Hi, Unfortunately I won't be able to mentor this summer.
But if others are able to mentor, I think most of my thoughts from last year still need work: https://sourceforge.net/p/apertium/mailman/apertium-stuff/thread/87r0i4qvnv.fsf%40fsfe.org/#msg58729086 More language data in particular is always relevant. Another idea that would be cool would be to have a way to query a small language model from transfer, e.g. when matching genitives we'd like to decide between keeping the genitive, rewriting with preposition (which preposition?) or rewriting to compound – these kinds of rules tend to end up with long lists, using little linguistic insight, ie. a better fit for learnt models than rules. Would it be possible to use a cpu-running lm trained on the target language and do something like <lm query="…"/> to decide which rewrite rule to apply; or to find the antecedent of some word? Would be an interesting project to find out how much help if any there is in a model that's small enough to work on the systems Apertium currently works on. best regards, Kevin Brubeck Unhammer _______________________________________________ Apertium-stuff mailing list Apertium-stuff@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/apertium-stuff
