|
|
|
Natural Language Applications
Hosted in conjunction
with the 11th Conference of the European Chapter of
the Association for Computational Linguistics Motivations
and Aims
Language
processing largely deals with multidimensional and highly structured forms of
information. Indeed, from the morphological up to the deep syntactic and semantic
levels, linguistic information is often described by structured data, making
the learning of the associated linguistic tasks more complex. Traditional
methods for the design of language applications involve the extraction of
features that map data representations to vectors of attributes/values.
Unfortunately, there is no methodology that helps in this feature modeling
problem. Consequently, in order to encode structured data, the designer has
to rely on his/her deep knowledge, expertise and intuition about the
linguistic phenomenon associated with the target structures. Recently,
approaches that attempt to alleviate such modeling complexity by directly
encoding structured data have been developed. Among other, kernel methods and
conditional random fields provide interesting properties. The former use
kernel functions to implicitly define richer feature spaces (e.g.
substructure spaces) whereas the latter allow the designer to directly encode
the probabilistic model on the structures. The promising aspects of such
approaches open new research directions: (a) the study
of their impact on the modeling of diverse natural language structures, (b)
their comparative assessment with traditional attribute-value models and (c)
the investigation of techniques which aim to improve their efficiency. Additionally,
the complementary study of mapping the classification function in structured
spaces is very interesting. Classification functions can be designed to
output structured data instead of simple values. In other words, the output
values may be interpreted as macro-labels which describe configurations and
dependencies over simpler components, e.g. parse trees or semantic
structures. ObjectivesThe main goal of this workshop is
to bring together researchers from different communities such as machine
learning, computational linguistics, information retrieval and data mining to
promote the discussion and development of new ideas and methods for the
effective exploitation of "structured data" for natural language
learning and applications. These latter include but are not restricted to:
We are particularly interested in
the following machine learning aspects:
Registration Information
on registration and registration fees are provided at the conference web
page. Final Program
Workshop Proceedings
Workshop ChairsRoberto Basili and
Alessandro
Moschitti (University of
Rome ”Tor Vergata”) Program Committee Nicola Cancedda (Xerox Research Centre
Europe, France) Nello Cristianini (University of California,
Davis , USA) Aron Culotta (University of Massachusetts
Amherst, USA) Walter Daelemans ( Marcello Federico ( Attilio Giordana ( Marko Grobelink (J. Stefan Institute, Fred Jelinek ( Thorsten Joachims ( Lluis Marquez (Universitat Politecnica de Catalunya, Spain) Giuseppe Riccardi (University of Trento, Italy) Dan Roth ( Alex Smola (National ICT Australia, ANU) Carlo Strapparava (ITC-Irst, Italy) John Shawe Taylor ( Ben Taskar ( Dimitry Zelenko (SRA international inc., USA) Proceedings and
Publications
Contacts have been established
with an International Publisher for the production of a Special Issue on an
International Journal or for a Book dedicated to the Workshop Topics. This
will include the extended versions of selected papers from the Workshop
Proceedings. Details on the post-workshop publication will be provided as
soon as possible on these pages.
Important dates
Further Information For any information, please contact: Alessandro
Moschitti Dept. of Computer Science, Systems and tel: +39 06 72597333
|
|
|||||||||||||||||
|
|
|
|
||||||||||||||||||
