Study One showed that a priming effect can be found between extremely strong collocates (MI > 8). It also suggested that collocational priming may exist
independently of whether collocates are salient enough to be listed as psychological associates, though some reservations remain about this conclusion. The present study aims to extend these findings by considering a) whether similar effects can be found for less extreme examples of significant collocation; b) whether this effect is indeed independent of association; and c) whether the effect is independent of mere semantic congruence.
The thinking behind question c) is that the collocating word pairs used in the previous study (e.g., death-penalty; elder-brother) form meaningful phrases, whereas the control pairs are often difficult to make sense of (e.g. east-penalty; leather-brother). It seems possible that the second part of a word pair which has a plausible meaning may
be recognised faster than the second part of a pair which is highly implausible, regardless of whether the pair forms a significant collocation, since semantic congruence alone will provide the subject with evidence that the target is indeed a word. It may be, then, that the priming effects found in Study One are due to mere semantic congruence, rather than to frequency of collocation.
To answer these three questions, the present study measures priming between three levels of collocation:
• moderately strong collocations which are not strong normative associates; • strong collocations which are not strong normative associates;
• strong collocations which are strong normative associates.
It uses for non-collocating controls only word combinations which are attested more than once in the BNC, and so which are presumably semantically plausible.
The inclusion of moderate strength collocations is intended to address question a); more rigorous tests of association are introduced (see below) to address question b); and the use of attested controls is designed to address question c)
Materials
16 collocating prime-target pairs were created for each of three conditions:
• Level 1 pairs are moderately strong collocations, having MI scores of 4-5 (Mdn = 4.47) and t-scores of 4-8 (Mdn = 5.52), and are not strong normative associates;
• Level 2 pairs are strong collocations, having MI scores of more than 6 (Mdn = 7.65) and t-scores of more than 7.5 (Mdn = 10.95), and are not strong
normative associates;
• Level 3 pairs are strong collocations, having MI scores of more than 5.5 (Mdn = 7.01) and t-scores of more than 6 (Mdn = 10.63), and are strong normative associates.
Frequency measures were again derived from BNC frequency data. To determine whether pairs were strong normative associates or not, two methods were used. First, as in Study One, the EAT was consulted. Pairs at levels 1 and 2 were deemed not to be strong associates only if neither the target word nor any word form derivationally or inflectionally related to the target was listed as an associate of the prime in the Thesaurus; similarly, neither the priming word nor any word form derivationally or inflectionally related to the prime was listed as an associate of the target. Pairs at level 3 were judged to be strong associates only if the target word was listed as either the first or second strongest associate of the prime in the Thesaurus and had a minimum association score of 10% (i.e. was supplied by 10 out of 100 respondents).
It was noted above that, because the EAT was elicited from a different population from that taking part in the present study (university students between 1968 and 1971), word pairs which are not attested as associates in the EAT may nevertheless be
strongly associated for our participants (university students in 2007). Similarly, some pairs which are prominent on the EAT may not be strong associates for these
participants. Moreover, because the EAT elicited only a single response from each participant, it is possible that some highly salient collocations do not feature on its listings. To ensure that the putatively ‘non-associated’ pairs in the current study were indeed not strong associates for our participants, and that the putatively associated pairs were, a second test for association was used in the present study. Two groups of 22 subjects (who did not take part in the main study) were each presented with 40 stimulus words (a different stimulus list for each group, giving a total of 80 stimuli) and asked to write down the first three words which came to mind on reading each stimulus. The subjects in these groups were taken from the same pool as those
participating in the main priming study (i.e. 2nd year undergraduate native speakers of British English enrolled in a modern English language course at the University of Nottingham), and so should provide a good indication of the likely associates of the main study participants. Moreover, by eliciting three associates for each stimulus, we may move a little beyond the very strongest associates. Pairs at Levels 1 and 2 were deemed not to be strong associates only if neither the target word nor any word forms derivationally-related to the target was supplied as an associate of the prime. Pairs at Level 3 were judged to be strong associates only if the target was supplied as an
associate of the prime by at least two respondents (the median association score was 43% - i.e. the association was given by 9.5 out of 22 subjects).
The 48 target nouns from the three collocation lists were also matched with 48 control primes. The control prime-target pairs were intended to be semantically plausible combinations which did not co-occur in the corpus with sufficient frequency to be considered collocations. These pairs occurred directly adjacently to each other in the between BNC two and four times, with MI scores of less than 2.5 and t-scores of less than 1.5. All pairs co-occurred within a +/- 4-word span of each other fewer than 10 times in the 100 million word corpus. Though they are not sufficiently frequent to count as collocations, therefore, the fact that they were all attested more than once in the corpus suggests that they are not likely to be semantically anomalous.
