A new study led by Dr Andrea Nini at The University of Manchester has found that a grammar-based approach to language analysis can match or outperform advanced AI systems in identifying who wrote a text. The method, called LambdaG, uses patterns in grammar and sentence construction rather than large-scale AI models, offering comparable accuracy with greater transparency and lower computational cost.

Key findings

• A grammar-based authorship analysis method matched or exceeded leading AI systems across most test datasets
• The approach outperformed several neural network-based authorship verification models
• Researchers tested the method across 12 real-world writing datasets including emails, forums and reviews
• The system is more transparent than many AI models because it shows which grammatical patterns informed decisions
• Researchers say the findings challenge assumptions that more complex AI always produces better results

What did the study find?

Researchers found that a relatively simple, linguistically grounded method can perform as well as - and in some cases better than - complex artificial intelligence systems in identifying authorship.

The study suggests that increasingly sophisticated AI is not always necessary for high-performing writing analysis, particularly when methods are designed around established principles of how language works.

How does the LambdaG method work?

The method, called LambdaG, analyses patterns in grammar rather than relying on large-scale machine learning models.

It builds a statistical profile of how an individual writes by measuring features such as function word usage (words like it, of and the), sentence structure, punctuation patterns and other grammatical habits.

The researchers say these features create a distinctive behavioural signature for each writer.

Why is this different from AI-based authorship analysis?

Many current authorship verification systems rely on complex AI models trained on vast datasets. While effective, these systems can be difficult to interpret, computationally expensive and hard to explain in high-stakes settings such as legal investigations. By contrast, LambdaG provides a transparent explanation of which grammatical features influenced its conclusions.

How accurate was the method?

Researchers tested LambdaG across 12 datasets designed to reflect real-world writing scenarios, including emails, online forum posts and consumer reviews.

In most cases, the method achieved higher accuracy than several established authorship verification systems, including neural network-based approaches.

Why does grammar reveal authorship?

The researchers argue that grammar acts as a behavioural signature, like how we write our signature or how we walk.

Over time, individuals develop unconscious habits in how they structure sentences and use language. These habits create identifiable linguistic patterns that can distinguish one writer from another.

What are the potential applications?

The researchers say the method could support work in:

• Forensic linguistics
• Criminal investigations
• Online abuse detection
• Academic integrity monitoring