The Concept of The Priming Based on Forster & Davis

Introduction

Priming is a phenomenon that occurs when exposure to one stimulus affects responses to a later stimulus (e.g., seeing a dog triggers buying dog food at the convenience store). Semantic priming is a type of priming where the prime (initial stimulus) is semantically—linguistically or logically—from the same category as the target (later stimulus) (Collins & Loftus, 1975). The theory of spreading activation explains semantic priming; the activation of a neural network causes activation of semantically-related (associative) neural networks (Collins & Loftus, 1975). The theory of spreading activation is from M.R. Quillian’s theory on semantic processing in memory (Quillian, 1967). Quillian theorized that when searching through memory, humans activate semantically-related networks (Quillian, 1966). Collins & Loftus applied Quillian’s theory to priming (1975). When a concept is primed, links to semantically-related networks are activated, speeding up processing and therefore the reaction when those semantically-related networks are accessed again within a short period of time (in a colloquial sense, those networks are already “warmed up” by the initial activation from priming). Morphemes, the smallest grammatical unit, could also be primes, but for the purposes of our experiment, the primes were complete words. Because of spreading activation, if a prime and target are semantically related, i.e. semantically primed, reaction to the target stimulus should be quicker. Forster & Davis expanded this theory to experiments explicitly examining differences between masked and unmasked priming (1984). Masked priming is when the prime is presented only fleetingly—not long enough for a person to consciously comprehend it—so a person is able to only process it subliminally. The prime is masked by sandwiching it in between two masks, such as a row of hashtags or another word, that conceal the prime.

Forster & Davis used short stimulus-onset asynchrony (SOA) so that participants would react to the target automatically, without employing more complex strategies (1984). In the same study, Forster & Davis found that when there was a repetitive prime to the target word—where the prime and target were the same word but were represented in different cases (i.e. lower and upper case)—participants reacted statistically significantly faster when responding, regardless of whether the target word was a real word or a made-up word for both unmasked and masked conditions. However, when differences were examined between prime words that are commonly used in everyday vocabulary (high-frequency words) and more obscure words (low-frequency words), Forster & Davis found that unmasked primes resulted in faster reaction time to targets than masked primes (1984). Thus, superliminal processing triggered by unmasked primes as opposed to subliminal processing must affect semantic network activation in a different way than masked priming, if masked priming affects semantic network activation at all.

Ratcliff, Gomez, & Perea answered this question with their theory on priming and the diffusion model (2013). The diffusion model states that when information is automatically processed and a reaction is required—manipulated with low SOAs—information is first encoded, processed, and then the subject comes up with a reaction; the overall process has those three parts to it (Ratcliff, 1978). Using semantic priming experiments, Ratcliff, Gomez, & Perea concluded that when participants were asked if the target was a word or nonword—with both masked and unmasked primes—when primes were semantically related to the target word, participants reacted faster when asked whether the target word was a real or made-up word. However, when the prime was unmasked, the prime appeared to affect both encoding and processing under the diffusion model; with masked primes, only encoding was affected (2013). Ratcliff, Gomez, & Perea theorized that this contrast results from differences in semantic network spreading activation; unmasked primes result in a greater amount of semantic relation, and thus result in greater activation of semantic networks. On the other hand, masked priming only activates semantic networks enough to decrease encoding time as participants already have a “running start” in encoding of the target word if they had subliminal knowledge of a related target word (Ratcliff, Gomez, & Perea, 2013). Ratcliff, Gomez, & Perea explained this disparity with differences in the quality of information received—unmasked priming resulted in a much greater quality of information received, and greater semantic network activation (2013). As a result, unmasked priming resulted in a more significant reduction of reaction time to targets that were semantically related to the prime in comparison to targets that were not semantically related to the prime, compared to masked priming (Ratcliff, Gomez, & Perea, 2013).

In this experiment, we wanted to see if unmasked priming reduced reaction times more compared to masked priming when primes were semantically related to the target word rather than semantically unrelated. Therefore, our experiment had participants react to either a semantically related word, unrelated word, or a nonword after being shown a masked or unmasked prime. Because semantically primed targets should already be triggered by spreading activation, and because unmasked primes offer greater quality of information than masked primes to increase spreading activation, we hypothesize that the semantic priming effect would be greater for unmasked primes than masked primes, resulting in a greater decrease in reaction time between the semantically related and unrelated conditions involving the unmasked prime than between the semantically related and unrelated conditions with the masked prime.

