search

kovvalsky / prove_SICK_NL

Prove Ducth NLI problems of SICK-NL with LangPro
MIT License
3 stars 0 forks source link

Conversion or parsing issue? #18

Open kovvalsky opened 3 years ago

kovvalsky commented 3 years ago

I want to make sure from where the wrong analysis come:

image

My alpino run returns this tree which doesn't have the issue: image

konstantinosKokos commented 3 years ago

The alpino parse is the problem, een and in een donker kammer attach to both man and twee vrouwen (notice the coindexing 1 and 2). The modifier attachment is possibly ok (there is real attachment ambiguity there), but the determiner is definitely wrong.

What the TLG conversion does is (1) push the determiner on the top of the conjunction (creating the corresponding abstractions) (2) push the modifier above the conjunction, since it applies to both conjuncts the transformation would be correct, if the original parse was.