<OT> New Posting: ROA-625

[roa at ruccs.rutgers.edu]

ROA 625-1003

Maximum Entropy Models and Stochastic Optimality Theory

Gerhard Jaeger <jaeger at zas.gwz-berlin.de>

Direct link: http://roa.rutgers.edu/view.php3?roa=625


Abstract:
The paper compares Stochastic OT (StOT) in the sense of
Paul Boersma's work with Maximum Entropy (ME) models. I
show that Boersma's Gradual Learning Algorithm (GLA) for
StOT is applicable to ME models as well. The combination
of the GLA with ME models is similar to StOT as far as empirical
coverage is concerned, but has several theoretical advantages.
Most importantly, GLA is guaranteed to converge if combined
with ME evaluation.

Comments: 
Keywords: stochastic OT
Areas: Computation,Learnability,Formal Analysis
Type: Manuscript

Direct link: http://roa.rutgers.edu/view.php3?roa=625