 |
| Figure 1 |
A few days ago, in the course of taking a look at reference 1, a paper about how some apparently similar things seem to be more memorable than others, and following up various leads, I came to want to know what was meant by something called ‘RTT’, apparently a well known psychologists’ abbreviation for ‘reinforcement, test and test again’. Bing turned up reference 2, all about a serious academic spat about reference 3, the work of a then young psychologist called Irvin Rock. More sixty years ago now. And in the course of digging up the paper at reference 3, by way of a bonus, Google also turned up the downloadable image of the typescript of a MA thesis from the University of Florida, reference 4, a thesis which takes Rock’s paper, then just a year old, as its starting point – with the author of this thesis still being alive at well, if more or less retired, and attached to Rutgers University, New Jersey. People who have put a lot of photographic work into the front page of their web site at reference 5. A place founded when an important chunk of what is now the US was still part of the UK.
To set the scene, the authors of reference 2 start with: ‘In 1957, Ghana obtained its independence, the Soviet Union launched Sputnik 2, the Wham-0 Company produced the first Frisbee, and the Ford Motor Company produced the Edsel (on what the company proclaimed was E-Day a take-off on D-Day 13 years earlier). The Edsel was going to make a splash, the Ford marketing people thought, and it did—just not the kind desired’.
So I now know about the Ford Edsel, named for the son of the original Henry Ford, well known in the world of product management for being a splendid example of how a very expensive product can go very badly wrong, with Ford only managing to sell getting on for 120,000 of them, nowhere near enough. A salutary tale taught ever since to students of business studies up and down the US.
In the same year, the then young and more or less unknown psychologist called Irvin Rock published the eight page paper at reference 3. By today’s standards, very readable. This despite the complete absence of diagrams, graphics, mathematical formulae or pictures, a paltry three references and only a very thin scatter of standard deviations.
A paper which suggested that memory might come in chunks, chunks which were either there not there, at a time when the prevailing wisdom was that these chunks came with a weight, a positive real number which told one how strong the memory was. Noting here that arriving in memory in chunks is one thing, getting at those chunks, retrieving those chunks back from memory, is quite another.
The great and the good of the world of psychology gathered for the kill of this presumptuous young upstart, who had not even the good grace to round out his paper with copious references to the work of his elders and betters.
The elements
 |
| Figure 2 |
The experiments which follow mostly use small sets of numbers and letters, things like ‘24’, ‘9’, ‘K’, ‘HH’ and ‘NYG’.
One might also use animals, flowers, famous faces, famous buildings, children’s toys, items of furniture and kitchen utensils. Perhaps the sort of thing you get in children’s alphabet toys, like that snapped above. Getting more complicated, one might have the names for such things rather than pictures of them. Only an option for subjects who can read – and preferably talk.
The members of any one set should be all the same kind of thing, equally familiar and recognisable. Mixed sets would be much harder, inter alia, making the learning experiments much longer, which would be likely to be a problem.
For present purposes, it was at first thought that it did not really matter which set was used, beyond the elements being readily recognised and easily named out loud. Recognition and naming was not what was being tested – which might disqualify sets like famous buildings, as you could not rely on all your subjects, university students though they may be, being familiar with such things.
Another problem is subjects having special subjects. So a student of botany might be an expert on flowers, which might well disturb the results of a test involving a set of flowers. Rock thought to avoid such problems by sticking with the numbers and letters introduced at the top of this section.
Two sorts of experiments
 |
| Figure 3 |
Two sorts of experiments are of present interest, free recall tests and associative memory tests.
In the first sort, the subjects are given a list of somethings to remember, perhaps as a list, perhaps one after the other, on cards or on a computer screen, a reasonably homogenous list of the sort snapped above left. Then, shortly after the list has been presented or withdrawn, they are asked to report as many members of the list as they can. Reports which are often biased towards the beginnings and ends of lists.
