Lecturers: Khalil Sima'an and Tejaswini Deoskar
Objectives How to deal with uncertainty and lack of knowledge when building models of human cognitive capacities such as language processing. This includes: -Statistical models for language processing -Language models for speech processing -Models for ambiguity resolution in POS tagging and syntactic parsing -Technological applications such as Statistical Machine Translation.
Contents When computational models of language processing are not constructed in a purely linguistic context, but aim at being relevant for psychological theory or for practical applications, they ought to be able to perform tasks like disambiguation and prediction. For this reason, increasingly many models take statistical properties of a sample corpus into account when they process new input. This course will give an overview of the most important techniques used in statistical language processing. The course starts out by giving a short introduction to probability theory, information theory and Bayesian learning, and continues with the following topics: n-gram statistics and Markov models, smoothing techniques (Good-Turing and Katz Discounting), Hidden-Markov Models (HMMs), application of HMMs to part-of-speech tagging, Stochastic Context-Free Grammars (SCFGs), stochastic parsing algorithms, Bilexical-Dependency models and Data-Oriented Parsing (DOP) models. The course will also touch on basic methods from unsupervised statistical learning (Expectation-Maximization) and basics of Statistical Machine Translation.
The course will consist of lectures (including guests) and student presentations of research papers from a list of papers provided by the lecturer (see next).