1 Reading skill and semantic transparency L. B. Feldman (The University at Albany-SUNY, Haskins Laboratories), P. Milin (University of Novi Sad, Tubingen University), F. M. del Prado Martín (University of California - Santa Barbara) K Cho (The University at Albany-SUNY) & H. Baayen (Tubingen University) ((( Haskins Labs))) National Institute Of Child Health and Development Grant HD-01994 to Haskins Laboratories 2 Semantically Transparent (S+): stem morpheme’s meaning recurs in complex and simple words farmer = farm + er buzzer = buzz + er Semantically Opaque (S-): stem morpheme’s meaning changes in the complex word corner NOT corn + er sadist NOT sad + ist Mapping between form and meaning is graded 3 Exhaustive decomposition: stem + affix stem occurs with a possible affix farmer = farm + er corner = corn + er ratify = rat + ify Partial decomposition: stem but no affix stem occurs in complex form w/ left over letters cornea NOT corn + ea rattan NOT rat+ tan Mapping between form and structure is graded 4 Form-with-meaning: form and meaning attributes of morphemes contribute concurrently to processing even at very short SOAs (34 ms). facilitation: batter-BAT> battery-BAT=battle-BAT Form-then-meaning: initial analysis is morpho-orthographic but semantically blind. (semantic processing arises later) facilitation: ratty-RAT = ratify-RAT > rattan-RAT Two perspectives on early morphological processing Forward masked priming at 48 ms SOA 5 unrelated UR unrelated opaque opaqueO Transparent transparentT UR-T 13 ms* UR-O 5 ms O-T 8 ms# Transparent Transparent unrelated Form-with-meaning at 34-100 ms SOAs Feldman et al., 2012 ms 6 unrelated UR unrelated opaque opaqueO Transparent transparentT UR-T 13 ms* UR-O 5 ms O-T 8 ms# Transparent Transparent unrelated Form-with-meaning at 34-100 ms SOAs Feldman et al., 2012 ms 7 Andrews and Lo 2013 SOA 48 ms Spelling high low Vocab high low 15 15 Overall, ss show a transparency effect facilitation: batter-BAT> battery-BAT Skill influences early morphological processing 8 Andrews and Lo 2013 SOA 48 ms Spelling high low Vocab high low N=15 N=15 Overall, ss show a transparency effect facilitation: batter-BAT> battery-BAT Skill influences early morphological processing 15: stronger semantic skill so T-O differ more than average 15: stronger form skill so UR-O differ more than average 9 Reading skill and semantic transparency Q1: Are transparency effects reliable at SOA 34 ms? Q2: Do vocabulary (meaning) and/or spelling (form) skill interact with transparency? Q3: If not, why did we think they did? 10 Targets with semantically transparent and opaque primes N = 63 11 12 Forward masked primed lexical decision task (Forster & Davis, 1984) Semantically related trial prime and target overlap Forward mask: ######## 500 ms batter Prime: 34 ms BAT TARGET: requires lexical decision; visible until response. 13 Forward masked primed lexical decision task (Forster & Davis, 1984) prime and target overlap Semantically unrelated trial ######## Forward mask: 500 ms Prime: 34 ms TARGET: requires lexical decision; visible until response battery BAT 14 RT ERR ANOVA Mean RT SOA 34 ms N= 73 ANOVA: Trans vs. Opaq vs. UR p < .001 Trans > Opaq p = .046 Hi spell Lo spell Hi V/Lo Sp Lo V/ Hi Sp Trans Opaq 626 639 0.04 0.05 O-T UR-T 13* 28* 18 24 8 32 UR 654 0.05 UR-O 15* 6 24 O-T UR-T 21 33 8 28 UR-O 12 20 15 Trans Opaq 626 639 0.04 0.05 UR 654 0.05 Hi spell Lo spell O-T UR-T 18 24 8 32 UR-O 6 24 Hi V/Lo Sp Lo V/ Hi Sp O-T UR-T 21 33 8 28 UR-O 12 20 RT ERR ANOVA Mean RT: spelling skill ANOVA Prime type p < .001 no main effect spelling p < .12 no interaction w/prime type 16 RT ERR ANOVA Mean RT: Vocabulary Hi spell ANOVA