Hidden Holes: topological aspects of language models

TL;DR

This paper investigates the topological structure of language model representations using algebraic topology, revealing differences between transformer and recurrent models and introducing a new measure called perforation.

Contribution

It introduces novel topological tools and a measure called perforation to analyze language model representations, highlighting differences between model architectures.

Findings

Transformers exhibit less topological complexity than recurrent models.

Topological patterns are consistent across natural languages but not synthetic data.

The study provides new mathematical insights into neural language model representations.

Abstract

We explore the topology of representation manifolds arising in autoregressive neural language models trained on raw text data. In order to study their properties, we introduce tools from computational algebraic topology, which we use as a basis for a measure of topological complexity, that we call perforation. Using this measure, we study the evolution of topological structure in GPT based large language models across depth and time during training. We then compare these to gated recurrent models, and show that the latter exhibit more topological complexity, with a distinct pattern of changes common to all natural languages but absent from synthetically generated data. The paper presents a detailed analysis of the representation manifolds derived by these models based on studying the shapes of vector clouds induced by them as they are conditioned on sentences from corpora of natural language text. The methods developed in this paper are novel in the field and based on mathematical apparatus that might be unfamiliar to the target audience. To help with that we introduce the minimum necessary theory, and provide additional visualizations in the appendices. The main contribution of the paper is a striking observation about the topological structure of the transformer as compared to LSTM based neural architectures. It suggests that further research into mathematical properties of these neural networks is necessary to understand the operation of large transformer language models. We hope this work inspires further explorations in this direction within the NLP community.