What do you learn from context? Probing for sentence structure in contextualized word representations

Abstract

Contextualized representation models such as ELMo (Peters et al., 2018a) and BERT (Devlin et al., 2018) have recently achieved state-of-the-art results on a diverse array of downstream NLP tasks. Building on recent token-level probing work, we introduce a novel edge probing task design and construct a broad suite of sub-sentence tasks derived from the traditional structured NLP pipeline. We probe word-level contextual representations from four recent models and investigate how they encode sentence structure across a range of syntactic, semantic, local, and long-range phenomena. We find that existing models trained on language modeling and translation produce strong representations for syntactic phenomena, but only offer comparably small improvements on semantic tasks over a non-contextual baseline.

ICLR 2019 camera-ready version, 17 pages including appendices

Cited by in corpus (34)

• Pre-trained Models for Natural Language Processing: A Survey
• SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems
• A Survey on Data Augmentation for Text Classification
• Revisiting Few-sample BERT Fine-tuning
• On the Explainability of Natural Language Processing Deep Models
• How Context Affects Language Models' Factual Predictions
• A Hierarchy of Limitations in Machine Learning
• exBERT: A Visual Analysis Tool to Explore Learned Representations in Transformers Models
• Differentiable Reasoning over a Virtual Knowledge Base
• Dynamic Prosody Generation for Speech Synthesis using Linguistics-Driven Acoustic Embedding Selection
• Extracting Sentence Embeddings from Pretrained Transformer Models
• Measuring Systematic Generalization in Neural Proof Generation with Transformers
• LMMS Reloaded: Transformer-based Sense Embeddings for Disambiguation and Beyond
• Probing BERT in Hyperbolic Spaces
• Finding Experts in Transformer Models
• Behind the Scene: Revealing the Secrets of Pre-trained Vision-and-Language Models
• The Life Cycle of Knowledge in Big Language Models: A Survey
• Why Do Masked Neural Language Models Still Need Common Sense Knowledge?
• Evaluation of taxonomic and neural embedding methods for calculating semantic similarity
• Grounding Representation Similarity with Statistical Testing
• Understanding Self-Attention of Self-Supervised Audio Transformers
• The Rediscovery Hypothesis: Language Models Need to Meet Linguistics
• AES Systems Are Both Overstable And Oversensitive: Explaining Why And Proposing Defenses
• A Pairwise Probe for Understanding BERT Fine-Tuning on Machine Reading Comprehension
• Attending to Entities for Better Text Understanding
• Textual Entailment for Effective Triple Validation in Object Prediction
• A Comparative Analysis of Knowledge-Intensive and Data-Intensive Semantic Parsers
• Self-conditioning pre-trained language models
• BERTnesia: Investigating the capture and forgetting of knowledge in BERT
• ISCAS at SemEval-2020 Task 5: Pre-trained Transformers for Counterfactual Statement Modeling
• Influence Patterns for Explaining Information Flow in BERT
• Does Chinese BERT Encode Word Structure?
• Duluth at SemEval-2020 Task 7: Using Surprise as a Key to Unlock Humorous Headlines
• Alleviating the Knowledge-Language Inconsistency: A Study for Deep Commonsense Knowledge