Interfacing WordNet-Affectwith OCC Model of Emotions

Alessandro ValituttiCarlo Strapparava

FBK-Irst - Istituto per la ricerca scientifica e tecnologica

Motivations

Affective analysis of text is a relatively newarea 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:A work in progress: Wordnet-Affect Wordnet-Affect andand

itsits integration withintegration with OCCOCC, an appraisal, an appraisalmodel of emotionsmodel of emotions

Affective lexical resources What an emotion is ?

Notoriously it is a difficult problem.Many approaches: facial expressions (Ekman), actiontendencies (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, essentialfor affective computing, computational humor,text analysis, etc.

It is a lexical repository of the direct affectivewords

The resource, named WordNet-Affect, startedfrom WordNet, through selection and labelingof synsets representing affective concepts.

WordNet

WordNet is an on-line lexical reference systemwhose design is inspired by psycholinguistictheories of human lexical memory

English nouns, verbs, adjectives and adverbsare organized into synonym sets (synsets),each representing one underlying lexicalconcept

IRST extensions: multilinguality and DomainLabels (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 setof 200 labels hierarchically organized

In WordNet-Affect we have an additionalhierarchy of affective domain labels(independent from the domain labels) withwhich the synsets representing affectiveconcepts 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)

a-label

emotion

affective

mood

trait

cognitive-state

physical-state

hedonic-signal

emotion-eliciting situation

emotional response

behavior

attitude

sensation

…. the hierarchy of the emotions…..

A-Labels and some examplesA-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 aspecialization of the a-label Emotion

Stative/Causative tagging.Concerning mainly the adjectivalinterpretation

Valence Tagging.Positive/Negative dimension

Emotional hierarchy

With respect to WN-Affect, we provided someadditional a-labels, hierarchically organizedstarting form the a-label Emotion

About 1637 words / 918 synsets

Stative/Causative tagging

An emotional adjective is called causative if itrefers to some emotion that is caused by themodified noun (e.g. “amusing movie”)

An emotional adjective is called stative if itrefers to some emotion owned or felt byentity denoted by the modified noun (e.g.“cheerful/happy boy”)

Valence tagging

Distinguishing synsets according to emotionalvalence

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 affectiveweight of generic words

The mechanism is based on similarity betweengeneric terms and affective lexical concepts

We estimated term similarity from a large scalecorpus (BNC ~ 100 millions of words)

Latent Semantic Analysis => dimensionality reductionoperated by Singular Value Decomposition on theterm-by-documents matrix

Homogeneous representations

In the Latent Semantic Space, we canrepresent in a homogeneous way Words Texts Synsets

Each text (and synsets) can be represented inthe LSA space exploiting a variation of thepseudo-document methodology=> summing up the normalized LSA vectorsof all the terms contained in it

LSA spaced3

d1

d2

synset = w1+ w2 + w3

term = w1

Similarity: cosine among vectors

Affective synset representation

Thus an affective synset (and then anemotional category) can be represented in theLatent Semantic Space

We can compute a similarity measure amongterms 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

Encouragementachievement

Devotionscholarship

Sympathyprofessor

Enthusiasmuniversity

Positive emotional categoryRelated emotional terms

Melancholyscholarship

Isolationstudy

Antipathyprofessor

Downheartednessuniversity

Negative emotional categoryRelated emotional terms

News titles

E.g. the affective weight of some news titles

protest#vAngerDead whale in Greenpeace protest

suffer#vSadnessRecord sales suffer steep decline

crash#vFearRomania: helicopter crash kills four people

pleasure#nJoyReview: `King Kong’ a giant pleasure

Word with highestaffective weightEmotionNews titles (Google-news)

Some resources and functionalities for dealingwith affective evaluative terms

An affective hierarchy as an extension ofWordNet-Affect lexical database, includingemotion, causative/stative and valencetagging

A semantic similarity mechanism acquired inan unsupervised way from a large corpus,providing relations among concepts andemotional categories

Summing up

WN-affect and OCC model

Role of the subject for indirect affectivewords

For example: help, revenge, victory arepositive or negative according to the role ofthe subject

