Does it still sound like her? Measuring authorial voice in fine-tuned models
Fluency is easy to score; resemblance to one writer is not. We describe a blind evaluation where authors pick their own model's output against generic baselines. [ABSTRACT TO CONFIRM]