Closed vcvpaiva closed 7 years ago
The Turku interface uses @arademaker's deduplicated sentences (which also include the original SICK phrases). Initially I only processed the derivative sentences. Anyway, I have updated the stats.xml
with the results from conll/sentences.conllu
.
I'm not sure if we should have included the original SICK sentences since they are not used in the entailment analysis, but maybe this is subject for a different issue.
Anyway, I have updated the stats.xml with the results from conll/sentences.conllu.
thanks!
and yes maybe the original sentences should've been kept apart: in one part of the experiment we focus on TE and on the other on conceptual semantics.
on the other hand there are appositives in the corpus that were not caught, e.g. https://github.com/own-pt/rte-sick/blob/master/pairs/1032.txt A person who is practicing snowboarding jumps into the air. agreed @livyreal ?
Appositives in SICK-PARSEY seem better. stats say 23. https://github.com/own-pt/rte-sick/blob/master/conll/sentences-parsey.conllu.stats.xml Turku interface provides them all, mostly correct.
some are clearly correct appositives like two young women , one with folded arms , look off screen four middle eastern children , three girls and one boy , climb on a grotto with a pink interior
some are debatable such as A group of children playing in a yard , a man in the background The back of a woman in costume ; little girl looks at her I would've thought conjunction both and some others.
One or two real errors: a small child climbs steps outdoors in a grassy area (no appos, climbs=verb, steps=noun, not vice-versa) blond child with dirty face holding yellow bottle with red cap , plants in background (root shouldn't be child, but holding.)
We have now '[A person] [who is practicing snowboarding] jumps into the air.' with rcmod(practicing, person)
. Not sure if rcmod
is the UD appos
.
We need to check the origin of the dependency tag set of Parsey McParseface. If it is the same as the original Stanford dependencies, then rcmod is:
rcmod: relative clause modifier
A relative clause modifier of an NP is a relative clause modifying the NP.
The relation points from the head noun of the NP to the head of the
relative clause, normally a verb.
“I saw the man you love” rcmod(man, love)
“I saw the book which you bought” rcmod(book,bought)
According to their paper, section 4.2 Dependency Parsing: Our model configuration is basically the same as the one originally proposed by Chen and Manning (2014) and then refined by Weiss et al. (2015).
@arademaker said:
Not sure if rcmod is the UD appos.
but we do have appositives in SICK-PARSEY, I don't get it.
EDIT: I got it now, @arademaker talked about the example that I was asking @livyreal if it was an appositive, for her. @fcbr showed it it's not, it's considered relative clause by SD.
if the creators of the corpus had succeed completely there should be no appositives in this corpus.
listing the ones marked by parsey in the interface, query <appos that seem to me correct:
The ones that don't look like appositives to me:
anyways 23 in 7.9K is a small number, closing this issue, hoping that this is small enough.
Stats say (line 120) 18 appositives, but the paper said there shouldn't be any.
Are the 18 sentences wrong? @livyreal help here?
the appositives in the Turku interface are:
anyways quite a few (8) more appositives in the Turku interface than in the stats, why?