SlideShare a Scribd company logo
1 of 17
AN EVOLVING SEMANTIC DATASET
FOR TRAINING AND EVALUATION OF
DISTRIBUTIONAL SEMANTIC
MODELS
E N R I C O S A N T U S , F R A N C E S Y U N G ,
A L E S S A N D R O L E N C I & C H U - R E N H U A N G
EVALution 1.0
Distributional Semantic Models
 Distributional Semantic Models
represent lexical meaning in vector spaces
by encoding corpora derived word co-
occurrences in vectors (Sahlgren, 2006).
 Distributional Hypothesis (Harris, 1954)
 “You shall know a word by the company it
keeps” (Firth, J. R. 1957:11).
Similarity
 DSMs are known to be particularly strong in
identifying semantic similarity between lexical
items, thanks to their geometric representation
(Zesch and Gurevych, 2006).
 Vector cosine: distance
as index of similarity
Many kinds of Similarity
 Lexical items are similar to each other in
many ways:
 cat is similar to lion  COORDINATES (under feline)
 cat is similar to animal  HYPONYM
 cat is similar to dog  ANTONYMS (or: PARANYMS)
 How to actually discriminate
the different types of similarity?
Discriminate Semantic Relations
 Several distributional approaches:
 Pattern based approaches (Hearst, 1992):
 word-pairs = seeds  collocations  patterns (training &
evaluation)
 Unsupervised distributional measures (Santus et al.,
2014; Lenci and Benotto, 2012)
 weighting the features (evaluation)
 Both the approaches rely on datasets containing
semantic relations, for training and/or evaluation
Datasets
 Test Of English as a Foreign Language (TOEFL)  80 multiple-choice questions
about SYN (Landauer and Dumais, 1997)
 Extended Graduate Record Examination (GRE)  Multiple-choice questions
about ANT (Mohammed et al., 2008)
 WordNet  Computational lexicon, developed by lexicographers, containing several
relations (HYPER, COORD, SYN, etc.) (Fellbaum, 1998)
 ConceptNet  Semantic network including WordNet and many other resources, plus
additional relations (UsedFor, Desires, etc.) (Liu and Singh, 2004)
 WordSim 353  Human ratings; “similarity” is left undefined and it contains several
kinds of paradigmatic relations (SIMIL) (Finkelstein et al., 2002)
 BLESS  Balanced resource, developed for evaluating DSMs. It contains several
relations (HYPER, COORD, MERO, EVENT, RANDOM, etc.) (Baroni and Lenci, 2011)
 Lenci/Benotto  Balanced resource based on human judgments (HYPER, SYN,
ANT) (Santus et al., 2014)
Why a new One?
 Benchmarks developed for purposes other than
DSMs training and evaluation.
 Most of the adopted benchmarks include:
 Task-specific resources (TOEFL, GRE)
 semantic relations defined according to the scope
 General-purpose resources (WordNet, ConceptNet)
 need to be inclusive and comprehensive, so inhomogeneous
 Relata and relations are given without additional
information (e.g. relation domain, word semantic
field, frequency, POS, etc.).
Example
 Consider the following pairs:
 key is a space
 relief is a damage
 silly is a child
 apple is a best
Example
 Consider the following pairs:
 key is a space  WordNet 4.0 (basketball)
 relief is a damage  WordNet 4.0 (law)
 silly is a child  WordNet 4.0 (hypernymy?)
 apple is a best  ConceptNet 5.0 (judgment)
 In a certain sense, these pairs are right. But how
representative are them?
Design
 PROTOTIPICAL PAIRS: Human judgments ensure that only
prototypical and reliable pairs are selected.
 HOMOGENEITY and DISCRIMINATIVE ANALYSIS: Relata
in the pairs should appear in more relations, in order to:
 increase homogeneity of data (e.g. not comparing dogs and apples)
 allow discriminative training and evaluation (analysis)
 BALANCING CRITERIA: Additional information allows
filtering the data according to the needs (e.g. semantic
criteria, statistical ones), both in training and evaluation
 We want to provide a balanced corpus NO!
 We want the user to be able to balance it according to
his/her criteria YES!
EVALution 1.0
 Freely downloadable dataset designed for the
training and the evaluation of DSMs
 7.5K pairs
 1.8K relata (63 of which: MWE)
 9 semantic relations
 10 types of additional information for PAIRS
 7 types of additional information for RELATA
Methodology
 Tuples were:
 extracted from ConceptNet 5.0 + WordNet 4.0 (8.8M pairs)
 filtered through automatic methods to exclude (13K pairs):
 useless pairs (i.e. !relevant relations, mirrors, !alpha char, etc.)
 pairs in other resources (i.e. BLESS and Lenci/Benotto).
 pairs which relata do not occur at least in 3 relations
 paraphrased: “W1 is a kind of W2”, “W1 is the opposite of W2”…
 judged through Crowdflower (7.5K pairs)
 5 subjects  1 (strongly disagree) to 5 (strongly agree)
 Threshold: 3 positive judgments (>3)
 annotated
 5 subjects  PAIRS  semantic tags
 2 subjects  RELATA  semantic tags
 Corpus-based info (frequency, POS, forms, etc.)
Relations, Pairs and Relata
Relation Pairs Relata Template Sentence
IsA 1880 1296 X is a kind of Y
Ant 1600 1144
X can be used as the opposite of
Y
Syn 1086 1019
X can be used with the same
meaning of Y
Mero
- PartOf
- MemberOf
- MadeOf
1003
654
32
317
978
599
52
327
X is…
…part of Y
…member of Y
…made of Y
Entailment 82 132 If X is true, then also Y is true
HasA
(possession)
544 460 X can have or can contain Y
HasProperty
(attribute)
1297 770 Y is to specify X
Additional Information
 Relata: Crowdflower (2 annotators) + Corpus (ukWac + Wackypedia)
 Semantic tags (basic, superordinate, event, time, object, etc.)
 Frequency
 Dominant POS / Distribution of POS
 Distribution of inflected/capitalized forms
 Pairs: Crowdflower (5 annotators) + ConceptNet 5.0
 Semantic tags (event, time, space, object, etc.)
 Paraphrases
 Judgments
 Source
 Score in the source, if available
Dataset Evaluation
Conclusions
 We have introduced EVALution 1.0, an evolving semantic
dataset designed for training and evaluation of DSMs.
 EVALution 1.0 vs. previous resources:
 prototypical pairs (i.e. human judgments);
 internal consistency (i.e. proportion term/SemRel);
 additional information (i.e. data filtering and analysis).
 Extensions include:
 Use of RDF (LEMON)
 Scripts for Data Analysis & Filtering
 Inclusion and Analysis of Rejected Pairs
 Extension of the
 # of pairs
 # and types of annotations
EVALution 1.0
 The resource is available at:
https://github.com/esantus

