The Low-Dimensional Linear Geometry of Contextualized Word Representations
The Low-Dimensional Linear Geometry of Contextualized Word Representations
复制标题
语境化单词表示的低维线性几何
DOI：
--
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Jacob Andreas
中科院分区：
文献类型：
--
作者：
Evan Hernandez;Jacob Andreas
黑盒探测模型可以从预先训练的单词表示中可靠地提取时态、数字和句法角色等语言特征。然而，这些特征在表征中编码的方式仍然知之甚少。本文系统地研究了埃尔莫和BERT中语境化词表征的线性几何。我们表明，各种语言特征（包括结构化的依赖关系）编码在低维子空间。然后，我们细化这个几何图片，显示有层次的子空间之间的关系编码一般的语言类别和更具体的，低维特征编码分布，而不是对齐到单个神经元。最后，我们证明了这些线性子空间与模型行为有因果关系，并且可以用于对BERT的输出分布进行细粒度操作。
Black-box probing models can reliably extract linguistic features like tense, number, and syntactic role from pretrained word representations. However, the manner in which these features are encoded in representations remains poorly understood. We present a systematic study of the linear geometry of contextualized word representations in ELMO and BERT. We show that a variety of linguistic features (including structured dependency relationships) are encoded in low-dimensional subspaces. We then refine this geometric picture, showing that there are hierarchical relations between the subspaces encoding general linguistic categories and more specific ones, and that low-dimensional feature encodings are distributed rather than aligned to individual neurons. Finally, we demonstrate that these linear subspaces are causally related to model behavior, and can be used to perform fine-grained manipulation of BERT’s output distribution.
DOI：
10.18653/v1/d17-1308
发表时间：
2017-09
期刊：
--
影响因子：
--
作者：
David Mimno;Laure Thompson
通讯作者：
David Mimno;Laure Thompson