Effects of orthography on speech production in a form-preparation paradigm

Four experiments investigated potential influences of spelling on single word speech production. A form-preparation paradigm that showed priming effects for words with initial form overlap was used to investigate whether words with form overlap, but different spelling (e.g., ‘‘camel’’-‘‘kidney’’) also show priming. Experiment 1 demonstrated that such words did not benefit from the form overlap, suggesting that the incongruent spelling disrupted the form-preparation effect. Experiment 2 replicated the first experiment with an independent set of items and an improved design, and once again showed a disruptive effect of spelling. To divert participants attention from the spelling of the targets, Experiment 3 was conducted entirely in the auditory domain, but yielded the same outcome as before. Experiment 4 showed that matching initial letters alone, in the absence of matching sounds (e.g., ‘‘cycle’’-‘‘cobra’’), did not produce priming. These findings raise the possibility that orthographic codes are mandatorily activated in speech production by literate speakers. 2003 Elsevier Science (USA). All rights reserved. Language is supported by various subsystems, including phonology, orthography, semantics, and syntax. One of the most fundamental (and well studied) issues in psycholinguistics is the extent to which these subsystems act in a modular fashion. On the modular approach, outputs of one subsystem serves as input to another, with no feedback. For instance, according to the modular view of speech production developed by Garrett (1975), and Levelt and colleagues (e.g., Levelt, Roelofs, & Meyer, 1999), a non-linguistic conceptual system that encodes a speaker s thoughts activates the relevant lexical–semantic representations (semantic representations associated with syntax, often referred to as lemmas), which in turn retrieve lexical–phonological codes. Central to this approach is the claim that information processing is unidirectional, and that the retrieval of phonology in no way contributes, via feedback, to the selection of lemmas. On a non-modular approach, by contrast, the initial activation of lexical–semantic codes (based on conceptual inputs) leads to partial activation of phonology, which in turn feeds back and constrains the final lemma selection (Dell, 1986). On a stronger version of this approach, all the various subsystems would become activated in any language task, and interactions would occur between all systems. For instance, in speech production, not only would phonology feed back on semantics, but orthographic information would become activated as well and constrain the speech production process, despite the fact that orthographic information should be irrelevant to the process of speaking (for such a theoretical framework in the area of visual word perception, see, e.g., Van Orden & Goldinger, 1994). Journal of Memory and Language 49 (2003) 119–132 Journal of Memory and Language www.elsevier.com/locate/jml Corresponding author. Fax: +44-117-928-8588. E-mail address: [email protected] (M.F. Damian). 1 In multilinguals, mutual co-activation of components might even occur between languages; see Dijkstra, Timmermans, and Schriefers (2000). 0749-596X/03/$ see front matter 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S0749-596X(03)00008-1 Most work relevant to the issue of modularity has focused on interactions between processes that are integral to the task at hand. So in the case of speech production, semantics and phonology are two essential steps in the process of converting a thought to an utterance, and the most common question is whether phonology feeds back on semantics. Various findings have been taken as evidence that such feedback indeed occurs (e.g., Dell & O Seaghda, 1992; Dell, Schwartz, Martin, Saffran, & Gagnon, 1997; but see, e.g., Levelt, 1992; Levelt et al., 1999, for a modular position). Similarly, in speech perception, the question is whether higher level information that must be accessed (e.g., complete phonological word forms, semantics, etc.) influences lower level perceptual processes (e.g., perception of phonemes). And again, various evidence suggests this is the case (e.g., Samuel, 1997), although there are modular interpretations of such findings as well (e.g., Norris, McQueen, & Cutler, 2000). In reading, the question is whether the identification of the visual structure of words (orthographic processing) is influenced by feedback from phonology or semantics, processes that are essential in the task of reading. Again, various evidence suggests that there is feedback from phonology (Stone, Vanhoy, & Van Orden, 1997), semantics (Hino & Lupker, 1996; Hino, Pexman, & Lupker, 2002; Pecher, 2001), and top-down processes within the orthographic system (McClelland & Rumelhart, 1981), although there are again different interpretations of these findings (cf. Grainger & Jacobs, 1994; Peereman, Content, & Bonin, 1998). The current study considers potential interactions between language subsystems when some are not strictly relevant; in particular, we consider the role of orthography in constraining speech production. To the extent there are interactions, it would reflect a strong version of non-modularity, with diverse forms of linguistic knowledge automatically accessed (regardless of their relevance) and constraining speech production. There is in fact a small literature relevant to testing this strong version—most notably, the role of orthographic processes in speech perception. And as reviewed below, the findings often support a strong interactive position. In a seminal study by Seidenberg and Tanenhaus (1979), participants either monitored lists of auditorily presented words for a target that rhymed with a cue word (Experiments 1 and 2), or performed rhyme judgements on auditorily presented pairs of words (Experiment 3). In both tasks, latencies were longer for cue–target pairs that rhymed but were orthographically dissimilar (rye-tie) than for those that were orthographically similar (pie-tie). Donnenwerth-Nolan, Tanenhaus, and Seidenberg (1981) replicated these findings while excluding the potential confound that orthographically similar cue–target pairs in Seidenberg and Tanenhaus study might have simply been more predictable than dissimilar ones. Both studies suggest that orthographic information influences performance in tasks that do not require access to visual codes. Taft and Hambly (1985) used a syllable monitoring task in which participants were asked to judge whether an auditorily presented syllable (such as ‘‘fin’’) occurred in an auditory target word (e.g., ‘‘finale’’). Of special interest were target words in which the key syllable was on the unstressed syllable, hence the vowel was reduced (‘‘metallic’’; /mEtælIk/). When syllables were presented in which the vowel of the syllable was of full value (e.g., /met/), participants tended to erroneously judge them as matching with the target word. This could be the case because the morphologically related word ‘‘metal’’ (/metl/) exists, in which the vowel is full; possibly the auditory system automatically accesses the underlying common morphological code, and match/mismatch decisions are based on these codes. However, error rates were similarly high if syllable and target were not morphologically related (e.g., ‘‘lag’’, /læg/-‘‘lagoon’’, /lEgun/). These results were attributed to an underlying orthographic code that was mandatorily activated in auditory processing. This code matched in both cases between syllable and target (‘‘met’’-‘‘metallic’’; ‘‘lag’’-‘‘lagoon’’) and obscured the phonetic mismatch between the full and the reduced vowel. Hence, spelling appeared to act as a powerful cue that induced participants to judge syllables and words as matching if their spelling also matched. Findings reported by Dijkstra, Roelofs, and Fieuws (1995) also appear to support the claim that orthographic codes are involved in speech perception. In a phoneme monitoring task performed on Dutch words, response latencies were affected by the spelling of the targets: when target phonemes had more than one possible spelling (e.g., /k/! ‘‘k’’ or ‘‘c’’), responses were faster when they were presented in their primary spelling (e.g., /k/ in ‘‘kabouter’’) than when presented in their secondary spelling (e.g., /k/ in ‘‘cabaret’’). These findings suggest that in addition to phonemic and phonological codes, orthographic codes also exert an influence in the phoneme monitoring task. Note, however, that Cutler, Treiman, and van Ooijen (1998) attributed this finding to participant strategies rather than automatic access to orthographic codes: they demonstrated that the effect disappeared when a large number of consistently spelled filler items was inserted that distracted participants attention from the spelling of the stimuli. The effects of orthography have also been investigated in priming paradigms: Jakimik, Cole, and Rudnicky (1985) asked participants to perform auditory lexical decisions to monosyllabic targets (e.g., ‘‘mess’’) that were preceded by multisyllabic auditory prime words (e.g., ‘‘message’’). Facilitation was obtained only when spelling and sound of prime and target coincided (as in the example above), but not when they were related by sound 120 M.F. Damian, J.S. Bowers / Journal of Memory and Language 49 (2003) 119–132