Interfacing WordNet-Affect
with OCC Model of Emotions
Alessandro Valitutti
Carlo Strapparava
FBK-Irst - Istituto per la ricerca scientifica e tecnologica
Motivations
 Affective analysis of text is a relatively new
area of research
 Important for many NLP applications




Opinion mining
Market analysis
Affective user interfaces
E-learning environments
Overview of the talk
 Brief outline of Wordnet-Affect
 The role of the subject for affective words
 A work in progress: Wordnet-Affect and
its integration with OCC, an appraisal
model of emotions
Affective lexical resources
 What an emotion is ?
 Notoriously
it is a difficult problem.
Many approaches: facial expressions (Ekman), action
tendencies (Frijda), physiological activity (Ax), …
 Emotions, of course, are not linguistic things
 However the most convenient access we have to
them is through the language
 Ortony et al. (1987) introduced the problem
Some affective lexical resources
 General Inquirer (Stone et al.)
 SentiWordNet (Esuli and Sebastiani)
 Affective Norms for English Words (ANEW)
(Bradley and Lang)
 WordNet Affect (Strapparava and Valitutti)
Affective semantic similarity
 All words can potentially convey affective




meaning
Even those not directly related to emotions can
evoke pleasant or painful experiences
Some of them are related to the individual
story
But for many others the affective power is part
of the collective imagination (e.g. mum, ghost,
war, …)
cfr. Ortony & Clore
 C. Strapparava and A. Valitutti and O. Stock “The Affective Weight
of Lexicon” Proceedings of LREC 2006
WordNet Affect
 We built an affective lexical resource, essential
for affective computing, computational humor,
text analysis, etc.
 It is a lexical repository of the direct affective
words
 The resource, named WordNet-Affect, started
from WordNet, through selection and labeling
of synsets representing affective concepts.
WordNet
 WordNet is an on-line lexical reference system
whose design is inspired by psycholinguistic
theories of human lexical memory
 English nouns, verbs, adjectives and adverbs
are organized into synonym sets (synsets),
each representing one underlying lexical
concept
 IRST extensions: multilinguality and Domain
Labels (WordNet Domains)
Analogy with WordNet domains
 In WordNet Domains each synset has been
annotated with a domain label (e.g. Sport,
Medicine, Politics) selected form a set
of 200 labels hierarchically organized
 In WordNet-Affect we have an additional
hierarchy of affective domain labels
(independent from the domain labels) with
which the synsets representing affective
concepts are annotated
Affective domain labels
 Ortony distinguishes between emotional and affective-not-emotional
words
 Among affective-not-emotional words many categories from the
literature (e.g. Johnson-Laird and Oatley, 1988 Ortony et al., 1988)
…. the hierarchy
of the emotions…..
mood
trait
emotion
cognitive-state
physical-state
a-label
hedonic-signal
affective
emotion-eliciting situation
emotional response
behavior
attitude
sensation
A-Labels and some examples
A-Label
Examples of Synsets
EMOTION
noun "anger#1", verb "fear#1"
MOOD
noun "animosity#1", adjective "amiable#1"
TRAIT
noun "aggressiveness#1", adjective "competitive#1"
COGNITIVE STATE
noun "confusion#2", adjective "dazed#2"
PHYSICAL STATE
noun "illness#1", adjective "all_in#1"
HEDONIC SIGNAL
noun "hurt#3", noun "suffering#4"
EMOTION-ELICITING SITUATION
noun "awkwardness#3", adjective "out_of_danger#1"
EMOTIONAL RESPONSE
noun "cold_sweat#1", verb "tremble#2"
BEHAVIOUR
noun "offense#1", adjective "inhibited#1"
ATTITUDE
noun "intolerance#1", noun "defensive#1"
SENSATION
noun "coldness#1", verb "feel#3"
Freely available (for research purposes) at
http://wndomains.itc.it
Emotions of WN-affect
 Specialization of the Emotional Hierarchy.
For the present work we provide a
specialization of the a-label Emotion
 Stative/Causative tagging.
Concerning mainly the adjectival
interpretation
 Valence Tagging.
Positive/Negative dimension
Emotional hierarchy
 With respect to WN-Affect, we provided some
additional a-labels, hierarchically organized
starting form the a-label Emotion
 About 1637 words / 918 synsets
Stative/Causative tagging
 An emotional adjective is called causative if it
refers to some emotion that is caused by the
modified noun (e.g. “amusing movie”)
 An emotional adjective is called stative if it
refers to some emotion owned or felt by
entity denoted by the modified noun (e.g.
“cheerful/happy boy”)
Valence tagging
 Distinguishing synsets according to emotional




