Fact · Speech recognition
75percent
In tests of word boundary marking by native Mandarin speakers, their agreement on where the word boundaries fall was about 75 percent, so the authors use character error rate as the usual metric for Mandarin speech recognition.
Checked against its source on October 17, 2026
In the source's own words
“agreement on the boundary locations occurs approximately 75% of the time”
Syllable language models for Mandarin speech recognition: Exploiting character language models, Xunying Liu et al., Journal of the Acoustical Society of America, 2013, Introduction.
Open the original- What it measures
- Mandarin Chinese is written with characters and no spaces, so a word must be inferred from context. The paper's authors cite this agreement level as the reason character error rate is the commonly used evaluation metric.
- What it does not show
- The agreement figure comes from tests of human word-boundary marking, not from recognizer output. It describes Mandarin in general, not any one dialect or dataset.
Source
Cited on
- Japanese, Chinese and Korean Speech to TextGuide
- Transcribe and Translate Audio in Other LanguagesGuide
- Do people speak faster in some languages?Question
- How do you transcribe Japanese, Chinese or Korean audio?Question
- Is speaking Chinese faster than typing pinyin?Question
- What is the difference between WER and character error rate?Question
- PinyinGlossary