Fact · Speech recognition
Many state-of-the-art inverse text normalization systems use hand-written weighted finite-state transducer grammars, because the task has extremely low tolerance to unrecoverable errors.
Checked against its source on October 17, 2026
In the source's own words
“Many state-of-the-art ITN systems use hand-written weighted finite-state transducer(WFST) grammars since this task has extremely low tolerance to unrecoverable errors.”
NeMo Inverse Text Normalization: From Development To Production, Yang Zhang et al., arXiv, 2021, Abstract.
Open the original (DOI 10.48550/arXiv.2104.05055)- What it measures
- The sentence explains why rule-based grammars remained common in ITN even as neural models spread through speech recognition.
- What it does not show
- Describes the state of the art in 2021. Neural ITN models have since been proposed, and the paper's own hybrid design adds a neural model to the grammar approach.