The role of phonology in visual word recognition and reading Marc Brysbaert Reading silently Recent skill (not before IX century; may require spaces between words) Takes some time in the development (children first read aloud; also high degree of learning problems and failure, certainly in English) Reading silently (cont.) Silent reading goes faster than reading aloud (250-300 words per minute, depending on reader, text, and goal) People remember more after silent reading than after reading aloud. Once mastered very powerful skill, because then reading becomes automatic (cf. Stroop effect) Reading silently (cont.) Not purely based on visual information inner voice phonological loop in working memory errors in proofreading particularly frequent for homophones and mute letters tongue-twister effect (e.g., “Boris burned the brown bread badly.”). Also in Chinese (Zhang & Perfetti, 1993) Addressed phonology vs. assembled phonology Given that phonological coding plays an important part in reading, where does the coding take place: before or after the word is recognised? Originally (1970s) many researchers thought “before” (implicit speech in reading) Gradually, shift to “after” or “a combination”. Addressed phonology vs. assembled phonology Important element: the development of the dual-route theory (Coltheart, 1978, 1993, 2001) Latest version: the DRC-model (Coltheart et al., 2001) DRC print Feature Representations Letter Representations Semantic Representations Orthographic Lexicon Phonological Lexicon Phoneme Representations speech Rule-Based Translation Addressed phonology vs. assembled phonology DRC = a weak phonological theory bulk of visual word recognition is orthographically based GPC-route is slow and serial (from the word beginning to the word end) GPC-route activates the wrong phonology for irregular words (e.g., “pint”) The position of irregularity effect (e.g., Roberts et al., 2003: more difference in naming times between “bind” and “bluff” (2nd position) than between “beige” and “bless” (3rd position) Addressed phonology vs. assembled phonology Addressed phonology = phonology activated on the basis of word representations in the orthographic lexicon Assembled phonology = phonology activated on the basis of direct lettersound correspondences (graphemephoneme conversions) Evidence for the importance of phonology in isolated visual word recognition Rubenstein et al. (1971): it takes longer to reject a pseudohomophone (“brane”) in a lexical decision task Van Orden (1987): many false alarms with homophones in semantic decision (e.g., “rows” is a flower), in particular with brief presentation duration (prevents spelling check) Evidence for the importance of phonology in isolated visual word recognition Lesch & Pollatsek (1993): it takes longer to decide that “sand-beech” are unrelated than that “sand-bench” are unrelated Same finding with “pillow-bead” Evidence for the importance of assembled phonology in visual word recognition Many of the findings thus far might be explained on the basis of addressed phonology. If we want to show the importance of assembled phonology, we have to work with non-words, that do not have a lexical representation Perfetti & Bell (1991): masked priming Perfetti & Bell (1991) three types of primes for target RATE: rait (pseudohomophone), ralt (graphemic control), busk (unrelated control) Procedure: prime in lower case (25, 35, 45, 55, or 65 ms) TARGET in upper case (30 ms) XXXXXX mask task = perceptual identification “which word was presented in capitals?” Perfetti & Bell (1991) Findings: Phonological priming is possible with pseudohomophones; so, it is non-lexical (i.e., assembled phonology) It takes some time before the phonological code is computed (45 ms) Ferrand & Grainger (1994) A further look at the time course of phonological activation and see whether this is the same for orthographic information French language: has many homophones, that can be written differently mert-MERE vs. mair-MERE vs. toul-MERE Ferrand & Grainger (1994) Procedure: ###### (500 ms) prime (14, 29, 43, 57 ms) TARGET (until lexical decision) Lexical decision is better than perceptual identification, because a more on-line task orthographic priming: mert vs. mair phonological priming: mair vs. toul Brysbaert (2001) To show that phonological priming is automatic, you have to create conditions where the use of the phonological code is negative Procedure of Perfetti & Bell (1991) with perceptual identification; 43 ms prime Two conditions with 60% fillers for which the targets were either preceded by pseudohomophonic primes (ieb-IEP) or by pseudohomophones of another word (gad-IEP) Lukatela & Turvey (1994) Phonological priming is not limited to form priming, you also find it for associative priming Prime duration 50 ms, word naming toad-FROG = towed-FROG = tode-FROG < tolled-FROG or tord-FROG Drieghe & Brysbaert, 2002 First replicated Lukatela & Turvey (57 ms) Drieghe & Brysbaert, 2002 Extended it to LDT (57 ms) Drieghe & Brysbaert, 2002 LDT( 258 ms) Strong phonological theories “. . .we take the primary and initial source of lexical activation in English to be phonological. The role of orthographic codes is then taken to be that of refining the lexical activation begun by phonology” (Lukatela & Turvey, 1994a, p. 108). Strong phonological theories (cont.) “Over the last two decades, a number of studies using brief-stimulus-presentation and masked-stimulus-presentation paradigms have reported phonological effects in visual word identification. . .. These effects have been taken as major evidence for a rapid, automatic, and obligatory phonological process during lexical access.” (Xu & Perfetti, 1999, p. 26). Strong phonological theories (cont.) “The consistent evidence for phonological computation, its role in lexical access when the minimality constraint is taken into account, the manner in which phonology is