Ev (not Eve) FedorenkoNov 28, 2022
Tiago Pimentel

The time we take to read a word depends on its predictability, i.e. its surprisal. However, we only know how surprising a word is after we see it. Our new paper investigates whether we anticipate words' surprisals to allocate reading times in advance :)

Joint work with Clara Meister, Ethan Wilcox, @roger_p_levy , @rdc
Paper: https://arxiv.org/abs/2211.14301
Code: https://github.com/rycolab/anticipation-on-reading-times

On the Effect of Anticipation on Reading Times

Over the past two decades, numerous studies have demonstrated how less predictable (i.e. higher surprisal) words take more time to read. In general, these previous studies implicitly assumed the reading process to be purely responsive: readers observe a new word and allocate time to read it as required. These results, however, are also compatible with a reading time that is anticipatory: readers could, e.g., allocate time to a future word based on their expectation about it. In this work, we examine the anticipatory nature of reading by looking at how people's predictions about upcoming material influence reading times. Specifically, we test anticipation by looking at the effects of surprisal and contextual entropy on four reading-time datasets: two self-paced and two eye-tracking. In three of four datasets tested, we find that the entropy predicts reading times as well as (or better than) the surprisal. We then hypothesise four cognitive mechanisms through which the contextual entropy could impact RTs -- three of which we design experiments to analyse. Overall, our results support a view of reading that is both anticipatory and responsive.

arXiv.org
Nov 28, 2022 at 11:23amWeb
Show threadTiago PimentelNov 28, 2022
First, we posit that a reader's expectation over the next word’s surprisal may affect its reading time (RT). This expectation has a clear operationalisation as the word's contextual entropy. When added besides the surprisal, the entropy helps predict RTs in ¾ analysed datasets.
Show threadTiago PimentelNov 28, 2022
Second, we posit that readers might predict a word’s surprisal in different ways, i.e., not necessarily minimising this predictions’ mean squared error. We generalise a family of skewed predictions using the Rényi contextual entropy.
Show threadTiago PimentelNov 28, 2022
A specific Rényi entropy (α=.5) is the most predictive of RTs in ¾ analysed datasets---this Rényi can be loosely interpreted as a soft version of the distributions' support. Perhaps surprisingly, this Renyi entropy is a *better* predictor than the surprisal in 2 datasets 😱
Show threadTiago PimentelNov 28, 2022
Third, we propose four mechanisms through which anticipation may impact reading: (i) word-skipping; (ii) budgeting (allocating RTs before reaching a word); (iii) preemptive processing; (iv) coping with uncertainty (additional processing loads when in high uncertainty contexts).
Show threadTiago PimentelNov 28, 2022

We then run several other experiments to analyse these mechanisms. We also discuss possible limitations due to approximating the surprisal and entropy with language models!

In short, our results strongly support the existence of anticipation effects on reading times! Check out our paper for more details :)