Skip to content

Dataset · AmazonScience

MASSIVE: intents and slots in 51 languages

AmazonScience/massive

Over 1M phrases in 51 languages with intent and slot annotations for NLU in voice assistants.

What's inside

MASSIVE is a parallel dataset of utterances in 51 languages. The annotation includes 60 intents and 55 slot types for intent prediction and slot annotation tasks. Created by localizing SLURP, a set of single-turn interactions with a voice assistant. The card claims over 1 million utterances; metadata gives a range of 100K–1M.

What you can do with it

  • Train an NLU model for a voice assistant
  • Evaluate a multilingual intent detector
  • Fine-tune XLM-R or mBERT on slots and intents
Tasks
Text classification
Size
100K–1M rows
Downloads this month
19.5K
Likes
108

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("AmazonScience/massive")