Steve Lupker1, Jimmie Zhang1, Jason Perry1 and Colin Davis2

1University of Western Ontario 2University of Bristol

Transposed letter nonwords like jugde are perceived as being more similar to their base word (i.e., JUDGE) than replacement letter nonwords like jupte are. In masked priming experiments, this similarity difference shows up in that jugde produces a larger priming for JUDGE than jupte does.

Potentially, the following bigrams can be activated: JU, JD, JG, JE, UD, UG, UE, DG, DE, GE In some versions of the model, JE is not activated In some versions of the model, _J and E_ are activated (e.g., SERIOL) These models can predict transposed-letter effects and other similar effects (e.g., relative position priming effects) based on the open-bigram units activated by the primes. However other models (e.g., SOLAR) can predict them too.

We reported that both a) cetupmor (versus cifagnar) and b) pmocretu (versus vifagnas) were a successful primes for COMPUTER, which SOLAR predicts but open-bigram models do not because the primes and targets share no open bigrams. More recently, demonstrations of larger transposed-letter effects than transposed-number or symbol effects suggest that letter processing may involve an extra level of representation, potentially, open-bigram units.

A basic principle of the open-bigram position is that similarity is maximal when two letter strings contain exactly the same letters in the same relative positions. Therefore, those types of primes should be extremely effective. Potential example: Prime: jupdgoe Target: JUDGE However, the JG, JE and UE bigrams may be too far apart to be activated by jupdgoe limiting its priming for JUDGE.

First-letter superset primes: xjudge-JUDGE Last-letter superset primes: judgex-JUDGE One-letter replacement primes: juxge-JUDGE Some open-bigram models’ similarity scores: 1.0, 1.0, .66 SERIOL’s similarity scores: .93, .93, .66 JU, JD, JG, JE, UD, UG, UE, DG, DE, GE SOLAR’s similarity scores: .88, .88, .88 SOLAR’s predicted priming effects: 39 ms, 39 ms, 36 ms

First-letter superset primes (xjudge-JUDGE): 27 ms Last-letter superset primes (judgex-JUDGE): 33 ms One-letter replacement primes (juxge, jxdge or judxe-JUDGE): 23 ms (compared to unpronounceable control primes) Note: 1) not our stimuli 2) prime length confound

Standard masked priming: Related Unrelated Priming First-letter 640 671 31 Last-letter 636 659 23 Replacement 639 682 43 No interaction (F1(2, 60) = 1.37, p > .20; F2 (2, 144) = 1.10, p > .30)

JUDGE

judge

##### 500 ms

33 ms

xjudge

55 ms

SOLAR’s predictions: xjudge-JUDGE: 64 ms judgex-JUDGE: 64 ms juxge-JUDGE: 59 ms

Masked sandwich priming: Related Unrelated Priming First-letter 662 701 39 Last-letter 644 696 52 Replacement 644 702 58 No interaction (F1(2, 60) = 0.63, p >.50; F2(2, 144) = 1.91, p >.10)

judge #####

JUDGE

xjugde

Reference stimulus - 1000 ms

50 ms

until response

Masked priming in same-different task: Related Unrelated Priming First-letter 548 570 22 Last-letter 531 570 39 Replacement 520 576 56 Interaction (F1(2, 60) = 6.79, p <.005; F2(2, 144) = 8.42, p <.001)

1) SOLAR does a reasonably good job of explaining masked priming, lexical decision data. 2) Similarity ratings from all models do a poor job of predicting masked priming, same-different data. 3) It seems likely that the codes used in the two tasks are different. How (and why) they’re different is far from clear. 4) With reference to the question we started with, we still haven’t been to produce any situations where open-bigram models do a better job of explaining priming data than other types of models. Therefore, we’re still searching for some evidence that orthographic coding actually involves the activation of open-bigram units.

Olessia Jouravlev Eric Stinchcombe Mark McPhedran Arum Jeong