Emergence of a High-Dimensional Abstraction Phase in Language Transformers

2024/05/24 by Emily Cheng, Cheng, Emily, Diego Doimo +11 · 12 citations
Social Sciences · Arts and Humanities · Computer Science · #Language and cultural evolution #Linguistics and Cultural Studies #Natural Language Processing Techniques

paper · pdf · doi:10.48550/arxiv.2405.15471

Abstract

A language model (LM) is a mapping from a linguistic context to an output token. However, much remains to be known about this mapping, including how its geometric properties relate to its function. We take a high-level geometric approach to its analysis, observing, across five pre-trained transformer-based LMs and three input datasets, a distinct phase characterized by high intrinsic dimensionality. During this phase, representations (1) correspond to the first full linguistic abstraction of the input; (2) are the first to viably transfer to downstream tasks; (3) predict each other across different LMs. Moreover, we find that an earlier onset of the phase strongly predicts better language modelling performance. In short, our results suggest that a central high-dimensionality phase underlies core linguistic processing in many common LM architectures.

Cited by

• Rethinking Intrinsic Dimension Estimation in Neural Representations
• The dimensionality of the Hopfield model
• From Topology to Retrieval: Decoding Embedding Spaces with Unified Signatures
• Geometry of Decision Making in Language Models
• Navigating through the hidden embedding space: steering LLMs to improve mental health assessment
• Controlling the Risk of Corrupted Contexts for Language Models via Early-Exiting
• Representation Potentials of Foundation Models for Multimodal Alignment: A Survey
• Probing Geometry of Next Token Prediction Using Cumulant Expansion of the Softmax Entropy
• Decoding Emotion in the Deep: A Systematic Study of How LLMs Represent, Retain, and Express Emotion
• Latent Thinking Optimization: Your Latent Reasoning Language Model Secretly Encodes Reward Signals in Its Latent Thoughts
• Towards Understanding the Shape of Representations in Protein Language Models
• The Generalization Ridge: Information Flow in Natural Language Generation

Related

• Bridging the Dimensional Chasm: Uncover Layer-wise Dimensional Reduction in Transformers through Token Correlation
• TRACE for Tracking the Emergence of Semantic Representations in Transformers
• Does Representation Matter? Exploring Intermediate Layers in Large Language Models
• Transformer Block Coupling and its Correlation with Generalization in LLMs
• Equivalent Linear Mappings of Large Language Models