Methods

Participants

We had 15 participants, with each group member recruiting 5 participants each. Participants were friends and acquaintances of the group members who are also college students at the Claremont Colleges. They were acquired by personally asking if they would participate in our study.

Materials and Procedure

For the experiment, each participant used a program called E-Prime on a Mac computer which was running Windows in Bootcamp. At the beginning of the experiment, each participant was shown a slide with a consent form and had to press the spacebar to give consent after reading it in order to proceed. Afterwards, a slide of experimental instructions appeared and the participant was asked to press any key once the participant was ready to begin the trials. Then, the participant went through 200 trials of two different types of primes—masked and unmasked. In the trials with the unmasked primes, the sequence of stimuli presented to the participant was a fixation cross for 1250 milliseconds, then a prime word for 45 milliseconds, then a blank screen for 250 seconds, and lastly the target. In the masked trials, the sequence of stimuli presented to the participant was a fixation cross for 1000 milliseconds, then a row of hashtags (the mask) for 250 milliseconds, then the prime for 45 milliseconds, then another mask with a row of hashtags for 250 milliseconds, and lastly the target. Otherwise, the trials for masked and unmasked primes were exactly the same. According to the experimental instructions, if the target was a word, the participant was supposed to press the key “f,” and if the target was not a word, the participant was supposed to press the key “j”. The target remained on the screen until the participant pressed either “f” or “j” in response. As soon as the participant pressed either “f” or “j” as an answer, the participant immediately moved on to the next trial. Trials from the different conditions (detailed below) were presented in random order to prevent the response from one trial to influence the response on another (e.g. numerous trials in a row that all had a real word as the target). The stimuli were shown for the amounts of time mentioned above in order to preserve the same trial time of 1545 milliseconds and the stimulus-onset asynchrony from the beginning of the prime until the beginning of the target of 300 milliseconds across both trial types. The SOAs were chosen to maintain automatic processing required for semantic network spreading activation to have an effect on response to the target word, and were drawn from various previous studies (Forster & Davis, 1984).

As previously stated, there were 200 total trials in the experiment, which was divided into 10 blocks, with 20 trials per block. After each block, the participant was given a slide of instructions that indicated a break and was instructed to continue on to the next block when ready by pressing any key. There were six conditions: masked related pairs, masked unrelated pairs, masked nonword pairs, unmasked related pairs, unmasked unrelated pairs, and unmasked nonword pairs. The masked related pairs condition and the unmasked related pairs condition contained 25 trials each. The masked unrelated pairs condition and the unmasked unrelated pairs condition contained 25 trials each. The masked nonword pairs condition and the unmasked nonword pairs condition contained 50 trials each. The order of the 200 trials were randomized so that they would be presented in a random order, and the time the participant took to press either the key “f” or “j” in response to whether the target was a word or not was recorded.

The stimulus consisted of a list of 200 prime-target word pairs. To create this list, we first made a list of 200 associated prime-target word pairs by utilizing Appendix A of the University of South Florida’s database of association norms (Nelson, McEvoy, & Schreiber, 1998) and ensured that our prime and target words did not appear more than once and that our prime words began with a variety of different letters and were of a reasonable length. Then, we randomly assigned each prime-target pair to one of our six experimental conditions (mask related, mask unrelated, mask nonword, unmasked related, unmasked unrelated, and unmask nonwords). The purpose of the randomization process was to not have systematic differences across trial types. Afterwards, we made a list of 100 nonsense words of reasonable length from the ARC Nonword database (Rastle, Harrington, & Coltheart, 2002). Subsequently, we replaced the second half of the target words of the prime-target word pairs with the nonwords. With 100 associated word-pairs and 100 nonword-target pairs, we then switched the order of 50 of the associated word-pairs so that they would become unrelated word-pairs and checked that the 50 unrelated pairs were truly unrelated. In the end, we had a stimuli list of 200 prime-target word pairs, consisting of 50 semantically related prime-target word pairs, 50 unrelated prime-target word pairs, and 100 nonword prime-target pairs. Half of those 200 prime-target word pairs (25 related prime-target word pairs, 25 unrelated prime-target word pairs, and 50 nonword prime-target pairs) had the prime masked in the experiment by having a row of hashtags that was two hashtags longer than the longest word in the stimulus list immediately precede and follow the prime, while the other half of the 200 word pairs were not masked in the experiment.

