[Mova](https://md.mova.elamur.ai/free/en/) · [this page in HTML](https://mova.elamur.ai/free/en/docs/papers/isabelle-1989-reversible-mt-systems/)

[home](https://md.mova.elamur.ai/free/en/) / [docs](https://md.mova.elamur.ai/free/en/docs/) / [papers](https://md.mova.elamur.ai/free/en/docs/papers/) / [Machine translation](https://md.mova.elamur.ai/free/en/docs/papers/mtrans/) / Towards Reversible MT Systems

# Towards Reversible MT Systems

**Authors:** [Pierre Isabelle](https://md.mova.elamur.ai/free/en/docs/papers/authors/pierre-isabelle/) · **Year:** 1989 · **Venue:** Proceedings of Machine Translation Summit II **Link:** [aclanthology.org/1989.mtsummit-1.33](https://aclanthology.org/1989.mtsummit-1.33/) · **Licence:** [CC BY-NC-SA 3.0](https://creativecommons.org/licenses/by-nc-sa/3.0/) — summary only

[Proceedings of Machine Translation Summit II](https://aclanthology.org/volumes/1989.mtsummit-1/)[licence](https://creativecommons.org/licenses/by-nc-sa/3.0/)

- [Publisher's page](https://aclanthology.org/1989.mtsummit-1.33/)
- [PDF](https://aclanthology.org/1989.mtsummit-1.33.pdf)

**Authors (1)**

1. Isabelle, Pierre

Citations and author metrics from OpenAlex (); authors with the same name are not always told apart.

## What it is about

The paper argues that translation competence is a relation between sets of source and target texts, and that most MT systems define only a one-way procedure, which hides the conditions of the relation. The examples are the French present, which is translated into English as one of two tenses depending on context, and the French second person, which is easy to translate one way and hard the other. The author argues that directional systems look simpler only from one side. He also argues that a grammar weakened to produce translations overgenerates, so the analysis side gets spurious ambiguities, and that the same grammar knowledge must filter both directions. The paper is an argument for reversibility; it reports no experiments.

## Ideas

- I1. Run the reversibility test as a gate: for each verse, read the source to the meta-world and write it back to the source language through the same tables, and count the verses that return exactly. The gap is a residual to report, twice, for determinism.
- I2. Keep the conditions of a choice on the side where they are learned. For the tense case, the reader records the progressive as an aspect feature on the meta token, and the encoder uses the progressive only when that feature is present. Count the cases where the reader lacks the feature.
- I3. A one-way control: learn a table only from Hebrew to English and measure how far it fails in the other direction. A table that passes both directions with no extra data is the evidence the paper asks for.

## For Mova

The code is `tools/translate` (`measure.rs` for the residual, `meta.rs` for the shared table) and the linearizer `en/src/lin.rs`, which handles the target side. The paper's main claim, that the grammar knowledge must filter in both directions, is the reason exact mode checks the output byte for byte.

Topics: [Machine translation](https://md.mova.elamur.ai/free/en/docs/papers/topics/machine-translation/)

## Cite

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

```
@inproceedings{isabelle1989reversible,
  title = {{Towards Reversible MT Systems}},
  author = {Isabelle, Pierre},
  booktitle = {Proceedings of Machine Translation Summit II},
  year = {1989},
  url = {https://aclanthology.org/1989.mtsummit-1.33/}
}
```

- [← previous Transfer and MT Modularity](https://md.mova.elamur.ai/free/en/docs/papers/isabelle-1986-transfer-mt-modularity/)
- [next → A Challenge Set Approach to Evaluating Machine Translation](https://md.mova.elamur.ai/free/en/docs/papers/isabelle-2017-challenge-set-approach-evaluating-machine/)
- [↑ back to Machine translation](https://md.mova.elamur.ai/free/en/docs/papers/mtrans/)