← Back to research

Generated-Text Detection and Quality Evaluation

We use spectral, statistical, and psycholinguistic signals to identify machine-generated text and evaluate generation quality.

Machine-generated text contains signals at statistical, spectral, and language-use levels. This project studies how stable those signals remain across models, tasks, and sampling methods.

Detection and Evaluation

Beyond determining whether text is machine-generated, we investigate the relationship between automatic metrics and human judgments of quality.

Methods

Our current methods combine frequency-domain analysis, statistical modeling, and psycholinguistic features, with an emphasis on robustness across domains.