The word victory can be used for expressingpride (related to the “winner”), but alsodisappointment (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 futureprobably the expressed emotion is hope

If victory refers to a past event, a possible emotion issatisfaction

Taking into account appraisal theories=> a process of cognitive evaluation of perceivedconditions

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, objectsand actions, expressing possible cause ofemotions

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 withthe affective hierarchy of WN-affect

We obtain a new organization of synsets witha 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 andOCC emotion nodes

3. Insertion of OCC type nodes4. Annotation of synsets with new affective

labels The same synset can have more labels Emotion labels are associated to appraisal labels

OCC Emotions

FORTUNES-OF-OTHERS•HappyFor•Pity•Gloating•Resentment

ATTRACTION•Liking•Disliking

PROSPECT-BASED•Hope•FearsConfirmed•Disappointment•Fear•Satisfaction•Relief

ATTRIBUTION•Admiration•Pride•Reproach•Shame

WELL-BEING/ATTRIBUTION•Gratitude•Remorse•Gratification•Anger

WELL-BEING•Distress•Joy

ATTRACTION/ATTRIBUTION•Hate•Love

OCC Positive Emotions

FORTUNES-OF-OTHERS•HappyFor•Pity•Gloating•Resentment

ATTRACTION•Liking•Disliking

PROSPECT-BASED•Hope•FearsConfirmed•Disappointment•Fear•Satisfaction•Relief

ATTRIBUTION•Admiration•Pride•Reproach•Shame

WELL-BEING/ATTRIBUTION•Gratitude•Remorse•Gratification•Anger

WELL-BEING•Distress•Joy

ATTRACTION/ATTRIBUTION•Hate•Love

OCC Negative Emotions

FORTUNES-OF-OTHERS•HappyFor•Pity•Gloating•Resentment

ATTRACTION•Liking•Disliking

PROSPECT-BASED•Hope•FearsConfirmed•Disappointment•Fear•Satisfaction•Relief

ATTRIBUTION•Admiration•Pride•Reproach•Shame

WELL-BEING/ATTRIBUTION•Gratitude•Remorse•Gratification•Anger

WELL-BEING•Distress•Joy

ATTRACTION/ATTRIBUTION•Hate•Love

Emotions from EventsEVENTS

SIMPLE EXPECTEDNESS

EXPECTED UNEXPECTED

BEFOREAFTER

POSITIVE NEGATIVE

HOPE FEAR

POSITIVE NEGATIVE

POSITIVE-SURPRISE

NEGATIVE-SURPRISE

DISCONFIRMED

CONFIRMED

POSITIVE NEGATIVE POSITIVE NEGATIVE

SATISFACTION FEARSCONFIRMED

RELIEF FRUSTRATION

Emotions from ActionsEVENTS

ACTOR

ACTEE

POSITIVE NEGATIVE

GRATIFICATION

SHAME

POSITIVE NEGATIVE

ADMIRATIONREPROACH

OBSERVER

POSITIVE ACTION NEGATIVE ACTION

POSITIVE NEGATIVE NEGATIVE

GRATITUDE INGRATITUDE RESENTMENT

Actions,

Events

Objects

Examples

Emotions arising from the evaluation of actions

Emotions arising from the evaluation of“fortune/misfortune” by other people (observers)

Positive-Emotion Negative-Emotion Fortune happy-for envy Misfortune gloat pity

Positive-Emotion Negative-Emotion actor Pride Shame actee gratitude resentment observer admiration reproach

Behavior

A Sketch from the Annotation

Encouragement

Actor

Actee

COMPASSION ADMIRATION

PRIDE

GRATITUDE

POSITIVE HOPE

SATISFACTION

Trait

A Sketch from the Annotation

Ambition

Subject

Others

PRIDE

ENTHUSIASM

ENVY

ADMIRATION

Quality

A Sketch from the Annotation

Vanity

Subject

Others

PRIDE

DISLIKE

Reproach

Possible applications

WN-Affect-OCC could be exploited to performimprovements in sentiment analysis andaffect 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 emotionselicited by expectedness of events)

Improvement of the synset annotation according todifferent 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 distinctionaccording to the valence values