Readers Are Parallel Processors

Abstract

Reading research has long endorsed the view that words are processed strictly one by one. The primary empirical test of this notion is the search for effects from upcoming words on readers' eye movements during sentence reading.Here we argue that no conclusions can be drawn from the absence of such effects, and that the serial versus parallel processing debate cannot be resolved without treading beyond the methodological scope of tracking eye movements. Recent considerations of how the brain organizes linguistic input have sparked key predictions in-and outside the realm of text reading, with ensuing research revealing phenomena that complicate the serial processing perspective. A case is made for parallelism, along with new methods to infer the cognitive architecture driving reading. The Serial Processing IdealIt is evident that when we read, we adopt a largely serial strategy: texts are processed from left to right and from the top downwards, with the eyes' fixation jumping from one word to the next. Should we cease to impose seriality on the linguistic plane that bombards our retina with so many words at once, we would squander the canonical order of words and so a key ingredient of linguistic communication. Moreover, simultaneous processing of multiple words might conceivably incite word-to-word interference; for instance, through confusion about which letters belong to which word. For these reasons we can safely claim that during text reading we should ideally be able to confine our attention (see Glossary) to single words. But to what extent can we live up to this ideal?In the course towards a full understanding of the reading system, the question of whether readers process multiple words simultaneously has retained much prominence. Precisely a decade ago Reichle, Liversedge, Pollatsek, and Rayner expressed in this journal the opinion that parallel word processing is implausible [1]. At the time, it had already become clear that serial and parallel processing frameworks were equally capable of accounting for eye movement behavior during text reading [2][3][4]. Therefore, Reichle et al.'s argument for serial processing was not necessarily data driven (although later empirical work was deemed to be in support of serial processing; see the next section), but rather one of parsimony: a serial processing system recognizes words in the intended order and does not mix-up information across words, given that only one word is attended at any time. By contrast, a parallel processing system might recognize words out of order, and 'if one were to simultaneously activate orthographic units for two words, this would produce noisy output corresponding to neither word' ([1], see p. 117).A decade later, we can but agree that a serial processing system would have the simplest time reading, yet recent research has yielded various reasons to believe that the reading process is in fact not so simple. In this Opinion article we show that the serial processing assumption holds only if the word recognition process is treated as a ...

Visual word recognition: Evidence for a serial bottleneck in lexical access

“…Interestingly, they sometimes report letters that were present in the other word. This type of "migration" error is sensitive to high-level lexical properties, and has been taken as evidence of parallel word recognition (McClelland & Mozer, 1986;Mozer, 1983;Snell & Grainger, 2019b). Again, however, the time between the onset of the words and the post-masks in those experiments was quite long, on the order of 200-500 ms.…”

“…The debate has recently extended beyond oculomotor measures during reading (Snell & Grainger, 2019b). For instance, several studies have shown that with relatively short displays (≤ 200 ms) word recognition performance is influenced by surrounding words and sentence context (Snell, Declerck, & Grainger, 2018a;Snell, Meeter, & Grainger, 2017).…”

Visual word recognition: evidence for a serial bottleneck in lexical access

“…Interestingly, they sometimes report letters that were present in the other word. This type of "migration" error is sensitive to high-level lexical properties, and has been taken as evidence of parallel word recognition (McClelland & Mozer, 1986;Mozer, 1983;Snell & Grainger, 2019). Again, however, the time between the onset of the words and the post-masks in those experiments was quite long, on the order of 200-500 ms.…”

“…The debate has recently extended beyond oculomotor measures during reading (Snell & Grainger, 2019). For instance, several studies have shown that with relatively short displays (<=200 ms) word recognition performance is influenced by surrounding words and sentence context (Snell, Declerck, & Grainger, 2018;Snell, Meeter, & Grainger, 2017).…”

How many words do we read per minute? A review and meta-analysis of reading rate

“…It has been argued that the letters of visual words up to a certain length can be processed in parallel, unlike the sequential unfolding of sounds in spoken language (Adelman, Marquis, & Sabatos-DeVito, 2010;Radeau, Morais, Mousty, Saerens, & Bertelson, 1992). There are also indications that more than one word can be processed simultaneously in reading (Snell & Grainger, 2019), and it has been argued that reading enables better prediction of upcoming words (Huettig & Pickering, 2019). All these factors help to understand why reading can be faster than listening.…”