2021/04/13 by Yichu Zhou, Zhou, Yichu, Vivek Srikumar +1 · 7 citations
Computer Science · #Artificial intelligence #Classifier (UML) #Computation and Language (cs.CL) #Computer science #Embedding #FOS: Computer and information sciences #Heuristic #Machine learning #Natural Language Processing Techniques #Natural language processing #Representation (politics) #Space (punctuation) #Task (project management) #Text Readability and Simplification #Topic Modeling #cs.CL
paper · pdf · doi:10.48550/arxiv.2104.05904
published in arXiv (Cornell University) (Cornell University) · NAACL 2021
arxiv created 2021/04/13 · openalex publication_date 2021/04/13 · arxiv updated 2021/04/14 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05
Understanding how linguistic structures are encoded in contextualized embedding could help explain their impressive performance across NLP@. Existing approaches for probing them usually call for training classifiers and use the accuracy, mutual information, or complexity as a proxy for the representation's goodness. In this work, we argue that doing so can be unreliable because different representations may need different classifiers. We develop a heuristic, DirectProbe, that directly studies the geometry of a representation by building upon the notion of a version space for a task. Experiments with several linguistic tasks and contextualized embeddings show that, even without training classifiers, DirectProbe can shine light into how an embedding space represents labels, and also anticipate classifier performance for the representation.