Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Synaesthesia and learning : a bidirectional relationship Watson, Marcus Robert 2013

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2013_fall_watson_marcus.pdf [ 1.91MB ]
Metadata
JSON: 24-1.0074011.json
JSON-LD: 24-1.0074011-ld.json
RDF/XML (Pretty): 24-1.0074011-rdf.xml
RDF/JSON: 24-1.0074011-rdf.json
Turtle: 24-1.0074011-turtle.txt
N-Triples: 24-1.0074011-rdf-ntriples.txt
Original Record: 24-1.0074011-source.json
Full Text
24-1.0074011-fulltext.txt
Citation
24-1.0074011.ris

Full Text

SYNAESTHESIA AND LEARNING A BIDIRECTIONAL RELATIONSHIP by MARCUS ROBERT WATSON B.Hum., Carleton University (2001) M. Phil., Simon Fraser University (2009) A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Psychology) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) July 2013 © Marcus Robert Watson, 2013 Abstract I present new evidence about the relationships between learning and synaesthesia, par-­‐ticularly grapheme-­colour synaesthesia, in which individuals experience letters andnumbers as coloured. As part of the largest survey of synaesthetic tendencies ever per-­‐formed, I show that second language acquisition can act as a trigger for the develop-­‐ment of synaesthesia, such that children who learn a second language in grade schoolare three times more likely to develop synaesthesia as native bilinguals. I also demon-­‐strate that previous reports of a sex bias in synaesthesia are almost certainly due to re-­‐sponse and compliance biases, rather than any real differences in the prevalence ofsynaesthesia between men and women. In a detailed examination of the in>luences oflearning on synaesthetic experiences, I show that synaesthetic colours are in>luenced byknowledge about letters’ shapes, frequencies, alphabetical order, phonology, and cate-­‐gorical qualities. Finally, I demonstrate that synaesthesia can itself be exploited inlearning. All these results are presented as supporting a developmental learning hypoth-­ esis	
  of	
  synaesthesia,	
  in	
  which	
  synaesthesia	
  develops,	
  at	
  least	
  in	
  part,	
  because	
  it	
  is	
  useful. ii Preface This thesis describes a number of studies of grapheme-­‐colour synaesthesia that tookplace at the University of British Columbia, Simon Fraser University, and Charles Univer-­‐sity in Prague. For the SFU/CU synaesthesia survey, which is the subject of Chapter 2and provided the data for Chapters 4 and 5, I participated in the research design fromthe start, in roughly equal collaboration with Kathleen Akins and Lyle Crawford. I did lit-­‐tle data collection, but performed all analyses solo, with helpful suggestions and inputfrom	
  collaborators.	
  All	
  the	
  writing	
  in	
  these	
  chapters	
  is	
  my	
  own.Chapter 3 is a slightly adapted version of a previously published paper, (Watson, M. R.,Akins, K. A., & Enns, J. T. (2012). Second-­‐order mappings in grapheme-­‐color synesthesia. Psychonomic Bulletin and Review, 19(2), 211-­‐217). The data for this paper was kindlyprovided by Michael Dixon and Jonathan Carriere of the University of Waterloo. The ini-­‐tial research question was collaboratively arrived at by the three co-­‐authors, all analyseswere	
  my	
  own,	
  and	
  I	
  was	
  the	
  principal	
  author	
  of	
  the	
  paper.Chapter 6 is also taken from a previously published paper (Watson, M. R., Blair, M. R.,Kozik, P., Akins, K. A., & Enns, J. T. (2012). Grapheme-­‐color synaesthesia bene>its rule-­‐based category learning. Consciousness and Cognition, 21, 1533-­‐1540). Here I was theprimary person involved in determining the research question and experimentalmethod, though all my collaborators made many useful suggestions and changes. Allanalysis	
  was	
  performed	
  by	
  myself,	
  and	
  I	
  was	
  the	
  primary	
  author	
  of	
  the	
  paper.The research described here was approved by the UBC and SFU Of>ices of ResearchEthics.	
  (UBC	
  BREB	
  #	
  H10-­‐00287,	
  SFU	
  #	
  39456). iii Table	
  of	
  Contents Abstract ii............................................................................................................................. Preface iii.............................................................................................................................. Table of Contents iv............................................................................................................. List of Tables vii................................................................................................................... List of Figures ix.................................................................................................................. Acknowledgements x........................................................................................................... Dedication xii........................................................................................................................ 1 Introduction 1........................................................................................................ 1.1  What is synaesthesia? 4................................................................................................ 1.1.1  Terminology and definitions 4................................................................................. 1.1.2  Operationalizing synaesthesia 5............................................................................... 1.1.3  Sub-types of synaesthesia 7..................................................................................... 1.2  Why are people synaesthetic? 9................................................................................... 1.2.1  The synaesthetic brain 9........................................................................................... 1.2.2  The synaesthetic genome 12.................................................................................... 1.2.3  The crucial role of learning in synaesthetic development 14................................... 1.3  How does learning change synaesthesia? 15................................................................ 1.3.1  Synaesthetic concurrents as fossils of learning processes 16.................................. 1.3.2  Semantic influences 16............................................................................................ 1.3.3  Common associative influences 17.......................................................................... 1.3.4  Universal influences? 17.......................................................................................... 1.3.5  Second-order influences 17...................................................................................... 1.4  Is synaesthesia good for anything? 19.......................................................................... 1.4.1  Synaesthesia and memory 19................................................................................... 1.4.2  Synaesthesia and creativity 20................................................................................. 1.4.3  Looking for synaesthetic styles of performance 21................................................. 1.5  An outline of this thesis 23........................................................................................... 2 Childhood learning and rates of synaesthesia—The prevalence of synaesthesia in Czech and English 26.................................................................................................... iv 2.1  Introduction 26............................................................................................................. 2.1.1  The development of literacy and the development of synaesthesia 27.................... 2.1.2  How common is synaesthesia? 30........................................................................... 2.1.3  Are there more female than male synaesthetes? 31................................................. 2.1.4  Outline of the study 33............................................................................................. 2.2  Methods 34................................................................................................................... 2.2.1  Phase I - Paper survey 34......................................................................................... 2.2.2  Phase II - Synesthesia Battery 35............................................................................. 2.3  Results 36..................................................................................................................... 2.3.1  Linguistic characteristics of the samples 36............................................................ 2.3.2  Higher rate of endorsing synaesthesia among Czechs 37........................................ 2.3.3  Higher rates of confirmed synaesthesia among Czechs 39...................................... 2.3.4  Higher rates among Czechs are due to late second-language acquisition 41........... 2.3.5  Grapheme stories and reading abilities are associated with reported but not confirmed synaesthesia 43............................................................................................... 2.3.6  A female bias for synaesthesia due to differences in compliance 45....................... 2.3.7  All forms of synaesthesia cluster together 48.......................................................... 2.4  Discussion 49................................................................................................................ 2.4.1  Overview of results 49............................................................................................. 2.4.2  Rates of synaesthesia are consistent with previous studies, likely under-estimates of true population rates 49................................................................................................... 2.4.3  Learning and synaesthesia 52.................................................................................. 2.4.4  Development or retention of synaesthesia 54.......................................................... 2.4.5  The generalizability of synaesthetic tendencies 54.................................................. 2.4.6  Sex bias in synaesthesia 55...................................................................................... 2.4.7  Problems with the self-report and consistency criteria? 56..................................... 2.4.8  Future directions 57................................................................................................. 3 Second-order mappings in grapheme-colour synaesthesia 58........................... 3.1  Introduction 58............................................................................................................. 3.2  Data preparation 61...................................................................................................... 3.3  Results 64..................................................................................................................... 3.3.1  Letter similarity measures predict different aspects of colour similarity 64............ 3.3.2  Letter shape and ordinality predict hue; letter frequency predicts luminance 65.... 3.3.3  Analyses of individual differences show that the mappings are independent 66.... 3.3.4  Which aspects of shape matter? 66.......................................................................... 3.4  Discussion 67................................................................................................................ 4 Higher-fidelity synaesthetic colour data increases strength of effects 70......... 4.1  Introduction 70............................................................................................................. 4.2  Participants 70.............................................................................................................. 4.3  Data preparation 71...................................................................................................... 4.4  Results 71..................................................................................................................... v 4.4.1  Shape-hue and ordinality-hue results replicate 72................................................... 4.4.2  The shape-hue and ordinality-hue effects are independent 73................................. 4.4.3  The same dimensions of shape predict hue distance 73........................................... 4.4.4  A frequency effect after all 74................................................................................. 4.5  Discussion 74................................................................................................................ 5 The structure of Czech synaesthesia 76............................................................... 5.1  Introduction 76............................................................................................................. 5.1.1  Unique features of Czech letters may affect synaesthetic colour 76....................... 5.1.2  Does phonological similarity map to synaesthetic colour similarity? 77................ 5.1.3  Alphabetical order vs. learning order in Czech 78................................................... 5.1.4  A special role for vowels? 80................................................................................... 5.1.5  Base-diacritical pairs 81........................................................................................... 5.1.6  I and Y 82................................................................................................................. 5.1.7  Outline of this study 82............................................................................................ 5.2  Participants 83.............................................................................................................. 5.3  Data preparation 83...................................................................................................... 5.4  Results 85..................................................................................................................... 5.4.1  Categorical second-order influences on synaesthetic colour in Czech 85............... 5.4.2  Notes on the remaining analyses 87......................................................................... 5.4.3  Shape-, ordinality-, and phonology-colour correlations in Czech 88...................... 5.4.4  Independent influences of shape, ordinality, and phonology 89.............................. 5.4.5  Shape-hue effect is strongest for vowel-vowel pairs, predictive dimensions for Czech and English overlap 90......................................................................................... 5.4.6  Ordinality effects with both hue and luminance, in opposite directions 92............. 5.4.7  Phonology-colour relations are strongest for vowel-like consonants 94................. 5.4.8  Is there any impact of learning order on synaesthetic colour? 95............................ 5.4.9  Chasing down the frequency-luminance effect 97................................................... 5.5  Discussion 97................................................................................................................ 5.5.1  Overview of results 97............................................................................................. 5.5.2  No learning order effect, a special role for sequences? 98...................................... 5.5.3  Ranking the influences on synaesthetic colour 99................................................... 5.5.4  A special role for vowels? 100................................................................................. 5.5.5  Ambiguous evidence for a hue/luminance split 100................................................ 6 Grapheme-colour synaesthesia benefits rule-based category learning 102..... 6.1  Introduction 102........................................................................................................... 6.2  Experiment 1 106......................................................................................................... 6.2.1  Participants 106........................................................................................................ 6.2.2  Displays and responses 107..................................................................................... 6.2.3  Category structure 107............................................................................................. 6.2.4  Foil stimulus sets 108............................................................................................... 6.2.5  Procedure 108.......................................................................................................... vi 6.2.6  Behavioral results 109.............................................................................................. 6.2.7  Self-report data 112.................................................................................................. 6.3  Experiment 2 113......................................................................................................... 6.3.1  Methods 114............................................................................................................. 6.3.2  Results 114............................................................................................................... 6.4  Discussion 115.............................................................................................................. 7 Conclusion 119....................................................................................................... 7.1  What do we know? 119................................................................................................ 7.1.1  Demonstrating 3 main points 119............................................................................ 7.1.2  Other findings 120................................................................................................... 7.2  What don’t we know? 122............................................................................................ 7.3  Why study synaesthesia? 124....................................................................................... Bibliography 126.................................................................................................................. Appendix 1 - Synaesthesia Survey 139............................................................................... Appendix 2 - Associations between reported synaesthesia types 141............................. Appendix 3 - Associations between confirmed synaesthesia types 142........................... vii List	
  of	
  Tables Table 2.1  Female:male relative rates of reported synaesthesia 46......................................... Table 2.2  Female:male relative rates of confirmed synaesthesia 46..................................... Table 3.1  Letter similarity measures used in the English study 62....................................... Table 3.2  Letter shape dimensions adapted from Gibson (1969) 63..................................... Table 3.3  Correlations between letter and colour similarity 64............................................. Table 4.1  Correlations between letter and colour similarity (new data set) 72..................... Table 5.1  The graphemes of the Czech alphabet 77.............................................................. Table 5.2  Three types of letter pairs in Czech 82.................................................................. Table 5.3  Letter similarity measures used in the Czech study 84.......................................... Table 5.4  Simple correlations between Czech letter similarity measures and colour distance 89.................................................................................................................................... Table 5.5  Summary of linear models predicting hue and luminance 89............................... viii List	
  of	
  Figures Figure 2.1  Differences in second-language acquisition between Czech and English speakers 37.................................................................................................................................... Figure 2.2  Reported rates of synaesthetic experiences for Czech and English speakers 39.. Figure 2.3  Confirmed rates of synaesthesia for Czech and English speakers 40.................. Figure 2.4  Rates of confirmed synaesthesia by age of second language acquisition 42....... Figure 2.5  Confirmed rates of synaesthesia among native Czech and English speakers who acquired a second language after age 1 43.............................................................................. Figure 2.6  Endorsement of questions concerning childhood reading and tendencies to tell stories involving graphemes 44.............................................................................................. Figure 2.7  Female:male relative rates across various stages of the experiment 47............... Figure 3.1  Scatterplots of three second-order mappings between letter similarity and synaesthetic colour similarity 65............................................................................................ Figure 5.1  Mean hue and luminance distances of all Czech letter pairs across all participants 86.................................................................................................................................... Figure 5.2  The shape-hue effect for Czech letters 90............................................................ Figure 5.3  The ordinality-hue effect in Czech synaesthesia 93............................................. Figure 5.4  The phonology-hue effect for vowel-consonant pairs alone 95........................... Figure 6.1  One of the stimulus sets 105................................................................................. Figure 6.2  Performance of participants in Experiment 1 111................................................ ix Acknowledgements I suppose it is standard to say that I couldn’t have completed this thesis without thework of many others, but it happens to be true. First of all, my PhD. supervisor Jim Ennsdid not just improve the speci>ic work I present here, but he taught me how to be a bet-­‐ter and smarter researcher overall. His clarity of vision and ability to mull over every as-­‐pect of an experimental design or a data set have repeatedly shown me how to get bet-­‐ter results and get more out of the results I do have. The >ive years I have spent workingwith him have given me the opportunity to see >irst hand what a scientist looks like, andI only hope to be able to live up to his standard in the rest of my career. He’s also a reallygood	
  guy.My M.A. supervisor Kathleen Akins has also been a central part of every project in thesepages, including being the P.I. on the work in Chapters 2, 4 and 5. She has an astonishingability to create a theoretical looking-­‐glass world, where what were previously out-­‐landish speculations begin to look normal, and what was previously sober and carefulthinking begins to look like sheer madness. I >ind myself here defending parts of a theo-­‐ry that she >irst described to me six or seven years ago in her of>ice, which initiallyseemed bizarre in the extreme but now seems to be one of the most plausible ways ofaccounting for half the >indings in our >ield. Her friendship and support well above andbeyond	
  the	
  call	
  of	
  duty	
  will	
  always	
  be	
  remembered.Mark Blair has also given me far more support of every kind than I ever deserved. Heonly directly collaborated on Chapter 6 in this thesis, but he has been a friend, mentor,collaborator	
  and	
  Starcraft	
  opponent	
  for	
  many	
  years	
  now.	
   x Jan Chromý single-­‐handedly took charge of the data collection in the Czech Republic forChapters 2 and 5, and has been incredibly patient with his substantially less-­‐organizedcollaborators in Vancouver. We had a far larger team at SFU, and so I owe many thanksto Lyle Crawford, David Gaber, Nazim Keven, Jason Leardi, Lydia Du Bois, Chris Spiker,Nicole Pernat, and others. Lyle Crawford was also a incredibly useful co-­‐author on thepaper that outlined the theoretical motivation for this thesis (Watson, Akins, &Crawford,	
  2010).	
  John Alderete developed the phonological similarity measures used in Chapter 5, andvery patiently explained to a linguistic ignoramus like myself how to think of them. Heand Martin Hahn were excellent sources of information on the Czech language in gener-­‐al, and helped design the appropriate analyses. David Eagleman kindly provided severalhours of his time to discuss theoretical issues with me, and re-­‐worked his SynesthesiaBattery	
  to	
  enable	
  our	
  Czech	
  study	
  to	
  go	
  forward.	
  My thesis examination committee also included Janet Werker, Peter Graf, Ara Norenza-­‐yan and Karon MacLean. Their questions, comments and suggestions have been tremen-­‐dously	
  helpful	
  in	
  making	
  this	
  thesis	
  more	
  readable	
  and	
  accurate.Several undergraduate RAs at UBC helped with these studies as well, in particular Chap-­‐ter 6: Pavel Kozik (who was also a co-­‐author of this chapter), Sean Povill, Sarah MacDon-­‐ald, Chris Yeh, Spencer Murch, K. L. Ta, and Igal Sterin all put in many thankless hoursrunning subjects and dealing with my various screw-­‐ups. Finally, the entire UBC VisionLab, past and present, deserves a great deal of thanks for repeatedly listening to mehalf-­‐coherently outline my latest results and future plans, and then asking me the ques-­‐tions	
  I	
  needed	
  to	
  hear	
  to	
  improve	
  my	
  thinking.Funding for this research came from a SSHRC Doctoral Fellowship to myself, a JamesMcDonnell Fellowship Centennial Award to Kathleen Akins, and NSERC DiscoveryGrants	
  to	
  Jim	
  Enns	
  and	
  Mark	
  Blair. xi Dedication This work is dedicated to my wife Cho. Without her love and support I would never havebeen able to survive almost ten years of graduate school, and our children William andJulia would not have developed into the happy people they are today. Here’s to the nextdecade,	
  and	
  the	
  ones	
  after	
  that.	
   Love,	
   M xii 1	
  	
  Introduction I cannot express it better than to say that a colored idea appears to[me]. [...] Particularly those things which form a simple series; e.g. numbers, the days of the week, the time periods of history and of human life, the letters of the alphabet, intervals of the musical scale,and	
  other	
  such	
  similar	
  things,	
  adopt	
  these	
  colors.These introduce themselves to the mind as if a series of visible ob-­ jects	
  in	
  dark	
  space,	
  formless	
  and	
  noticeably	
  of	
  different	
  colors.[...]In the alphabet, A and E are vermillion, A however is morecinnabar, E is more inclined to rose; I is white; O orange; U black; Ue (ü) gray; C pale-­‐ash-­‐colored; D yellow; F dark gray; H is bluishash-­‐colored; K nearly dark green (uncertain); M and N white; Sdark-­‐blue;	
  W	
  brown.(Jewanski,	
  Day,	
  &	
  Ward,	
  2009,	
  pp.	
  297-­‐298,	
  italics	
  in	
  original)This quotation comes from the >irst published account of synaesthesia, from the doctoraldissertation of George Tobias Ludwig Sachs in 1812, but it could easily have been writ-­‐ten by any of millions of people alive today. The >inal paragraph speci>ically describeshis grapheme-­colour synaesthesia, in which individuals experience letters or numbers ashaving colours. There are numerous other varieties, including such oddities as swim-­‐ming styles that have colours (Nikolic, Jurgens, Rothen, Meier, & Mroczko, 2011), wordsthat have tastes (Cytowic, 1993), calendars and number sequences that lie along convo-­‐luted three-­‐dimensional paths in one’s personal space (Sagiv, Simner, Collins, Butter-­‐worth, & Ward, 2006), letters and numbers that have well-­‐de>ined personalities andgenders (Amin et al., 2011), music that has colour and texture (Head, 2006; Ward, 1 Tsakanikos, & Bray, 2006), and even orgasms that have colours (cf. Novich, Cheng, & Ea-­‐gleman,	
  2011).	
  In the century after Sachs’ account was published, synaesthesia became a popular topicof research, attracting the attention of several important >igures in early psychology (e.g.Binet & Philippe, 1892; Calkins, 1893; Claparède, 1900; Flournoy, 1892; Galton, 1883).It fell out of favour for much of the 20th century, likely because a condition in which in-­‐dividuals describe unusual internal states without obvious behavioural correlates wasincoherent according to the dominant behaviourist framework. Publications on synaes-­‐thesia slowed to a trickle (with notable exceptions such as Marks, 1975) from themid-­‐1930’s until Richard Cytowic's work in the 1980’s (Cytowic, 1988, 1989a, 1989b;Cytowic & Wood, 1982a, 1982b). His push to bring synaesthesia back to the scienti>icmainstream came at exactly the right time, and other researchers slowly began to pickup the topic. Behaviourism was long gone, there was a renewed interest in consciousstates, and researchers could tackle synaesthesia with the new tools of cognitive neuro-­‐science, and with new insights from the study of sensory development and other unusu-­‐al conditions such as autism or phantom limb syndrome (e.g. Baron-­‐Cohen & Harrison,1997; Baron-­‐Cohen, Wyke, & Binnie, 1987; Maurer, 1993; Paulesu et al., 1995; Ra-­‐machandran & Rogers-­‐Ramachandran, 1996). Since the turn of the millennium, the>loodgates have truly opened, with research groups all over the world studying every as-­‐pect of synaesthesia using the full range of tools and methodologies of modern cognitivescience,	
  producing	
  almost	
  500	
  publications	
  over	
  the	
  past	
  ten	
  years.My collaborators and I have been a small part of this >lood of new research for the pastsix years, and this thesis brings together our work. I have been particularly interested inthe bidirectional relationship between synaesthesia and learning: how learning is a nec-­‐essary part of the development of synaesthesia, and how synaesthesia might itself beuseful for learning. This interest arose because both directions of this relationship arenecessary parts of a developmental learning hypothesis of synaesthesia developed in col-­‐laboration with Kathleen Akins and Lyle Crawford, which states that synaesthesia devel-­‐ops, at least in part, as a strategic aid to overcoming a number of learning challenges in 2 childhood (Watson, Akins, & Crawford, 2010). We initially thought of this as an entirelynew idea, but later discovered that important aspects of it had been sketched over acentury	
  ago	
  (Calkins,	
  1893).This thesis presents evidence for three critical components of the developmentallearning	
  hypothesis:1. Synaesthesia typically develops as part of a dif>icult learningprocess in which the synaesthete learns a category structurewhose members become the triggers of synaestheticexperiences.2. Synaesthetic experiences are shaped by this learning process,such that they encode a wide range of information about thelearned	
  domain.	
  3. Synaesthesia is exploited on a variety of memory, learning, andcreative tasks, leading to “synaesthetic styles” of performance onthese	
  tasks.The developmental learning hypothesis connects these three claims in a causal chain.Synaesthesia develops in response to various learning challenges (#1) because it is use-­‐ful for these challenges (#3). Synaesthetic experiences are shaped by various aspects ofthe learned domain (#2) because they developed as part of a learning strategy to ac-­‐quire	
  knowledge	
  of	
  this	
  domain	
  (#1	
  and	
  #3).Establishing this causal chain would prove the developmental learning hypothesis, butis well beyond the scope of this thesis, as it would require a comprehensive set of devel-­‐opmental studies. Rather, my collaborators and I have provided new evidence for eachof	
  the	
  three	
  claims,	
  which	
  is	
  presented	
  in	
  the	
  research	
  chapters	
  of	
  this	
  thesis.In the remainder of this Introduction I want to give the reader the background and con-­‐text necessary to judge the work I present in the research chapters. I begin with anoverview of the current debates over how to de>ine and operationalize synaesthesia,and then turn to what other research has already established about the three claimsabout	
  synaesthesia	
  and	
  learning.	
   3 1.1	
  	
  What	
  is	
  synaesthesia? 1.1.1	
  	
  Terminology	
  and	
  deOinitionsResearchers usually use the term inducer to refer to the “trigger” of synaesthetic experi-­‐ences, and the term concurrent to refer to these unusual experiences themselves. ThusGeorge Sachs has letters as inducers and colours as concurrents. The different varietiesof synaesthesia are typically named using the formula inducer-­concurrent, and so wespeak of grapheme-­colour, music-­colour, or word-­taste synaesthesias (although this for-­‐mula may be falling out of favour, cf. “coloured sequence synaesthesia” from Novich,Cheng,	
  &	
  Eagleman,	
  2011).	
  Coloured	
  inducers	
  are	
  often	
  referred	
  to	
  as	
  photisms.The word synaesthesia means "union of the senses", and it was generally thought of inthese terms until quite recently. Typically, synaesthesia was de>ined as a case of unusualassociations between sensory modalities, so for example you might have the sense ofhearing (music) leading to visual experiences (colour). This cross-­‐modal de>inition wascommonplace despite the fact that grapheme-­‐colour synaesthesia, by far the most stud-­‐ied variety, blatantly contradicts it, as both graphemes and colours are visual. More re-­‐cently it has become commonplace to address this issue by complicating the de>initionslightly, e.g. stating that the concurrent experiences are “in another modality [or in] adifferent aspect of the same sensory modality” as the inducer (Asher et al., 2009, p.279), which, while more accurate, is so general as to be almost useless. (Are there con-­‐sistent	
  sensory	
  associations	
  which	
  are	
  not	
  synaesthetic,	
  according	
  to	
  this	
  de>inition?)Furthermore, neither synaesthetic inducers nor concurrents are necessarily sensory atall. Grapheme-­‐colour synaesthetes, for example, can experience colours correspondingto the answers of mathematical problems, even when these answers are nowhere in thephysical stimulus (Dixon, Smilek, Cudahy, & Merikle, 2000; Smilek, Dixon, Cudahy, &Merikle, 2002a). It is also universally acknowledged that colours for graphemes do nottypically vary with font or case (although there may be atypical synaesthetes in this re-­‐spect, cf. Ramachandran & Hubbard, 2001a), suggesting that the inducer is not a simple 4 sensory stimulus (consider, e.g., how little the shapes of A and a have in common). Sev-­‐eral researchers now explicitly reject the notion that there need be any straightforwardsensory aspects to synaesthesia at all, arguing that it is generally triggered by “higher-­‐level” conceptual or linguistic constructs (e.g. Jürgens & Nikolic, 2012; Simner, 2012a).And in the case of varieties of synaesthesia such as graphemes with personalities, it isnot	
  clear	
  that	
  the	
  concurrent	
  is	
  sensory	
  either.	
  The original de>inition is clearly unsatisfactory, then, but nothing has arisen to take itsplace. Indeed, a recent exchange of papers has a number of the leading researchers inthe >ield agreeing with each other that we do not know how to de>ine synaesthesia (Co-­‐hen Kadosh & Terhune, 2012; Eagleman, 2012; Simner, 2012a; Simner, 2012b). I wouldargue that this is a feature, not a bug, of modern synaesthesia research. Synaesthesia is arelatively rare and still rather poorly-­‐understood phenomenon, and we run far less riskof needless errors if we do not impose arti>icial restrictions on what does and does notcount as “real” synaesthesia, instead letting the data itself slowly shape our categoryboundaries. 1.1.2	
  	
  Operationalizing	
  synaesthesiaWhile there is no accepted conceptual de>inition of synaesthesia, there are more-­‐or-­‐lesswidely-­‐accepted criteria for establishing that a given person is or is not synaesthetic forthe purposes of research. There are three broad types of such operationalizing criteria:tests	
  of	
  self-­report,	
  consistency,	
  and	
  automaticity.The self-­‐report criterion is the most most basic, indicating merely that participants self-­‐identify as having synaesthetic experiences. That is, when asked questions like thosefrom Appendix 1, e.g. “When you see, hear, or think about certain letters or numbers, doyou see or feel any colours?”, people who answer “no” are not typically considered assynaesthetes. Some criterion of self-­‐identi>ication is almost universal in synaesthesia re-­‐search, in fact it is so standard that it is frequently not explicitly mentioned by re-­‐searchers. Nevertheless, it is a crucial test that eliminates roughly 80% of the popula-­‐ 5 tion from consideration as a synaesthete (see Chapter 2 for rates of self-­‐report ofsynaesthetic	
  experiences).The second criterion uses high performance on a consistency test of inducer-­‐concurrentassociations as a marker of synaesthesia. In these tests, participants are presented witha series of synaesthetic inducers, usually in random order, and are asked to report theirconcurrents for each one. At some later time, they are asked to perform the same taskagain, usually in a new order, and a measure of consistency is taken between responsesto the two tests. Those individuals who meet a certain threshold of consistency aredeemed to be synaesthetic. The speci>ic details vary quite widely. For example, consis-­‐tency tests for grapheme-­‐colour synaesthesia might ask participants to report the nameof the colour they experience for a given letter (Baron-­‐Cohen, Burt, Smith-­‐Laittan, Harri-­‐son, & Bolton, 1996), or to choose the best match to their concurrent from a small sam-­‐ple of colours (Simner et al., 2006), or to choose the speci>ic shade of their experiencefrom the >16,000,000 colours available on a standard computer monitor (Eagleman, Ka-­‐gan, Nelson, Sagaram, & Sarma, 2007). The test-­‐retest interval varies from a matter ofseconds (Eagleman, Kagan, Nelson, Sagaram, & Sarma, 2007) to many months (Simneret al., 2006). The precise consistency threshold and de>inition of consistency differ be-­‐tween studies as well. Typically, less than 5% of the population meets this more strin-­‐gent	
  requirement.Another common way of verifying synaesthesia is to test for the automaticity of the in-­‐ducer-­‐concurrent relationship. Typically, such tests employ a modi>ied Stroop paradigm,in which participants are asked to name either the physical colour of a stimulus or theirsynaesthetic colour for this stimulus, where the physical colour is either congruent orincongruent with the synaesthetic colour (Dixon, Smilek, Cudahy, & Merikle, 2000; Mills,Boteler, & Oliver, 1999). Usually there is a strong bene>it of congruency for responsetime. This is the least commonly-­‐used criterion of synaesthesia, likely because custom-­‐generating	
  the	
  stimuli	
  for	
  each	
  participant	
  is	
  time-­‐consuming.	
   6 Many researchers hope for a clear neurobiological operationalization of synaesthesia(e.g. Simner, 2012a), but for the time being none exists, and so we are left with the threecriteria of self-­‐report, consistency and automaticity of synaesthetic experiences. It is im-­‐portant to note that these criteria are independent of each other, not just logically butalso in practice. Thus a constant irritation for synaesthesia researchers is that the largemajority of individuals who report synaesthetic experiences do not meet the consisten-­‐cy criterion (see Chapter 2). Further, the automaticity criterion can be met without theself-­‐report criterion, as the implicit learning of novel letter-­‐colour associations can leadto performance differences on Stroop-­‐type tasks without any conscious awareness ofthese associations (Colizoli, Murre, Rouw, Karniel, & Witthoft, 2012; Kusnir & Thut,2012). After more extensive associative training over the course of weeks, participantsmay report experiences that sound somewhat similar to conscious synaesthetic concur-­‐rents, and this is associated with a stronger Stroop effect, but this has not been system-­‐atically explored beyond one study that did not directly address synaesthesia (MacLeod&	
  Dunbar,	
  1988).	
  There are theoretical and practical limitations to these criteria, then, but they are theonly ones we have. The studies reported here, like the majority of recent work onsynaesthesia, use the self-­‐report and consistency criteria: anyone who says they havesynaesthetic experiences and is able to consistently reproduce highly similar colours forthe same stimuli is treated as synaesthetic. Details of the speci>ic questions asked andconsistency	
  test	
  used	
  are	
  found	
  in	
  Chapter	
  2	
  and	
  Appendix	
  1. 1.1.3	
  	
  Sub-­types	
  of	
  synaesthesiaThere are varieties of synaesthesia that co-­‐occur in individuals more than others, lead-­‐ing Novich, Cheng, and Eagleman (2011) to identify >ive distinct sub-­‐types of synaesthe-­‐sia that cluster in this way. For example, any type of what Novich et al. refer to as coloured sequence synaesthesia—grapheme-­‐colour, weekday-­‐colour, etc—is much morelikely to co-­‐occur in one individual with another form of coloured sequence synaesthe-­‐sia, but is not nearly as likely to co-­‐occur with other clusters of synaesthesias, which in-­‐ 7 clude coloured sensations (e.g. touch-­‐colour or orgasm-­‐colour), spatial sequences (e.g.calendar or number forms in personal space), coloured music, and synaesthesias with non-­visual sequelae (e.g. word-­‐taste or sound-­‐smell). Nevertheless, individuals withcoloured sequence synaesthesias are far more likely to have, e.g., spatial sequences thanindividuals with no other forms of synaesthesia (Brang & Ramachandran, 2011; Sagiv,Simner, Collins, Butterworth, & Ward, 2006, see also Chapter 2). So synaesthesias ofwidely different types appear to be somewhat related, but there are at least >ive distinctsets of synaesthesias that are especially tightly linked together, for as-­‐yet unknown rea-­‐sons. This places a potentially important limit on the results in the research chapters ofthis thesis, as virtually all the analyses reported here deal with some form of colouredsequence	
  synaesthesia.Researchers have also tried to sub-­‐divide synaesthesia based on differences in phenom-­‐enological reports. In particular, there is a widely used distinction between projectorand associator synaesthetes , where projectors are supposed to experience their concur-­‐rents as spatially located outside the body, while associators are supposed to experiencethem without a location in particular. Several researchers have shown that these phe-­‐nomenological reports correlate with differences in performance (e.g. Dixon, Smilek, &Merikle, 2004) and neurophysiology (e.g. Rouw & Scholte, 2010). However other re-­‐searchers, including me, have had substantially more dif>iculty in establishing whichgroup their participants should be classi>ied in. I am confused by several of the ques-­‐tions on a supposedly clear questionnaire to distinguish between projectors and associ-­‐ators (Skelton, Ludwig, & Mohr, 2009), and after using it with a number of participants Ifound that they had the same confusion, and that responses were not consistent acrossrepeated presentations of the questionnaire. Thus I gave up attempting to classify myparticipants in this manner. This is not to imply that the classi>ication has no merit, sim-­‐ply that there is still work to be done before it is usable by all researchers. There is evi-­‐dence that the divisions run further than projector/associator, and that we should cat-­‐egorize at least four different phenomenologies of concurrents (Ward, Li, Salih, & Sagiv, 8 2007), but for the time being I do not differentiate projectors, associators, or any otherphenomenological	
  sub-­‐types	
  of	
  synaesthetes. 1.2	
  	
  Why	
  are	
  people	
  synaesthetic?If one wants to understand why there are synaesthetes, there are at least two inter-­‐twined questions that are really being asked. The more proximal question asks what itis about synaesthetes’ brains that causes inducers to give rise to synaesthetic concur-­‐rents: what is the neurophysiology of synaesthesia? The more distal question asks howit	
  is	
  that	
  such	
  brains	
  came	
  about:	
  how	
  does	
  synaesthesia	
  develop? 1.2.1	
  	
  The	
  synaesthetic	
  brainAll major theories of the neurophysiology of synaesthesia agree that inducers and con-­‐currents are represented by activation in distinct populations of neurons in bothsynaesthetes and non-­‐synaesthetes. What makes synaesthetes unusual, according tothese theories, is that inducer-­‐related activation in area A leads to concurrent-­‐relatedactivation in area B, a causal link that does not occur in non-­‐synaesthetes. There are es-­‐sentially	
  three	
  broad	
  types	
  of	
  theories	
  about	
  how	
  this	
  happens.	
  First, cross-­activation theorists propose that inducer areas have unusually strong directconnections to concurrent areas among synaesthetes, allowing inducer activations totrigger concurrent activations. Thus synaesthesia stems from a breakdown, or at least areduction, in neural modularity (Baron-­‐Cohen, Harrison, Goldstein, & Wyke, 1993; Mau-­‐rer, 1993; Ramachandran & Hubbard, 2001b). These theorists have particularly concen-­‐trated on grapheme-­‐colour synaesthesia, noting that the so-­‐called visual word form area(VWFA) in the fusiform gyrus is right next to the so-­‐called “colour area” V4, whichwould	
  mean	
  that	
  the	
  unusual	
  connectivity	
  in	
  synaesthesia	
  could	
  be	
  highly	
  localized.	
   Re-­entrant processing theories propose a two-­‐step connection, starting with an area thatprocesses sensory features of the inducer, which projects to an area that processes theconcepts or meanings associated with the inducer, which in turn back-­‐projects to senso-­‐ry areas that represent the concurrent (Smilek, Dixon, Cudahy, & Merikle, 2001). Thus a 9 critical difference between re-­‐entrant and cross-­‐activation models is that in the formerthe synaesthetic experience is determined by the meaning of the inducer rather than itssensory features. Once again, the model has been most thoroughly >leshed-­‐out with re-­‐gards to grapheme-­‐colour synaesthesia. As with the cross-­‐activation model, the inducerand concurrent areas in this case are presumed to be VWFA and V4, respectively, whilethe posterior-­‐inferior-­‐temporal region is suggested as the area representing graphemes’meanings	
  (Smilek,	
  Dixon,	
  Cudahy,	
  &	
  Merikle,	
  2001).	
  	
