Towards a Similarity-adjusted Surprisal Theory

TL;DR

This paper introduces similarity-adjusted surprisal, a new measure that incorporates word similarity into predictability models, improving the prediction of reading times and understanding of language comprehension.

Contribution

It extends surprisal theory by integrating word similarity through information value, providing a more nuanced measure of cognitive effort during language processing.

Findings

Similarity-adjusted surprisal improves prediction of reading times.

The measure reduces to standard surprisal when words are distinct.

Experimental results show added predictive power over traditional surprisal.

Abstract

Surprisal theory posits that the cognitive effort required to comprehend a word is determined by its contextual predictability, quantified as surprisal. Traditionally, surprisal theory treats words as distinct entities, overlooking any potential similarity between them. Giulianelli et al. (2023) address this limitation by introducing information value, a measure of predictability designed to account for similarities between communicative units. Our work leverages Ricotta and Szeidl's (2006) diversity index to extend surprisal into a metric that we term similarity-adjusted surprisal, exposing a mathematical relationship between surprisal and information value. Similarity-adjusted surprisal aligns with information value when considering graded similarities and reduces to standard surprisal when words are treated as distinct. Experimental results with reading time data indicate that similarity-adjusted surprisal adds predictive power beyond standard surprisal for certain datasets, suggesting it serves as a complementary measure of comprehension effort.