More Related Content

Similar to Evolving Semantic Dataset for Training and Evaluating Distributional Semantic Models (EVALution 1.0

Taxonomies in Search
Taxonomies in SearchTaxonomies in Search
Taxonomies in SearchTSoholt
 
Qualitative differences between human behvaioral data and co-occurrence mode...
Qualitative differences between  human behvaioral data and co-occurrence mode...Qualitative differences between  human behvaioral data and co-occurrence mode...
Qualitative differences between human behvaioral data and co-occurrence mode...Gabriel Recchia
 
Introduction to Distributional Semantics
Introduction to Distributional SemanticsIntroduction to Distributional Semantics
Introduction to Distributional SemanticsAndre Freitas
 
SemEval-2012 Task 6: A Pilot on Semantic Textual Similarity
SemEval-2012 Task 6: A Pilot on Semantic Textual SimilaritySemEval-2012 Task 6: A Pilot on Semantic Textual Similarity
SemEval-2012 Task 6: A Pilot on Semantic Textual Similaritypathsproject
 
Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...
Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...
Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...Nathan Mathis
 
Interactive Analysis of Word Vector Embeddings
Interactive Analysis of Word Vector EmbeddingsInteractive Analysis of Word Vector Embeddings
Interactive Analysis of Word Vector Embeddingsgleicher
 
Text Analytics for Semantic Computing
Text Analytics for Semantic ComputingText Analytics for Semantic Computing
Text Analytics for Semantic ComputingMeena Nagarajan
 
Zouaq wole2013
Zouaq wole2013Zouaq wole2013
Zouaq wole2013Amal Zouaq
 
Leveraging Sentiment to Compute Word Similarity
Leveraging Sentiment to Compute Word SimilarityLeveraging Sentiment to Compute Word Similarity
Leveraging Sentiment to Compute Word SimilaritySubhabrata Mukherjee
 
Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Bhaskar Mitra
 
Taxonomy Development and Digital Projects
Taxonomy Development and Digital ProjectsTaxonomy Development and Digital Projects
Taxonomy Development and Digital Projects daniela barbosa
 
The Geometry of Learning
The Geometry of LearningThe Geometry of Learning
The Geometry of Learningfridolin.wild
 
