Reading different writing systems: The grapholinguistic equilibrium hypothesis Mark S. Seidenberg University of Wisconsin-Madison Golden era for reading research! Not just English; many writing systems, languages Lots.
Download ReportTranscript Reading different writing systems: The grapholinguistic equilibrium hypothesis Mark S. Seidenberg University of Wisconsin-Madison Golden era for reading research! Not just English; many writing systems, languages Lots.
Reading different writing systems: The grapholinguistic equilibrium hypothesis Mark S. Seidenberg University of Wisconsin-Madison Golden era for reading research! Not just English; many writing systems, languages Lots of progress! One of the big success stories in cognitive science/neurosci (links to education: not so good, at least in US) My own work Brain circuits Research at UW, Medical College of Wisconsin, Haskins Labs (CT) Computational models Connectionist models that simulate detailed aspects of acquisition, skilled performance. Dyslexia = anomalies in how system develops Behavior Children, adults Normal, dyslexic English, Serbian, Chinese, other writing systems For today’s talk, I tried to pick a topic that is of interest to this aud Writing Systems and Reading • Do properties of writing systems affect – – – – Skilled reading Learning to read Brain circuits that support reading Dyslexia We have this framework…. context meaning spelling sound 1. Mappings between codes are statistical, not categorical 2. Ouput determined by multiple constraints 3. Division of labor among components varies between writing systems between individuals Other models? There are some. Not the time or place to do comparisons. But, DRC Doesn’t Read Correctly And CDP+ Can’d Do Pronunciation, and other stuff Post-hoc fitting of models to data. Only allows models to fit individual studies of a phenomenon, s “incremental, nested”? No, not actually. But that’s a different talk Impact of writing systems: an area where dual-route models have little to say • Fitting models to writing systems/languages • Each gets a different model, different parameters • Similarities/differences built in: have to know them already • No learning • No semantics • No “why” But that’s a different talk It’s a golden era for reading research but it’s taken a while for cross-linguistic issues to come into focus Most research: it’s about the properties of writing systems Orthographic depth I don’t think this is quite right. It’s about writing systems and the languages they represent There are tradeoffs between writing systems and languages There is Grapholinguistic Equilibrium Confidential: I don’t actually know how every writing system in the world For example, I don’t know český jazyk Grapholinguistic Equilibrium is a hypothesis. Most of the evidence is circumstantial. Not much direct experimental evidence. Let’s do an experiment here! Now! When you hearthi s Ask: is it true of český jazyk? If it is, great. If it isn’t, I’ll go So: semantics orthography phonology Writing affords routes to meaning! ALL writing Whether your word is or or PICTURE obrázek An early division of labor theory: Orthographic Depth semantics orthography phonology Orthographic depth hypothesis: shallow: more orth-phon-sem deep: more orth-sem English: both 1980s Katz, Turvey, Haskins Labs Among alphabetic writing systems, English is unusual many inconsistencies unlike Finnish, Italian, Russian, Korean, Czech others orth-->phon is a big issue for English learners not for everyone else We don’t want theories of reading to be based on the outlier data It’s true that written English differs from shallower alphabetic sy Assumptions derived from English may not be valid. Findings differ in important respects. But, there are no “outlier” orthographies. Just: Different tradeoffs between writing systems and languages Is English an outlier? For example, Learning to read: are shallow orthographies e Case study: Welsh vs. English • Welsh: shallow English: deep • Different schools, same communities • Natural controls for SES etc. (These are older studies, Marketa.) Ellis & Hooper, 2001: Welsh-reading 7 year olds correctly name twice as many words as English readers Percent Correctly Pronounced Spencer & Hanley, 2003: 6 year olds 100 90 80 70 60 50 40 30 20 10 0 Out of 30 items/ condition Welsh English Words Nonwords Similar findings in other languages Italian Spanish German French Finnish Serbian Turkish Albanian others Handbook of Orthography and Literacy, Joshi & Aron (Eds.), Erlbaum 2006 Seymour et al. (2003) Ziegler et al. (2010) Czech? Issue: These studies equate “reading” with “reading aloud” Question: What is the relationship between reading aloud and comprehension? Not tested or not tested in detail. The word “comprehension” does not occur in this article. Why not? 1. Many studies of English show that learning orth-phon is hard Reading aloud is related to comprehension skill 2. Therefore, writing systems that make it easier to learn orth-phon should be easier to learn to read = comprehend Case where thinking was too tied to studies of English. But Reading Aloud ≠ Reading I shall demonstrate… 1. Dissociations of reading aloud and comprehension Good reading aloud Zero comprehension Bar Mitzvah Languages For the Bar/Bat Mitzvah, the boy/girl • • • • Must be able to read Hebrew aloud Do not have to comprehend Can be done if the writing system is shallow Which vowelled Hebrew is. Welsh: also a very good Bar Mitzvah language! Do Shallow Orthographies Promote Better Comprehension? Not in the Welsh-English studies