Computational Semantic Lexicon of French Verbs of Emotion

The study of emotion verbs in French presents a unique challenge for natural language processing (NLP) and computational linguistics. Unlike concrete verbs that denote physical actions, emotion verbs describe internal states, psychological reactions, and interpersonal causalities. Constructing a computational semantic lexicon for these verbs requires a structured approach that captures both syntactic behavior and fine-grained semantic nuances.

The Complexity of Emotion Verbs

French emotion verbs, such as aimer (to love), effrayer (to frighten), or admirer (to admire), are notoriously difficult to classify because of their bidirectional nature. They often participate in a causative alternation: the subject can be the stimulus (the cause of the emotion) or the experiencer (the person feeling the emotion). A robust computational lexicon must distinguish between these roles to accurately interpret the sentiment polarity of a sentence.

Structural Components of the Lexicon

A comprehensive lexicon for French emotion verbs relies on several key parameters:

• Valence and Arousal: Assigning numerical values based on the Circumplex Model of Affect.
• Syntactic Frames: Mapping the verbs to specific structures, such as [Sujet] effraie [Objet] versus [Sujet] s'effraie de [Complement].
• Selectional Preferences: Defining the semantic types of the subjects and objects (e.g., whether the stimulus is an event, an object, or an animate being).

Applications in NLP

The development of this lexicon serves multiple downstream tasks. In sentiment analysis, knowing the specific directionality of an emotion verb allows for the correct identification of the target of the sentiment. Furthermore, in dialogue systems, understanding these verbs enables agents to detect the emotional state of a user and respond with appropriate empathy, improving the human-computer interaction loop.

Challenges and Future Directions

One of the primary challenges remains the polysemy of French verbs. Many verbs of emotion possess neutral or physical meanings in specific contexts. Future work involves integrating large-scale transformer-based models with the static lexicon to create hybrid systems that benefit from both high-level contextual awareness and precise, structured linguistic knowledge.

As French is a language with rich morphology, the lexicon must also account for reflexive structures (the se- particle), which frequently signal the internalization of the emotion. Mapping these morphological variations to consistent semantic tags ensures that the computational model remains reliable across diverse texts, from literary corpora to social media discourse.