Results

In this experiment, a subject’s response to the target was correct if the subject pressed “f” when the target was a word and “j” when the target was a nonword. A correct answer had an accuracy value of 1 and an incorrect answer had an accuracy value of 0. The overall accuracy across all conditions and all subjects was 97.8% out of 200 trials.

The masked nonword and unmasked nonword conditions were excluded from the analysis because they were irrelevant for our hypothesis. They were a side task for the subjects to do during the experiment, and what we really wanted to focus on was the semantically related and unrelated trials. Inaccurate trials and trials with response times above 2000 ms were excluded from the analysis. For masked word related trials, 5.1% of trials were excluded; for masked word unrelated trials, 6.1% of trials were excluded; for unmasked word related trials, 9.1% of trials were excluded; for unmasked word unrelated trials, 5.8% of trials were excluded.

After exclusion of inaccurate and outlier trials, the mean reaction time to masked word related trials was 715.24 ms (SD = 150.88); the mean reaction time to masked word unrelated trials was 738.80 ms (SD = 206.89); the mean reaction time to unmasked word related trials was 694.45 ms (SD = 161.27); the mean reaction time to unmasked word unrelated trials was 768.60 ms (SD = 238.51).

Paired one-tail t-tests were run on both masked and unmasked conditions to find out whether there were statistically significant differences between participant response (answering whether target was word or nonword) reaction times when the prime was semantically related or not. For the masked conditions, we found out that t1 = -0.9446 with df = 14. The one-tail p-value was p = 0.1804. This result was not significant at the 5% level. For the unmasked conditions, t2 = -1.8087 with df = 14. The one-tail p-value was p = 0.0460, making this result statistically significant at the 5% level.

Discussion

From our results and the bar graph above, we can see that the average response time to word related trials is faster than that of word unrelated trials in both the masked and unmasked conditions, and that the difference in average response times for the unmasked trials is larger than that for the masked trials; this result is consistent with our hypothesis and with experiments done by Ratcliff, Gomez, & Perea (2013). The t-tests we performed show that the response time difference between masked word related trials and masked word unrelated trials is not statistically significant at a one-tailed 5% level. In contrast, the response time difference between unmasked word related trials and unmasked word unrelated trials is statistically significant at a one-tailed 5% level. These results and observation from the bar graph indicate that there is a semantic priming effect under both masked and unmasked conditions, and this result also supports our hypothesis that the semantic priming effect is larger for unmasked prime words than for masked prime words. However, according to the one-tailed t tests we ran, we cannot conclude that the semantic priming effect for masked primes is statistically significant at the 5% level. In other words, we cannot reject the null hypothesis that there is no semantic priming effect for masked primes. The theory about the quality of information from masked and unmasked primes that motivated our hypothesis could explain this result; information from masked semantically primed perceived subliminally activated just enough semantic network spreading that participants had a “jump start” in encoding. On the other hand, unmasked semantically-related primes perceived supraliminally gave activated semantic networks enough information for “jump starts” during both the encoding and processing stages of the diffusion model (Ratcliff, Gomez, & Perea, 2013). However, there are a couple points of caution about this result, despite its conformity with our hypothesis and previous theory. This outcome might also be explained by the small number of participants we had, and increasing the number of participants in future experiments may help in generating more significant results. Excluding trials with extremely fast reaction times could also alter our result. Moreover, the three non-native English speakers among the 15 participants could have introduced a bias into the outcome since they may not process English words as quickly as native speakers do or may process them differently since their associations between English words may be weaker or different. Therefore, a question that could be addressed in follow-up research is whether there is a difference in how the semantic priming effect works between native English speakers and non-native English speakers.