A Geometric Notion of Causal Probing

arXiv:2307.15054v3 Announce Type: replace
Abstract: The linear subspace hypothesis (Bolukbasi et al., 2016) states that, in a language model's representation space, all information about a concept such as verbal number is encoded in a linear subspace. Prior work has relied on auxiliary classification tasks to identify and evaluate candidate subspaces that might give support for this hypothesis. We instead give a set of intrinsic criteria which characterize an ideal linear concept subspace and enable us to identify the subspace using …

abstract arxiv classification concept cs.cl hypothesis identify information language language model linear notion prior representation space support tasks type verbal work