Spoken Language Understanding on Internal Traffic Dataset (regular test)
9.3ICER Improvement (%)transformer-based E2E SLU
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| transformer-based E2E SLUPre-training strategy=BERT, AC Pretraining, Num. Params.=115M2021.06 | 9.3 | 37.3 | 42.8 | |
| transformer-based E2E SLUPre-training strategy=No Pretraining, Num. Params.=115M2021.06 | 9.1 | 37.1 | 40.5 | |
| Multitask Transformer BaselineNum. Params.=101M2021.06 | 4 | -2.7 | -0.1 | |
| Multitask BiLSTM BaselineNum. Params.=67.3M2021.06 | 0.5 | 0.5 | -1.1 | |
| Multitask LSTM BaselineNum. Params.=45.9M2021.06 | 0 | 0 | 0 |