framECO

The Idea

Every day we experience much more complex emotions than those we can most easily give a name to. In literary and philosophical texts we find descriptions of emotions, or feelings, or moods that go beyond traditional models. So it is necessary to go beyond what already exists to formalize without trivializing, therefore in a sophisticated way, the humanistic perspectives in order to study and exploit them more effectively.

So the purpose of this project is twofold:

1. to identify the theories of current emotions, how to give a formal characterization by bringing examples from the literature and from other projects on the subject;
2. to propose an ontology of those emotions to which we have not yet given a name and for which there is no taxonomy or classification, thus giving formal characterization to these new, emerging concepts. It is a question of creating a formal model (i.e. an ontology of the contexts of emotion) to describe the emerging emotions (more precisely, their contexts). Therefore, showing what is in the state of the art and what is missing to develop a model instead that allows both to describe an emotion starting from a vocabulary, and, possibly in the future, to recognize it in the text.

What's Already Out There: the Need to Reuse and not Reinvent the Wheel

This section presents the existing models of description of emotions, as well as some theories of cognitive-affective science that underlie these models plus a particular case of a theoretical model that does not yet have a formalization.

The purpose of this part of the project is to identify what is there, what can be reused and what is missing in defining an emotion. If most of the existing models share the same principles and if they are exhaustive and consistent both from a formal and content point of view.

Existing Ontologies

1. EMO / MFOEM Emotion Ontology

Developed by the OBO (Open Biological and Biomedical Ontology) Foundry group, which is a community which develops interoperable ontologies for the biological sciences.

It’s an ontology of affective phenomena such as emotions, moods, appraisals and subjective feelings, designed to support interdisciplinary research by providing unified annotations. The ontology is a domain specialisation of the broader Mental Functioning Ontology.

The emotion process:

• is an affective process;
• has agent an organism;
• involves:
  • an appraisal process which rises a a cognitive representation which represents an evaluation of the relevance of some triggering object or event to the organism
  • an emotional behavioral process: behaviour of the organism in response to the appraisal, which includes the
  • characteristic facial expressions for particular emotion types
  • a physiological response to emotion process: a bodily process which encompasses all the neurophysiological changes caused by the appraisal, which take place in the central nervous system (CNS), neuro-endocrine system (NES) and autonomous nervous system (ANS)
• occurs with consciousness;
• realizes a mental disposition;
• has an object;
• gives rise to an action tendency.

So the basic idea is that every emotion process involves these 4 + 1 components:

1. appraisal: e.g. the appraisal of suddenness gives rise to the emotional process which is called surprise;
2. subjective emotional feeling: an affective representation that the organism has about its own emotion (representations are always about something), e.g. feeling nervous;
3. action tendency: a disposition to behaviour that inheres in an organism by virtue of the physical changes brought about by an emotion process. e.g., smiling;
4. physiological response: bodily process which encompasses all the neuro-physiological changes caused by the emotion, e.g. shivering, becomig pale, heart beating at a fast race;
5. emotional behavioral process: not an actual component but rather an outcome of the emotion process. It’s the behavior of an organism in response to the emotion and as a result of the emotional action tendency which I referred to in the previous slide.

The emotion process is linked to but distinguished from: an emotional personality trait, an affective representation, and emotional action tendency, and an emotion disposition.

The Emotion Ontology has been developed within the Basic Formal Ontology framework, to support interdisciplinary research into emotions and other affective phenomena in fields such as cognitive neuroscience, behavioral psychology, psychiatry, literary analysis, and artificial intelligence. The ontology provides a common language effectively, since the terms and relations in an ontology are provided with logical definitions which enable reasoning with the data.

2. VEO: a model for representing visual cues of emotions

The background here are the healthcare services, particularly patient-provider interaction, which often involve highly emotional situations, and it is important for physicians to understand and respond to their patients’ emotions to best ensure their well-being.

This ontology is based on the Ortony, Clore, & Collins’ models of emotions ("the OCC model"), which differentiates 22 emotions depending on the psychological scenario that causes the emotion and the subsequent affective reactions that appear. The OCC model proposes that emotions are comprised of a collection of behaviors rather than independent entities that then cause the behaviors. This model describes emotions based on situations rather than on patterns of physiology, neurology, experience, expression, and motivation is more straightforward and reliable for computers to understand. Additionally, the OCC model organizes emotions into three categories: those concerning consequences of events, actions of agents, and aspects of objects.

VEO is aligned with EMO of MFOEM, although it does not consider all the emotions categorized in the latter.

Emotions are defined by valence, hierarchically and semantically (e.g. Satisfaction is a subclass of Joy).

Finally, the emotion is defined as "an affective state of conciousness"

3. WN-AFFECT: a WordNet extension

While not an emotion ontology, WN-AFFECT is an extension of the WordNet ontology with annotations that describe the emotional valence of words based on the W3C lists of emotions. In particular, to some synsets where assigned affective labels such us ‘emotion’, ‘mood’, ‘trait’ + extension of a set of additional a-labels (called emotional categories), hierarchically organized. In a second stage, synsets were distinguished according to emotional valence, defining four additional a-labels: positive, negative, ambiguous, and neutral. Another important property is the stative/causative dimension: an emotional adjective is said causative if it refers to some emotion that is caused by the entity represented by the modified noun (e.g. amusing movie); while an emotional adjective is said stative if it refers to the emotion owned or felt by the subject denoted by the modified noun (e.g. cheerful/happy boy).

4. EmOCA: Emotion Ontology for Context Awareness

EmOCA is an emotion ontology for describing and reasoning on emotion context in order to improve emotion detection based on bodily expression. Context is incorporated into the two-factor theory of emotion (bodily reaction plus cognitive input) to demonstrate the importance of context in the emotion experience. In attempting to determine emotion felt by another person, the bodily expression of their emotion is the only evidence directly available, eg, “John looks angry”.

The motivation behind the ontology creation is to bring context into the emotion-modulating cognitive input, eg, we know that John is a generally calm person, so we can conclude from expression (anger) plus context (calm) that John is not only angry, but that “John must be furious”. The choince of the ontology as a tool is due to the aim of bringing context into the implementation of the emotion detection process.

Two-factor theory: this theory proposes that after perceiving a stimulus (internal eg, souvenir or external eg, real world object), a person first experiences an immediate bodily reaction, and secondly a cognitive process is launched which uses the bodily expression and cognitive valuation to define their emotion.

Emotions used in the ontologies are Elkmann’s 6 basic emotions

5. EmotionsOnto: an Ontology for Developing Affective Applications

EmotionsOnto is a generic ontology for describing emotions and their detection and expression systems taking contextual and multimodal elements into account. Once formalized as an ontology, the knowledge about emotions can be used in order to make computers more personalised and adapted to users’ needs. The ontology has been validated and evaluated by means of an applications based on a emotions-aware Tangible User Interface (TUI).

