Connectionists: CFP: HLT-NAACL 2006 Workshop on Joint Inference for NLP

[Charles Sutton]

Call for Papers: Extended Deadline!
======================================================================
JOINT INFERENCE FOR NATURAL LANGUAGE PROCESSING

Workshop at HLT/NAACL 2006, in New York City
June 8, 2006
http://purl.oclc.org/NET/workshops/jinlp2006/

NEW SUBMISSION DEADLINE: March 8, 2006
======================================================================


In NLP there has been increasing interest in moving away from systems
that make chains of local decisions independently, and instead toward
systems that make multiple decisions jointly using global information.
For example, NLP tasks are often solved by a pipeline of processing
steps (from speech, to translation, to entity extraction, relation
extraction, coreference and summarization)---each of which locally
chooses its output to be passed to the next step.  However, we can
avoid accumulating cascading errors by joint decoding across the
pipeline---capturing uncertainty and multiple hypotheses throughout.
The use of lattices in speech recognition is well-established, but
recently there has been more interest in larger, more complex joint
inference, such as joint ASR and MT, and joint extraction and
coreference.

The trend toward joint decisions using global information also appears
at a smaller scale.  For example, the benefit of discriminative
reranking is that it can efficiently exploit global features of the
output space.  Also, recent sequence models, such as CRFs and
Maximum-margin Markov networks, are trained to optimize a global
objective function over the space of all sequences, leveraging global
features of the input.

The main challenge in applying joint methods more widely throughout
NLP is that they are more complex and more expensive than local
approaches.  Various models and approximate inference algorithms have
been used to maintain efficiency, such as beam search, reranking,
simulated annealing, and belief propagation, but much work remains in
understanding which methods are best for particular applications, or
which new techniques could be brought to bear.

The goal of this workshop is to explore techniques for joint
processing for NLP tasks that involve multiple, interrelated
decisions.  Themes of the workshop include:

* Practical examples of joint models in NLP.  Applications to
    traditionally hard NLP problems, including speech and machine
    translation, are encouraged.

* Inference methods for joint approaches, including message-passing
    algorithms, discriminative reranking, sampling methods, propagation
    of n-best lattices, and linear programming.

* What kinds of global features tend to have the most impact in joint
    approaches?

* An intriguing property of joint models is that they have the
    potential to integrate information from multiple sources,
    (e.g. top-down information helping low-level processing).  What
    kinds of higher-level information are useful in NLP tasks?

* Comparison of local methods for training and inference, such as
    those based on local classifiers, and global approaches such as CRFs
    and Maximum-margin Markov Networks.

* When is it appropriate to use a joint model, and when do simpler,
    more independent approaches suffice?

* Training techniques for joint approaches. Training local classifiers
    is often more efficient training global approaches, and sometimes it
    is possible to use local training, but joint decision-making at test
    time.  When are such hybrid techniques expected work well?  What are
    the trade-offs between accuracy and training time?

Potential participants are encouraged to submit papers on these  
topics, and
on others related to joint decision-making in NLP.


IMPORTANT DATES

* Paper submissions due:  Wednesday, March 8
* Notification of accepted papers: Thursday, April 21
* Camera ready papers due: Wednesday, May 3
* Workshop: June 8, 2006


FORMAT OF PAPERS

If you wish to present at the workshop, submit a paper of no more than
8 pages in two column format, following the HLT/NAACL style (see
http://nlp.cs.nyu.edu/hlt-naacl06/cfp.html).  Proceedings will be
published in conjunction with the main HLT/NAACL proceedings.
The web site for workshop submissions is
http://www.softconf.com/start/HLT-WS06-JINLP/submit.html
Authors who cannot submit a PDF file electronically should contact the
organizers.


ORGANIZERS

Charles Sutton, University of Massachusetts
Andrew McCallum, University of Massachusetts
Jeff Bilmes, University of Washington


PROGRAM COMMITTEE

Razvan Bunescu, University of Texas
Bill Byrne, University of Cambridge
Xavier Carreras,  Technical University of Catalonia
Ozgur Cetin, University of California
David Chiang, University of Maryland
Michael Collins, Massachusetts Institute of Technology
Hal Daume, University of Southern California
Eric Fosler-Lussier, The Ohio State University
Dan Gildea, University of Rochester
Ralph Grishman, New York University
Eric Horvitz, Microsoft Research
Katrin Kirchhoff, University of Washington
Philipp Koehn, University of Edinburgh
Shankar Kumar, Google
Chris Manning, Stanford University
Lluís Màrquez,  Technical University of Catalonia
Gideon Mann, University of Massachusetts
Erik McDermott,  NTT Communication Science Laboratories
Ray Mooney, University of Texas
Franz Och, Google
Kishore Papineni, IBM TJ Watson Research Center
Brian Roark, Oregon Graduate Institute
Dan Roth, University of Illinois
Salim Roukos, IBM TJ Watson Research Center
Koichi Shinoda, Tokyo Institute of Technology
Noah Smith, Johns Hopkins University
Andreas Stolcke, SRI International
Ben Taskar, Unversity of California