Fwd: CfP Semantics of the Long Tail workshop at IJCNLP 2017

It is a pleasure to invite you to the workshop SLT-1: Semantics of the Long
Tail
<https://www.google.com/url?q=http://www.understandinglanguagebymachines.org/semantics-of-the-long-tail/&sa=D&ust=1494255281396000&usg=AFQjCNFochPRp7OrC3VbVM4RzzET9iVxNg>.
The workshop is co-located with the IJCNLP 2017
<https://www.google.com/url?q=http://ijcnlp2017.org&sa=D&ust=1494255281397000&usg=AFQjCNFasdrWAuDFo747FTmoHAIFwd47tg>
conference,
which is organized by The Asian Federation of Natural Language Processing.
IJCNLP 2017 will take place in Taipei, Taiwan from the 27th of November to
the 1st of December 2017.

The use of language expressions and their meanings follow a Zipfian
distribution, featuring a small amount of very frequent observations and a
very long tail of low frequent observations. As a result of this, the data
we use for learning and testing systems also exhibit overfitting to the
head of the distribution. It is thus not surprising that statistical
approaches automatically exploit the distributional preference and
dominance of the most ‘popular’ interpretations for disambiguation and
reference tasks in NLP: most of the test interpretations in a task
correspond to the majority interpretations of the training instances. But
what about the long-tail cases? How well are systems capable of
semantically interpreting less and low-frequent cases? Surprisingly, humans
do not experience significant problems determining that a long-tail
interpretation applies in a specific text. But how can semantic NLP systems
intelligently deal with sparse cases? Little attention has been devoted to
how systems should solve interpretation tasks for local and perhaps unknown
event and entity instances, which are described in sparse documents, but
might not be observed in any training set or knowledge base. Potentially,
this would require new representations and deeper processing than the ones
that work well on the head, which involves reading between the lines, e.g.
textual entailment and robust (common sense) reasoning.

It is the goal of this workshop to bring researchers together to share
their experiences and results in dealing with long tail semantics: i.e.
interpretation of low-frequent or rare events, entities and relations for
which there is little training data. We welcome papers that describe:
(error) analyses of semantic tasks with respect to head and long-tail
distributions, evaluation of different methods on head and long-tail cases,
new methods to interpret long-tail phenomena, and the role of context to
prime long-tail cases over head interpretations. We are interested in how
knowledge and data can be acquired to counterbalance popular data and
interpretations and how to make semantic tasks sustainable over time when
the world and the interpretation space changes.


Topics of interest

Topics include, but are not limited to, the following:

   1. System Performance


   1. How can we define the head and the tail for each semantic NLP task?
   2. Which evaluation metrics are needed to gain insight into system
   performance on the tail?
   3. Do existing datasets suffice to gain insight into tail performance?
   What kind of benchmarks are needed to better track progress in processing
   the tail?


   1. Data and Knowledge Requirements


   1. What kind and amount of (annotated) data is needed?
   2. Do customary knowledge sources (e.g. DBpedia, BabelNet, and WordNet)
   suffice?
   3. Do we need massive local knowledge resources to represent the world
   and all its contexts?


   1. Methods and linguistic representations


   1. Are the methods and representations needed for the tail different
   from the ones for the head?
   2. How can we transfer models developed for the head to make them
   appropriate for modeling the tail?
   3. How can the recent advances in deep neural networks and matrix
   factorization be directed to accomplish this?


   1. Contextual adaptation


   1. How to build systems that can switch between contexts of time, topic,
   and location (e.g. how to build systems that can adapt to new or past long
   tail realities)?


Guide for authors
The deadline to submit papers is September 5, 2017.
Paper submissions for IJCNLP will be handled by the Softconf START system
(submission link will be provided once available).
The program chairs will release both Latex and Microsoft Word templates
soon.


Prospective authors should submit an extended abstract of 2 pages in
length, excluding references, and will be asked to extend it to a long or
short paper upon acceptance. For more information on short papers (5 pages
of content + 2 pages for references) and long papers (9 pages of content +
2 pages for references), we refer to: http://ijcnlp2017.org/site
/page.aspx?pid=148&sid=1133&lang=en
<https://www.google.com/url?q=http://ijcnlp2017.org/site/page.aspx?pid%3D148%26sid%3D1133%26lang%3Den&sa=D&ust=1494255281412000&usg=AFQjCNGDvsG-y2sdBbPf6DrNqTgUAOy3qQ>



Important Dates
Deadline for submission: September 5, 2017
Notification of acceptance: September 30, 2017
Deadline for final paper submission: October 10, 2017

Workshop Chairs

Initiators:

Piek Vossen (Vrije Universiteit Amsterdam)

Marten Postma (Vrije Universiteit Amsterdam)

Filip Ilievski (Vrije Universiteit Amsterdam)

Co-organizers:

Martha Palmer (University of Colorado Boulder)

Chris Welty (Google)

Ivan Titov (University of Edinburgh)

Eduard Hovy (Carnegie Mellon University)

Eneko Agirre (University of the Basque Country)

Philipp Cimiano (University of Bielefeld)

Frank van Harmelen (Vrije Universiteit Amsterdam)

Key-Sun Choi (Korea Advanced Institute of Science and Technology)

For any enquiries regarding the programme, please contact: m.c.postma@vu.nl
 or f.ilievski@vu.nl
For all general enquiries, please contact: m.c.postma@vu.nl or f
..ilievski@vu.nl

We look forward to seeing you at SLT-1.

Sincerely,
The Organizing committee of SLT-1