First assumption: the focus are emotion detection and expression systems instead of modelling internal emotion processing in humans. Second assumption: the emotion is conceived as Emergent Emotion (full-blown) in Scherer-Douglas-Cowie list of affective states, defined as: "states where the person’s whole system is caught up in the way they react to a particular person or situation". Aspects involved are: valence, impulses to act or express, distinctive changes in the body, short duration. EmotionsOnto is based on a generic model geared towards capturing the entities that take part in the Emergent Emotion process.

The ontology can help implementing emotion-aware applications based on a wider range of theoretical approaches. This flexibility and wide applicability of EmotionsOnto is in great part due to the fact that it is capable of modeling contexts. In order to do that, the DOLCE upper ontology is reused and extended, particularly the Description and Situation concepts. Descriptions correspond to the representations for the situations, which then trigger and are associated with emotions. Descriptive Ontology for Linguistic and Cognitive Engineering has also been chosen for cognitive aspects. Situations associated with emotions are stated to be modeled with FrameNet. FrameNet can be easily connected with DOLCE since the concept Frame appears as a subclass of Description.

6. SOCAM affective extension (Ontology Based Affective Context Representation)

Focusing on the context, this is an ontology-based representation of the affective states for context aware applications which allows expressing the complex relations that are among the affective states and between these and the other context elements.

Emotions are subdivided in 2 main subclasses: basic and secondary. The Secondary has as subclasses a number of items that application-specific. Activation-evaluation model: each secondary state has a membership degree value for each of the basic state considered. The membership degree is expressed for each of the two following axes: valence (how positive, negative or neutral the state is) and activation (how dynamic the state is). The third axis measures the intensity of the state as in using values as low, medium, high. Since the affective states are considered to be permanently changing, a representation of time for an affective state is provided too.

7. Emotional Cues Ontology

It's a multimodal ontology: it is an ontology based on emotional cues that uses media properties from different sources to model emotion. It uses a generic approach to modeling emotional cues, so it is theoretical-system independent.

It presents 3 modules representing three layers of emotions' detection or production: the emotion module, the emotional cue module, and the media module:

• emotion module: defines emotions as represented with emotional cues;
• emotional cue module: describes external emotional representations in terms of media properties (change of tone has media: voice/sound). Every emotion can have one or more emotional cues, represented in different media. Emotional cues can be simple (verbal, conductal, physiological) or complex (integration of two or more cues);
• media module: describes basic media properties important for emotional cues. This relation among emotional cues and media properties can be parameterized, enabling different settings for different emotional cues. Media is described with either simple media properties (e.g., intensity of sound), or derived (e.g., variations in the intensity of sound).

The RECOGNITION OF EMOTIONS is BOTTOM-UP PROCESS: it is possible to define emotions as composed of complex emotional cues, which are composed of other simple emotional cues, which are based on some media properties.

The PRODUCTION OF EMOTIONS is TOP-DOWN PROCESS: we detect simple media properties, and then derive complex media properties. By analyzing complex media properties, we get simple emotional cues, and by combining these cues, we get complex emotional cues, which represent some emotion.

Existing models

1. Underlying theories

For most of the ontologies or vocabularies proposed there are some theories underlying them. The most relevant are:

• Categorical emotional models: emotions are defined in discrete categories, such as happiness, sadness etc;
• Dimensional emotional model: emotions are defined according to dimensions, such as valence, arousal and control;
  • Lang's dimensional emotional model: Lang also proposed analyzing of emotions according to three systems involved in the expressions of emotions: subjective or verbal information, behavioral or conductal, psycho-physiological answers.
• The OCC model (Ortony, Clore, & Collins’ models of emotions): states that the strength of a given emotion primarily depends on the events, agents, or objects in the environment of the agent exhibiting the emotion;
• Elkmann’s 6 basic emotions: through a series of studies, Ekman found a high agreement across members of diverse Western and Eastern literate cultures on selecting emotional labels that fit facial expressions. Expressions he found to be universal included those indicating wrath, grossness, fear, joy, loneliness, and shock;
• Scherer-Douglas-Cowie list of affective states: Attitudes, Established emotion, Emergent Emotion (full-blown), Emergent Emotion (suppressed), Moods, Partial emotion (topic shifting), Partial emotion (simmering), Stance towards person, Stance towards object/situation, Interpersonal bonds, Altered state of arousal, Altered state of control, Altered state of seriousness and Emotionless.

2. Theories without a formal model yet

Among all the emotion theories that have not been translated to a formal model, there is one that is more interesting because of its significant conceptual difference: the Theory of Constructed Emotion.

Presented by psychology professor and neuroscientist Dr. Lisa Feldman Barrett in her book How Emotions Are Made, it contradicts many of our most firmly held ideas about how human emotions work. Indeed, some of this unusual statements can bu used to build a Conceptual Model since they are the most relevant for the theory. For example, it argues that:

• Emotions cannot be detected through facial expressions or any other physiological measurement;
• Emotions don’t have any Expressor.
• There are no "universal" emotions across people, nations, or cultures. Each one of us constructs their own emotions based on their personal experiences, although we can share it with others;
• All emotions are subjective and based on personal experiences. We can share emotions with others.
• Emotions are not “reactions” to external events: the experience of an emotion is a “simulation” or prediction of the appropriate way for the body to react to an event. The experience of an emotion is a prediction of the brain of what it thinks it might happen next;
• All experiences of an emotion are predictions. Predictions are about how to react to an event.
• Emotions are concepts that are constructed by the brain;
• All emotions are concepts. Concept is a necessary condition for Emotion.
• Emotions are concepts built by the mind out of pieces of sensory data, knowledge and history of social interactions.
• Emotions are concepts built by the mind out of pieces of sensory data, knowledge and history of social interactions.

Glossary of terms: Emotion, Concept, Experiencer, Personal Experience, Emotion Experience, Prediction, Bodily Reaction, Event, Mind, Sensory Data, Knowledge, Social Interactions.

