Jack FitzGerald, Christopher Hench, Charith Peris, Scott Mackie, Kay Rottmann, Ana Sanchez, Aaron Nash, Liam Urbach, Vishesh Kakarala, Richa Singh, Swetha Ranganath, Laurie Crist, Misha Britan, Wouter Leeuwis, Gokhan Tur, Prem Natarajan
We present the MASSIVE dataset--Multilingual Amazon Slu resource package (SLURP) for Slot-filling, Intent classification, and Virtual assistant Evaluation. MASSIVE contains 1M realistic, parallel, labeled virtual assistant utterances spanning 51 languages, 18 domains, 60 intents, and 55 slots. MASSIVE was created by tasking professional translators to localize the English-only SLURP dataset into 50 typologically diverse languages from 29 genera. We also present modeling results on XLM-R and mT5, including exact match accuracy, intent classification accuracy, and slot-filling F1 score. We have released our dataset, modeling code, and models publicly.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Intent Classification | MASSIVE | Intent Accuracy | 86.1 | mT5 Base (encoder-only) |
| Intent Classification | MASSIVE | Intent Accuracy | 85.3 | mT5 Base (text-to-text) |
| Intent Classification | MASSIVE | Intent Accuracy | 85.1 | XLM-R Base |
| Slot Filling | MASSIVE | Slot F1 Score | 83.6 | XLM-R Base |
| Slot Filling | MASSIVE | Slot F1 Score | 82.2 | mT5 Base (encoder-only) |
| Slot Filling | MASSIVE | Slot F1 Score | 81.3 | mT5 Base (text-to-text) |
| Slot Filling | MASSIVE | Slot F1 Score | 64.2 | XLM-R Base |
| Slot Filling | MASSIVE | Slot F1 Score | 56.9 | mT5 Base (encoder-only) |
| Slot Filling | MASSIVE | Slot F1 Score | 50.6 | mT5 Base (text-to-text) |