Computer Science 248/448
Statistical Speech and Language Processing
Fall 2017
|
|
Instructor:
Dan Gildea (Office hours: Mon 2:00-3:00pm, Wed 1:00-2:00pm or by appointment)
TA: Linfeng Song
Location: Tu/Th 11:05am-12:20pm, CSB 601.
Homeworks
Recommended resources:
Syllabus
| On | we will cover | which means that after class you will understand | if before class you have read |
| 8/30 |
Hidden Markov Models |
viterbi, forward
| JM 6 |
| 9/5 |
Perceptron for Tagging |
NLP = ML + DP
| Collins 2002 |
| 9/7 |
CRFs |
f - E[f]
| McCallum 2001 |
| 9/12 |
Max Margin Learning |
subgradients
| Tsochantaridis 2004, Ratliff 2006 |
| 9/14 |
EM for HMM |
| JM 6 |
| 9/19 |
latent CRF |
| |
| 9/21 |
N-grams |
Good-Turing, Katz, Kneser-Ney
| |
| 9/26 |
Context Free Grammars |
highest prob string, infinite trees
| JM 13 |
| 9/28 |
Context Free Grammars |
viterbi, inside-outside
| |
| 10/3 |
Context Free Grammars |
viterbi, inside-outside
| |
| 10/5 |
Parsing |
minimum risk decoding, crf parsing
| Finkel 08 |
| 10/10 |
Fall Break |
| |
| 10/12 |
Parsing |
em, latent crf, max-margin
| |
| 10/17 |
Treebanks |
right node raising
| JM 12 |
| 10/19 |
Search |
beam search, A*
| |
| 10/24 |
Dependency parsing |
| |
| 10/26 |
Review |
| |
| 10/31 |
Midterm |
|
|
| 11/2 |
Midterm Solutions |
| |
| 11/7 |
Machine Translation
| model 1, 2, 3
| Brown 1993 |
| 11/9 |
Machine Translation
| decoding
| Koehn et al. 2003 Galley 2004, Chiang 05 |
| 11/14 |
Neural Machine Translation |
LSTM
| Sutskever 2014, Bahdanau 2014 |
| 11/16 |
Machine Translation |
minimum risk training
| Papineni 2002, Shen 2016 |
| 11/21 |
Semantics |
word2vec
| Mikolov 2013, Pennington 2014 |
| 11/28 |
Semantics |
lambda calculus
| JM 20, Zettlemoyer 2005, Liang 2011 |
| 11/30 | Deep Learning |
neural symbolic machines
| Liang et al., 2017, |
| 12/5 |
Deep Learning for parsing |
tabularization
| Socher 2013, Shi 2017 |
| 12/7 |
Something fun |
expected products of counts
| Pauls 2009 |
| 12/12 |
no class |
| |
Final Exam: Thursday, December 21, 4-7pm, 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, 2017
|