Modeling Linguistic Violence: An Ontology-Based Framework for the Computational Analysis of Violence Manifested in Language

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Social Sciences & Behavioral Studies · Depth: Expert, quick

Summary

A new framework addresses the conceptual ambiguity among terms like "hate speech," "toxic speech," and "dangerous speech," which currently hinder research and automated content moderation. This framework introduces a unified ontology that differentiates various forms of linguistic violence, including verbal aggression and cyberbullying, based on attributes such as target, intent, and rhetorical hallmarks. It also proposes a computational methodology to operationalize this ontology using a multi-stage NLP pipeline. This pipeline employs semantic analysis, specifically Semantic Role Labeling and Named Entity Recognition, to break down speech acts into core components like target and action. This component-based classification aims to robustly distinguish nuanced phenomena, such as general insults versus targeted identity-based attacks, and is particularly beneficial for low-resource languages like Portuguese due to its reliance on multilingual semantic models.

Key takeaway

For AI Engineers developing content moderation systems, this framework offers a robust approach to classifying linguistic violence. You should consider integrating an ontology-based, multi-stage NLP pipeline that leverages semantic role labeling and named entity recognition to move beyond lexical cues, especially when dealing with implicit harm or low-resource languages like Portuguese.

Key insights

An ontology and multi-stage NLP pipeline clarify and classify linguistic violence beyond lexical cues.

Principles

• Linguistic violence requires nuanced, attribute-based differentiation.
• Semantic analysis improves classification of implicit harm.

Method

The proposed method uses a multi-stage NLP pipeline with Semantic Role Labeling and Named Entity Recognition to deconstruct speech acts into components (target, action) for granular classification based on an established ontology.

In practice

• Apply semantic analysis to differentiate implicit harm.
• Utilize multilingual models for low-resource languages.

Topics

• Linguistic Violence
• Ontology-Based Framework
• Semantic Analysis
• NLP Pipeline
• Automated Moderation

Best for: Research Scientist, AI Engineer, AI Product Manager, AI Scientist, NLP Engineer, AI Ethicist

Related on AIssential

• Measuring and Mitigating Toxicity in Large Language Models: A Comprehensive Replication Study — Inference-time toxicity mitigation is effective for explicit toxicity but struggles with implicit hate speech and incurs significant latency.
• From Annotated Clinical Narratives to Ontology: Structuring Brazilian Portuguese Clinical Data — A new ontology structures Brazilian Portuguese clinical narratives, enhancing NLP interoperability and semantic expressiveness.
• Evaluating FrameNet-Based Semantic Modeling for Gender-Based Violence Detection in Clinical Records — FrameNet-based semantic annotation significantly improves GBV detection in clinical text over categorical data alone.
• The Register Gap: A Meaning Intelligence Framework for Nigerian Public Discourse — AI systems fail on Nigerian discourse due to context failure, not translation, requiring deeper meaning intelligence.
• Multi-Stage Training for Abusive Comment Detection in Indic Languages — Multi-stage training with language preprocessing and model ensembles improves abusive comment detection while minimizing false positives.
• Revisiting Semantic Role Labeling: Efficient Structured Inference with Dependency-Informed Analysis — A new SRL framework offers explicit predicate-argument structure and 10x faster inference with modern encoders.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.