It was this that Fisher (1976), mentioned earlier, took as a jumping off point in examining the effect of existing associations between information in a retrieval cue and an already encoded event. Fisher believed that, based on the Bartlett’s (1932) work that recall and recognition were similar processes which differed based on the individual features of the cues available- recall using a preexperimental association; recognition using a nominal copy of the target. Fisher continued by saying that it was expected that the preexperimental association from the target item to the information in the episodic event would be the feature that most strongly determines the ability to recognise the original event, with recall using the opposite relationship (the relationship from the cue to the target). Fisher tested these by having subjects learn a set of word pairs (a cue word and a target word) with distinct relationships between the cue and targets insofar that the cue was either easy or difficult to produce from the target, and the target was either easy or difficult to produce from the cue. It was the second set, in which the relationship from the target to the cue was manipulated, that Fisher believed would affect recognition ability.
The reason that fisher believed this particular relationship would effect because of the backwards association. Rabinowits et al (1977) later elaborated in this same idea, providing the explanation that failure of recognition could be due to the failure of backwards retrieval- a target of a word pair is not recognised due to the inability to retrieve the first word of the pair when the target is given as a cue (as is standard in recognition tests). This would suggest, therefore, that subjects were observing the target word in the context of the paired word and thus were using it to aid in recognition. Fisher’s ideas on this phenomenon extended further- he believed that the specific relationship of the target to the cue was responsible for the strength of the recognition performance and this could be manipulated, as outlined in his work discussed above.
There was, however, some possible elements missing from Fisher’s own work. While he did use both strong and weak target to cue relationships (as well as cue to target relationships, as he was simultaneously assessing recall) he did not factor in a control condition and thus did not test to see if there was a true effect of the cue word in the first place. This is the point at which the present experiment begins.
The purpose of the current experiment is to test the theory established by Fisher (1979), confirming the positive effects of backwards association- specifically high target to cue relatedness- in recognition ability. However, expanding on the original method, this experiment aims to test the true effect of this theory by implementing an additional between subjects condition which will have no cue words- thus acting as a control condition. Based on previous work in this area, it is safe to assume that the present experiment will confirm Fisher’s own results. More specifically, however, there are two main predictions for the outcome of this experiment.
Primarily, based on much of the work outlined above, it is predicted that recognition performance will increase significantly when there is a high degree of relatedness from the target to the cue. Conversely, there will be a poorer, possibly inhibited ability to recognise target words when the relationship from the target to the cue is low.
Secondly, when no cue word is present at all- and instead only the target word is displayed, there will be significantly worse performance than when there is a cue word present with a strong target to cue relationship. It is likely that all words will be subject to a flatness of effect, and thus all be recognised equally well at test.
Method
Participants
Data was collected from a sample of 48 participants, drawn from a broad population which met no specific criteria. The sample had a mean (range) age of 25.1 (20-53); 23 were male and 25 female. All participants had English as their first language and were capable readers- no participants suffered from reading, or other learning difficulties.
Apparatus
All materials in this instance were displayed on a 16” laptop screen, specifically an Acer “Aspire 6935”, and displaying them using Microsoft Office PowerPoint. An English 10p coin was used to randomise assignment between the cues present and the cues absent conditions.
Materials
The materials in this experiment were divided into three specific experiments, and thus designed appropriately for each.
The first experiment- the set inducing task- consisted of 20 word pairs taken from Tulving & Thomson (1973) with a cue word and a target word in each pair. The word pairs came in two forms- high association or low association pairs, varying by the strength by which the cue word would evoke the target word in a free association task. For example the target word “light” could be paired with the strong cue “dark” or the weak cue “head”. In the set inducing task word pairs were displayed side by side with the cue word in lower case, and the target word in upper case. They were positioned in the centre of the screen in bold black size 44 font, for 5 seconds each- before automatically moving onto the next word. The set inducing task also had a second slideshow used to test participants. The test was similar to the first display, but instead showed only the cue word in the centre of the screen and each word appeared on screen for 8 seconds before automatically moving onto the next slide- and in this instance, each cue word was numbered to aid subjects with correctly recording their answers. Finally for the set inducing task, an answer sheet was provided reiterating the instructions presented on screen at the beginning of the task, as well as 20 numbered blank lines on which the answers could be recorded. Presentation style, including presentation duration, was based on Fisher’s (1979) experiment, which was modified for this experiment.
