Talk:Earley-based structural transfer for Apertium

The current transfer assumes lemma/PoS/sub-pos (and possibly lexical selection) are all completely disambiguated. This reduces parse ambiguity _a lot_. It might even reduce it so much that it's not worth it to do parse selection, but simply choose the first, longest match, similar to what transfer currently does. This might make writing a recursive parse module a bit simpler since the Earley parser wouldn't have to keep track of shorter matches, and would not have to do anything after parsing to select between them. It's an empirical question how well that would work though (most likely it would work a lot better for pairs that use syntactic/dependency CG's). --unhammer 13:20, 2 October 2013 (UTC)