  Finally, disinhibited feedback theories propose that the connections between inducerand concurrent areas are no different between synaesthetes and non-­‐synaesthetes, butthat there are differences in the feedback from other areas that modulate the signalspassing from inducer to concurrent areas (Grossenbacher, 1997; Grossenbacher &Lovelace, 2001). This could either take the form of less inhibitory feedback or more ex-­‐citatory feedback, in either case making it possible for signals to pass from inducer toconcurrent	
  areas	
  that	
  would	
  otherwise	
  be	
  too	
  weak	
  to	
  do	
  so.	
  Thus far none of these models has been con>irmed by actual neurophysiological studies.As Hubbard, Brang, and Ramachandran (2011) point out, the cross-­‐activation theoryhas probably received the most direct empirical support, but this is patchy at best, andthe re-­‐entrant processing and disinhibited feedback models have not been as carefullyinvestigated. This is not to say that there is a lack of neurophysiological studies, just thattheir	
  results	
  do	
  not	
  map	
  neatly	
  on	
  to	
  any	
  of	
  the	
  theories.There have now been a number of functional neuroimaging studies showing increasedactivation of various brain areas during synaesthetic experiences (Aleman, Rutten, Sit-­‐skoorn, Dautzenberg, & Ramsey, 2001; Brang, Hubbard, Coulson, Huang, & Ramachan-­‐dran, 2010; Cohen Kadosh, Kadosh, & Henik, 2007; Hubbard, Arman, Ramachandran, &Boynton, 2005; Laeng, Hugdahl, & Specht, 2011; Nunn et al., 2002; Paulesu et al., 1995;Rich et al., 2006; Rouw & Scholte, 2010; Weiss, Zilles, & Fink, 2005), structural studiesshowing increased connectivity in various brain areas among synaesthetes (e.g. Banissyet al., 2012; Hänggi, Beeli, Oechslin, & Jancke, 2008; Rouw & Scholte, 2007; Weiss & 10 Fink, 2009), and EEG/MEG studies showing differences in the time course of neural ac-­‐tivity associated with synaesthesia (e.g. Barnett et al., 2008b; Brang, Hubbard, Coulson,Huang, & Ramachandran, 2010; Jäncke, Rogenmoser, Meyer, & Elmer, 2012). (For a re-­‐view of the brain areas associated with synaesthesia, see Rouw, Scholte, & Colizoli,2011.)There are two main ways in which these data fail to map easily on to theoretical predic-­‐tions. First, results are highly heterogeneous. While there has been some overlap, manydifferences exist between the various studies, for instance several studies have foundthat when synaesthetes with coloured concurrents are presented with their inducersthey show activation in V4 (e.g. Brang, Hubbard, Coulson, Huang, & Ramachandran,2010; Hubbard, Arman, Ramachandran, & Boynton, 2005; Nunn et al., 2002) but othershave found no such activation (e.g. Rich et al., 2006; Paulesu et al., 1995; Rouw &Scholte, 2010; Weiss, Zilles, & Fink, 2005). Second, many of the areas that do seem to bereliably associated with synaesthesia, such as the precentral gyrus (Laeng, Hugdahl, &Specht, 2011; Nunn et al., 2002; Paulesu et al., 1995; Rouw & Scholte, 2010; Weiss,Zilles, & Fink, 2005), and frontal-­‐parietal networks (Laeng, Hugdahl, & Specht, 2011;Rouw	
  &	
  Scholte,	
  2010),	
  are	
  not	
  predicted	
  by	
  any	
  of	
  the	
  three	
  models.The heterogeneity of results is to be expected, given both the wide variety of experimen-­‐tal tasks and tools used and the frequently small sample sizes employed. The hetero-­‐geneity of synaesthesia itself is also a serious issue: these studies employ different typesof synaesthetes, both in terms of their inducers and concurrents (e.g. grapheme-­‐colourvs. music-­‐colour), and in terms of their self-­‐reported phenomenology (some studiesseparate associators from projectors, others do not). Until a larger number of experi-­‐ments are run using consistent methods, it will be hard to conduct appropriate meta-­‐analyses	
  that	
  allow	
  for	
  these	
  inconsistencies	
  to	
  be	
  sorted	
  out.The activation of areas that are not predicted by any neurophysiological theory ofsynaesthesia is also not terribly surprising, and does not indicate that any of these theo-­‐ries	
  are	
  false.	
  Rather,	
  it	
  merely	
  shows	
  that	
  they	
  are	
  incomplete. 11 The developmental learning hypothesis is neutral with regards to the neurological un-­‐derpinnings of synaesthesia. However it is certainly compatible with an important rolefor attention and executive functions, as hinted at by the involvement of frontal-­‐parietalnetworks in synaesthetic experiences (Laeng, Hugdahl, & Specht, 2011; Rouw & Scholte,2010). 1.2.2	
  	
  The	
  synaesthetic	
  genomeWhatever the speci>ics of synaesthetic neurophysiology, how do synaesthetes’ brains getto be that way? A genetic explanation for synaesthetic development has been an attrac-­‐tive idea for many researchers. The developmental learning hypothesis is not in con>lictwith such explanations per se. However it would be hard to reconcile it with a simple ge-­‐netic cause of synaesthesia, in which a single gene or group of genes reliably causessynaesthesia to develop (such as is the case for, e.g., Huntington’s chorea), since, the de-­‐velopmental learning hypothesis includes an important role for learning in synaestheticdevelopment.Such simple genetic explanations have been relatively popular, however. There are prob-­‐ably two main reasons for this, one arising from theory and one from evidence. First, anumber of researchers have proposed a relatively simple genetic mechanism underlyingthe high degree of connectivity that is the basis of the cross-­‐activation and re-­‐entrantfeedback theories. They suggest that this connectivity may be present in all of us atbirth, but deteriorates in the standard process of neural pruning. Synaesthetes, on theother hand, may have a mutation in a gene that controls neural pruning, preventing itfrom occurring to the same extent as in non-­‐synaesthetes, leading to the unusualconnectivity (Baron-­‐Cohen, 1996; Maurer, 1993; Ramachandran & Hubbard, 2001b). Ofcourse this account requires an explanation of why the gene is only selectively ex-­‐pressed in particular regions of cortex, which may complicate the genetic storysomewhat.	
  The more evidence-­‐based reason why researchers began speculating about simple ge-­‐netic causes for synaesthesia comes from the results of familial studies. Since the earli-­‐ 12 est days of synaesthesia research, scientists have noted that it runs in families (Galton,1883) and appeared to be strongly linked to sex. One well-­‐cited study found a fe-­‐male:male ratio of 6:1, further >inding that these synaesthetes had a ratio of female:malefamily members (synaesthetic or not) of 8:1 (Baron-­‐Cohen, Burt, Smith-­‐Laittan, Harri-­‐son, & Bolton, 1996). Furthermore, almost all reports of synaesthesia within families in-­‐volve the trait being passed along the maternal line (Barnett et al., 2008a; Baron-­‐Cohen,Burt, Smith-­‐Laittan, Harrison, & Bolton, 1996; Ward & Simner, 2005). Such skewed ra-­‐tios require explanation, and a popular hypothesis was that synaesthesia might be an x-­‐linked dominant trait with lethality in males (Bailey & Johnson, 1997; Baron-­‐Cohen,Burt, Smith-­‐Laittan, Harrison, & Bolton, 1996), which might explain both the female biasamong	
  synaesthetes	
  and	
  the	
  lack	
  of	
  male	
  family	
  members	
  of	
  synaesthetes.Two studies effectively ended speculation about the lethality in males of any putative“synaesthesia gene”, using much larger samples of families, and >inding no difference inthe number of male and female family members of synaesthetes (Barnett et al., 2008a;Ward & Simner, 2005). Both studies still found a larger number of female than malesynaesthetes, in one case a ratio of 6:1 (Barnett et al., 2008a), and in the other a ratio of2:1 (Ward & Simner, 2005). The 2:1 ratio was smaller than previous estimates, andthere was evidence that even this lower ratio was likely too high due to systematic un-­‐der-­‐reporting of synaesthesia by men. A later, better-­‐controlled and larger, study (Simn-­‐er et al., 2006) found no evidence of a female bias at all, and it was argued that the fe-­‐male bias in previous results was largely, if not solely due to differences in response andcompliance biases between the sexes (see Chapter 2 for a more complete account of thisissue). However it has never yet been >irmly established that these response and com-­‐pliance	
  biases	
  exist.It seems unlikely, then, that there is a simple x-­‐linked genetic cause of synaesthesia,however it is clear that it does run in families (Barnett et al., 2008a; Baron-­‐Cohen, Burt,Smith-­‐Laittan, Harrison, & Bolton, 1996; Ward & Simner, 2005), suggesting that a genet-­‐ic component of some kind is at play. Direct comparisons of DNA between synaesthetesand non-­‐synaesthetes have found several candidate chromosonal regions (Asher et al., 13 2009; Tomson et al., 2011), but these differ between studies, and the same genetic fac-­‐tors are not present in all synaesthetes within either study, suggesting that the geneticin>luence	
  on	
  synaesthesia	
  is	
  highly	
  polygenic	
  and	
  variable.	
  	
   1.2.3	
  	
  The	
  crucial	
  role	
  of	
  learning	
  in	
  synaesthetic	
  developmentMost common synaesthetic inducers are culturally transmitted by processes that in-­‐volve considerable time and effort on the part of the learner, and often formal instruc-­‐tion. In the >irst paragraph of this thesis I mentioned swimming styles, words, calendars,letters and numbers, music, and orgasms, all but the last of which is clearly learned (andeven there one might debate the point). This bias towards learned inducers is not coin-­‐cidental, and is found in more formal analyses. For example, of the >ive sub-­‐groups ofsynaesthesia identi>ied by Novich and colleagues (2011) that were described in the pre-­‐vious section, three of them, which constitute the large majority of cases, exclusively in-­‐volve learned inducers. Day (2005) lists several dozen types of synaesthesia, includingmany examples where the inducers are not obviously learned, and certainly not learneddeliberately or via formal instruction (e.g. orgasms, smells/tastes, personalities, envi-­‐ronmental sounds, temperature). However >ive of the six most prevalent types ofsynaesthesia on Day's list involve learned inducers, and these account for the vast ma-­‐jority of his cases. Large-­‐scale surveys of synaesthesia (Rich, Bradshaw, & Mattingley,2005; Simner et al., 2006 see also Chapter 2 of this thesis) also show an overwhelmingmajority	
  of	
  cases	
  involving	
  learned	
  inducers.A simple conclusion follows from this: synaesthesia normally only develops as part of a formal learning process. The need to explain this places a serious constraint upon genet-­‐ic and neurological theories of synaesthetic development. I agree with Cytowic and Ea-­‐gleman (2009) that so far none of these theories even attempts to do so, because re-­‐searchers have not generally acknowledged this crucial role of learning, at least not tothe the extent that it actually informs their theories (for notable and welcome excep-­‐tions to this trend, see, e.g. Simner, Harrold, Creed, Monro, & Foulkes, 2009; Witthoft &Winawer,	
  2013). 14 There is only one published account that directly studies the development of synaesthe-­‐sia in children (Simner, Harrold, Creed, Monro, & Foulkes, 2009). Here a large number(N = 615) of children ages 6-­‐7 were given a modi>ied version of a letter-­‐colour consis-­‐tency test that required them to select a colour for each letter twice, and then a year lat-­‐er were given the same test. Synaesthetes were identi>ied as those who showed a highdegree of consistency in their colours both within each test and across both tests. Therewas a clear developmental trajectory here: synaesthetes' mean number of consistently-­‐coloured	
  letters	
  was	
  approximately	
  11	
  on	
  the	
  >irst	
  test	
  and	
  16	
  on	
  the	
  second.These results demonstrate that synaesthetic associations coalesce over a lengthy periodtime that roughly coincides with the development of reading and writing. At 6 years oldthe average grapheme-­‐colour synaesthete has consistent colours for less than half theletters in the alphabet. One year later, this rises to slightly over half. Clearly, these chil-­‐dren have a long way to go before they reach the consistency levels of adult synaes-­‐thetes,	
  who	
  frequently	
  have	
  100%	
  consistent	
  colours.Like Simner et al. (2009), Cytowic and Eagleman (2009, Table 2.2) suggest that the de-­‐velopment of synaesthesia coincides with the development of literacy, but they focus onan earlier stage, namely when children >irst start learning their letters (generally from34-­‐48 months). If grapheme-­‐colour synaesthesia begins to develop with the >irst acqui-­‐sition of letters, then given Simner et al.’s (2009) results, the development of grapheme-­‐colour synaesthesia is slow indeed, taking shape over the course of at least six years,and likely much longer. As we will see in the next section, there is evidence that this isexactly what happens, with different stages of learning about letters affecting the devel-­‐opment	
  of	
  synaesthesia,	
  leaving	
  behind	
  traces	
  in	
  the	
  synaesthetic	
  colours	
  themselves. 1.3	
  	
  How	
  does	
  learning	
  change	
  synaesthesia?We have just seen that synaesthesia generally only develops as part of an explicitlearning process, and that this development takes years. What do we know of the in>lu-­‐ences	
  on	
  this	
  development?	
   15 1.3.1	
  	
  Synaesthetic	
  concurrents	
  as	
  fossils	
  of	
  learning	
  processesIn terms of direct observation, virtually nothing. Simner and colleagues are followingthe progress of the synaesthetes identi>ied in their childhood study (Simner, Harrold,Creed, Monro, & Foulkes, 2009), but as of yet no further data have been published. How-­‐ever there are a number of papers that describe colour regularities found across adultsynaesthetes, particularly grapheme-­‐colour synaesthetes, and I argue that many ofthese regularities can be thought of as perceptual/cognitive "fossils": traces laid down asa result of the in>luences on synaesthetic development. Like real fossils, these can tell usa	
  great	
  deal	
  about	
  the	
  environments	
  in	
  which	
  they	
  were	
  formed. 1.3.2	
  	
  Semantic	
  inOluencesOne class of these involves cases where inducers and concurrents have a common se-­‐mantic content. For example, G is often green for English grapheme-­‐colour synaesthetes,and in general the >irst letters of common colour words are often associated with thecolours named by these words (Barnett et al., 2008a; Rich, Bradshaw, & Mattingley,2005; Simner et al., 2005; Simner, Harrold, Creed, Monro, & Foulkes, 2009). Similarly, Dis often brown, which may re>lect the fact that D is often taught to English speakers asthe >irst letter of "dog", an animal that is stereotypically brown (Rich, Bradshaw, & Mat-­‐tingley,	
  2005).	
  Some of these associations may not be learned until quite late in life. For instance onesynaesthete reports that after learning the meaning of the word phthalocyanine (a typeof blue-­‐green dye), the colour of the letters within it changed from largely purple andpink to blue and green (Curtis, 1998). (This colour change was only within the contextof the word itself -­‐ one complicating factor that this thesis ignores entirely is that wordsoften have their own colours that are somewhat independent of the colours of the let-­‐ters	
  making	
  them	
  up.) 16 1.3.3	
  	
  Common	
  associative	
  inOluencesAnother in>luence on the development of synaesthesia is standard associative learning:some inducer-­‐concurrent associations are formed as a result of the synaesthete beingexposed to these associations in the environment. The letter and number colours ofsome grapheme-­‐colour synaesthetes are derived from those found on toys they playedwith as young children (Hancock, 2006; Witthoft & Winawer, 2006; Witthoft & Winawer,2013), others report that they are identical to those used on the wall of their kinder-­‐garten (Colizoli, Murre, Rouw, Karniel, & Witthoft, 2012), and other childhood associa-­‐tions have been noted for many years (Calkins, 1893). It should be noted that such easi-­‐ly-­‐determinable associations are relatively rare in the literature, and often attempts to>ind them result in failure. For example, an in-­‐depth look at the colours of letters in chil-­‐dren’s books published in Australia between 1862-­‐1989 failed to >ind evidence of astrong connection to the letter colours of a large sample (N=150) of synaesthetes whogrew up there, although there was some evidence that number colours had been in>lu-­‐enced by the colours used in a popular method of math teaching during the 1950’s and1960’s	
  (Rich,	
  Bradshaw,	
  &	
  Mattingley,	
  2005).	
   1.3.4	
  	
  Universal	
  inOluences?There are also common letter-­‐colour associations that may be universal among bothsynaesthetes and non-­‐synaesthetes. For example, I and O are most often white forsynaesthetes, and X is often black (Simner et al., 2005), and these shape-­‐colour corre-­‐spondences are found among non-­‐synaesthetes as early as age 2, long before they havestarted	
  reading	
  (Spector	
  &	
  Maurer,	
  2008;	
  Spector	
  &	
  Maurer,	
  2011).	
   1.3.5	
  	
  Second-­order	
  inOluencesThe semantic, associative, and possibly innate in>luences on synaesthetic inducers de-­‐scribed above are all examples of Oirst-­order mappings, in which a single element of onedomain is related to a single element of another. For example, the individual letter G isrelated to a particular shade of green. These are to be distinguished from second-­order 17 mappings, “relations between relations”, which, as Chapter 3 explains, are also found insynaesthesia. Here it is not a single element being mapped from one domain to a singleelement within another domain, but rather a pattern or relationship within one domainbeing mapped to a relationship within another domain. For example, synaesthetes havea general tendency to associate similarly-­‐shaped letters, such as E and F, with similarcolours (Brang, Rouw, Ramachandran, & Coulson, 2011; Eagleman, 2010; Jürgens &Nikolic, 2012, see also Chapters 3-­‐5). Here, a relationship of similarity within the do-­‐main of shape is mapped on to a relationship of similarity within the domain of colour.These mappings can exist independently of >irst-­‐order relations -­‐ thus E might haveentirely different colours for different synaesthetes, but if its colour for one individual issimilar to F's colour for the same individual, and so on for other synaesthetes, thenthere	
  is	
  a	
  second-­‐order	
  relationship	
  between	
  shape	
  and	
  synaesthetic	
  colour	
  similarity.There are also second-­‐order pitch-­‐luminance and pitch hue mappings in music-­‐coloursynaesthesia, such that higher pitches tend to be associated with brighter colours(Marks, 1975) and quartertones tend to be associated with colours that are closer to themidpoint of the two adjacent semitones (Head, 2006). Letters and numbers that aremore frequently seen in print tend to be associated with brighter colours (Beeli, Esslen,& Jäncke, 2007; Cohen Kadosh, Henik, & Walsh, 2007; Simner & Ward, 2008; Smilek,Carriere, Dixon, & Merikle, 2007), more saturated colours (Beeli, Esslen, & Jäncke,2007), and colours whose names are more commonly used (Rich, Bradshaw, & Matting-­‐ley, 2005; Simner et al., 2005). Letters that appear earlier in the alphabet tend to havemore distinct colours from each other than do letters that appear later in the alphabet(Eagleman, 2010, see also Chapter 3). Finally, a number of new second-­‐order >indingsare	
  detailed	
  in	
  Chapters	
  3-­‐5.Synaesthetic concurrents, then, contain numerous traces of the factors that in>luencedtheir development. To put it another way, synaesthetic concurrents encode a wide varie-­‐ty of information about their inducers, both >irst-­‐order information about speci>ic in-­‐ducers and second-­‐order information about the relationships between these inducers.Some of this information is learned right at the start of literacy development (e.g. the 18 shape relationships between letters), some of it quite a bit later (the most extremeexample presented here being the colour change due to learning the meaning of ph-­‐thalocyanine,	
  but	
  Chapter	
  5	
  will	
  provide	
  further	
  examples).	
   1.4	
  	
  Is	
  synaesthesia	
  good	
  for	
  anything?In popular culture, synaesthesia is often portrayed as a superpower, sometimes quiteliterally, (Anonymous, n.d.). It seems intuitive to many people that the kinds of experi-­‐ences synaesthetes describe ought to be useful in some way. Speaking somewhat moreformally, synaesthetes' experiences of their inducers are different from non-­‐synaes-­‐thetes, in that they have additional associations with these inducers that the rest of usdo not. Do these additional experiences provide any bene>it to the synaesthete? Re-­‐searchers have suggested that this is the case for well over a century (Calkins, 1893),but very little controlled research was conducted into the utility of synaesthesia untilthe	
  turn	
  of	
  the	
  millennium. 1.4.1	
  	
  Synaesthesia	
  and	
  memoryAnecdotal reports of synaesthesia's utility for memory are common. Synaesthetes oftenreport using their concurrents in everyday life to assist with remembering names, tele-­‐phone numbers or the spellings of words (Chapter 2 in Cytowic, 2002, provides severalexamples), and they tend to self-­‐report better than average memories (Yaro & Ward,2007). Several savants also report that their synaesthesia is an integral part of their as-­‐tounding recall, allowing them to memorize pi to over 20,000 decimal places, perfectlyrecall a list of random words on a surprise test 20 years after the initial encoding, ormemorize several 50-­‐digit matrices in a matter of minutes, retaining them for months(respectively, Bor, Billington, & Baron-­‐Cohen, 2007; Luria, 1968; Smilek, Dixon, Cudahy,& Merikle, 2002b). Research has con>irmed that synaesthesia is associated with mildlyenhanced memory for certain stimuli, such as lists of words, but not generally with trulyexceptional abilities (Rothen, Meier, & Ward, 2012 provide a detailed overview of thework on synaesthesia and memory), and there is some evidence that this enhancementis due to the synaesthetic concurrents themselves -­‐ e.g. synaesthetes' memory for let-­‐19 ters can be impaired if these letters are physically coloured incongruently with theirsynaesthetic colour (Radvansky, Gibson, & McNerney, 2011). Thus while synaesthesiamay be an integral part of certain exceptional memory abilities, such abilities are not anintegral	
  part	
  of	
  synaesthesia.At present it is unclear how much of the mild memory bene>it associated with synaes-­‐thesia is due to the synaesthetic experiences themselves. Studies initially found thatgrapheme-­‐colour synaesthetes have enhanced memory for colours (Yaro & Ward, 2007)and that number-­‐form synaesthetes have enhanced visuospatial memories (Simner,Mayo, & Spiller, 2009), but that these memory advantages were domain-­‐speci>ic suchthat, e.g., number-­‐form synaesthetes did not have any enhancement for colour memory.The suggestion was that this would give synaesthetes a memory advantage for thosestimuli that induce synaesthesia, but not for other stimuli. However the overall evidencefor this is murky at best, as memory advantages can be found on stimuli that do not in-­‐duce synaesthesia, and some stimuli that do induce synaesthesia do not have a corre-­‐sponding	
  memory	
  bene>it	
  (Rothen,	
  Meier,	
  &	
  Ward,	
  2012).	
  	
   1.4.2	
  	
  Synaesthesia	
  and	
  creativityA large number of artists are self-­‐reported synaesthetes. There are, for instance, de-­‐tailed accounts of the synaesthesia of the painters Vassily Kandinsky (Ione, 2004) andDavid Hockney (Cytowic, 2002); the composers Alexander Scriabin (Peacock, 1985) andOliver Messiaen (Bernard, 1986); the novelist Vladimir Nabokov (Nabokov, 1989); andthe pop/rap musicians Pharrell Williams (Seaberg, 2012) and Kanye West (Anonymous,2011). Many less famous synaesthetes also offer anecdotal reports of unusually strongartistic	
  interest	
  or	
  ability.	
  This connection between synaesthesia and creativity has some empirical corroboration.The rate of synaesthesia among >ine arts students is 3-­‐4 times as high as that in the gen-­‐eral population (Rothen & Meier, 2010b), and synaesthetes are much more likely to beemployed in artistic professions than non-­‐synaesthetes (Rich, Bradshaw, & Mattingley,2005; Ward, Thompson-­‐Lake, Ely, & Kaminski, 2008). Furthermore, synaesthetes tend 20 to score unusually high on the Remote Associates Test, a common measure of creativity(Ward, Thompson-­‐Lake, Ely, & Kaminski, 2008), and members of the general populationwho score highly on this test also tend to have unusually high consistencies on tests ofassociations between colour and tones, vowel sounds, and emotional words, althoughno formal test of synaesthesia was used in this study (Dailey, Martindale, & Borkum,1997).	
  As with the memory studies, however, care should be taken not to over-­‐interpret theseresults: synaesthetes do not out-­‐perform controls on all measures of creativity, andtheir performance is higher than average, but not extraordinarily so. Once again, it is asyet unclear whether synaesthetic experiences themselves are actually exploited inenhanced creativity, or whether synaesthetes are simply more involved in the arts be-­‐cause their experiences of, e.g., unusual colours lead them to be interested in colour forits	
  own	
  sake	
  	
  (Ward,	
  Thompson-­‐Lake,	
  Ely,	
  &	
  Kaminski,	
  2008). 1.4.3	
  	
  Looking	
  for	
  synaesthetic	
  styles	
  of	
  performanceThere is some evidence, then, that synaesthesia contributes to enhanced memory, artis-­‐tic abilities, and other skills, albeit usually to a moderate extent. However synaesthesiaserving a useful function does not necessarily entail that synaesthetic performance oneveryday tasks is superior to that of non-­‐synaesthetes. Rather, it might be that synaes-­‐thesia enables different methods of learning, creative production, and so forth, but thatnon-­‐synaesthetic methods may be just as effective. For instance, many synaesthetes re-­‐port that their synaesthesia contributes to their memory, but others do not, and there isno difference on standard memory tasks between these two groups (Rothen & Meier,2010a).The question, then, is not whether synaesthetes are better than non-­‐synaesthetes at agiven task, but rather whether synaesthetes perform this task in a unique manner thatexploits their synaesthesia. In this case synaesthesia would still be useful, but not neces-­‐sarily	
  superior. 21 Anecdotal reports of a "synaesthetic approach" to memory, learning, or creative endeav-­‐ours	
  are	
  extremely	
  common.	
  For	
  example,Speci>ic numbers interact with other numbers in different waysand their personalities meshed in speci>ic ways as well. When itdidn’t mesh or color correctly, I knew that I had a) done somethingwrong or b) had seen a new kind of math. For example, a word like‘pottery’ has some of its personality from the double ‘t’ and ‘y’. If Ileft off a ‘t’ or ‘y’, or spelled it with an ‘ie’, the personality was off.[...] With music, I could easily tell if a note was off on my viola or ifa classmate played the wrong note. It was just the personalities ofthe	
  notes.	
  (J.	
  Alsaied,	
  personal	
  communication,	
  March	
  18,	
  2013)Another compelling anecdote comes from a synaesthete who reports that when shetried to learn the piano, she discovered that she had three different and inconsistentsets of colours for letters (i.e. the names of the notes), for each of her >ingers, and formusical pitches, which made it entirely impossible for her to succeed (Pautzke, 2010).More generally, many of the artists noted previously report using their synaesthesia intheir artwork (Anonymous, 2011; Bernard, 1986; Cytowic, 2002; Ione, 2004; Pautzke,2010), and as previously described many individuals report using their synaesthesia tohelp	
  with	
  memory	
  and	
  learning	
  (cf.	
  Cytowic,	
  2002).There is little experimental evidence for different synaesthetic approaches to tasks, butthis is unsurprising, as very little research into this area has been undertaken. Someearly work provided in-­‐depth descriptions of a synaesthete's unique approach to a vari-­‐ety of memory and learning tasks, but this was using the method of introspection andwas never veri>ied in any other way (Wheeler & Cutsforth, 1921, 1922, 1925). More re-­‐cently, a synaesthetic savant was shown to actually perform worse than controls at re-­‐membering a matrix of numbers when they were coloured incongruently with hersynaesthetic experiences (Smilek, Dixon, Cudahy, & Merikle, 2002b). One should notethat this congruency effect is not universal: it was observed in a group of child synaes-­‐thetes (Green & Goswami, 2008) but was not observed with larger groups of synaes-­‐thetes whose memory for digits is only slightly above average (Yaro & Ward, 2007) orno higher than non-­‐synaesthetes (Rothen & Meier, 2009). One interesting interpretation 22 of these results is that synaesthesia can be exploited for unusual performance, but neednot be. This suggests that if synaesthesia is useful, as the developmental learning hy-­‐pothesis	
  supposes,	
  it	
  may	
  be	
  useful	
  in	
  different	
  ways	
  for	
  different	
  synaesthetes. 1.5	
  	
  An	
  outline	
  of	
  this	
  thesisEach of the research chapters of this thesis is devoted to providing further evidence forone	
  of	
  the	
  three	
  points	
  introduced	
  earlier:1. Synaesthesia typically develops as part of a dif>icult learningprocess in which the synaesthete learns a category structurewhose members become the triggers of synaestheticexperiences.2. Synaesthetic experiences are shaped by this learning process,such that they encode a wide range of information about thelearned	
  domain.	
  3. Synaesthesia is exploited on a variety of memory, learning, andcreative tasks, leading to “synaesthetic styles” of performance onthese	
  tasks.I take it that point 1, while not always acknowledged, is uncontroversial: virtually allsynaesthetic inducers are only acquired through a lengthy process of learning, usually ina classroom. However Chapter 2, which presents the largest survey on synaesthetic ten-­‐dencies yet performed, and the >irst covering more than one linguistic environment,demonstrates that learning's role in the development of synaesthesia is far strongerthan previously known. I show that second language learners who are bilingual from in-­‐fancy are three time less likely to develop synaesthesia (of any kind) than those wholearn a second language in grade school. Thus the age at which inducers are learned andthe manner in which learning takes place can determine whether or not one developssynaesthesia, or at least whether one maintains it into adulthood. Furthermore, sincethis determination occurs in grade school, this pushes the development of synaesthesiaquite a bit later than is generally assumed. This chapter also establishes that the previ-­‐ously-­‐reported gender bias in synaesthesia is almost certainly entirely due to differ-­‐ences in reporting rates and compliance with experimental protocols, rather than to ac-­‐ 23 tual differences in synaesthetic experiences, which further weakens one of the originalmotivations	
  for	
  assuming	
  a	
  strong	
  genetic	
  cause	
  for	
  synaesthesia.There is fairly extensive evidence for point 2, and we have already reviewed the ways inwhich synaesthetic concurrents are modi>ied by various learned characteristics of theinducing domain. This evidence, however, has generally been presented in a piecemealfashion, with most research focussing on one or two ways in which inducer characteris-­‐tics map on to concurrent characteristics, almost always from a >irst-­‐order perspective.In Chapters 3-­‐5 I simultaneously examine multiple second-­‐order in>luences ongrapheme-­‐colour synaesthetes' colours, showing how different aspects of these coloursare responsive to different relationships between letters, and how these in>luences areindependent of each other. Chapters 3 (previously published as Watson, Akins, & Enns,2012) and 4 outline these effects among two different groups of English synaesthetes,while Chapter 5 shifts the focus to Czech grapheme-­‐colour synaesthesia, showing thatCzech synaesthetic colours are highly responsive to a extremely detailed set of charac-­‐teristics of their graphemes. The Czech data also shows that quite sophisticated knowl-­‐edge about these graphemes affects their synaesthetic colours, showing that the in>lu-­‐ence of learning about letters on the development of synaesthesia must continue wellinto	
  the	
  primary	
  school	
  years.As we have seen, there is a great deal of anecdotal evidence for point 3, but less directsupport for it in the laboratory. Chapter 6 outlines a categorization learning study (pre-­‐viously published as Watson, Blair, Kozik, Akins, & Enns, 2012) showing not only thatsynaesthetes perform differently from non-­‐synaesthetes looking at the same stimuli(and far better than these non-­‐synaesthetes), but also that these differences in perfor-­‐mance arise because the synaesthetes exploit their synaesthetic colours to succeed, andnot because of general group differences such as motivation or memory advantages.This is also the >irst time that synaesthesia has been shown to be useful for a dif>icultlearning	
  task	
  that	
  involves	
  consciously	
  coordinating	
  multiple	
  pieces	
  of	
  information. 24 Finally, the Conclusion will attempt to tie all these strands back together, and assess theplausibility of the developmental learning hypothesis of synaesthesia in light of the newevidence	
  I	
  have	
  presented. 25 2	
  	
  Childhood	
  learning	
  and	
  rates	
  of	
   synaesthesia—The	
  prevalence	
  of	
   synaesthesia	
  in	
  Czech	
  and	
  English1 2.1	
  	
  IntroductionIf synaesthesia develops in response to childhood learning challenges, then some differ-­‐ences in these challenges ought to correspond with differences in the development ofsynaesthesia. One way this might manifest itself would be if synaesthesia were more (orless) likely to develop in children who face a particular type of learning challenge. In thiscase, one would expect to >ind differences in rates of synaesthesia in adults that corre-­‐spond to differences in these challenges. The present study looks for evidence of exactlythis, by measuring rates of synaesthesia among adults and seeing if these rates are asso-­‐ciated with differences in the particular learning challenges these adults had faced aschildren. Of course since synaesthesia is a relatively rare condition, testing this requiresa fairly large study. Thus it was that over the course of four years we found ourselvesconducting by far the largest study of synaesthetic tendencies yet performed (N = 1. The data presented in this chapter and Chapters 4 and 5 was collected over several years at Simon Fraser University and Charles University in the Czech Republic. I participated in the research design from the start, in roughly equal collaboration with Kathleen Akins and Lyle Crawford. Data collection was coordinated by Jan Chromý at Charles University and by myself, Kathleen Akins, Lyle Crawford and a number of research assistants at SFU. I wrote the chapters myself and performed all analyses solo, with helpful suggestions and input from collaborators. 26 11,664) at Charles University in Prague, Czech Republic, and Simon Fraser University inBurnaby,	
  British	
  Columbia. 2.1.1	
  	
  The	
  development	
  of	
  literacy	
  and	
  the	
  development	
  of	
  synaesthesiaThe particular learning challenge, or more accurately group of challenges, of interest inthis study is that associated with becoming literate. We know that grapheme-­‐coloursynaesthesia begins developing prior to age 6, and continues to develop after age 8(Simner, Harrold, Creed, Monro, & Foulkes, 2009), meaning that much, if not all, of itsdevelopment overlaps with the development of literacy. We also know that synaestheticcolours are in>luenced by a wide range of learned properties of letters (Beeli, Esslen, &Jäncke, 2007; Brang, Rouw, Ramachandran, & Coulson, 2011; Eagleman, 2010; Jürgens &Nikolic, 2012; Simner et al., 2005; Simner & Ward, 2008; Smilek, Carriere, Dixon, &Merikle, 2007; see also Chapters 3-­‐5), meaning that the development of grapheme-­‐colour synaesthesia is in>luenced by the development of literacy. What is unclear is howfar this in>luence goes. Is it merely that speci>ic grapheme-­‐colour associations arechanged in the course of becoming literate, as is already known, or is there a deeper in-­‐>luence, such that factors that change the process of becoming literate can change thelikelihood of developing grapheme-­‐colour synaesthesia in the >irst place? This study ex-­‐plores	
  the	
  second	
  possibility.Differences in learning might affect rates of synaesthesia in a number of ways. We offertwo competing hypotheses for how this might occur, the complexity and simplicity hy-­ potheses. The complexity hypothesis states that synaesthetic associations are more like-­‐ly to develop when children are faced with more complex or dif>icult tasks, because theywould be most useful when task demands are high. The simplicity task, as the name im-­‐plies, states that synaesthesia is more likely to develop when faced with a simpler task,simpler either because it is intrinsically easier or because the child understands someaspect of the task that makes it simpler for them. Using synaesthesia as a strategy mightsimply require cognitive resources that are not available when the learning task is toochallenging, or might only be feasible when the child has enough conceptual under-­‐ 27 standing of the inducer domain. For instance, one might need to have a high degree ofmetalinguistic awareness, understanding what letters are and how an alphabet works,in	
  order	
  to	
  develop	
  grapheme-­‐colour	
  synaesthesia.There are three speci>ic factors that we reasoned might reliably indicate differences inthe dif>iculty of becoming literate in ways that might also affect the development ofsynaesthesia. These include the orthographic transparency of the language that one islearning to write (how orderly and simple the relationship between phonemes andwritten graphemes is), the acquisition of second languages by the children who arelearning to write, and any unusual strengths or weaknesses that a child has with read-­‐ing or writing such as learning to read very early or, conversely, dyslexia. For all threefactors, we hypothesized they would affect the rates of grapheme-­‐colour but not othertypes of synaesthesia, since the letters that induce grapheme-­‐colour synaesthesia are,obviously,	
  a	
  critical	
  component	
  of	
  literacy.Our interest in orthographic transparency led to the study being run in the Czech Re-­‐public and Canada. Czech and English are interesting comparison cases because their al-­‐phabets are highly similar, but the relationship between these alphabets and thephonology of the spoken language differs greatly. Czech is highly orthographically trans-­‐parent—there is almost a one-­‐to-­‐one mapping between letter identities and their corre-­‐sponding phonemes (see Chapter 5 for a more complete description of Czech orthogra-­‐phy). English, on the other hand, is about as orthographically opaque as is possible—each letter can produce several different phonemes, and each phoneme can be producedby many different combinations of letters. Learning phoneme-­‐letter correspondences, acrucial part of learning to read, is a more dif>icult task in orthographically opaque lan-­‐guages such as English than in orthographically transparent languages (Ellis et al., 2004;Seymour, Aro, & Erskine, 2003). According to the complexity hypothesis, then, thereought to be more English-­‐speakers than Czech-­‐speakers resorting to unusual strategiessuch as using synaesthetic associations in order to learn the dif>icult English orthogra-­‐ 28 phy, and thus to higher rates of grapheme-­‐colour synaesthesia among English speakers,while	
  the	
  simplicity	
  hypothesis	
  predicts	
  the	
  reverse.It is well-­‐established that second language acquisition has associated bene>its that affectthe development of literacy. In particular, bilingual children have advantages in execu-­‐tive function and metalinguistic knowledge. Either of these, according to the simplicityhypothesis, could lead to higher rates of synaesthesia among bilinguals (and accordingto the complexity hypothesis could lead to higher rates of synaesthesia among themonolinguals who do not have these bene>its). Thus the survey asked participants to in-­‐dicate what languages they spoke and when they began learning them, enabling us toverify if the acquisition of second languages affects the prevalence of grapheme-­‐coloursynaesthesia.Finally, reading ability might affect rates of grapheme-­‐colour synaesthesia in at leasttwo different ways. First, according to the complexity hypothesis, someone who foundreading particularly dif>icult might be more likely to develop synaesthesia. Conversely,someone with unusually strong reading abilities might be more likely to developgrapheme-­‐colour synaesthesia according to the simplicity hypothesis. Cytowic and Ea-­‐gleman (2009) also suggest that early reading might be associated with synaesthesia.We asked all participants if they learned to read unusually early (before kindergarten)and	
  also	
  asked	
  several	
  questions	
  about	
  reading	
  dif>iculties	
  in	
  childhood	
  or	
  at	
  present.	
  This is, clearly, a highly exploratory study, and before starting it was apparent that wemight very well not >ind any in>luence of orthographic transparency, second languageacquisition, or reading ability on rates of synaesthesia. However testing for these threeeffects required a large-­‐scale survey of synaesthetic tendencies, indeed the largest yetperformed, and so would produce the highest-­‐quality data on the epidemiology ofsynaesthesia. In particular, we expected to provide clear answers to two long-­‐standingepidemiological questions: how common is synaesthesia, and are there more femalethan	
  male	
  synaesthetes? 29 2.1.2	
  	