The second and third experiments use identical sets of word pairs to one another, although some materials differed. Primarily they shared a new set of word pairs, reusing Fisher’s (1979) word lists- with 48 words making up four distinct sets: the “High-High”(Hi-Hi) group in which the target and the cue words both illicit a strong response for one another in free association tests; the “Low-Low” (Lo-Lo) group in which both illicit a low response for one another in free association tests, the “High-Low” (Hi-Lo) group in which the cue will illicit the target regularly, but the opposing relationship is weak; the “Low-High”(Lo-Hi) group which is the reverse of the Hi-Lo condition. These words are outlined in table 1, below.
Table 1: Main task word pairs, divided up by group.
Word pairs in the presentation phase were displayed in an identical fashion to the set inducing task, again lasting 5 seconds per word pair. An alternative will also be available presenting only the target words in the centre of the screen, relevant to the “target only” condition- but is otherwise identical. Words were presented in a way that evenly distributed their order of appearance-in this case, a Hi-Hi word first, followed by a Hi-Lo word, then finally a Lo-Lo followed by a Lo-Hi- this would then be repeated until all words are displayed once. In the presentation phase of the recall/recognition task there were also 6 extra word pairs- 3 shown at the start; 3 at the end- used to eliminate the recency and primacy effects. These 6 words were taken from the Tulving & Thomson (1973) word sets-3 of which were strong and weak pairs respectively.
There are two further, distinct, materials for the recall test. First of all, a power point presentation similar to the presentation phase of the set inducing task showing only the 48 cue words from Fisher (1979) in the same order as the respective word pairs were displayed in the presentation phase. Like in the set inducing task, the items in this phase were numbered- again, to aid participants correctly fill in the answer sheet. The answer sheet was identical to that of the set inducing task, but with the high number of trials the answer spaces were positioned in three columns. Once again, the on screen instructions from the beginning of the task were reiterated at the top of the answer sheet.
Finally, for the recognition phase of the experiment, there is once again a specific answer sheet. Clear instructions were at the top, followed by the 48 target words from the presentation period for the main task mixed together with 144 foil words- arranged in 4 columns. The foil words were hand selected based on word frequency, all of which were rated as being ranked in approximately the top 2,000 for word frequency- with word frequency data being taken from Thorndike & Lorge (1944).
Design
This experiment used a three factor mixed-design ANOVA. Each of the three factors had two levels. The first factor here is cue word presence- which encompasses two levels: Cues present, and cues not present respectively. The cue presence factor was a between subjects factor.
The second factor is the relationship of the cue words to the target words, which once again has two levels: a high level of relatedness and a low level of relatedness. This factor was a within subjects factor.
The third and final factor was the relatedness of the target words to the cue words- and has two levels matching the cue to target factor: high relatedness and low relatedness. This factor was also a within subjects factor.
In terms of assignment to conditions, every second participant was part of either the cues present or cues absent condition at random- randomised with a flip of a coin- with the following participant being assigned to the opposite condition.
The ‘target to cue’ and ‘cue to target’ conditions were counterbalanced in such a way that in both the presentation and test phases only one of each word pair type (HiHi/ HiLo/ LoHi/ LoLo) was presented once in a pattern comprised of all word types occurring in a block of four word pair trials.
To further counterbalance the trials, half of all participants in both the cues absent and cues present conditions were presented with the word pairs in reverse order. Moreover a second answer sheet for the recognition test, on which the words were in a completely different and random order, was distributed evenly among all conditions- and distributed evenly between both those seeing the original or the reverse ordered presentation.
In order to fully account for false alarms, all subjects are instructed to make exactly 48 “recognised” responses in the recognition test, ensuring that all participants have an equal opportunity to succeed- marking all words, including foils, as “recognised” would produce a 100% score, but would also be essentially worthless in terms of this experiment.
Procedure
All participants began this series of experiments by carefully reading the instructions for the first task- the set inducing task. Instructions were as identical as possible to those given by Fisher (1979), in which they were instructed to remember the target word- in upper case- in the context of the cue word. Participants were given the opportunity to ask any questions, and continued with the set inducing task as outlined above- seeing an automatic presentation of the 20 words pairs. Following this, participants were given another set of instructions, and at this point also given the appropriate answer sheet- and when they were fully ready, they began the test phase of the set inducing task- which once again began with full instructions to write target words when they see corresponding cue words, and a chance to ask questions. This was followed by an automatic show of cue words and when concluded, participants moved onto the next stage.
