Problems with the second generation architecture and new trends in MT

What it is about

A tutorial abstract. It lists the drawbacks of the second-generation rule-based architecture: development effort, robustness, scalability and translation quality, with comparison to commercial systems. Its main subject is example-based MT, also called analogy-based or memory-based MT, which translates by finding the nearest aligned examples in a corpus. The range runs from translation memories inside a traditional system, through hybrid systems, to wholly corpus-based designs. The local text holds only the abstract and the background reading list, not the body of the tutorial. The card is therefore based on the abstract, and any idea below should be checked against the full paper before use.

Ideas

• I1. Make the example fallback explicit: when the encoder has no chunk for a meta token in its most specific context, look up the nearest aligned verse examples for the same concept and tag, and count these fallbacks as their own class in the residual.
• I2. Copy baseline for verses that are already in the Bible pairs: report the exact-match rate of a translation memory lookup in measure.rs next to the translator's score, so that any gain is compared with what reuse alone gives.
• I3. Split the residual between a structural part (the rule stack) and a lexical-choice part (example lookup or counts): measure which part holds most of the bits in each class, and decide where each part is used.

For Mova

The hybrid view matches the translator's split between structure and lexical choice. The translation memory baseline is the simplest honest control for the Bible pairs. The example fallback is already partly present in the standard encoder's context order, so the new work is to count it. The text read is only an abstract, so no claim is made about the tutorial's own evidence.

Topics: Machine translation

Cite

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

@inproceedings{somers1995problems,
  title = {{Problems with the second generation architecture and new trends in MT}},
  author = {Somers, Harold},
  booktitle = {Proceedings of Machine Translation Summit V},
  year = {1995},
  url = {https://aclanthology.org/1995.mtsummit-1.4/}
}

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/somers-1995-problems-second-generation-architecture-new/