  How	
  common	
  is	
  synaesthesia?Numerous surveys of synaesthetic tendencies have been made over the past 130 years(Baron-­‐Cohen, Burt, Smith-­‐Laittan, Harrison, & Bolton, 1996; Calkins, 1893; Cytowic,1993; Cytowic, 1997; Domino, 1989; Galton, 1883; Niccolai, 2012; Ramachandran &Hubbard, 2001a; Rich, Bradshaw, & Mattingley, 2005; Rose, 1909; Rothen & Meier,2010b; Simner et al., 2006; Simner, Harrold, Creed, Monro, & Foulkes, 2009; Ulich, 1957;Ward & Simner, 2005), but there is little agreement among them. Estimates of the preva-­‐lence of grapheme-­‐colour synaesthesia, for example, range from 0.05% (Baron-­‐Cohen,Burt,	
  Smith-­‐Laittan,	
  Harrison,	
  &	
  Bolton,	
  1996)	
  to	
  13%	
  (Calkins,	
  1893).	
  Simner and colleagues (2006) argue that these differences are due to two key method-­‐ological >laws, one or both of which are found in almost all these studies. Some studiesare far too liberal, in that they simply ask a large group of people whether they makesynaesthetic associations and take them at their word, meaning they use the self-­‐reportcriterion for synaesthesia and nothing else. Most recent studies avoid this problem byusing a more stringent consistency criterion, however most of them do not randomlysample the population, instead relying on self-­‐referral by synaesthetes, canvassing sub-­‐jects by means such as newspaper advertisements. This makes it likely that their ratesare far too conservative, as it is very unlikely that anything other than a small minorityof	
  newspaper	
  readers	
  would	
  respond	
  to	
  such	
  a	
  advertisement.To date there have only been two studies that avoid both of these problems. Simner et al. (2006) directly asked a random sample of individuals about their synaesthetic ten-­‐dencies, thereby avoiding the self-­‐referral problem, and immediately followed this upwith rigorous tests of consistency, >inding an overall prevalence of approximately 4.4%for all varieties of synaesthesia. Rothen & Meier (2010b) provided a grapheme-­‐colourconsistency test to everyone in their sample, >inding a prevalence of 2% in the generalpopulation and 7% among >ine-­‐arts students. However even with samples as large as500, these studies are somewhat underpowered to establish precise rates: for examplethe University study of Simner et al. (2006) >inds that grapheme-­‐colour synaesthesia 30 has an estimated prevalence of 1.8%, but the 95% con>idence interval for this estimateranges from 0.6-­‐3.0%. This is still vastly better than the 0.05-­‐13.0% range establishedfrom former conservative and liberal estimates (cf., respectively, Baron-­‐Cohen, Burt,Smith-­‐Laittan, Harrison, & Bolton, 1996; Calkins, 1893). However, given that the upperbound of Simner et al.’s con>idence interval is 5 times its lower bound, it is clear thatanyone interested in comparing rates across groups, as we are in the present study,would be unable to >ind anything other than exceptionally large effects (such as thosefound	
  by	
  Rothen	
  &	
  Meier,	
  2010b).Our survey was handed out to a random sample of students at Charles University, andthe online Synesthesia Battery (Eagleman, Kagan, Nelson, Sagaram, & Sarma, 2007) pro-­‐vided a rigorous consistency test. Thus, like Simner et al. (2006) and Rothen & Meier(2010b), the present study avoids both the liberal and conservative >laws of previousepidemiological studies, and its much larger sample size allows for a higher degree ofprecision	
  in	
  the	
  results. 2.1.3	
  	
  Are	
  there	
  more	
  female	
  than	
  male	
  synaesthetes?Several studies have reported a strong female bias among synaesthetes, with reportedrelative rates of synaesthesia associated with being female as high as 6 (Baron-­‐Cohen,Burt, Smith-­‐Laittan, Harrison, & Bolton, 1996). Ward and Simner (2005) cast doubt onthis by conducting a much larger familial study (85 families compared to 6 in Baron-­‐Co-­‐hen, Burt, Smith-­‐Laittan, Harrison, & Bolton, 1996), in which they found a far smaller fe-­‐male:male bias than in previous studies. Their initial group of 85 synaesthetes wasentirely composed of self-­‐referred synaesthetes, and was strongly female-­‐biased (4:1).These participants identi>ied a second group of 58 family members who they knew tobe synaesthetic as a result of directly speaking to them about the topic, and the ratio offemale:male synaesthetes was recalculated using both groups to be 2:1. This halving ofthe degree of female bias suggests that the initial bias of 4:1 was largely due to womenbeing more likely to self-­‐refer themselves, which means that even the lower ratio of 2:1is likely too high, for two reasons. First, as Ward and Simner (2005) point out, a female 31 referral bias would mean that male synaesthetes who were the sole synaesthetic mem-­‐bers of their families were far less likely to come to the attention of the study, as theycould only be members of the >irst group, since they could not be identi>ied as synaes-­‐thetic by other family members, yet their lower rates of self-­‐referral would make themless likely to be part of this >irst group. They do not discuss a second potential reasonwhy their 2:1 ratio may be too high, namely that a female self-­‐reporting bias might alsoaffect the likelihood of family members reporting synaesthesia to each other, not simplyto researchers, which could skew the composition of their second group of familymembers.	
  A more recent large-­‐scale familial study employing a similar methodology (Barnett etal., 2008a) still found a 6:1 female:male ratio of con>irmed synaesthetes, but this ratio ishighly doubtful. First, as just discussed, male synaesthetes who are the sole synaestheticmembers of their families would probably be less likely to come to the attention of thestudy. Second, the proportion of family members who were able to be contacted by theresearchers differed between genders, with 43% of female relatives uncontacted, com-­‐pared	
  to	
  69%	
  of	
  male	
  relatives.	
  Ward and Simner (2005) conclude that accurately determining the ratio of female:malesynaesthetes can only be done using a large-­‐scale study that does not rely on self-­‐refer-­‐ral. They proceeded to carry out such a study (Simner et al., 2006), which found a ratioof female:male synaesthetes of 1.1:1, which was not signi>icant. Thus they concludedthat previous reports of female biases were largely, if not entirely, due to a female self-­‐referral bias rather than to real differences in the prevalence of synaesthesia, but didnot entirely rule out the possibility of a small female bias. Simner et al.’s (2009) laterstudy of the development of synaesthesia in childhood found a female:male ratio of1.6:1, which was also non-­‐signi>icant due to a small sample size, but suggestivenevertheless.Our study may be able to shed more light on this question. As sample sizes are muchlarger than any previous study, we have the potential to >ind small but signi>icant effects. 32 Furthermore, the study mixes random sampling and self-­‐referral in a way that may al-­‐low for a female bias to be observed in action. In the >irst phase of the study, a papersurvey is handed out to a random sample of students in undergraduate university class-­‐es, but in the second phase of the study, participants are invited to register for an onlineconsistency test (Eagleman, Kagan, Nelson, Sagaram, & Sarma, 2007) and complete it tobe con>irmed as synaesthetes. Hence differences in willingness to comply with the ex-­‐perimental protocol should have relatively little impact on the >irst phase of the study,but could potentially affect the second phase. Thus we can not only determine sex bias-­‐es in rates of self-­‐reported and con>irmed synaesthesia, but we can also verify if there isa difference in compliance between men and women during the second phase, whichwill provide further evidence for the source of any gender differences in rates ofsynaesthesia. 2.1.4	
  	
  Outline	
  of	
  the	
  studyThe study consisted of two phases. First, a paper survey (see Appendix 1) was given touniversity students at Charles University (hereafter CU) in the Czech Republic and Si-­‐mon Fraser University (hereafter SFU) in Canada. This survey included descriptions of anumber of synaesthetic experiences and asked respondents to indicate if they had theseor similar experiences, as well as a number of questions about factors that we thoughtmight relate to synaesthesia (e.g. second-­‐language acquisition, reading ability, and gen-­‐der). Respondents who answered positively were then invited to participate in the sec-­‐ond phase of the study by taking the online consistency tests at the Synesthesia Battery(www.synesthete.org,	
  Eagleman,	
  Kagan,	
  Nelson,	
  Sagaram,	
  &	
  Sarma,	
  2007).	
  Analyses consisted of calculating the reported and con>irmed rates of the various typesof synaesthesia, and determining what other factors were associated with each of thesetypes. In particular, we veri>ied whether there were prevalence differences due to nativelanguage (which, given our two populations, corresponds to orthographic transparen-­‐cy), second language acquisition, reading ability, and gender. We also determined thedegree to which the various types of synaesthesia tend to cluster together in individu-­‐ 33 als, by calculating whether having a given type of synaesthesia makes one more or lesslikely	
  to	
  have	
  another	
  type. 2.2	
  	
  Methods 2.2.1	
  	
  Phase	
  I	
  -­	
  Paper	
  surveyThe paper survey (see Appendix 1) included questions about gender, handedness, lan-­‐guages spoken and the age of acquisition of these languages; six questions asking if par-­‐ticipants had speci>ic synaesthetic experiences (e.g. “When you see, hear, or think aboutcertain letters or numbers, do you see or feel any colours? Example: There is somethingyellow about the letter G."); one open-­‐ended question inviting participants to record anysynaesthetic experiences not covered by the other questions; one question about usingletters or numbers as characters in childhood stories; and four questions on readingabilities. We also asked participants for their email address in order to invite them toparticipate	
  in	
  phase	
  2.	
  The	
  survey	
  was	
  in	
  English	
  at	
  SFU	
  and	
  in	
  Czech	
  at	
  CU.Surveys were handed out to students in undergraduate university classes at the start ofthe period. A brief presentation was given describing synaesthesia and explaining thatparticipation was strictly voluntary and was not tied to their course performance. Ap-­‐proximately 10 minutes after the surveys had been handed out, they were collected. Itwas not possible to keep a precise count of the number of students who chose to com-­‐plete the survey, but we estimate that well over 95% completed and returned the surveyin	
  both	
  universities.	
  5001 students from CU and 6663 students from SFU returned a completed survey. Ofthese, 3431 CU (69%) and 6084 SFU (91%) students provided an email address, whichwas necessary for the second phase of the study. A wide variety of courses were sam-­‐pled in both universities, primarily from Arts and Social Sciences faculties. 69% of CUand	
  61%	
  of	
  SFU	
  respondents	
  were	
  female. 34 2.2.2	
  	
  Phase	
  II	
  -­	
  Synesthesia	
  Battery2394 CU and 2054 SFU respondents (48% and 31%, respectively, of the total survey re-­‐spondents) reported that they experienced colored letters, numbers, weekdays, months,or sounds. Of these, 1862 CU and 1881 SFU respondents also provided an email address(78% and 92%, respectively, of those who reported synaesthesia). Each of these individ-­‐uals was emailed and invited to participate in the veri>ication stage of the study usingthe online Synesthesia Battery (Eagleman, Kagan, Nelson, Sagaram, & Sarma, 2007),which includes consistency tests for a wide range of synaesthesias. Those who reportedother forms of potential synaesthesia were not contacted, as we did not have easily-­‐available online tests of consistency. In exchange for registering, each Battery partici-­‐pant was entered into a lottery for 10,000 Czech Crowns or 500 Canadian dollars (bothapproximately 500 USD) with a greater than 1% chance of winning. 355 CU and 302SFU	
  participants	
  registered	
  for	
  the	
  Battery	
  (thus	
  we	
  ran	
  4	
  lotteries	
  in	
  each	
  country).After registering on the Synesthesia Battery website, participants were presented with achecklist of synaesthesia types, and instructed to select any types that they may have. Ifthey selected a form of synaesthesia for which the Battery includes a consistency test,then they were given this test. These tests are described in detail in Eagleman et al.(2007) but in brief, they require participants to choose a color for each inducer in ran-­‐dom order, then repeat this process twice more. The similarity between the threecolours assigned to each inducer is calculated as a distance in RGB space, and the aver-­‐age across all inducers is calculated. This mean distance is the participant’s consistencyscore for a test, and each participant will generate one consistency score for each of thetests they complete. Eagleman et al. (2007) suggest that genuine synaesthetes tend tohave	
  a	
  consistency	
  score	
  below	
  1,	
  which	
  is	
  the	
  threshold	
  we	
  adopted. 35 2.3	
  	
  Results 2.3.1	
  	
  Linguistic	
  characteristics	
  of	
  the	
  samplesThere were large differences between the linguistic capabilities reported by students atCU and SFU (see Figure 2.1). Broadly speaking, the CU population was almost entirelycomposed of native Czech or Slovak speakers who learned multiple other European lan-­‐guages in grade school. (Czech and Slovak are closely related, largely mutually intelligi-­‐ble languages that use identical alphabets.) The SFU population was far more heteroge-­‐neous, containing roughly equal proportions of respondents who were bilingual frombirth, who learned a second language from kindergarten on, and who were monolingual.There are increases in the acquisition of a second language in the SFU population atages 5, 10, and 14+, which likely correspond, respectively, to large numbers of childrenentering second-­‐language immersion programs in kindergarten, to the beginning of theBritish Columbia elementary second-­‐language programs in grade 5 (BC Ministry of Edu-­‐cation, 2001), and to second-­‐language learning in high school and beyond. Unlike the CUsample, SFU multilinguals spoke languages from every corner of the globe. We adopteda conservative de>inition of “native” languages: any language reported as having beenacquired	
  after	
  their	
  second	
  birthday	
  was	
  considered	
  non-­‐native.	
  4727 CU respondents (95%) reported speaking Czech or Slovak as a native language,compared to only 4156 SFU participants (62%) speaking English as a native language.There were also far fewer distinct languages reported by CU (59) than SFU (158) stu-­‐dents, with only 10 of the CU languages being non-­‐European, compared to 123 at SFU.CU students tended to be more multilingual, with only 5 students (0%) reporting them-­‐selves to be monolingual, compared to 1507 SFU students (23%), and a mean number oflanguages spoken of 3.5 compared to 2.3. However there were far fewer nativemultilin-­‐guals in the CU sample, with only 102 students (2%) who reported speaking at least 2languages	
  before	
  age	
  2,	
  compared	
  to	
  1427	
  (21%)	
  at	
  SFU. 36 Figure	
  2.1	
  	
  Differences	
  in	
  second	
  language	
  acquisition	
  between	
  Czech	
  and	
  English	
  speakers.Unless otherwise stated, all proportions reported during the remainder of these resultsare taken from the 4727 native speakers of Czech or Slovak (CU) and the 4156 nativespeakers	
  of	
  English	
  (SFU).	
   2.3.2	
  	
  Higher	
  rate	
  of	
  endorsing	
  synaesthesia	
  among	
  CzechsFigure 2.2 presents the proportions of native-­‐speaking survey respondents who report-­‐ed synaesthetic experiences on the paper survey. In general, reported rates were high,with each form of synaesthesia described on the survey being endorsed by 5-­‐40% of re-­‐spondents at both universities, and 5% reporting other types of experience that theyfelt might qualify as synaesthetic. The vast majority of responses to the “other types” 0" 200" 400" 600" 800" 1000" 1200" 1400" 1600" 0" 1" 2" 3" 4" 5" 6" 7" 8" 9" 10 " 11 " 12 " 13 " 14 +" M on oli ng ua l" N um be r' of 'N a+ ve 'S pe ak er s' Age'of'Second'Language'Acquisi+on' Czech" English" 37 question appeared to be consistent with those reported in other surveys of the range ofsynaesthetic experiences (e.g. Day, 2005). Most were associations between particularcategories (e.g. school subjects, seasons, tastes, bodily actions, car models, letters, mu-­‐sic) and sensory experiences (colors, sounds, tastes, smells, shapes, sizes, textures, andtemperatures). There were also many reports of personi>ications of inanimate objects,as well as of arrangements of various categories in personal space. These reports, how-­‐ever, were not veri>ied any further, and neither were the reports of word-­‐taste, number-­‐form,	
  and	
  grapheme-­‐personality	
  synaesthesia.	
  All varieties of synaesthesia save the “other” category were endorsed by a greater pro-­‐portion of CU than SFU respondents, a difference that was signi>icant in all cases savegrapheme-­‐personality synaesthesia (as the error bars in Figure 2.2 are 95% C.I.’s, anycases where they do not cross are signi>icantly different). It may be worth noting thatdifferences in the proportions of letter-­‐color and number-­‐color synaesthesia were sole-­‐ly due to differences in the proportion of participants who reported both types ofsynaesthesia. That is, there were no signi>icant differences between the proportion ofrespondents endorsing letter-­‐color but not number-­‐color synaesthesia (CU: 10.0%, SFU:8.5%), or the converse (CU: 5.1%, SFU: 4.5%), but there was a large difference betweenthe proportion who reported experiencing both (CU: 6.0%, SFU: 1.5%), and this differ-­‐ence	
  was	
  the	
  main	
  contributor	
  to	
  the	
  overall	
  differences	
  between	
  the	
  two	
  rates. 38 0%# 5%# 10%# 15%# 20%# 25%# 30%# 35%# 40%# #Le +e r# #N um be r# #W ee kd ay /M on th # #So un d# #W or d= Ta ste # #N um be r#F or m# #G rap he me #Pe rso na lity # #O th er # Pe rc en ta ge )o f)s ur ve y) re sp on de nt s) Czech# English# Figure 2.2 Reported rates of synaesthetic experiences for native Czech and English speakers. Error barsindicate	
  +/-­‐	
  95%	
  C.I.s. 2.3.3	
  	
  Higher	
  rates	
  of	
  conOirmed	
  synaesthesia	
  among	
  CzechsFigure 2.3 presents the proportions of synaesthesias con>irmed by the Synesthesia Bat-­‐tery among native speakers, which ranged from less than 0.05% to 1.8%. The same gen-­‐eral trend of higher rates of Czech synaesthesia was observed: absolute proportionswere higher among the CU sample than SFU for all varieties of synaesthesia tested onthe battery save chord-­‐color synaesthesia, and signi>icantly higher in all cases save let-­‐ter-­‐colour and scale-­‐colour synaesthesia. Interestingly, the differences in the rates of re-­‐ported sound-­‐color synaesthesia (see Figure 2.2) appeared to be driven largely by dif-­‐ferences in instrument-­‐color synaesthesia. (Since the paper survey asked about “sound-­‐color” synaesthesia in general, and the Battery only includes tests for chords, instru-­‐ments, and scales, it is possible that there are differences in the rates of synaestheticcolors experienced in response to speech or other sounds, but this could not be veri-­‐>ied.) There was a high degree of crossover between the various types of synaesthesia,39 with most synaesthetes being con>irmed as having multiple types (see section 2.3.7 andAppendices 2 and 3 for a complete discussion of this). We found an overall prevalence of3.7%	
  synaesthetes	
  in	
  the	
  CU	
  sample	
  and	
  2.3%	
  in	
  the	
  SFU	
  sample. Figure 2.3 Con>irmed rates of synaesthesia for native Czech and English speakers. Error bars indicate +/-­‐95% C.I.s. “Unreported” synaesthetes are those who reported that they did not have the type of synaes-­‐thesia	
  in	
  question	
  on	
  the	
  survey,	
  but	
  tested	
  positively	
  for	
  it	
  on	
  the	
  Synesthesia	
  Battery.An unanticipated >inding was that there were a relatively large number of con>irmedsynaesthetes who reported on the initial paper survey that they did not have the type ofsynaesthesia in question, as shown in Figure 2.3. For instance, 19.0% (CU) and 30.0%(SFU) of con>irmed letter-­‐color synaesthetes selected the response “No, I do not haveexperiences like this”, when asked about associations between letters and colors on thesurvey. Since only participants who reported synaesthetic experiences were asked toregister for the Synesthesia Battery, this means that all of these participants reportedhaving at least one other type of synaesthetic experience. (E.g. 7 of the 8 con>irmed CUletter-­‐color synaesthetes who reported not experiencing colored letters did report ex-­‐periencing weekday or month color synaesthesia, and 3 of them reported experiencingnumber-­‐color	
  synaesthesia.)	
   0.0%$ 0.5%$ 1.0%$ 1.5%$ 2.0%$ 2.5%$ Le *e r$ Nu mb er $ W ee kd ay $ M on th $ Ch or ds $ Ins tru me nt s$ Sc ale s$ Pe rc en ta ge )o f)s ur ve y) re sp on de nt s) Czech$ English$ Unreported$ 40 2.3.4	
  	
  Higher	
  rates	
  among	
  Czechs	
  are	
  due	
  to	
  late	
  second-­language	
  acquisitionWhat explains the greater prevalence of synaesthesia among Czechs? The second factorwe suggested might impact rates of synaesthesia is second-­‐language acquisition. Recallthat virtually all the Czech speakers learned a second language, but not as a nativespeaker, whereas English speakers were almost evenly split between monolinguals, na-­‐tive multi-­‐linguals, and non-­‐native multi-­‐linguals (see Figure 2.1). If rates of synaesthe-­‐sia are higher among non-­‐native multilinguals than among monolinguals or native mul-­‐tilinguals, then this could explain the higher rates among Czech speakers. Figure 2.4shows the rates of con>irmed synaesthesia among these three second-­‐language groups.In order to maximize power, this analysis was performed using all 11,664 survey partic-­‐ipants from both samples, regardless of whether they were native speakers or not. Fur-­‐thermore, since the greater rates among Czechs were found in almost all varieties ofsynaesthesia we tested, we combined all varieties into a single measure, which indicat-­‐ed if a participant was con>irmed as having any one of the varieties of synaesthesiashown in Figure 2.3. Figure 2.4 demonstrates a clear prevalence difference between sec-­‐ond-­‐language groups. In particular, those who learn a second language between theages of 2-­‐12 are 2-­‐4 times as likely to have synaesthesia as native bilinguals, a differencewhich is signi>icant between ages 5-­‐10. An identical analysis using only SFU participantsyielded the same pattern of results. (This could not be meaningfully performed usingonly CU participants due to the extremely low numbers of Czech monolinguals and na-­‐tive	
  multi-­‐linguals.)	
   41 Figure 2.4 Rates of con>irmed synaesthesia by age of second language acquisition, for all survey partici-­‐pants	
  in	
  both	
  countries.	
  Error	
  bars	
  indicate	
  +/-­‐	
  95%	
  C.I.s.This means than any comparison of all CU participants to all SFU participants, as in Fig-­‐ure 2.3, would confound at least two factors: native language and age of second-­‐lan-­‐guage acquisition. Since there were effectively no native multilingual or monolingualCzechs, the only way to test for an effect of native language without this confound was totest only non-­‐native multilinguals (i.e. to eliminate roughly 2/3 of the English speakersfrom the analysis). When we did this, restricting our sample to native speakers of Czechor English who learned a second language after their second birthday, no signi>icant dif-­‐ferences	
  between	
  the	
  two	
  native	
  language	
  groups	
  were	
  found	
  (Figure	
  2.5).	
   0%# 1%# 2%# 3%# 4%# 5%# 6%# 7%# 0# 1# 2# 3# 4# 5# 6# 7# 8# 9# 10 # 11 # 12 # 13 # 14 +# M on oli ng ua l# Pe rc en ta ge )C on fir m ed )S yn ae st he te s) (a ny )t yp e) ) Age)of)Second)Language)Acquisi<on) 42 Figure 2.5 Con>irmed rates of synaesthesia among native Czech and English participants who acquired asecond	
  language	
  after	
  age	
  1.	
  Error	
  bars	
  indicate	
  +/-­‐	
  95%	
  C.I.s. 2.3.5 Grapheme stories and reading abilities are associated with reported but not conOirmed	
  synaesthesiaFigure 2.6 shows participants’ endorsement of the >inal >ive statements on the papersurvey, which included one statement concerning telling stories involving graphemes,and four concerning reading ability. CU participants were more likely to report tellinggrapheme stories (24.4%, SFU: 17.3%), while SFU participants were more likely to re-­‐port learning to read before kindergarten (CU: 43.5%, SFU: 48.8%) and also more likelyto report having had special assistance with reading as a child (CU:13.2%, SFU: 20.3%)and having reading dif>iculties as an adult (CU: 1.9%, SFU: 6.3%). Equal proportions ofrespondents endorsed having been formally diagnosed with a reading disorder as achild	
  (CU:	
  6.6%,	
  SFU:	
  6.2%).	
   0.0%$ 0.5%$ 1.0%$ 1.5%$ 2.0%$ 2.5%$ Le*er$ Number$ Weekday$ Month$ Chords$ Instruments$ Scale$ Pe rc en ta ge )C on fir m ed )S yn ae st he te s) Czech$ English$ 43 0%# 10%# 20%# 30%# 40%# 50%# 60%# #Grapheme#Stories# #Pre7K#Reader# #Help#With# Reading# #Reading#Disorder# #Reading#Problem# Now# Pe rc en ta ge )o f)s ur ve y) re sp on de nt s) Czech# English# Figure 2.6 Endorsement of questions concerning childhood reading and tendencies to tell stories in-­‐volving	
  graphemes,	
  in	
  native	
  Czech	
  and	
  English	
  speakers.	
  Error	
  bars	
  indicate	
  +/-­‐	
  95%	
  C.I.s.Positive responses to several of these questions were strongly associated with reportsof synaesthesia, but none with con>irmed synaesthesia. Participants who reportedtelling stories with graphemes as characters were more likely to report all forms ofsynaesthesia in both the CU and SFU samples. Reports of learning to read before kinder-­‐garten or having reading dif>iculties, on the other hand, were also associated with re-­‐ports	
  of	
  most	
  forms	
  of	
  synaesthesia,	
  but	
  only	
  in	
  the	
  SFU	
  sample.	
  These qualitative descriptions were supported by a series of Fisher's exact tests. Foreach variety of synaesthesia and each question about grapheme stories or reading, weconstructed a 2-­‐by-­‐2 contingency table, where one dimension separated those whoendorsed the variety of synaesthesia (or were con>irmed as having it on the Battery)from those who did not, and one separated those who endorsed the question from thosewho did not. Due to very low cell values, the three questions concerning reading dif>i-­‐culty were combined into a single variable: anyone who reported having at least one ofthese	
  dif>iculties	
  was	
  designated	
  as	
  having	
  indicated	
  a	
  reading	
  problem.This left one question asking about grapheme stories, one about learning to read beforekindergarten, and one about reading problems in general. Endorsements of 8 varieties44 of synaesthesia included on the survey were compared to each of these 3 questions, andso p-­‐values were Bonferroni-­‐corrected by multiplying them by 24. All forms of synaes-­‐thesia were positively associated with reporting having told stories involvinggraphemes as a child (all ps < .001 in both samples), save for the open-­‐ended questionabout "other" types of synaesthesia which was not signi>icant even prior to Bonferronicorrection in the CU sample (p > .1), but was marginally signi>icant after correction inthe SFU sample (p = .06). Reports of early reading were associated with reports of allforms of synaesthesia in the SFU sample (all ps < .05) save for number-­‐forms and "oth-­‐er" varieties (both ps > .9), whereas in the CU sample early reading was signi>icantly as-­‐sociated with sound-­‐colour synaesthesia (p < 001), and marginally associated with re-­‐ports of letter-­‐colour (p = .07), number-­‐colour (p = .08) and word-­‐taste (p = .08)synaesthesia. Reports of reading problems were associated with all forms of synaesthe-­‐sia in the SFU sample (all ps < .05), but none in the CU sample (all ps > .2). None of thesequestions were associated with con>irmed synaesthesias in either sample (all ps > .1).(It should be kept in mind that since there are far fewer con>irmed than reportedsynaesthetes,	
  power	
  is	
  clearly	
  an	
  issue	
  here.)The survey also included two questions concerning gender (CU: 69% female, SFU: 63%)and handedness (CU: 89.4% right-­‐handed, SFU: 87.3%). The associations between gen-­‐der and synaesthesia are described in the following section. Handedness was not associ-­‐ated with any form of synaesthesia, reported or con>irmed (all ps > .1 beforecorrection). 2.3.6	
  	
  A	
  female	
  bias	
  for	
  synaesthesia	
  due	
  to	
  differences	
  in	
  complianceWomen were both more likely to report synaesthesia and to be con>irmed as havingsynaesthesia, a trend which was true for virtually all varieties of synaesthesia in thestudy, but was much stronger among CU than SFU participants. However, the bias incon>irmed rates was almost entirely due to women being more likely to comply with ourrequest to take part in the second part of the study, suggesting that response and com-­‐pliance	
  biases	
  were	
  the	
  true	
  source	
  of	
  the	
  sex	
  differences	
  we	
  found. 45 Table 2.1 shows the female:male bias for each of the forms of synaesthesia reported onthe paper survey, while Table 2.2 shows the bias for forms of synaesthesia con>irmed bythe battery. In general, there was a far stronger female bias among CU participants, withonly two reported rates and one con>irmed rate showing a signi>icant bias among SFUparticipants, although in almost every case there was a non-­‐signi>icant trend towardsone.	
   Table	
  2.1	
  	
  Female:male	
  relative	
  rates	
  of	
  reported	
  synaesthesia Letter-­‐Colour Number-­‐Colour Weekday/Month-­‐Colour Sound-­‐Colour Word-­‐Taste Number	
  Forms Grapheme-­‐Personality OtherCzech 1.50*** 1.55*** 1.49*** 1.13	
  . 1.24*** 1.08 1.18* 0.95English 0.97 1.05 1.22* 1.09 1.08 1.07 1.21** 1.08 P-­‐values	
  are	
  computed	
  using	
  Fisher’s	
  exact	
  method.	
  All	
  p	
  values	
  are	
  Bonferroni	
  corrected	
  and	
  >	
  .1,	
  ex-­‐cept:	
  .	
  p	
  <	
  .1,	
  *	
  p	
  <	
  .05,	
  **	
  p	
  <	
  .01,	
  ***	
  p	
  <	
  .001 Table	
  2.2	
  	
  Female:male	
  relative	
  rates	
  of	
  con>irmed	
  synaesthesiaLetter-­‐Colour Number-­‐Colour Weekday-­‐Colour Month-­‐Colour Chord-­‐Colour Instrument-­‐Colour Scale-­‐ColourCzech 3.51* 2.32	
  . 3.13*** 2.21** 0.45 3.31** 3.13English 3.09* 1.19 1.85 1.68 0.20 4.16 0.59 P-­‐values	
  are	
  computed	
  using	
  Fisher’s	
  exact	
  method.	
  All	
  p	
  values	
  are	
  Bonferroni	
  corrected	
  and	
  >	
  .1,	
  ex-­‐cept:	
  .	
  p	
  <	
  .1,	
  *	
  p	
  <	
  .05,	
  **	
  p	
  <	
  .01,	
  ***	
  p	
  <	
  .001The degree of female bias increased sharply from reported to con>irmed rates for almostevery form of synaesthesia in both groups. One possible explanation is that there was ahigher rate of false reporting of synaesthesia among males, or at least a higher propor-­‐tion of men who reported synaesthetic experiences that are not consistent over timeand hence not con>irmable by the Synesthesia Battery. Another possibility, however, isthat men were less likely to comply with Phase 2 of the experiment, meaning that theywould	
  never	
  have	
  been	
  tested	
  in	
  the	
  >irst	
  place.	
  In order to test this we examined the female:male relative rates at various stages of theexperiment. Speci>ically, we looked at rates of those who reported synaesthesia, of thosewho registered for the Synesthesia Battery after being invited, of those who completedat least one test on the Battery, and >inally of those who were con>irmed synaesthetic by46 the Battery. Since the female bias was present in both samples and across virtually allforms of synaesthesia, all 11,664 participants were used in order to maximize power(results were qualitatively identical when run with either sample alone), and a singlemeasure was used to indicate if a participant reported or was con>irmed as having anyone of the seven varieties of synaesthesia tested on the Battery, as in the analysis of sec-­‐ond-­‐language acquisition shown in Figure 2.4. Con>idence intervals for relative rateswere	
  calculated	
  using	
  an	
  online	
  relative	
  risk	
  calculator	
  (MedCalc	
  Software,	
  2013).Figure 2.7 shows the results. There was a small female:male bias (1.28) for reportingsynaesthesia, then a larger bias (1.67) for registering for the Battery after reportingsynaesthesia. No signi>icant bias existed for either completing a Battery test after regis-­‐tering for the Battery (1.09) or being con>irmed as synaesthetic after completing a test(1.01). This means that the increase in female:male bias between reported and con-­‐>irmed synaesthesia was entirely due to attrition: men were less likely to comply withour request to register for the Battery. Of those who did register for the Battery, equalrates of men and women were con>irmed synaesthetic. Thus we have no evidence for afemale	
  bias	
  in	
  synaesthesia	
  beyond	
  a	
  mild	
  difference	
  in	
  initially	
  reported	
  rates. Figure 2.7 Female:male relative rates across various stages of the experiment. There is female bias forreporting synaesthesia and registering for the Synesthesia Battery, but almost identical female:male ratesof completing the Battery and being con>irmed as synaesthetic. Error bars are +/-­‐ 95% con>idenceintervals 0.8$ 1$ 1.2$ 1.4$ 1.6$ 1.8$ 2$ 2.2$ Reported$ Synaesthesia$ Registered$for$ Ba:ery$ Completed$ Ba:ery$ Confirmed$ Synaesthesia$ Fe m al e: M al e$ Re la Bv e$ Ra te $ 47 2.3.7	
  	
  All	
  forms	
  of	
  synaesthesia	
  cluster	
  togetherVirtually every form of synaesthesia, whether reported in Phase 1 of the study or con-­‐>irmed in Phase 2, was associated with every other form of synaesthesia, in both the CUand SFU populations. In other words, individuals with only one variety of synaesthesia,either	
  reported	
  or	
  con>irmed,	
  were	
  extremely	
  rare	
  in	
  these	
  data.	
  This qualitative description is supported by a series of Fisher’s exact tests. First, each ofthe eight forms of synaesthesia asked about on the survey (including “other” as a singletype) was compared to each of the 7 other forms, by constructing a 2-­‐by-­‐2 contingencytable, where each dimension separated those who endorsed one form of synaesthesiafrom those who did not. As there were 56 total tests, p-­‐values were Bonferroni-­‐correct-­‐ed by multiplying them by 56. All p-­‐values were still signi>icant below the .001 level,save for the comparison between endorsements of grapheme-­‐personality and “other”types of synaesthesia in the SFU sample, where p = .004. All these differences were dueto a positive association, such that the relative rate of reporting one type of synaesthesiawas 1.5-­‐4.9 times higher among those who had reported another type than it wasamong those who did not report any other type. There were no apparent differences be-­‐tween the CU and SFU samples in these results. (See Appendix 2 for a complete table ofrelative	
  rates.)Results were similar for the seven varieties of synaesthesia con>irmed on the Battery.Here there were 42 total tests, and so p-­‐values were Bonferroni-­‐corrected by multiply-­‐ing them by 42. Letter-­‐, number-­‐, weekday-­‐, month, and instrument-­‐colour synaesthe-­‐sias were all positively associated with each other in both the CU and SFU samples (all ps < .001), chord-­‐colour synaesthesia was positively associated with only month-­‐coloursynaesthesia in both samples (both ps < .05), while scale-­‐colour synaesthesia was posi-­‐tively associated with letter-­‐, number-­‐, and instrument-­‐colour synaesthesia in the CUsample (all ps < .01) and with letter-­‐ and month-­‐colour synaesthesia in the SFU sample(both ps < .05). Effect sizes were frankly enormous, with relative rates for the signi>icantcomparisons ranging from 23.9-­‐185.9. Given that scale-­‐colour and chord-­‐colour synaes-­‐ 48 thesia were con>irmed in such tiny percentages of both samples (see Figure 2.3), itseems likely that a failure to >ind positive associations with some other varieties ofsynaesthesia was simply a power issue. Further evidence in support of this claim comesfrom the fact that of 8 non-­‐signi>icant comparisons in the CU sample and 9 in the SFUsample, 4 and 6 (respectively) were signi>icant prior to the Bonferroni corrections, andthere was no overlap between the 4 remaining non-­‐signi>icant comparisons in the CUsample and the 3 in the SFU sample. (See Appendix 3 for a complete table of relativerates.) 2.4	
  	