assembled from print and shaped into a detailed representation, and the basic role of phonological structures in conveying meaning all suggest that the role of phonology is more important than dual-route models have assumed.” (Frost, 1998, p.95) Strong phonological theories (cont.) Brysbaert (2001) “Now that the existence of mandatory prelexical phonology assembly has been demonstrated, the logical next question is what this code looks like.” Drieghe & Brysbaert (2002) “ Our data add further support to the strong phonological theory of visual word recognition, which claims that the stored lexico-semantic information requires a phonological access code.” Strong phonological theories (cont.) Coltheart, Rastle, Perry, Langdon, & Ziegler (2001) “Potential problems for the DRC model: Masked phonological priming effects” “… there currently exist some difficulties concerning exactly what the effects are that would need to be simulated. …” “Hence the implementation of a computational account of masking effects in the DRC would need to be accompanied by considerable further empirical work…” Rastle & Brysbaert (2006) Despite the previous evidence many researchers still not convinced about the importance of phonological coding in English Rastle & Brysbaert: meta-analysis of previous research in English + two new, fully controlled studies Task = lexical decision (stronger than naming + can be modelled in DRC) First new Lexical Decision Experiment Two types of primes : phonological (pharm FARM; korce - COARSE) and graphemic controls (gharm - FARM; roipe - COARSE) Selected from the ARC Nonword Database (Rastle et al., 2002) Same number of overlapping letters both on matching and non-matching positions phonological primes do not activate the targets to a higher degree in any component of DRC First new Lexical Decision Experiment (cont.) 112 word trials and 112 non-word trials (also with phononological and graphemic control primes) 42 participants presentation with DMDX (Forster & Forster, 2003) trial : ######## 500 ms prime 58 ms TARGET until response First new Lexical Decision Experiment (cont.) Results: phon contr effect 603 ms 5.8% 617 ms 7.5% 14 ms 1.7% No effect of orthographic similarity prime - target Second new Lexical Decision Experiment (rationale) In Experiment 1 (and all published experiments) only word trials preceded by pseudohomophones of existing words pharm FARM gharm FARM whone WONE sowd GOWD pharm FARM gharm FARM phite FITE biss BUSS • Phonology uninformative both for word/non-word decision and for the target that will follow a particular type of prime Second new Lexical Decision Experiment (cont.) 112 words and 112 pseudo-homophones (both with phononological and graphemic control primes) 80 participants procedure same as in Experiment 1 (SOA = 58 ms) after the experiment, session run again and this time participants tried to indicate whether the prime had been a pseudohomophone or a control (at chance) Second new Lexical Decision Experiment (cont.) Results: phon contr effect 634 ms 5.8% 643 ms 6.4% 9 ms 0.6% No effect of orthographic similarity prime - target Conclusions Rastle & Brysbaert (2006) Despite some justified concerns about the previous evidence, masked phonological priming effect in lexical decision is real The effect is rather small (d = .30) The effect does not depend on the orthographic similarity of prime and target Is this evidence against a weak phonological model like DRC? Conclusions Rastle & Brysbaert (2006) DRC simulates LDT by looking either at the orthographic activation of the most active word node or at the total activity in the orthographic lexicon. Different simulations show that it is impossible to find a parameter set that at the same time predicts phonological priming and correct reading of irregular words. Conclusions Rastle & Brysbaert (2006) The situation looks much better when we look at the activity of the phonological lexicon. There we see clear phonological priming. However, is it possible to make a word/ pseudohomophone decision on the basis of this lexicon? (researchers always assumed this was not possible) Conclusions Rastle & Brysbaert (2006) In a weak phonological model, the activation of phonology is much stronger for a word than for a pseudohomophone, because a word also activates the phonology via the orthographic lexicon. So, the masked phonological priming effect is not really evidence against a weak phonological model. A new challenge: The transposed letter priming effect How to reconcile the findings with transposed letters (fiary-TALES) with the use of phonology? Perea & Carreiras (2006): Is it also possible to have transposed letter priming with pseudohomophones? In Spanish “v” and “b” sound the same; so “rebolucion” is a pseudohomophone of “revolucion” A new challenge: The transposed letter priming effect Will “relubocion” then also prime “REVOLUCION”? 50 ms priming, LDT results reloducion-REVOLUCION reluvocion-REVOLUCION relubocion-REVOLUCION 585 ms 570 ms 585 ms Conclusion: TL-effect is orthographic A new challenge: The transposed letter priming effect Grainger et al. (2006) Letter position effects are part of the orthographic route (cf. Grainger & Ferrand’s findings of the time course of orthography and phonology) Therefore, letter position effects should be stronger for short prime durations than for long prime durations 33 ms prime duration vs. 83 ms A new challenge: The transposed letter priming effect (Grainger et al., 2006) 33 ms prime duration slne-SILENCE: brma-SILENCE: 577 ms 597 ms 83 ms prime duration slne-SILENCE: brma-SILENCE: 613 ms 610 ms Reading list Frost, R. (1998). Toward a strong phonological theory of visual word recognition: True issues and false trails. Psychological Bulletin, 123, 71-99. Rastle, K. & Brysbaert, M. (2006). Masked phonological priming effects in English: Are they real? Do they matter? Cognitive Psychology, 53, 97-145.