In this study, we introduce a novel approach to understanding the emotive content present in ancient literature, specifically focusing on the first Book of Homer's Iliad. Recognizing the challenges inherent in interpreting emotions from ancient texts, we developed a Domain-Specific Language (DSL) tailored for this purpose. This DSL not only allows for the annotation of basic sentiments such as positive, negative, neutral, or mixed but also facilitates the identification and categorization of specific emotions. To ensure the relevance and applicability of our annotations, we mapped the extracted emotions to some authoritative domain ontologies. This mapping process aids in bridging the gap between ancient emotional expressions and contemporary understanding. Our preliminary results, which we discuss in detail, highlight the potential of our approach in ofering deeper insights into the emotional landscape of ancient texts. We believe that our methodology can serve as a foundation for future studies aiming to decode emotions in historical literature.
A group of 15 volunteers (students and scholars) are asked to annotate the first book of the Iliad on the Euporia [ 12] web platform developed by CoPhiLab.1 and they are provided with detailed guidelines, based on the syntax and semantics of the Domain-Specific Language (DSL) created for this task. A DSL is a formal language, usually defined by a Context-Free Grammar, which is compact and familiar to the user because it is optimized for limited purposes. Each participant annotates the same portion of the text, resulting in the first book being annotated in its entirety 15 times. As shown in Fig. 3, an annotation is constituted by a reference to the Homeric text and a sequence of one or more structured sentences, ending with a punctuation mark. If we consider the first line of the bottom box, 1.11e¯tímasen (which means: [he, i.e. Agamemnon] dishonored, disdained [Chryses]) is the reference to the text, followed by two sentences: a lexical observation and a situational observation. The structure of a sentence extends the syntax of the scripting language turtle: it is constituted by subject(s), predicate(s) and object(s), each of which can be preceded by hashtags to categorise them; the object(s) can be followed by a recipient (or cause) of the emotion. In our example, the first sentence is the lexical observation: #lex this expresses #quite_neg shame, humiliation. In this case the annotator asserts that e¯tímasen expresses two quite negative emotions: shame and humiliation. The second sentence is the situational observation: #character Chryses feels #neg humiliation. As pointed out by Luraghi and Sausa [6], an emotion is a complex experience that involves an experiencer (the subject), an expertum (the emotion) and a stimulus (the cause of the emotion, that can be mentioned or not). The classifiers for experiencers are: #character (an Iliadic hero), #narrator (that can be the_poet, or an internal narrator), #recipient (that can be the_ancient_audience or the_contemporary_audience), and #annotator (that is always expressed by “I”). In this way, multiple perspectives can be captured: e.g. diferent characters may experience diferent emotions in the same scene.
The polarity of the emotions are graded on a scale of seven degrees from #very_neg to #very_pos.
The annotations are parsed with a compiler compiler2 and converted in XML or json to facilitate data analysis.
1https://cophilab.ilc.cnr.it/ A prototype of the platform, which is an app for eXist-db, is available at https://github.com/CoPhi/euporia 2We use the compiler compiler ANTLR: https://www.antlr.org/ x‘ Because the annotators are free to add new emotion terms to their list, we structure these terms a posteriori by mapping them to the following ontologies: the Emotion Ontology (MFOEM), the Semanticscience Integrated Ontology (SIO), Visualized Emotion Ontology (VEO), and the National Cancer Institute Thesaurus ontology (NCIT ).
MFOEM [13] applies a cognitive perspective and builds upon MF (Mental Functioning) and BFO (Basic Format Ontology) and considers afective phenomena, e. g. emotions, moods, appraisals as well as subjective feelings.
SIO [14] uses a simpler approach but assigns positive and negative polarities to the emotions.
VEO [15] builds on MFOEM, but aims at the visual representation of emotions.
NCIT [16] applies a clinical perspective and distinguishes between emotions and feelings.
We take into account also the Time Event Ontology, TEO, [17] to shape temporal aspects of the annotations, such as the phases of a complex event that involves multiple emotions.
