Micha Elsner

My thesis focused on discourse coherence-- the way a document or conversation is structured to provide context for new information. I constructed models looking at where and how entities (things in the world) are mentioned in a text. I also showed that these models can be used to disentangle the different threads of conversation going on in a crowded chat room.

I got my Phd. from Brown University in 2011, advised by Eugene Charniak, with Mark Johnson and Regina Barzilay as committee members. At Brown, I worked in the Brown Laboratory for Linguistic Information Processing (BLLIP).

I graduated from the University of Rochester in 2005 with degrees in Computer Science and Classics. I got my MS from Brown in 2007.

Publications

• Micha Elsner and Deepak Santhanam. Learning to Fuse Disparate Sentences. Workshop on Monolingual Text-to-Text Generation (T2T 2011), Portland Oregon. [PDF][Slides (PDF)]
• Micha Elsner and Eugene Charniak. Extending the Entity Grid with Entity-specific Features. Proceedings of the Association for Computational Linguistics (ACL 2011), Portland, Oregon. [PDF] [Slides (PDF)]
• Micha Elsner and Eugene Charniak. Disentangling Chat with Local Coherence Models. Proceedings of the Association for Computational Linguistics (ACL 2011), Portland, Oregon. [PDF] [Poster (PDF)] [Thesis defense slides (PDF)]
• Micha Elsner and Eugene Charniak. Disentangling Chat. Computational Linguistics 36(3), September 2010. [PDF]
• Micha Elsner and Eugene Charniak. The Same-head Heuristic for Coreference. Proceedings of the Association for Computational Linguistics (ACL 2010), Uppsala, Sweden. [PDF] [Poster (PDF)] [Slides (PDF)]
  (This is a short version of the same-head work.)
• Micha Elsner and Eugene Charniak. The Same-head Heuristic for Coreference. Northeast Student Conference on Artificial Intelligence (NESCAI 2010), Amherst, Massachusetts. [PDF] [Poster (PDF)]
  (This is a full-length version of the same-head work.)
• Micha Elsner and Warren Schudy. Bounding and Comparing Methods for Correlation Clustering Beyond ILP. NAACL-HLT 2009 Workshop on Integer Linear Programming for Natural Language Processing (ILP-NLP 2009), Boulder, Colorado. [PDF] [Slides (PDF)]
• Micha Elsner, Eugene Charniak, and Mark Johnson. Structured Generative Models for Unsupervised Named-Entity Clustering. Proceedings of the Conference on Human Language Technology and North American chapter of the Association for Computational Linguistics (HLT-NAACL 2009), Boulder, Colorado. [PDF] [Slides (PDF)]
• Eugene Charniak and Micha Elsner. EM Works for Pronoun Anaphora Resolution. Proceedings of the Conference of the European Chapter of the Association for Computational Linguistics (EACL 2009), Athens, Greece. [PDF]
• Micha Elsner and Eugene Charniak. You Talking to Me? A Corpus and Algorithm for Conversation Disentanglement. Proceedings of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2008), Columbus, Ohio. [PDF] [Slides (PDF)]
• Micha Elsner and Eugene Charniak. Coreference-inspired Coherence Modeling. Proceedings of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2008), Columbus, Ohio. [PDF] [Poster (PDF)]
• Micha Elsner, Joseph Austerweil, and Eugene Charniak. A Unified Local and Global Model for Discourse Coherence. Proceedings of the Conference on Human Language Technology and North American chapter of the Association for Computational Linguistics (HLT-NAACL 2007), Rochester, New York. [PDF] [Slides (PDF)]
  Note: this publication contains a bug affecting development results. A short explanation has been attached to the beginning of the PDF.
• Eugene Charniak, Mark Johnson, Micha Elsner, Joseph Austerweil, David Ellis, Isaac Haxton, Catherine Hill, Shrivaths Iyengar, Jeremy Moore, Michael Pozar, and Theresa Vu. Multilevel Coarse-to-fine PCFG Parsing. Proceedings of the Conference on Human Language Technology and North American chapter of the Association for Computational Linguistics (HLT-NAACL 2006), Brooklyn, New York. [PDF] [Slides (PDF)]
• Micha Elsner, Mary Swift, James Allen and Daniel Gildea. Online Statistics for a Unification-Based Dialogue Parser. Proceedings of the Ninth International Workshop on Parsing Technologies (IWPT 2005), Vancouver. [PDF] [Poster (PDF)]
• Thomas Kollar, Jonathan Schmid, Eric Meisner, Micha Elsner, Diana Calarese, Chikita Purav, Chris Brown, Jenine Turner, Dasun Peramunage, Gautam Altekar and Victoria Sweetser. Mabel: Extending Human Interaction and Robot Rescue Designs. AAAI Mobile Robot Competition 2003: Papers from the AAAI Workshop (ed. Smart, Smart, Bugajska), Acapulco. [PDF]

Thesis

• Generalizing Local Coherence Modeling. Brown University, January 2011. [PDF] [Defense slides (PDF)]
  Primarily based on work from NAACL-07, ACL-08a and b, ILP-NLP-09, CL-10, and ACL-11a and b.