Conceptual model:

Summing up table

Ontology/Vocabulary	First-level classification	Second-level classification	Source of emotions definitions	Hierarchy among emotions	Valence	Other criterion for definition / relevant features	Focus	Scope/Application context
EMO/MFOEM	Distinction of emotion process from mood, trait, emotion disposition, subjective emotion feeling.	Emotion processes are classified in 33 different classes.	OCEAS (Oxford Companion to Emotion and the Affective Sciences).	Yes. Some emotions are subclasses of others (e.g., positive/negative surprise).	Yes. Positive/Negative.	The emotion process involves 5 components: appraisal, subjective emotional feeling, action tendency, physiological response and emotional behavioral process. Occurs with consciousness, has an object and realizes a mental disposition. Emotions as collections of independent entities that then cause the behaviors.	Modeling internal emotion processing in humans.	Bio-medical. Ontology is designed to support interdisciplinary research by providing unified annotations. Eamples of use cases: ‘sentiment mining’.
VEO	Not stated clearly but since it is stated that the Emotion class in the VEO is equivalent to the emotion process class in EMO, first-level classification is supposed to be the same.	25 emotions defined.	OCC model (psychological constructivist approach).	Yes. An Emotion is divided into either a Positive Emotion or a Negative Emotion subclass, which then can be further divided into Approving/Disapproving, Liking/Disliking, and Pleased/Displeased subclasses. Then, emotions are categorized into one or more of these subclasses in accordance with the OCC model.	Yes. Positive/Negative.	Emotions are defined depending on the psychological scenario that causes the emotion and the subsequent affective reactions that appear. Thus they are organized into 3 categories: those concerning consequences, actions and aspects of objects. Emotions are comprised of a collection of behaviors.	Modeling internal emotion processing in humans.	Clinical. It’s planned to use VEO in patient-facing software tools to enhance interactions between patients and providers in a clinical environment. Developed together with emotions visualizations.
WN-AFFECT	Distinction of emotion from mood, trait, behavior, attitute, sensation etc.	Assignment of emotional categories to synsets.	Based on the W3C lists of emotions.	Yes. Emotional categories are hierarchically organized.	Yes. Positive, negative, ambiguous and neutral.	Stative/causative dimension: an emotional adjective is said causative if it refers to some emotion that is caused by the entity represented by the modified noun (e.g. amusing movie); while an emotional adjective is said stative if it refers to the emotion owned or felt by the subject denoted by the modified noun (e.g. cheerful/happy boy).	Word valence.	Extension of the WordNet ontology with annotations that describe the emotional valence of words (synsets).
EmOCA	Emotions are clearly distinguished from stimulus, impact and trait.	6 emotions defined.	Two-factor theory of emotions (after perceiving a stimulus, a person first experiences an immediate bodily reaction, and secondly a cognitive process is risen which uses the bodily expression and cognitive valuation to define their emotion) + Elkmann 6 basic emotions.	No hierarchy among emotion categories.	Yes. Numerical valence of an emotion is retrieved/assigned with Russel model (continuous model).	Emotion context: incorporation of context into the two-factor theory of emotion (bodily reaction plus cognitive input) → importance of context in the emotion experience. Arousal is also calculated with Russel model. .	Emotion context.	Improve emotion detection based on bodily expression. The ontology can be used to temper emotion determination based on facial expression alone.
EmotionsOnto	Emotion is distinguished from sensation, perception, description and memory (and situation and stimulus).	Not stated.	None: it’s a generic ontology independent from any theoretical approach. But it adopts the Emergent Emotion (full-blown) definition.	Not stated clearly.	Yes. Positive or negative.	Reuse of DOLCE upper ontology and FrameNet: frame semantics for context.	Emotion detection and expression systems instead of modelling internal emotion processing in humans.	Emotion-aware applications, emotion expression and recognition systems an social networks, make computers more personalised and adapted to users’ needs.
Constructed Emotion Theory	Emotion is a concept, thus it is distinguished from reaction, experience, perception, description, situation and stimulus.	Not stated.	How Emotions Are Made by Lisa Feldman Barrett.	N.d.	Not stated clearly, but possibly yes. Positive or negative.	Emotions are concept defined by the brain, which is driven by simulations, which we are indeed experiencing.	Birth and development of emotions + what are emotions and what is emotional experience.	Theoretical, relevant in the affective science and neuroscience domain.

Let's Clear Things Up: Reasoning Patterns Behind Existing Ontologies' Knowledge Patterns

Formalization of the Theory of Constructed Emotion

Starting from the Theory of Constructed Emotion, it is possible to define a OWL and DL formalization starting from the conceptual model (using CemoT as a hypothetical prefix):

Clarification of the formalization of existing and mentioned ontologies

In this section the formalization regarding the concept of "emotion" is made explicit for each ontology considered earlier. Where possible, the formalization of the definition of an emotion is provided too.

1. EMO/MFOEM

1. Definition of Emotion (for 3 of the main class expressions and axioms for Emotion class):

Other class expressions relevant for the description of the Emotion class:
subclassof: hasPart some 'emotional behavioural process', hasPart some 'physiological response to emotion process', 'has occurrent part' some consciousness, realizes some 'mental disposition', 'has participant at all times' some 'extended organism'

1. Definition of an Emotion (which is a subclass of the class "emotion process"):

Interest: Interest is a positive emotion which motivates learning, exploration and curiosity. Feelings of interest are associated with the novelty and complexity of stimuli. [Source: OCEAS]

2. VEO

1. Definition of Emotion:

"an affective state of consciousness" + equivalent to the emotion process class in EMO

1. Definition of an Emotion:

Interest: Interest is liking an unfamiliar aspect (of an object) [Source: OCC Model]

3. EmOCA

1. Definition of Emotion:

Emotion is defined by Coordinate, which has subclasses Radius and Component. Component has 2 subclasses: Arousal and Valence. A person has trait Trait which has impact Impact which pertains to Component.

4. EmotionsOnto

1. Definition of Emotion:

an EmergentEmotion is a subclass of Event. The class “EmergentEmotion” is not linked with any other class.

This lacking definition provided by the ontology is different from the definitions stated in the paper (López et al., 2014):

• Description triggers EmergentEmotion, and EmotionExpressionSystem has input EmergentEmotion. Perception hasOutput Description.
• Emergent Emotion: states where the agent’s whole emotional system is caught up in the way they react to a particular situation [Douglas-Cowie et al., 2006], which is "internally" represented using a description

5. SOCAM affective extension (Ontology Based Affective Context Representation)

1. Definition of Emotion:

Emotion is an Affective State. State has subclasses: Affective, Mental and Physiological. Affective has subclasses Basic, Current and Secondary.
[Basic ∪ Secondary] HasSensitivity Sensitivity. Sensitivity isOwnedBy Person. Person isInTheState Current. Person isLocatedIn Location. Current isInducedBy Location or Time.

6. Emotional Cues Ontology

1. Definition of Emotion:

Emotion has external manifestation EmotionalCue. EmotionalCue has mediarepresentation MediaProperty. EmotionalCue has subclasses: SimpleEmotionalCue and ComplexEmotionalCue.

Frame-based Alignment of Existing Ontologies: Approaching FrameNet

Since we do not have the complete structure for each ontology mentioned, and almost no logical formalization for them, a complete and exhaustive alignment is not possible, but an attempt may be useful to detect elements in common in the definitions of the concept of emotion. Considering the lack of axiomatization and of declaration of logical types defined within the ontologies collected, the approach chosen to find correspondences between ontology entities is frame-based: we want to express the intensional meaning of the models, abstracting from logical types (Aspirino et al., 2017). Thus, we will follow the method explained in: Luigi Asprino, Valentina Presutti, Aldo Gangemi, and Paolo Ciancarini. 2017. Frame-based ontology alignment. In Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI'17). AAAI Press, 4905–4906.

