PDF] Low-Dimensional Structure in the Space of Language Representations is Reflected in Brain Responses

An encoder-decoder transfer learning method from computer vision is adapted to investigate the structure among 100 different feature spaces extracted from hidden representations of various networks trained on language tasks and reveals a low-dimensional structure where language models and translation models smoothly interpolate between word embeddings, syntactic and semantic tasks, and future word embedDings. How related are the representations learned by neural language models, translation models, and language tagging tasks? We answer this question by adapting an encoder-decoder transfer learning method from computer vision to investigate the structure among 100 different feature spaces extracted from hidden representations of various networks trained on language tasks. This method reveals a low-dimensional structure where language models and translation models smoothly interpolate between word embeddings, syntactic and semantic tasks, and future word embeddings. We call this low-dimensional structure a language representation embedding because it encodes the relationships between representations needed to process language for a variety of NLP tasks. We find that this representation embedding can predict how well each individual feature space maps to human brain responses to natural language stimuli recorded using fMRI. Additionally, we find that the principal dimension of this structure can be used to create a metric which highlights the brain's natural language processing hierarchy. This suggests that the embedding captures some part of the brain's natural language representation structure.

Leila Wehbe's Homepage

Integrating Machine Learning with Human Knowledge - ScienceDirect

PDF) Low-Dimensional Structure in the Space of Language Representations is Reflected in Brain Responses

PDF) The representation of space in the brain

The Brain Basis of Language Processing: From Structure to Function

The representation of space in the brain - ScienceDirect

Abstract representations emerge naturally in neural networks trained to perform multiple tasks

A Low Dimensional Description of Globally Coupled Heterogeneous Neural Networks of Excitatory and Inhibitory Neurons

Considerations for constructing a protein sequence database for metaproteomics - Computational and Structural Biotechnology Journal