search

nickyringland / nested_named_entities

63 stars 7 forks source link

One mention two type #5

Closed iofu728 closed 4 years ago

iofu728 commented 4 years ago

After re-read the paper, I don't know how to evaluation multi-type? every type must predict? In my mind, the hypergraph model can't do that thing.

In the error analysis, I find a strange result.

like

6,7 DATE|3,3 CARDINAL|7,7 DAY|1,1 INDEX|4,4 UNIT|1,1 NAME|3,4 PERCENT|1,1 MEDIA|6,7 EVENT

https://github.com/nickyringland/nested_named_entities/blob/f2128eadd3a5bfae2a4e19c5c6df1feb4a5aeb00/ACL2019%20Paper/Resources/no-text/23/wsj_2300.mrg#L83

There are many mention links to two entity types. I'm confused, It's not nested phenomenon.

It's 1612 mentions have multi-type in the test set.

Again, wish you guys best luck in your ACL submission. ; )

iofu728 commented 4 years ago

I understand the problem. There are also happen in GENIA.