Tech Reports

• Micha Elsner and Eugene Charniak. A Generative Discourse-New Model for Text Coherence. Technical Report CS-07-04, Brown University. [PDF]

Talks

• Dialogue Structure in Microtext. Invited talk, AAAI-11 Workshop on Analyzing Microtext, Aug. 8, 2011, San Francisco. [Slides (PDF)]
• Generalizing Local Coherence Modeling. Thesis defense, Jan. 10, 2011, Brown Univ. (Also delivered as ILCC Seminar, Feb. 4, 2011, University of Edinburgh.) [Slides (PDF)]
• Learning to Fuse Disparate Sentences. Invited talk, Nov. 15, 2010, Columbia Univ. (extended version of the PIRE talk) [Slides (PDF)]
• Learning to Fuse Disparate Sentences. PIRE grant meeting, July 15, 2010, Uppsala. [Slides (PDF)]
• Debugging Samplers: Making MCMC Work in Practice. Tutorial for Machine Learning Reading Group, Jul. 8, 2010, Brown Univ. [Slides (PDF)] [Example code]
• Reference Patterns for Discourse Coherence. Thesis proposal, May 10, 2010, Brown Univ. [Slides (PDF)]
• The Same-head heuristic for Coreference. Invited talk, Jan. 20, 2010, Stanford Univ. [Slides (PDF)]
• Learning Maximum-entropy Models of Salience via EM. Pattern theory reading group, Sept. 30, 2009, Brown Univ. [Slides (PDF)]
• Entity-based Coherence: Going Off the Grid. Invited talk, Mar. 4, 2009, Univ. of Pennsylvania. [Slides (PDF)]
• The Dangling Conversation: A Corpus and Algorithm for Conversation Disentanglement (extended version of ACL 2008 talk). Invited talk, Jan. 21, 2009, Univ. of Maryland. [Slides (PDF)]
• Given/New Information and the Discourse Coherence Problem. Invited talk, Oct. 10, 2007, MIT. [Slides (PDF)]

Software

• Brown Coherence Toolkit: software for a variety of local coherence models, now including the extended entity grid, and test applications for ordering and chat disentanglement (C++).
  New version 1.0 as of 2011!
  [Bitbucket]
• Sentence Fusion Software: software for preprocessing, training and running our English sentence fusion system (C++/Python). By me and Deepak Santhanam.
  [Bitbucket]
• Waterworks: Python utility package, including ClusterMetrics library for evaluating clusterings. Mostly by David McClosky.
  [Python Package Index]
• Correlation Clustering System: framework for creating and analyzing datasets (Python), heuristic solvers, LP, ILP and SDP bounding systems (C++).
  This is the release version; the evaluation code requires Waterworks.
  README, [tgz]
  You may also want the data matrices we constructed for 20 newsgroups: [tgz]
• Unsupervised Pronoun Anaphora System: EM learner, pre-trained model (newswire) and pronoun resolver (C++).
  Eugene wrote this software, so although I'm pleased to answer questions on it, I don't know the gory innards in detail.
  [tgz]
  If you're planning to use this software, you should consider Shane Bergsma's NADA non-referential pronoun detector as a preprocess. The reported results demonstrate significant improvements over our built-in non-referential detector.
• IRC Chat Data and Disentanglement Model: annotated IRC chat data, annotation software (Java), analysis and disentanglement model (Python).
  README, [tgz]

Teaching

• Guest lecture, Human and Machine Learning: Machine learning doesn't work: the no-free-lunch theorem, the curse of dimensionality, bias and variance of estimators.
• TA for CS 1460: Introduction to Computational Linguistics:
  prepared assignment 1 (language modeling), helped write midterm/final questions
  guest lectures: Expectation/Maximization for IBM model 1; EM using the Inside/Outside algorithm for PCFGS; Improving PCFGs with tree annotation and state-splitting
  grading, office hours and support
• Sheridan Center teaching certificate 1

People I've Worked With

• When I started out at Rochester, I worked as an annotator for Joel Tetreault.
• My undergraduate research was supervised by James Allen and Mary Swift.
• Joe Austerweil worked with me when he was an undergrad at Brown. He's now at Berkeley working for Tom Griffiths.
• In spring 2008, I studied at Universität des Saarlandes under the PIRE grant. I did some work (which I'm currently not pursuing) with Caroline Sporleder on event detection in the GENIA corpus.
• I've done some work on correlation clustering with my housemate Warren Schudy.
• I've worked on sentence fusion with Deepak Santhanam, leading to our T2T 2011 paper.
• I did some work on TSGs with Ben Swanson.

Service

• Faculty-Graduate Liason (FGL), Brown CS, 2008-9.
• Brown/MIT/Harvard NLP 2009 Summit co-organizer
• 2006-7: Leader of Machine Learning Reading Group
• Reviewing:
  • NAACL 2009-10
  • NAACL 2010 student research workshop
  • ACL 2008-9
  • ACL 2010-11 student research workshop
  • AAAI 2010
  • EMNLP 2010-11
  • JAIR
  • COLING 2008, 2010
  • NESCAI 2007-8, 2010
  • ICML 2007 co-reviewer
  • AISTATS 2011
  • IJCAI 2011 senior PC member
  • AAAI-11 workshop on microtext, PC

melsner0@gmail.com

Informatics Forum

10 Crichton St.

Edinburgh, UK EH8 9AB

melsner0 (skype)