Participants were next, once again, given full instructions as to what they would see next- either a list of 48 cue-target word pairs, or a similar list of 48 isolated target words- generally similar to the set inducing task in terms of presentation. Following this initial presentation, half of all participants would perform the recall test- being shown the cue words in a method that replicates the test phase of the set inducing task exactly, other than length.
Participants who did not do the recall test, and those who had completed the recall test, were then immediately given the recognition task answer sheet- with instructions at the top asking them to make exactly 48 choices as to which words they had seen in the second presentation phase- in an unlimited amount of time. Completing this answer sheet, participants were given an explanation of the experiment, and were free to leave.
Results
For each subject, a total score for each item type (HiHi, HiLo, LoHi and LoLo types) was calculated separately. Means and standard deviations of scores for each item type is given in Table 1, separated into the Cues present and Cues absent conditions.
Table 1 Mean scores in each individual word type category.
Standard deviations appear in brackets.
A three way mixed design ANOVA was performed on the sum of scores of each participant by condition, with 2 within subjects factors and 1 between subjects factor.
There was not a significant three way interaction between cue presence, cue to target relationship and target to cue relationship, F(1,46)=0.24, p<0.62.
There was not a significant main effect of cue presence, F(1,46)= 0.35, p<0.553.
There was not a significant main effect of cue to target relatedness, F(1,46)=3.93, p<0.053.
There was, however, a significant main effect of target to cue relatedness, F(1,46)=16.7, p<0.001.
There was not a significant interaction between cue presence and cue to target relatedness, F(1,46)=0.03, p<0.863.
There was not a significant interaction between cue presence and target to cue relatedness, F(1,46)=0.001, p<0.999.
There was not a significant interaction between cue to target relatedness, and target to cue relatedness, F(1,46)=0.24, p<0.623.
The results for this experiment certainly show a number of distinct trends, although not significant, and thus at least a vague glance at the possible patterns of effects is possible. Primarily, the large main effect for target to cue relatedness is distinctly reflective of the hypothesis, which based predications on the work of Fisher (1979). By looking at the strength of this particular effect over the other cues present conditions it would be easy to assume that Fisher’s own work had been reiterated in these results. However, the same distinct pattern of results reflects some peculiar findings in regards to what was expected and thus outlined in the hypothesis of the current experiment- specifically, the peculiarity arises rank order of the results with the LoHi condition showing the strongest performance, followed by the HiHi condition, and then the LoLo condition- and finally the HiLo condition. While the order of these scores matches Fisher’s in the cues present condition- as expected, the same pattern is also present in the cues absent condition. This would suggest the absence of an effect for cue presence, rendering Fisher’s points about backwards association and target to cue relatedness in recognition moot.
On the other hand, it is interesting that the only significant interaction was the main effect of cue to target relatedness. While this provides interesting insight into the effects, it is unfortunately only one half of the results: this effect did not differ in tandem with cue word presence and thus means that the cue words were largely irrelevant to performance.
Discussion
The results of the current experiment reveal a number of interesting findings. Primarily, in a stark contradiction to the results of Fisher (1979), there was not a significant effect of word relationship on the outcome of recognition performance scores at test. This, unfortunately, is contrary to the hypothesised result in this experiment- though there was a significant main effect of target to cue, which was predicted by Fisher and thus does fit, at least partially, with Fisher’s model. From this it is fair to assume that, based on this set of results, that the specific method of recognition Fisher suggested- the process of backwards association from the target to the cue- was not present here, and furthermore, may not be an effect that occurs within the wider context of a persons cognitive psychology. Further to this point, it might be possible to draw the conclusion that recognition and recall are not processes which interact independentlym but interact with the same existing encoded memory trace. However, the trend of the results did match those of Fisher- with the target to cue relationship items having, overall, the highest score (represented by the HiHi and LoHi groups), and the reverse true for the low target to cue relationship items showing the opposite effect.
Secondly, and equally relevant to the hypothesis, it is very clear from the results of this experiment that there is absolutely no effect present that causes any difference between the cues present and the cues not present conditions. Much like the first point, outlined above, this is a direct contradiction of Fisher’s theory- as, without a cue word present, it would be unlikely that the word is recognised in such a way that is aided by the pre-existing contextual information (in this case, the cue word) of the word (the target word) that is available. This, of course, further highlights the fact that recognition occurs in some way that differs from what was expected- and is certainly not a phenomenon in which contextual information is relevant- which, interestingly, was suggested by several theories outlined in the introduction prior to, and after, the work of Fisher- highlighting the idea that contextual information is used for recall performance, as memories are recalled based on contextually relevant information, where as recognition has been confirmed by this experiment to be dependent on a copy cue (a cue which is identical to the target), this prompting a positive or negative recognition response as apposed to a contextual investigation of the cue.
