COMS W4705: Natural Language Processing

[Main] | [General Information] | [Problem Sets]



Instructor: Michael Collins
Time & Location: Tues & Thurs 4.10-5.25, 1024 Mudd
Office Hours:Thursdays 2-3pm, CEPSR (Schapiro) 723

TAs: Please send all questions to nlpfall2012.columbia at gmail.
Alexander Rush [ar3148] (office hours Tuesday 1.30-2.30, in 701 CEPSR)
Yuan Du [yd2234] (office hours Monday 1-3, in TA room in Mudd)
Dhinesh Dhanasekaran [dkd2110] (office hours Wednesday 10-12, in TA room in Mudd)

Announcements:


Lectures:
Date Topic References
9/4 Introduction and Overview
9/6 Language Modeling Notes on language modeling (required reading)
9/11 Tagging, and Hidden Markov Models Notes on HMMs
(Required reading)
9/13 Tagging, and Hidden Markov Models (continued)
9/18 Parsing, context-free grammars, and probabilistic CFGs Note on PCFGs (required reading)
9/20 Parsing, context-free grammars, and probabilistic CFGs (continued)
9/25 Lexicalized probabilistic CFGs
9/25 Lexicalized probabilistic CFGs (continued) Note on Lexicalized PCFGs (required reading)
10/2 Guest lecture by Nizar Habash
10/4 Machine translation part 1 (Note: we didn't cover the final section, on evaluation using BLEU, but I've kept it in the slides in case it's of interest.)
10/9 Machine translation part 2 Note on IBM Models 1 and 2 (required reading)
10/11 Phrase-based translation models Note on phrase-based models (required reading)
Slides from the tutorial by Philipp Koehn
10/16 Phrase-based translation models: the decoding algorithm
10/18 Mid-term (in class)
10/23 Reordering for statistical MT
10/25 Log-linear models Note on log-linear models (required reading).
11/1 Log-linear tagging (MEMMs)
11/8 Global linear models
11/13 Global linear models part II
11/15 Global linear models part III
11/20 Guest lecture: Joint Decoding Tutorial on dual decomposition
11/27 The Brown word-clustering algorithm
11/29 Semi-supervised learning for word-sense disambiguation, and cotraining for named-entity detection
12/3 The EM algorithm for Naive Bayes Notes on the EM algorithm for Naive Bayes (Sections 4 and 6 provide useful technical background, but can be safely skipped.)