Glitter: Visualizing Lexical Surprisal for Readability in Administrative Texts

TL;DR

Problem addressed: readability of administrative texts can impede access to justice. Approach: Glitter uses next-token probabilities to compute lexical surprisal $S(w_t|context) = -\\log p(w_t|w_{<t})$ and entropy $H(X) = -\\sum_i p_i \\log p_i$, visualizing results to aid authors. Key contributions: an end-to-end pipeline for per-token surprisal estimation in bureaucratic prose, a GLITTER-inspired visualization, and open-source tooling supporting multiple architectures. Significance: supports clearer writing for non-expert readers and provides a foundation for systematic readability evaluation across large legal corpora.

Abstract

This work investigates how measuring information entropy of text can be used to estimate its readability. We propose a visualization framework that can be used to approximate information entropy of text using multiple language models and visualize the result. The end goal is to use this method to estimate and improve readability and clarity of administrative or bureaucratic texts. Our toolset is available as a libre software on https://github.com/ufal/Glitter.

Figures (2)

• Figure 1: Glitter web user interface in light mode
• Figure 2: Glittered administrative text