In order to do a frame-based alignment, we will use FrameNet: a project housed at the International Computer Science Institute in Berkeley, California which produces an electronic resource based on frame semantics. Frame Semantics (Fillmore, 1982) is a formal theory of meaning based on the idea that human can better understand the meaning of a single word by knowing the relational knowledge associated to that word (Aspirino et al., 2017). A semantic frame can be thought of as a conceptual structure describing an event, relation, or object and the participants in it. Frame elements (FE) provide additional information to the semantic structure of a sentence. Each frame has a number of core and non-core FEs which can be thought of as semantic roles. Indeed, a frame can be considered as “unit of meaning” for ontologies, and can be exploited as a means for representing the intensional meaning of the entities.

In this ontology alignment we will focus only on the entity of "emotion", to find out if there is something that all these models have in common and which is therefore factually fundamental in describing an emotion – in order to distinguish the essential elements to describe the meaning of the concept (frame) from the merely descriptive ones.

The purpose of this alignment, however, is not strictly that of making the entities of different ontologies correspond, but rather that of identifying the most important frames (and their respective elements) to define the concept of emotion, i.e. those common to several annotations of the entity it represents in any ontology taken into consideration. Eventually, frames evoked by these entities will be compared to the frame elements of the Emotion frame in FrameNet.

1. Selecting frames evoked by annotations

In this first step, for each ontology selected from the bibliographic search, the words that evoke the frames representing the meaning of emotion are extracted from the annotation of the entity "emotion". When neither the annotation nor the comment were present, the definition was taken from the refering journal article. Taking into account the entity's comment and annotation allows us to disambiguate the meaning of the word. The selection of the evoked frames takes place exploiting the frames FrameNet as mentioned, but also the synsets of WordNet. WordNet is a large lexical database of English: nouns, verbs, adjectives and adverbs are grouped into sets of cognitive synonyms (synsets), each expressing a distinct concept. Synsets are interlinked by means of conceptual-semantic and lexical relations. WordNet has been used to find synonyms / hyponyms / hypernyms of keywords retrieved in the annotation of the entity, in the case that exact keyword is not present on FrameNet.

Ontology	Entity corresponding to the concept of “emotion” in NL	Comment or annotation of the entity	Frames evoked (choosing among FrameNet’s ones)
EMO/MFOEM	emotion process	An emotion process is a mental process that is a synchronized aggregate of constituent mental processes including an appraisal process, which is valenced, has an object, and gives rise to an action tendency.	Mental_activity, Assessing (has not-annotated lexical unit: appraisal), Process, Inclination, Response, Body_movement, Position_on_a_scale, Degree
EMO/MFOEM	Emotion [stated as equivalent to the emotion process class in EMO]	An affective state of consciousness	Awareness(has not-annotated lexical unit: consciousness), Awareness_status, Awareness_situation, State
EmOCA	Emotion	After perceiving a stimulus (internal or external), a person first experiences an immediate bodily reaction, and secondly a cognitive process is launched which uses the bodily expression and cognitive valuation to define their emotion. [definition from the article]	Categorization, Stimulate_emotion, Stimulus_focus, Cause_to_start, Response, Biological_mechanisms, Body_movement, Assessing
EmotionsOnto	EmergentEmotion	A state where the person’s whole system is caught up in the way they react to a particular person or situation. Aspects involved: valence, impulses to act or express, distinctive changes in the body, short duration. [definition from the article]	State, Response, Stimulate_emotion, Degree, Timespan
SOCAM affective extension	Affective [subclass of State]	Emotion is an affective state which has some valence, activation and intensity. A person can be in the state of a current emotion, which can be induced by location or time. [definition inferred from the article]	State, Degree, Timespan
Emotional Cues Ontology	Emotion	It is possible to define emotions as composed of complex emotional cues, which are composed of other simple emotional cues, which are based on some media properties. [definition from the article]	Body_movement, Facial_expression
CemoT	Emotion	Emotions are concept built by the mind driven by simulations, based on the personal experience, knowledge and history of social interactions of the experiencer. They are subjective but can be shared with others. The emotion is content of the emotion experience.	Awareness (conception is a LU of the frame), Entity, Feeling, Experiencer_focused_emotion, Mental_activity

2. Mapping frames and ontologies

In order to identify the effective mapping between ontologies and frames, we go through ontology entities and for each entity we compute any possible mapping between the entity and the frames selected in the previous step (i.e. those evoked by its annotations). In frame semantics, a frame is characterized by its roles (also called frame elements) and each element possibly define the semantic type of the individual that can play that role in the frame.

Frame	Definition	Role [and Semantic Type]
Mental activity	In this frame, a Sentient_entity has some activity of the mind operating on a particular Content or about a particular Topic. The particular activity may be perceptual, emotional, or more generally cognitive. This non-lexical frame is intended primarily for inheritance.	Content (Content), Expressor, Sentient_entity (Sentient), Topic, Degree (Degree), Evidence, Manner (Manner), Paradigm, Role, Time (Time)
Assessing	An Assessor examines a Phenomenon to figure out its Value according to some Feature of the Phenomenon. This Value is a factor in determining the acceptability of the Phenomenon. In some cases, a Method (implicitly involving an Assessor) is used to determine the Phenomenon's Value.	Assessor (Sentient), Feature, Medium, Method, Phenomenon, Beneficiary, Circumstances, Co-participant, Degree, Depictive, Duration, Evidence, Explanation, Frequency, Manner (Manner), Means (State_of_affairs), Place (Locative_relation), Purpose (State_of_affairs), Result, Standard, Time, Value
Process	This frame describes a complex event which lasts some amount of time, consisting of a beginning stage, a stage where the process is ongoing, and a finish or end.	Process, Duration, Explanation, Manner (Manner), Place (Locative_relation), Result, Time (Time)
Inclination	Due to internal forces/properties, an Entity is inclined towards a particular Behavior, which may be to perform an action or to maintain a state. The underlying force is seen as being beyond the Entity's control. The Degree to which the Entity is inclined towards the Behavior may also be specified as well as the Basis for the inclination.	Entity, Behavior, Basis, Containing_event, Degree
Response	An Agent performs a Response action in consequence of a Trigger event. In many cases, a non-agentive Responding_entity causes the Response after the Trigger occurs.	Agent (Sentient), Responding_entity (Agent), Trigger, Response, Dimension, Manner (Manner), Place (Locative_relation), Purpose (State_of_affairs), Role, Time (Time)
Body_movement	This frame contains words for motions or actions an Agent performs using some part of his/her body.	Agent (Sentient), Body_part (Body_part), Addressee, Area, Cognate_event, Coordinated_event, Degree, Depictive, Duration (Duration), External_cause, Goal [Goal], Internal_cause, Manner (Manner), Message (Message)...
Degree	LUs in this frame modify a gradable attribute and describe intensities at the extreme positions on a scale.	Gradable_attribute
Awareness	A Cognizer has a piece of Content in their model of the world. The Content is not necessarily present due to immediate perception, but usually, rather, due to deduction from perceivables.	Cognizer (Sentient), Content (Content), Expressor, Topic, Degree, Evidence, Explanation, Manner, Paradigm, Role, Time
Categorization	A Cognizer construes an Item as belonging to a certain Category.	Cognizer (Sentient), Category, Criteria, Item, Circumstances, Concessive, Explanation, Manner (Manner), Means (State_of_affairs), Time (Time)
Stimulate_emotion	Some phenomenon (the Stimulus) provokes a particular emotion in an Experiencer.	Experiencer (Sentient), Stimulus, Circumstances, Degree, Depictive, Explanation, Manner (Manner), Means (State_of_affairs), Result, Time (Time)
State	An Entity persists in a stable situation called a State	Entity, State
Entity	This frame is for words that denote highly schematic entities.	Entity, Constituent_parts, Formational_cause, Name, Type

