Unverified50% confidenceFactExact time
Normalized WER means that uniform text normalization (such as number transcription, punctuation removal, and case unification) is applied to both the reference text and the system output before scoring, while unnormalized WER preserves differences in the original format.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBuilding a standardized prompt template library (such as having a validated template each for text summarization, code review, translation polishing, etc.), where only the variable parts need to be replaced when used, saves time and ensures consistent output quality.70% similarUnverifiedWER is calculated as (number of substituted words + number of deleted words + number of inserted words) / total number of reference words, and is extremely sensitive to text formatting (case, punctuation, abbreviations, spelling variants), so normalization is usually required before evaluation.70% similarUnverifiedNormalizing documents to Markdown preserves text structure, but the conversion process may lose layout information or complex table structures.69% similarUnverifiedByte-level preservation is a key technical feature in the field of document editing, meaning the editor only rewrites the parts actually changed, while the rest of the content remains unchanged at the binary level.69% similarUnverifiedDeterministic format normalization is a technical approach that handles format issues through string repair, avoiding LLM retries, and applies only to layout-drift-type format issues, not semantic errors.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1153262API
curl https://kongchang.com/api/v1/knowledge/claims/1153262MCP
get_claim(id=1153262)