As a whole, this experiment seems to largely disprove Fisher’s theory that recognition occurs when using a cue to aid the initial encoding process and thus aid recognition with a stronger target to cue relationship. Instead, not only is the strength of the target to cue relationship not beneficial to recognition, but the presence of the cue is entirely unnecessary.
There are a number of issues, methodologically speaking, that would be necessary to address if this experiment were to be repeated- and thus have possibly affected the reliability of the data in this experiment. The first and most striking methodological flaw is the way in which word frequency have been used. Initially clear from the information available in Fisher’s original words list- as outlined in his 1979 paper- that there may be discrepancies in the word frequencies between the groups. Using Thorndike & Lorge’s 1944 word frequency data, Fisher outlined all word groups as having “approximately equal word frequency”- and even went on to investigate the groups in a controlled condition to test for this kind of artefact- and found that there was no significant effect between the four groups. However, this minor investigation used only ten subjects- a low number which was highly likely to have negatively impacted the significance of this trial, thus leading to the false conclusion that this particular artefact was insignificant to the overall findings. In reality, the average word frequency for the groups does differ: The HiHi group has an average word frequency rating of roughly 75 per million; the LoLo group has an average frequency of roughly 84 per million; the LoHi word group has an average frequency of 61 per million; the HiLo group has an average frequency of around 93 per million. With this clear difference, it would be easy to understand why certain groups might be more easily recognised overall- thus eliminating the reliability of the results, which may vary based on the differences in the frequency as apposed to being an effect of the specific relationships between the word pairs in each group. To follow on from this point, words with differing word frequency are easier for recognition tasks. Specifically, Morton (1969) suggested that word recognition units- logogens- had a ‘resting’ level of activity as well as a threshold which determined the amount of evidence necessary to activate it. In this theory, Morton suggested that this threshold was determined by a number of things including recency of encounter, but more importantly to this particular study- general word frequency. Thus, because of the distinct differences in word frequency, the thresholds will be largely dissimilar, and would be the process which the results actually reflect. Alternatively, an established fact of psychology is that lower word frequency words would be better recognised. Words that are more distinct would be more easily recognise as they will be more obvious because of their abnormality within memory- and this kind of distinctiveness would be caused by having not encountered a word on a regular basis, thus a lower word frequency (Seymour, 1987).
Further considering Morton’s logogen model, and the impact of recency, it is worth considering this as an effect aswell. As part of the current experiment, half of all participants performed a lengthy recall test prior to the recognition task. While this was counterbalanced thoroughly it is possible that, using this method, the effect of this prolonged task prior to the test phase might be looked at- it is entirely possible that, even in the current experiment, the averages were changed significantly because of this, and this may well offer insight into the processes of recognition (Flammer & Lüthi, 1991). Moreover, by examining the independent results in both the recognition and recall tasks, one might be able to further isolate evidence as to whether or not, like Fisher described, recall and recognition are independent processes have interacting elements, or if both work entirely in isolation.
It is plausible that, during the course of this experiment, there were a number of occurrences of ‘false remembering’- believing to have recognised a word which was not there (Tulving, 1973). In the current experiment it is possible that this occurred as a result of the semantic similarity between some of the foil words, and cue words which were a part of the recognition test, and as such participants will have falsely recognised a word because, while making a backward association to the cue from the target, the recognition of the cue as well as being paired with a sensible target would mean that it would be easy, in some cases, to make a logical inference that a word was one seen previously when it had not. This phenomenon may be of interest to future research, especially those that want to take this specific task and theory further, examining how and why backwards associations occur, and the seemingly fallible systems that surround them.
