Description

This course provides a unified view of grammars, automata, parsing and
inference from HMMs and finite state machines, through (P)CFGs to MaxEnt
or log-linear models for context-sensitive grammar.   It covers both
categorical and stochastic approaches.  The course starts with finite
state machines and HMMs, then moves to (Probabilistic) Context-Free
Grammars and (Stochastic) Push-down Automata, and ends with MaxEnt or
log-linear models and Context-Sensitive Grammars.

It also covers inference methods for the stochastic approaches, focusing
mainly on Maximum Likelihood estimation.  The course covers estimation
from visible training data, and estimation from partially labelled and
unlabelled training data using Expectation Maximization, including the
Forward-Backward Algorithm and the Inside-Outside Algorithm.