Visualizing and Measuring the Geometry of BERT

Abstract

Transformer architectures show significant promise for natural language processing. Given that a single pretrained model can be fine-tuned to perform well on many different tasks, these networks appear to extract generally useful linguistic features. A natural question is how such networks represent this information internally. This paper describes qualitative and quantitative investigations of one particularly effective model, BERT. At a high level, linguistic features seem to be represented in separate semantic and syntactic subspaces. We find evidence of a fine-grained geometric representation of word senses. We also present empirical descriptions of syntactic representations in both attention matrices and individual word embeddings, as well as a mathematical argument to explain the geometry of these representations.

arXiv comment8 pages, 5 figures

References in corpus (6)

• Scikit-learn: Machine Learning in Python
• Distributed Representations of Words and Phrases and their Compositionality
• UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
• Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)
• What you can cram into a single vector: Probing sentence embeddings for linguistic properties
• Visualizing Attention in Transformer-Based Language Representation Models

Cited by in corpus (17)

• TabTransformer: Tabular Data Modeling Using Contextual Embeddings
• COCO-LM: Correcting and Contrasting Text Sequences for Language Model Pretraining
• Machine Learning for Scent: Learning Generalizable Perceptual Representations of Small Molecules
• Enhancing the Transformer with Explicit Relational Encoding for Math Problem Solving
• Capturing Evolution in Word Usage: Just Add More Clusters?
• A Systematic Analysis of Morphological Content in BERT Models for Multiple Languages
• CommonsenseVIS: Visualizing and Understanding Commonsense Reasoning Capabilities of Natural Language Models
• HUBERT Untangles BERT to Improve Transfer across NLP Tasks
• IsoScore: Measuring the Uniformity of Embedding Space Utilization
• Integrating Multimodal Information in Large Pretrained Transformers
• Combating the Curse of Multilinguality in Cross-Lingual WSD by Aligning Sparse Contextualized Word Representations
• The Rediscovery Hypothesis: Language Models Need to Meet Linguistics
• Exploring Linguistic Properties of Monolingual BERTs with Typological Classification among Languages
• How Far Does BERT Look At:Distance-based Clustering and Analysis of BERT${}^{\prime }$s Attention
• Positional Artefacts Propagate Through Masked Language Model Embeddings
• Unsupervised Word Polysemy Quantification with Multiresolution Grids of Contextual Embeddings
• Circles are like Ellipses, or Ellipses are like Circles? Measuring the Degree of Asymmetry of Static and Contextual Embeddings and the Implications to Representation Learning