Another Look at the Ranking Problem in OT
Giorgio Magri (MIT)
Optimality Theory (OT) looks prima facie like a rather exotic combinatorial
framework, that does not seem to have any close correspondent within
Learning Theory. For this reason, various scholars are currently exploring
variants of standard OT that replace ''strict domination'' with ''additive
interaction'', and thus fall within the general class of ''linear
well studied in mainstream Learning Theory. Some recent examples of
this line of
research are Hayes & Wilson (2008), Coetzee & Pater (2008), Boersma & Pater
(2007, 2008)}, Goldwater & Johnson (2003), Potts et al. (2006), etcetera. In
this seminar, I will argue that this departure from standard OT is not needed.
In fact, I will present a simple trick that allows us to reinterpret within
standard OT many results and methods from the theory of linear models. I will
illustrate my proposal by concentrating on the case study of algorithms
Ranking problem that perform both promotion and demotion.