Closed spyysalo closed 12 years ago
Currently, entity IDs are separately managed for each semantic type. (just a C++ map container :-)
The output above shows that you used the "standoff" option, not the "brat" option for output. Do you think that using unique IDs regardless of their semantic types is necessary for only the "brat" option, or all output formats?
I think unique IDs would be a benefit for all output options. Miwa-san is currently planning to use NERsuite in an extraction pipeline using the "standoff" output format and would hope to be able to avoid duplicate IDs without running a separate script, if possible.
While I am trying to add this functionality today, I found that Sampo added this already.
@priancho : are you sure? If you're referring to 2775f229429343edeee61f56bac1ab5cedcbf91f, that appears to apply to brat-flavored standoff only.
Oh, sorry about my mistake. I am now working on this. brat output option will use unique entity IDs regardless of its semantic types soon :-)
Now the brat output option (-o brat) generates unique IDs for all entities regardless of their semantic types. It also counts the IDs in document level, whereas other options (-o conll, -o standoff) still use IDs in sentence level.
@priancho : thanks, but I think this issue actually applies to the -o standoff
option, not to the -o brat
one. From the original:
When run with the -o standoff option, NERsuite output contains duplicate IDs (within a single input document).
Hi, sorry for my mistake. I applied the same functionality for the standoff format output :-)
Great, thanks!
S
On Mon, Jun 25, 2012 at 7:29 PM, Han-Cheol Cho < reply@reply.github.com
wrote:
Hi, sorry for my mistake. I applied the same functionality for the standoff format output :-)
Reply to this email directly or view it on GitHub: https://github.com/nlplab/nersuite/issues/9#issuecomment-6544134
When run with the
-o standoff
option, NERsuite output contains duplicate IDs (within a single input document). For example (for an AnEM model):Entity IDs should preferably be unique for each input document.