Visually Analyzing Contextualized Embeddings
This provides a tool for researchers and practitioners in NLP to better understand language models, though it is incremental as it builds on existing probing techniques.
The paper tackles the problem of analyzing what contextualized embeddings from language models learn by introducing an unsupervised visual exploration method based on clustering and textual structure, which user feedback indicates helps discover various linguistic structures.
In this paper we introduce a method for visually analyzing contextualized embeddings produced by deep neural network-based language models. Our approach is inspired by linguistic probes for natural language processing, where tasks are designed to probe language models for linguistic structure, such as parts-of-speech and named entities. These approaches are largely confirmatory, however, only enabling a user to test for information known a priori. In this work, we eschew supervised probing tasks, and advocate for unsupervised probes, coupled with visual exploration techniques, to assess what is learned by language models. Specifically, we cluster contextualized embeddings produced from a large text corpus, and introduce a visualization design based on this clustering and textual structure - cluster co-occurrences, cluster spans, and cluster-word membership - to help elicit the functionality of, and relationship between, individual clusters. User feedback highlights the benefits of our design in discovering different types of linguistic structures.