Visual crowding refers to our failure to identify an object when it is surrounded by other objects (Bouma, 1970; Martelli et al., 2009; Whitney & Levi, 2011). This phenomenon impacts basic features such as orientation, color, and spatial frequency (Greenwood & Parsons, 2020; Kewan-Khalayly and Yashar, 2022; Shechter & Yashar, 2021; Yashar et al., 2019), and extends to more higher-level objects such as faces and letters (Louie et al., 2007). Crowding is most pronounced at the visual periphery, e.g., at the parafovea (5°–8°), which plays an important role in preprocessing upcoming words during reading (Schotter et al., 2012). Thus, it sets significant constraints on the visual-orthographic processes involved in reading (Grainger et al., 2016). Crowding slows down reading speed (Pelli et al., 2007) and has been linked to reading difficulties (Joo et al., 2018) and developmental dyslexia (e.g., Bertoni et al., 2019; but see Doron et al., 2015). Thus, understanding the factors that mitigate the detrimental effect of crowding has both theoretical and practical implications.

Over the years, various factors have been shown to reduce crowding interference. These factors include mental capacities, such as visual attention (Herzog et al., 2015; Manassi et al., 2012) and perceptual learning (Hussain et al., 2012; Yashar et al., 2015), as well as stimulus-related elements like grouping (Herzog et al., 2015; Manassi et al., 2012) and object configuration (Jimenez et al., 2022). However, their practical use is limited, as they are not readily available or easily manipulated in a natural environment. Therefore, a deeper understanding of environmental regularities, also known as statistical learning (Frost et al., 2019), may be key in our ability to adapt to the natural environment. Yet, whether environmental statistics and context can help mitigate crowding interference remains unknown. We address this issue in the context of letter recognition.

Regarding higher-order contextual factors in reading, there is well-replicated evidence, at least in alphabetic scripts, that gazed-centered letter recognition is better when it appears within the context of a (familiar) word compared to isolation (word superiority effect) or a pseudoword (lexicality effect). These phenomena have been explained by the orthographic context of the word (e.g., McClelland & Rumelhart, 1981; Reicher, 1969). That is, the orthographic context (i.e., a familiar word) supplies more information compared to the no-context condition (i.e., isolated letter), hence makes the letter within a word more predictable and more resistant to interference than a letter presented alone (Reicher, 1969). McClelland and Rumelhart (1981) explained this finding by their interactive-activation model which postulates that visual word perception incorporates simultaneously processing of “bottom-up” and “top-down” input. Thus, in the case of a familiar letter string, facilitatory activation will be generated from the feature level to the letter level and from the letter to the word level, concurrently with activation from the word level to the letter level.

An alternative to the parallel activation model has been proposed by Pelli et al. (2003), who postulate that recognition involves feedforward connections in a hierarchical and sequential process. This begins with feature perception, progresses to letter identification, and culminates in interpreting combinations of letters. According to this model, words are unreadable unless each of their letters is separately identifiable. Despite their differences, both models emphasize the importance of orthographic processing, which involves the extraction of the identity and position of letters within a string (Grainger et al., 2016).

The involvement of orthographic processes in parafoveal processing in reading (e.g., Bouma, 1971, 1973; also see Schotter et al., 2012) is supported by studies that demonstrate a reduction in parafoveal crowding interference on letter recognition when the target letter is embedded in a real word compared to a pseudoword, the well-known lexicality effect (Bouma & Legein, 1977; Martelli et al., 2005). For instance, Martelli et al. (2005) examined word recognition, specifically focusing on familiarity and crowding effects, in three conditions: isolated letters, letters embedded within a three-letter string consisting of a real word (e.g., ace) or a pseudoword (e.g., aca). Along with replicating the word superiority effect in central vision, they found an opposite pattern in the visual periphery, with surrounding letters causing hindrance in performance due to crowding. Nonetheless, performance improved when letters were embedded in a real word as opposed to pseudowords – thus confirming the “lexicality” effect at both the fovea and parafovea (Martelli et al., 2005).

However, lexicality is only part of the story as words and pseudowords not only differ in their lexical properties, but also sub-lexically in terms of the probabilities of the word-internal sequence of letters (n-grams). In a given orthography, the probability of a specific letter can be determined by the preceding and following letters within the text. For example, in English, the letter c frequently precedes the word-final letter e (e.g., ace, once, ice, etc.), but c rarely precedes a word-final a. Consequently, certain bigrams (strings of two letters) and trigrams (strings of three letters) have higher probabilities than others (independently of their lexical status), with their frequency in the orthography dictating this probability. In reading research, investigators have explored the role of transitional probability by examining the impact of statistical learning on reading skills and deficits (see Frost et al., 2019). However, no studies have yet directly investigated the influence of transitional probability on letter recognition in print. Here, we address this issue by investigating the independent effects of lexicality (words versus pseudowords) and bigram and trigram frequencies on letter recognition at the parafovea in a crowded display of letters.

Previous studies have suggested a domain-specific neural mechanism for reading, located in the left hemisphere (Dehaene, 2005; Ossowski & Behrmann, 2015). According to the neuronal recycling hypothesis, this lateralization to the left hemisphere explains the findings of higher accuracy in the right hemifield in reading tasks in multiple scripts (e.g., White et al., 2020), including Hebrew (Ibrahim & Eviatar, 2009). Here, we further examined the assumed left hemisphere advantage by focusing on letter recognition under varied crowded conditions.

Our study also addresses concerns that research based on English and Western European alphabets may not necessarily allow generalizations regarding universal reading and language processing phenomena (e.g., Huettig & Ferreira, 2022; Share, 2008, 2021). Here, we conducted our investigation in Hebrew, a non-European language written in a non-alphabetic right-to-left script. We predict two language universal effects. First, we expect a right hemifield advantage due to left hemisphere language and reading circuits. Second, we anticipate a lexicality effect similar to that found in English. Crucially, if the lexicality effect in Hebrew can be largely explained by sublexical transitional probabilities, we predict that bigram frequencies will account for a significant portion of performance variation. Furthermore, language specific factors, mainly the right-to-left reading direction of Hebrew that extends the reading span more to the left (Pollatsek et al., 1981), may be reflected by any difference between the effect of the right bigram frequency and the left bigram frequency.