Filtered-Corpus Training (FiCT)

Language models learn unlike coordination without direct exposure, study finds

Researchers have investigated whether language models can learn to understand unlike coordination, a linguistic structure where elements of different categories are joined, without direct exposure in their training data…