1. Serial Processing in the Assembly of Phonology from Print
Kathy Rastle
Royal Holloway University of London

2. Collaborators Max Coltheart, Macquarie University
Derek Besner & Martha Roberts, University of Waterloo
Jelena Havelka, University of Kent
Taeko Wydell, Brunel University
Most of this work is summarized in:
Rastle & Coltheart (2006). Is there serial processing in the reading system; and are there local representations? In S. Andrews (Ed.) From inkmarks to ideas: Current issues in lexical processing. NYC: Psychology Press.

3. Reading aloud: a dual-route theory
print
Early Sources of Empirical Support
Regularity effect, and its interaction with frequency.
Surface and phonological dyslexia.
‘Route emphasis’
Lexical (dictionary) route
Nonlexical (rule-based) route
speech

4. A turn toward PDP modelling
Orthographic Units
Hidden Units
Phonological Units
Seidenberg & McClelland, 1989
“…empirical and theoretical constraints cannot be identified within dual-route models because of the essentially vague and unconstrained manner in which the models are stated.” (Seidenberg, 1985)
1980s
Empirical jitters (Campbell & Besner, 1981; Glushko, 1979; Kay & Marcel, 1981; Rosson, 1983).
Growing dissatisfaction with lack of theoretical specificity.

5. A turn toward PDP modelling
Orthographic Units
Hidden Units
Phonological Units
Seidenberg & McClelland, 1989
Captured statistical structure of O-P mapping.
No rules, no lexicon.
Shown to be insufficient, now superseded (Plaut et al., 1996; Harm & Seidenberg, 2004).
Models aspire to certain a priori ‘principles’ – e.g., distributed representation, strictly parallel processing.
PDP models cannot offer a sufficient account of orthography-phonology translation because they define in advance that processing must occur strictly in parallel.

6. The DRC model Feature Analysis Letter Analysis
print
speech
Feature Analysis
Letter Analysis
Orthographic input lexicon
Phonological output lexicon
Phoneme system
GPC translation
Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001, Psychological Review

7. The DRC model /p I n t/ Feature Analysis Letter Analysis
print
Feature Analysis
Letter Analysis
GPC translation
/p I n t/
Phoneme system
speech

8. The DRC model /p I n t/ Empirical Phenomena
Position of Irregularity Effect (Coltheart & Rastle, 1994; Rastle & Coltheart, 1999)
Position of Bivalence Effect (Havelka & Rastle, 2005)
Position-Sensitive Stroop Effect (Coltheart et al., 1999; Bibi et al., 2000)
Length by Lexicality Effect (Weekes, 1997)
Length by Script Effect (Wydell et al., in prep)
Exaggerated Length Effect in Surface Dyslexia (Gold et al., 2005)
Whammy Effect (Rastle & Coltheart, 1998)
Length by Language Effect (Ziegler et al., 2001)
Onset Effect in Masked Form Priming (Forster & Davis, 1991; Kinoshita, 2003)
print
Feature Analysis
Letter Analysis
GPC translation
/p I n t/
Phoneme system
speech

9. The Position of Irregularity Effect
print
speech
Feature Analysis
Letter Analysis
Orthographic input lexicon
Phonological output lexicon
Phoneme system
GPC translation
If phonological assembly operates serially, then incorrect information that arrives early should hurt performance more than information that arrives late.

10. The Position of Irregularity Effect
Position Irregular Regular Cost
1 chaos comic
2 goody garlic
3 stifle staple
4 wrestle wrinkle
5 debris dismay
Coltheart & Rastle, 1994

11. The Position of Irregularity Effect
Position Irregular Regular Cost
1 chaos comic 59 ms
2 goody garlic 32 ms
3 stifle staple 28 ms
4 wrestle wrinkle 12 ms
5 debris dismay 8 ms
Coltheart & Rastle, 1994

12. The Position of Irregularity Effect: Replication with Monosyllables
Position Irregular Regular Cost (Ss) Cost (DRC)
1 thyme tempt
2 butch batch
3 climb cliff

