Verified80% confidenceFactExact time
Word Error Rate (WER) is calculated by dividing the sum of substitution, deletion, and insertion errors between the recognition output and the reference text by the total number of words in the reference text.
4
Sources
80%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWER 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.79% similarUnverifiedWER treats all word errors equally, applying the same penalty regardless of whether the error involves a function word or a key entity word.64% similarUnverifiedWER calculation is based on the edit distance (Levenshtein distance) algorithm, which uses dynamic programming to find the optimal alignment path between the reference text and the hypothesis text.61% similarUnverifiedFor a published 406M-parameter Fusion-in-Decoder specialized model, the ROUGE-1 score dropped sharply by 6.30 points when the input was switched from truncated long text to 2,000-word retrieved segments.61% similarUnverifiedCombining perplexity, burstiness, and word-frequency anomaly analysis significantly improves AI text detection accuracy compared to any single method.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1003004API
curl https://kongchang.com/api/v1/knowledge/claims/1003004MCP
get_claim(id=1003004)