Frost suggests that transposed-letter priming is obtained in some languages and not others because of the variable rigidity of the writing systems, which evolved to optimally transmit phonological and semantic information, given pre-established language-specific phonological spaces. We comment here on differences in the time course of phonological and semantic processing between languages within the Indo-European, alphabetic cluster, German and English, as supported by early indicators of processing obtained from eye-tracking measures. Additionally, we consider logographic Chinese. In particular, we focus on parafoveal preview benefit (PB) obtained by using the gaze-contingent boundary technique (Rayner Reference Rayner1975), in which a preview at the parafoveal target-word location is visible only until the gaze crosses an invisible boundary before the target; participating cognitive processes can be inferred by measuring target processing as a function of the relationship between preview and target.

In English, phonologically and orthographically related previews have repeatedly been shown to produce PB (Balota et al. Reference Balota, Pollatsek and Rayner1985; Inhoff Reference Inhoff1989; Rayner et al. Reference Rayner, McConkie and Ehrlich1978; Reference Rayner, McConkie and Zola1980; Reference Rayner, Well, Pollatsek and Bertera1982), whereas no evidence for semantic PB has been found in either sentence reading (Altarriba et al. Reference Altarriba, Kambe, Pollatsek and Rayner2001; Rayner et al. Reference Rayner, Balota and Pollatsek1986) or naming (Rayner et al. Reference Rayner, McConkie and Zola1980)–despite significant semantic priming in a standard (foveal) priming task (Rayner et al. Reference Rayner, Balota and Pollatsek1986). This has led Rayner et al. (Reference Rayner, White, Kambe, Miller, Liversedge, Hyönä, Radach and Deubel2003) to conclude that “the basis for the robust parafoveal preview benefit obtained in numerous studies is not any type of semantic code” (pp. 229–30). In contrast, in German, we have repeatedly obtained semantic PB by using a parafoveal fast-priming technique (Hohenstein et al. Reference Hohenstein, Laubrock and Kliegl2010), as well as the more traditional boundary technique (Hohenstein & Kliegl Reference Hohenstein, Kliegl, Vitu, Castet and Goffart2011).

What could be the reason for this striking difference between two related languages? Further considering that PB can be obtained from word n+2 in German (Kliegl et al. Reference Kliegl, Risse and Laubrock2007), we believe that extraction of parafoveal information is easier in German, mainly because of orthographic consistency: In German, the correspondence between graphemes and phonemes is relatively high, whereas the relation is rather opaque in English because of its very complex phonological space. Orthographic consistency likely affects both the strategies and the mechanisms of word processing.

There is good evidence that phonological codes are routinely co-activated during reading and can influence semantic processing (McCutchen & Perfetti Reference McCutchen and Perfetti1982; van Orden Reference Van Orden1987; but see Rastle & Brysbaert Reference Rastle and Brysbaert2006). This does not necessarily imply a strong phonological account (Frost Reference Frost1998) in which semantic access is through phonology; alternatives are either feedback from phonology to semantics as in the dual-route cascaded (DRC) model (Coltheart et al. Reference Coltheart, Rastle, Perry, Langdon and Ziegler2001), or a triangular model (Seidenberg & McClelland Reference Seidenberg and McClelland1989) in which both orthography and phonology activate semantics. Critically, however, the pattern of empirical PB effects suggests a language-differential distribution of limited resources between phonological and semantic activation. Given that orthography codes for both semantic and phonological information, in a language with opaque spelling-to-sound relations like English, more cognitive resources are occupied by phonological decoding, and hence are unavailable for decoding of semantic meaning.

