VANCOUVER WELCOMES YOU! MINIMALIST METONYMY RESOLUTION THE ROAD MAP - - PowerPoint PPT Presentation
MILAN GRITTA, MOHAMMAD TAHER PILEHVAR, NUT LIMSOPATHAM, NIGEL COLLIER VANCOUVER WELCOMES YOU! MINIMALIST METONYMY RESOLUTION THE ROAD MAP THE PROBLEM + BRIEF HISTORY PREWIN MODEL RELOCAR FEATURE SELECTION MINIMALIST NN NEW DATASET RESULTS
MILAN GRITTA, MOHAMMAD TAHER PILEHVAR, NUT LIMSOPATHAM, NIGEL COLLIER
THE PROBLEM + BRIEF HISTORY RESULTS + SUMMARY
PREWIN FEATURE SELECTION MODEL MINIMALIST NN RELOCAR NEW DATASET
THE PROBLEM + BRIEF HISTORY RESULTS + SUMMARY
PREWIN FEATURE SELECTION MODEL MINIMALIST NN RELOCAR NEW DATASET
GEOGRAPHICAL PARSING PHD RESEARCH
WHAT IS A (METONYMIC) LOCATION?
MOSCOW TO DISCUSS EBOLA RISKS IN WEST AFRICA.. ALUMINIUM AND GAS TRADING PICKED UP IN MOSCOW. LONDON CONSTITUENCY VOTED TO REMAIN IN THE EU. LONDON VOTED TO STAY IN THE EUROPEAN UNION..
METONYMY ~20% OF ALL LOCATIONS (BNC, WIKIPEDIA)
VERY BRIEF HISTORY OF METONYMY RESOLUTION
▸Nissim et al. (2002, 03, 05)
SEMEVAL (2007) Shared Task
▸Brun et al. (2007) 85.1% ▸Farkas et al. (2007) 85.2% ▸Nastase et al. (2009, 2012) 86.1% ▸Nastase et al. (2013) 86.2% ▸Zhang et al. (2015) 86.5**
Custom Databases Handmade Features Proprietary Parser Co-occurrence lists Levin’s Verb Classes Collocations list Trigger Word Lists MASCARA additional data Parsing BNC Parsing Wikipedia FrameNet VerbNet Multiple Parsers Selectional Preferences Global Context
Head modifier (dependency) PMW (singular, plural) No of words in PMW Determiner of PMW No of gram. roles of PMW POS Tags Manual annotations: syntactical/grammatical
Naïve Bayes ME Classifier Decision List Unsupervised Context Similarity SVM
** not peer reviewed, only in preprint
THE PROBLEM + BRIEF HISTORY RESULTS + SUMMARY
PREWIN FEATURE SELECTION MODEL MINIMALIST NN RELOCAR NEW DATASET
BASELINE 5, 10, 50 WINDOW
COLLOBERT ET AL. (2011), MIKOLOV ET AL. (2014)
THE PROBLEM + BRIEF HISTORY RESULTS + SUMMARY
PREWIN FEATURE SELECTION MODEL MINIMALIST NN RELOCAR NEW DATASET
THE PROBLEM + BRIEF HISTORY RESULTS + SUMMARY
RELOCAR NEW DATASET PREWIN FEATURE SELECTION MODEL MINIMALIST NN
ANNOTATED AT CAMBRIDGE UNI FACULTY FOR MODERN AND MEDIEVAL LANGUAGES
TRAIN 1,026 TEST 1,000 LITERAL 49% METONYMIC 49% MIXED 2% Germany, US and France talk climate science,
THE PROBLEM + BRIEF HISTORY RESULTS + SUMMARY
RELOCAR NEW DATASET PREWIN FEATURE SELECTION MODEL MINIMALIST NN
81.3 81.9 81.3 83.1
50 58 66 74 82 90 Base 5 Base 10 Paragraph PreWin
SEMEVAL (SOTA 86.2%)
81.4 81.3 80 83.6
50 58 66 74 82 90 Base 5 Base 10 Paragraph PreWin
RELOCAR (SOTA 84.8%)
ENSEMBLE 84.6% ENSEMBLE 84.8%
50 58 66 74 82 90
Trained on Semeval Trained on Relocar Trained on CONLL
SEMEVAL TEST DATA
50 58 66 74 82 90
Trained on Semeval Trained on Relocar Trained on CONLL
RELOCAR TEST DATA
50 58 66 74 82 90
Trained on Semeval Trained on Relocar Trained on CONLL
SEMEVAL TEST DATA
50 58 66 74 82 90
Trained on Semeval Trained on Relocar Trained on CONLL
RELOCAR TEST DATA
74 76 78 80 82 84 86 88 90
1 2 3 4 5 6 7 8 9 10 Prewin Window Size Comparison CONLL Semeval Relocar
GITHUB FOR CODE & DATA & INSTRUCTIONS.
DATA.CAM.AC.UK REPOSITORY.CAM.AC.UK
RELOCAR MORE TEST DATA PREWIN FEATURE SELECTION MODEL MINIMALIST NN