  Discussion 2.4.1	
  	
  Overview	
  of	
  resultsThe prevalence of synaesthesia differs markedly between our samples of native Czechand English speakers, with Czechs reporting and being con>irmed as having higher ratesof almost all forms of synaesthesia—an overall prevalence of 3.7% synaesthetes as op-­‐posed to 2.3% among our native English speakers. However this difference does notstem from differences in orthographic transparency, or any other difference betweenthe Czech and English languages per se. Rather, it is tied to second language acquisition:those who learn a second language as a native speaker (before age 2) are three times less likely to be synaesthetes as those who learn a second language later in life, andthere are far more native bilinguals found in the SFU sample. Reading ability may beanother in>luence on the development of synaesthesia, but this result is far strongeramong our English than Czech speakers, and is only measurable in rates of reported, notcon>irmed, synaesthesia. We also >ind a female bias among our reported and con>irmedsynaesthetes, however further analysis demonstrates that the con>irmed bias is almostentirely due to differential rates of attrition between men and women: men are less like-­‐ly than women to participate in the second phase of our study. Finally, all forms ofsynaesthesia are highly associated, such that having one type means one is much morelikely	
  to	
  have	
  any	
  other	
  type. 2.4.2 Rates of synaesthesia are consistent with previous studies, likely under-­49 estimates	
  of	
  true	
  population	
  ratesThe con>irmed rates of synaesthesia established here are within the general range ofthose from the two previous studies that use a large sample not dependent on self-­‐re-­‐ferral and use a formal test of synaesthetic consistency (Rothen & Meier, 2010b; Simneret al., 2006). More speci>ically, four types of synaesthesia that we determine rates for(letter-­‐, number-­‐, weekday-­‐, and month-­‐colour) were also tested for in the Universitystudy of Simner et al. (2006), and both the CU and SFU rates for these four types are allwithin the 95% con>idence intervals of this study’s results. Thus it is clear that Simner et al. (2006) were correct to argue that appropriate methodologies are the key to con-­‐sistent results across different studies. By avoiding the liberal bias of only using the self-­‐report criterion as a test of synaesthesia, and the conservative bias of depending on self-­‐referral to obtain subjects, a clearer picture of the epidemiological pro>ile of synaesthe-­‐sia can be established. (The fact that both SFU and CU rates are within these con>idenceintervals, however, also establishes that previous studies lacked the power to >ind therelatively	
  large	
  group	
  effects	
  we	
  >ind	
  in	
  the	
  present	
  study.)While our study’s large sample size allows for a high degree of con>idence in the results,the proportion of con>irmed synaesthesia should be interpreted as lower bounds on itstrue prevalence, rather than estimates of the actual rate. There are two reasons for this.First, attrition or non-­‐compliance at various stages of the experiment meant that wereundoubtedly some genuine synaesthetes who were never tested. A large number of sur-­‐vey respondents did not provide an email address (CU: 31.4%, SFU: 7.1%), meaning thatthey could not be contacted to participate in the second phase of the study. Of those whowere contacted, only approximately one quarter registered for the Synesthesia Battery(CU: 25.6%, SFU: 21.2%), and of those who registered for the Battery, many did notcomplete a single consistency test (CU: 19.4%, SFU: 27.1%), even though we used a veryliberal criterion for “completing” a test: participants simply had to assign any colors atall to at least one inducer on each of the three occasions that this inducer was presentedto	
  them	
  during	
  the	
  test.	
   50 True synaesthetes would likely be far more interested in the study, and thus more will-­‐ing to comply with its various stages. Some evidence in favour of this comes from thefact that the rate of endorsing some form of grapheme-­‐color synaesthesia was far loweramong participants who did not provide an email address (CU: 13.1%, SFU: 9.2%) thanamong those who did (CU: 24.8%, SFU: 14.5%), and the same pattern holds for all otherforms of synaesthesia on the survey. We have followed up informally with a number ofthe participants who responded positively to questions about synaesthetic experienceson the survey but did not register for or complete the Battery, all of whom indicated thatthey misunderstood the initial question or that they were referring to one-­‐off experi-­‐ences rather than consistent experiences across their lifetime. This is consistent withnumerous anecdotal reports from other synaesthesia researchers, as well as the >indingof Simner et al. (2006) that the majority of individuals who respond positively to ques-­‐tions about synaesthetic experiences are unable to successfully complete consistencytests. Furthermore, we have already noted that our results are within the ballpark ofprevious studies, in particular our rates for grapheme-­‐colour synaesthesia are close tothose established in the Museum study of Simner et al. (2006) and in the control studyof Rothen & Meier (2010b), both of which were designed such that attrition was not anissue (all participants completed a consistency test). Nevertheless, there are almost cer-­‐tainly some genuine synaesthetes who, for whatever reason, did not provide an emailaddress,	
  or	
  did	
  not	
  register	
  for	
  or	
  complete	
  the	
  Battery.	
  The second reason why our rates should be interpreted as lower bounds stems from thefact that we found a surprisingly large number of con>irmed synaesthetes who reportedon the survey that they did not have the form of synaesthesia they were con>irmed ashaving. This raises important questions about the appropriate de>inition and opera-­‐tionalization of synaesthesia, as clearly the self-­‐report criterion that is used by almostall studies of synaesthesia is not only prone to false positives (in that most individualswho meet the criterion do not meet the consistency criterion) but also to false negatives(in that some people who do not meet the criterion do meet the consistency criterion).A more pressing concern for interpreting these results, however, is that it is unclear how 51 many of those who report no synaesthetic associations at all would be able to make con-­‐sistent	
  associations.	
  We	
  see	
  no	
  reason	
  to	
  think	
  that	
  this	
  group	
  is	
  insigni>icant.Clearly the best way of solving these issues would be to run a similarly-­‐sized studywhere all participants were given consistency tests, regardless of their self-­‐reported ex-­‐periences. For the time being, we will simply have to accept that the true rates ofsynaesthesia are likely somewhat higher than those established in this study and in theUniversity study of Simner et al (2006), which also only gave consistency tests to thosewho	
  reported	
  synaesthesia. 2.4.3	
  	
  Learning	
  and	
  synaesthesiaOne of the hypotheses driving this study was that the orthographic opaqueness of Eng-­‐lish would result in more English than Czech grapheme-­‐colour synaesthetes, but not af-­‐fect other varieties of synaesthesia. Clearly, this hypothesis is false, both in terms of thedirection of the effects and their speci>icity. Czechs, not English speakers, are more likelyto be synaesthetic, and this is true across synaesthesia in general, not restricted to thegrapheme-­‐colour variety. And in perhaps the most intriguing >inding from the presentstudy, we >ind that this group difference is associated with second-­‐language learning:those who learn a second language later in life are more likely to develop synaesthesia,of virtually any type. This single factor appears to entirely account for the differencesbetween	
  our	
  two	
  language	
  groups.This result was, to say the least, a surprise. Bilingualism is associated with stronger ex-­‐ecutive functioning and a higher degree of metalinguistic awareness (Bialystok & Barac,2012), which might either lead to greater or lesser rates of grapheme-­‐colour synaesthe-­‐sia (supporting the simplicity and complexity hypotheses, respectively). Indeed, our ini-­‐tial reason for asking about the age of acquisition of second languages was simply torule out those who learned a second language as an adult. However there is no signi>i-­‐cant difference between the rates of synaesthesia between bilinguals and monolinguals,rather	
  the	
  difference	
  is	
  between	
  native	
  and	
  non-­native	
  bilinguals.	
   52 Unfortunately, any attempts to explain this >inding will suffer from a serious lack of harddata. Most research on bilingualism tends to be performed on non-­‐native bilinguals, andmost research on native bilinguals compares their performance to monolinguals, not tonon-­‐native bilinguals (cf. Werker & Byers-­‐Heinlein, 2008). The research that does exist,in a nutshell, shows that non-­‐native bilinguals are faced with a harder task than nativebilinguals, and that they have a number of corresponding neurophysiological and be-­‐havioral differences. In general, the brain areas activated by the use of second languagesdo not differ with the age of acquisition, but children who acquire their second lan-­‐guages later show more variable and greater activation in these areas (Bloch et al.,2009), and are less able to suppress the involuntary switching of attention (Ortiz-­‐Mantilla, Choudhury, Alvarez, & Benasich, 2010). Furthermore, if participants are youngenough, then studies comparing monolinguals and native bilinguals are effectively alsocomparing non-­‐native bilinguals, since some of the monolinguals will go on to learn asecond language later in life. There are several reliable differences of note here. In anutshell, young native bilinguals tend to have a smaller vocabulary in each of their twolanguages (e.g. Bialystock & Herman, 1999; Cobo-­‐Lewis, Pearson, Eilers, & Umbel,2002), a greater degree of phonological awareness (Bialystock, Luk, & Kwan, 2005), andstronger	
  executive	
  functions	
  (Kovács	
  &	
  Mehler,	
  2009).	
  Non-­‐native bilinguals are, by de>inition, faced with a dif>icult learning task that monolin-­‐guals do not face: namely learning a new language, and more speci>ically learning toread and write in this new language. Native bilinguals will generally also become liter-­‐ate in both their languages, but they have the distinct advantage of already being >luentspeakers of each language, and also a higher degree of phonological awareness and ex-­‐ecutive function that will likely assist in becoming literate. Thus learning to read andwrite in both languages may be signi>icantly easier for native than non-­‐native bilinguals,and since monolinguals do not have a second language to learn, this allows them to bedifferentiated from both bilingual groups. This explanation, then, is consistent with thecomplexity hypothesis and not with the simplicity hypothesis. However more researchis	
  clearly	
  needed	
  in	
  this	
  case. 53 2.4.4	
  	
  Development	
  or	
  retention	
  of	
  synaesthesiaAn important question is whether the bilingualism effects we observe are due to non-­‐native bilinguals being more likely to develop synaesthesia, or less likely to lose it. Thischapter has generally advanced the idea that differences in adult rates of synaesthesiaare due to differences in developing synaesthesia, but there is no a priori reason whythese effects could not arise due to differences in retaining an already-­‐existing synaes-­‐thesia. The idea that the utility of synaesthesia might cause either its development orsimply its retention is an old one (Calkins, 1893), and the latter possibility is consistentwith the notion that synaesthesia is a normal part of development (Maurer, 1993). Thepresent	
  data,	
  of	
  course,	
  do	
  not	
  allow	
  these	
  two	
  possibilities	
  to	
  be	
  decided. 2.4.5	
  	
  The	
  generalizability	
  of	
  synaesthetic	
  tendenciesWe initially hypothesized that group differences would only be found for grapheme-­‐colour synaesthesia, but instead found a general difference between groups for almostall varieties of synaesthesia, and a strong tendency for all types of synaesthesia that wetest for to cluster with each other. This is consistent with Novich et al.’s (2011) reportsof clusters of synaesthetic types, since almost all the varieties of synaesthesia con>irmedon the Battery are what they refer to as coloured-­‐sequence synaesthesia. Novich et al.(2011) provide a neurological explanation of this clustering, suggesting that coloured-­‐sequence synaesthesias are based in unusual connectivity found within the neural net-­‐works responsible for coding sequence information, located in the middle temporalgyrus and temporoparietal junction in the right hemisphere, and the inferior frontalgyrus in the left hemisphere (Pariyadath, Plitt, Churchill, & Eagleman, 2012). We wouldlike to offer another, learning-­‐based, hypothesis for this clustering: perhaps oncesynaesthesia has been employed in solving a given task, it is likely to be used for other,conceptually similar tasks. Of course this is not meant as an alternative to the neurologi-­‐cal	
  account	
  of	
  Novich	
  et	
  al.	
  (2011),	
  and	
  both	
  factors	
  might	
  play	
  an	
  important	
  role. 54 2.4.6	
  	
  Sex	
  bias	
  in	
  synaesthesiaOur results >irmly establish that women are more likely than men to report synaestheticexperiences, and to voluntarily follow experimental procedures testing for synaesthesia(at least in the North American and European contexts in which these studies have tak-­‐en place). It is technically possible that these differential rates of attrition and non-­‐com-­‐pliance between men and women are due to more men realizing that their false reportsof synaesthesia will be uncovered by rigorous consistency testing, and thus feeling toouncomfortable to proceed. However this seems overly complex, especially in the face ofthe alternative explanation that women in our societies are simply more likely to followrequests and instructions, particularly when these involve self-­‐disclosure of things thatmay be embarrassing or at least uncomfortable (cf. Dindia & Allen, 1992; Simner et al.,2006;	
  Ward	
  &	
  Simner,	
  2005).Like the two previous studies that have shown a smaller sex bias in synaesthesia thanwas previously thought to exist (Simner et al., 2006; Ward & Simner, 2005), our >indingsshow a small female bias that cannot be explained on the basis of attrition (an overallrelative ratio associated with being female of 1.28 for reporting synaesthesia). It is pos-­‐sible that this re>lects a genuine sex difference in prevalence, but once more it seems farmore parsimonious to explain this bias on the basis of a self-­‐disclosure bias: malesynaesthetes may be less likely than women to report synaesthesia even when asked di-­ rectly about it. The only appropriate way of falsifying this hypothesis, once again, is alarge-­‐scale study that requires all participants to take a consistency test regardless ofself-­‐report. Until such time, we can be con>ident based on these results that if there is agenuine	
  sex	
  bias	
  in	
  synaesthesia,	
  it	
  is	
  a	
  minor	
  one.If men are less likely to report synaesthesia even when asked directly about it, this castsdoubt upon reports that synaesthesia is predominantly passed down via the maternalline (Barnett et al., 2008a; Baron-­‐Cohen, Burt, Smith-­‐Laittan, Harrison, & Bolton, 1996;Ward & Simner, 2005). All these studies screened participants using a short question-­‐naire, and did no further testing on those who responded negatively. Thus it could very 55 well be that there are far more synaesthetic fathers of synaesthetes than previouslythought.In general, then, these results strongly imply that there is no true sex bias in synaesthe-­‐sia, which further weakens one of the original motivations for assuming a simple genet-­‐ic	
  cause	
  of	
  synaesthesia. 2.4.7	
  	
  Problems	
  with	
  the	
  self-­report	
  and	
  consistency	
  criteria?Two aspects of our results raise questions about the suitability of the self-­‐report criteri-­‐on for synaesthesia. First, as suggested above, it is very likely the case that men are lesslikely to report synaesthesia even when asked directly about it. Second, a large fractionof our con>irmed synaesthetes denied having the type of synaesthesia on the initial sur-­‐vey which they were eventually con>irmed to have by the consistency criterion. Boththese results indicate that self-­‐reported synaesthetic experience is not necessary tohave highly consistent inducer-­‐concurrent relationships. Conscious experience has typi-­‐cally been taken as a fundamental aspect of synaesthesia, but we suggest that this is notnecessarily	
  true.Furthermore, there may be reason to think that the consistency criterion is not an ap-­‐propriate test for all forms of synaesthesia. Roughly the same proportion of people re-­‐port colour experiences induced by letters and by music, but the consistency test ispassed by far more of those who report letter colours than music colours. It is possiblethat individuals interpret questions about music-­‐induced colours in a more metaphori-­‐cal manner than questions about letter-­‐induced colours, and thus that questions aboutmusic-­‐colour synaesthetes will be answered far more liberally. However it seems equal-­‐ly plausible that music tends to induce colour experiences that are highly context-­‐de-­‐pendent, and thus not amenable to simple consistency tests. This may be the real expla-­‐nation for the very low rates of con>irmed music-­‐colour synaesthesia in this and otherstudies	
  (e.g.	
  Novich,	
  Cheng,	
  &	
  Eagleman,	
  2011). 56 2.4.8	
  	
  Future	
  directionsThe epidemiology of synaesthesia is still somewhat mysterious. There are strong indica-­‐tions that the standard criteria used to con>irm synaesthesia have important >laws. Italso appears that in order to >irmly establish both the precise rates of synaesthesia inthe population and the nonexistence of a female bias among synaesthetes, a large-­‐scalestudy that does not screen participants based on self-­‐report will be necessary. And newresearch will also be needed to con>irm or refute the hypothesis that the greater rates ofsynaesthesia among non-­‐native bilinguals are due to the increased demands they facewhile becoming literate in their second language, and to determine if these greater ratesare due to the development or retention of synaesthesia. However our results have shedlight on some mysteries. Most importantly, they provide a strong reason to think thatthe nature of the learning challenges faced in childhood have a large impact on the de-­‐velopment	
  of	
  synaesthesia. 57 3	
  	
  Second-­order	
  mappings	
  in	
  grapheme-­ colour	
  synaesthesia2 3.1	
  	
  IntroductionDespite an explosion of research on grapheme–colour synaesthesia over the past twodecades, little is known about how these associations are made. Why does Jane see theletter M as a deep purple, while John associates the same letter with forest green? Herewe verify that there are several different sources of synaesthetic associations, and weinvestigate both how they interact with each other and what aspects of synaestheticcolour	
  they	
  in>luence.To date, synaesthesia research has documented a number of regularities in thegrapheme–colour pairs of individuals. For example, English speakers often associate theletter B with blue or brown, G with green, and so on for the >irst letters of other com-­‐mon colour names (Barnett et al., 2008a; Rich, Bradshaw, & Mattingley, 2005; Simner etal., 2005). Similarly, some synaesthetes have adopted the colours of letter-­‐shaped fridgemagnets used in their childhoods (Witthoft & Winawer, 2006; Witthoft & Winawer,2013). These are regularities in Oirst-­order relations—that is, between nonrelational 2. This chapter is a slightly adapted version of a previously published paper (Watson, Akins, & Enns, 2012). The data for this paper was kindly provided by Michael Dixon and Jonathan Carriere of the University of Waterloo. The initial research question was collaboratively arrived at by the three co-authors, all analyses were my own, and I was the principal author of the paper. 58 properties of a letter (such as its shape or name) and dimensions of synaesthetic coloursuch	
  as	
  hue	
  and	
  lightness	
  (Day,	
  2005).A parallel line of research has begun to investigate grapheme–colour pairings by lookingfor second-­order relations, or “relations between relations.” For example, letters withsimilar shapes, such as E and F, tend to be associated with synaesthetic colours that aresimilar in hue (Brang, Rouw, Ramachandran, & Coulson, 2011; Eagleman, 2010; Jürgens&	
  Nikolic,	
  2012;	
  Brang,	
  Rouw,	
  Ramachandran,	
  &	
  Coulson,	
  2010).Here there is a correlation between two relations: A relation of similarity in the domainof letter shape is correlated with a relation of similarity in the domain of synaestheticcolour. Importantly, second-­‐order relations can exist independently of >irst-­‐order pair-­‐ings. That is, two synaesthetes may each assign different colours to E, but so long aseach individual’s colour for F is similar to that individual’s colour for E, this constitutesa second-­‐order relation between letter shape and synaesthetic colour. Thus, second-­‐or-­‐der letter–colour associations may not be apparent when looking at >irst-­‐orderrelations.A variety of second-­‐order in>luences on synaesthetic colour have been demonstrated.Marks (1975) noted that music-­‐colour synaesthetes often associate higher pitches withbrighter colours. In grapheme-­‐colour synaesthesia, numerals and letters that appearmore frequently in print tend to be associated with brighter (Beeli, Esslen, & Jäncke,2007; Cohen Kadosh, Henik, & Walsh, 2007; Simner & Ward, 2008; Smilek, Carriere,Dixon, & Merikle, 2007) and more saturated (Beeli, Esslen, & Jäncke, 2007) colours.More frequent letters also tend to be associated with colours whose names are morecommon in spoken language (Rich, Bradshaw, & Mattingley, 2005; Simner et al., 2005).Each of these results has been reported as a >irst-­‐order relation (correlations between absolute values on two dimensions), but they all imply second-­‐order relations (correla-­‐tions between differences in values on two dimensions). For example, the fact that more-­‐frequent letters have brighter colours implies that letters that differ greatly in terms offrequency	
  will	
  also	
  differ	
  in	
  terms	
  of	
  their	
  brightness. 59 Two recent results have come directly from second-­‐order analyses. First, as notedabove, letters with similar shapes appear to be associated with similar synaestheticcolours (Brang, Rouw, Ramachandran, & Coulson, 2011; Eagleman, 2010; Jürgens &Nikolic, 2012; Brang, Rouw, Ramachandran, & Coulson, 2010). Second, letters early inthe alphabet tend to have colours that are quite distinct from each other, whereas laterletters tend to have colours that are more similar to those of nearby letters (Eagleman,2010). On Eagleman’s view, this pattern stems from the order in which children learntheir letters. The >irst letter learned is associated with an idiosyncratic colour; the nextletter is associated with a colour that is easily distinguishable from the >irst; and eachsubsequently learned letter is associated with a colour as distinct as possible from thosealready assigned. With each letter learned, however, the range of distinct colour choicesis diminished, and inevitably, letters learned later are associated with colours similar tosome of those associated with earlier letters. Note that this interpretation implies a rela-­‐tion between letter ordinality and synaesthetic colour that is similar to Weber’s fraction.In brief, a pair of letters that appear early in the alphabet (e.g., A and D) will be assignedcolours that are more distinctive than will a pair of letters later in the alphabet (e.g., Sand V), even though they are equal numbers of steps apart in absolute units (three, inthis example). Such a >inding requires a second-­‐order perspective: When one looks atabsolute hue assignments, no relation with ordinality is found (Cohen Kadosh, Henik, &Walsh,	
  2007;	
  Smilek,	
  Carriere,	
  Dixon,	
  &	
  Merikle,	
  2007).In line with these >indings, we prefer to analyze second-­‐order relations amonggrapheme–colour pairs. Our primary motivation is that strong second-­‐order mappings(with weaker >irst-­‐order mappings) have often been observed in human perceptionmore generally (e.g., we remember melodies, not absolute pitch, in music; facial con>igu-­‐rations, not speci>ic facial features, in vision; and words, not phonemes, in language). Asecondary motivation is that second-­‐order analyses allow for the easy investigation ofthe property of hue. Because luminance and saturation are one-­‐dimensional propertiesof colours, they can be used in correlations or other linear analyses. Hue requires atleast two dimensions, however, in order to be speci>ied (e.g., blue-­‐yellow or red-­‐green), 60 which makes it impossible to compute a simple correlation between hue and any othermeasure. Differences between hues, on the other hand, are one-­‐dimensional, and thusamenable	
  to	
  linear	
  analysis.In the present study, we compared the colours assigned to letters by a large group ofsynaesthetes (N=54) with a wide variety of letter similarity measures taken from non-­‐synaesthetic individuals. We sought to determine how different aspects of letter similar-­‐ity (e.g., shape, order, and frequency) are related to synaesthetic colours, and how theseeffects relate to each other. We were especially interested in how the various aspects ofletter similarity might be related to two dimensions of colour—namely, luminance andhue (Beeli, Esslen, & Jäncke, 2007). As noted above, differences in letter frequency havebeen shown to correspond to differences in luminance, while differences in hue havegenerally been overlooked, possibly because researchers have been looking for >irst-­‐or-­‐der relations. How letter shape and ordinality map separately onto luminance and hueremains	
  an	
  open	
  question. 3.2	
  	
  Data	
  preparationThe RGB colour values of each letter were provided by 54 con>irmed grapheme–coloursynaesthetes (Smilek, Carriere, Dixon, & Merikle, 2007). These values were recoded intoCieLab colour space, which more accurately describes human colour discriminationsand allows for the separation of colour into luminance and hue components. There are325 possible letter pairs (not including doubles of the same letter), and for each of thesepairs we computed separate values for colour distance (Euclidean distance in CieLabspace), luminance distance (distance along the CieLab L-­‐axis), and hue distance (dis-­‐tance in the CieLab ab plane). These values were averages of the distances across all 54synaesthetes. 61 Table	
  3.1	
  	
  Letter	
  similarity	
  measures	
  used	
  in	
  the	
  English	
  studiesSimilarity	
  Measure DescriptionShape	
  difference Euclidean	
  distance	
  in	
  an	
  11-­‐dimensional	
  space	
  de>ined	
  using	
  the	
  ba-­‐sic	
  letter	
  shape	
  features	
  from	
  Gibson	
  (Gibson,	
  1969)Frequency	
  difference Difference	
  of	
  two	
  letters’	
  frequencies	
  divided	
  by	
  the	
  sum	
  of	
  their	
  fre-­‐quencies	
  (Lewand,	
  2000)Ordinality	
  difference Difference	
  of	
  two	
  letters’	
  positions	
  in	
  the	
  alphabet	
  divided	
  by	
  the	
  sumof	
  their	
  positionsRatings	
  	
  	
  A	
  (similarity) Similarity	
  ratings	
  of	
  uppercase	
  letters	
  (Boles	
  &	
  Clifford,	
  1989)	
  	
  	
  B	
  (similarity) Similarity	
  ratings	
  of	
  lowercase	
  letters	
  (Boles	
  &	
  Clifford,	
  1989)	
  	
  	
  C	
  (difference) Difference	
  ratings	
  of	
  uppercase	
  letters	
  (Podgorny	
  &	
  Garner,	
  1979)Discrimination	
  RT Reaction	
  time	
  on	
  a	
  same–different	
  discrimination	
  task	
  for	
  uppercase	
  letter	
  pairs	
  (Podgorny	
  &	
  Garner,	
  1979)Confusion	
  	
  	
  A Chance	
  of	
  confusing	
  two	
  brie>ly	
  presented	
  uppercase	
  letters	
  on	
  a	
  let-­‐ter-­‐naming	
  task	
  (Gilmore,	
  Hersh,	
  Caramazza,	
  &	
  Grif>in,	
  1979)	
  	
  	
  B Chance	
  of	
  confusing	
  two	
  uppercase	
  letters	
  (in	
  Keepsake	
  font)	
  pre-­‐sented	
  at	
  low	
  intensity	
  on	
  a	
  letter-­‐naming	
  task	
  (Gupta,	
  Geyer,	
  &	
  Maalouf,	
  1983)	
  	
  	
  C Chance	
  of	
  confusing	
  two	
  uppercase	
  letters	
  (dot-­‐matrix	
  font)	
  present-­‐ed	
  at	
  low	
  intensity	
  on	
  a	
  letter-­‐naming	
  task	
  (Gupta,	
  Geyer,	
  &	
  Maalouf,	
  1983)Letter	
  Name	
  Similarity Number	
  of	
  shared	
  phonemes	
  in	
  two	
  letter	
  names	
  (e.g.,	
  “bee”	
  and	
  “dee”	
  have	
  1	
  shared	
  phoneme,	
  /i/)A total of 11 measures of letter similarity were derived for comparisons with thesynaesthetic colour data (see Table 3.1). Shape difference is the Euclidean distance in aletter-­‐shape similarity space generated from 11 basic letter-­‐shape features (Gibson,1969), such as the presence or absence of a diagonal line (see Table 3.2). Frequency dif-­ ference and ordinality difference are the differences between the frequencies (Lewand,2000) and positions in the alphabet of two letters, divided by their sum. Letter name similarity consists of the number of shared phonemes in the English names of two let-­‐ters; for instance, the names of the letters B and D share one phoneme, /i/, and hencewould have a letter name similarity of 1 (Ward & Simner, 2003). These are examples ofthe familiar Weber fraction that describes perceived difference in numerous psy-­‐chophysical domains. The remaining measures were previously published behavioral 62 data on letter similarity, and thus may have been in>luenced by letter shape, frequency,order of acquisition, and (potentially) many other factors. These measures include dis-­ crimination RTs, from a same–different task in which the subjects were brie>ly presentedwith letter pairs (Podgorny & Garner, 1979); comparison ratings of letter similarity ordifference (Boles & Clifford, 1989; Podgorny & Garner, 1979); and confusion, from letter-­‐naming tasks using degraded stimuli (Gilmore, Hersh, Caramazza, & Grif>in, 1979; Gup-­‐ta,	
  Geyer,	
  &	
  Maalouf,	
  1983).Except where noted, all subsequent analyses were performed after binning the 325 let-­‐ter pairs into 65 bins that each included >ive letter pairs. Bins were determined by themean colour distance of each letter pair across all 54 synaesthetes, such that the >irstbin contained the >ive pairs whose two letters were, on average, most similar in colour,and the last bin contained the >ive pairs whose two letters were, on average, most dis-­‐similar	
  in	
  colour. Table 3.2 Letter shape dimensions adapted from Gibson (1969). Some dimensions’ names have beenchanged for clarity, and two dimensions—right and left diagonal—have been combined into a single diag-­‐onal	
  dimension.Shape	
  Dimension LettersStraight	
  Horizontal A	
  E	
  F	
  G	
  H	
  L	
  T	
  ZStraight	
  Vertical B	
  D	
  E	
  F	
  H	
  I	
  K	
  L	
  M	
  N	
  P	
  R	
  T	
  YClosed	
  Curve B	
  D	
  O	
  P	
  Q	
  RUpward-­‐Opening	
  Curve J	
  UHorizontal-­‐Opening	
  Curve C	
  G	
  J	
  SIntersection A	
  B	
  E	
  F	
  H	
  I	
  K	
  P	
  Q	
  R	
  T	
  XRepeated	
  Element B	
  E	
  M	
  S	
  WSymmetry A	
  B	
  C	
  D	
  E	
  H	
  I	
  K	
  M	
  O	
  T	
  U	
  V	
  W	
  X	
  YVertical	
  Discontinuity A	
  F	
  H	
  I	
  K	
  M	
  N	
  P	
  R	
  T	
  YHorizontal	
  Discontinuity E	
  F	
  L	
  T	
  ZDiagonal A	
  K	
  M	
  N	
  Q	
  R	
  V	
  W	
  X	
  Y	
  Z 63 3.3	
  	
  Results 3.3.1	
  	
  Letter	
  similarity	
  measures	
  predict	
  different	
  aspects	
  of	
  colour	
  similarityWe computed the simple correlations of all of the letter similarity measures with colour,luminance, and hue distance (see Table 3.3). Since multiple correlations were run, wecorrected the p values, multiplying each by 7, as seven distinct types of measures werebeing compared with each of the colour distance measures. We also used Spearman’srho	
  for	
  the	
  correlations	
  involving	
  the	
  three	
  ratings,	
  since	
  they	
  are	
  ordinal	
  measures.Colour distance and hue distance were both correlated with shape difference, ordinalitydifference, and Letter Confusion B. Luminance distance was correlated with frequencydifference, with Rating C and marginally correlated with Letter Confusion C. Thus, thereappears to be a split between those aspects of letter similarity that predict synaestheticluminance	
  and	
  hue. Table	
  3.3	
  	
  Correlations	
  between	
  letter	
  and	
  colour	
  similaritySimilarity	
  Measure Colour	
  Distance Luminance	
  Distance Hue	
  DistanceShape	
  difference 0.48*** 0.07 0.49***Frequency	
  difference 0.06 0.34* 0.01Ordinality	
  difference 0.37* 0.02 0.39**Ratings	
  	
  	
  A	
  (similarity) -­‐0.20 -­‐0.28 -­‐0.19	
  	
  	
  B	
  (similarity) -­‐0.22 -­‐0.29 -­‐0.20	
  	
  	
  C	
  (difference) 0.23 0.35* 0.21Discrimination	
  RT -­‐0.21 -­‐0.26 -­‐0.19Confusion	
  	
  	
  A 0.04 -­‐0.23 0.07	
  	
  	
  B -­‐0.34* -­‐0.26 -­‐0.32	
  .	
  	
  	
  C -­‐0.27 -­‐0.31	
  . -­‐0.24Letter	
  Name	
  Similarity -­‐0.02 -­‐0.00 -­‐0.02Correlations	
  with	
  the	
  ratings	
  use	
  Spearman’s	
  rho.	
  All	
  p	
  values	
  are	
  Bonferroni	
  corrected	
  and	
  >	
  .1,	
  except:	
  .	
  p	
  <	
  .1,	
  *	
  p	
  <	
  .05,	
  **	
  p	
  <	
  .01,	
  ***	
  p	
  <	
  .001 64 3.3.2	
  	
  Letter	
  shape	
  and	
  ordinality	
  predict	
  hue;	
  letter	
  frequency	
  predicts	
  luminanceAll of the correlations described above can be accounted for in terms of only three map-­‐pings, shown in Figure 3.1. A >irst mapping involves letter shape and synaesthetic hue, asecond involves letter ordinality and hue, and a third involves letter frequency and lu-­‐minance. A regression model using only shape difference and ordinality difference topredict hue distance (R2 0 .31, p < .001) did not explain less variance than one using all11 letter similarity measures to predict hue distance (p > .05). However, removing ei-­‐ther shape difference or ordinality difference from the reduced model resulted in signi>-­‐icantly less explained variance (p < .05 in both cases). As Confusion B was also signi>i-­‐cantly correlated with hue distance (see Figure 3.3), we tried adding it to this reducedmodel, but it did not explain any variance independently of shape and ordinality differ-­‐ence (p > .9). Similarly, a regression model using frequency difference as the sole predic-­‐tor of luminance distance (R2 0 .12, p < .01) did not differ from a model using all 11 simi-­‐larity measures as predictors (p > .1). We also tried a two-­‐predictor model that includedRating C, as this was correlated with luminance distance, but it did not explain any vari-­‐ance	
  independent	
  of	
  frequency	
  difference	
  (p	
  >	
  .1). Figure 3.1 Scatterplots of three second-­‐order mappings between letter similarity and synesthetic color.The x-­‐axes denote differences between letter pairs in (a) letter shape, (b) letter ordinality, and (c) letterfrequency. The y-­‐axes for panels (a) and (b) denote distance in terms of synesthetic hue; in panel (c), they-­‐axis denotes distance in terms of synesthetic luminance. The 65 data points in each plot were obtainedby binning 325 letter pairs into >ive-­‐pair bins and then averaging over 54 synesthetes (from Smilek et al.,2007) 40# 50# 60# 70# 80# 90# 100# 1.5# 1.8# 2.1# 2.4# 2.7# H ue $D is ta nc e$ Shape$Difference$ 40# 50# 60# 70# 80# 90# 100# 0.1# 0.3# 0.5# 0.7# H ue $D is ta nc e$ Ordinality$Difference$ 22# 24# 26# 28# 30# 32# 34# 36# 0.1# 0.3# 0.5# 0.7# 0.9# Lu m in an ce $D is ta nc e$ Frequency$Difference$ a# c#b# 65 3.3.3	
  	
  Analyses	
  of	
  individual	
  differences	
  show	
  that	
  the	
  mappings	
  are	
  independentWe computed the correlations for each of the mappings in Figure 3.1 at the level of indi-­‐vidual synaesthetes. This revealed that these correlations were positive for a majority ofthe synaesthetes (81%, 67%, and 54% for panels (a), (b), and (c), respectively, in Fig. 1])Overall, 28% of the synaesthetes had positive correlations for all three mappings, 46%had positive correlations for two of the mappings, 19% had a positive correlation foronly one mapping, and the remaining 7% had no positive correlations for any of thethree mappings. Critically, there were no hints of correlations between any of thesemappings, as tested by coding the presence or absence of each mapping as 0 or 1 foreach synaesthete, or by correlating the rank order of synaesthetes on each mapping, asdetermined	
  by	
  the	
  magnitude	
  of	
  their	
  individual	
  correlations	
  (all	
  ps	
  >	
  .2). 3.3.4	
  	
  Which	
  aspects	
  of	
  shape	
  matter?The shape difference measure was further subdivided into 11 dimensions of shape thatare important in letter identi>ication (Gibson, 1969). Only two of these dimensions weresigni>icantly correlated with hue distance (after a Bonferroni correction)—namely, dis-­‐tance along the closed curve and repeated element dimensions (rs = .41 and .38, ps = .01and .02, respectively; all other ps > .1). However, a model that used only these twodimensions to predict hue distance explained less variance than did a model using all 11dimensions (p < .05). In the complete 11-­‐predictor model, the only variables that madea signi>icant independent contribution were distance along the closed curve, repeatedelement, and diagonal dimensions (all ps < .01). Thus, we tried a model using thesethree variables to predict hue distance (R2 = .38, p < .001), and found that it did not pre-­‐dict less variance than the complete 11-­‐predictor model (p > .1), but removing any oneof these three dimensions from the model resulted in less explained variance (all ps< .05). Further study will be needed to determine why these features are especially im-­‐portant	
  to	
  synaesthetes. 66 3.4	
  	
