Regularities impact eye movements during book reading

A story about statistical learning in a story: Regularities impact eye movements during book reading

Joshua Snell, Jan Theeuwes

Highlights

•

How statistical regularities impact our eye movements during text reading is unknown.

•

Here we investigated the influence of regularities at the level of sentence structures.

•

Repeatedly encountering given combinations of word lengths or syntax eases reading.

•

Learning curves are steeper for more frequent structures.

•

Effects were observed in both English and Dutch reading.

Abstract

A wealth of research attests to the key role of statistical learning in the acquisition and execution of skilled reading. Little is known, however, about how regularities impact the way readers navigate through their linguistic environment. While previous studies have mostly gauged the recognition of single words, oculomotor processes are likely influenced by multiple words at once. With these premises in mind, we performed analyses on the GECO book reading corpus to determine whether repeatedly encountering a given sentence structure improves oculomotor control. In the reading materials we labeled structures on the basis of both low- and high-level properties: respectively word length combinations (e.g., a 2-letter word followed by a 6-letter word followed by a 4-letter word) and syntactic structures (e.g., an article followed by a noun followed by a verb). Our analyses show that repeatedly encountering a structure leads to fewer and shorter fixations, and fewer corrective saccades. Critically, learning curves are steeper for structures that have a higher overall frequency, hence evidencing true statistical learning over and above readers’ general tendency to accelerate as they progress through the book. Further, data from Dutch-English bilingual readers suggest that these types of learning occur across languages and at various levels of proficiency. We surmise that the reading system is tuned to statistical regularities pertaining not just to single words but also combinations of words. These regularities impact both linguistic processing and oculomotor control.

Keywords

Reading

Statistical learning

Oculomotor control

Syntactic processing

Book reading

Introduction

It is evident that any language, both oral and written, is rife with statistical regularities. Perhaps unsurprisingly, it has been suggested that language acquisition itself largely depends on one’s ability to detect these (e.g., Chater and Manning, 2006, Kuhl, 2004; see Siegelman, Bogaerts, Christiansen, & Frost, 2017, for an extensive review). Statistical learning bears prominence not only during language acquisition, but also in the way highly proficient language users process linguistic input. The most straightforward example is that regular words are recognized faster than irregular words (e.g., White, Warrington, McGowan, & Paterson, 2015). Additionally, predictable words—i.e., words that more regularly succeed a given word or clause—are recognized faster than unpredictable words (e.g., Ehrlich and Rayner, 1981, Rayner and Well, 1996, Kretzschmar et al., 2015); and learned associations between a given context and its constituent words may similarly facilitate recognition (McDonald & Shillcock, 2001).

While the above examples pertain to linguistic processing, language processing is not a purely linguistic matter. Our ability to read, for instance, depends not only on linguistic processing but also on attentional- and oculomotor control. One might reasonably claim that spoken language can be ‘passively’ processed; but this does not hold for written language, where the receiving party—the reader—has to pro-actively move the eyes from word to word to garner the message being conveyed. It is precisely this pro-active component of (written) language processing for which we do not yet know to what extent statistical learning plays a role. Simply put, although there are accounts of how statistical regularities influence word recognition, much less is known about how statistical regularities influence the way we navigate through our linguistic environment. This is the starting point of the present work.

This paper addresses the following question: Do statistical regularities affect oculomotor control during text reading? It should be noted that the developing reader might be explicitly instructed about certain regularities, such as which letters belong to which sounds (and when and why there are exceptions); but a myriad of other regularities are likely implicitly learned along the course of encountering large amounts of text. For example, readers will gradually learn on which syllables to put lexical stress, and that (English) words may often end with double letters (e.g. ‘boss will miss’) but will never begin with double letters (Arciuli & Simpson, 2011). Concerning our question of interest, readers are not explicitly instructed on how to move their eyes from word to word; but they nevertheless do learn to optimize word targeting strategies, evidenced by the fact that average fixation landing positions gradually shift towards the word center during the first years of learning to read (Ducrot, Pynte, Ghio, & Lété, 2013). Furthermore, saccade amplitudes are strongly influenced by word length (e.g., McConkie et al., 1988, Rayner, 1998).

Despite these strategies, saccades are error-prone even in the most experienced readers. Readers quite frequently execute a regressive saccade very quickly after the fixation onset, indicative of a correction after initially overshooting the target location (Radach et al., 1999, Rayner, 1998, Vitu and McConkie, 1998). Additionally, the distribution of initial landing positions (ILPs) is generally quite diffuse, and especially so in poor readers (e.g., Gagl, Hawelka, & Hutzler, 2014). In short, readers do not have perfect oculomotor control—and the issue investigated here is whether certain statistical regularities help improve this. Below, we will discuss, respectively, the potential impact of low-level visual regularities (i.e. combinations of word lengths) and high-level linguistic regularities (i.e., syntactic structures).

Download - full article pdf

Source : 

Department of Experimental and Applied Psychology, Vrije Universiteit Amsterdam, Netherlands

Institute of Brain and Behavior Amsterdam, Netherlands

Received 19 August 2019, Revised 28 April 2020, Accepted 29 April 2020, Available online 12 May 2020.