LEXICALIZED PARSING FOR DIFFERENT DOMAINS Laura Rimell and Stephen - PowerPoint PPT Presentation

Daniel C. Müller LEXICALIZED PARSING FOR DIFFERENT DOMAINS Laura Rimell and Stephen Clark 10.12.2009

Hypothesis 2 � parser adaption in the context of lexicalized grammar � � according to two different domains Daniel C. Müller - 10.12.2009

domains 3 � Biomedical domain � Questions of question answering Daniel C. Müller - 10.12.2009

Lexicalized parser 4 � POS-Tagging based on Penn Tree Bank � Combinatory Categorial Grammar + manual annotation Daniel C. Müller - 10.12.2009

Lexicalized parser 5 � POS-Tagging based on Penn Tree Bank � POS Tag: � 50 grammatical labels indicating part of speech � Each word one POS Tag Daniel C. Müller - 10.12.2009

Lexicalized parser 6 � POS-Tagging based on Penn Tree Bank � Combinatory Categorial Grammar + manual annotation � lexical categorization (super-tagger) � 425 categories � Each word at least one category � Containing subcategorial information � Complex categories like (S\NP)/NP means: returns S\NP when applied to NP Daniel C. Müller - 10.12.2009

Lexicalized parser 8 � POS-Tagging based on Penn Tree Bank � Combinatory Categorial Grammar + manual annotation � lexical categorization (super-tagger) � derivation (hierarchy) � Lexicalized categories + combinatory rules Viterbi packed chart representation best derivation Daniel C. Müller - 10.12.2009

Lexicalized parser 9 � Example I drink coffee NP (S\NP)/NP NP . (S\NP)/NP needs a NP to the right NP S\NP S\NP needs a NP to the left S Daniel C. Müller - 10.12.2009

Motivation 10 � creating new training data at the lower levels of representation � better POS tagging better categorization � reduce annotation overhead Daniel C. Müller - 10.12.2009

Experiments 11 � Training resources � Baseline � Wall Street Journal Sections 02-21 of CCGbank � Biomedical domain � POS tagger: gold-standard POS tags from GENIA � Lexical categories: � rst1,000 sentences of GENIA � parser evaluation: BioInfer � Evaluation set: Pyysalo et al. (2007b) � Question domain � Questions beginning with the word What, from the TREC 9-12 competitions: manually POS tagged & annotated with lexical categories Daniel C. Müller - 10.12.2009

Experiments 12 � Results � POS-Tagger % WSJ 02-21 Retrained Sec.00 96.7 - Biomedical 93.4 98.7 Question 92.2 97.1 Daniel C. Müller - 10.12.2009

Experiments 13 � Results � Supertagger % Original Retrained Retrained pipeline POS POS & Super Sec.00 91.5 - - Biomedical 89.0 91.2 93.0 Question 71.6 74.0 92.1 Daniel C. Müller - 10.12.2009

Experiments 14 � Results � Parser evaluation % Original new new pipeline POS POS & Super Biomedical 76.0 80.4 81.5 Question 64.4 69.4 86.6 Daniel C. Müller - 10.12.2009

Analysis 15 � Comparing to WSJ: � Biomedical domain: + similar syntactic structure � vocabulary & foreign words � long noun phrases � Question domain: + vocabulary � words with different distribution of POS in source domain � different syntactic structure Daniel C. Müller - 10.12.2009

Analysis 16 � POS tagging � Biomedical domain: � nouns and adjectives (801 NN + 268 JJ errors) very long noun phrases and unknown words like � major histocompatibility complex class II molecules � � Question domain: � wh-determiners (129 errors) only one occurrence in WSJ 02-21 Daniel C. Müller - 10.12.2009

Analysis 17 � POS tagging � Biomedical domain: � nouns and adjectives (801 NN + 268 JJ errors) very long noun phrases and unknown words � Question domain: � wh-determiners (129 errors) (S/(S/NP))/N: � What Liverpool club spawned the Beatles? � S/(S\NP) : � What are the colors of the German � ag? � much more errors but related syntactic structure Daniel C. Müller - 10.12.2009

Analysis 18 � Syntactic differences � Unknown POS n-gram rate % WJS 02-21 New training sets 3-grams 5-grams 3-grams 5-grams Sec.00 0.4 12.1 - - Biomedical 0.7 10.9 0.5 9.2 Question 3.6 22.0 0.7 7.4 Daniel C. Müller - 10.12.2009