  DiscussionThese results con>irm that three distinct aspects of letter similarity have a second-­‐orderin>luence on synaesthetic colour assignments. The shape, frequency, and ordinality ofindividual letters in>luence the colours assigned to them by synaesthetes, and thesethree effects are completely independent of each other: For instance, an individual witha strong shape-­‐hue association may or may not have a strong frequency-­‐luminance as-­‐sociation. Finally, each of these mappings is con>ined to a particular dimension of colourspace: Letter shape and ordinality are associated with hue, while frequency is associat-­‐ed	
  with	
  luminance.Brang et al.’s (2011) cascaded cross-­‐tuning model of synaesthesia states that shape–colour associations are the result of the coactivation of contiguous brain areas in thefusiform gyrus that represent letter form and colour. This model does not currently ac-­‐count for the other relations we found, nor for the fact that each relation is con>ined to aparticular dimension of colour. Instead of looking for an explanation at the level ofshared neurons, we offer two complementary hypotheses for these >indings, both ofwhich revitalize an old hypothesis of Calkins (1893): that synaesthetic associations mayarise for strategic reasons. (We remain agnostic as to whether those who employ suchstrategies	
  are	
  consciously	
  aware	
  of	
  doing	
  so.)First, associating letter shapes and identities with hue might aid learning to read, butassociating them with luminance might compromise reading performance. In vision, acommon strategy is to process hue and luminance separately (Gheorghiu & Kingdom,2006, 2007; Kingdom, Beauce, & Hunter, 2004; Kingdom & Kasrai, 2006; Liebe, Fischer,Logothetis, & Rainer, 2009; Nagai & Uchikawa, 2009; Shimono, Shiori, & Yaguchi, 2009),because each dimension provides different information about the environment (Hansen& Gegenfurtner, 2009). For example, a vital part of vision is to differentiate shadowsfrom material objects. Since shadows are de>ined by differences in luminance, whereas 67 objects usually differ from their background in both luminance and hue, it follows thathue	
  edges	
  are	
  a	
  more	
  reliable	
  cue	
  to	
  object	
  boundaries	
  than	
  are	
  luminance	
  edges.A similar moral applies in reading. Graphemes are usually presented as dark, achromat-­‐ic elements on a lighter background, and thus are usually processed entirely on the ba-­‐sis of luminance contrast. Second-­‐order relations between synaesthetic hue and shapecould provide an additional source of information to be exploited for such tasks as lettersegmentation, identi>ication, place-­‐holding for visual saccades, search for letters, main-­‐taining letter order in short-­‐term memory, and so forth. However, similar mappings be-­‐tween synaesthetic luminance and shape might interfere with the luminance-­‐sensitivechannels responsible for letter shape perception, and so could con>lict with natural vari-­‐ations in luminance from the font and from illumination. Thus, synaesthetes may exploitinformation about letter identity encoded in synaesthetic hue, in addition to the sys-­‐tems that they share with nonsynaesthetes, which use luminance contrast in the variouscognitive	
  operations	
  involved	
  in	
  reading.A complementary hypothesis for mapping hue and luminance to separate aspects of let-­‐ter identity stems from differences in the ways that humans use hue and luminance torepresent information. Take map reading as an example. De>ining regions by hue typi-­‐cally allows for faster and more accurate judgments of categorical distinctions than doesde>ining them by luminance, while luminance scales afford advantages for judgmentsabout relative quantity or continuous magnitudes (Breslow, Trafton, McCurry, & Rat-­‐wani, 2010). This likely re>lects the fact that variations in luminance have an underlyingcontinuity, from dark to light, while hues are perceived categorically. As letter frequencyvaries along a continuum, then, it maps naturally to luminance. Letter shapes, on theother hand, are perceived categorically (Boles & Clifford, 1989), and thus map naturallyto hue. Since letter ordinality also varies continuously, one might think that it should beassociated with luminance. Recall, however, that we use letter ordinality as a rough in-­‐dex of the order of learning of individual letters, which are themselves seen as categori-­‐ 68 cal objects (Eagleman, 2010), so the association between ordinality and hue is also con-­‐sistent	
  with	
  this	
  hypothesis.Previous research has reported a number of >irst-­‐order synaesthetic colour associa-­‐tions; for instance, the letters used to begin common colour words are frequently asso-­‐ciated with the colours named by these words, and the letters O and I are almost alwaysblack, white, or gray (Barnett et al., 2008a; Day, 2005; Rich, Bradshaw, & Mattingley,2005; Simner et al., 2005). We stress that the >inding of second-­‐order relations is com-­‐plementary to such results. It appears that a wide range of factors, of both the >irst andsecond orders, can potentially in>luence letter–colour mappings, as a good deal of vari-­‐ance in both luminance and hue still remains unexplained. Our analysis of individual dif-­‐ferences suggests that these factors often co-­‐exist within individual synaesthetes. Thatis, any given letter-­‐colour mapping might be in>luenced by a particular factor, but a dif-­‐ferent letter (or the same letter for a different synaesthete) is quite likely to be colouredaccording	
  to	
  a	
  different	
  factor.In summary, examining relations involving differences between letters and their as-­‐signed colours has allowed us to directly compare and contrast multiple in>luences onsynaesthetic associations. The >inding that second-­‐order relations are pervasive insynaesthesia is further evidence for the view that synaesthesia builds on normal mecha-­‐nisms (Barnett et al., 2008a; Simner et al., 2005). Though most of us may not reliably as-­‐sociate letters with colours, those of us who do tend to use principles common to othersensory	
  and	
  cognitive	
  domains. 69 4	
  	
  Higher-­Ridelity	
  synaesthetic	
  colour	
   data	
  increases	
  strength	
  of	
  effects3 4.1	
  	
  IntroductionThe previous chapter demonstrated a number of second-­‐order relationships betweenvarious properties of letters and their synaesthetic colours. Since analyzing these data,however, I have obtained a similarly-­‐sized database of colours from different synaes-­‐thetes. In this chapter I re-­‐run the same analyses as in Chapter 3 on this new data set.This is partly in order to verify that the pattern of results is similar, as Chapter 3 was anexploratory study that found a number of unexpected results, and so one might ques-­‐tion if these results generalize. Furthermore, it is probable that the colours in the newdata set have a higher >idelity to synaesthetic experience, as they are chosen from allpossible colours on a standard computer screen, which vary freely in hue, saturation,and luminance, unlike the colours chosen by participants in Chapter 3, which, in orderto speed up colour selection, were all fully saturated. Thus it is possible that there willbe	
  stronger	
  effect	
  sizes	
  using	
  the	
  present	
  data	
  set. 4.2	
  	
  ParticipantsPotential participants were identi>ied on the basis of their responses to an ongoinglarge-­‐scale survey about synaesthetic tendencies at Simon Fraser University in BritishColumbia, Canada, or were self-­‐referred after viewing advertisements on various web-­‐sites (e.g. Craigslist) or on bulletin boards in the Vancouver area (e.g. at universities, cof-­‐ 3. See Footnote 1 on p. 26 for authorship details. 70 fee shops, libraries, community centres, etc). All potential synaesthetes were invited totake the online synaesthesia Battery (Eagleman, Kagan, Nelson, Sagaram, & Sarma,2007), a standard co>irmation test for synaesthesia that veri>ies that the colour associa-­‐tions made with a variety of inducers (letters, sounds, weekdays, etc) are consistentacross repeated presentations. 48 participants obtained a consistency score below 1 onthe letter-­‐colour consistency test, which is considered a strong indicator of genuinesynaesthesia. 4.3	
  	
  Data	
  preparationWhile completing the Battery, each participant assigned an RGB colour to each letter 3times. For each letter, we took the >inal (third) colour assigned to that letter as itscanonical synaesthetic colour for that individual, on the assumption that by the third at-­‐tempt, participants would have fully understood the colour-­‐choosing interface. Allanalyses use these canonical colours. Participants had the option of choosing “nocolour” for particular letters, and if they did this on any trial that letter was ignored inall	
  subsequent	
  analyses	
  for	
  that	
  participant.As in Chapter 3, all RGB colours were recoded into CieLab colour space, and each partic-­‐ipant’s 26 letter colours were converted into three different distance measures: colour distance, luminance distance, and hue distance. These values were then averaged acrossall 48 synaesthetes. The same 11 measures of letter similarity were used for compar-­‐isons with the synaesthetic colour data (see Table 3.1). Also as in Chapter 3, all datawere binned from 325 letter pairs into 65 bins that each included >ive letter pairs, or-­‐dered	
  according	
  to	
  the	
  mean	
  colour	
  distance	
  across	
  all	
  participants. 4.4	
  	
  ResultsChapter 3 used four types of analyses. First, simple correlations between the 3measures of colour distance and the 11 letter similarity measures revealed that lu-­‐minance and hue were correlated with different aspects of letter similarity. Second, lin-­‐ear models revealed that three effects drove the simple correlations: a relationship be-­‐ 71 tween shape and hue, between ordinality and hue, and between frequency andluminance. Third, analyses of individual differences showed that these three effectswere independent of each other, such that the likelihood of individuals displaying any ofthem is unaffected by whether they display any of the others. Finally, by furtherbreaking down shape difference into 11 different dimensions, it was shown that onlythree of these dimensions (closed curve, repeated element, and diagonal) were relatedto	
  differences	
  in	
  hue.	
  Each	
  of	
  these	
  analyses	
  is	
  reproduced	
  below. 4.4.1	
  	
  Shape-­hue	
  and	
  ordinality-­hue	
  results	
  replicate Table	
  4.1	
  	
  Correlations between letter and colour similarity (new data set)Similarity	
  Measure Colour	
  Distance Luminance	
  Distance Hue	
  DistanceShape	
  difference 0.47*** 0.21	
  . 0.45***Frequency	
  difference -­‐0.05 0.09 -­‐0.06Ordinality	
  difference 0.64*** 0.02 0.66***Ratings	
  	
  	
  A	
  (similarity) -­‐0.35** -­‐0.18 -­‐0.32**	
  	
  	
  B	
  (similarity) -­‐0.10 -­‐0.07 -­‐0.10	
  	
  	
  C	
  (difference) 0.35** 0.23	
  . 0.31*Discrimination	
  RT -­‐0.40*** -­‐0.20 -­‐0.39**Confusion	
  	
  	
  A -­‐0.14 -­‐0.17 -­‐0.11	
  	
  	
  B -­‐0.35** 0.04 -­‐0.36**	
  	
  	
  C -­‐0.36** -­‐0.12 -­‐0.35**Letter	
  Name	
  Similarity 0.04 -­‐0.20 0.09Correlations	
  with	
  the	
  ratings	
  use	
  Spearman’s	
  rho.	
  All	
  p	
  values	
  are	
  Bonferroni	
  corrected	
  and	
  >	
  .1,	
  except:	
  .	
  p	
  <	
  .1,	
  *	
  p	
  <	
  .05,	
  **	
  p	
  <	
  .01,	
  ***	
  p	
  <	
  .001Table 4.1 displays the simple correlations between the three dimensions of colour dis-­‐tance and the 11 dimensions of letter similarity in the new data set. A comparison withTable 3.3 on p. 64 reveals a high degree of similarity between the two sets of results.Once again, there were strong correlations between both Colour Distance and Hue Dis-­‐tance with Shape Difference and Ordinality Difference. Furthermore, the overall patternof positive and negative correlations was preserved quite closely. Indeed, all of the cor-­‐ 72 relations with an absolute magnitude of over 0.1 on Table 4.1 were in the same direc-­‐tion on Table 3.3, and vice versa. In general, the correlations on Table 4.1 are stronger,with twice as many reaching signi>icance (14 as opposed to 7 on Table 3.3). One cleardifference between the two data sets, however, is that the correlation between lu-­‐minance	
  and	
  frequency	
  did	
  not	
  approach	
  signi>icance	
  in	
  the	
  new	
  data	
  set.	
  A model using shape and ordinality difference to predict hue distance (R2 = .57, p< .001) did not predict any less variance than a model using all 11 letter similaritymeasures as predictors (p > .5). However removing either shape or ordinality from thetwo-­‐predictor model led to less variance explained (p < .001 in both cases). Thus, as inChapter 3, it seems safe to say that any correlations between the various similaritymeasures and hue are entirely due to shape and ordinality effects. Furthermore, thismodel explained almost twice as much variance in hue distance as the same model us-­‐ing data from Chapter 3. The model using all 11 measures to predict luminance differ-­‐ence, on the other hand, barely approached signi>icance (R2 = .26, p = .09). Thus lu-­‐minance in the present data set was not well-­‐predicted by the similarity measures,whether	
  we	
  use	
  simple	
  correlations	
  or	
  a	
  linear	
  model. 4.4.2	
  	
  The	
  shape-­hue	
  and	
  ordinality-­hue	
  effects	
  are	
  independentIn terms of individual differences, the shape-­‐hue and ordinality-­‐hue correlations wereboth positive for 81% of participants. However these were not necessarily the same par-­‐ticipants: 65% of participants showed both effects, while 15% showed only one or theother. As in Chapter 3, there was no hint of a correlation between these mappings, astested by coding the presence or absence of each mapping as 0 or 1 for each synaes-­‐thete, or by correlating the rank order of synaesthetes on each mapping, as determinedby	
  the	
  magnitude	
  of	
  their	
  individual	
  correlations	
  (both	
  ps	
  >	
  .2). 4.4.3	
  	
  The	
  same	
  dimensions	
  of	
  shape	
  predict	
  hue	
  distanceFinally, separating the shape difference measure into distance along 11 separate dimen-­‐sions led to very similar results as in Chapter 3. Distance along three of these dimen-­‐sions was signi>icantly correlated with hue (after a Bonferroni correction): the repeated 73 element (r = .63, p < .001), diagonal (r = .45, p < .01) and upward-­opening curve (r = -­‐.36, p < .05) dimensions. Two of these three dimensions—repeated element and diagonal—were also identi>ied as important predictors in Chapter 3. The third dimension identi-­‐>ied in that chapter was the closed curve dimension, which does not have a simple corre-­‐lation with hue distance in this data set. However it is a marginal contributor to thecomplete 11-­‐predictor model (p = .05), with the only other two contributors to thismodel being distance along the repeated element (p < .001) and diagonal dimensions (p< .01). Since it had already been identi>ied as a dimension of interest in Chapter 3, itseemed	
  worthwhile	
  to	
  investigate	
  its	
  role	
  as	
  a	
  predictor	
  further.A 4-­‐predictor model using distance along the repeated element, diagonal, upward-­‐opening curve, and closed curve dimensions (R2 = .59, p < .001) explained no less vari-­‐ance than the complete 11-­‐predictor model (p > .5). Distance along each of these dimen-­‐sions was a signi>icant contributor to the 4-­‐predictor model (all ps ≤ .05), and removingany	
  of	
  them	
  from	
  this	
  model	
  results	
  in	
  less	
  variance	
  explained	
  (all	
  ps	
  ≤	
  .05).	
   4.4.4	
  	
  A	
  frequency	
  effect	
  after	
  allThe lack of a frequency-­‐luminance effect came as something of a surprise, given that it isone of the more replicated results in the literature (Beeli, Esslen, & Jäncke, 2007; CohenKadosh, Henik, & Walsh, 2007; Simner &Ward, 2008; Smilek, Carriere, Dixon, & Merikle,2007; Watson, Akins, & Enns, 2012). As there was prior reason to suspect that such aneffect exists, I re-­‐ran the frequency-­‐luminance correlation using un-­‐binned data, in or-­‐der to obtain higher power. This effect was signi>icant (r = .25, p < .001), showing thatthere is indeed a frequency-­‐luminance effect. It is surprising that the magnitude of thiseffect increased substantially with unbinned data, as the opposite is normally the case(and	
  is	
  for	
  both	
  the	
  shape-­‐hue	
  and	
  ordinality-­‐hue	
  relationships).	
   4.5	
  	
  DiscussionWith only one exception, the results from Chapter 3 are entirely replicated in the newdata set, with larger effect sizes. We can be very con>ident that the synaesthetic hue of 74 letters is strongly in>luenced both by the shape of these letters and by their positions inthe alphabet, and that this applies to synaesthetes in general, rather than simply being apeculiarity of those participants in the >irst study. It is also clear that the strength ofthese effects are independent of each other, such that a synaesthete whose hues arestrongly in>luenced by letter shape is no more likely than any other synaesthete to havea	
  strong	
  relationship	
  between	
  hue	
  and	
  alphabetical	
  order.	
  The speci>ic dimensions of shape—repeated element, diagonal and closed curve—identi>ied as driving the shape-­‐hue effect in Chapter 3 are also sign>icant predictors ofhue distance in the present data. Thus it seems likely that there is something aboutthese dimensions that are of particular importance to synaesthetes, but it is not clearwhat this is. Diagonals and curves are fairly low-­‐level visual features, but the notion ofrepeated elements (as in the letters S and M) is not, and other low-­‐level elements ofshape such as vertical and horizontal lines do not in>luence hue. Furthermore, one othershape dimension—upward-­‐opening curve—is an important predictor of hue in thepresent	
  data	
  but	
  not	
  in	
  Chapter	
  3.	
  Further	
  research	
  is	
  needed.The only effect that is not replicated with the present data is the correlation betweenfrequency and luminance, but this appears to be due to two factors: lower power as aresult of binning data, and possibly the presence of speci>ic groups of letters that havedifferent relationships with luminance, and thus whose effects might cancel out afterbinning.	
  Finally, the synaesthetes in the present data set chose colours that varied in saturation,which I suggested might increase the >idelity of these colours. The larger effect sizesfound in this chapter support this suggestion, and also open the door to increasing pow-­‐er by using raw data without binning. The next chapter takes advantage of this in a se-­‐ries	
  of	
  novel	
  analyses. 75 5	
  	
  The	
  structure	
  of	
  Czech	
  synaesthesia4 5.1	
  	
  IntroductionWe have now established a number of independent relationships between letter simi-­‐larity and synaesthetic colour similarity among English language speakers. This chaptershifts focus to a different linguistic and cultural context, examining similarity relationsbetween the synaesthetic colours of the 41 letter-­‐colour synaesthetes identi>ied atCharles University in the Czech Republic (see Chapter 2). The Czech language and edu-­‐cational system have a number of unique qualities that enable several important exten-­‐sions	
  of	
  our	
  understanding	
  of	
  the	
  in>luences	
  of	
  learning	
  on	
  synaesthetic	
  colour. 5.1.1	
  	
  Unique	
  features	
  of	
  Czech	
  letters	
  may	
  affect	
  synaesthetic	
  colourCzech uses an alphabet that is very similar to English (see Table 5.1). Thus the visualstimuli that induce Czech grapheme-­‐colour synaesthesia are more or less the same as inthe previous chapters, and indeed the same as in the vast majority of published studies,almost all of which involve languages that use some variation of the Latin alphabet (forsome welcome exceptions to this trend, see Asano & Yokosawa, 2011, 2012; Mills et al.,2002; Simner, Hung, & Shillcock, 2011). This similarity of inducers enables testing ofmany of the same effects presented in Chapters 3 and 4, but the many differences be-­‐tween	
  Czech	
  and	
  English	
  allow	
  us	
  to	
  branch	
  out. 4. See Footnote 1 on p. 26 for authorship details. 76 Table 5.1 The graphemes of the Czech alphabet including all diacriticals, their alphabetical positions, andorder	
  of	
  learning.	
  Letter AlphabetPosition LearningGroup Letter AlphabetPosition LearningGroup A 1 1 N 17 3 Á 1 1 Ň 18 8 B 2 5 O 19 1 C 3 5 Ó 19 1 Č 4 5 P 20 3 D 5 4 R 21 4 Ď 6 8 Ř 22 5 E 7 1 S 23 2 É 7 1 Š 24 5 Ě 7 7 T 25 3 F 8 6 Ť 26 8 G 9 6 U 27 1 H 10 5 Ú 27 1 CH 11 5 Ů 27 1 I 12 1 V 28 4 Í 12 1 Y 29 1 J 13 3 Ý 29 1 K 14 4 Z 30 4 L 15 2 Ž 31 5 M 16 2 5.1.2	
  	
  Does	
  phonological	
  similarity	
  map	
  to	
  synaesthetic	
  colour	
  similarity?The >irst difference to consider between Czech and English is the much higher degree oforthographic transparency of Czech letters. As explained in Chapter 2, in a perfectlytransparent language, each grapheme represents only one sound and each sound isproduced by only one grapheme (some exceptions to this in Czech are discussed below).This means that the phonological similarity of Czech letters can be measured relativelysimply. Doing this is impossible in a language as orthographically opaque as English,where the closest one can come is likely something like the phoneme co-­‐occurencescore used in Chapters 3 and 4, which only measures the similarity of letter names, not 77 the sounds they represent. Given that various types of letter similarity affect synaesthet-­‐ic colour, and acoustic properties can also affect synaesthetic colour (e.g. the stressedsyllable in an English word often determines its synaesthetic colour, Simner, Glover, &Mowat, 2006), it seems likely that phonological similarity could also affect synaestheticcolours. This might be especially true in an orthographically transparent language likeCzech,	
  where	
  phonology	
  and	
  letter	
  identity	
  are	
  almost	
  perfectly	
  mapped	
  to	
  each	
  other. 5.1.3	
  	
  Alphabetical	
  order	
  vs.	
  learning	
  order	
  in	
  CzechAnother important difference between Czech and English is that Czech students typical-­‐ly learn their letters in a set order that is entirely unrelated to alphabetical position.This enables a clear test of the learning order explanation of the ordinality-­‐hue effect.Recall that in Chapter 3, it was suggested that the ordinality-­‐hue effect is the result ofthe order in which letters are learned. If this is true, then there should be a relationshipbetween Czech learning order and synaesthetic hue. If, on the other hand, there is a re-­‐lationship between alphabetical order and hue, but no relationship between learningorder and hue, this would be very strong evidence against the learning order hypothe-­‐sis.	
  (Of	
  course	
  >inding	
  both	
  effects,	
  or	
  neither,	
  would	
  only	
  confuse	
  matters	
  more.)The alphabetical order of the Czech letters is presented in Table 5.1. This order may ap-­‐pear haphazard to an English observer, but it has a (relatively) simple phonemic justi>i-­‐cation. All the vowels have both a short and a long form, where the latter is indicated bythe accent known as the čárka (which occurs over the >irst A in the word “čárka”). Thečárka merely indicates a longer duration of pronunciation, not the underlying vowelquality, and vowels with čárkas are not considered as separate letters in their own right(e.g. they are ignored when determining the order of words in a dictionary or phonebook). The small circle over Ů (known as the kroužek, or “small circle”) lengthens thesound of U in exactly the same way as a čárka (Ú always occurs at the beginning ofwords, Ů always occurs within them), and so is not considered a unique letter either.Czech orthography also contains the háček (the diacritical over the C in the words “čár-­‐ka” and “háček”), which can be applied to several consonants and also to E. It indicates 78 a palatalized consonant or, to the English speaking ear, a 'softening' of the consonant.The phonemes differ as a result of this palatalization, and consonants with a háček areconsidered letters in their own right, occupying their own places in the Czech alphabetand dictionary. Ě is a special case, as the háček here does not modify the vowel’s sound,but rather indicates that the preceding consonant is itself palatalized, and so is not con-­‐sidered as a separate letter from E. Finally, the letter pair CH (the sound ⟨χ⟩) is also con-­‐sidered	
  as	
  a	
  unique	
  letter.Table 5.1 also gives the learning order for the Czech letters, adapted from eight recent>irst-­‐grade textbooks (Březinová, Havel, & Stadlerová, 2007; Ladová, Holas, & Staudková,2011; Melichárková, Štĕpán, & Švecová, 2008; Mikulenková & Mladý, 2004; Miku-­‐lenková, Mladý, & Forman, 2004; Nováčková, 2010; Potůčková, 2010; Žáček &Zmatlíková, 2010). As noted above, Czechs learn their letters in a highly regular order.Grade 1 textbooks present the letters in a number of distinct groups, where the lettersin each group are presented on the same page or two-­‐page spread of the textbook.While the order of letter presentation within each group varies substantially betweentexts, the composition and order of the groups themselves is highly consistent (withsome exceptions, noted below), and the pedagogical method is highly similar across theCzech Republic. The vowels and their accented forms are always taught >irst (with theexception of Ě), followed by groups of consonants. Each consonant is always introducedas a modi>ier of the vowels, thus a child learns the letter T by learning the nonsense syl-­‐lables ta, te, ti, to, tu. By the end of Grade 1 the child is expected to be able to read anyCzech text >luently, albeit without comprehension, which is of course impossible in alanguage	
  as	
  orthographically	
  opaque	
  as	
  English.These learning groups should be taken as a fairly close approximation of any givenCzech student’s order of learning the letters, rather than a perfect replication of this or-­‐der. Aside from well-­‐known phenomena such as children learning the letters of their>irst name prior to other letters (Justice, Pence, Bowles, & Wiggins, 2006), the learninggroup to which some letters belong differs slightly between textbooks. For example, Y issometimes presented later than the other vowels, and S is sometimes a member of79 group 3, not group 2. Further, the consonants in groups 4 and 5 are not generally pre-­‐sented as a single group on a two-­‐page spread, but rather are each presented on a singlepage, in an order that varies widely between textbooks (though the letters in group 4are learned before the letters in group 5). Aside from small differences such as these,however,	
  the	
  learning	
  order	
  is	
  remarkably	
  constant. 5.1.4	
  	
  A	
  special	
  role	
  for	
  vowels?Phonological similarity, alphabetical order and learning order apply across all letterrs,but there are also a number of discrete categories among Czech letters that could in>lu-­‐ence the synaesthetic colours. In each case, the prediction is that letters that are catego-­‐rized	
  together	
  will	
  be	
  closer	
  in	
  synaesthetic	
  colour.To begin with, vowels and consonants are clearly distinguished in Czech, as they are inEnglish. But this distinction may be far more salient for Czech than for English speakers,for a number of reasons. There are only >ive vowel sounds in Czech and at least twice asmany in English (the precise number depending on dialect), and thus less phonologicalvariation within the Czech vowels, which might make the vowel class easier to catego-­‐rize. Furthermore, as described above, consonants are always learned from the start ascomponents of vowel-­‐consonant morphemes, whereas the vowels are taught on theirown, and each vowel is re-­‐presented as part of learning each consonant. This specialtreatment of vowels might also increase the salience of vowels as a special class. Finally,the vowels are the >irst letters Czech children learn (see Table 5.1), which could serve tofurther	
  increase	
  the	
  strength	
  of	
  the	
  vowel	
  category.	
  Vowels being learned >irst also enables another test of the learning order hypothesis.The previous chapters have shown that letter similarity often leads to synaestheticcolour similarity, which would predict a clustering of the vowels in colour space, sincefor Czechs they are highly similar in phonology, function within the writing system, andpedagogy. However according to the learning order hypothesis, letters learned earlierare generally further apart in colour space. These two hypotheses, then, make exactlyopposite	
  predictions	
  about	
  vowels’	
  colour	
  relationships. 80 5.1.5	
  	
  Base-­diacritical	
  pairsPerhaps the most obvious categories in Czech are the ones formed by base letters andtheir diacritical variations. An intuitive hypothesis is that these would tend to be veryclose in colour, if not identical, since the base letters and their variations are so similar.An important question, however, is exactly what type of similarity matters here. Thereare at least three ways in which the base pairs can be similar to their diacritical varia-­‐tions, which do not apply equally to all base-­‐diacritical pairs. These include similarity interms of shape, identity, and phonology. By choosing comparisons carefully, it may bepossible to determine the priority of these three types of similarity in terms of in>luenc-­‐ing	
  synaesthetic	
  colour.First, bases and their diacritical forms are highly similar in shape, and we have seen inChapters 3 and 4 that shape similarity is generally correlated with hue similarity. Shapesimilarity then, would seem to apply more or less equally to all base-­‐diacritical pairs.Second, as discussed above, vowels with čárkas are considered to be variants of thesame letter as their base forms, whereas consonants with háčeks are distinct letters(see Table 5.1). This, then, might cause vowel-­‐čárka pairs to be closer together in colourthan consonant-­‐háček pairs. Third, there is a phonological relationship between basesand their diacriticals, but the precise nature of this relationship differs between čárkasand háčeks. Vowels, as noted above, are merely lengthened by their čárkas, with nochange in vowel quality, whereas háčeks indicate a different place of articulation for thephoneme. This change is rule-­‐governed (palatalization), but it is more signi>icant thanthe lengthening of vowels by the čárkas. Thus if phonological similarity drives synaes-­‐thetic colour similarity, we might again expect vowel-­‐čárka pairs to be closer together incolour	
  than	
  consonant-­‐háček	
  pairs.Merely comparing vowel-­‐čárka and consonant-­‐háček pairs, then, confounds similarity interms of abstract identity and in terms of phonology, since vowel-­‐čárka pairs are moresimilar in both ways. However a third type of category within Czech letters might allowthese two types of similarity to be disentangled. The members of each pair of voiced-­‐un-­‐ 81 voiced consonants (D-­‐T, G-­‐C, H-­CH, V-­‐F, and Z-­‐S) bear very little shape similarity to eachother, and are not considered to have the same letter identity, but their phonological re-­‐lationship is quite similar to that between consonants and their háčeked forms. Thus ifphonological relationships in>luence letter similarity, one would expect to >ind that pairsof voiced-­‐unvoiced consonants are somewhat closer to each other than they are to theother consonants. Thus by comparing how clustered vowel-­‐čárka pairs are from theother vowels, and how clustered consonant-­‐háček and voiced-­‐unvoiced pairs are fromthe other consonants, one can determine which of the three types of similarity affectsynaesthetic	
  colour.	
  	
  Table	
  5.2	
  summarizes	
  the	
  possibilities. Table 5.2 Three types of letter pairs in Czech, three ways in which the letters in these pairs are similar toeach other, and three predictions about what ought to happen if each type of similarity affects synaesthet-­‐ic	
  colour. Similar	
  shapes Same	
  letter	
  identity Similar	
  phonologyVowel-­‐čárka	
  pairs(A-­‐Á,	
  E-­‐É,	
  I-­‐Í,	
  O-­‐Ó,	
  U-­‐Ú,	
  Y-­‐Ý) Yes Yes Identical	
  phonemeConsonant-­‐háček	
  pairs(C-­‐Č,	
  D-­‐Ď,	
  N-­‐Ň,	
  R-­‐Ř,	
  S-­‐Š,	
  T-­‐Ť,	
  Z-­‐Ž) Yes No Different	
  phonemeVoiced-­‐unvoiced	
  pairs(D-­‐T,	
  G-­‐C,	
  H-­‐CH,	
  V-­‐F,	
  Z-­‐S) No No Different	
  phonemePrediction	
  for	
  synaesthetic	
  colour	
  (assuming	
  similarity	
  maps	
  to	
  similarity). Vowel-­čárka	
  and	
  consonant-­háček	
  pairs	
  should	
  cluster,	
   voiced-­unvoiced	
   pairs	
  should	
  not. Vowel-­čárka	
  pairs	
   should	
  cluster,	
  con-­ sonant-­háček	
  pairs	
   and	
  voiced-­unvoiced	
   pairs	
  should	
  not. All	
  three	
  types	
  of	
   pairs	
  should	
  cluster,	
   vowel-­čárka	
  pairs	
   closest	
  of	
  all. 5.1.6	
  	
  I	
  and	
  YOne >inal category that may in>luence Czech synaesthetic colour stems from one of therare exceptions to the orthographic transparency of Czech. I and Y both represent thephoneme ⟨ɪ⟩, and Í and Ý both represent the phoneme ⟨iː⟩. This phonemic identity, then,might translate into colour similarity, with I and Y (and their čárka forms) closer to eachother	
  than	
  they	
  are	
  to	
  the	
  other	
  vowels. 5.1.7	
  	
  Outline	
  of	
  this	
  studyThe remainder of this chapter explores each potential in>luence on Czech synaestheticcolour in turn. First, I look for evidence of each of the potential categorical effects just82 described. I then return to the correlational analyses that were the centerpieces ofChapters 3 and 4, incorporating novel measures of letter similarity derived from Czechletters’	
  frequency,	
  alphabetical	
  order,	
  learning	
  order,	
  and	
  phonology.	
   5.2	
  	
  Participants41 native Czech-­‐speaking letter-­‐colour synaesthetes were identi>ied as part of theCharles University survey described in Chapter 2. As with the English-­‐speaking partici-­‐pants in Chapter 4, all participants were con>irmed as having highly-­‐consistentgrapheme colours by the Synesthesia Battery (Eagleman, Kagan, Nelson, Sagaram, &Sarma,	
  2007).	
   5.3	
  	
  Data	
  preparationAs in Chapter 4, the >inal (third) colour assigned to each letter on the Synesthesia Bat-­‐tery was used in all analyses, and if a participant chose “no colour” for any trial, that let-­‐ter was removed from all analyses for that participant. Once again, participants’ lettercolours were transformed from RGB to CieLAB coordinates, and three separate colourdistance measures were computed: Colour, Luminance, and Hue distance. Mean dis-­‐tance scores for each letter pair were computed by averaging across all participants. Asin Chapter 4, participants had the option of selecting “no colour” for particular letters, inwhich	
  case	
  their	
  data	
  was	
  ignored	
  for	
  that	
  letter.	
  Five letter similarity measures were computed, summarized in Table 5.3. The Shape Dif-­ ference measure was adapted from the one used in Chapters 3 and 4 (Gibson, 1969),with the addition of three extra dimensions: one representing the presence or absenceof a čárka (´), one the presence or absence of a háček (ˇ), and one the presence or ab-­‐sence of the kroužek (˚). Also as in Chapters 3 and 4, Frequency and Ordinality Differencescores were computed as Weber fractions (absolute value of the difference between twoletters divided by the sum). Data for the frequency fractions were taken from the overallfrequency of each grapheme in a large corpus of Czech texts, as reported by Králik(1983). Ordinality was computed using Czech dictionary order, as given in Table 5.1. 83 Learning Order Difference was another Weber fraction, computed using the learninggroups presented in Table 5.1. Finally, Phonological Similarity of Czech letter pairs wascomputed using the SimilarityCalculator PERL script (Albright, 2006). This character-­‐izes similarity according to the system of Frisch (Frisch, 1996; as used in, e.g., Frisch,Pierrehumbert, & Broe, 2004), which de>ines the similarity of two phonological seg-­‐ments as the number of natural classes they share over the sum of all shared and non-­‐shared natural classes. A complete set of feature values for the Czech letters was cus-­‐tom-­‐developed for this study by a collaborator (John Alderete, SFU) using the Uni>iedFeature Theory of Clements & Hume (1995), which speci>ies each phonological segmentin	
  terms	
  of	
  19	
  articulatory	
  features. Table	
  5.3	
  	
  Letter	
  similarity	
  measures	
  used	
  in	
  the	
  Czech	
  study.Similarity	
  Measure DescriptionShape	
  difference Euclidean	
  distance	
  in	
  an	
  14-­‐dimensional	
  space	
  de>ined	
  using	
  the	
  ba-­‐sic	
  letter	
  shape	
  features	
  from	
  Gibson	
  (1969)	
  shown	
  in	
  Table	
  3.2,	
  with	
  the	
  addition	
  of	
  čárka	
  (´),	
  háček	
  (ˇ),	
  and	
  kroužek	
  (˚)	
  features.Frequency	
  difference Difference	
  of	
  two	
  graphemes’	
  frequencies	
  in	
  Czech	
  divided	
  by	
  the	
  sum	
  of	
  their	
  frequencies	
  (from	
  Králik,	
  1983).Ordinality	
  difference Difference	
  of	
  two	
  graphemes’	
  positions	
  in	
  the	
  alphabet	
  divided	
  by	
  the	
  sum	
  of	
  their	
  positions	
  (see	
  Table	
  5.1).Learning	
  order	
  difference Difference	
  in	
  two	
  graphemes’	
  learning	
  order	
  group	
  in	
  Czech,	
  dividedby	
  the	
  sum	
  of	
  these	
  groups	
  (see	
  Table	
  5.1).Phonological	
  similarity Phonological	
  similarity	
  between	
  Czech	
  letters.Participants provided colours for 38 letters: all 23 base letters plus all their diacriticalforms (not including CH due to experimenter error). There are 703 possible letter pairsmade from these 38 letters, and thus each similarity measure has 703 points. UnlikeChapters 3 and 4, no binning was performed on the data. This allows for increased pow-­‐er, but at the expense of effect sizes, which should be kept in mind when makingcomparisons	
  to	
  previous	
  chapters. 84 5.4	
  	
  Results 5.4.1	
  	
  Categorical	
  second-­order	
  inOluences	
  on	
  synaesthetic	
  colour	
  in	
  CzechOur Czech synaesthetes formed several categorical clusters of graphemes in colourspace, such that members of a given category were closer together than they were toother graphemes (see Figure 5.1). This is supported by a series of t-­‐tests comparing themean distances of the categories noted in the Introduction. As is immediately obviousupon glancing at Figure 5.1, there was a bimodal distribution of letter pair distances incolour space, such that bases and their diacritical variations were much closer to eachother than to other letters, in terms of both hue (t656 = -­‐16.33, p < .001) and luminance(t656 = -­‐14.61, p < .001). The base-­‐diacritical pairs could be further sub-­‐divided: vowelswith čárkas were closer together than consonants with háčeks (Hue: t11 = -­‐3.42, p < .01,Luminance:	
  t11	
  =	
  -­‐3.60,	
  p	
  <	
  .01).	
  The bimodal nature of these data meant that base-­‐diacritical pairs had to be removedfrom all analyses not speci>ically pertaining to them, because any other effects couldeasily be swamped by the strong clustering of base-­‐diacritical pairs. This is particularlytrue of the correlational analyses in the next sections, since base-­‐diacritical pairs werealso very close in terms of shape, ordinality, learning order, and phonology, which con-­‐stitute most of the measures to be correlated with colour. Thus the 17 base-­‐diacriticalpairs were removed from the remainder of analyses, leaving 686 grapheme pairs intotal.Among these 686 pairs, there was a less visually obvious, but equally signi>icant, clus-­‐tering of vowels, which were closer to each other than they were to the consonants(Hue: t415 = -­‐3.34, p < .001, Luminance: t415 = -­‐4.60, p < .001). Conversely, the consonantstended to be slightly further apart in hue from each than they were from the vowels (t603=	
  3.20,	
  p	
  <	
  .01),	
  but	
  were	
  not	
  clustered	
  in	
  luminance	
  (p	
  >	
  .4).Breaking things down further, there was at least one sub-­‐cluster within the vowels: Iand Y (and their accented forms) are closer together than they were to the other vowels 85 in hue (t79 = -­‐3.68, p < .001), but very slightly further apart in luminance (t79 = 2.54, p< .05). The only potential letter clusters mentioned in the Introduction that do not ap-­‐pear to impact synaesthetic colour were pairs of voiced-­‐unvoiced consonants, whichwere no closer together in either hue or luminance than they were to other consonants(both	
  ps	
  >	
  .25,	
  not	
  shown	
  in	
  Figure	
  5.1). Figure 5.1 Mean hue and luminance distances of all Czech letter pairs across all participants, groupedinto various categories. The “Vowel-­‐Other” pairs include E-­‐Ě, É-­‐Ě, U-­‐Ů,and Ú-­‐Ů. The I-­‐Y pairs are also in-­‐cluded	
  in	
  analyses	
  of	
  the	
  Vowel-­‐Vowel	
  group. 0" 20" 40" 60" 80" 100" 120" 0" 5" 10" 15" 20" 25" 30" 35" 40" H ue $D is ta nc e$ Luminance$Distance$ Vowel/Čárka" Vowel/Other" Consonant/Háček" I/Y" Vowel/Vowel" Consonant/Consonant" Vowel/Consonant" 86 With one possible exception, English synaesthetes did not appear to cluster any of thesecategories in colour space. While none of the tests involving base pairs with diacriticalscould be performed for the English alphabet, those tests that were meaningful in Eng-­‐lish showed no signi>icant effects in the colour data used in Chapter 4 save, curiouslyenough, for consonants being slightly closer together in luminance than they were tovowels	
  (t308	
  =	
  -­‐2.80,	
  p	
  <	
  .01;	
  all	
  other	
  ps	
  >	
  .25). 5.4.2	
  	
