A Transformer and Prototype-based Interpretable Model for Contextual Sarcasm Detection

2025-03-14

AI Summary

This work addresses the challenge of detecting contextual irony, where literal sentiment contradicts the speaker’s true intent. We propose an end-to-end interpretable fusion model that jointly leverages a Transformer-based language model and prototypical networks. To explicitly model sentiment consistency, we introduce sentiment embeddings and design an inconsistency loss grounded in sentiment prototypes—enabling direct generation of natural-language explanations without requiring post-hoc interpretability techniques. The architecture thus achieves intrinsic interpretability alongside strong discriminative performance. Evaluated on three benchmark public datasets, our model surpasses state-of-the-art methods in detection accuracy. Moreover, it produces high-fidelity, instance-specific explanations, significantly enhancing both the credibility and human comprehensibility of irony identification.

Technology Category

• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Computer Vision: Interpretability, Explainability, and Transparency
• Philosophy and Ethics of AI: Accountability, Interpretability & Explainability

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Sarcasm detection, with its figurative nature, poses unique challenges for affective systems designed to perform sentiment analysis. While these systems typically perform well at identifying direct expressions of emotion, they struggle with sarcasm's inherent contradiction between literal and intended sentiment. Since transformer-based language models (LMs) are known for their efficient ability to capture contextual meanings, we propose a method that leverages LMs and prototype-based networks, enhanced by sentiment embeddings to conduct interpretable sarcasm detection. Our approach is intrinsically interpretable without extra post-hoc interpretability techniques. We test our model on three public benchmark datasets and show that our model outperforms the current state-of-the-art. At the same time, the prototypical layer enhances the model's inherent interpretability by generating explanations through similar examples in the reference time. Furthermore, we demonstrate the effectiveness of incongruity loss in the ablation study, which we construct using sentiment prototypes.

Problem

• Detecting sarcasm in text using contextual models
• Improving interpretability in sarcasm detection systems
• Enhancing sentiment analysis with prototype-based networks

Innovation

• Transformer-based language models for contextual sarcasm detection
• Prototype-based networks enhance model interpretability
• Sentiment embeddings and incongruity loss improve performance