No very common or very rare words were used as targets or as primes: all words used occurred in the BNC between 3,000 and 30,000 times; placing them well outside the top 300 word forms in the corpus and well within the top 3,500 (Leech, Rayson, & Wilson, 2001). All words were one or two syllables (four to seven letters) in length.
The collocating and control primes were combined into two counterbalanced lists – referred to below as Set 1 and Set 2 - such that eight collocating pairs from each level were included in each list and targets which were matched with their collocating prime in one list were matched with their control prime in the other. No prime or target word was used more than once in either list. A single set of 48 prime-non-word pairs was also added to both lists. Non-words were items of four to seven letters, generated using the ARC Nonword Database (Rastle et al., 2002). Primes were items which appeared in the BNC between 3,000 and 30,000 times and which were attested to be used as pre-modifiers but which had not been used elsewhere in the experiment. The final materials are shown in Appendix Aii.
Participants
40 undergraduate students at the University of Nottingham participating in a course in modern English language. All were native speakers of British English.
Procedure
Participants were randomly assigned to one of two groups. The first group was tested on Set 1 only, the second group on Set 2 only. Participants were tested individually in a quiet room. Presentation of the stimuli and recording of the reaction times were controlled by Psychology Software Tools’ E-Prime software and items were displayed on a CRT monitor. On each trial, a fixation point (‘+’) was presented, centred on the screen, for 1,500ms. This was replaced with a priming word, which was presented in lowercase letters for 600ms. The prime was then immediately replaced by the target, in uppercase letters. The target stayed on the screen until the participant made a response. Following the response, the screen went blank for 1,000ms before the onset of the next trial. Participants were instructed to press the right button on a button-box if the string was a word and the left button if it was not. They were told to make this decision as quickly as possible. Reaction times were measured from target onset to response. Participants received 10 practice trials, there was then a break, at which point participants were invited to begin the trial proper in their own time. They were then presented with the appropriate list of 96 trial items, presented in random order.
Results and discussion
Reaction times of less that 250ms and greater and 1250ms (1.33% of the total) were excluded from analysis. Mean accuracy by participant, collapsed across conditions was 97%. There was no effect of condition upon accuracy. Average reaction times for collocations and non-collocations at each of the three levels are shown in Table 9. Reaction times were normally distributed within conditions for Levels 1 and 2 but not for Level 3. Paired samples t-tests (for Levels 1 and 2) and a Wilcoxon signed rank test (for Level 3) revealed no significant differences at the p < .05 level between reaction times for the collocating and non-collocating conditions at any level.
Table 9: average reactions times in each condition
collocations non-collocations Level 1 Mean RT (ms) 514 519
Level 2 Mean RT (ms) 516 530
Level 3 Median RT (ms) 479 490
Study One appeared to show that target words are recognised more quickly when they follow collocating primes than when they follow non-collocating primes. However,
that study compared collocating prime-target pairs only with semantically-
incongruous pairs (e.g. death-penalty vs. east-penalty). This left open the possibility that the advantage seen was due not to the relatively higher frequency of the
collocating pairs but to the fact that they were semantically congruous, whereas the control items were, in general, not. The present results suggest that this may indeed have been the case. High frequency collocations – regardless of whether they are likely to be strong associates - appear not to demonstrate any statistically robust priming effect in comparison to low frequency but semantically plausible word pairs.
Some concerns might be raised by our failure to find any priming between pairs which are likely to be strong psychological associates (i.e. Level 3 pairs). As we have seen, priming between such pairs has been extensively documented. However, this finding appears to have been based entirely on comparisons of the sort seen in Study One – i.e. comparison between associates and semantically incongruous word pairs. All of the associative priming studies with which I am familiar and which report their method of creating ‘unrelated pairs’ have followed the lead of Meyer and Schvanevendlt’s
original paper (1971) by interchanging words from the associated items such that there are “no obvious associations within the resulting pairs” (1971, p. 228). These studies tend not to list the items used in the control condition, but Meyer and Schvaneveldt provide the illustrative examples of BREAD-BUTTER and DOCTOR-NURSE being re-paired to BREAD-DOCTOR and NURSE-BUTTER. As in Study One, then, what the associative priming literature demonstrates is an advantage of associated pairs over pairs which are highly incongruous. The findings in this literature are therefore entirely consistent with the results reported here.
We can conclude, then, that though it is possible to observe collocational priming in experimental set-ups which use semantically incongruous control items (such as that in Study One), the effect is not replicated when more plausible controls are used. This applies even to very high frequency collocations which are not normative associates. Since much of the interest of collocation is in the distinction between collocating word pairs and pairs which are plausible but uncommon, these considerations suggest that the traditional priming paradigm may not be a helpful way to study collocations in the mind.