Analysis 19 � Syntactic differences � Unknown POS n-gram rate � Number of 20 most frequent POS n-grams 3-grams 5-grams Sec.00 18 19 Biomedical 16 13 Question 8 5 Daniel C. Müller - 10.12.2009

Analysis 20 � Syntactic differences � Unknown POS n-gram rate � Number of 20 most frequent POS n-grams � POS Trigrams � Biomedical domain: � Domination of NPs and PPs � Question domain: � Beginning with WP VBZ like � What is � � Ending with VB . � Daniel C. Müller - 10.12.2009

Analysis 21 � Syntactic differences � Unknown POS n-gram rate � Number of 20 most frequent POS n-grams � POS Trigrams � Number of rare or unseen lexical categories Daniel C. Müller - 10.12.2009

Conclusion 22 � Biomedical domain � Question domain � need for accurate parsing � long and difficult sentences � uniform sentences � many POS tag errors � less related syntax with supertagging with POS tagging parser adaption successful! Daniel C. Müller - 10.12.2009

References 23 � Laura Rimell, Stephen Clark. 2008. Adapting a Lexicalized-Grammar Parser to Contrasting Domains. EMNLP 2008. � Julia Hockenmaiers. 2007. Expressive Grammar Formalisms for Natural Language: Theory and Applications. Lecture 16: Extracting a CCG from the Penn Treebank . � Julia Hockenmaiers. 2005 CCGBank User � s Manual Daniel C. Müller - 10.12.2009

LEXICALIZED PARSING FOR DIFFERENT DOMAINS Laura Rimell and Stephen - PowerPoint PPT Presentation

Daniel C. Mller LEXICALIZED PARSING FOR DIFFERENT DOMAINS Laura Rimell and Stephen Clark 10.12.2009 Hypothesis 2 parser adaption in the context of lexicalized grammar according to two different domains Daniel C. Mller

Introduction to Bottom-Up Parsing Shift-reduce parsing The LR parsing algorithm

CSC 4181 Compiler Construction Parsing 1 1 Outline Top-down v.s. Bottom-up Top-down parsing

Lexicalized Probabilistic Context-Free Grammars Michael Collins, Columbia University Overview

Some Experiments on Indicators of Parsing Complexity for Lexicalized Grammars Anoop Sarkar, Fei

Overview Introduction Lexicalized TAG, Advantages of parsing with LTAG Parsing LTAGs

Last class Dependency parsing and logistic regression Dependency parsing: a fully lexicalized

Robust Incremental Neural Semantic Graph Parsing Jan Buys and Phil Blunsom Dependency Parsing vs

Basic Parsing Algorithms Chart Parsing Seminar Recent Advances in Parsing Technology WS

Data-oriented Parsing with Lexicalized Tree Insertion Grammars Gnter Neumann LT-lab, DFKI

Natural Language Processing Other Syntactic Models Parsing IV Dan Klein UC Berkeley Dependency

Simple, Lexicalized Choice of Translation Timing for Simultaneous Speech Translation Tomoki

Exploring Lexicalized Features for Coreference Resolution Anders Bj orkelund and Pierre Nugues

Lagrangian Based Approaches for Lexicalized Tree Adjoining Grammar Parsing Caio Corro

Models of Human Parsing Experimental Data 2 Informatics 2A: Lecture 22 Eye-tracking Reading

Outline LR Parsing Review of bottom-up parsing LALR Parser Generators Computing the

Generalised Parsing and Combinator Parsing A Happy Marriage? L. Thomas van Binsbergen

Multiple Sequence Alignment based on Ch. 6 from Biological Sequence Analysis by R. Durbin et

IMMUNOTHERAPY AND RADIOTHERAPY FOR MELANOMA BRAIN METASTASES: IS THERE A SYNERGISM? Di Brina

Antigen Presentation K.J. Goodrum Department of Biomedical Sciences Ohio University 2005 MHC II

Depression: A Training for Community Health Workers in Iowa December 11, 2017 Presented by Iowa

Balancing Selection and Beyond: Machine learning approaches for determining selection scenarios

Genomics Genomics extravaganza extravaganza Genomics Genomics overview overview Genomics

1 Natural selection The conditions for natural selection: 1. variation among individuals

Part 2: Stress By Liz Witherspoon and Julie Elliott of Simplicity Coaching STRESS MEETING STRESS

Explore More Topics

Sambuz

Useful Links

Newsletter

Mail Us