WORKSHOP: PRIOR KNOWLEDGE FOR TEXT AND LANGUAGE PROCESSING

9 July 2008, Helsinki, in conjunction with the ICML/UAI/COLT

Web page: http://prior-knowledge-language-ws.wikidot.com (please monitor this page for updates)

CONTEXT: The workshop is part of the Thematic Programme "Leveraging Complex Prior Knowledge for Learning" of the PASCAL-2 European Network of Excellence starting in March 2008.

GOALS: The aim of the workshop is to present and discuss recent advances in machine learning approaches to text and natural language processing that capitalize on rich prior knowledge models in these domains.

MOTIVATION: Traditionally, in Machine Learning, a strong focus has been put on data-driven methods that assume little a priori knowledge on the part of the learning mechanism. Such techniques have proven quite effective not only for simple pattern recognition tasks, but also, more surprisingly, for such tasks as language modeling in speech recognition using basic n-gram models. However, when the structures to be learned become more complex, even large training sets become sparse relative to the task, and this sparsity can only be mitigated if some prior knowledge comes into play to constrain the space of fitted models. We currently see a strong emerging trend in the field of machine learning for text and language processing to incorporate such prior knowledge for instance in language modeling (e.g. through non-parametric Bayesian priors) or in document modeling (e.g. through hierarchical graphical models). There are complementary attempts in the field of statistical computational linguistics (e.g in statistical machine translation) to build hybrid systems that do not rely uniquely on corpus data, but also exploit some form of a priori grammatical knowledge, bridging the gap between purely data-oriented approaches and the traditional purely rule-based approaches, that do not rely on automatic corpus training, but only indirectly on human observations about linguistic data. The domain of text and language processing thus appears as a very promising field for studying the interactions between prior knowledge and raw training data, and this workshop aims at providing a forum for discussing recent theoretical and practical advances in this area.

TOPICS: The workshop aims at presenting a diversity of viewpoints on prior knowledge for language and text processing. Discussion of the following topics, techniques and issues is encouraged (non-limitative):

* Prior knowledge for language modeling, parsing, translation

* Topic modeling for document analysis and retrieval

* Parametric and non-parametric Bayesian models in NLP

* Graphical models embodying structural knowledge of texts

* Complex features/kernels that incorporate linguistic knowledge; kernels built from generative models

* Limitations of purely data-driven learning techniques for text and language applications; performance gains due to incorporation of prior knowledge

* Typology of different forms of prior knowledge for NLP (knowledge embodied in generative Bayesian models, in MDL models, in ILP/logical models, in linguistic features, in representational frameworks, in grammatical rules)

* Formal principles for combining rule-based and data-based approaches to NLP

* Linguistic science and cognitive models as sources of prior knowledge

FORMAT: The workshop will consist of a mix of submitted papers, invited talks, and discussion/panels in which different viewpoints will be emphasized.

CALL FOR PAPERS: Researchers interested in presenting their work at the workshop should send an email (preferably plain text or pdf attachment) to ws_pktlp@xrce.xerox.com with the following information:

TITLE
AUTHORS
ABSTRACT (corresponding to approximately two plain text pages)

Note: In case you experience problem with the above email alias, please contact: marc (dot) dymetman (at) xrce (dot) xerox (dot) com

We expect speakers to provide a final version of their paper before end of June for inclusion on the workshop home page, and authors will be encouraged to read the included papers prior to the meeting. A compiled set of papers will be distributed as working notes at the workshop.

DATES:

Abstract submission deadline: 30 April 2008
Notification to authors: 15 May 2008
Final version: 30 June 2008
Workshop: 9 July 2008

INVITED PRESENTATIONS AND PANELISTS (partial list, TBC):

* David Blei
* Pedro Domingos
* Peter Grünwald
* Mark Johnson
* Dan Melamed
* Massimiliano Pontil

ORGANIZERS:

* Guillaume Bouchard: guillaume (dot) bouchard (at) xrce (dot) xerox (dot) com
* Hal Daumé III: hal (at) cs (dot) utah (dot) edu
* Marc Dymetman (main contact): marc (dot) dymetman (at) xrce (dot) xerox (dot) com
* Yee Whye Teh: yeewhye (at) gmail (dot) com