[Corpora-List] SemEval 2015 Task 18: Broad-Coverage Semantic Dependency Parsing

[Stephan Oepen]

[with apologies for cross-posting]

We would like to kindly invite participants to Task 18 at SemEval 2015:

  Broad-Coverage Semantic Dependency Parsing
    http://alt.qcri.org/semeval2015/task18

This task seeks to stimulate the statistical parsing community to move
towards graph-structured target representations, to facilitate semantic
dependency parsing, i.e. an analysis of ‘Who did What to Whom’.  In the
task setup, there is similarity to both data-driven dependency parsing
and semantic role labeling.  Besides practitioners from these ‘classic’
tasks, we further hope to attract participants with a general interest
in natural language processing in terms of graph-shaped structures.

Please see the attachment for a sample of semantic dependency graphs as
used in this task.  Here, the goal is to incorporate all content words
in one semantic dependency graph, where bi-lexical dependencies encode
predicate–argument relations, and argument sharing is reflected through
re-entrancies in the graph, for example representing control relations.

As training data, we provide about 750,000 tokens of English newspaper
text annotated with gold-standard semantic dependency graphs of three
different schools (as exemplified in the attachment: DM, PAS, and PSD).

All data is aligned at the sentence and token levels and distributed in
a uniform file format.  Test data will be drawn from the same genre, as
well as from a different domain and text type.  In addition to the core
task on semantic dependency parsing for English, we will offer optional
tasks on (a) parsing Chinese, Czech, and Spanish, using the same types
of target representations and comparable volumes of training data, and
on (b) predicate (or semantic frame) disambiguation, for the DM and PSD
target representations.

The task offers three tracks: closed, open, and gold; participants are
free to submit to any non-empty subset of tracks.  In the closed track,
systems can be trained only on the data provided for the task, whereas
the open track encourages the use of additional resources, for example
a syntactic parser.  Finally, the gold track aims to shed light on the
role of syntax for this task, where idealized, gold-standard syntactic
analyses will be provided in a broad variety of frameworks.

Following the common SemEval 2015 schedule, training data is available
now, and evaluation will be in mid-December 2014.  Please see the task
web pages for further details:

  http://alt.qcri.org/semeval2015/task18

We ask all interested parties to self-subscribe to the mailing list for
this task: the subscription link is available from the above web page.

Please do not hesitate to contact the task organizers for questions or
clarifications, using the email address provided on the task web pages.

Dan Flickinger, Jan Hajič, Angelina Ivanova, Marco Kuhlmann,
Yusuke Miyao, Stephan Oepen, and Daniel Zeman
-------------- next part --------------
A non-text attachment was scrubbed...
Name: samples.pdf
Type: application/pdf
Size: 33306 bytes
Desc: not available
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20140810/f443b6a1/attachment.pdf>
-------------- next part --------------
_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora