Linguistics 575: MRS in Applications

Spring quarter, 2016

Course Info

Instructor Info



The English Resource Grammar (Flickinger 2000, 2011) is a broad-coverage precision grammar for English, written in HPSG (Pollard and Sag 1994) and producing semantic representations in the format of Minimal Recursion Semantics (Copestake et al 2005). It encompasses analyses of a wide range of phenomena in English, and a key piece of each analysis is the design of the resulting semantic representation. The MRS representations are built compositionally by the grammar and represent a significant abstraction away from the surface string.

The goal of this seminar is to explore how the MRS representations can be used to inform semantically-sensitive NLP tasks, such as anaphora resolution, event detection, or relation extraction. We will begin with an overview of MRS, and then move on to an exploration of candidate tasks and how to create machine learning features from MRSs to augment existing solutions to those tasks. Term projects (which may be done in pairs) will involve selecting an existing annotated data set for a semantically-sensitive task as well as an existing baseline solution and then attempting to improve on the baseline by adding MRS-based features.

Prereqs: This is a hands-on course that presupposes sufficient knowledge of NLP systems to work with and augment existing solutions. Students should have taken Ling 570 (or equivalent) and ideally also Ling 571/572 or be concurrently enrolled in those courses. Ling 566 may be beneficial, but is not required.

Note: To request academic accommodations due to a disability, please contact Disabled Student Services, 448 Schmitz, 206-543-8924 (V/TTY). If you have a letter from Disabled Student Services indicating that you have a disability which requires academic accommodations, please present the letter to the instructor so we can discuss the accommodations you might need in this class.


Schedule of Topics and Assignments (still subject to change)

3/30 Introduction, organization
Why use semantics?
The DELPH-IN ecology
Bender 2013: Ch 9
Bender et al 2015
4/4     Sample MRS output from the ERG
4/6 Minimal Recursion Semantics Copestake et al 2005
DELPH-IN wiki page on EDS
KWLA: K and W due
4/13 Syntactic features in shared tasks 2-3 papers from the list below under "Papers about Tasks", or others that you propose  
4/20 Target task/baseline system presentations Wicke Monteverde;
4/27 Target task/baseline system presentations Team BioNLP;
5/2     Target task/baseline system descriptions
5/4 Evaluation and Error Analysis Discussion prep  
5/9     Evaulation plans
5/11 MRS Feature Design Zhang et al (ms) (Sec 5 & 6)
Flickinger et al 2013: main page, Basics, then choose 2-3 phenomena pages to read
Kramer and Gordon 2014
5/18 MRS Feature Design Lien and Kouylekov 2015
Tanaka et al 2007, Packard et al 2014
5/23     Feature design
5/25 Term project presentations LaTerza; Preddy; Lane+Horn  
6/1 Term project presentations Garnick; Shintani; Wicke-Monteverde  
6/3     KWLA papers due
6/9     Final projects due 11pm


General background

Papers about tasks

Clinical Temp Eval

Risk Factor Identification (i2b2 2014)


Ontology Extraction

Robust Textual Entailment/Semantic Textual Similarity

Coreference resolution

Question Answering

Discourse Processing

Robust Textual Entailment/Semantic Textual Similarity

Sentiment Analysis


Word Sense Disambiguation

MRS feature design

ebender at u dot washington dot edu
Last modified: 4/7/16