[KongchangAI]
Benchmark

sWuggy

A zero-shot spoken language understanding evaluation task used to test the lexical-level linguistic knowledge acquired by speech models.

Timeline (last 90 days)

Nov 3

On the three tasks sWuggy, sBLIMP, and tSC, GSLM2 narrows the gap between speech models and text models by about two orders of magnitude.

Unverified50%
Nov 3

The sWuggy task examines lexical and phonological knowledge, requiring the model to judge which of a real pseudoword and a nonword with one phoneme changed is more like a real word.

Unverified50%

All Facts (2)

Source Articles