Grammar-based method LambdaG matches or exceeds AI in authorship analysis, study finds
At The University of Manchester, researchers have created a grammar-focused authorship analysis technique known as LambdaG, which matches or surpasses sophisticated AI systems in determining the authors of texts. Under the leadership of Dr. Andrea Nini, the study evaluated LambdaG across 12 datasets, such as emails and consumer reviews, and achieved superior accuracy compared to traditional authorship verification tools. This method discerns linguistic patterns tied to individual writing styles by analyzing features like function word frequency and sentence composition. LambdaG's clarity reveals the grammatical patterns that influenced its conclusions. Its potential uses span forensic linguistics and monitoring academic integrity. The findings were published in Humanities and Social Sciences Communications, DOI: https://doi.org/10.1057/s41599-025-06340-3, disputing the notion that more complex AI guarantees improved outcomes.
Key facts
- LambdaG method uses grammar patterns rather than AI models for authorship analysis
- Method matched or outperformed AI systems across most test datasets
- Tested across 12 real-world writing datasets including emails and forum posts
- Provides transparent explanations of which grammatical features influenced conclusions
- Developed by researchers at The University of Manchester led by Dr Andrea Nini
- Published in Humanities and Social Sciences Communications journal
- Potential applications in forensic linguistics and criminal investigations
- Challenges assumption that complex AI is always necessary for high performance
Entities
Institutions
- The University of Manchester
- Humanities and Social Sciences Communications
Locations
- Manchester
- United Kingdom