In the original model proposed for this type of alignment, it is now necessary to compute the similarity of the entities with the evoked frames, its elements, and its semantic types: an ontology entity may correspond to one of these components defined in the evoked frames. However, this is not the case. These frame-based specifications in our case it is useful more than anything else to identify the most common frames and roles, and therefore most significant for the meaning of emotion among the different ontologies.

Conclusion

Anyway, comparing frame-based specifications of ontology entities – which is a step of the alignment thought to finally and definitely align entities – could be useful for our purposes of retrieving most relevant frames and frame elements. Summing up:

1. An emotion is mainly defined as a State;
2. An emotion involves Assessing (and possibly Awareness or Categorization), a Response – such as a Body_movement – or an Inclination [to respond]. Since it involves or appear as a Response, it is also linked with a stimulus (Emotion_stimulus).

Designing FramECO

Why do we need something different?

Comparing existing models in a frame-based alignment, after having explained the implicit formalizations (as far as possible), it makes evident the discrepancies between definitions of the concept of "emotion" coming from different sources. Existing ontologies do not allow us to accurately describe emotions, especially those outside the norm (emerging, anonymous or meta-emotions), as they are still linked to restricted or outdated cognitive theories. We therefore need a model that is innovative from the point of view of contents, or, better, in defining a concept of emotion suitable for identifying also uncommon emotions, not yet part of a taxonomy or psychological theory. At the same time, one could think of a model that could be the formal definition of new theories on emotions, such as that of professor and scientist Lisa Feldman Barrett.

The focus therefore becomes the context of an emotion, so the best formalization to express a concept based on its semantic context of meaning is certainly the frame semantics. The privileged tool for a frame-based ontology formalization is FrameNet, but there are many other relevant resources in the context of frame semantics (e.g., FRED). But even an extraordinary tool like FrameNet appears insufficient, in its current state, if it is used for ontology design.

In fact, although in FrameNet there is a distinction between core elements, which are essential to the expression of the meaning of the frame, and non-core elements, which are merely descriptive; there is no logical formalization which allows to distinguish universal axioms from existential ones, which indeed allows to clarify which elements are strictly necessary for the definition of emotions and which are not, and of the latter which are their domains and ranges or subclasses. In short, the aim is to try to exploit a linguistic resource such as FrameNet for the design of a proper frame-based ontology.

Another reason why a tool like FrameNet is preferable to other more traditional ones is that it allows us to express a network concept of emotion and emotions that is much more interesting than the simple division by types, according to which complex emotions are compositions of the emotions of basic and basic emotions are defined diametrically on the basis of positive and negative valence. In other words, instead of giving us a purely hierarchical organization of knowledge (which in any case exists in part, just think of the fact that some elements of the frame are hereditary), there is also partially a reticular structure that allows us to formulate definitions and wider models.

Domain and application tasks: the domain and the goal of the ontology

We are considering affective sciences as primary domain (emotions are the object of study of affective sciences such as cognitive psychology), but eventually also literature (because on the one hand, the vocabulary on which FramECO is built comes from the humanistic field; and secondly because a potential use of the resource would be in the context of the sentiment analysis of literary texts).

The application task can be summarized as formally defining unfamiliar/anonymous emotion.

Competency questions

Competency questions to highlight and analyse the most relevant features of the subject: these are the questions we want to be answered through SPARQL queries. The CQs results listed a range of different characteristics our module needed to take into account in order to effectively model it.

What is an (unfamiliar) emotion?

What is the context of an emotion? Which are the elements necessarily defining an emotion? Which elements make an emotion detectable (of its context)?

Can an emotion be valenced?

What’s in FrameNet

FrameNet Emotion and Feeling frames:

The close relationship of meaning between the concept of "emotion" and the concept of "feeling" is ambiguous, and it seems to resume the WordNet definitions:

• emotion: any strong feeling
• feeling: the experiencing of affective and emotional states

Indeed, feeling is the direct hypernim (term with broader meaning) of emotion, but the feeling synset is defined according to emotion, which is its direct hyponym (specific instance of the hypernym). This may suggest a distinction between the emotion as a phenomenon, and the feeling as the subjective (as “of the subject”) experiencing of the emotion itself. In FrameNet, “emotion” appear as one of the lexical units of the “feeling” frame, but there are plenty of frames regarding emotion as an independent concept from feeling.

Other hyponims of feeling – so other sister terms of emotion: mood (temper, mood, humor: a characteristic (habitual or relatively temporary) state of feeling); affect (the conscious subjective aspect of feeling or emotion); and sentiment (tender, romantic, or nostalgic feeling or emotion).

Among all the frames related to the concept of “emotion” in some way, two of them appear to be more relevant because among the lexical units related to them there are the traditional and well-know emotions:

As we can see, most of the elements of the proposed frames are synonymous or semantically similar to the keywords used in existing models to describe the concept of emotion.

Some emotions can be found as lexical units also for two other frames: Emotions_of_mental_activity and Emotions_by_stimulus. Then there are some frames whose lexical units are mostly verbs that arouse emotions: Cause_emotion and Stimulate_emotion. Finally, the last frames related to emotions are the following: Emotion_heat and Emotions_by_possibility.

Conclusion of what’s on FrameNet

There are many frames related to the concept of emotion, some more consistent and relevant than others on a formal and characterization level. Emotions such as happiness, sadness, anger appear as lexical units of some frames. However, it lacks an underlying organicity and consistency, as well as an appropriate formalization.

Dictionary for Obscure sorrows: what is it and how can it help building a more comprehensive model for emotions

The Dictionary of Obscure Sorrows is an online resource, which is a “compendium of all the aches, demons, vibes, joys and urges that roam the wilderness of the psychological interior. The author’s mission is to harpoon, bag and tag wild sorrows then release them gently back into the subconscious.” (from the website). Among all the new words which are entries of the dictionary, some of them could be considered as emotions: something that we might feel at some point in our lives or about some topic, an event. Since they are all new words anyway, we could define them as “anonymous emotions” or “unfamiliar emotions” – since we may have already experienced some of them, but they are not the well-known ones. Three of them are reported as significant examples:

Onism
n. the frustration of being stuck in just one body, that inhabits only one place at a time, which is like standing in front of the departures screen at an airport, flickering over with strange place names like other people’s passwords, each representing one more thing you’ll never get to see before you die — and all because, as the arrow on the map helpfully points out, you are here.

This first example represents all those emotions, described in the dictionary, which appear to be specifications of already defined, known, existing emotions.

Midding
v. intr. feeling the tranquil pleasure of being near a gathering but not quite in it—hovering on the perimeter of a campfire, chatting outside a party while others dance inside, resting your head in the backseat of a car listening to your friends chatting up front—feeling blissfully invisible yet still fully included, safe in the knowledge that everyone is together and everyone is okay, with all the thrill of being there without the burden of having to be.

This second example instead represents those emotions that are not specifications of pre-existing emotions, but are described as "feeling".

Ecstatic shock
n. the surge of energy upon catching a glance from someone you like—a thrill that starts in your stomach, arcs up through your lungs and flashes into a spontaneous smile—which scrambles your ungrounded circuits and tempts you to chase that feeling with a kite and a key.

Finally, this third example represents all definitions that are implicitly emotions or involve emotions.

For the sake of brevity and necessity, we will focus here only on the second type of "anonymous emotion" that is present in the Dictionary of Obscure Sorrows: before we model emotions that appear as specifications of existing emotions, we want to try to model those that are independent of them.

A new frame for Emotion: FEs and key concepts (glossary of terms)

Let’s see if they have some elements in common, i.e., considering the definitions in order to define a possible Unfamiliar_emotion frame in FrameNet, we can try to spot the frame elements and frame relations. The definition is inspired by those present on FrameNet, while the roles proposed as core or non-core are also taken from FrameNet, but modified to better suit our purpose and to have a unique and consistent definition of these elements (while in FrameNet themselves FEs have different definitions in different frames).

Chrysalism
n. [feeling] the amniotic tranquility of being indoors during a thunderstorm, listening to waves of rain pattering against the roof like an argument upstairs, whose muffled words are unintelligible but whose crackling release of built-up tension you understand perfectly.
Midding
v. intr. feeling the tranquil pleasure of being near a gathering but not quite in it — hovering on the perimeter of a campfire, chatting outside a party while others dance inside, resting your head in the backseat of a car listening to your friends chatting up front — feeling blissfully invisible yet still fully included, safe in the knowledge that everyone is together and everyone is okay, with all the thrill of being there without the burden of having to be.
Rückkehrunruhe
n. the feeling of returning home after an immersive trip only to find it fading rapidly from your awareness — to the extent you have to keep reminding yourself that it happened at all, even though it felt so vivid just days ago — which makes you wish you could smoothly cross-dissolve back into everyday life, or just hold the shutter open indefinitely and let one scene become superimposed on the next, so all your days would run together and you’d never have to call cut.

Compared with those of well-known emotions:

Happiness
n. the state of feeling pleasure or satisfaction.
Frustration
n. the feeling of being annoyed or less confident because you cannot achieve what you want, or something that makes you feel like this.
Surprise
an unexpected event, or the feeling caused when something unexpected happens.

We can propose a frame:

Emotion

An Experiencer is in a particular emotional Affective State with respect to some Circumstances under which the Experiencer experiences the emotion. Circumstances do not give the specific content for the emotion, but the Content can often be inferred from the Circumstances. The Content may refer to an actual, current state of affairs, but quite often it refers to a general situation which causes the emotion.

FEs:

Core:

Experiencer [Exp], Semantic Type: Sentient
The Experiencer is the person or sentient entity that experiences or feels the emotion or other internal state.

Circumstances [cir]
The Circumstances FE encodes the circumstances or conditions under which the Experiencer experiences the emotion. Unlike Content it does not give the specific content of the emotion, but the Content can often be inferred from the Circumstances.
When returning home, I always feel like it is fading from my awareness.

Non-core:

Content [Cont], Semantic Type: Content
Content is what the Experiencer's feelings or experiences are directed towards or based upon. The Content differs from a stimulus because the Content is not construed as being directly responsible for causing the emotion.
I am afraid of spiders. My enjoyment of the movie was considerably impaired by the seven-foot guy sitting in front of me.
Smithers takes great pleasure in collecting matchboxes.
Sergio derives great pleasure from smoking.
I’m feeling nostalgic about the past that I’ll never be able to live.

Event
The Event is the occasion or happening that Experiencers in a certain emotional state participate in.
At the wedding I was happy when I was close to the group of guests but not among them.

Expressor
The Frame Element Expressor marks expressions that indicate a body part, gesture or other expression of the Experiencer that reflects his or her emotional state. They describe a presentation of the experience or emotion denoted by the adjective or noun.
The delight on her face was mixed with astonishment.

Stimulus
The Stimulus is the person, event, or state of affairs that evokes the emotional response in the Experiencer.
Pietro’s return home caused happiness in Teresa yesterday.

Degree
The Degree is the degree to which the Experiencer feels the emotion.
Smoking gives me extreme pleasure.

Manner
Any description of the way in which the Experiencer experiences the Emotion which is not covered by more specific FEs, including secondary effects (quietly, loudly), and general descriptions comparing events (the same way). Manner may also describe a state of the Experiencer that affects the details of the emotional experience.
At the wedding, Anna felt sad the same way she felt at the funeral.

Valence
Valence represents that dimension of a feeling as positive or negative, pleasant or unpleasant, good or bad.
Clara had a good feeling about the audition.

Frame-frame Relations:

inherits from: Affective State [State]
The Affective State is the abstract noun that describes a more lasting experience by the Experiencer.

What it is new and what it is not: to reuse or not reuse?

In formulating a possible definition for the Unfamiliar_emotion frame, among the frames already present on FrameNet the one most similar to what we would like to describe is the Experiencer_focused_emotion frame. As the name of the frame suggests, this definition of emotion moves away from the more well-known ones that focus on the stimulus that causes the emotion and the behavior that follows. For this reason, since this frame is more suitable for our purposes, we have reused some of its FEs.

For what concern already mentioned ontologies, we cannot directly reuse one of them since they represents because they are models conceived and designed focusing on modeling the neurological / biological / physiological process in which the emotion is supposed to consist, or focusing on the external manifestation of the emotion on the subject. Since this is not our focus, and neither our scope is the medical-clinical application of framECO, we can only recover some concepts from these existing ontologies, but not consistently integrate them into framECO.