This reasoning is supported by developmental differences: Children learning to read languages with low transparency have to acquire a much larger set of ambiguous orthographic-phonological relations, resulting in slower learning (Goswami et al. Reference Goswami, Ziegler, Dalton and Schneider2001; Reference Goswami, Ziegler, Dalton and Schneider2003; Landerl et al. Reference Landerl, Wimmer and Frith1997; Wimmer & Goswami Reference Wimmer and Goswami1994). In a large study comparing 13 European orthographies, the rate of development of basic decoding skills was slowest in English (Seymour et al. Reference Seymour, Aro and Erskine2003). Using pseudo-homophone priming, Goswami et al. (Reference Goswami, Ziegler, Dalton and Schneider2001) demonstrated that for German but not English children the activation of phonological information is relatively automatic. Landerl et al. (Reference Landerl, Wimmer and Frith1997) concluded that consistency of grapheme–phoneme relations affects the working-memory demands of recoding.

Results from other alphabetic languages with more transparent orthographies are not currently decisive. In very transparent Spanish, despite many reports of parafoveal semantic priming (Fuentes & Tudela Reference Fuentes and Tudela1992; Fuentes et al. Reference Fuentes, Carmona, Agis and Catena1994; Ortells et al. Reference Ortells, Abad, Noguera and Lupiáñez2001), a study using the boundary paradigm failed to find semantic PB (Altarriba et al. Reference Altarriba, Kambe, Pollatsek and Rayner2001). However, that study used bilingual readers of Spanish and English and always employed cross-language priming; thus, an opaque word was either preview or target. We are unaware of a direct study of semantic PB in highly transparent Finnish. Semantically unassociated emotional previews did not produce preview benefit or cost (Hyönä & Häikiö Reference Hyönä and Häikiö2005). However, Finnish has many compound words, and semantic PB can be obtained from the second constituent of a compound word (White et al. Reference White, Bertram and Hyönä2008). Finnish words tend to be long, and acuity limitations might weaken parafoveal effects. Note that also for morphological information, closely related to semantics, there is no evidence for parafoveal processing during the reading of English (Inhoff Reference Inhoff1989; Kambe Reference Kambe2004; Lima Reference Lima1987), whereas morphological PB effects are reliable in morphologically rich Hebrew (Deutsch et al. Reference Deutsch, Frost, Pollatsek and Rayner2000; Reference Deutsch, Frost, Peleg, Pollatsek and Rayner2003; Reference Deutsch, Frost, Pollatsek and Rayner2005).

In general, to the degree that phonological decoding is easy in alphabetic languages, we expect semantic PB effects to emerge, although they might not be quite as strong as in some non-alphabetic (logographic) languages like Chinese, in which “semantic information comes first” (sect. 3.2.1, para. 2), as reflected in the position of semantic and phonetic radicals in compound characters. Semantic decoding is fast and phonological decoding relatively slow in Chinese (Zhou & Marslen-Wilson Reference Zhou and Marslen-Wilson2000). Consequently, there is fairly unequivocal evidence for semantic PB (Yan et al. Reference Yan, Richter, Shu and Kliegl2009; in press; Yen et al. Reference Yen, Tsai, Tzeng and Hung2008). In fact, semantic PB is even larger than phonological PB in Chinese (Yan et al. Reference Yan, Richter, Shu and Kliegl2009), and the latter is somewhat delayed (Liu et al. Reference Liu, Inhoff, Ye and Wu2002; Pollatsek et al. Reference Pollatsek, Tan and Rayner2000; Tsai et al. Reference Tsai, Lee, Tzeng, Hung and Yan2004).

Processes of semantic and phonological activation not only share resources, but also differ in timing, with relative speed and timing a function of the writing system. Whereas phonological processing has a head start in alphabetic languages, semantic processing may start drawing resources earlier in logographic systems. Computational models of eye-movement control during reading (Engbert et al. Reference Engbert, Longtin and Kliegl2002; Reference Engbert, Nuthmann, Richter and Kliegl2005; Reichle et al. Reference Reichle, Pollatsek, Fisher and Rayner1998; Reference Reichle, Rayner and Pollatsek2003) could provide a good starting point for modeling such differences. More generally, in the spirit of the target article, the pattern of results calls for more language-comparative studies using comparable material, methods, and design to arrive at a universal model of reading.