Unverified50% confidenceFactExact time
TRACE splits the author's linguistic features into three granularities—author-level lexical habits, contact-specific phrasing, and temporal expression estimation—and feeds each into a frozen language model.
1
Sources
50%
Confidence
Long-term
Relevance
11/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAuthor-conditioned style transfer refers to a natural language processing task that conditions on a certain author's writing fingerprint (word preferences, syntactic complexity, punctuation habits, sense of rhythm, etc.) to guide a large model to imitate the target style while preserving semantic content.70% similarUnverifiedResearchers refer to the high-frequency stylistic patterns of large language models as LLM stylistic fingerprints, with different models having their own characteristic expressions.68% similarUnverifiedThe authors cancel out the confounding effects of fluency and text quality by reading the same sentence with two different models.67% similarUnverifiedThe authors used a byte-level neural network sandhi splitter and lemmatizer for preprocessing, then trained word vectors separately for each period.66% similarUnverifiedThe signals the author lists for splitting away from a single prompt include: the script is non-negotiable, the call does several things in sequence, the tool list is growing, you keep writing 'critical', and the stages are numbered.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1178606API
curl https://kongchang.com/api/v1/knowledge/claims/1178606MCP
get_claim(id=1178606)