Introduction to development of lexical databases
Introduction to development of lexical databasesIntroduction to development of lexical databases
Introduction to development of lexical databasesMuhammad Shoaib Chaudhary
 
2010-04-29-swnj-pcls-presentation
2010-04-29-swnj-pcls-presentation2010-04-29-swnj-pcls-presentation
2010-04-29-swnj-pcls-presentationDouglas Randall
 
Using topic modelling frameworks for NLP and semantic search
Using topic modelling frameworks for NLP and semantic searchUsing topic modelling frameworks for NLP and semantic search
Using topic modelling frameworks for NLP and semantic searchDawn Anderson MSc DigM
 
IDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATA
IDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATAIDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATA
IDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATAijistjournal
 
Identifying the semantic relations on
Identifying the semantic relations onIdentifying the semantic relations on
Identifying the semantic relations onijistjournal
 

Similar to Evolving Semantic Dataset for Training and Evaluating Distributional Semantic Models (EVALution 1.0 (20)

Feb20 mayo-webinar-21feb2012
Feb20 mayo-webinar-21feb2012Feb20 mayo-webinar-21feb2012
Feb20 mayo-webinar-21feb2012
 
Taxonomies in Search
Taxonomies in SearchTaxonomies in Search
Taxonomies in Search
 
Qualitative differences between human behvaioral data and co-occurrence mode...
Qualitative differences between  human behvaioral data and co-occurrence mode...Qualitative differences between  human behvaioral data and co-occurrence mode...
Qualitative differences between human behvaioral data and co-occurrence mode...
 
Frsad overview
Frsad overviewFrsad overview
Frsad overview
 
Introduction to Distributional Semantics
Introduction to Distributional SemanticsIntroduction to Distributional Semantics
Introduction to Distributional Semantics
 
SemEval-2012 Task 6: A Pilot on Semantic Textual Similarity
SemEval-2012 Task 6: A Pilot on Semantic Textual SimilaritySemEval-2012 Task 6: A Pilot on Semantic Textual Similarity
SemEval-2012 Task 6: A Pilot on Semantic Textual Similarity
 
Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...
Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...
Analyzing The Semantic Types Of Claims And Premises In An Online Persuasive F...
 
Interactive Analysis of Word Vector Embeddings
Interactive Analysis of Word Vector EmbeddingsInteractive Analysis of Word Vector Embeddings
Interactive Analysis of Word Vector Embeddings
 
Text Analytics for Semantic Computing
Text Analytics for Semantic ComputingText Analytics for Semantic Computing
Text Analytics for Semantic Computing
 
Measuring Similarity Between Contexts and Concepts
Measuring Similarity Between Contexts and ConceptsMeasuring Similarity Between Contexts and Concepts
Measuring Similarity Between Contexts and Concepts
 
Zouaq wole2013
Zouaq wole2013Zouaq wole2013
Zouaq wole2013
 
Leveraging Sentiment to Compute Word Similarity
Leveraging Sentiment to Compute Word SimilarityLeveraging Sentiment to Compute Word Similarity
Leveraging Sentiment to Compute Word Similarity
 
Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)
 
Taxonomy Development and Digital Projects
Taxonomy Development and Digital ProjectsTaxonomy Development and Digital Projects
Taxonomy Development and Digital Projects
 
The Geometry of Learning
The Geometry of LearningThe Geometry of Learning
The Geometry of Learning
 
Introduction to development of lexical databases
Introduction to development of lexical databasesIntroduction to development of lexical databases
Introduction to development of lexical databases
 
2010-04-29-swnj-pcls-presentation
2010-04-29-swnj-pcls-presentation2010-04-29-swnj-pcls-presentation
2010-04-29-swnj-pcls-presentation
 
Using topic modelling frameworks for NLP and semantic search
Using topic modelling frameworks for NLP and semantic searchUsing topic modelling frameworks for NLP and semantic search
Using topic modelling frameworks for NLP and semantic search
 
IDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATA
IDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATAIDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATA
IDENTIFYING THE SEMANTIC RELATIONS ON UNSTRUCTURED DATA
 
Identifying the semantic relations on
Identifying the semantic relations onIdentifying the semantic relations on
Identifying the semantic relations on
 

Recently uploaded

꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Callshivangimorya083
 
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...Suhani Kapoor
 
Invezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz1
 
Low Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service Bhilai
Low Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service BhilaiLow Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service Bhilai
Low Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service BhilaiSuhani Kapoor
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPramod Kumar Srivastava
 
Ukraine War presentation: KNOW THE BASICS
Ukraine War presentation: KNOW THE BASICSUkraine War presentation: KNOW THE BASICS
Ukraine War presentation: KNOW THE BASICSAishani27
 
