Identifying and interpreting non-aligned human conceptual representations using language modeling