ufal / treex

Treex NLP framework
33 stars 6 forks source link

German harmonization (Tiger to Prague) broken #63

Open dan-zeman opened 7 years ago

dan-zeman commented 7 years ago

(Reported by @michnov .)

treex -Lde -Sorig Read::Treex from='/net/data/hamledt/de/treex/dev/001.treex.gz' HamleDT::DE::Harmonize Write::Treex to='hamledt_001.streex'

Sentence 6 (for instance). Coordination is not harmonized, neither is Tiger-style prepositional phrase. Probable cause: phrase-based refactorization of transformation code by @dan-zeman in early 2016. Last Prague-style data were released in HamleDT 3.0, August 2015, still transformed by the old code.

Possible workaround: use the code from the time HamleDT 3.0 was released, i.e. https://github.com/ufal/treex/tree/HamleDT_3.0