代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改
代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改
代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改atducpo
 
FESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfFESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfMarinCaroMartnezBerg
 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...Suhani Kapoor
 
From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...Florian Roscheck
 
Digi Khata Problem along complete plan.pptx
Digi Khata Problem along complete plan.pptxDigi Khata Problem along complete plan.pptx
Digi Khata Problem along complete plan.pptxTanveerAhmed817946
 
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...dajasot375
 
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...Pooja Nehwal
 
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Serviceranjana rawat
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingNeil Barnes
 
VIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service Amravati
VIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service AmravatiVIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service Amravati
VIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service AmravatiSuhani Kapoor
 
Log Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxLog Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxJohnnyPlasten
 
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...Suhani Kapoor
 

Recently uploaded (20)

VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
 
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
 
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
 
Decoding Loan Approval: Predictive Modeling in Action
Decoding Loan Approval: Predictive Modeling in ActionDecoding Loan Approval: Predictive Modeling in Action
Decoding Loan Approval: Predictive Modeling in Action
 
Invezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signals
 
Low Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service Bhilai
Low Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service BhilaiLow Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service Bhilai
Low Rate Call Girls Bhilai Anika 8250192130 Independent Escort Service Bhilai
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
 
Ukraine War presentation: KNOW THE BASICS
Ukraine War presentation: KNOW THE BASICSUkraine War presentation: KNOW THE BASICS
Ukraine War presentation: KNOW THE BASICS
 
代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改
代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改
代办国外大学文凭《原版美国UCLA文凭证书》加州大学洛杉矶分校毕业证制作成绩单修改
 
FESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfFESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdf
 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
 
From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...
 
Digi Khata Problem along complete plan.pptx
Digi Khata Problem along complete plan.pptxDigi Khata Problem along complete plan.pptx
Digi Khata Problem along complete plan.pptx
 
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
Indian Call Girls in Abu Dhabi O5286O24O8 Call Girls in Abu Dhabi By Independ...
 
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
 
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
(PARI) Call Girls Wanowrie ( 7001035870 ) HI-Fi Pune Escorts Service
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data Storytelling
 
VIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service Amravati
VIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service AmravatiVIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service Amravati
VIP Call Girls in Amravati Aarohi 8250192130 Independent Escort Service Amravati
 
Log Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxLog Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptx
 
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
 

Evolving Semantic Dataset for Training and Evaluating Distributional Semantic Models (EVALution 1.0

