DAHL: Domain Adaptation via Hierarchical Lexicons

The proposal is to implement a state of the are sub-word unit approach to NLP. This is likely to involve letters, byte pair encodings or explicit sub-word representations. The key here is that the approach is independent of fixed vocabularies. Given a vocabulary independent approach to NLP, we will implement the same hierarchy including phonetics for ASR. The result will be a fundamentally vocabulary independent ASR / NLP hybrid.
Idiap Research Institute
Industrial
Nov 01, 2019
Oct 31, 2021