Parsing/sdp
Background
The Semantic Dependency Parsing (SDP) initiative seeks to aid the development of data-driven parsing systems whose outputs are general graphs (i.e. transcend rooted trees). These target representations have become (somewhat) popular through two semantic dependency parsing competitions as part of the 2014 and 2015 Semantic Evaluation Exercises (SemEval). For additional background, please see the SDP web site and the companion site for one of the frameworks represented under the SDP umbrella.
Available Data
As of late 2018, there have been three public releases of the
SDP graphs.
(1) The full SDP collection includes annotations that are
derivative of the venerable Penn Treebank (PTB) and the Penn
Chinese Treebank (CTB) and, thus, is distributed via the
Linguistic Data Consortium as
LDC2016T10.
This version is installed as
.../data/parsing/sdp/2016T10/ and must only
be used by NLPL users whose home institution holds LDC
licenses to the PTB and CTB (which likely is true for
most NLPL partner sites).
Subsets of the SDP graphs that are free of historic
licensing constraints have been released as
Open SDP versions
(2) 1.1 and
(3) 1.2.
These fully open-source releases include the
English SDP graphs in the DELPH-IN MRS Dependencies (DM)
format and the Czech graphs in the Prague Semantic
Dependencies (PSD) format.
They are installed as
.../data/parsing/11/ and
.../data/parsing/12/, respectively.