  • 1. AN EVOLVING SEMANTIC DATASET FOR TRAINING AND EVALUATION OF DISTRIBUTIONAL SEMANTIC MODELS E N R I C O S A N T U S , F R A N C E S Y U N G , A L E S S A N D R O L E N C I & C H U - R E N H U A N G EVALution 1.0
  • 2. Distributional Semantic Models  Distributional Semantic Models represent lexical meaning in vector spaces by encoding corpora derived word co- occurrences in vectors (Sahlgren, 2006).  Distributional Hypothesis (Harris, 1954)  “You shall know a word by the company it keeps” (Firth, J. R. 1957:11).
  • 3. Similarity  DSMs are known to be particularly strong in identifying semantic similarity between lexical items, thanks to their geometric representation (Zesch and Gurevych, 2006).  Vector cosine: distance as index of similarity
  • 4. Many kinds of Similarity  Lexical items are similar to each other in many ways:  cat is similar to lion  COORDINATES (under feline)  cat is similar to animal  HYPONYM  cat is similar to dog  ANTONYMS (or: PARANYMS)  How to actually discriminate the different types of similarity?
  • 5. Discriminate Semantic Relations  Several distributional approaches:  Pattern based approaches (Hearst, 1992):  word-pairs = seeds  collocations  patterns (training & evaluation)  Unsupervised distributional measures (Santus et al., 2014; Lenci and Benotto, 2012)  weighting the features (evaluation)  Both the approaches rely on datasets containing semantic relations, for training and/or evaluation
  • 6. Datasets  Test Of English as a Foreign Language (TOEFL)  80 multiple-choice questions about SYN (Landauer and Dumais, 1997)  Extended Graduate Record Examination (GRE)  Multiple-choice questions about ANT (Mohammed et al., 2008)  WordNet  Computational lexicon, developed by lexicographers, containing several relations (HYPER, COORD, SYN, etc.) (Fellbaum, 1998)  ConceptNet  Semantic network including WordNet and many other resources, plus additional relations (UsedFor, Desires, etc.) (Liu and Singh, 2004)  WordSim 353  Human ratings; “similarity” is left undefined and it contains several kinds of paradigmatic relations (SIMIL) (Finkelstein et al., 2002)  BLESS  Balanced resource, developed for evaluating DSMs. It contains several relations (HYPER, COORD, MERO, EVENT, RANDOM, etc.) (Baroni and Lenci, 2011)  Lenci/Benotto  Balanced resource based on human judgments (HYPER, SYN, ANT) (Santus et al., 2014)
  • 7. Why a new One?  Benchmarks developed for purposes other than DSMs training and evaluation.  Most of the adopted benchmarks include:  Task-specific resources (TOEFL, GRE)  semantic relations defined according to the scope  General-purpose resources (WordNet, ConceptNet)  need to be inclusive and comprehensive, so inhomogeneous  Relata and relations are given without additional information (e.g. relation domain, word semantic field, frequency, POS, etc.).
  • 8. Example  Consider the following pairs:  key is a space  relief is a damage  silly is a child  apple is a best
  • 9. Example  Consider the following pairs:  key is a space  WordNet 4.0 (basketball)  relief is a damage  WordNet 4.0 (law)  silly is a child  WordNet 4.0 (hypernymy?)  apple is a best  ConceptNet 5.0 (judgment)  In a certain sense, these pairs are right. But how representative are them?
  • 10. Design  PROTOTIPICAL PAIRS: Human judgments ensure that only prototypical and reliable pairs are selected.  HOMOGENEITY and DISCRIMINATIVE ANALYSIS: Relata in the pairs should appear in more relations, in order to:  increase homogeneity of data (e.g. not comparing dogs and apples)  allow discriminative training and evaluation (analysis)  BALANCING CRITERIA: Additional information allows filtering the data according to the needs (e.g. semantic criteria, statistical ones), both in training and evaluation  We want to provide a balanced corpus NO!  We want the user to be able to balance it according to his/her criteria YES!
  • 11. EVALution 1.0  Freely downloadable dataset designed for the training and the evaluation of DSMs  7.5K pairs  1.8K relata (63 of which: MWE)  9 semantic relations  10 types of additional information for PAIRS  7 types of additional information for RELATA
  • 12. Methodology  Tuples were:  extracted from ConceptNet 5.0 + WordNet 4.0 (8.8M pairs)  filtered through automatic methods to exclude (13K pairs):  useless pairs (i.e. !relevant relations, mirrors, !alpha char, etc.)  pairs in other resources (i.e. BLESS and Lenci/Benotto).  pairs which relata do not occur at least in 3 relations  paraphrased: “W1 is a kind of W2”, “W1 is the opposite of W2”…  judged through Crowdflower (7.5K pairs)  5 subjects  1 (strongly disagree) to 5 (strongly agree)  Threshold: 3 positive judgments (>3)  annotated  5 subjects  PAIRS  semantic tags  2 subjects  RELATA  semantic tags  Corpus-based info (frequency, POS, forms, etc.)
  • 13. Relations, Pairs and Relata Relation Pairs Relata Template Sentence IsA 1880 1296 X is a kind of Y Ant 1600 1144 X can be used as the opposite of Y Syn 1086 1019 X can be used with the same meaning of Y Mero - PartOf - MemberOf - MadeOf 1003 654 32 317 978 599 52 327 X is… …part of Y …member of Y …made of Y Entailment 82 132 If X is true, then also Y is true HasA (possession) 544 460 X can have or can contain Y HasProperty (attribute) 1297 770 Y is to specify X
  • 14. Additional Information  Relata: Crowdflower (2 annotators) + Corpus (ukWac + Wackypedia)  Semantic tags (basic, superordinate, event, time, object, etc.)  Frequency  Dominant POS / Distribution of POS  Distribution of inflected/capitalized forms  Pairs: Crowdflower (5 annotators) + ConceptNet 5.0  Semantic tags (event, time, space, object, etc.)  Paraphrases  Judgments  Source  Score in the source, if available
  • 16. Conclusions  We have introduced EVALution 1.0, an evolving semantic dataset designed for training and evaluation of DSMs.  EVALution 1.0 vs. previous resources:  prototypical pairs (i.e. human judgments);  internal consistency (i.e. proportion term/SemRel);  additional information (i.e. data filtering and analysis).  Extensions include:  Use of RDF (LEMON)  Scripts for Data Analysis & Filtering  Inclusion and Analysis of Rejected Pairs  Extension of the  # of pairs  # and types of annotations
  • 17. EVALution 1.0  The resource is available at: https://github.com/esantus