Can anyone point me to the publications describing the techniques used by the currently available parsers to parse ranking? I know that we have mainly two methods for treebanking: the classic and the FFTB. But in both cases, treebanks provided data for training a ranking model, right? What are the methods used for that? From the ACE parameter --maxent I infer that it uses a maximum entropy model. What about LKB? What paper described the maxent implementation in ACE?
I don’t know enough to be very specific but I think, publications by Kristina Toutanova and coauthors, such as this one.
Thank you!!!
There is some information here:
http://wiki.delph-in.net/LogonModeling
I think Erik’s thesis (linked in the above page) is the best description. There are some more papers here:
http://wiki.delph-in.net/RedwoodsTop
We have some notes on how to train using the FFTB:
http://wiki.delph-in.net/IndraTreebanking
If you are interested in using semantics in the model we have some papers:
Xiaocheng Yin, Jungjae Kim, Zinaida Pozen and Francis Bond (2014)
Parse Ranking with Semantic Dependencies and WordNet. In Proceedings of the 7th Global WordNet Conference (GWC 2014) Tartu. pp 186–193
Sanae Fujita, Francis Bond, Takaaki Tanaka and Stephan Oepen (2010).
Exploiting Semantic Information for HPSG Parse Selection. Research on Language and Computation, 8(1), pp 1–22.
But these were never folded into a tool.
1 Like