no main effect vocabulary (vocab) Lo spell no interaction w/prime type Hi V Lo V Trans Opaq 626 639 0.04 0.05 UR 654 0.05 O-T UR-T 18 24 8 32 UR-O 6 24 O-T UR-T 15 23 11 33 UR-O 8 22 17 ANOVA Mean RT: High Vocabulary and low spelling (N=15) vs. Low Vocabulary and high spelling (N=15) pattern replicates Andrews & Lo RT ERR Hi spell Lo spell Hi V/Lo Sp Lo V/ Hi Sp Trans Opaq 626 639 0.04 0.05 O-T UR-T 28 13 18 24 8 32 UR 654 0.05 UR-O 15 6 24 O-T UR-T 21 33 8 28 UR-O 12 20 18 Generalized additive model (GAMM) 1. Predict response RT from predictor variables as in general linear model 2. Works with mixed effect of factors, covariates (numeric predictors), random effects simultaneously 3. Works with non-linear effects (smooths) and interactions (tensor products) -Smooths for ss over trials -Smooths for items 4. AIC includes penalty as each smooth function is added (wiggle vs. fit) 5. Can give better fit than other models 19 GAMM analysis: Principal Component Analysis Frequency HAL K-F Form-related number of orthographic neighbors number of phonological neighbors word length 20 fast slow Log HAL freq; Log K-F Differences between targets (primes ns) high scores on PC1 pair longer words and words w/ small(er) # of neighbors 21 RESULTS: Transparency effect SOA 34 ms (SE) GAM N= 73 non-linear interaction of vocabulary size by spelling proficiency … consistent w/ claims of Andrews and Lo 2013 Prime type 22 Ss vary over trials Generalized additive mixed model (GAM) smooths for random byparticipant variation over trials 23 Ss vary over trials Generalized additive mixed model (GAM) smoothes for random byparticipant variation over trials. Inclusion of participant variation over trials in model renders interaction of vocabulary size and spelling proficiency unnecessary! 24 Properly modeled by-participant random variation leaves little evidence for skill effects AIC values of goodness-of-fit (small is better) 25 FINAL MODEL: transparency, nonlinear PCs and ss/trial, target effects but no skill effects AIC values of goodness-of-fit (small is better) 26 Ramscar, Hendrix, Shaoul, Milin, & Baayen (2014): "psychometric vocabulary measures are virtually guaranteed to fail (...) because they attempt to extrapolate vocabulary sizes from sets of test words that are biased towards frequent types (Raven, 1965; Heim, 1970; Wechsler, 1997)". Reliable estimation of vocabulary sizes from small samples is mathematically impossible (Baayen, 2001). Individual differences vs. “skill” effect? 27 34 ms SOA unrelated opaque transparent Transparent Transparent unrelated Q1: Form-with-meaning at 34-100 ms SOAs 28 34 ms SOA unrelated opaque transparent Transparent Transparent unrelated Q2: Reading skill plays a minor role when individual differences are are modeled 29 Q3: Apparent “skill effect” derives from: differences between participants within a session “Interaction” with skill O-T Hi V/Lo Sp 21 Lo V/ Hi Sp 8 UR-T 33 28 UR-O 12 20 random smooths for trials 30 Form-with-meaning acct of word recognition -Even early in word recognition (34ms) -Not tied to reading skill CONCLUSION: -characteristic of all readers despite their idiosyncrasies 31 Thank-you Form-with-meaning Even early in word recognition (34ms) Not tied to reading skill 32 unrelated UR + Opaque UR-opaque Opaque Transparent Transparent Transparent Transparent Transparent Transparent FORM-WITH-MEANING: EVEN AT SHORT SOAS Transparent Form-with-meaning: even at short SOAs 33 Vector representation of variables in PCs Frequency HAL frequency Brown frequency Form-related number of orthographic neighbors number of phonological neighbors word length variables Log HAL freq; Log K-F
© Copyright 2026 Paperzz