valence
Positive emotions (joy#1, enthusiasm#1),
Negative emotions (fear#1, horror#1),
Ambiguous, when the valence depends on
the context (surprise#1) ,
Neutral, when the synset is considered
affective but not characterized by valence
(indifference#1)
Corpus-based affective semantic similarity
 An example of corpus-based application
 We needed a technique for evaluating the affective
weight of generic words
 The mechanism is based on similarity between
generic terms and affective lexical concepts
 We estimated term similarity from a large scale
corpus (BNC ~ 100 millions of words)
 Latent Semantic Analysis => dimensionality reduction
operated by Singular Value Decomposition on the
term-by-documents matrix
Homogeneous representations
 In the Latent Semantic Space, we can
represent in a homogeneous way



Words
Texts
Synsets
 Each text (and synsets) can be represented in
the LSA space exploiting a variation of the
pseudo-document methodology
=> summing up the normalized LSA vectors
of all the terms contained in it
LSA space
d3
synset = w1+ w2 + w3
d1
term = w1
d2
Similarity: cosine among vectors
Affective synset representation
 Thus an affective synset (and then an
emotional category) can be represented in the
Latent Semantic Space
 We can compute a similarity measure among
terms and affective categories
 Ex. the term “gift” is highly related (in BNC)
with the emotional categories:




Love (with positive valence)
Compassion (with negative valence)
Surprise (with ambiguous valence)
Indifference (with neutral valence)
An example: university
Related emotional terms
university
Positive emotional category
Enthusiasm
professor
Sympathy
scholarship
Devotion
achievement
Encouragement
Related emotional terms
university
Negative emotional category
Downheartedness
professor
Antipathy
study
Isolation
scholarship
Melancholy
News titles
 E.g. the affective weight of some news titles
Emotion
Word with highest
affective weight
Review: `King Kong’ a giant pleasure
Joy
pleasure#n
Romania: helicopter crash kills four people
Fear
crash#v
Sadness
suffer#v
Anger
protest#v
News titles (Google-news)
Record sales suffer steep decline
Dead whale in Greenpeace protest
Summing up
 Some resources and functionalities for dealing
with affective evaluative terms
 An affective hierarchy as an extension of
WordNet-Affect lexical database, including
emotion, causative/stative and valence
tagging
 A semantic similarity mechanism acquired in
an unsupervised way from a large corpus,
providing relations among concepts and
emotional categories
WN-affect and OCC model
 Role of the subject for indirect affective
words
 For example: help, revenge, victory are
positive or negative according to the role of
the subject
 The word victory can be used for expressing
pride (related to the “winner”), but also
disappointment (related to the “loser”)
WN-affect and OCC model
 Another contextual information that can matter:
temporal aspects
 If victory is a future (desired) event in the future
probably the expressed emotion is hope
 If victory refers to a past event, a possible emotion is
satisfaction
 Taking into account appraisal theories
=> a process of cognitive evaluation of perceived
conditions
WN-affect and OCC model
 An appraisal model of emotions
 OCC (from Ortony, Clore and Collins, 1988)
 In OCC emotions are classified according to
some main categories such as events, objects
and actions, expressing possible cause of
emotions
 Contextual information:


different subject or roles in an actions
as a future development: considering different
temporal conditions
WN-affect and OCC model
 We rearrange OCC and we integrate it with
the affective hierarchy of WN-affect
 We obtain a new organization of synsets with
a set of additional labels called OCC-labels
Integrating WN-Affect and OCC
1. Rearrangement of OCC hierarchy:

Valence inheritance
2. Mapping between WN-Affect categories and
OCC emotion nodes
3. Insertion of OCC type nodes
4. Annotation of synsets with new affective
labels


