I thought this paper was really interesting. They did things that are very similar to what I’m currently working on. So, this is a bit longer than most of my posts about research papers.

Ok, this isn’t actually a paper, it’s a book. But I just finished it, and it’s a really good book, I thought.

This is the last of H&K’s coreference resolution papers, and as you can see, it’s not actually about coference resolution, per se. I include it in the group because essentially all they do is take their NAACL 2010 system and modify it slightly to perform a template-filling task for information extraction. That’s very similar to the kind of extension I’m looking at, though I’m taking a pretty different approach than they used.

This won the best paper award at NAACL that year, and I saw the presentation. I was pretty excited about the work then, and it looks like I still am, because I’m working on extending it. I have the code that ran this system, and I’m working on getting it to run with my data, then I can add some of my ideas to it.

This is the second of H&K’s coreference resolution papers. In a complete departure from their previous work, the system in this paper is completely deterministic and works with just a few syntactic and semantic cues. Basically, they come up with a candidate set of antecedents for each mention and select the one that is closest in the parse tree.