Can the dual-route cascaded computational model of reading offer a valid account of the masked onset priming effect?

Abstract

The masked onset priming effect (MOPE) refers to the empirical finding that target naming is faster when the target (SIB) is preceded by a briefly presented masked prime that starts with the same letter/phoneme (suf) than when it does not (mof; Kinoshita, 2000, Experiment 1). The dual-route cascaded (DRC) computational model of reading (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001) has offered an explanation for how the MOPE might occur in humans. However, there has been some empirical discrepancy regarding whether for nonword items the effect is limited to the first-letter/phoneme overlap between primes and targets or whether orthographic/phonological priming effects occur beyond the first letter/phoneme. Experiment 1 tested these two possibilities. The human results, which were successfully simulated by the DRC model, showed priming beyond the first letter/phoneme. Nevertheless, two recent versions of the DRC model made different predictions regarding the nature of these priming effects. Experiment 2 examined whether it is facilitatory, inhibitory, or both, in order to adjudicate between the two versions of the model. The human results showed that primes exert both facilitatory and inhibitory effects.

Keywords:

• Reading aloud
• Masked onset priming effect
• Dual-route cascaded computational model of reading

Acknowledgments

We would like to thank Ken Forster and an anonymous reviewer for valuable comments and suggestions.

Notes

¹ For the DRC 1.1.4 model to show a MOPE with three-letter-long CVC nonword prime–target pairs, the minimum prime duration had to be 40 cycles.

² In human experiments that use the three-field paradigm masking procedure, which is the procedure used in the studies that we tried to simulate with the DRC model in the present study, a forward mask is presented first, followed by the prime, followed by the target. The target acts as a backward mask to the prime so that the prime's letters are not visible to the participants at target onset. In order to simulate backward masking effects with the DRC model in the same way as they occur with humans, the activation of the prime's letters at the letter level was completely switched off at target onset so that the letters of the prime would not be available to the model when the target was presented.

³ Given that different components of the human reading system are likely to be involved in word versus nonword reading (e.g., lexical vs. nonlexical route for word and nonword reading, respectively) we only refer here to the discrepancies among empirical studies of masked priming that used nonword stimuli.

⁴ It is worth noting that although the orthographic priming effect in that study was significant in the subject analysis, it was not significant in the item analysis, and therefore the effect was not consistent across items.

⁵ The results from this experiment also show an advantage of initial over final overlap between the prime and the target, which agrees with the idea that a nonlexical serial mechanism must be processing the prime in a left-to-right manner.

⁶ The differences between the two versions are fully documented at: http://www.maccs.mq.edu.au/~ssaunder/DRC/DRC-Differences.pdf

⁷ On the same site the differences between this version and the original DRC version—that is, DRC 1.0 (Coltheart et al., 2001)—are fully documented. Also, a document entitled Incremental Modelling reports results from the simulations (with DRC 1.2) of all the benchmark effects on reading aloud (as listed in Perry, Ziegler, & Zorzi, 2007, p. 301) that DRC 1.0 could simulate. Both for the simulations of these benchmark effects and for the simulations of the experiments reported in the present paper the default parameters installed in the downloadable model were used.

⁸ The findings from a study that Finkbeiner, Almeida, and Caramazza (2006) conducted showed that nonletter distractors (e.g., ^∧% ∼ *#) engage a bilaterally distributed mechanism responsible for detecting letter shapes: the first stage in the reading process, as the authors suggest on page 1098 of the corresponding paper.

⁹ Coltheart, Davelaar, Jonasson, and Besner (1977) defined neighbourhood size (N) as the number of words differing by a single letter from the stimulus, preserving letter positions—for example, worse and house are orthographic neighbours of horse.

¹⁰ Due to an oversight, two pairs in the unrelated condition shared the same phoneme in the same position—that is, dys–PIV and pym–VIC.

¹¹ In the item analysis prime type was a between-groups factor, and therefore a univariate analysis of variance was carried out with prime type (one-letter, two-letters, unrelated) and list (A, B, and C) as fixed factors.

¹² When we looked at the individual RTs we noticed that 16 out of 24 participants showed a naming-latency advantage in the two-letters overlap condition in comparison with the one-letter overlap condition. Similarly, 47 out of 81 targets yielded faster naming latencies in the two-letters overlap condition than in the one-letter overlap condition.

¹³ The reason we used a prime duration of 50 ms as a within-condition baseline is that, to our knowledge, all studies in the literature that reported a MOPE with nonword items used prime durations no shorter than 50 ms. Therefore, it is unknown whether the effect would still be present at shorter prime durations. We are aware of the fact that at prime durations as long as 70 and 90 ms, there is the possibility that primes become visible to the participants, and therefore the priming effects observed at such long prime durations, as well as their underlying mechanisms, might well be strategically affected. However, we would like to remind the reader that the only aim of Experiment 2 was to compare the trend (facilitatory, inhibitory, or both) of the three prime-type conditions as they develop in time between the human data and the computational data in order to adjudicate between two versions of the DRC model that offer a different explanation for how priming effects arise in humans. Given that consciousness does not affect the model's performance, if conscious perception of the primes at the long prime durations affects human behaviour in a way that would particularly influence the nature of the observed priming effects, then we should not be able to find a pattern of results in the computational data that could possibly match that of the human data.

¹⁴ The reason we blocked the items by prime type and not by prime duration was that, in a different experiment that we carried out in our laboratory in order to investigate whether the MOPE is orthographic or phonological in nature, we used three prime durations (30, 50, and 70 ms), which were presented in pure blocks in a within-subjects design. The results showed that the priming effects observed at the prime duration of 50 ms, in particular, were greatly influenced by whether the prime duration of the preceding block of trials was short (30 ms) or long (70 ms), indicating carry-over effects between the blocks.

¹⁵ Although this effect was not statistically significant it is worth pointing out that some minor facilitation (5 ms) was observed as prime duration increased from 50 to 70 ms, while the effect remained flat between 70 and 90 ms. The reason why this effect deserves some attention is discussed in the Discussion section.

¹⁶ Due to our experimental design, where each participant saw each target three times, each time preceded by a different type of prime and a different prime duration, the item analysis could only be carried out across participants.

¹⁷ We would like to thank Ken Forster for pointing out this alternative explanation of the flat effect observed in the one-letter overlap condition.

¹⁸ This parameter serves for simulating different levels of speed in reading aloud. In particular, when the value of this parameter is high, self-paced reading is simulated, while when it is low, speeded naming is simulated.

¹⁹ We would like to point out that the only reason we used very long prime durations in the two versions of the DRC model was that we wanted to observe whether the pattern of results that the computational data show in each of the three prime-type conditions as prime duration increases is similar to the corresponding pattern of results that the human data show so as to determine whether priming effects in DRC arise in the same way as they do in humans. Evidently, we do not intend to use such long prime durations for generally simulating priming effects with the DRC model, because if the experimental stimuli consisted of word items, for example, it is very likely that at such long prime durations the model would pronounce the primes instead of the targets in many cases. Thus, the use of long prime durations in the model are limited to the simulations of the human data from Experiment 2.

²⁰ It is worth pointing out that the RTs in the two-letters overlap condition did not increase between the short and the medium prime durations.

²¹ The reason the F values in the reported simulation results are so high is that there was hardly any variance in the model's RTs across items.

²² Since the priming effects in DRC 1.1.4 only arise because of the interference from the unrelated prime to the target, there was no point in trying to modify any of the parameters in order to simulate the human data from Experiment 2, because the model would still be unable to generate a facilitatory priming effect.