Idea
A multilingual claim normalization platform converting informal social media posts into clear statements for fact-checkers and media organizations
Research Paper
Core Innovation
This paper introduces a hybrid approach combining fine-tuned Small Language Models for languages with training data and Large Language Model prompting for zero-shot languages. This method enables effective claim normalization across twenty languages, outperforming prior single-model or monolingual approaches. The system's adaptability to zero-shot languages is a key advance for multilingual fact-checking.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: growing demand for automated fact-checking and multilingual content moderation globally.
Potential Customers & Pain Points
- Fact-Checking Organizations Needing Automated Claim Verification
- Social Media Platforms Seeking Content Moderation Tools
- News Agencies Requiring Accurate Information Summarization
- AI Developers Building Multilingual NLP Solutions
- Governments Monitoring Misinformation
Business Model
SaaS platform offering API access and custom integrations for media, social platforms, and fact-checking services with tiered pricing based on usage and language support.
Competitive Landscape
- ClaimBuster
- Full Fact
- Factmata
Implementation Challenges
- Data scarcity for low-resource languages
- Integration complexity with existing fact-checking workflows
- Dependence on LLM access and costs
Validation Strategy
- Pilot deployment with fact-checking organizations in multiple languages
- User feedback collection to refine normalization accuracy
- Benchmarking against existing claim normalization tools
Research Paper Overview
AKCIT-FN at CheckThat! 2025: Switching Fine-Tuned SLMs and LLM Prompting for Multilingual Claim Normalization
Summary
This paper presents a system for claim normalization, converting informal social media posts into concise, self-contained statements to aid automated fact-checking. The approach uses fine-tuned Small Language Models for supervised languages and Large Language Model prompting for zero-shot languages across twenty languages. The system achieved top-three rankings in fifteen languages, demonstrating strong performance especially in zero-shot scenarios. All code and configurations are publicly available.