To better clarify the diferences between the diferent ontologies, we shall now describe how Anger is classified in each: • MFOEM, anger < emotion < afective process < mental process < bodily process < process < occurrent < entity; • SIO, anger < disgust < hostility < negative emo
tion < emotion < behaviour < process < entity; • VEO, anger < emotion < afective process < mental process < behaviour < action < bodily process < process < occurrent < continuant < entity < consequence < aspect; • NCIT, anger < emotion < mental process < neurologic process < organismal process < biological process.
We decided to use MFOEM as the main reference ontology, because it is more suitable for our purposes.In fact, both MFOEM and our ontologies have a cognitive perspective. We then proceeded to map the list of emotions extracted from the annotations to MFOEM, SIO, VEO and NCIT.
Only 7 of these emotions - anger, fear, hope, joy,
sadness, satisfaction - were included in all the ontologies of
emotions, but other 32 terms can be mapped to one or
another of them. Besides 7 unfound terms, all the
remaining terms were classifiable through MultiWordNet 3, but
as hyponyms of the following synsets: feeling (
with values present, future; • agent, meaning who perceives the emotion, with the values oneself, external. There is also one instance - submission - of oneself toward external; • valence, with the values, positive, negative, am
biguous; • consequence, i.e. when the result of the emotion will take place and what type of result will be, with the values unpleasant, pleasant, expected, unexpected, actual, future. In one instance - suspense - it was not possible to determine the efect of the emotion. For example, empathy is classified as somebody’s reaction for an actual consequence to an event happened toward another agent Therefore, it is perceived in the present by oneself with a negative and it has an actual, unpleasant consequence. Figure 2: Frequency of annotated emotions
By comparing the list of emotions in Pavlopoulos et al. [1] and the list extracted from the current annotations, 2 terms are missing: guilt, loneliness and 10 are new en- regarded encouragement, gratitude, and shock. In 62% of tries: scorn, threat, acknowledgement, warlike, sadness, the annotations of anger, the most frequently annotated betrayal, contempt, disrespect, emotion, rage. emotion, the polarity was negative (very negative in 5, quite negative in one) while in the rest it was neutral. As 4. Current results can be seen in Figure 4, the number of polarity-carrying emotions per verse varies. One of the highest peak was The most frequently annotated emotion was anger (Fig- observed in verse 474, where Apollo is satisfied by the ure 4), with 97 occurrences. The emotions of respect, ag- song sang to him by the Achaeans. All annotations were gression, and fear followed with less than 50 occurrences positive. each. On the other hand, the most infrequent annotations Currently only 4 annotators out of 15 have completed their tasks.
anger aggrreesspseiocnt
fear request sorrow
fury empathy
threat admiration awe worry piety disdain hurmepiliraotaiocnh cominpjaussstiiocne
grudge comshpalaminet joy love bravery sadness affection justice expectation
rage anxiety
plea sarcasm acknowledgement disrespect hope pride subrmemisosriosne rejection desire pity relief revenge intimidation appdeiaspseamsseionnt impiety hostility
pain willingness distress
guilt suspicion confidence satisfaction lust hate aanwnaoryeannecses
irony repulsion approval question cowraertdreicaet
honour being_moved exasperation
range self_pity ambcoivnatleemncpet blame courage dishonour
discord homesickness plead hostile regret aggrevation
jealousy disdaisgarepepmroevnatl
doubt threatening self_denial irritation kindness solidarity patience mockery bemupsleeamseunret agreement passive_aggresiveness
misery unemotional provocation insecurity indignation contentment
shock gratitude encouragement occurrences1 50 00
The DSL context-free grammar, the reference text of the first book of the Iliad encoded in XML-TEI, the updated annotations and the script to convert the DSL in XML are available at https://github.com/CoPhi/ emohomer.
We are grateful to Giota Nikolopoulou and to the students and scholars that are annotating the emotional aspects related to the first book of the Iliad. This study developed a domain-specific language for the annotation of emotive content of the first Book of Homer’s Iliad. We mapped the list of emotions we extracted from the annotations and we discussed the results. Next steps comprise the study of polarity for more emotions, an exploration of verses with contradicting polarity, and sentiment analysis based on the subject’s role (annotator, character, or narrator). Also, we plan to employ more annotators, in order to measure inter-annotator agreement, and study verses which provoke consistent and diverse emotions to the diferent annotators.