13. The Position of Irregularity Effect: Replication with Monosyllables
Position Irregular Regular Cost (Ss) Cost (DRC)
1 thyme tempt ms cycles
2 butch batch ms cycles
3 climb cliff ms cycles
Rastle & Coltheart, 1999

14. The Position of Irregularity Effect: Replication at Positions 2 and 3
Position Irregular Regular Cost (Ss) Cost (DRC)
2 rind rhyme
3 crow claw

15. The Position of Irregularity Effect: Replication at Positions 2 and 3
Position Irregular Regular Cost (Ss) Cost (DRC)
2 rind rhyme ms cycles
3 crow claw ms cycles
This body of findings is consistent with the claim that one component of print-to-sound translation (ie., nonlexical assembly) operates in a serial manner.
Roberts, Rastle, Coltheart, & Besner, 2003

16. The Position of Bivalence Effect
Serbian writing system
Adults equally competent in reading and writing both alphabets.
Completely consistent: Say what you read, and write what you hear.

17. The Position of Bivalence Effect
Serbian word types
Unique words - unique and common letters, can be read only in one alphabet
Bivalent words – common and ambiguous letters. Read as a word in one alphabet and as a nonword in another.
KOBAC – “hawk” in Roman, pronounceable nonword in Cyrillic
HOBOCT – “news” in Cyrillic, pronounceable nonword in Roman
Bivalent words read aloud more slowly than unique words.

18. The Position of Bivalence Effect: KOBAC example
print
Feature Analysis
KOBAC
Letter Analysis
Orthographic input lexicon
GPC translation
K O B A C
Semantics
Phonological output lexicon
/kobac/
Phoneme system k o b x a c x
/kobac/
speech

19. The Position of Bivalence Effect
Rom-Biv Cyr-Biv Rom-Cost Cyr-Cost
Initial pakt paka
Final kamp atap
Each bivalent word matched to a unique control. Cost of bivalence = bivalent-unique.

20. The Position of Bivalence Effect
Rom-Biv Cyr-Biv Rom-Cost Cyr-Cost
Initial pakt paka ms ms
Final kamp atap ms ms
Each bivalent word matched to a unique control. Cost of bivalence = bivalent-unique.
Consistent with serial, letter-by-letter phonological assembly system. Inconsistent with any model that operates solely in parallel.
Havelka & Rastle, 2004

21. The Position-Sensitive Stroop Effect
RED
Semantic Stroop:
Phonological Stroop:
SKY
PUT
DUP
QZT
Even for pronounceable nonwords, phonology is generated automatically from print, even if detrimental to performance.
Coltheart, Woollams, Kinoshita, & Perry, 1999

22. The Position-Sensitive Stroop Effect
Name the Colour:
print
Feature Analysis
KIT
RAT
Letter Analysis
POD
Orthographic input lexicon
Semantic System
RED
GPC translation
Phonological output lexicon
Relative to a phonologically-unrelated word, words sharing phonemes with the target colour will produce a benefit. The greatest benefit will occur when the phoneme shared is early.
Phoneme system
speech
Coltheart et al. (1999)

23. The Position-Sensitive Stroop Effect
Position of Overlap
Initial Final
Critical (e.g., RAT, POD) 557 ms (127.55) 571 ms (131.48)
Control (e.g., KIT) 587 ms (131.76) 583 ms (132.54)
Colour naming faster when word+colour share a sound, and effect is larger when shared phoneme is early than when it is late. Replicated in Hebrew (but effect is reversed).
Consistent with serial, letter-by-letter phonological assembly system. Inconsistent with any model that operates solely in parallel.
Coltheart et al. (1999)

24. Interim Summary Position of Irregularity Effect
The cost of GPC irregularity declines monotonically and linearly across the position of irregularity, with earlier irregularities being most costly.
Position of Bivalence Effect
The cost of bivalence in Serbian reading is greater when the ambiguous letter occurs at the beginning of the word than when it occurs at the end of the word.
Position-Sensitive Stroop Effect
Colour naming is facilitated by a phonologically-related distractor, and this effect is more pronounced when the phonological overlap occurs at the beginning of the word than when it occurs at the end of the word.

