Open zhuang-li opened 7 years ago
The 2013 paper indeed does not address the first issue. Their follow-up paper in 2014 (http://www.aclweb.org/anthology/P14-1133) should address this issue by allowing non-adjacent logical form parts to combine.
For the second issue, I think their "bridging" technique should be able to create the second film.film.director
(and also the and
). After creating (film.film.director Christopher_Nolan)
and Tom_Hanks
, the bridging rule would find all relations BRIDGE such that (and (film.film.director join Christopher_Nolan) (BRIDGE Tom_Hanks))
successfully executes, and film.film.director
will be one of the possible BRIDGEs.
Hi, I am conducting research based on the semantic parsing paper of EMNLP 2013.
However, I found that there are several cases that sempre can't solve.
How the sempre solve the long dependent predicates? For example,
In this case, the alignment should be:
and the parsing rules:
Thus the logic form could be
(film.film.director join Christopher_Nolan) intersect (film.film.actor join Tom_Hanks)
. The paper proposed that you used a common bottom-up parser. Assuming I use a CYK parser,film.film.director
andChristopher_Nolan
are non-consecutive fragments so they can't be assembled together as in this case.Second case,
The logic form could be
(film.film.director join Christopher_Nolan) intersect (film.film.director join Tom_Hanks)
. However, theand
won't be aligned to 'film.film.director' and thefilm.film.director
won't be joined with two entities. So how to produce such logic form?Maybe there is something that I missed when reading this paper. Have you already proposed methods that can deal with such cases?
Thank you in advance for resolving my issue.