ANCHOR: An English-to-Chinese Transcreation Method for Transferring Check-Worthiness Estimation Data

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Translating English annotated data into Chinese can supply resources for check-worthiness estimation, but entity omission, shifts in claim scope and added information during translation may make the source label inapplicable to the translated text. We propose ANCHOR (Adaptive Naturalization for Check-worthiness-Oriented Rewriting), a transcreation method for transferring English check-worthiness data into Chinese. ANCHOR first classifies the cultural dependence of each text. Highly culture-dependent texts follow a domestication path that adapts expressions while preserving factual scope, assertion strength and pragmatic function; the remaining texts follow a foreignization path built on retrieved bilingual demonstrations, candidate generation and self-supervised reranking. On three English check-worthiness datasets, NLP4IF, ClaimBuster and ViClaim, we evaluate general translation quality and check-worthiness preservation separately. Against the majority judgment of six human evaluators, cultural dependence classification reaches an accuracy of 0.8225 on 400 texts, and seven evaluators rated 100 domesticated translations at a mean of 4.386 out of 5 for meaning accuracy, cultural relevance and fluency. On 320 stratified samples, three evaluators judged whether each translation still supports checking the source claim: ANCHOR preserves the source label in 99.38% of cases, tied with DeepL for the highest among eight translation models, and preserves every negative sample. In an ablation against direct translation by GPT-5.2, ANCHOR raises this rate by 1.88 percentage points and makes fewer errors in names, quantities and entities. These results show that ANCHOR offers a practical way to reuse English annotated resources for building Chinese check-worthiness data. Content note. Some examples presented in this paper contain profanity and offensive language. These expressions are retained where necessary to preserve the linguistic characteristics of the original data.