MASSIVE: A 1M-Example Multilingual Natural Language Understanding Dataset with 51 Typologically-Diverse Languages

TL;DR

The paper introduces MASSIVE, a large-scale multilingual dataset with 1 million virtual assistant utterances across 51 languages, enabling improved multilingual natural language understanding and evaluation.

Contribution

It provides a new, extensive multilingual dataset with diverse languages and domains, along with baseline modeling results and publicly released resources.

Findings

XLM-R and mT5 achieve competitive accuracy on the dataset.

The dataset covers 51 languages, 18 domains, 60 intents, and 55 slots.

Public release of dataset and models facilitates future research.

Abstract

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.