[developers] Fwd: Treebanks with MRS
bond at ieee.org
Fri Jul 21 07:28:11 CEST 2017
I have been asked about the current state of non-English treebanks. I
have answered about Hinoki (9,128 treebanked sentences out of 15,000,
slightly old grammar). Are there any other treebanks people think it may
be interesting to learn grammars off?
I will try and summarize replies on the wiki.
---------- Forwarded message ----------
From: David R. Mortensen <dmortens at cs.cmu.edu>
Date: Thu, Jul 20, 2017 at 10:51 PM
Subject: Treebanks with MRS
To: bond at ieee.org, fcbond at ntu.edu.sg
We met at CMU earlier this year. It was very interesting to talk to you. It
turns out that I now have some HPSG/MRS related questions for you: I'm
participating in a JSALT workshop (https://www.lti.cs.cmu.edu/
2017-jelinek-workshop) on neural machine translation and we have become
interested in MRS as a semantic representation for NMT. For English, we
have a rich resource in form of DeepBank. We have found that ERG is not
robust enough to do our MRS parsing but Buys and Blunsom (
https://arxiv.org/abs/1704.07092) have demonstrated a robust
transition-based parser for MRS that they trained on a subset of DeepBank,
and we may do something similar.
We are now interested in other resources that contain significant amounts
of MRS. I noticed that you mention MRS in a paper on the Hinoko Treebank.
Is this treebank available? Do you know how many MRSs it contains? Do you
know of other treebanks that contain significant numbers of
hand-annotated/corrected MRS representations, especially for non-English
David R. Mortensen
Language Technologies Institute
Carnegie Mellon University
Francis Bond <http://www3.ntu.edu.sg/home/fcbond/>
Division of Linguistics and Multilingual Studies
Nanyang Technological University
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the developers