Visual complexity in orthographic learning: Modeling learning across writing system variations Li-Yun Chang a , David C. Plaut b , and Charles A. Perfetti c a National Taiwan Normal University; b Carnegie Mellon University; c University of Pittsburgh ABSTRACT The visual complexity of orthographies varies across writing systems. Prior research has shown that complexity strongly influences the initial stage of reading development: the perceptual learning of grapheme forms. This study presents a computational simulation that examines the degree to which visual complexity leads to grapheme learning difficulty. We trained each of 131 identical neural networks to learn the structure of a different orthography and demonstrated a strong, positive association between net- work learning difficulty and multiple dimensions of grapheme complexity. We also tested the model’s performance against grapheme complexity effects on behavioral same/different judgments. Although the model was broadly consistent with human performance in how processing difficulty depended on the complexity of the tested orthography, as well as its relationship to viewers’ first-language orthography, discrepancies provided insight into important limitations of the model. We discuss how visual complexity can be a factor leading to reading difficulty across writing systems. Introduction The study of reading development across writing systems has focused primarily on the principles governing mapping between graphemes and various linguistic units such as phonemes, syllables, and morphemes. Grapheme encoding itself has received relatively less attention. Graphemes are the basic units that distinguish among a language’s written morphemes (e.g., single letters and letter combinations in alphabets/abjads, akshara in alphasyllabaries/syllabaries, and characters in morpho- syllabaries). Accurate, stable orthographic representations are required for associations to be reliably learned between visual forms and other aspects of language (Perfetti & Hart, 2002). Representations of the visual forms of graphemes are thus a critical beginning point of reading. Visual complexity influences the development of orthographic representations, thus contributing to difficulty in learning to read. Orthographies with visually complex graphemes are also likely to contain a larger grapheme inventory, making learning that much more difficult (e.g., Nag, 2011; Nag & Snowling, 2012; Nag, Treiman, & Snowling, 2010). Here, we use computational modeling as a tool to examine the relationship between the visual complexity of graphemes and learning difficulty across writing systems. The visual demands of grapheme processing, driven by the size of the grapheme inventory and the corresponding complexity of the graphemes, can pose a challenge to beginning learners. Empirical studies covering a wide range of orthographies have demonstrated that grapheme com- plexity is negatively correlated with grapheme identification efficiency (Liu, Chen, Liu, & Fu, 2012; CONTACT David C. Plaut plaut@cmu.edu Department of Psychology, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213-3890. Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/hssr. © 2015 Society for the Scientific Study of Reading SCIENTIFIC STUDIES OF READING http://dx.doi.org/10.1080/10888438.2015.1104688 Downloaded by [David Plaut] at 15:56 01 December 2015