What distinguishes this frame, and therefore this concept of emerging emotion, from the concept of emotion that emerges from the traditional models that we have collected and from those proposed in different frames by FrameNet:

1. emotions are not necessarily defined by valence or degree in a valenced way: that is mainly because unfamiliar emotions are not defined by a clear valence, either positive or negative; even if traditional well-known emotions might be – that’s why “degree” and “valence” are proposed as non-core elements;
2. emotions are not hierarchically ordered between them: there are no basic emotions from which complex ones should derive;
3. they are not necessarily caused by a Stimulus and the Content of the emotion is not necessarily expressed, understood as the general situation that causes the emotion – rather, Stimulus and Content are proposed as non-core elements;
4. the main focus is not the external or physical expression of the emotion, but rather the emotion itself;
5. the emotion as a state or process is not distinguished by the subjective emotion feeling.

What this definition of emotion has in common with those proposed by existing models:

1. it is an affective state, which is a state: most of the theoretical models proposed define emotions as states;
2. can bevalenced;
3. can have a stimulus and can have content
4. can have a physical expressor of the emotion: this expressor has been defined in MFOEM as the action tendency rose by the emotion; by EmOCA as immediate bodily reaction to a stimulus; and by Emotional Cues ontology as the emotional cues based on media properties which characterize emotions.

Talking about the distinction between emotion, feeling and mood, existing models do not agree, and are inconsistent even in themselves. This theoretical confusion mentioned earlier, added to the fact that in framECO we do not want to distinguish perception or experience from the physical-biological phenomenon, it leads us to overcome this distinction. Indeed, we ought not make formal distinctions between emotion and feeling, but of course an emotion should be defined as an (affective) state. On the other hand, we can neither distinguish a framECO-defined emotion from a mood, since traditionally [like in wordnet and mfoem] emotions are distinguished from moods because the former necessarily have a content or object. Here, we want to overcome this definition to include unfamiliar emotions, which are not necessarily directed towards an object or risen from a content. However, a distinction could still be made based on the fact that an emotion has a circumstance while a mood might not be defined according to some circumstances, an emotion is usually stronger than a mood, and a mood is usually habitual or characteristic or less temporary (long-lasting) than an emotion (WordNet, MFOEM).

Thus we can say that the definition of this frame might be suitable also to describe traditional well-know emotions, thanks to the non-core elements, which conceptually are closer to old and established theoretical models for emotion description. In any case, one of the goal of this project is to be theoretically independent from any theory of emotions, so thanks to the alignment proposed earlier, we can grasp the elements used to describe emotions that are common to more than one model. The ultimate goal of framECO is to define the context both of traditional and unfamiliar emotions.

Even if we are now reusing existing ontologies, such as an upper ontology, we still reuse one of the most common OWL design pattern: enumerated class, which will be explained in depth later.

From frame to OWL ontology class: organizing concepts in classes and hierarchies among classes, relations in properties and individuals

Conceptual model of framECO

Our goal now is therefore to rigorously formalize this definition, defining which elements of the frame are necessary for its definition and which ones can be accessories. It is a distinction that goes beyond that between core elements and non-core elements, and goes in the direction of a descriptive logic of the first order. In other words, it is a question of proposing an axiomatic distinction, which is currently lacking in FrameNet, between those elements that are always there and those that could be there. This logical definition will serve as foundations for building the ontology, since framECO is defined in OWL-DL language (Description Logic Semantics, also Direct Semantics, is chosen as semantics for OWL – rather than RDF/XML syntax) and built with Protégé.

Before getting to the heart of the logical formalization, a premix is necessary. Initially, the idea was to translate FrameNet’s frames into classes, frame-frame relations into binary relations between classes, and frame elements as fillers. However, frame-frame relations are often not consistent or well-defined; we are not so much interest in many other frames a part from Emotion, State and few others; and since they can all be considered as stand-alone concepts, both core and non-core elements can be defined as classes (and not just fillers of existential restrictions) of framECO ontology.

Defining core elements and non-core elements as classes

framECO will be composed of classes that define the emotion, either necessarily or not: AffectiveState, Experiencer, Circumstance, Content, Event, Expressor, Stimulus, Degree, Manner, Valence, State (superclass);

Differences between classes derived from core elements and classes derived from non-core elements

Core elements are classes linked with Emotion class with existential restriction:

When we assign a property to a class using restrictions, it means that this property is necessary for defining that class. Using existential restriction means that having circumstances, an affective state and an experiencer is necessary for a thing to be an Emotion; or if a thing is an Emotion it must always have a Circumstance, a State and an Experiencer. Existential restrictions on the three core elements:
hasCircumstance some Circumstance
hasExperiencer some Experiencer
These are the class expressions to whom is subclass the defined class Emotion, that for now is a primitive class, because they are necessary to define that class, but not sufficient for themselves (if they were necessary and sufficient conditions, the class Emotion would be equivalent to these class expressions). With this we have stated that for every instance of the class Emotion there must be a hasCircumstance property instance to an instance of Circumstances and a hasExperiencer property instance to an instance of Experiencer.
However, we can say that Emotion is equivalent to the intersection of AffectiveState, and the two class expressions mentioned above (since we suppose that other AffectiveState can have an experiencer, but only emotion is defined by having both an experiencer and a circumstance). So emotion can be considered as a defined class since it is equivalent to: AffectiveState and (hasExperiencer some Experiencer) and (hasCircumstance some Circumstance). With this we have stated that if we have an individual, and that individual is a member of the class “AffectiveState”, and hasExperiencer some Experiencer and hasCircumstance some Circumstance; that means that individual is an Emotion.

The Emotion defined class can be finally defined:

Defining non-core elements

Non-core elements are defined either as ranges of properties with domain Emotion (thus, when these properties are used, we can infer that the thing is an emotion); or as ranges of properties with a domain that is the union of the Emotion class and another class from the Emotion class (elements that are also present in other frames besides emotion)

Some properties (regarding both core and non-core elements) are functional property

A functional property is a property that can have only one (unique) value y for each instance x, i.e. there cannot be two distinct values y1 and y2 such that the pairs (x,y1) and (x,y2) are both instances of this property. Thus, to add precision to framECO we can state that both :hasExperiencer and :hasCircumstance are functional properties for instance.

Cardinality

All emotions have exactly one experiencer. So we need to add cardinality to the hasExperiencer property as a restriction for the Emotion class.

Adding other constraints on the properties to define a class by its description

Apart from the existential restrictions specified above to define the Emotion class, the Valence class is also defined according to a description – but this time not a description based on constraints: the enumeration. Valence is a defined and enumerated class since its three possible individuals are specified (Positive, Negative and Neutral) and its asserted as equivalent to the class expression that is the set of the individuals.

Final Knowledge Graph

Ontology Documentation

To read the full documentation of framECO ontology, visit the documentation page.

