Computer Science 248/448

Statistical Speech and Language Processing

Fall 2015

Instructor: Dan Gildea (Office hours: Wed 2:00-3:00pm or by appointment CSB 609)
TA: Xiaochang Peng (CSB 627)
Location: Tu/Th 11:05am-12:20pm, CSB 601.

Homeworks

Text: Jurafsky and Martin, Speech and Language Processing, 2nd edition.

Syllabus

Onwe will coverwhich means that after class you will understandif before class you have read
9/1 Perceptron for Tagging NLP = ML + DP Collins 2002
9/3 Hidden Markov Models viterbi, forward JM 6
9/8 Variational Bayes digamma, expected log probs JM 5; Johnson 2007
9/10 CRFs f - E[f] McCallum 2001
9/15 N-grams Good-Turing, Katz, Kneser-Ney JM 4
9/17 Signal Processing for Speech MFCC JM 9
9/22 Forced Alignment Deep learning Hinton 2012
9/24 Large Vocabulary Speech Recognition beam search, multistack decoding JM 10
9/29 Max Margin Learning subgradients Tsochantaridis 2004, Ratliff 2006
10/1 Context Free Grammars highest prob string, infinite trees JM 13
10/6 Fall Break
10/8 Treebanks right node raising JM 12
10/13 Parsing cyk, earley, posterior decoding, inside-outside JM 14
10/15 Real-World Parsing lexicalization, markovization, reranking JM 15; Charniak 2005
10/20 Discriminative Methods tree CRF, early update perceptron Finkel 08, Huang 2012
10/22 Review
10/27 Midterm
10/29 Midterm Solutions
11/3 Machine Translation source-channel, model 1 JM 25
11/5 Machine Translation model 2, 3, hmm Brown 1993
11/10 Machine Translation decoding Koehn et al. 2003
11/12 Machine Translation syntax-based MT Galley 2004, Chiang 05
11/17 Machine Translation syntax-based decoding
11/19 Machine Translation Parameter Optimization Och 2003
11/24 Semantics lambda calculus JM 20, Zettlemoyer 2005
12/1 Semantics word2vec Liang 2011, Mikolov 2013
12/3 Deep Learning Socher 2013, Sutskever 2014
12/8 Something fun LSTM, image to text Greff 2015, Karpathy 2015
12/10 Review come to class with questions!
Final Exam: Saturday, December 19, 4pm, CSB 601.

Grading

  • Final exam: 35%
  • Homeworks: 35%
  • Midterm: 25%
  • Class participation: 5%
  • No late homework
  • Programming assignments must be in python, and must run on class account

gildea @ cs rochester edu
December 7, 2015