Fact · Languages and accents
2.1percent out-of-vocabulary rate
In Shin and colleagues' 2013 Russian system, a sub-word language model cut the out-of-vocabulary rate from 4.8 percent to 2.1 percent.
Checked against its source on October 17, 2026
In the source's own words
“the use of the sub-word based language model reduces the vocabulary size from 120k to 40k and the OOV rate from 4.8% to 2.1%”
Maximum Entropy Language Modeling for Russian ASR, Evgeniy Shin et al., Proceedings of IWSLT 2013, 2013, Abstract.
Open the original- What it measures
- Out-of-vocabulary words are words the recognizer has never seen, so it cannot write them correctly. The smaller vocabulary is the trade-off.
- What it does not show
- A research system from 2013 on Russian broadcast-style test data. The figure shows the method works, not what modern tools achieve.