NLPwin parses AMR

Established: March 17, 2015

Publications

Overview

The Logical Form analysis produced by the NLPwin parser is very close in spirit to the level of semantic representation defined in AMR, Abstract Meaning Representation. The “NLPwin parses AMR” project is a conversion from LF to AMR in order to facilitate 1) evaluation of the NLPwin LF and 2) contribution the ongoing discussion of the specification of AMR. In this project, we include publications, as well as links to our LF training data converted to AMR and to the LF-AMR parser for English.

NLPwin AMR parser

We make the output of the NLPwin LF to AMR conversion for English available through MSR SPLAT, the statistical processing and language analysis toolkit. We also have parsers for French, German, Spanish, and Japanese available on request.

NLPwin Logical Form development file in AMR format

To support the development of Logical Form in NLPwin, we curated sets of sentences that include target phenomena that Logical Form is designed to handle. We share these development files below, both as sentence-only files, as well as files with AMR-similar output. The sentence file lf-all-good.txt includes only those sentences for which the LF produces the optimal output, which is then converted to AMR. Thus, the AMR output in these files is not manually annotated (“gold” annotation) and may therefore not conform to the current AMR spec in all regards, but might be considered AMR “silver” output, suitable for further manual annotation to be fully conformant with the AMR spec.

LF-all-good: A file containing sentences for which the LF is known to be good

  • Example: Sentences only with comments
  • Example: Human-readable AMR-like output produced by NLPwin
  • Example: Machine-readable AMR-like output produced by NLPwin

LF-almost-there: A file containing sentences for which the LF is mostly good but there are still some points:

  • Example: Sentences only with comments
  • Example: Human-readable AMR-like output produced by NLPwin
  • Example: Machine-readable AMR-like output produced by NLPwin

100 sentences sampled from XXXX, used in comparison published in Vanderwende, Menezes and Quirk (2016):

People