Ellis and Hooper Pronunciation Welsh > English Comprehension English > Welsh* Hanley et al. Comprehension English > Welsh Correlation between pronunciation, comprehension: English highly significant Welsh n.s. “This result suggests that a transparent orthography does not confer any advantages as far as reading comprehension is concerned. As comprehension is clearly the goal of reading this finding is potentially reassuring for teachers of English.” Hanley et al. 2004 What about other Bar Mitzvah Languages? Turkish: Aydin Durgunoğlu has looked at both reading aloud and comprehension in detail “Phonological awareness and decoding develop rapidly in both young and adult readers of Turkish because of the transparent orthography and the special characteristics of phonology and morphology. However, reading comprehension is still a problem.” Durgunoğlu, 2006 Also true of other shallow orthographies? Czech? 2. People comprehend words they cannot pronounce correctly English speakers all have (or had) words of this sort in our vocabularies. Egregious Piquant Suave Rapport Quay Non-pareil Automata Chaos Coitus URANUS “URINE-OUS?” If we tested my reading aloud, I might perform more poorly than Welsh readers too. 3. How shallow are shallow orthographies? Writing systems are not transcriptions of speech. Information relevant to pronunciation is left out. Creates limit on strictly orth-phon-sem processing. Example: Serbo-Croatian, the original “shallow” orthography Grapheme-phoneme correspondences easy, but not sufficient Pronunciation requires more syllabic stress: pitch accent: ZATvori prisons RIBA fish vs. to scrub zatVORi to shut LUK onion vs. arch PROIZvodi proizVODi products to produce A lot like English! CONduct conDUCT Czech? 4. What prevents people from learning orth->sem? Even in shallow orthographies? Harm & Seidenberg (2004) division of labor model learned Orth-phon-sem Orth-sem At the same time. Maybe people do too. 5. If shallow is so GREAT, what about Hebrew? It’s shallow all right… … but they leave out the vowels! 6. And what about the spoken language? • Writing systems differ • So do the languages they represent Comprehension depends on both! Gough, Simple view of reading Decoding X Spoken language comprehension Orthographic depth Morphological complexity DEEPER Morphologically simple SHALLOWER Morphologically complex English, Chinese Finnish, Serbian, Italian, Russian Albanian, Welsh, Spanish, etc. Czech? Why would this relation hold, in general? Consider Serbo-Croatian • They get the spelling-sound correspondences for free • But the morphology is very complex! 3 genders masc fem neuter Czech 2 numbers sing plural 7 cases nominative genitive dative accusative instrumental locative vocative Mirkovic, Seidenberg, Joanisse (Cognitive Science, 2011) Model of learning Serbian inflectional system Now: imagine learning to read Serbian if, as in English, many letters had multiple pronunciations A few consonants like C and G Each vowel represents many sounds This additional level of complexity would make the system vastly more difficult to learn. Too hard! Contrast: Learning to read in English • • NOT: one spelling - one sound – But, irregulars are mostly short, high-frequency words – And not arbitrary: HAVE is not “glorp” The inflectional system is trivial – Number on nouns, tense and number on verbs – Makes words shorter too Grapholinguistic Equilibrium The simple view of writing systems and reading: G = orth opacity x linguistic complexity) English: high opacity, low complexity Serbian: low opacity, high complexity Languages/writing systems tend to keep G constant. In other words Languages get the writing systems they deserve (why English spelling reform is pointless) Even more broadly Writing systems provide cues about sound and meaning Late Hieroglyphics Hememu = “humanity” Sound cues sound meaning + Meaning cues (man, woman, many) Chinese “Mother” Semantic cue “radical” Sound cue “phonetic” In Hebrew and Arabic K-T-B In English Redhea d Blockhea d Deadheads Morphemes = convergence of sound and meaning In Serbo-Croatian all related to “advisor” Lemmas = strong semantic cues In Czech Conclusions • Most comparative research on reading has focused on reading aloud – definitely easier in shallow orthographies • However, comprehension depends on knowledge of spoken language • Spoken languages vary in “morphological depth” and other ways • Tradeoffs between properties of writing systems and languages = grapholinguistic equilibrium • Writing systems are codes for conveying sound + meaning, universally. context meaning spelling sound Thanks for listening! Are Some Writing Systems Easier to Learn to Read? We won’t know without taking into account properties of spoken language But it doesn’t look like it. Every theory/model must have these basic elements But, additional assumptions: • Each code is learned,constrained by other codes; context meaning spelling sound • Interactivity, not modularity; • Information encoded by collections of units, etc. Not specific to reading. What is different about “dual-route” models (e.g., DRC, CDP+): But, for what? Also: Models don’t address computation of meaning! In the US there are many people who are poor at pronouncing words and nonwords aloud: Millions of readers taught by Whole Language Method! X No no no! “Dual-route” models are different!!! Two routes to phonology Lexical route phonology orthography Nonlexical route Not about computing meaning! Other talk! University of Wisconsin-Madison A birthplace of American psychology (1886) NRC rankings, 2010 DRC ̬