Masked priming paradigm. A prime is presented briefly between a forward mask consisting of XXXXXX or ####### and a target word. Although participants report not seeing the prime, it facilitates recognition of the target word if there is overlap between the prime and target. The overlap can be orthographic, phonological, morphological, or semantic.
Stronger evidence for the involvement of assembled phonology in visual word recognition in English comes from experiments using masked priming. In masked priming (Figure 4.1), a target word is preceded by a prime, presented so briefly (~50 milliseconds) that participants report not seeing it. Still, participants are influenced by the prime: A target word is processed more efficiently when the prime is related to it than when it is not. Perfetti and Bell (1991) observed more priming of target words (e.g., rate) with homophonic pseudowords as primes (e.g., rait) than with orthographic control primes (e.g., ralt) that share the same letter overlap with the target but not the same overlap in sound. Because the primes were pseudowords, the effect could only be explained on the basis of assembled phonology.
Phonology‐mediated masked priming has been observed in many studies, at least when prime duration is slightly longer than 50 ms and when there is a considerable difference in phonology between the homophonic prime and the orthographic control prime (for review, see Rastle & Brysbaert, 2006). The activation of assembled phonology seems to be automatic: Phonological priming is still evident when strategic effects are minimized by reducing the proportion of phonological priming trials to only 9% (Xu & Perfetti, 1999). Another important demonstration came from Lukatela and Turvey (1994) who reported priming of the target word FROG by the pseudoword tode. This shows that written words can activate their meaning through assembled phonology (tode – toad – FROG).
In summary, there is good evidence that phonological information is activated during word reading, and that the activation can be in the form of assembled phonology as well as addressed phonology.
Phonology in words with inconsistent mappings
English has inconsistencies in the mappings from orthography to phonology. The grapheme ea can refer to the phoneme /i/ (as in bead) or /ε/ (as in head). Which of these is activated by assembled phonology?
One assumption might be that that there is a set of grapheme‐phoneme correspondence rules and only one phoneme is activated (see Rastle & Coltheart, 1999, for a set of rules). In the example of ea, the grapheme‐phoneme conversion rule would be ea ‐> /i/, on the basis that this is the most frequent conversion in English monosyllabic words containing the grapheme ea. On the basis of assembled phonology, the word head would be pronounced /hid/. Words with a conflict between assembled and addressed phonology are called irregular words (because they do not follow the rules).
The word‐naming task has been used to investigate conflicts between assembled and addressed phonology. Participants are presented with printed words and asked to name them aloud as rapidly as possible. As predicted, Rastle and Coltheart (1999) observed longer naming times for irregular words (hearts) than for regular words (hounds). In addition, the effect was stronger when the irregularity was at the beginning of the word (aisle) than toward the end (swap). Rastle and Coltheart interpreted this as evidence for the hypothesis that assembled phonology is a serial process going from word beginning to word end.
The assumption that there is a set of grapheme‐phoneme correspondence rules has been contested, however. Glushko (1979) observed that participants took longer to name the word wave than wade, even though both contained the same vowel grapheme and both followed the ae ‐> /ā/ rule. Glushko argued that the difference arose because all words ending on –ade are pronounced in the same way, whereas this is not true for the words ending on –ave (think of have). According to Glushko, what is important is not whether a grapheme‐phoneme conversion follows a rule, but the extent to which the conversion is consistent across words.
Further research found that in words with inconsistent mappings, not one but all possible phonemes become activated to some extent. This would mean that the written word bead activates both the phonology /bid/ and /bεd/; similarly, the assembled phonology of head includes both /hid/ and /hεd/. Evidence comes from Lesch and Pollatsek (1993) who observed that participants often erroneously indicate that the written words PILLOW and BEAD are semantically related. This is only possible if the visual word bead also activates the phonological code /bεd/.
More evidence for the automatic activation of several possible phonological codes from orthography comes from research with bilinguals. If assembled phonology is computed automatically (Xu & Perfetti, 1999), the question then arises as to which phonology is computed in bilinguals who have mastered several alphabetic languages. Take a Dutch‐English bilingual for whom the Dutch word meel (flower) sounds like male and the English word male has the same Dutch rhyme pronunciation as the word finale. Which phonological forms are activated when the bilingual person reads a text? The phonology of the first language (Dutch)? Or the phonology of the language being read, in which case the activation of assembled phonology cannot be automatic but is under strategic control?
A series of studies show that bilinguals activate the codes of both languages when they are reading (Brysbaert et al., 1999; Friesen et al., 2020; Gollan et al., 1997; Nakayama et al., 2012; Timmer et al., 2014; van Wijnendaele & Brysbaert, 2002; Wu, & Thierry, 2010; Zhou et al., 2010). For instance, Brysbaert et al. (1999) reported that in bilinguals it is possible to prime a second‐language word with a homophonic prime from the first language (so, the English target word male is primed by the Dutch word meel in a Dutch‐English bilingual). It is also possible to prime the word female with the Dutch pseudohomophone fiemeel. Van Wijnendaele and Brysbaert (2002) further observed that priming also works from the second language to the mother tongue. In proficient Dutch‐English bilinguals it is possible to prime the Dutch word meel [flower] with the English word male, and the Dutch target word penseel (paintbrush) with the English pseudohomophone pensale. This pattern is found when participants have no knowledge that masked primes are being presented in their second language while they are responding to words in their native language. Gollan et al. (1997) and Nakayama et al. (2012) even reported phonological priming across languages with different scripts, such as Hebrew‐English or Kana‐English.
All in all, there is strong evidence that in alphabetic writing systems, phonology is assembled automatically in the early stages of visual word recognition, and that various possible phonological codes are co‐assembled that can in turn contribute to visual word recognition and reading.
Is there need for an orthographic code in visual word recognition?
In the previous section, we discussed issues related to inconsistencies in grapheme‐phoneme mappings in English. But what happens when words are fully consistent (as in kiss)? Or in languages without inconsistencies? If the written code is fully transparent, is there still a need for speakers to store information about visual word forms in the brain, or is it more economical to assemble phonology and then access stored representations of spoken words?
The idea that visual word recognition may be fully mediated by assembled phonology was defended by Frost (1998) in the strong phonological theory. Frost gave a list of arguments why such an organization would be likely in transparent languages, and might even apply to English, despite the inconsistent mappings for some words. In a way, the strong phonological theory returned to the original question asked by Huey (1908).
Evidence against the strong phonological hypothesis was
