This study proposes two novel normalisation techniques, Square Root Correction and Hapax Correction, to derive likelihood ratios from the authorship verification method LambdaG without requiring a separate calibration model. These methods are designed to mitigate the overestimation of evidential strength caused by long or highly repetitive texts.
- Evaluated against logistic regression calibration across fifteen corpora and text lengths ranging from 100 to 9,500 tokens using log-likelihood ratio cost (Cllr).
- The proposed methods achieve performance comparable to logistic regression calibration.
- The Hapax Correction outperforms logistic regression calibration in approximately 45% of tests weighted by corpora.
- Performance is more frequently close (within 5%) when the Hapax Correction is outperformed compared to the reverse comparison.
Eliminating the need for a calibration model reduces data requirements, time, and complexity, thereby increasing the accessibility and transparency of forensic text comparison.