Snips Built In Intents: 328 Utterances Across 10 NLU Classes

Snips Built In Intents consists of 328 English utterances across 10 intent classes, released by Snips in December 2016. This expert-generated dataset was created to benchmark the performance of natural language understanding systems and voice assistants.

Use Cases

• Benchmarking NLU models on intent classification tasks using the 10 provided intent classes
• Evaluating text classification accuracy for voice assistant utterances
• Testing few-shot learning performance on specific intent labels

Strengths

• Expert-generated utterances and labels
• Cited in Arxiv 1805.10190
• 10 distinct intent categories for classification

Limitations

• Small sample size of 328 records
• Temporal staleness as data originates from 2016
• Limited to 10 intent classes

Provenance

Source

Snips (Sonos NLU Benchmark)

Collection Method

Expert generated

Time Range

2016

Freshness

Last updated in 2024, but data originates from 2016.

Geography

United States