Get Started
Research questionHow can we tell whether generated text matches an author's style across pragmatic contexts?Common lexical, embedding-based, and language-model-based metrics may disagree with human judgments and overlook how authorial style changes with pragmatic context. This makes it difficult to determine whether a personalized output genuinely resembles its intended author.
AI
Evaluation & Benchmarks
Natural Language Processing
Latest papersRecent research connected to this question, newest first.Evaluating Style-Personalized Text Generation: Challenges and DirectionsThe evidence concerns style-personalized text generation evaluated with BLEU, embedding-based metrics, and LLM-as-judge approaches. The study uses a style discrimination benchmark covering eight writing tasks and three settings: domain discrimination, authorship attribution, and distinguishing personalized from non-personalized LLM output; its findings about metric ensembles are limited to these evaluations.research paper · Sep 1, 2026
Related questions
How can few-shot authorship style transfer match an arbitrary author’s context-dependent style while preserving meaning?How can we reliably evaluate specialised style in LLM-generated Chinese legal text?How can we distinguish AI assistance from AI authorship when the prompting dialogue is unavailable?How can activation steering represent and control multidimensional authorship style without dedicated training?
Home
Topics
Search
Library