Future research might consider changing various methodological aspects of this test, if that experiment were to investigate the same specific points as the current study. One overall change might be to change the word pairs in a number of ways. First of all, it would be beneficial to update the target words so that they have word frequency scores which are as close to one another as possible, unlike those used here, which has various flaws outlined above. Secondly, the probability of preexperimental eliciting a target from a cue could find advantages in using a more rigorously controlled measure of what percentage constitutes a “high” relationship and what constitutes a “low” relationship from the target to the cue or vice-versa. This, however, is only an ideal- while moving towards this goal would be a positive step, forming word pair lists which are perfect would likely be wildly impractical, in part due to the fact that it might be near impossible to reach such exacting standards . A second modification might be to update the word lists more literally- by using word frequency data that is more modern, and thus more applicable a measure of current word frequencies as they relate to the information consumed in modern society (this is especially important considering the emergence of the internet, and the fact that it is largely text based, as well as modern changes in media consumption habits as well as information presented as a result of changes in popular culture and important social figures).
Another possible continuation of this methodology might be to use it, and the associated theories, in parallel with Morton’s Logogen Model, and examine it by modifying this methodology to incorporate elements of priming in regards to presentation recency as well as word frequency. By doing this to investigate recognition, it might be possible to expand upon Fisher’s model of recognition and recall by creating individual models of recognition and recall that depict the associated processes working within them, as well as the affecting factors, as apposed to the broad summary of the processes which was made available by Fisher in his original discussion of the experiment.
References
Baddely, A. (1991). Human memory: Theory and Practice. United Kingdom: Lawrence Erlbaun Associates Ltd.
Bartlett, F. C. (1932). Remembering. Cambridge, England: Cambridge University Press
Eysenck, M. W. & Keane, M. T (2005) Cognitive Psychology: A Students Handbook (5th Edition). New York, NY: Psychology Press
Fisher, R. P. (1979). Retrieval operations in cued recall and recognition. Memory and Cognition, Vol. 7, 224-231.
Flammer, A. & Lüthi, R. (1991). The mental representations of the text. In G Denhière & J-P Rossi, Text and Text Processing. (pp. 45-59). North Holland: Elsevier Science Publishers B.V.
Gregg, V. (1976). Word frequency, recognition and recall. In J. Brown (Ed.), Recognition and recall. Chichester: John Wiley & Sons Ltd.
Groeger J. A. (1997) Memory & Remembering: Everyday Memory in Context, Essex, UK: Wesley Longman Limited
Mandler. G. (1980). Recognising: The judgment of previous occurrence. Psychological Review, Vol. 87, 252-271
Mandler, G. & Boeck, W. (1974). Retieval processes in recognition. Memory and Cognition, Vol. 2, 613-615
Mandler, G.,Pearlstone. Z,. & Koopmans, H.S (1969) Effects of organization and semantic similarity on recall and recognition. Journal of Verbal Learning and Verbal Bebaviour, Vol. 8, 410-423
Morton, J. (1969). Interaction of information in word recognition. Psychological Review, Vol 76, 165-178.
Muter, P. (1978). Recognition failiure of recallable words in semantic memory. Memory & Cognition, Vol. 6, 9-12.
Rabinowitz, J. C., Mandler, G., & Barsalou, L. W. (1977) Recognition failiure: Another case of retrieval failure. Journal of Verbal Learning and Verbal Behaviour, Vol. 16, 639-663.
Schulman, A.I. (1976). Memory for rare words previously rated for familiarity. Journal of Experimental Psychology: Human Learning and Memor, Vol. 2, 301-307
Seymour, P. H. K. (1987). Word Recognition Processes. An analysis based on format distortion effects. In J. R. Beech & A. M. Colley. Cognitive approaches to reading. (pp. 31-55) New York, USA: John Wiley & Sons Ltd.
Thorndike, E. L., & Lorge, I. (1944). The teacher’s word book of 30,000 words. New York, NY: Columbia University Press
Tulving, E. (1973) Elements of Episodic Memory, New York, NY: Oxford University Press.
Tulving, E. Thomson, D. M. (1973) Encoding specificity and retrieval processes in episodic memory. Psychological Review, Vol. 80, 352-373
Appendix
Appendix 1: Set inducing task word pairs (Source: Tulving & Thomson, 1973)
Cue Target word
hot COLD
head LIGHT
want NEED
cheese GREEN
small LARGE
sun DAY
sky BLUE
cave WET
lake WATER
noise WIND
chair TABLE
command MAN
eat FOOD
home SWEET
infant BABY
exist BEING
tobacco SMOKE
art GIRL
bake CAKE
door RED
Appendix 2: Experimental Foil words (Source: Thorndike, 1944)
Appendix 3: Primacy/Recency words (Source: Tulving & Thomson, 1973)
Cue Target word
dumb STUPID
hope HIGH
leaf TREE
tennis BALL
swift GO
cloth SHEEP