The same synset can have more labels
Emotion labels are associated to appraisal labels
OCC Emotions
ATTRACTION
•Liking
•Disliking
WELL-BEING
•Distress
•Joy
PROSPECT-BASED
•Hope
•FearsConfirmed
•Disappointment
•Fear
•Satisfaction
•Relief
WELL-BEING/ATTRIBUTION
•Gratitude
•Remorse
•Gratification
•Anger
ATTRIBUTION
•Admiration
•Pride
•Reproach
•Shame
ATTRACTION/ATTRIBUTION
•Hate
•Love
FORTUNES-OF-OTHERS
•HappyFor
•Pity
•Gloating
•Resentment
OCC Positive Emotions
ATTRACTION
•Liking
•Disliking
WELL-BEING
•Distress
•Joy
PROSPECT-BASED
•Hope
•FearsConfirmed
•Disappointment
•Fear
•Satisfaction
•Relief
WELL-BEING/ATTRIBUTION
•Gratitude
•Remorse
•Gratification
•Anger
ATTRIBUTION
•Admiration
•Pride
•Reproach
•Shame
ATTRACTION/ATTRIBUTION
•Hate
•Love
FORTUNES-OF-OTHERS
•HappyFor
•Pity
•Gloating
•Resentment
OCC Negative Emotions
ATTRACTION
•Liking
•Disliking
WELL-BEING
•Distress
•Joy
PROSPECT-BASED
•Hope
•FearsConfirmed
•Disappointment
•Fear
•Satisfaction
•Relief
WELL-BEING/ATTRIBUTION
•Gratitude
•Remorse
•Gratification
•Anger
ATTRIBUTION
•Admiration
•Pride
•Reproach
•Shame
ATTRACTION/ATTRIBUTION
•Hate
•Love
FORTUNES-OF-OTHERS
•HappyFor
•Pity
•Gloating
•Resentment
Emotions from Events
EVENTS
SIMPLE
EXPECTEDNESS
UNEXPECTED
EXPECTED
BEFORE
POSITIVE
AFTER
POSITIVE
NEGATIVE
POSITIVESURPRISE
CONFIRMED
HOPE
FEAR
POSITIVE
SATISFACTION
NEGATIVE
NEGATIVESURPRISE
DISCONFIRMED
NEGATIVE
POSITIVE
FEARS
CONFIRMED
RELIEF
NEGATIVE
FRUSTRATION
Emotions from Actions
EVENTS
ACTOR
POSITIVE
OBSERVER
POSITIVE
NEGATIVE
ACTEE
ADMIRATION
REPROACH
SHAME
GRATIFICATION
POSITIVE ACTION
POSITIVE
GRATITUDE
NEGATIVE
NEGATIVE
INGRATITUDE
NEGATIVE ACTION
NEGATIVE
RESENTMENT
Actions,
Events
Objects
Examples
 Emotions arising from the evaluation of actions
actor
actee
observer
Positive-Emotion
Pride
gratitude
admiration
Negative-Emotion
Shame
resentment
reproach
 Emotions arising from the evaluation of
“fortune/misfortune” by other people (observers)
Fortune
Misfortune
Positive-Emotion
happy-for
gloat
Negative-Emotion
envy
pity
A Sketch from the Annotation
Behavior
COMPASSION
ADMIRATION
GRATITUDE
Actor
PRIDE
Encouragement
Actee
SATISFACTION
POSITIVE HOPE
A Sketch from the Annotation
Trait
PRIDE
Subject
Ambition
ENTHUSIASM
ADMIRATION
Others
ENVY
A Sketch from the Annotation
Quality
PRIDE
Subject
Vanity
Reproach
Others
DISLIKE
Possible applications
 WN-Affect-OCC could be exploited to perform
improvements in sentiment analysis and
affect sensing
 identification of the emotion-holder
 identification of the emotion-target
 OCC analysis of the emotion-target
 “I need help” => sadness
 “Your help was useful” => gratitude
Future Work
 Still a work in progress
 Extension of annotation to synsets with pos
“adjective”, “verb”, and “adverb”
 Not only subjectivity but also other appraisal features
(e.g. temporal features in the case of emotions
elicited by expectedness of events)
 Improvement of the synset annotation according to
different sources of stereotypical knowledge (e.g.
Latent Semantic Analysis)
Conclusions
 Still a work in progress
 Development of WN-Affect-OCC as extension of WordNet-Affect
with the integration of OCC model of emotions.
 Tagging of (indirect affective) words referring to cause of
emotions (through the process of cognitive appraisal)
 Introduction of appraisal variables that can be disambiguated
from the analysis of textual context (e.g. subjectivity:
actor/actee/others)
 Improvement of the emotional hierarchy and distinction
according to the valence values