  Notes	
  on	
  the	
  remaining	
  analysesAs described above, the 17 base-­‐diacritical pairs were removed from all remaininganalyses. However preliminary analyses revealed that this was not always enough, andthat some important effects could not be uncovered without removing diacriticals fromanalyses entirely. Consider that among the 686 grapheme pairs there are twice as manypairs involving some version of A as pairs involving some version of B, since A can bemodi>ied by the čárka, but there is no version of Bwith a háček. E and Uwould be repre-­‐sented three times as often as a letter like B, since they each have two diacritical varia-­‐tions. Since diacritical variations tend to be very similar in colour to their base forms(see Figure 5.1), including them in the data set effectively magni>ies the importance ofthe colours of letters that have diacritical versions, giving them two or three times asmuch potential in>luence over the results as their unmodi>ied cousins. One way of pro-­‐ceeding that could avoid this would be to remove diacriticals entirely from all analyses,leaving only the 23 base letters. This would be far from ideal, however, both because itgreatly reduces power and because the diacriticals are interesting in their own right.Thus, the remainder of analyses in this chapter were performed twice, once on the com-­‐plete data set of 686 grapheme pairs (not including the 17 base-­‐diacritical pairs), andonce on the 253 pairs of base letters only. For the sake of brevity, however, results willonly be reported for the complete set of 686 pairs, except in those instances wherethere are important differences. Where the smaller data set is not explicitly mentioned,the reader can safely assume that the qualitative pattern of results was the same as in 87 the larger set (i.e. all signi>icant effects in either data set were signi>icant, or at leastmarginal,	
  in	
  the	
  other,	
  and	
  in	
  the	
  same	
  direction).Another wrinkle that became apparent during preliminary analyses is that a number ofeffects were different for vowel-­‐vowel, consonant-­‐consonant, and vowel-­‐consonantpairs. Thus the in-­‐depth analyses of each important relationship between letter andcolour similarity split the letter pairs into these three groups, although it should be keptin mind that this reduced power once more. The lack of an effect for one (or all three)groups does not necessarily mean that they did not contribute to an effect that appliesover the entire set of letter pairs. However when strong differences between the groupswere found, these may be indicative of some of the important processes that underlieCzech	
  synaesthetic	
  development. 5.4.3	
  	
  Shape-­,	
  ordinality-­,	
  and	
  phonology-­colour	
  correlations	
  in	
  CzechAs in Chapters 3 and 4, simple correlations were calculated between the various lettersimilarity measures and colour distance measures, multiplying p-­‐values by 15 to com-­‐pensate for the number of tests. For the simple correlations, the pattern of results wasvery similar for both the complete data set including diacriticals and the reduced baseletter only data set.(see Table 5.4). As one would expect, correlations in the smaller baseletter data set were generally less signi>icant, but were all in the same general range asthe results from the larger set. In both sets, shape distance was correlated with colourdistance, and this was solely due to a correlation with hue distance, as in Chapters 3 and4. Both ordinality difference and phonological similarity were correlated with colourdistance in the larger data set, but these correlations arose from relationships with bothluminance and hue. In the base letter set, none of the correlations with ordinality differ-­‐ence were signi>icant, although both the ordinality-­‐luminance and ordinality-­‐hue corre-­‐lations were marginally signi>icant. Furthermore, the correlation between phonologicalsimilarity and luminance distance was no longer signi>icant, although given that its ab-­‐solute magnitude increased from 0.11 to 0.14, this was likely a power issue (without theBonferroni	
  correction,	
  it	
  was	
  signi>icant:	
  p	
  =	
  .03).	
   88 Table 5.4 Simple correlations between Czech letter similarity measures and colour distance, calculatedusing	
  all	
  38	
  graphemes	
  (after	
  removing	
  the	
  base-­‐diacritical	
  pairs)	
  and	
  the	
  23	
  base	
  letters	
  only.All	
  Graphemes	
  (no	
  base-­‐diacriticals)(38	
  graphemes,	
  683	
  pairs) Base	
  Letters	
  Only(23	
  graphemes,	
  253	
  pairs)Similarity	
  Measure ColourDistance LuminanceDistance HueDistance ColourDistance LuminanceDistance HueDistanceShape	
  difference 0.21*** 0.03 0.21*** 0.19* 0.07 0.19*Frequency	
  difference -­‐0.07 0.02 -­‐0.09 0.09 0.04 0.07Ordinality	
  difference 0.16*** -­‐0.19*** 0.21*** 0.12 -­‐0.16	
  . 0.16	
  .Learning	
  order	
  difference 0.05 0.10 0.01 -­‐0.02 0.06 -­‐0.04Phonological	
  similarity -­‐0.15*** -­‐0.11* -­‐0.14** -­‐0.21** -­‐0.14 -­‐0.20**All	
  p-­‐values	
  are	
  Bonferroni	
  corrected	
  and	
  >	
  .1,	
  except:	
  .	
  p	
  <	
  .1,	
  *	
  p	
  <	
  .05,	
  **	
  p	
  <	
  .01,	
  ***	
  p	
  <	
  .001. 5.4.4	
  	
  Independent	
  inOluences	
  of	
  shape,	
  ordinality,	
  and	
  phonologyThe various effects reported in Table 5.4 were independent of each other. That is, eachof the aspects of letter similarity that are signi>icantly (or marginally) correlated withsynaesthetic luminance or hue accounted for different portions of the variance in lu-­‐minance or hue distance. This is supported by two linear models, one which predictedhue distance on the basis of shape difference, ordinality difference and phonologicalsimilarity, and the other which predicted luminance distance on the basis of ordinalitydifference and phonological similarity (Table 5.5). All the letter similarity measureswere	
  signi>icant	
  contributors	
  to	
  their	
  models. Table 5.5 Summary of linear models predicting hue and luminance, giving t-­‐values of each predictor and R2	
  of	
  the	
  models.Difference	
  Measure Luminance	
  Modelt-­‐values Hue	
  Modelt-­‐valuesShape	
  Difference 	
  	
  	
  	
  	
  	
  	
  N/A 6.04***Ordinality	
  Difference -­‐4.74*** 6.26***Phonological	
  Difference -­‐2.47* -­‐4.81***Model	
  R2 0.04*** 0.12***All	
  p-­‐values	
  are	
  	
  >	
  .1,	
  except:	
  .	
  p	
  <	
  .1,	
  *	
  p	
  <	
  .05,	
  **	
  p	
  <	
  .01,	
  ***	
  p	
  <	
  .001.In terms of absolute effect sizes, both models were fairly anemic. Keep in mind, however,that effect sizes would be larger with binned data as is used in Chapters 3 and 4. More 89 importantly, these models establish that the effects were independent of each other, andso a complete account of in>luences on synaesthetic colour in Czech needs to considereach of these three letter similarity measures separately. The following three sectionsconsider	
  each	
  of	
  these	
  effects	
  in	
  turn. 5.4.5 Shape-­hue effect is strongest for vowel-­vowel pairs, predictive dimensions for Czech	
  and	
  English	
  overlap Figure 5.2 The shape-­‐hue effect for Czech letters (not including base-­‐diacritical pairs). In order to im-­‐prove legibility, the Shape Difference values for Consonant-­‐Consonant pairs have been shifted slightly tothe	
  left,	
  and	
  the	
  values	
  for	
  Vowel-­‐Consonant	
  pairs	
  have	
  been	
  shifted	
  slightly	
  to	
  the	
  right.The shape-­‐hue effect was due to a general trend across all letter pairs, rather than to afew outliers (see Figure 5.2). The effect was much stronger among vowel-­‐vowel pairs (r 30# 40# 50# 60# 70# 80# 90# 100# 0.5# 1# 1.5# 2# 2.5# 3# H ue $D is ta nc e$ Shape$Difference$ Vowel2Vowel# Consonant2Consonant# Vowel2Consonant# 90 = .63, p < .001), but was still signi>icant for consonant-­‐consonant pairs (r = .16, p < .01)and	
  for	
  vowel-­‐consonant	
  pairs	
  (r	
  =	
  .16,	
  p	
  <	
  .01).	
  While the shape-­‐hue effect had virtually the same strength for both data sets (see Table5.4), the two differed markedly in terms of their usefulness for determining the dimen-­‐sions of shape that actually drove the overall effect. Indeed, this could only be done us-­‐ing the smaller data set that excluded the diacriticals. This is supported by a series oflinear	
  models,	
  described	
  below.The data set using all 38 graphemes did not allow for a sensible interpretation of thedimensions of shape that in>luence synaesthetic hue. Distance along all but 4 (Straight Vertical, Closed Curve, Diagonal, and Háček) of the 14 shape dimensions had a signi>icantsimple correlation with hue distance, even after a Bonferroni correction (all ps < .01).And all but 5 (Straight Horizontal, Straight Vertical, Symmetry, Diagonal, and Háček)were signi>icant contributors to a model predicting hue distance on the basis of all theshape dimensions (R2 = .29, p < .001). Without compelling theoretical reasons to selectfrom among the 9 or 10 dimensions that were signi>icantly associated with Czechsynaesthetic hue, it was impossible to reduce this models further, and one can bereasonably certain that some of the variance it explained was simply due to the sheernumber of predictors in play. Note that the Diagonal and Close Curve dimensions, whichwere among the few non-­‐signi>icant predictors of shape here, were important predic-­‐tors	
  in	
  both	
  sets	
  of	
  English	
  data	
  from	
  Chapters	
  3	
  and	
  4.However the smaller data set including only the base letters produced more inter-­‐pretable results. Bonferroni-­‐corrected simple correlations between distance along thevarious shape dimensions and hue distance revealed only three sign>icant correlations: upward-­opening curve (r = -­‐.20, p < .05), repeated element (r = .29, p < .001), and hori-­ zontal discontinuity (r = .24, p < .01). These three dimensions were also the only sign>i-­‐cant predictors (ps < .01, all other ps > .05) in a model predicting hue using all 11 shapedimensions (R2 = .20, p < .001). (Note that this model has 11 dimensions instead of 14,as the three diacritical dimensions had no predictive value for the base letters.) A model 91 using only these three dimensions to predict hue distance (R2 = .16, p < .001) explainedonly 3% less variance than the complete 11-­‐dimensional model (see Table 4.1), but thisdifference was nevertheless marginally signi>icant (p = .08). Using these three dimen-­‐sions alone as predictors in the complete data set explained almost as much variance asthe complete 14-­‐predictor model (R2 = .24, p < .001), but the remaining 5% of variancewas a signi>icant difference between the 3-­‐predictor and 14-­‐predictor models (p< .001), again indicating that some other dimension of letter shape was an importantpredictor. Thus these three dimensions explain almost all the variance in hue spaceattributable to shape similarity for Czechs. The repeated element dimension was a con-­‐tributor to the reduced shape models for the English data in Chapters 3 and 4, while theupward-­‐opening curve dimension was a contributor in Chapter 4, indicating a partialoverlap between the dimensions of shape that in>luence synaesthetic colour in all threesamples.	
  For the sake of completeness, an attempt was made to account for the remaining differ-­‐ence between the reduced model and the complete 11-­‐predictor one. Three dimensionswere marginal contributors to the 11-­‐predictor model (diagonal, horizontal-­opening curve, and horizontal discontinuity), and adding any one of these to the 3-­‐predictor mod-­‐el explained no less variance than the complete model (all ps > .2), while adding anyother dimension to the 3-­‐predictor model resulted in a model that was still marginallydifferent from the complete model (.05 < p < .1 in all cases). While the diagonal dimen-­‐sion was important in both sets of English data, there was no principled reason to preferit over the other two marginal predictors of hue distance, and so the present data didnot	
  support	
  any	
  further	
  conclusions. 5.4.6	
  	
  Ordinality	
  effects	
  with	
  both	
  hue	
  and	
  luminance,	
  in	
  opposite	
  directionsThe ordinality-­‐hue effect, like the shape-­‐hue effect, was strongest for vowel-­‐vowel pairs(r = .63, p < .001), and weaker for vowel-­‐consonant pairs (r = .32, p < .001) (see Figure5.3). Unlike the shape-­‐hue effect, consonant-­‐consonant pairs made no contribution tothe overall ordinality-­‐hue effect (p > .3), indeed they trended in the opposite direction. 92 The ordinality-­‐luminance effect, on the other hand, was only signi>icant for vowel-­‐con-­‐sonant pairs (r = -­‐.25, p < .001, other ps > .3) although the trend for vowel-­‐vowel andconsonant-­‐consonant	
  pairs	
  was	
  still	
  negative. Figure	
  5.3	
  	
  The	
  ordinality-­‐hue	
  effect	
  in	
  Czech	
  synaesthesia.No further reductions of the ordinality effects are possible at present. It seemed that theordinality-­‐luminance effect might be driven solely by the unusual brightness of I and O,which have often been reported as preponderantly white (cf. Simner et al., 2005), whichwould lead to a large luminance distance with other letters. In the Czech data, I, Í, O and Ó were all among the 5 brightest letters, with a mean CIE L value of 85, compared to anaverage of 71 for all other letters (t37 = 3.84, p < .001). However after removing them 30# 40# 50# 60# 70# 80# 90# 100# 0# 0.2# 0.4# 0.6# 0.8# 1# H ue $D is ta nc e$ Ordinality$Difference$ Vowel2Vowel# Consonant2Consonant# Vowel2Consonant# 93 from the data set, the ordinality-­‐luminance effect was still signi>icant for vowel-­‐conso-­‐nant	
  pairs	
  (r	
  =	
  -­‐.19,	
  p	
  <	
  .01).	
  No	
  other	
  potential	
  outliers	
  were	
  apparent	
  in	
  these	
  data. 5.4.7	
  	
  Phonology-­colour	
  relations	
  are	
  strongest	
  for	
  vowel-­like	
  consonantsThe correlation between phonological similarity and hue was signi>icant for both thecomplete and the base-­‐letter-­‐only data sets, but when these were decomposed alongvowel-­‐consonant lines only one signi>icant correlation was found, for vowel-­‐consonantpairs in the base-­‐letter-­‐only data set (r = -­‐.26, p < .01). The phonology-­‐luminance corre-­‐lation disappeared altogether when the data were decomposed into the vowel/conso-­‐nant	
  groups	
  (all	
  ps	
  >	
  .05).	
  Of course this does not mean that the phonology-­‐hue and phonology-­‐luminance effectswere spurious, simply that these effects were likely weak general trends across all letterpairs, and carving the data into three groups lowered power. However carving up thedata in this way still proved valuable, as further investigation of the one signi>icant cor-­‐relation revealed a very clean effect of phonological similarity between vowels and con-­‐sonants. In Figure 5.4, the vowel-­‐consonant pairs are presented by themselves, and fur-­‐ther split into 3 sub-­‐groups: all pairs where either the consonant or the vowel has adiacritical, all pairs involving a consonant other than J or V, and all pairs involving J or V.This scatterplot allows us to see two trends clearly. First, among the base pairs withoutdiacriticals, the phonology-­‐hue effect was driven almost entirely by J and V, which arethe closest base consonants in phonology to all the vowels. These two letters appearedto have been pulled towards the vowels by their relatively high phonological similarity,leading to the strong phonology-­‐hue effect among the vowel-­‐consonant pairs. Removingpairs involving J or V from the base-­‐letter-­‐only data set rendered the phonology-­‐hue ef-­‐fect among vowel-­‐consonant pairs insigni>icant (r = -­‐.14, p = .19), although there wasstill	
  a	
  signi>icant	
  effect	
  across	
  all	
  base	
  letter	
  pairs	
  (r	
  =	
  -­‐.13,	
  p	
  <	
  .05).Second, there was no phonology-­‐hue effect for the vowel-­‐consonant pairs if the pairs in-­‐cluding diacriticals are included, despite the fact that several diacriticals had a far high-­‐er degree of phonological similarity to the vowels than either J or V. Presumably, this 94 was because the categorical similarity between bases and diacriticals (see Figure 5.1)overrode the phonological similarity between these vowel-­‐consonant pairs involving di-­‐acriticals. This would explain why the absolute magnitude of the correlation betweenphonological similarity and colour distance went up when the diacriticals wereremoved	
  from	
  the	
  data	
  set	
  (see	
  Table	
  5.4). Figure 5.4 The phonology-­‐hue effect for vowel-­‐consonant pairs alone, demonstrating the special role of Jand	
  V,	
  and	
  the	
  lack	
  of	
  an	
  effect	
  among	
  pairs	
  that	
  include	
  diacriticals.Further investigations did not uncover any other outliers or unusual patterns in eitherphonology-­‐hue	
  or	
  phonology-­‐luminance	
  relationships.	
   5.4.8	
  	
  Is	
  there	
  any	
  impact	
  of	
  learning	
  order	
  on	
  synaesthetic	
  colour?The learning order hypothesis was not well-­‐supported by the data thus far. The vowels,which are learned >irst by Czechs, were closer together in both hue and luminance than 40# 50# 60# 70# 80# 90# 100# 0# 0.1# 0.2# 0.3# 0.4# 0.5# 0.6# 0.7# 0.8# 0.9# H ue $D is ta nc e$ Phonological$Similarity$ Consonant3Vowel#With#Diacri>cal# Base#Consonant#(Not#J/V)#3#Base#Vowel# J/V#3#Vowel# Base#LeFer#Pairs#Only# All#Consonant3Vowel#Pairs# 95 they were to the subsequently-­‐learned consonants, exactly opposite to the prediction ofthe hypothesis. Furthermore, there was no correlation between learning order and hueor luminance, while there was also a moderate correlation between alphabetical orderand	
  hue,	
  despite	
  the	
  fact	
  that	
  Czechs	
  do	
  not	
  learn	
  their	
  letters	
  in	
  alphabetical	
  order.In order to leave no stone unturned, the correlational analyses between learning orderdifference and hue and luminance distance were re-­‐run, splitting the data set into thethree consonant/vowel groups. Surprisingly, there were highly signi>icant effects. Therewas no relationship with either hue or luminance among the consonant-­‐consonant pairs(all ps > .1), nor among the consonant-­‐vowel pairs when the base letter only data setwas used (both ps > .5). However when the diacriticals were included, there was a rela-­‐tionship between learning order distance and hue distance among the vowel-­‐consonantpairs (p < .05, r = -­‐.13) the vowel-­‐vowel pairs (p < .001, r = .40), as well as a relationshipbetween learning order distance and luminance distance among the vowel-­‐vowel pairs(p	
  <	
  .05,	
  r	
  =	
  -­‐.26;	
  p	
  >	
  .5	
  for	
  vowel-­‐consonant	
  pairs).	
  It quickly became apparent that all the apparent relationships here were driven solelyby Ě, the only vowel that is not learned at the beginning of Czech letter-­‐learning (see Ta-­‐ble 5.1). Removing pairs including Ě from the data set eliminated all effects (all ps > .1).The reason for this is easy to understand in the case of vowel-­‐consonant pairs. Becausevowels, including Ě, clustered together in hue and luminance (see Figure 5.1), then Ěwould tend to be far away in hue and luminance from the consonants that it was close toin learning order, producing the negative relationship between learning order differenceand hue distance among the vowel-­‐consonant pairs. With the vowel-­‐vowel pairs, how-­‐ever, the role of Ě was due to two effects we did not originally anticipate: Ě was furtherin hue from the other vowels, on average, than the other vowels with diacriticals werefrom these vowels, and closer in luminance (hue: p < .001, t64 = -­‐3.99, luminance: p < .05, t64 = 2.29). Note that no form of p-­‐value correction was used here, and it is quite possi-­‐ble that either of these effects is spurious. Certainly there is a sensible explanation forthe hue effect: Ě is categorically distinct from vowels with čárkas or kroužeks, in that itindicates a palatalization of the immediately preceding consonant, as opposed to merely 96 lengthening the vowel. Thus it may be that its separation in hue was due to another cat-­‐egorical effect like those shown in Figure 5.1. However this does not explain the factthat it is closer together in luminance. Whatever the reason for the effects involving Ě’s,these analyses give no reason to think that there is any genuine impact of learning orderon	
  synaesthetic	
  colour	
  in	
  Czech. 5.4.9	
  	
  Chasing	
  down	
  the	
  frequency-­luminance	
  effectThe lack of a frequency-­‐luminance effect (see Table 5.4) once again came as a surprise,although this time it could not be blamed on low power alone, since the data are notbinned as they were in Chapters 3 and 4. One further analysis did indicate that such aneffect existed, but only as a >irst-­‐order effect, unlike all the other effects described in thischapter. This was by calculating, for each subject, the magnitude of the correlation be-­‐tween the raw luminance values of each letter (not distances between letter pairs) andthe raw frequency of each letter. Only one of these correlations was signi>icant on itsown (p < .05 with no Bonferroni correction), but on average, the absolute magnitudes ofthe individual correlations were above 0 (p < .01, t40 = 3.17). This was not the case whenindividual correlations were calculated between frequency difference and luminancedistance (p = .20), nor when correlations were calculated with either dimension of hue(both ps > .1). Thus there was a very small frequency-­‐luminance effect in these data, butonly	
  at	
  a	
  >irst-­‐order	
  level. 5.5	
  	
  Discussion 5.5.1	
  	
  Overview	
  of	
  resultsCzech grapheme-­‐colour syanesthetes, like their English counterparts, are in>luenced bya wide range of learned letter properties as they develop their letter-­‐colour associa-­‐tions. To begin with, they cluster their letters categorically within colour space. Base let-­‐ters are extremely close to their diacritical variations, and among these pairs, vowel-­‐čár-­‐ka pairs are closer to each other than consonant-­‐háček pairs. Vowels are slightly closerto each other than they are to consonants, and within the vowels, I, Y and their accented 97 forms are closer to each other than they are to the other vowels. As with English synaes-­‐thetes, they tend to associate similarly-­‐shaped letters with similar hues, and less similarcolours to letters that are earlier in the alphabet. Furthermore, similar-­‐sounding letterstend to be more similarly-­‐coloured, both in luminance and hue. These three effects alldiffer for different groups of letters: the shape-­‐hue and ordinality-­‐hue relationships areparticularly strong for vowel-­‐vowel pairs, and the phonology-­‐hue and phonology-­‐lu-­‐minance relationship are much stronger for vowel-­‐consonant pairs, speci>ically thoseinvolving the consonants that are most vowel-­‐like. Unlike English synaesthetes, there isno second-­‐order relationship between letter frequency and synaesthetic luminance,though there is a small >irst-­‐order effect. Finally, there is no hint of a relationship be-­‐tween	
  the	
  order	
  in	
  which	
  letters	
  are	
  learned	
  and	
  their	
  synaesthetic	
  colours. 5.5.2	
  	
  No	
  learning	
  order	
  effect,	
  a	
  special	
  role	
  for	
  sequences?These results force the rejection of the learning order hypothesis, which states thatsynaesthetes tend to assign more distinct colours to letters earlier in the alphabet be-­‐cause this is (roughly) the order they learn these letters in. Czechs do not learn their let-­‐ters in alphabetical order, but the learning order of Czech letters is not even close to be-­‐ing correlated with synaesthetic colour, while alphabetical order is. Furthermore, Czechvowels, which are the >irst letters Czech children learn, are clustered together in bothhue and luminance, while the learning order hypothesis suggests that they should bedriven widely apart. One can account for the ordinality-­‐hue effect among both Czechand English speakers by positing a single mechanism related to ordinality itself, ratherthan learning order. That is, ordinality maps on to synaesthetic colour because ordinali-­‐ty is particularly salient to synaesthetes. This is consistent with the position argued byEagleman and colleagues in recent years, who have suggested that grapheme-­‐coloursynaesthesia is a sub-­‐type of “coloured sequence synaesthesia” (Novich, Cheng, & Eagle-­‐man,	
  2011;	
  Pariyadath,	
  Plitt,	
  Churchill,	
  &	
  Eagleman,	
  2012;	
  Tomson	
  et	
  al.,	
  2011).	
  As the name implies, coloured sequence synaesthesias are those in which individualsassociate colours with items that are habitually learned as members of a sequence. This 98 may be the most common type of synaesthesia, and appears to be only weakly related toforms involving colour but not sequences, such as pain-­‐colour or orgasm-­‐colour, orforms involving sequences but not colour, such as spatial forms for numbers or time(Novich, Cheng, & Eagleman, 2011). Neuroimaging work in non-­‐synaesthetes suggestthat over-­‐learned sequences, unlike linguistic items in general, are predominantlyprocessed in the right hemisphere, speci>ically in the middle temporal gyrus and inferi-­‐or parietal lobe (Pariyadath, Plitt, Churchill, & Eagleman, 2012). Several neuroimagingstudies have shown these areas activated in synaesthetic perception involving sequen-­‐tial stimuli, however several others have not (for a review see Rouw, Scholte, & Colizoli,2011). These Czech data can be seen as further support for the idea that there is some-­‐thing	
  special	
  about	
  sequences	
  for	
  synaesthetes. 5.5.3	
  	
  Ranking	
  the	
  inOluences	
  on	
  synaesthetic	
  colourThese results also allow a tentative ranking of some of the in>luences on synaestheticcolour. For example, it appears that phonological similarity is of less importance than vi-­‐sual or ordinal similarity. Two results point towards this conclusion. First, consonant-­‐háček pairs are tightly clustered in colour space, but pairs of voiced-­‐unvoiced conso-­‐nants are not. Both groups have similar phonological relationships within each letterpair, but only the consonant-­‐háček pairs are similar in shape and ordinality, suggestingthat one or both of these two types of similarity accounts for the consonant-­‐háček ef-­‐fect. Second, there is a relatively strong phonology-­‐colour relationship for vowel-­‐conso-­‐nant pairs, due to the pair of vowel-­‐like consonants J and V being “pulled” towards thevowel cluster. However this effect vanishes when letters with diacriticals are included inthe analysis. Despite the fact that many of these letters are much more vowel-­‐like thaneither J or V (Figure 5.4), there is no hint that this affects their synaesthetic colours. Isuggest this is because their colour is far more determined by their strong relationship(in terms of shape, ordinality, or abstract identity) with their base letters. While phonol-­‐ogy clearly affects synaesthetic colour, it is trumped by shape and ordinality in thepresent	
  data. 99 This is in keeping with a report of a native English-­‐speaking synaesthete who learnedRussian in high school, long after her English letter colours had stabilized. Her coloursfor Cyrillic letters were strongly in>luenced by their visual and phonological similarity toEnglish letters, however in a case where a Cyrillic letter was visually similar to one Eng-­‐lish letter and phonologically similar to another, it usually took the colour of the similar-­‐ly-­‐shaped	
  letter	
  (Mills	
  et	
  al.,	
  2002).	
  The relative importance of shape and ordinality cannot be determined from these data.The shape-­‐colour effect in the present data is slightly stronger in absolute magnitude,but the ordinality-­‐hue effect is stronger for the English synaesthetes in Chapter 4. Vow-­‐el-­‐čárka pairs are closer in hue than consonant-­‐háček pairs, and they are closer in termsof ordinality while being (arguably) no closer in terms of shape. However this does notshow that ordinality is in any way overriding the effects of shape, simply that it can addto	
  it.	
  These	
  data,	
  then,	
  do	
  not	
  allow	
  the	
  shape	
  and	
  ordinality	
  to	
  be	
  disentangled.	
   5.5.4	
  	
  A	
  special	
  role	
  for	
  vowels?There seems to be a special role for vowels in Czech synaesthesia. Vowels are somewhatclustered together in colour space, but they also seem to be the prime movers of severalof the other similarity relationships. For instance, the shape-­‐hue and ordinality-­‐hue re-­‐lationships are far stronger for vowel-­‐vowel pairs than they are for consonant-­‐conso-­‐nant pairs, indeed of the effects discussed in this chapter, only the shape-­‐hue effect has adetectable impact on consonant-­‐consonant pairs at all. Furthermore, the phonology-­‐hueeffect	
  appears	
  to	
  be	
  driven	
  largely	
  by	
  the	
  similarity	
  of	
  J	
  and	
  V	
  to	
  the	
  vowels.	
   5.5.5	
  	
  Ambiguous	
  evidence	
  for	
  a	
  hue/luminance	
  splitEvidence of a hue/luminance split is more ambiguous in the Czech data than those fromthe English synaesthetes in Chapters 3 and 4. As in the English data, there is a shape-­‐hue effect but no shape-­‐luminance effect, and a frequency-­‐luminance but no frequency-­‐hue effect. Further, while there are both ordinality-­‐hue and ordinality-­‐luminance effectsof roughly equal magnitude, they are in opposite directions, indicating that they cannotbe due to the same factors. Thus for the similarity dimensions that were tested among 100 English synaesthetes, we replicate the >inding of a hue/luminance split. However, all thecategorical effects shown in Figure 5.1 are more or less equivalent in hue and lu-­‐minance, save that I and Y are only clustered within hue space. Also, the phonology-­‐hueand phonology-­‐luminance effects seem essentially indistinguishable from each other,especially among the base letter pairs where they are strongest. Further work is neededto determine why the hue/luminance split occurs for some types of similarity but notothers.What these results establish beyond the shadow of a doubt is that synaesthetic coloursof Czech speakers re>lect a number of learned properties of letters. Thus Czechgrapheme-­‐colour synaesthesia, like English grapheme-­‐colour synaesthesia, encodes asurprising	
  amount	
  of	
  information	
  about	
  its	
  inducer	
  domain. 101 6	
  	
  Grapheme-­colour	
  synaesthesia	
   beneRits	
  rule-­based	
  category	
  learning5 6.1	
  	
  IntroductionWe have now seen strong evidence for the effect of learning on synaesthesia. It has beenestablished that the likelihood of developing (or retaining) synaesthesia is dependentupon childhood learning challenges (Chapter 2), and that speci>ic synaesthetic associa-­‐tions encode a great variety of information about the inducer domain (Chapters 3-­‐5).Now we examine the other direction of the relationship, verifying if synaesthesia can beuseful	
  for	
  learning.Is synaesthesia good for anything? The suggestion that it has some utility goes back wellover a century (Calkins, 1893; Calkins, 1895) and recent work has begun to con>irmthis. Grapheme-­colour synaesthetes, who experience letters and numerals as having spe-­‐ci>ic colours, have episodic memory advantages for letters and words (Radvansky, Gib-­‐son, & McNerney, 2011; Rothen & Meier, 2010a; Yaro & Ward, 2007), calendar-­form synaesthetes, who experience dates as located in peripersonal space, have advantagesfor remembering events and dates (Simner, Mayo, & Spiller, 2009), and several varietiesof synaesthesia are associated with enhanced perceptual discrimination (Banissy,Walsh, & Ward, 2009; Saenz & Koch, 2008). However it remains an open question 5. This chapter is slightly adapted from a previously published paper (Watson, Blair, Kozik, Akins, & Enns, 2012). I was the primary person involved in determining the research question and experimental method, though all my collaborators made many useful suggestions and changes. All analyses were performed by myself, and I was the primary author of the paper. 102 whether synaesthesia can be exploited for more sophisticated and abstract forms oflearning (Brang & Ramachandran, 2011). Here we answer this question in the af>irma-­‐tive	
  for	
  rule-­‐based	
  category	
  learning.We investigated whether grapheme-­‐color synaesthetes are able to use synaestheticcolours on a dif>icult category learning task. We show that synaesthetes viewing blackletters use their internally-­‐generated colours during this task in much the same way asnon-­‐synaesthetes viewing genuinely coloured stimuli. Thus synaesthesia can be a toolused	
  in	
  learning	
  novel	
  abstractions.Participants learned to classify stimuli according to a rule-­‐based category structure.Such learning is hypothesized to involve an explicit reasoning process in which hypothe-­‐ses are maintained in working memory, individual stimuli are attended to and catego-­‐rized according to the currently active hypothesis, and this hypothesis is eitherstrengthened or modi>ied on the basis of subsequent feedback (Ashby & Maddox, 2005).The particular 4-­‐category structure we created was structurally similar to one used byMaddox, Filoteo, Hejl, and Ing (2004), in that the category rules conjoin two distinctpieces of information. Such conjunctive rules are frequently taught in primary school,for example when learning English phonetics (e.g. a vowel followed by a consonant hasa short pronunciation, unless the consonant is immediately followed by the letter ‘e’, inwhich case the >irst vowel has a long pronunciation), in mathematics (e.g. a number isprime if it can be divided by 1 and not by any other number), or in the sciences (e.g. amammal	
  is	
  an	
  animal	
  with	
  warm	
  blood	
  that	
  gives	
  birth	
  to	
  live	
  young).Stimuli were pairs of graphemes (see Figure 6.1a) whose category membership couldbe determined by simple rules involving the order and associated colours of graphemes,e.g. ‘‘Members of category 1 contain a green followed by a pink grapheme’’. As synaes-­‐thetes’ colours are idiosyncratic, a different stimulus set was generated for each synaes-­‐thetic participant. Participants who discovered the colour rules were expected to bemore accurate on the initial Category Learning task than those who did not. Other parti-­‐cipants would have to resort to more complex rules based on all possible combinations 103 of the eight graphemes in the stimulus set, to use explicit memorization of 16 stimulus-­‐category pairs, or to resort to more idiosyncratic strategies – e.g. treating letter pairs asacronyms	
  for	
  words	
  or	
  phrases	
  with	
  personal	
  meaning. 104  a. Category Learning Stimuli   1    2      3    4  GH   GT   4H   4T  GK   G6   4K   46  AH   AT   YH   YT  AK   A6   YK   Y6 b. Transfer Test Stimuli   1    2      3    4  3K   AJ   YP   4J  3P   36   BH   B6    3J   BP c. Recognition Test Stimuli - Foils   1    2      3    4  3H   GJ   4P   BT  GP   3T   BK   BJ  AP      YJ d. Recognition Test Stimuli - Novel Stimuli WD Z5 FR 9S UF Q8 Figure 6.1 One of the stimulus sets, based on the color assignments of one of the synaesthetes, for thevarious measures in the experiment. Synaesthetes and members of the Control-­‐Achromatic group wouldhave been presented with these stimuli in black. (a) Stimuli used during the Category learning task andRecognition Test, arranged so that the color rules are obvious. (b) The 10 stimuli used during the TransferTest. (c) The 10 Foil Stimuli used during the Recognition Test. (d) The six completely novel stimuli usedduring	
  the	
  Recognition	
  Test.This initial task was followed by a Transfer Test and Recognition Test, both designed toverify if participants were using colour rules. Immediately following the categorylearning task, participants completed 10 Transfer Test trials, in which novel stimuli thatfollowed the same colour rules were presented (Figure 6.1b). Participants who had usedcolour rules previously ought to be able to apply them to these novel stimuli, whereasthose who used other strategies should be at chance. This Transfer Test was followed bya Recognition Test, on which the opposite pattern of results was expected. Here partici-­‐ 105 pants were presented with grapheme pairs and asked if they had been presented previ-­‐ously in the experiment. These stimuli included all 16 stimuli from the CategoryLearning task (Figure 6.1a), 10 novel Foil Stimuli that also followed the colour rules(Figure 6.1c), and six additional stimuli with no colours or identities in common withany others used during the experiment (Figure 6.1d). Subjects who had used colourrules would be expected to confuse the 10 Foil Stimuli with those previously presentedin the category learning task and Transfer Test. However those using alternative ruleswould	
  be	
  expected	
  to	
  correctly	
  reject	
  more	
  of	
  them.Three groups participated in the study. A Synaesthete group viewing achromatic stimuliwas compared with non-­‐synaesthetes viewing either the same achromatic stimuli (Con-­‐trol-­‐Achromatic) or stimuli that were coloured according to synaesthetic colour assign-­‐ments (Control-­‐colour). Thus if synaesthetic colours can be used in rule-­‐based catego-­‐rization tasks, we expect the Synaesthete group to perform better than the Control-­‐Achromatic group on the category learning task and the Transfer Test, but worse on theRecognition Test. Comparing the Synaesthete and Control-­‐colour groups allows us to in-­‐fer further similarities and differences between synaesthetic and normal colourperception. 6.2	
  	
  Experiment	
  1 6.2.1	
  	
  ParticipantsTen grapheme-­‐colour synaesthetes participated in the study and were rewarded with$10 (CAN). All synaesthetes’ grapheme-­‐colour associations were veri>ied as consistentby the online Synaesthesia Battery (Eagleman, Kagan, Nelson, Sagaram, & Sarma, 2007),with a mean consistency score of .70, and a mean accuracy score of 89% on the Speed-­‐Congruency Test. Eighty-­‐ six non-­‐synaesthetes were recruited from undergraduate psy-­‐chology classes at the University of British Columbia. Six of these participants wereremoved from all analyses for performing at chance, leaving 80 non-­‐synaesthetic partic-­‐ 106 ipants. Eight of these participants were randomly assigned to each synaesthete’s stimu-­‐lus	
  set,	
  4	
  to	
  the	
  Achromatic	
  and	
  4	
  to	
  the	
  colour	
  condition. 6.2.2	
  	
  Displays	
  and	
  responsesStimuli consisted of grapheme pairs presented in Arial font, each grapheme occupyingapproximately 2.5 cm2 (3.6° of visual angle at a distance of 40 cm). Graphemes were ei-­‐ther all black (Achromatic condition) or coloured as the synaesthete reported them(colour condition). Each Category Learning and Transfer Test trial began with a >ixationcross (approximately 1.5 cm2, or 2.1° v.a.) at the center of the screen for 400–800 ms,followed by a stimulus at the center of the screen and four response boxes near the bot-­‐tom, labeled with the digits 1–4. Participants selected one box with a mouse click, andwere given feedback in the form of the incorrect response boxes disappearing. Partici-­‐pants responded to the feedback by clicking on the correct box, and the next trial began.Recognition Test trials had identical displays, save that there were only two re-­‐ sponseboxes,	
  labeled	
  ‘‘Yes’’	
  and	
  ‘‘No’’,	
  and	
  no	
  performance	
  feedback	
  was	
  given. 6.2.3	
  	
