lesley, and Jonathan Slocum. 1985. Machine Translation: Viewpoints from Both Sides. Working Paper LRC-85-3. Austin TX: Linguistics Research Center. find
7. White, John S. 1987. The research environment in the METAL project. In Nirenburg (1987), 225-246. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/bennett-1988-methodological-considerations-metal-project/