Muted: Multilingual Targeted Offensive Speech Identification and Visualization

2023-12-18 · Source: Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Muted is a system developed for Multilingual Targeted Offensive Speech Identification and Visualization. This tool is capable of leveraging any transformer-based HAP-classification model to accurately identify toxic spans within text, a significant feature being its ability to do so without requiring additional fine-tuning of the base model. Furthermore, Muted incorporates the spaCy library to precisely identify the specific targets and arguments for words that are predicted by attention heatmaps. This comprehensive approach enables a detailed analysis of offensive language, moving beyond simple detection to understand the context, direction, and linguistic components of harmful speech across multiple languages.

Key takeaway

For NLP Engineers developing offensive speech detection systems, Muted offers a compelling approach to identify targeted toxic content without the overhead of fine-tuning existing transformer models. You can utilize its ability to integrate with any HAP-classification model and use spaCy for granular target identification. This allows for more precise content moderation and deeper insights into harmful speech patterns, potentially streamlining development and deployment efforts.

Key insights

Muted identifies targeted offensive speech using transformer models and spaCy, requiring no fine-tuning for toxic span detection.

Principles

• Reuses existing HAP-classification models.
• Requires no model fine-tuning.
• Pinpoints specific speech targets.

Method

Muted employs transformer-based HAP-classification models to detect toxic spans. Subsequently, it uses spaCy and attention heatmaps to identify specific targets and arguments within the predicted offensive words.

In practice

• Analyze multilingual offensive content.
• Enhance content moderation systems.
• Visualize speech targets and arguments.

Topics

• Offensive Speech Detection
• Multilingual NLP
• Transformer Models
• spaCy
• Content Moderation
• Attention Heatmaps

Best for: Research Scientist, NLP Engineer, AI Engineer, AI Scientist

Related on AIssential

• A Multitask Transformer for Offensive Language Detection and Target Identification in HateBR — Multitask learning improves offensive language detection consistency but may reduce fine-grained target attribution.
• Intro to NLP with spaCy (1): Detecting programming languages — spaCy's linguistic features enable robust rule-based NLP systems beyond simple string matching.
• Holy $#!t: Are popular toxicity models simply profanity detectors? — Toxicity models often misclassify positive profanity due to inadequate training data and contextual understanding.
• 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.
• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Explain the Flag: Contextualizing Hate Speech Beyond Censorship — A hybrid LLM and vocabulary approach improves hate speech detection and explanation across multiple languages.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai.