On-ramp

From NLPWiki

... to the ALFA project.

Contents 1 April 28, 2009 2 May 5, 2009 3 May 12, 2009 4 May 19, 2009 5 May 26, 2009 6 May 31 - June 6, 2009 7 June 9, 2009 8 June 16, 2009

April 28, 2009

(continued on 4/30/09)

• Topic: Welcome
• Topic: Intro. to Probability Theory
  • Presenter: Eric Ringger
• Reading assignment: Manning & Schuetze 2.1, 2.2, 3, 4
• Reading assignment: Russell & Norvig 14.1-14.4
• Homework: https://cswiki.cs.byu.edu/cs479/index.php/Homework_0.1
• Optional homework: https://cswiki.cs.byu.edu/cs479/index.php/Homework_0.2

May 5, 2009

• Topic: Word Sense Disambiguation as motivation for Feature Engineering
  • Presenter: Eric Ringger
• Topic: Feature Engineering Console
  • Presenter: Josh Hansen
• Topic: Maximum Entropy Models
  • Presenter: Peter McClanahan
• Reading assignment: M&S 7, M&S 16
• Optional reading assignment: Berger's MaxEnt tutorial
• Homework: https://cswiki.cs.byu.edu/cs479/index.php/Project_2.2
  • BUT: Use the Feature Engineering Console! http://nlp.cs.byu.edu/mediawiki-private/index.php/Feature_Engineering_Console (on the Private wiki -- BYU NLP only -- requires authentication)
  • Write as little extra code as possible. Possible exceptions: new feature templates/extractors.
  • Work with Josh Hansen if you want to improve the FEC itself.

May 12, 2009

• Topic: Active Learning
  • Presenter: Robbie Haertel
• Reading assignment: Survey of Active Learning by Burr Settles
• Homework:
  • Implement one active learning selection function
  • Reference: http://nlp.cs.byu.edu/mediawiki/index.php/Using_the_active_learner
  • Plot learning curve for chosen function, versus random, using Gnuplot or Excel
  • Work with Robbie Haertel to bring the plotting code back to life

May 19, 2009

• Topic: Sequence Labeling
  • Presenter: George Busby
• Reading assignment: M&S 9, M&S 10
• Reading assignment: Paper by Toutanova & Manning on MEMMs
• Optional reading assignment: Paper on TnT by Brants
• Homework:
  • Continue on Active Learning experiments
  • Focus on PNP classification task
    • Plot means of multiple (around 5) random runs over # of iterations.
    • Would be interesting to plot variances of multiple random runs over # of iterations.
  • Try POS tagging task with small batch size B and number of iterations N, such that N x B is approx. 300

May 26, 2009

• Topic: Intro. to StatNLP Code-base
  • Presenter: Robbie Haertel
• Reading assignment: our ALFA Paper at the LAW
• Reading assignment: Tomanek et al., "An Approach to Text Corpus Construction which Cuts Annotation Costs and Maintains Reusability of Annotated Data"
• Homework: https://cswiki.cs.byu.edu/cs479/index.php/Project_3.1

May 31 - June 6, 2009

• NAACL HLT 2009
• Workshop on Active Learning for NLP

June 9, 2009

• Topic: Named Entity Recognition
  • Presenter: Eric Ringger
• Reading assignment: Klein et al. paper from 2003: "Named Entity Recognition with Character-Level Models"
• Reading assignment: McCallum et al. paper in CoNLL 2003: "Early Results for Named Entity Recognition with Conditional Random Fields, Feature Induction and Web-Enhanced Lexicons"
• Reading assignment: Ratinov and Roth paper in CoNLL 2009: "Design Challenges and Misconceptions in Named Entity Recognition"
• Homework:
  • Data: CoNLL 2003 Named Entity shared task data set: http://www.cnts.ua.ac.be/conll2003/ner/
  • Baseline: dictionary look-up method on CoNLL named entity recognition shared task data
    • dictionary is simply the list of named entities extracted from the training set
  • Baseline: MEMM for Named Entity Recognition on the CoNLL data
    • Improve on this by doing error analysis and feature engineering, as you did for the POS tagging task
  • Run both methods (dictionary look-up and MEMM) on noisy OCR data
  • Coordinate with Thomas Packer for noisy OCR data (esp. the labeled dev test set)
    • Private wiki site for the noisy OCR data: http://nlp.cs.byu.edu/mediawiki-private/index.php/Ancestry_dot_Com
  • Pick one 3rd-party tool (distinct from other students) from the list of open source tools on Wikipedia: http://en.wikipedia.org/wiki/Named_entity_recognition
    • Prefer one of the following:
      • Stanford Named Entity tagger
      • CCG group at UIUC: Named Entity + semantic role-labeling tagger
      • Mallet from U. Mass. Amherst
  • Run 3rd-party tool on CoNLL data and noisy OCR data
  • Report results

June 16, 2009

• Topic: User Study and Regression Results
  • Presenter: Kevin Seppi
  • Reading assignment: our LREC Paper
  • Reading assignment: Shilpa Arora et al.'s paper at ALNLP 2009
• Homework:
  • Repro. regression results from LREC paper using R
  • Apply methods from Arora's study to our data using SVM Regression
    • over-all cost model
    • per-subject cost model
    • per-subject-type cost model