Connectionists: Letter-to-phoneme machine lerning challenge

[Bob Damper]

Connectionist may be interested in the following ...

*******

Dear PASCAL researchers,

Welcome to the Letter-to-Phoneme Conversion Challenge -- PRONALSYL

http://www.pascal-network.org/Challenges/PRONALSYL/
email:	rid at ecs.soton.ac.uk
	Yannick.Marchand at nrc-cnrc.gc.ca

Part of the EU Network of Excellence PASCAL Challenge Program.

Participation is open to all.

The objective of the Challenge is to advance the state of the art in
letter-to-phoneme conversion.  This `automatic pronunciation' problem is
not only important in its own right (e.g., as a component part of many
speech technologies) but has also served as a benchmark for machine
learning methodologies for many years (cf. Sejnowski and Rosenberg's
famous NETtalk). 

Although PASCAL is centrally concerned with machine learning (ML), we are
also very keen that at least some participants employ expert knowledge
based approaches (i.e., manually-written linguists' rules) to allow us
to compare the performance of the two very different paradigms.

This is not a competition in the sense of having winners and losers.
Rather the motivation is to learn from our joint efforts.  Hence, we
are keen for at least some participants to try out very simple methods
(e.g., naive Bayes classifier) that can be used as a `baseline' for
comparative purposes.

The scientific goals are:

* To understand better the nature of the conversion problem, how this
varies across languages, and how and why different machine learning
techniques might be more or less appropriate for this task.

* To improve on the best automatic pronunciation result(s) so far
achieved, either by use of a novel ML approach or by improving an existing
one (or perhaps using manually-written rules).

* To understand better the nature of strongly related problems
(letter-phoneme string alignment, syllabification, morphemic
decomposition, stress assignment, ...) that might impact on the
capabilities of a letter-to-phoneme conversion system.

* To advance machine learning methodology in general.

It is intended that discussion and preliminary results will be presented
in the Pascal Challenge Workshop in Venice (10-12 April 2006), and
final reporting will be at NIPS in December 2006.

Program Committee:

Walter Daelemans, University of Antwerp, Belgium
Bob Damper, University of Southampton, UK (Co-Chair)
Kjell Gustafson, Acapela and KTH, Stockholm, Sweden
Yannick Marchand, National Research Council, Canada (Co-Chair)
Francois Yvon, ENST, Paris, France

Please go to the website for further details of PRONALSYL and available
datasets:

	http://www.pascal-network.org/Challenges/PRONALSYL/

We are looking forward to an interesting challenge!

The Program Committee.