Fusing Stylometric and Embedding Systems to Estimate Authorship Likelihood Ratios in Japanese

In three linesFirst application of likelihood ratio framework for authorship attribution in Japanese. Fusion of stylometric systems and embeddings from pre-trained language models on ~1000-character blog excerpts. Fused system improves discriminability (log-likelihood-ratio cost: 0.32484) while maintaining excellent calibration.

Summary generated by Claude — human-verified