Sample of possible SPARQL queries

What is an (unfamiliar) emotion?

SELECT ?thing
WHERE { framECO:Emotion rdfs:subClassOf ?thing }

Result:

AffectiveState

What is the context of an emotion? Which are the elements necessarily defining an emotion? Which elements make an emotion detectable (of its context)?

SELECT ?class
WHERE { framECO:Emotion ?p ?class .
VALUES ?p { rdfs:subClassOf owl:equivalentClass } }

Result:

AffectiveState
ObjectIntersectionOf(AffectiveState ObjectSomeValuesFrom(hasCircumstance Circumstance))

Can an emotion be valenced?

SELECT ?valence
WHERE { ?valence rdf:type framECO:Valence. }

Result:

Neutral
Positive
Negative

FramECO for Sentiment Analysis

It is possible to think of a further categorization of emotions, which can somehow distinguish the unfamiliar or anonymous ones from the well-known and precisely defined ones. It is a question of distinguishing emotions from each other not by value but by determining elements for the onset or content of the emotion itself, and therefore conceptually closer to FrameNet frames.

Looking at the frame-frame relations in FrameNet could help understanding where to locate this frame for unfamiliar emotions in the frames hierarchy:

In this case the definition of emotion that is the focus of framECO would correspond to a frame, which could be called Unfamiliar_Emotion or Circumstanced_Emotion, which would turn out to be just a perspective on the Emotions frame, next to the Experiencer_focused_emotion and Stimulus_Emotion frames.

Furthermore, in this way, the distinction between core and non-core elements would be overcome as necessarily present to define an emotion or not. It is a useful distinction in an abstract model of emotions, but in natural language, given also the complexity of emotion as an expressed concept, it is much more likely that emotions can be defined each time with different elements, and that there never are any of constants. So in an ontology that has textual annotation as its purpose, it may be preferable to substitute the distinction between the elements that are instantiated and that are not to the core-non-core distinction.

The absence of core elements intended as elements necessary for the definition of emotion does not preclude the possibility of detecting the emotion in the text, which could in any case occur through the verb "feeling", "state" and their declensions. Indeed, for the application of framECO in sentiment analysis - or in general in text annotation activities - a possible issue that should be overcome concerns the distinction between feeling and emotion.

This kind of ontology would be more suitable in (semi-)automated parsing of natural language sentences, similarly to tools like FRED or other existing projects or methods (e.g., Jan Scheffczyk et al., Ontology-based Reasoning about Lexical Resources).

Basing the modification of framECO for annotation purposes on the cited work, the new version of the ontology would be:

And a parsing could be similar to:

Conclusions and References

An emotion has a complex scenario in which it is inserted, and therefore being able to formalize it and make it come out automatically or transform a resource into a formal resource to enhance humanistic resources (such as the Dictionary for Obscure Sorrows) without trivializing them, to make them potential also in the computational field.

Ontologies are needed for classification as well as for automatic parsing of text in natural language for disciplines such as sentiment analysis, but in both cases the formalization must be consistent and the contents relevant.

FramECO for now is only the first step in this direction.

References

• López, Juan & Gil, Rosa & García, Roberto & Collazos, César. (2014). EmotionsOnto: an Ontology for Developing Affective Applications. JOURNAL OF UNIVERSAL COMPUTER SCIENCE. 20. 1813-1828.
• Janna Hastings, Werner Ceusters, Barry Smith, and Kevin Mulligan, The Emotion Ontology: Enabling Interdisciplinary Research in the Affective Sciences, in Proceedings of Modeling and Using Context, 7th International and Interdisciplinary Conference CONTEXT 2011, Karlsruhe, Germany, September 26-30, Springer, 2011.
• Josef Ruppenhofer, Michael Ellsworth, Miriam R. L. Petruck, Christopher R. Johnson, Collin F. Baker. Jan Scheffczyk, FrameNet II: Extended Theory and Practice, 2016.
• Jan SCHEFFCZYK, Adam PEASE, Michael ELLSWORTH, Linking FrameNet to the Suggested Upper Merged Ontology, in Formal Ontology in Information Systems, B. Bennett and C. Fellbaum (Eds.), IOS Press, 2.
• Luigi Asprino, Valentina Presutti, Aldo Gangemi, Paolo Ciancarini, Frame-Based Ontology Alignment, in Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17), 2017, Association for the Advancement of Artificial Intelligence
• Stefano Baccianella, Andrea Esuli, and Fabrizio Sebastiani, SENTIWORDNET 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining
• Paulo Hauck, Regina Braga, Fernanda Campos, Tiago Torrent, Ely Matos, José Maria N. David, Supporting FrameNet Project with Semantic Web technologies
• Josef Ruppenhofer, The treatment of emotion vocabulary inFrameNet: Past, present and future developments
• Rebecca Lin, Muhammad “Tuan” Amith, Chen Liang, Rui Duan, Yong Chen and Cui Tao, Visualized Emotion Ontology: a model for representing visual cues of emotions.From The 2nd International Workshop on Semantics-Powered Data Analytics Kansas City, MO, USA. 13 November 2017. https://doi.org/10.1186/s12911-018-0634-6
• Franck Berthelon, Peter Sander. Emotion Ontology for Context Awareness. Coginfocom 2013 - 4th IEEE Conference on Cognitive Infocommunicaitons, Dec 2013, Budapest, Hungary. ￿hal-00908543
• López, Juan & Gil, Rosa & García, Roberto & Collazos, César. (2014). EmotionsOnto: an Ontology for Developing Affective Applications. JOURNAL OF UNIVERSAL COMPUTER SCIENCE. 20. 1813-1828.
• Kuderna-Iulian Benta, Anca Rarău, and Marcel Cremene. 2007. Ontology Based Affective Context Representation. In Proceedings of the 2007 Euro American conference on Telematics and information systems (EATIS '07). Association for Computing Machinery, New York, NY, USA, Article 46, 1–9. DOI:https://doi.org/10.1145/1352694.1352741
• Obrenovic, Z. & Garay-Vitoria, Nestor & López, Juan & Fajardo, Inmaculada & Cearreta, Idoia. (2005). An Ontology for Description of Emotional Cues. 505-512. 10.1007/11573548_65.
• Mathieu Y.Y. (2005) Annotation of Emotions and Feelings in Texts. In: Tao J., Tan T., Picard R.W. (eds) Affective Computing and Intelligent Interaction. ACII 2005. Lecture Notes in Computer Science, vol 3784. Springer, Berlin, Heidelberg. https://doi.org/10.1007/11573548_45
• Lisa Feldman Barrett, How Emotions Are Made: The Secret Life of the Brain, Houghton Mifflin Harcourt, 2017
• David Sander, Klaus Scherer, Oxford Companion to Emotion and the Affective Sciences, Oxford University Press; Reprint edizione, 2014.