The Low-Dimensional Linear Geometry of Contextualized Word Representations
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.
Code (0)
등록된 구현이 없습니다.
Methods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
How Contextual are Contextualized Word Representations? Comparing the Geometry of BERT, ELMo, and GPT-2 Embeddings
Replacing static word embeddings with contextualized word representations has yielded significant improvements on many NLP tasks. However, just how contextual are the contextualized representations produced by models suc…
Word EmbeddingsCX-ST-RNM at SemEval-2019 Task 3: Fusion of Recurrent Neural Networks Based on Contextualized and Static Word Representations for Contextual Emotion Detection
In this paper, I describe a fusion model combining contextualized and static word representations for approaching the EmoContext task in the SemEval 2019 competition. The model is based on two Recurrent Neural Networks, …
Frequency-based Distortions in Contextualized Word Embeddings
How does word frequency in pre-training data affect the behavior of similarity metrics in contextualized BERT embeddings? Are there systematic ways in which some word relationships are exaggerated or understated? In this…
Semantic SimilaritySemantic Textual SimilarityWord EmbeddingsImproved Word Sense Disambiguation Using Pre-Trained Contextualized Word Representations
Contextualized word representations are able to give different representations for the same word in different contexts, and they have been shown to be effective in downstream natural language processing tasks, such as qu…
named-entity-recognitionNamed Entity RecognitionNamed Entity Recognition (NER)Question Answering+3Contrastive Learning for Low Resource Machine Translation
Representation learning plays a vital role in natural language processing tasks. More recent works study the geometry of the representation space for each layer of pre-trained language models. They find that the context…
Contrastive LearningData AugmentationLanguage ModelingLanguage Modelling+3