  Category	
  structureA category structure of 16 letter pair stimuli was created for each of the 10 synaes-­‐thetes. These were generated from eight graphemes, which were associated with fourdistinct colours (see Figure 6.1a), organized such that each colour appeared only in theleft or right position. Stimuli were assigned to one of four categories on the basis of sim-­‐ple conjunctive colour rules, as illustrated in Figure 6.1a that can be easily applied in a2-­‐stage hierarchy. For example, one could begin a trial by looking at the left-­‐ hand letter,and narrowing down the possible responses to categories 1 and 2 if the letter is blue or3 and 4 if it is red. Then the colour of the right hand letter could be used to determinewhich of the two remaining options is correct, since the possible responses are 1 and 3if this letter is orange or 2 and 4 if it is green. Of course these colour rules would be un-­‐available to non-­‐ synaesthetes viewing achromatic letters, and we expected their perfor-­‐mance	
  to	
  suffer	
  accordingly. 107 6.2.4	
  	
  Foil	
  stimulus	
  setsIn addition to the stimulus sets shown to participants during Category Learning, weconstructed stimulus sets following the same colour rules, but including novelgraphemes, for use in the subsequent Transfer and Recognition Tests, illustrated in Fig-­‐ure 6.1b and c. Within each set, four additional graphemes were used, each associatedwith one of the four colours from the learning phase. Combined with the originalgraphemes, this allowed for the construction of 20 more grapheme pairs (>ive new stim-­‐uli in each category) that followed the same colour mapping as in the learning task. Tenof these stimuli were randomly selected to appear in the Transfer Test, and the other 10appeared during the Recognition Test. Again, participants in the Achromatic conditionsaw	
  the	
  same	
  letter	
  pairs,	
  but	
  coloured	
  black. 6.2.5	
  	
  ProcedureCategory Learning consisted of 256 trials in total, divided into eight blocks of 32 trials.Each block contained each of the 16 stimuli presented twice in random order. On eachtrial participants indicated which category a stimulus belonged to and were given feed-­‐back as described above. Immediately following these eight blocks, participants com-­‐pleted 10 Transfer Test trials that were identical in format, except that the stimuli weredrawn from the Foil Stimuli. Other than the sudden appearance of novel stimuli, partici-­‐pants	
  were	
  given	
  no	
  indication	
  that	
  anything	
  was	
  different	
  on	
  these	
  trials.Participants then completed 32 Recognition Test trials, where they were asked to indi-­‐cate if they had seen a particular stimulus previously during the experiment. They indi-­‐cated their response by clicking on one of two boxes, labeled ‘‘Yes’’ and ‘‘No’’, and werenot given any feedback. The stimuli presented in this phase consisted of all 16 originalstimuli, the 10 Foil Stimuli that had not been used in the Transfer Test, and six newgrapheme pairs unrelated to any of the other stimuli in the experiment. Thus, half of thestimuli in the Recognition Test had been seen previously and half had not. We were par-­‐ 108 ticularly interested in participants’ responses for the 10 Foil Stimuli, as someone payingattention	
  to	
  colour	
  might	
  be	
  expected	
  to	
  make	
  False	
  Recognition	
  errors	
  on	
  these	
  trials.Finally, participants were asked to write down any strategies they used during the Cate-­‐gory	
  Learning	
  phase	
  of	
  the	
  experiment. 6.2.6	
  	
  Behavioral	
  resultsFor each participant, we computed four scores. Categorization Accuracy was the meanaccuracy over each block of the category learning task, and response times (RTs) werealso recorded during these blocks. Transfer Accuracy was the mean accuracy over the10 Transfer Test trials. False Recognition was the inverse of the mean accuracy over the10 Recognition Test trials that used the Foil Stimuli. (Recognition Test accuracy for theother	
  stimuli	
  was	
  over	
  95%	
  for	
  all	
  groups,	
  and	
  so	
  was	
  not	
  analyzed	
  further.)The results were qualitatively very simple. First, accuracy on the Category Learning taskwas higher for those with access to colour information, whether these colours weresynaesthetic or real (see Figure 6.2a), although synaesthetes learned somewhat moreslowly than controls viewing real colours. Second, participants looking at achromaticletters were slower to make decisions, whether they were synaesthetes or controls, andthe synaesthetes were generally slowest of all. Third, access to colour information alsoimproved participants’ ability to generalize to novel stimuli on the Transfer Test, al-­‐though real colours provided more of an advantage than synaesthetic colours (see Fig-­‐ure 6.2b). Finally, participants with access to colour information were prone to FalseRecognition of the Foil Stimuli during the Recognition Test, but those without colourwere	
  able	
  to	
  correctly	
  re-­‐	
  ject	
  most	
  of	
  these	
  stimuli	
  (see	
  Figure	
  6.2c).These qualitative descriptions are supported by analyses of variances (ANOVAs) andpost hoc group comparisons. To begin with, Categorization Accuracy and RT were thedependent measures in two-­‐way ANOVAs using Group as a between-­‐subjects factor withthree levels (Synaesthete, Control-­‐Achromatic, and Control-­‐colour), and Epoch (1–4,each composed of two experimental blocks) as a within-­‐subjects factor. In both cases,there were signi>icant main effects of Group (Categorization Accuracy: F2,87 = 12.1, η2 = 109 .22, MSE = .11, p < .001; RT: F2,87 = 11.5, η2 = .21, MSE = 2.1, p < .001) and Epoch (F3,261 =146.5, 31.4, η2 = .61, .26, MSE = .01, .42, respectively; both ps < .001), as well as Block byEpoch interactions (F6,261 = 2.5, 2.3, η2 = .02, .04, MSE = .01, .42, respectively; both ps< .05). These were followed by tests of the simple main effect of Group at each of thefour levels of Epoch, which all indicated group differences (F2,261 between 5.0 and 19.3, η2 between .04 and .12, MSE for accuracy between .06 and .14, for RT between 2.1 and8.1, all ps < .01; except for RT on epoch 4, where F2,261 = 3.2, η2 = .02, MSE = 1.3, p = .04).The Tukey–Kramer method was used to determine which groups were signi>icantly dif-­‐ferent from each other on each of the four epochs, and these results are describedbelow.In the case of Categorization Accuracy, as shown in Figure 6.2a, the interaction stemsfrom the Synaesthete group improving at a faster rate (a rise of over 40% from epochs 1to 4) than either control group (both of whom improve by approximately 30%). TheControl-­‐colour group outperforms the Control-­‐Achromatic group by 15–20% through-­‐out the experiment, while the Synaesthete group begins by performing similarly to theControl-­‐Achromatic group on the >irst epoch, but on Epochs 2–4 is signi>icantly more ac-­‐curate than the Control-­‐Achromatic group, and not distinguishable from the Control-­‐colour	
  group.In the case of RT, the interaction also stems from the Synaesthete group improving at afaster rate (an overall gain of 1.7 s from an initial RT of 4.9 s in Epoch 1) than either con-­‐trol group (the Control-­‐colour and Control-­‐Achromatic groups improve by 0.6 s from 2.1s and by 0.8 s from 2.8 s, respectively). Despite this greater improvement, the Synaes-­‐thete group was slower than both control groups on all epochs save for epoch 4, whereit was not distinguishable from the Control-­‐Achromatic group. The Control-­‐colour groupwas faster than both other groups on all epochs save for epoch 3, where it was not dis-­‐tinguishable	
  from	
  the	
  Control-­‐Achromatic	
  group. 110 0.2$ 0.4$ 0.6$ 0.8$ 1$ Epoch$1$ Epoch$2$ Epoch$3$ Epoch$4$ M ea n% A cc ur ac y% Categoriza1on%Accuracy% Synaesthete$ Control9Color$ Control9Achroma<c$ 0.2$ 0.4$ 0.6$ 0.8$ 1$ Synaesthete$ Control$$$$$$$$ Color$ Control$ Achroma<c$ M ea n% A cc ur ac y% Transfer%Accuracy% 0.2$ 0.4$ 0.6$ 0.8$ 1$ Synaesthete$ Control$$$$$$$$ Color$ Control$ Achroma<c$ M ea n% Er ro r% Ra te % False%Recogni1on% a$ b$ c$ *$ *$ *$ *$ *$ *$ *$ **$ **$ Figure 6.2 Performance of participants in Experiment 1. (a) Accuracy over the course of the CategoryLearning task (each epoch = 64 trials), (b) accuracy over the 10 Transfer Test trials in which participantscategorized foils that follow the same color mapping rules, and (c) error rate over the 10 Recognition Testtrials involving novel Foil Stimuli. Error bars indicate plus/minus one standard error of the mean. Aster-­‐isks	
  indicate	
  signi>icant	
  group	
  differences.	
  *:	
  p	
  <	
  .05,	
  **:	
  p	
  <	
  .01. 111 The remaining two measures (Transfer Accuracy and False Recognition) were depen-­‐dent variables in one-­‐way ANOVAs using Group as a between-­‐subjects factor. Levene’stest showed a violation of the homogeneity of variance assumption for Transfer Accura-­‐cy (F2,87 = 4.5, p < .05) so Welch’s statistic was used. Both ANOVAs were signi>icant(Transfer Accuracy: F2,22.2 = 39.4, η2 = .46, MSE = .06, p < .001; False Recognition: F2,87 =24.2, η2 = .36,MSE = .04, p < .001), indicating group differences on each of the measures.Following up with Tukey–Kramer revealed that the Control-­‐colour group had the high-­‐est Transfer Accuracy (78%), then the Synaesthete group (56%), followed by the Con-­‐trol-­‐Achromatic group (32%), and all 3 between-­‐group comparisons were signi>icant(all ps < .05). Finally, the Synaesthete and Control-­‐colour groups both performed poorlyon the Recognition Test trials using Foil Stimuli (False Recognition of 48% and 55%, re-­‐spectively) whereas the Control-­‐ Achromatic group made far fewer errors (24%) thaneither	
  (both	
  ps	
  <	
  .01). 6.2.7	
  	
  Self-­report	
  dataThe data from participants’ reports of their own strategies also support the notion thatgroup performance differences stemmed from the availability of colour-­‐based rules. Re-­‐viewing these reports revealed a number of common strategies, including the use ofacronyms (mentioned by 14% of participants), memorization (60%), various forms ofmathematical reasoning (11%), the use of colour information (49%), and explicit de-­‐scriptions of the formal category structure (22%). Fisher’s Exact Test was used to see ifthe proportions of participants reporting each strategy differed between groups. Thiswas the case for memorization (p = .002), the use of colour (p < .001) and describing thestructure (p < .001), but not for acronyms (p > .9) or math (p > .25). The Control-­‐Achro-­‐matic group was the source of all three group differences, as removing this group result-­‐ed in no signi>icant differences between the Synaesthete and Control-­‐colour groups (all ps > .5). Speci>ically, the Control Achromatic group was more likely to report memoriza-­‐tion (80%) than the Synaesthete (50%) or Control-­‐colour (43%) groups, and less likely 112 to report the use of colour (0% vs. 80% and 90%, respectively) or to describe the cate-­‐gory	
  structure	
  (3%	
  vs.	
  40%	
  and	
  38%,	
  respectively).Finally, we veri>ied whether these strategies were connected to performance using >iveANOVAs, each using Group and one of the strategies described above as between-­‐sub-­‐jects factors, and accuracy on the >inal category learning epoch as the dependentmeasure. There were main effects of describing the category structure (F = 7.1, η2 = .08, MSE = .04, p < .01) and mathematical reasoning (F = 6.8, η2 = .05,MSE = .04, p < .01), andan interaction between the use of colour information and group (F = 5.2, η2 = .05, MSE =.04, p < .05). No other main effects of strategy or interactions were signi>icant (all ps >.05). The two main effects of strategy were due to participants who described the cate-­‐gory structure performing better than those who did not (mean accuracy on Epoch 4 =94% vs. 73%) and those who used mathematical reasoning performing worse thanthose who did not (60% vs. 80%). The group by colour interaction was followed withtests of the simple main effect of colour, which was marginally signi>icant for the Con-­‐trol-­‐colour group (F = 3.8, η2 = .04,MSE = .15, p = .06), but not for the Synaesthete group(p > .25). Among Control-­‐colour participants, Tukey–Kramer revealed that participantswho used colour had a higher accuracy than those who did not (87% vs. 59%, respec-­‐tively). As only two synaesthetes did not report using colour, the lack of a main effect islikely	
  uninformative	
  for	
  this	
  group. 6.3	
  	
  Experiment	
  2The overall pattern of results from Experiment 1 is consistent with the claim thatsynaesthetes can exploit their colour experiences during category learning, in much thesame manner as non-­‐synaesthetes viewing real colours. Indeed, the only accuracy dif-­‐ferences between the Synaesthete and Control-­‐colour groups were that the synaes-­‐thetes were slightly slower to learn the category structure, and somewhat less accurateon the Transfer Task. But it was still possible that the superior performance of theSynaesthete and Control-­‐colour groups was not due to their using colour rules per se.For instance, it is possible that colour by itself makes the categorization task easier to 113 learn, irrespective of any category rules: perhaps it is simply easier to memorize letterpairs when they are coloured, and hence easier to apply mnemonic strategies to learnthe category structure. To see if this was the case, we ran a second experiment usingnew subjects, identical to Experiment 1 save that colours were no longer diagnostic ofcategory membership. If the results from Experiment 1 were indeed due to participantsin the Synaesthete and Control-­‐colour groups making category decisions on the basis ofcolour rules, then they should not be able to do this in Experiment 2, and we should not>ind	
  the	
  group	
  differences	
  we	
  found	
  in	
  Experiment	
  1. 6.3.1	
  	
  MethodsThe experimental procedure was identical to Experiment 1, save that similarly-­‐colouredletters were not grouped in the same categories, so colours were no longer diagnostic ofcategory membership. Eight synaesthetes participated in the experiment, whosegrapheme-­‐colour associations were veri>ied as consistent by the Synethesia Battery(mean consistency score: .66, mean Speed-­‐Congruency accuracy: 87%), along with 68non-­‐synaesthetic controls, four of whom were eliminated from the analysis as randomresponders, leaving 64 non-­‐synaesthetes who were randomly assigned to a particularsynaesthete’s stimulus set, and to a colour or Achromatic condition, as in Experiment 1.None	
  of	
  the	
  participants	
  were	
  in	
  Experiment	
  1.As in Experiment 1, stimuli were composed of pairs of graphemes, made from eightgraphemes with four distinct colours. These were organized into four categories of fourstimuli each. However graphemes with each of the four colours appeared in at leastthree of the four categories, and at least once on the left and once on the right-­‐hand sideof	
  different	
  stimuli.	
  Thus	
  colour	
  was	
  entirely	
  useless	
  for	
  categorization. 6.3.2	
  	
  ResultsIn brief, the three groups perform similarly to each other on all tasks, with only one ex-­‐ception. Furthermore, all three groups also perform very similarly to the Control-­‐Achro-­‐matic group from Experiment 1. Thus it is clear that the group differences we found inExperiment	
  1	
  are	
  not	
  due	
  to	
  colour	
  per	
  se,	
  but	
  to	
  its	
  use	
  in	
  a	
  system	
  of	
  rules. 114 To support this conclusion, we performed the same analyses on the same variables as inExperiment 1. No group differences or interactions were found (all ps > .3) save forFalse Recognition (F = 5.4, η2 = .14, MSE = .05, p < .01). Tukey’s HSD revealed that thisgroup difference was due to the Control-­‐colour group performing signi>icantly worse (p< .01) than the Control-­‐Achromatic group (False Recognition of 57% and 28%, respec-­‐tively). Furthermore, with the exception of the Control-­‐colour group’s False Recognition,all groups’ mean performance on all measures was within the 95% con>idence intervalof	
  the	
  performance	
  of	
  the	
  Control-­‐Achromatic	
  group	
  on	
  Experiment	
  1. 6.4	
  	
  DiscussionThese results demonstrate that synaesthetes can learn rule-­‐based categories using in-­‐ternally-­‐generated synaesthetic colours. Moreover, they do this similarly to non-­‐synaes-­‐thetic individuals using physical colours. Both synaesthetes and non-­‐synaesthetic parti-­‐cipants viewing coloured stimuli learned to categorize more successfully than non-­‐synaesthetes viewing achromatic stimuli, were able to generalize to novel stimuli on thetransfer task, and were unable to correctly reject Foil Stimuli in a Recognition Test, indi-­‐cating that their memory for individual grapheme identities was impaired. Further-­‐more, these participants were also likely to give explicit reports indicating that theyused the colour information and understood the category structure, unlike non-­‐synaes-­‐thetes viewing achromatic stimuli, and giving these reports was correlated with higheraccuracy. Taken together, these >indings demonstrate that synaesthetes can exploit theirgrapheme colours to learn a rule-­‐based category structure similar to those taught in avariety	
  of	
  domains.More detailed analyses showed some performance differences between synaesthetesand non-­‐synaesthetes viewing physically coloured stimuli. First, synaesthetes learnedmore slowly. Though their performance for most of the experiment was comparable tonon-­‐synaesthetes viewing real colours, their accuracy was lower at the start of the ex-­‐periment. We suggest that this is because experiences of synaesthetic colours may be 115 somewhat less vivid than experiences of real colours, which might delay ruleacquisition.Second, synaesthetes were not as successful in transferring their learning to novel stim-­‐uli. This might also be explained by less vivid synaesthetic experiences. Alternatively, acomment made by a synaesthetic participant may shed light on this result. He indicatedthat when viewing the stimuli, he did not experience two different colours, but saw asingle colour for the pair as a whole, typically the colour of the grapheme that seemedmore ‘‘dominant’’ than the other. Indeed, many grapheme-­‐colour synaesthetes experi-­‐ence single colours for words, often determined by the colour of an individual letter(Simner, Glover, & Mowat, 2006; Ward, Simner, & Auyeung, 2005). This may account forthe lower accuracy of synaesthetes on the transfer task, although it does not mitigatethe critical >inding that their accuracy was almost twice that of non-­‐synaesthetes view-­‐ing	
  achromatic	
  grapheme	
  pairs.Third, synaesthetes were slower to respond than participants viewing real colours.There are at least two ways of accounting for this result. First several researchers arguethat synaesthetic colours cannot be induced without the conscious recognition of thegrapheme (e.g. Laeng, 2009). This would imply that the Synaesthete group ought to re-­‐spond at least as slowly as the Control-­‐Achromatic group, which is what we >ind. An al-­‐ternative is that the process of establishing which letter in a pair is dominant, as de-­‐scribed in the previous paragraph, may take some time to resolve itself. The presentdata does not provide enough evidence to decide whether one or both of these is thetrue	
  source	
  of	
  the	
  reaction	
  time	
  differences.How well might these results generalize to other tasks? Stimuli in Experiment 1 werespeci>ically tailored to each synaesthete such that their personal colour associationswould be maximally informative for distinguishing between the four categories. Itseems remarkably unlikely that this could happen by chance. Thus one would be justi-­‐>ied in asking whether our results have any meaning outside the laboratory. Are the ap-­‐parently arbitrary associations synaesthetes make between graphemes and colours ac-­‐ 116 tually any use in learning or using the rules of, for example, spelling, mathematics orphonetics?Our data do not directly address this question, but there is reason to think that synaes-­‐thetic colours could provide a signi>icant bene>it to such rule use. Many of the explicitrules we learn in everyday life – including all the examples given in Section 1 – are sin-­‐gle rules that do not require combining with other rules in a hierarchical fashion, as thecolour rules in this study do. Any rule that involves a speci>ic combination of letters –e.g. ‘‘I before E except after C’’ – is one that involves a speci>ic combination of colours fora grapheme-­‐colour synaesthete. Provided the synaesthete’s colours for these letters aredistinguishable from each other, this could provide a cue to aid in learning and applyingthe rule. The same is true for any numerical rule in math – e.g. any number ending in 5is divisible by 5. The present study shows that with less than 30 min training, synaes-­‐thetes can >lexibly employ their colours to learn and use a complex and abstract set ofintertwined rules. We see no reason why they could not do the same for simpler rules inthe	
  classroom	
  or	
  in	
  the	
  rest	
  of	
  daily	
  life.Of course establishing that this is possible is one thing, verifying that it occurs under na-­‐tural conditions is another matter. There are no published studies that test this hypothe-­‐sis. There are anecdotal reports, and several savants attribute their astounding memoryand mathematical skills to synaesthesia (Bor, Billington, & Baron-­‐Cohen, 2007; Luria,1968), but it remains uncon>irmed whether the average synaesthete employs theircolours in this manner. It appears that synaesthetic photisms in>luence mathematicalprocessing (Ghirardelli, Mills, Zilioli, Bailey, & Kretschmar, 2010), but the nature of thisin>luence is far from clear. Determining if colours are actually being used to representrules	
  in	
  mathematics,	
  spelling,	
  or	
  in	
  other	
  domains	
  is	
  a	
  crucial	
  next	
  step.Finally, the rule-­‐based categorization task used here is generally considered to involveexplicitly conscious processes that operate in a fundamentally different manner fromthe processes used in statistical or implicit learning (Reber, 1993). Further experimentscould more directly test whether the sub-­‐personal mechanisms that underlie implicit 117 learning can also exploit synaesthetic colour information. If this is the case, the potentialutility of synaesthesia for learning is even wider, given the ubiquity of implicit learningthroughout	
  life.Previous work has demonstrated that the colours synaesthetes associate with lettersare in>luenced by a number of learned properties of these letters (Beeli, Esslen, &Jäncke, 2007; Cohen Kadosh, Henik, & Walsh, 2007; Day, 2005; Rich, Bradshaw, & Mat-­‐tingley, 2005; Simner et al., 2005; Simner & Ward, 2008; Watson, Akins, & Enns, 2012).Here, we demonstrate the reverse: synaesthetic colours can in>luence learning aboutletters. Further exploration of the interactions between synaesthesia and learning islikely to be the source of new understanding about the nature of this fascinatingphenomenon. 118 7	
  	
  Conclusion 7.1	
  	
  What	
  do	
  we	
  know? 7.1.1	
  	
  Demonstrating	
  3	
  main	
  pointsThe work described in this thesis was inspired by the developmental learning hypothe-­‐sis of synaesthesia, according to which synaesthesia develops, at least in part, as astrategic method of solving various learning challenges in childhood. The thesis set outto	
  defend	
  three	
  points	
  which	
  underpin	
  this	
  hypothesis.	
  These	
  are:1. Synaesthesia typically develops as part of a dif>icult learningprocess in which the synaesthete learns a category structurewhose members become the triggers of synaestheticexperiences.2. Synaesthetic experiences are shaped by this learning process,such that they encode a wide range of information about thelearned	
  domain.	
  3. Synaesthesia is exploited on a variety of memory, learning, andcreative tasks, leading to “synaesthetic styles” of performance onthese	
  tasks.In the Introduction, I presented what we already know about these three points, and inthe	
  research	
  chapters	
  I	
  expanded	
  on	
  this	
  by	
  presenting	
  new	
  evidence	
  supporting	
  them.For point 1, the literature already shows that the majority of synaesthetic inducers arelearned categories, and that the development of grapheme-­‐colour synaesthesia takesplace over a period of several years that overlap with the development of literacy. Chap-­‐ter 2 expanded on this, by showing that a particular learning challenge—the acquisitionof a second language in grade school—is associated with a higher likelihood of becom-­‐ing	
  synaesthetic.	
   119 Point 2 is supported by a number of previous >indings demonstrating that synaestheticconcurrents are shaped by a wide range of learned information about the inducer do-­‐main. In Chapters 3-­‐5, I expanded on this, showing that various different kinds oflearned information independently in>luence the colours of grapheme-­‐colour synaes-­‐thetes. In Czech, this includes quite high-­‐level categorical knowledge about the role dif-­‐ferent letters and diacriticals play within the language. This knowledge is not availableuntil >irst grade at the earliest, showing that synaesthetic development is mediated byhigh-­‐level	
  learning	
  over	
  the	
  course	
  of	
  several	
  years.Finally, point 3 has been largely accepted by researchers, but generally supported byanecdotal evidence. Chapter 6 provides the >irst clear proof that synaesthetic associa-­‐tions can be useful on a dif>icult learning task. Synaesthetes in this category learningstudy spontaneously began using their colours as guides to the category structure in avery similar manner to non-­‐synaesthetes looking at real colours, and both groups signi>-­‐icantly out-­‐performed non-­‐synaesthetes who did not have access to colour information.If synaesthetes can learn to do this in 20-­‐40 minutes seated in front of a computer, whoknows what they can do over the course of decades as they navigate the various tasksassociated	
  with	
  literacy? 7.1.2	
  	
  Other	
  OindingsIn the course of gathering evidence for the three main points guiding this thesis, my col-­‐laborators and I uncovered a number of other fascinating >indings, which may or maynot directly relate to the developmental learning hypothesis. Here I outline some of themost	
  interesting	
  of	
  these	
  >indings.Chapter 2 demonstrated that women are far more likely to cooperate with experimenterrequests in synaesthesia studies, demonstrating that previously-­‐reported female biasesin synaesthesia are almost certainly due to cooperation with the experimental protocol,rather than any actual differences in prevalence between the sexes. It is probably nowsafe to assume that there is at most a very small in>luence of sex on synaesthetic preva-­‐lence, and very likely no in>luence at all. If there is no in>luence, this suggests that the fe-­‐ 120 male compliance bias extends all the way to answering direct questions about synaes-­‐thesia: men may be less likely to self-­‐report as synaesthetic even if they are given adescription of synaesthesia and asked if this applies to them. If this is the case, then allprevious reports of sex differences in synaesthesia should be treated very cautiously, inparticular the >inding of a strong bias for transmission down the maternal line. Thiscasts doubt upon one of the main reasons for believing in a simple genetic account ofsynaesthetic development, which in turn makes something like the developmentallearning	
  hypothesis	
  far	
  more	
  plausible.	
  Another important result from Chapter 2 is the >inding of a large proportion of individu-­‐als who pass the consistency tests for various kinds of synaesthesia, but initially deniedhaving the forms of synaesthesia in question. This con>irms that the self-­‐report and con-­‐sistency criteria for synaesthesia are fully independent of each other, and shows that wecannot simply screen synaesthetes based on self-­‐report. In the Introduction I reviewedthe current consensus that we do not have a good theoretical de>inition of synaesthesia,and these data suggest that there are serious issues with our operationalizations aswell.	
  Chapter 2’s results are also consistent with previous reports of synaesthetic “clustering”.We >ind that the effect of second language acquisition is not con>ined to grapheme-­‐colour synaesthesia, as originally hypothesized, but rather to all varieties of synaesthe-­‐sia we test for. This, it is argued, may be the result of generalizing a particular learningstrategy initially developed during literacy acquisition to other, conceptually similar,problems.One fascinating >inding from Chapters 3-­‐5 is that almost all the effects we uncover arecon>ined to either luminance or hue, but not both. In Chapter 3 it was argued that thissplit is useful. Categorical information can be most usefully mapped onto hue space,partly because this is more natural for humans, but also because this would not inter-­‐fere with the luminance-­‐based processes we use to actually identify letter categories.This suggestion is complicated somewhat by the Czech >indings from Chapter 5. While 121 the same hue/luminance splits were found for several mappings, the strongest categori-­‐cal mappings (the base-­‐diacritical pairs) were found in both luminance and hue, as wasthe	
  in>luence	
  of	
  phonology.The Czech data used in Chapter 5 also enabled a strong test of the learning order hy-­‐pothesis, which stated that the ordinality-­‐hue effect, in which letters earlier in the al-­‐phabet are further apart in hue, was due to the earlier letters having been learnt >irst.Since Czech letters are not learned in alphabetical order, this hypothesis could be direct-­‐ly tested, and it was conclusively rejected. Instead, it appears that there is somethingspecial about ordinality itself, which may provide support for Novich et al.’s notion ofcoloured	
  sequence	
  synaesthesias.The data from Chapter 5 also allowed a tentative ranking of some of the in>luences onsynaesthetic colour, suggesting that letter shape and alphabetical order (and possibly anabstract notion of letter identity) are stronger determinants of colour than phonology.This might be taken as evidence that grapheme-­‐colour synaesthesia does not have itsearliest roots in a phoneme-­‐colour synaesthesia, but rather results speci>ically from thelearning	
  problems	
  associated	
  with	
  the	
  development	
  of	
  literacy.Finally, in Chapter 6 we saw that though the synaesthetes in our study exploited theircolours in a similar manner to non-­‐synaesthetes viewing real colours, their responsetimes were far slower. This may be an indication that the induction of synaesthetic expe-­‐riences is a slow process, which might be taken as support for the re-­‐entrant processingor disinhibited feedback theories of the neurophysiology of synaesthesia. As always,more	
  investigation	
  is	
  needed. 7.2	
  	
  What	
  don’t	
  we	
  know?I take it that the three points that were the organizing principles of this thesis have been>irmly established. This does not, it should be noted, prove the developmental learninghypothesis per se. In order to do this, one would need to connect the points in the causalchain described in the Introduction, which is far beyond the scope of this thesis. That is, 122 one would have to show that synaesthesia develops in the course of learning (point 1) because it can be exploited to achieve learning success (point 3), and that the particularaspects of the inducer domain which are encoded within synaesthetic concurrents(point 2) are encoded because they are useful in this manner. I hope that this causalchain seems at least somewhat plausible now. For example, it seems intuitive that en-­‐coding categorical information about vowels and consonants might be particularly use-­‐ful in learning to read. However plausibility is a long way from proof, and a great deal ofinnovative research would need to take place before this causal chain could be estab-­‐lished.	
  What	
  might	
  this	
  research	
  look	
  like?One step would be to verify if children with synaesthetic colours are exploiting them insome way. Consider the case of the categorical distinction between vowels and conso-­‐nants. Would a child whose synaesthetic colours make this distinction clear be impairedon a vowel/consonant discrimination task if the letters were presented in incongruentcolours? This would at least provide preliminary evidence that their method of makingthis	
  discrimination	
  in	
  some	
  way	
  exploited	
  their	
  colours.A great deal of reading performance involves the recognition of dozens of common bi-­‐grams and trigrams. Colour-­‐based “markers” for letters might be a particularly goodmethod of recognizing such common groups of letters, allowing for improved acquisi-­‐tion and retention. Another possible study, then, would be to see, for individual synaes-­‐thetes, which of the common bigrams and trigrams are particularly salient, given theirpersonal letter colours, and then to see if they have a particular advantage at recogniz-­‐ing,	
  discriminating	
  or	
  otherwise	
  interacting	
  with	
  these	
  groups	
  of	
  letters	
  above	
  others.An important longitudinal question concerns what happens to those children who atsome point have synaesthetic tendencies, but who lose these as they get older. Mightthese be individuals for whom letter colours were particularly uninteresting, or forwhom the development of literacy was accomplished in a manner that did not exploittheir	
  synaesthesia? 123 Finally, the research presented here has been strongly slanted towards grapheme-­‐colour synaesthesia, and letter-­‐colour synaesthesia in particular. However we have seenstrong evidence that this emphasis does not re>lect the tendencies among synaesthetes,who generally have multiple forms of synaesthesia. Understanding how synaestheticcolours (or other concurrents) might help with learning these other inducer domains isan	
  important	
  and	
  dif>icult	
  task. 7.3	
  	
