BLiMP: The Benchmark of Linguistic Minimal Pairs for English
It provides a tool for evaluating language models on grammatical phenomena, addressing a need for researchers in natural language processing, though it is incremental as it builds on existing benchmark practices.
The paper introduced BLiMP, a benchmark of 67 sub-datasets with 1000 minimal pairs each to evaluate language models' knowledge of English grammar, finding that state-of-the-art models reliably handle morphological contrasts but struggle with semantic restrictions and subtle syntactic phenomena.
We introduce The Benchmark of Linguistic Minimal Pairs (shortened to BLiMP), a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars, and aggregate human agreement with the labels is 96.4%. We use it to evaluate n-gram, LSTM, and Transformer (GPT-2 and Transformer-XL) LMs. We find that state-of-the-art models identify morphological contrasts reliably, but they struggle with semantic restrictions on the distribution of quantifiers and negative polarity items and subtle syntactic phenomena such as extraction islands.