ufal / treex

Treex NLP framework
33 stars 6 forks source link

Fixing T2A::EN::AddArticles #25

Closed tuetschek closed 8 years ago

tuetschek commented 8 years ago

The block T2A::EN::AddArticles now just assigns articles according to the definiteness grammateme and does not try to guess them (which was a legacy function, designed for CS-EN translation only).

The article guessing rules have been moved to CS-EN transfer (T2T::CS2EN::AddDefiniteness).

In addition, several fixes in Czech and language-independent analysis were needed to keep the performance approximately similar.

All X-to-EN language pairs (except CS-EN) should receive gains from this, since they have articles and it is safer to copy the definiteness grammateme from the source language than to try and guess.