  Why	
  study	
  synaesthesia?Supposing the developmental learning hypothesis is true, one can imagine a great num-­‐ber of ways in which the kinds of studies described here might have practical bene>its.For instance, if colour-­‐coding information is useful for synaesthetes, might it be equallyuseful for non-­‐synaesthetes? Would it help with second-­‐language learning among olderchildren or adults? Anything that could reduce the drudgery and increase the successrate of second-­‐language learning would be of obvious utility to many. Other similar pos-­‐sibilities	
  are	
  fairly	
  easy	
  to	
  think	
  of.Aside from its potential practical utility, it is often suggested that synaesthesia is ofgreat theoretical importance. Researchers often describe the importance of synaesthe-­‐sia research in terms of its utility for understanding the mysteries of consciousness,multimodal associations, the heterogeneity of perception across individuals, or otherprofound aspects of the human mind. One question which might have real theoreticalimportance, for example, concerns the great preponderance of spatial and coloured con-­‐currents in synaesthesia. Though there are synaesthetes who taste words, or hearsounds when they see particular stimuli, they are a tiny minority compared to thosewho experience members of categories as coloured or as located in personal space. Whyare space and colour so important? Understanding this might be of real importance tounderstanding the structure of our own minds, which is arguably the purpose ofpsychology.However, while I agree that such theoretical and practical considerations are important,they are not the ones that motivated me to begin studying synaesthesia, and I suspect124 this is true of most researchers. I study synaesthesia for its own sake. It is seriouslyweird, and endlessly fascinating. I argue that this is a perfectly good reason to devotetime and energy to a research topic. Basic research cannot be guided purely by short-­‐term utilitarian concerns, especially if a society’s long-­term interests are utilitarian. Byde>inition, unexpected results that genuinely shift paradigms and provide the greatestlong-­‐term bene>its to society cannot come from “safe” areas of research, because all thatwe	
  mean	
  by	
  “safe”	
  is	
  that	
  an	
  area	
  of	
  research	
  is	
  expected	
  to	
  produce	
  useful	
  results.	
  Frankly, while I agree that synaesthesia research can shed light on consciousness or oth-­‐er important issues, it is not at all clear to me that it is more likely to do this than re-­‐search into virtually any other conscious phenomenon. And while colour-­‐codinglearning materials might improve retention, I suspect that raising teacher’s salarieswould have a far stronger effect. However science needs to keep poking into the weirdand fascinating, because every now and then when we turn over one of these pebbles,we	
  >ind	
  it	
  is	
  actually	
  a	
  diamond. 125 Bibliography Albright, A. (2006). SimilarityCalculator.pl. Retrieved from http://web.mit.edu/albright/ www/software/SimilarityCalculator.zip Aleman, A., Rutten, G. J. M., Sitskoorn, M. M., Dautzenberg, G., & Ramsey, N. F. (2001). Activation of striate Cortex in the absence of visual stimulation: an fMRI study of synesthesia. Neuroreport, 12(13), 2827-2830. Amin, M., Olu-Lafe, O., Claessen, L. E., Sobczak-Edmans, M., Ward, J., Williams, A. L., & Sagiv, N. (2011). Understanding grapheme personification: A social synaesthesia? Journal of Neuropsychology, 5, 255-282. Anonymous. (2011). Kanye West behind the scenes interview. YouTube. Retrieved March 15, 2013, from http://www.youtube.com/watch?v=UhXKhSer2ic Anonymous. (n.d.). List of superhero characters with synaesthesia powers. Comic Vine. Re- trieved March 12, 2013, from http://www.comicvine.com/synaesthesia/41-129/ Asano, M., & Yokosawa, K. (2011). Synesthetic colors are elicited by sound quality in Ja- panese synesthetes. Consciousness and Cognition, 20(4), 1816-1823. Asano, M., & Yokosawa, K. (2012). Synesthetic colors for Japanese late acquired graphemes. Consciousness and Cognition, 21(2), 983-993. doi:10.1016/ j.concog.2012.02.005 Ashby, F. G., & Maddox, W. T. (2005). Human category learning. Annual Review of Psy- chology, 56, 149-178. Asher, J. E., . . . Monaco, A. P. (2009). A Whole-Genome Scan and Fine-Mapping Linkage Study of Auditory-Visual Synesthesia Reveals Evidence of Linkage to Chromosomes 2q24, 5q33, 6p12, and 12p12. American Journal of Human Genetics, 84(2), 279-285. Bailey, M. E. S., & Johnson, K. J. (1997). Synaesthesia: Is a genetic analysis feasible? In S. Baron-Cohen & J. E. Harrison. Oxford: Basil Blackwell. Banissy, M. J., Stewart, L., Muggleton, N. G., Griffiths, T. D., Walsh, V. Y., Ward, J., & Kanai, R. (2012). Grapheme-color and tone-color synesthesia is associated with struc- tural brain changes in visual regions implicated in color, form, and motion. Cognitive Neuroscience, 3(1), 29-35. 126 Banissy, M. J., Walsh, V., & Ward, J. (2009). Enhanced sensory perception in synaesthesia. Experimental Brain Research, 196(4), 565-571. Barnett, K. J., Finucane, C., Asher, J. E., Bargary, G., Corvin, A. P., Newell, F. N., & Mitchell, K. J. (2008a). Familial patterns and the origins of individual differences in synaesthesia. Cognition, 106(2), 871-893. Barnett, K. J., Foxe, J. J., Molholm, S., Kelly, S. P., Shalgi, S., Mitchell, K. J., & Newell, F. N. (2008b). Differences in early sensory-perceptual processing in synesthesia: A visu- al evoked potential study. Neuroimage. doi:10.1016/j.Neuroimage.2008.07.028 Baron-Cohen, S. (1996). Is there a normal phase of synaesthesia in development? Psyche: An Interdisciplinary Journal of Research on Consciousness, 2(2). Baron-Cohen, S., Burt, L., Smith-Laittan, F., Harrison, J., & Bolton, P. (1996). Synaesthesia: Prevalence and familiality. Perception, 25(9), 1073-1079. Baron-Cohen, S., Harrison, J., Goldstein, L. H., & Wyke, M. A. (1993). Coloured speech perception: Is synaesthesia what happens when modularity breaks down? Perception, 22(4), 419-426. Baron-Cohen, S., & Harrison, J. E. (Eds.). (1997). Synaesthesia: Classic and Contemporary readings. Oxford: Basil Blackwell. Baron-Cohen, S., Wyke, M. A., & Binnie, C. (1987). Hearing words and seeing colours: an experimental investigation of a case of synaesthesia. Perception, 16(6), 761-767. Education, B. C. M. o. (2001). Core French 5 to 12. Ministry of Education, BC. Beeli, G., Esslen, M., & Jäncke, L. (2007). Frequency correlates in grapheme-color synaes- thesia. Psychological Science, 18(9), 788-792. Bernard, J. W. (1986). Messiaen’s Synaesthesia, The Correspondence Between Color and Sound Structure in his Music. Music Perception, 4(1), 41-68. Bialystock, E., & Herman, J. (1999). Does bilingualism matter for early literacy? Bilingual- ism: Language and Cognition, 2(1), 35-44. Bialystock, E., Luk, G., & Kwan, E. (2005). Bilingualism, biliteracy, and learning to read: Interactions among languages and writing systems. Scientific Studies of Reading, 9(1), 43-61. Bialystok, E., & Barac, R. (2012). Emerging bilingualism: dissociating advantages for met- alinguistic awareness and executive control. Cognition, 122(1), 67-73. doi:10.1016/ j.cognition.2011.08.003 Binet, A., & Philippe, J. (1892). Êtude sur un nouveau cas d’audition colorée. Revue Philosophique de la France et de l’étranger, 33, 461-464. Bloch, C., . . . Nitsch, C. (2009). The age of second language acquisition determines the vari- ability in activation elicited by narration in three languages in Broca’s and Wer- nicke’s area. Neuropsychologia, 47(3), 625-633. doi:10.1016/j.neuropsycholo- gia.2008.11.009 127 Boles, D. B., & Clifford, J. E. (1989). An upper- and lowercase alphabetic similarity matrix, with derived generation similarity values. Behavior Research Methods, Instruments, & Computers, 21(6), 579-586. Bor, D., Billington, J., & Baron-Cohen, S. (2007). Savant memory for digits in a case of synaesthesia and Asperger syndrome is related to hyperactivity in the lateral pre- frontal Cortex. Neurocase, 13(5-6), 311-319. Brang, D., Hubbard, E. M., Coulson, S., Huang, M., & Ramachandran, V. S. (2010). Magne- toencephalography reveals early activation of V4 in grapheme-color synesthesia. Neuroimage, 53(1), 268-274. Brang, D., & Ramachandran, V. S. (2011). Survival of the synesthesia gene: Why do people hear colors and taste words? PLoS Biology, 9(11), e1001205. doi:10.1371/jour- nal.pbio.1001205.g002 Brang, D., Rouw, R., Ramachandran, V. S., & Coulson, S. (2010). Similar letters, similar hues: Shape-color isomorphism in grapheme-color synaesthesia. Proceedings from 2010 Meeting of the UK Synaesthesia Association, Brighton, U.K. Brang, D., Rouw, R., Ramachandran, V. S., & Coulson, S. (2011). Similarly shaped letters evoke similar colors in grapheme-color synesthesia. Neuropsychologia, 49(5), 1355-1358. doi:10.1016/j.Neuropsychologia.2011.01.002 Breslow, L. A., Trafton, J. G., McCurry, J. M., & Ratwani, R. M. (2010). An algorithm for generating color scales for both categorical and ordinal coding. Color Research and Application, 35(1), 18-28. doi:10.1002/col.v35:1 Březinová, L., Havel, J., & Stadlerová, H. (2007). Živá Abeceda, Učebnice pro 1. ročník zák- ladní školy [The Living ABC’s: Program in reading for the first year of school]. Pizeň: Fraus. Calkins, M. W. (1893). A statistical study of pseudo-chromesthesia and of mental-forms. American Journal of Psychology, 5, 439-464. Calkins, M. W. (1895). Synaesthesia. American Journal of Psychology, 7, 90-107. Claparède, E. (1900). Sur l’Audition Colorée. Revue Philosophique de la France et de l’étranger, 49, 515-521. Clements, G. N., & Hume, E. V. (1995). The internal organization of speech sounds. In J. A. Goldsmith (Ed.), The Handbook of Phonological Theory (pp. 245-306). Cambridge, MA: Blackwell. Cobo-Lewis, A. B., Pearson, P. Z., Eilers, R. E., & Umbel, V. C. (2002). Effects of bilingual- ism and bilingual education on oral and written Spanish skills: A multifactor study of standardized test outcomes. In D. K. Oller & R. E. Eilers (Eds.), Language and litera- cy in bilingual children (pp. 98-117). Clevedon, UK: Multiligual Matters. Cohen Kadosh, R., Henik, A., & Walsh, V. (2007). Small is bright and big is dark in synaes- thesia. Current Biology, 17(19), R834-R835. 128 Cohen Kadosh, R., Kadosh, K. C., & Henik, A. (2007). The neuronal correlate of bidirection- al synesthesia: A combined event-related potential and functional magnetic resonance Imaging study. Journal of Cognitive Neuropsychology, 19(12), 2050-2059. Cohen Kadosh, R., & Terhune, D. B. (2012). Redefining synaesthesia? British Journal of Psychology, 103, 20-23. Colizoli, O., Murre, J. M. J., Rouw, R., Karniel, A., & Witthoft, N. (2012). Pseudo-Synesthe- sia through Reading Books with Colored Letters. PLoS One, 7(6), 1-10. Curtis, C. (1998). Letter-color synaesthesia. OtherThings.com. Retrieved April 12, 2013, from otherthings.com/uw/syn Cytowic, R. E. (1988). Tasting colors, smelling sounds - Neurological clues to a confounding condition. The Sciences, 32-37. Cytowic, R. E. (1989a). Synesthesia: A Union of the Senses (First ed.). New York: Springer- Verlag. Cytowic, R. E. (1989b). Synesthesia and mapping of subjective sensory dimensions. Neurol- ogy, 39(6), 849-850. Cytowic, R. E. (1993). The Man Who Tasted Shapes. New York: G.P. Putnam’s Sons. Cytowic, R. E. (1997). Synaesthesia: phenomenology and neuropsychology - a review of cur- rent knowledge. In S. Baron-Cohen & J. E. Harrison(pp. 17-39). Oxford: Basil Blackwell. Cytowic, R. E. (2002). Synesthesia: A Union of the Senses (Second ed.). Cambridge Massa- chusetts: The MIT Press. Cytowic, R. E., & Eagleman, D. M. (2009). Wednesday is Indigo Blue: Discovering the Brain of Synesthesia. Boston, MA: MIT Press. Cytowic, R. E., & Wood, F. B. (1982a). Synesthesia I. A review of major theories and their brain basis. Brain and Cognition, 1, 23-35. Cytowic, R. E., & Wood, F. B. (1982b). Synesthesia II. Psychophysical relations in the synesthesia of geometrically shaped taste and colored hearing. Brain and Cognition, 1, 36-49. Dailey, A., Martindale, C., & Borkum, J. (1997). Creativity, synesthesia, and physiognomic perception. Creativity Research Journal, 10(1), 1-8. Day, S. A. (2005). Some demographic and socio-cultural aspects of synesthesia. In L. C. Robertson & N. Sagiv(pp. 11-33). Oxford: Oxford University Press. Dindia, K., & Allen, M. (1992). Sex differences in self-disclosure: A meta-analysis. Psycho- logical Bulletin, 112(1), 106-124. Dixon, M. J., Smilek, D., Cudahy, C., & Merikle, P. M. (2000). Five plus two equals yellow: Mental arithmetic in people with synaesthesia is not coloured by visual experience. Nature, 406(679), 365-365. 129 Dixon, M. J., Smilek, D., & Merikle, P. M. (2004). Not all synaesthetes are created equal: Projector versus associator synaesthetes. Cognitive, Affective, & Behavioral Neuro- science, 4, 335-343. Domino, G. (1989). Synesthesia and creativity in fine arts students: An empirical look. Cre- ativity Research Journal, 2, 17-29. Eagleman, D. M. (2010). What has large scale analysis taught us? Proceedings from 2010 Meeting of the UK Synaesthesia Association, Brighton, U.K. Eagleman, D. M. (2012). Synaesthesia in its protean guises. British Journal of Psychology, 103, 16-19. Eagleman, D. M., Kagan, A. D., Nelson, S. S., Sagaram, D., & Sarma, A. K. (2007). A standardized test battery for the study of synesthesia. Journal of Neuroscience Meth- ods, 159(1), 139-145. Ellis, N. C., . . . Petalas, M. (2004). The effects of orthographic depth on learning to read al- phabetic, syllabic, and logographic scripts. Reading Research Quarterly, 39(4), 438-468. Flournoy, T. (1892). L’audition colorée. Archives des Sciences Physiques et Naturelles, 28, 505-508. Frisch, S. A. (1996). Similarity and Frequency in Phonology. PhD. Northwestern University, Evanston, Il. Frisch, S. A., Pierrehumbert, J. B., & Broe, M. B. (2004). Similarity avoidance and the OCP. Natural Language and Linguistic Theory, 22, 179-228. Galton, F. (1883). Inquiries into Human Faculty and Its Developmment. Gheorghiu, E., & Kingdom, F. A. A. (2006). Luminace-contrast properties of contour-shape processing revealed through the shape-frequency after-effect. Vision Research, 46(21), 3603-3615. Gheorghiu, E., & Kingdom, F. A. A. (2007). Chromatic tuning of contour-shape mechanisms revealed through the shape-frequency and shape-amplitude after-effects. Vision Re- search, 47(14), 1935-1949. doi:10.1016/j.visres.2007.03.010 Ghirardelli, T. G., Mills, C. B., Zilioli, M. K. C., Bailey, L. P., & Kretschmar, P. K. (2010). Synesthesia affects verification of simple arithmetic equations. Journal of General Psychology, 137(2), 175-189. Gibson, E. J. (1969). Principles of Perceptual Learning and Development. New Jersey: Prentice-Hall, Inc. Gilmore, G. C., Hersh, H., Caramazza, A., & Griffin, J. (1979). Multidimensional letter simi- larity derived from recognition errors. Perception & Psychophysics, 25(5), 425-431. Green, J. A. K., & Goswami, U. (2008). Synesthesia and number cognition in children. Cog- nition, 106(1), 463-473. 130 Grossenbacher, P. G. (1997). Perception and sensory information in synesthetic experience. In S. Baron-Cohen & J. E. Harrison(pp. 148-172). Oxford: Basil Blackwell. Grossenbacher, P. G., & Lovelace, C. T. (2001). Mechanisms of synesthesia: cognitive and physiological constraints. Trends in Cognitive Sciences, 5(1), 36-41. Gupta, S. M., Geyer, L. H., & Maalouf, J. A. (1983). Effect of font and medium on recogni- tion/confusion. SIGCHI conference on Human Factors in Computing Systems, 144-149. Hancock, P. (2006). Monozygotic twins’ colour-number association: A case study. Cortex, 42(2), 147-150. Hänggi, J., Beeli, G., Oechslin, M. S., & Jancke, L. (2008). The multiple synaesthete E.S.: neuroanatomical basis of interval-taste and tone-colour synaesthesia. Neuroimage, 43(2), 192-203. doi:10.1016/j.Neuroimage.2008.07.018 Hansen, T., & Gegenfurtner, K. R. (2009). Independence of color and luminance edges in na- tural scenes. Visual Neuroscience, 26(1), 35-49. Head, P. D. T. (2006). Synaesthesia: Pitch-colour isomorphism in RGB-space? Cortex, 42(2), 164-174. Hubbard, E. M., Arman, A. C., Ramachandran, V. S., & Boynton, G. M. (2005). Individual differences among grapheme-color synesthetes: Brain-behavior correlations. Neuron, 45(6), 975-985. Hubbard, E. M., Brang, D., & Ramachandran, V. S. (2011). The cross-activation theory at 10. Journal of Neuropsychology, 5(2), 152-177. doi:10.1111/ j.1748-6653.2011.02014.x Ione, A. (2004). Klee and Kandinsky: Polyphonic painting, chromatic chords, and synaesthe- sia. Journal of Consciousness Studies, 11(3-4), 148-158. Jäncke, L., Rogenmoser, L., Meyer, M., & Elmer, S. (2012). Pre-attentive modulation of brain responses to tones in coloured-hearing synesthetes. BMC Neuroscience, 13(1), 1-15. doi:10.1186/1471-2202-13-151 Jewanski, J., Day, S. A., & Ward, J. (2009). A Colorful Albino: The First Documented Case of Synaesthesia, by Georg Tobias Ludwig Sachs in 1812. Journal of the History of the Neurosciences, 18(3), 293-303. Jürgens, U. M., & Nikolic, D. (2012). Ideasthesia: Conceptual processes assign similar colours to similar shapes. Translational Neuroscience, 3(1), 22-27. Justice, L. M., Pence, K., Bowles, R. B., & Wiggins, A. (2006). An investigation of four hy- potheses concerning the order by which 4-year-old children learn the alphabet letters. Early Childhood Research Quarterly, 21(3), 374-389. doi:10.1016/ j.ecresq.2006.07.010 Kingdom, F. A. A., Beauce, C., & Hunter, L. (2004). Colour vision brings clarity to shadows. Perception, 33(8), 907-914. 131 Kingdom, F. A. A., & Kasrai, R. (2006). Colour unmasks dark targets in complex displays. Vision Research, 46(6-7), 814-822. doi:10.1016/j.visres.2005.08.018 Kovács, Á. M., & Mehler, J. (2009). Cognitive gains in 7-month-old bilingual infants. PNAS, 106(16), 6556-6560. Králik, J. (1983). Statistika českých grafémů s vyuźitím moderní výpočetní techniky [Statis- tics of the Czech graphemes with the aid of modern computational techniques]. Slove a Slovesnost [Word and Literature], 44, 295-304. Kusnir, F., & Thut, G. (2012). Formation of automatic letter–colour associations in non- synaesthetes through likelihood manipulation of letter–colour pairings. Neuropsy- chologia, 50(14), 3641-3652. doi:10.1016/j.neuropsychologia.2012.09.032 Ladová, A., Holas, M., & Staudková, H. (2011). Živá Abeceda, Učebnice pro 1. ročník zák- ladních škol pro vzdĕlávací obor Český jazyk a literature [The Living ABC’s: Pro- gram in reading for the first year of school with Czech rhymes and literature]. Praha: Alter. Laeng, B. (2009). Searching through synaesthetic colors. Attention, Perception, & Psy- chophysics, 71(7), 1461-1467. Laeng, B., Hugdahl, K., & Specht, K. (2011). The neural correlate of colour distances re- vealed with competing synaesthetic and real colours. Cortex, 47(3), 320-331. doi:10.1016/j.Cortex.2009.09.004 Lewand, R. (2000). Cryptological Mathematics. Washington: The Mathematical Association of America. Liebe, S., Fischer, E., Logothetis, N., & Rainer, G. (2009). Color and shape interactions in the recognition of natural scenes by human and monkey observers. Journal of Vision, 9(5), 1-16. Luria, A. R. (1968). The Mind of a Mnemonist. New York: Basic Books. MacLeod, C. M., & Dunbar, K. (1988). Training and Stroop-like interference: Evidence for a continuum of automaticity. Journal of Experimental Psychology: Learning, Memory and Cognition, 14(1), 126-135. Maddox, W. T., Filoteo, J. V., Hejl, K. D., & Ing, A. D. (2004). Category number impacts rule-based but not information-integration category learning: Further evidence for dissociable category-learning systems. Journal of Experimental Psychology: Learning, Memory and Cognition, 30(1), 1-19. doi:10.1037/0278-7393.30.1.227 Marks, L. E. (1975). On colored-hearing synesthesia: Cross-modal translations of sensory dimensions. Psychological Bulletin, 82(3), 303-331. Maurer, D. (1993). Neonatal synesthesia: Implications for the processing of speech and faces. In B. de Boysson-Bardies, S. de Schonen, P. Jusczyk, P. MacNeilage, & J. Mor- ton(pp. 109-124). New York: Springer. Retrieved from http://direct.bl.uk/research/ 26/64/EN004127304.html?source=googlescholar&oi=docdel 132 MedCalc Software. (2013). Relative Risk. Medcalc. Retrieved April 14, 2013, from http:/ /www.medcalc.org/calc/relative_risk.php Melichárková, I., Štĕpán, L., & Švecová, L. (2008). Slabikář [The Syllable Book]. Liberec: Dialog. Mikulenková, H., & Mladý, R. (2004). Příručka k vyučování čtení, psaní a literature v prvním ročníku základní školy [Teaching Reading to 6 year olds, with literature for the first year of school]. Olomouc: Prodos. Mikulenková, H., Mladý, R., & Forman, M. (2004). Slabikář [The Syllable Book]. Olomouc: Prodos. Mills, C. B., Boteler, E. H., & Oliver, G. K. (1999). Digit synaesthesia: A case study using a Stroop-type test. Cognitive Neuropsychology, 16(2), 181-191. Mills, C. B., Viguers, M. L., Edelson, S. K., Thomas, A. T., Simon-Dack, S. L., & Innis, J. A. (2002). The color of two alphabets for a multilingual synesthete. Perception, 31(11), 1371-1394. doi:10.1068/p3429 Nabokov, V. (1989). Speak, Memory: An Autobiography Revisited. New York: Random House, Inc. Nagai, T., & Uchikawa, K. (2009). Different hue coding underlying figure segregation and region detection tasks. Journal of Vision, 9(9), 1-19. Niccolai, V. (2012). Modality and Variability of Synesthetic Experience. The American Jour- nal of Psychology, 125(1), 81-94. doi:10.5406/amerjpsyc.125.1.0081 Nikolic, D., Jurgens, U. M., Rothen, N., Meier, B., & Mroczko, A. (2011). Swimming-style synesthesia. Cortex, 47(7), 874-879. doi:10.1016/j.Cortex.2011.02.008 Nováčková, O. (2010). Živá Abeceda, Učebnice vytvořená v souladu s RVP ZV [The Living ABC’s: Program for reading and writing for the first year at school]. Brno: Nova Skola. Novich, S., Cheng, S., & Eagleman, D. M. (2011). Is synaesthesia one condition or many? A large-scale analysis reveals subgroups. Journal of Neuropsychology, 5(2), 353-371. doi:10.1111/j.1748-6653.2011.02015.x Nunn, J. A., . . . Gray, J. A. (2002). Functional magnetic resonance imaging of synesthesia: activation of V4/V8 by spoken words. Nature Neuroscience, 5(4), 371-375. Ortiz-Mantilla, S., Choudhury, N., Alvarez, B., & Benasich, A. A. (2010). Involuntary switching of attention mediates differences in event-related responses to complex tones between early and late Spanish-English bilinguals. Brain Res, 1362, 78-92. doi:10.1016/j.brainres.2010.09.031 Pariyadath, V., Plitt, M. H., Churchill, S. J., & Eagleman, D. M. (2012). Why overlearned se- quences are special: distinct neural networks for ordinal sequences. Frontiers in Human Neuroscience, 6, 1-9. doi:10.3389/fnhum.2012.00328/abstract 133 Paulesu, E., . . . Frith, C. D. (1995). The physiology of coloured hearing: A PET activation study of colour-word synaesthesia. Brain, 118, 661-676. Pautzke, R. (2010). ‘Making sense’ messing around with black boxes: Synaesthesia and learning - How to master the Theremin without notes. Proceedings from 2010 Meet- ing of the UK Synaesthesia Association, Brighton, UK. Peacock, K. (1985). Synaesthetic perception: Alexander Scriabin’s color hearing. Music Per- ception, 2(4), 483-506. Podgorny, P., & Garner, W. R. (1979). Reaction time as a measure of inter- and intraobject visual similarity: Letters of the alphabet. Perception & Psychophysics, 26(1), 37-52. Potůčková, J. (2010). Živá Abeceda pro 1. ročník základní školy [The Living ABC’s: Pro- gram #1 for the first year of school]. Brno: Studio 11. Radvansky, G. A., Gibson, B. S., & McNerney, M. W. (2011). Synesthesia and memory: Color congruency, von Restorff, and false memory effects. Journal of Experimental Psychology: Learning, Memory and Cognition, 37(1), 219-229. Ramachandran, V. S., & Hubbard, E. M. (2001a). Synaesthesia--A Window into Perception, Thought and Language. Journal of Consciousness Studies, 8(12), 3-34. Ramachandran, V. S., & Hubbard, E. M. (2001b). Psychophysical investigations into the neural basis of synaesthesia. Proceedings of the Royal Society of London Series B - Biological Sciences, 268(1470), 979-983. Ramachandran, V. S., & Rogers-Ramachandran, D. (1996). Synaesthesia in phantom limbs induced with mirrors. Proceedings of the Royal Society of London Series B - Biologi- cal Sciences, 263(1369), 377-386. Reber, A. S. (1993). Implicit Learning and Tacit Knowledge: An Essay on the Cognitive Un- conscious. Oxford: Oxford University Press. Rich, A. N., Bradshaw, J. L., & Mattingley, J. B. (2005). A systematic, large-scale study of synaesthesia: Implications for the role of early experience in lexical-colour associa- tions. Cognition, 98, 53-84. Rich, A. N., Williams, M. A., Puce, A., Syngeniotis, A., Howard, M., McGlone, F., & Mat- tingley, J. B. (2006). Neural correlates of imagined and synaesthetic colours. Neu- ropsychologia, 44(1), 2918-2925. Rose, K. B. (1909). Some statistics on synaesthesia. American Journal of Psychology, 20(3), 447. Rothen, N., & Meier, B. (2009). Do synesthetes have a general advantage in visual search and episodic memory? A case for group studies. PLoS One, 4(4), e5037. doi:10.1371/ journal.pone.0005037 Rothen, N., & Meier, B. (2010a). Grapheme-colour synaesthesia yields an ordinary rather than extraordinary memory advantage: Evidence from a group study. Memory, 18(3), 258-264. 134 Rothen, N., & Meier, B. (2010b). Higher prevalence of synaesthesia in art students. Percep- tion, 39(5), 718-720. Rothen, N., Meier, B., & Ward, J. (2012). Enhanced memory ability: Insights from synaes- thesia. Neuroscience and Biobehavioral Reviews, 36(8), 1952-1963. doi:10.1016/ j.neubiorev.2012.05.004 Rouw, R., & Scholte, H. S. (2007). Increased structural connectivity in grapheme-color synesthesia. Nature Neuroscience, 10(6), 792-797. Retrieved from http:/ /www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Cita- tion&list_uids=17515901 Rouw, R., & Scholte, H. S. (2010). Neural Basis of Individual Differences in Synesthetic Ex- periences. Journal of Neuroscience, 30(18), 6205-6213. Rouw, R., Scholte, H. S., & Colizoli, O. (2011). Brain areas involved in synaesthesia: A re- view. Journal of Neuropsychology, 5(2), 214-242. doi:10.1111/ j.1748-6653.2011.02006.x Saenz, M., & Koch, C. (2008). The sound of change: visually-induced auditory synesthesia. Current Biology, 18(15), R650-R651. doi:10.1016/j.cub.2008.06.014 Sagiv, N., Simner, J., Collins, J., Butterworth, B., & Ward, J. (2006). What is the relationship between synaesthesia and visuo-spatial number forms? Cognition, 101, 114-128. Seaberg, M. (2012). Synesthetes: “People of the future”. Psychology Today. Retrieved March 16, 2013, from http://www.psychologytoday.com/blog/tasting-the-universe/201203/ synesthetes-people-the-future Seymour, P. H. K., Aro, M., & Erskine, J. M. (2003). Foundation literacy acquisition in Eu- ropean orthographies. British Journal of Psychology, 94, 143-174. Shimono, K., Shiori, S., & Yaguchi, H. (2009). Psychophysical evidence for a purely binocu- lar color system. Vision Research, 49(2), 202-210. Simner, J. (2012a). Defining synaesthesia. British Journal of Psychology, 103, 1-15. Simner, J. (2012b). Defining synaesthesia: A response to two excellent commentaries. British Journal of Psychology, 103, 24-27. Simner, J., Glover, L., & Mowat, A. (2006). Linguistic determinants of word colouring in grapheme-colour synaesthesia. Cortex, 42(2), 281-289. Simner, J., Harrold, J., Creed, H., Monro, L., & Foulkes, L. (2009). Early detection of mark- ers for synaesthesia in childhood populations. Brain, 132(Pt 1), 57-64. doi:10.1093/ brain/awn292 Simner, J., Hung, W. Y., & Shillcock, R. (2011). Synaesthesia in a logographic language: The colouring of Chinese characters and Pinyin/Bopomo spellings. Consciousness and Cognition, 20(4), 1376-1392. Simner, J., Mayo, N., & Spiller, M. J. (2009). A foundation for savantism? Visuo-spatial synaesthetes present with cognitive benefits. Cortex, 45(10), 1246-1260. 135 Simner, J., . . . Ward, J. (2006). Synaesthesia: The prevalence of atypical cross-modal experi- ences. Perception, 35, 1024-1033. Simner, J., & Ward, J. (2008). Synaesthesia, color terms, and color space - Color claims came from color names in Beeli, Esslen, and Jancke (2007). Psychological Science, 19(4), 412-414. Simner, J., Ward, J., Lanz, M., Jansari, A., Noonan, K., Glover, L., & Oakley, D. A. (2005). Non-random associations of graphemes to colours in synaesthetic and non-synaes- thetic populations. Cognitive Neuropsychology, 22, 1069-1085. Skelton, R., Ludwig, C., & Mohr, C. (2009). A novel, illustrated questionnaire to distinguish projector and associator synaesthetes. Cortex, 45(6), 721-729. Smilek, D., Carriere, J. S. A., Dixon, M. J., & Merikle, P. M. (2007). Grapheme frequency and color luminance in grapheme-color synaesthesia. Psychological Science, 18(9), 793-795. Smilek, D., Dixon, M. J., Cudahy, C., & Merikle, P. M. (2001). Synaesthetic photisms influ- ence visual perception. Journal of Cognitive Neuroscience, 13, 930-936. Smilek, D., Dixon, M. J., Cudahy, C., & Merikle, P. M. (2002a). Concept driven color expe- riences in digit-color synesthesia. Brain and Cognition, 48, 570-573. Smilek, D., Dixon, M. J., Cudahy, C., & Merikle, P. M. (2002b). Synesthetic color experi- ences influence memory. Psychological Science, 13, 548-552. Spector, F., & Maurer, D. (2008). The colour of Os: Naturally biased associations between shape and colour. Perception, 37(6), 841-847. doi:10.1068/p5830 Spector, F., & Maurer, D. (2011). The colors of the alphabet: Naturally-biased associations between shape and color. Journal of Experimental Psychology: Human Perception and Performance, 37(2), 484-495. doi:10.1037/a0021437 Tomson, S. N., . . . Eagleman, D. M. (2011). The genetics of colored sequence synesthesia: Suggestive evidence of linkage to 16q and genetic heterogeneity for the condition. Behavioural Brain Research, 223(1), 48-52. doi:10.1016/j.bbr.2011.03.071 Ulich, E. (1957). Synästhesie und geschlecht - Empirische Untersuchungen über die geschlechtsspezifische Häufigkeit des Auftretens synästhetischer Phänomene [Synes- thesia and gender - empirical studies on the gender frequency of occurrence of synaesthetic phenomena]. Zeitschrift Fur Experimentelle und Angewandte Psychol- gie, 4, 31-52. Ward, J., Li, R., Salih, S., & Sagiv, N. (2007). Varieties of grapheme-colour synaesthesia: A new theory of phenomenological and behavioural differences. Consciousness and Cognition, 16(4), 913-931. Ward, J., & Simner, J. (2003). Lexical-gustatory synaesthesia: Linguistic and conceptual fac- tors. Cognition, 89(3), 237-261. 136 Ward, J., & Simner, J. (2005). Is synaesthesia an X-linked dominant trait with lethality in males? Perception, 34(5), 611-623. Ward, J., Simner, J., & Auyeung, V. (2005). A comparison of lexical-gustatory and grapheme-colour synaesthesia. Cognitive Neuropsychology, 22, 28-41. Ward, J., Thompson-Lake, D., Ely, R., & Kaminski, F. (2008). Synaesthesia, creativity and art: What is the link? British Journal of Psychology, 99, 127-141. Ward, J., Tsakanikos, E., & Bray, A. (2006). Synaesthesia for Reading and Playing Musical Notes. Neurocase, 12, 27-34. Watson, M. R., Akins, K., & Crawford, L. (2010). The developmental learning hypothesis of synaesthesia - A Summary. Studie z aplikované lingvistiky/Studies in Applied Linguis- tics, 1. Watson, M. R., Akins, K. A., & Enns, J. T. (2012). Second-order mappings in grapheme-col- or synesthesia. Psychonomic Bulletin and Review, 19(2), 211-217. doi:10.3758/ s13423-011-0208-4 Watson, M. R., Blair, M. R., Kozik, P., Akins, K. A., & Enns, J. T. (2012). Grapheme-color synaesthesia benefits rule-based category learning. Consciousness and Cognition, 21, 1533-1540. doi:10.1016/j.concog.2012.06.004 Weiss, P. H., & Fink, G. R. (2009). Grapheme-colour synaesthetes show increased grey mat- ter volumes of parietal and fusiform cortex. Brain, 132(Pt 1), 65-70. doi:10.1093/ brain/awn304 Weiss, P. H., Zilles, K., & Fink, G. R. (2005). When visual perception causes feeling: Enhanced cross-modal processing in grapheme-color synesthesia. Neuroimage, 28(4), 859-868. doi:10.1016/j.neuroimage.2005.06.052 Werker, J. F., & Byers-Heinlein, K. (2008). Bilingualism in infancy: first steps in perception and comprehension. Trends in Cognitive Sciences, 12(4), 144-151. Wheeler, R. H., & Cutsforth, T. D. (1921). The role of synaesthesia in learning. Journal of Experimental Psychology, 4, 448-468. Wheeler, R. H., & Cutsforth, T. D. (1922). Synaesthesia and meaning. American Journal of Psychology, 33, 361-384. Wheeler, R. H., & Cutsforth, T. D. (1925). Synæsthesia in the development of the concept. Journal of Experimental Psychology, 8, 149-159. Witthoft, N., & Winawer, J. (2006). Synesthetic colors determined by having colored refrig- erator magnets in childhood. Cortex, 42(2), 175-183. Witthoft, N., & Winawer, J. (2013). Learning, memory, and synesthesia. Psychological Sci- ence. doi:10.1177/0956797612452573 Yaro, C., & Ward, J. (2007). Searching for Shereshevskii: What is superior about the memo- ry of synaesthetes? The Quarterly Journal of Experimental Psychology, 60(5), 681-695. 137 Žáček, J., & Zmatlíková, H. (2010). Slabikář [The Syllable Book]. Praha: Prodos. 138 Appendix	
  1	
  -­	
  Synaesthesia	
  Survey Page 1 of 2  Email:
__________________________________________________________________
 
 Gender:
 
 
 
 
 Handedness:
 ____
Male
 
 
 
 
 ____
Left‐handed
 ____
Female
 
 
 
 ____
Right‐handed
 
 
 
 
 
 
 ____
Both
 
 First
language
(the
one
you
learned
to
understand
first)(if
you
learned
two
languages
 from
birth,
please
name
both
languages):
_____________________________________
 
 Have
you
learned
any
other
languages?

What
age
were
you
when
you
learned
them?
 
 Language:
______________________________


Age:
________
 Language:
______________________________


Age:
________
 Language:
______________________________


Age:
________
 
 
 When
you
see,
hear
or
think
about
certain
letters
or
numbers,
do
you
see
or
feel
any
 colours?

(Example:
There
is
something
yellow
about
the
letter
G.)
 ____
No,
I
do
not
have
experiences
like
this.
 ____
Yes,
I
have
experiences
like
this
with
letters.
 ____
Yes,
I
have
experiences
like
this
with
numbers.
 
 When
you
see,
hear
or
think
about
the
names
of
the
days
of
the
week
or
months,
do
you
 see
or
feel
any
colours?

(Example:
The
name
“Monday”
is
purple.)
 ____
No,
I
do
not
have
experiences
like
this.
 ____
Yes,
I
have
experiences
like
this.
 
 When
you
hear
certain
sounds,
do
you
see
or
feel
any
colours?

(Examples:
Car
horns
are
 blue.

The
musical
note
C‐sharp
feels
dark
green.

The
sound
of
a
guitar
seems
pink.)
 ____
No,
I
do
not
have
experiences
like
this.
 ____
Yes,
I
have
experiences
like
this.
 
 Do
any
words
seem
to
have
tastes
or
smells?

(Example:
The
word
"elephant"
tastes
like
 strawberries.)
 ____
No,
I
do
not
have
experiences
like
this.
 ____
Yes,
I
have
experiences
like
this.
 
 Do
any
numbers,
hours,
days
of
the
week,
or
months
seem
to
have
a
location
in
space
 around
you?

(Example:
September
is
always
three
feet
to
my
left.)
 ____
No,
I
do
not
have
experiences
like
this.
 ____
Yes,
I
have
experiences
like
this.
 
 Turn
Over
→
 139 Page 2 of 2  
 Do
you
think
of
any
numbers
or
letters
as
having
a
personality,
gender,
or
age?

 (Example:
4
is
a
friendly,
older
lady.)
 ____
No,
I
do
not
have
experiences
like
this.
 ____
Yes,
I
have
experiences
like
this.
 
 These
questions
have
described
some
different
kinds
of
synaesthesia,
but
there
are
 other
kinds,
too.

Do
you
think
you
might
have
another
kind
that
we
did
not
ask
about?
 
____
No.
 ____
Yes.
Describe:________________________________________________________
 
 Do
you
remember
telling
yourself
stories
with
letters
or
numbers
as
characters
when
 you
were
a
child?
 ____
No,
I
do
not
remember
telling
myself
stories
like
this.
 ____
Yes,
I
remember
telling
myself
stories
like
this.
 
 Did
you
learn
to
read
before
you
went
to
kindergarten
(or
whatever
kind
of
school
you
 went
to
at
age
5)?
 _____I
do
not
know.
 _____Yes,
I
learned
to
read
before
I
went
to
kindergarten.
 _____No,
I
learned
to
read
after
I
went
to
kindergarten.
 
 As
a
child,
were
you
ever
given
extra
help
with
reading,
writing
or
spelling?

For
 example,
did
you
ever
have
one‐on‐one
lessons
with
a
special
teacher?
 ____I
do
not
remember.
 ____
No,
I
was
not.
 ____
Yes,
I
was.
 
 As
a
child,
were
you
ever
told
you
had
dyslexia
or
some
other
problem
with
reading?
 ____I
do
not
remember.
 ____
No,
I
was
not.
 ____
Yes,
I
was.
 
 Do
you
feel
that
reading
or
spelling
is
difficult
for
you
now?
 ____
I
am
not
sure.
 ____
No,
these
things
are
not
difficult.
 ____
Yes,
these
things
are
difficult.
 
 
 140 Appendix	
  2	
  -­	
  Associations	
  between	
   reported	
  synaesthesia	
  types Letter Number Wkdy/Mth Sound Word-­‐Taste Number	
  Fm OtherLetter-­‐Colour 3.00 3.54 2.70 2.26 2.34 2.68Number-­‐Colour 4.79 4.45 3.86 2.33 2.27 2.55Weekday/Month-­‐Colour 4.11 4.91 4.24 2.83 2.95 2.50Sound-­‐Colour 2.07 2.33 1.96 3.65 3.40 3.59Word-­‐Taste 1.63 1.62 1.60 2.59 2.99 2.10Number	
  Forms 1.57 1.58 1.45 1.66 1.54 2.04Other	
  Types 2.56 2.69 1.79 2.27 1.46 1.99Relative	
  rates	
  of	
  simultaneous	
  occurrence	
  of	
  different	
  forms	
  of	
  reported	
  synaesthesia	
  (how	
  much	
  more	
  likely	
  someone	
  who	
  reports	
  synaesthesia	
  of	
  Type	
  A	
  is	
  to	
  also	
  report	
  synaesthesia	
  of	
  Type	
  B	
  than	
  some-­‐one	
  who	
  does	
  not	
  report	
  synaesthesia	
  of	
  Type	
  A).	
  The	
  upper	
  triangle	
  is	
  taken	
  from	
  the	
  SFU	
  sample,	
  the	
  lower	
  from	
  the	
  CU	
  sample.	
  P-­‐values	
  are	
  all	
  <.001	
  after	
  Bonferroni	
  correction. 141 Appendix	
  3	
  -­	
  Associations	
  between	
   conRirmed	
  synaesthesia	
  types Letter Number Weekday Month Chord Scale InstrumentLetter-­‐Colour 66.53*** 25.75*** 36.96*** 44.35 79.84* 133.06Number-­‐Colour 65.98*** 52.86*** 49.22*** 0.00 41.65 41.65Weekday-­‐Colour 47.07*** 49.70*** 93.51*** 37.11 66.79 0.00Month-­‐Colour 50.18*** 43.66*** 52.11*** 89.35* 148.91* 89.35Chord-­‐Colour 0.00 35.26* 0.00 54.31 0.00 346.00Scale-­‐Colour 25.94** 23.90** 34.96 41.19 111.55 172.83Instrument-­‐Colour 48.15 27.23 13.72 13.56 0.00 79.71**Relative	
  rates	
  of	
  simultaneous	
  occurrence	
  of	
  different	
  forms	
  of	
  con>irmed	
  synaesthesia	
  (how	
  much	
  morelikely	
  someone	
  who	
  has	
  synaesthesia	
  of	
  Type	
  A	
  is	
  to	
  also	
  have	
  synaesthesia	
  of	
  Type	
  B	
  than	
  someone	
  who	
  does	
  not	
  have	
  synaesthesia	
  of	
  Type	
  A).	
  The	
  upper	
  triangle	
  is	
  taken	
  from	
  the	
  SFU	
  sample,	
  the	
  lower	
  from	
  the	
  CU	
  sample.	
  P-­‐values	
  are	
  all	
  Bonferroni-­‐corrected,	
  and	
  >	
  .05	
  except:	
  ***	
  p	
  <	
  .001,	
  **	
  p	
  <	
  .01,	
  *	
  p	
  <	
  .05 142

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0074011/manifest

Comment

Related Items