25. The Length by Lexicality Effect
A serial, letter-by-letter, nonlexical translation system predicts:
An effect of length (number of letters) on reading aloud that is particularly apparent for nonwords.
Weekes (1997) examined the effect of length on word and nonword reading aloud latency.
-100 high-frequency words
-100 low-frequency words
-100 nonwords

26. The Length by Lexicality Effect
Human Data (Weekes, 1997)

27. The Length by Lexicality Effect
Human Data (Weekes, 1997)
DRC Data

28. The Exaggerated Length Effect in Surface Dyslexia
print
speech
Feature Analysis
Letter Analysis
Orthographic input lexicon
Phonological output lexicon
Phoneme system
GPC translation
Surface Dyslexia
Spared reading aloud of nonwords
Impaired reading aloud of irregular words: e.g., yacht ->/y{Jt/; pint->/pInt/; have-> /h1v/, etc.

29. The Exaggerated Length Effect in Surface Dyslexia
print
speech
Feature Analysis
Letter Analysis
Orthographic input lexicon
Phonological output lexicon
Phoneme system
GPC translation
Patients who accomplish their reading largely through the nonlexical route should show a length effect that extends to regular words (unlike unimpaired readers).

30. The Exaggerated Length Effect in Surface Dyslexia
Group 3/4 letters 5/6 letters Diff
Healthy controls ms 678 ms 9 ms
Alzheimer’s patients 803 ms 812 ms 9 ms
Sem. dementia w/ surface dyslexia

31. The Exaggerated Length Effect in Surface Dyslexia
Group 3/4 letters 5/6 letters Diff
Healthy controls ms 678 ms 9 ms
Alzheimer’s patients 803 ms 812 ms 9 ms
Sem. dementia w/ surface dyslexia 844 ms 918 ms 74 ms
Consistent with serial, letter-by-letter phonological assembly system. Inconsistent with any model that operates solely in parallel.
Gold et al., 2005

32. The Length by Script Effect
Japanese Writing System
Kanji (金魚 )
Hiragana (きんぎょ )
Katakana (キンギョ)
Considering only Hiragana and Katakana.
Some words only ever appear in one of these two scripts.
These words can always be transcribed into the other script.
Pronunciation identical but orthographic nonwords.
Transcriptions should show an exaggerated length effect.

33. The Length by Script Effect
Length (mora) Typical Script Transcribed Script ms ms ms ms
Difference 89 ms

34. The Length by Script Effect
Length (mora) Typical Script Transcribed Script
ms ms
ms ms
ms ms
ms ms
Difference 89 ms ms
Consistent with serial, letter-by-letter phonological assembly system. Inconsistent with any model that operates solely in parallel.
Wydell, Rastle, Coltheart, & Besner, in prep

35. The Length by Script Effect
Typical Transcribed
Katakana LF (y=30.6x+599) Katakana LF (y=53.8x+626)
Katakana HF (y=25.6x+531) Katakana HF (y=59x+546)
Hiragana LF (y=31.7x+598) Hiragana LF (y=46.9x+604)
Hiragana HF (y=32.4x+526) Hiragana HF (y= )

36. Interim Summary Length by Lexicality Effect
Reading aloud latency increases with length of the letter string, but only if the letter string is a nonword. There is no length effect for regular monosyllabic words.
Exaggerated Length Effect in Surface Dyslexia
Surface dyslexics (thought to read via the nonlexical route) show an exaggerated length effect when reading regular words.
Length by Script Effect
The length effect in Japanese reading aloud is exaggerated if words are transcribed into a script in which they do not normally occur (and are thus read as nonwords).

37. Conclusions
One component of print-to-sound translation (nonlexical assembly) operates in a serial manner.
These six phenomena can be accounted for (and have largely been simulated) by the DRC model of reading, one component of which operates in a serial, letter-by-letter, manner.
These six phenomena challenge an a priori ‘principle’ of PDP models. PDP models define in advance of the empirical evidence that processing is Parallel, leaving them unable to find the correct serial solution.
My view: so much for the principles. Evidence-based approaches to modeling human cognition are preferable to faith-based approaches.