In the second sort, the subjects are given a list of pairs, with the first element of each pair (often called the stimulus) usually being taken from one set and the second (often called the response) from another. The idea is that they remember the pairings, the associations. Then when the experimenter runs through the first elements, the subject tries to recall the second elements. Hopefully the subject is cued for the second element by the first element. While the experimenter is trying to find out something about how such associations get into memory, hence associative memory test.
Usually, there is no link between the first and second element; the first element gives no information about the second. Although in the case of alphabet drills with children, the first element might be a letter and the second element might be something whose name starts with that letter.
Subjects might use tricks to help them with the associations. So, Rock (as reported in the present paper), talked of subjects linking lion to stripe by going via tiger. Lion is already associated to tiger, and tiger is already associated to stripe, so many subjects find it helps to remember this particular association by adding tiger to the mix. Some of the mnemonists of references 8 and 9, people who crop up in these pages from time to time, talk of doing something of the sort. Of using an intermediate object as a stepping stone to building an association between two otherwise unrelated objects.
These experiments
In Rock’s work of 1957 and the flurry of activity which followed, it is mostly experiments of the second sort.
Each experiment consists of a number of trials. The subjects (S) are divided by the experimenter (E) into two groups: the experimental group (who do something special) and the control group (who do the regular thing, who are the baseline). The interest lies in the difference, if any, between the performance of these two groups.
The number of subjects seems to be quite small, say less than thirty taking the two groups together. The is some rudimentary balancing of the groups so that their performance on these sorts of tasks is roughly the same.
We suppose that we have a pool of stimulus-response pairs, and the idea is that subjects learn to link, to associate the first element (the stimulus) of each pair with the second (the response). In this work, the elements are either one or two digit numbers (for example, 57) or one, two or three letter groups of letters (for example, DEK). For our experiment, N of these pairs are randomly selected from the pool. N might, for example, be ten.
Protocols are laid down for things like reuse of elements and reuse of pairs.
 |
| Figure 4 |
Each trial consists of two parts. In the first part, E shows S the N pairs, one after the other. In this work the two members of each pair are mounted left and right on filing cards, say five inches wide by three inches high. Maybe the elements are written on envelopes which can be fitted over the ends of the cards, as in the figure above. And then changed around as needed. In the second part, E shows S the left hand member of each pair in turn and S is required to supply the right hand member, being given some number of seconds to do this. A right answer is scored 1, a wrong answer or no answer is scored 0. There is no feedback.
E starts out with a few familiarisation trials before starting the experiment proper, using elements which do not appear in the experiment proper. E then does trials with a new set of pairs, carrying on until S achieves a score of N, a clear round, as it were. Sometimes paraphrased here as ‘until criterion’. Repeat for all the subjects. Make special provision for subjects who do make the criterion in reasonable time.
In the control group, the same N pairs are used in every trial, perhaps just varying the order.
In the experimental group, the pairs may be varied. Or to be more exact, in the original Rock experiments, the pairs which S gets wrong are replaced with new pairs, the present point being to test whether the control group gets an advantage from repetition, which it would not if learning is indeed all or nothing. If associations are just there or not there, not built up over time.
Protocols are laid down for the speed at which all this is done and for the intervals between the various parts of the whole.
The famous paper
 |
| Figure 5 |
 |
| Figure 6 |
So the 1957 paper was all about learning associations from lists of this sort, with this kind of learning having then been a hot subject for more than fifty years, with the figure above being an important exhibit. A figure which shows how learning of some small block of data, in this case a short list of stimulus-response pairs, starts off fast and rapidly flattens off. The idea being that the subject after being shown the list by the experimenter a few times gets to remember the responses (perhaps a small number) which are associated with each of the stimuli (perhaps picture of animals). The sort of curve that, we are told, pops up all over the psychology of learning, of both humans and other animals.
But why? What models of learning and memory would generate nice smooth curves like this? On the face of it, one might have guessed that gradual learning was more likely to generate a nice smooth curve than all-or-nothing learning, and that was the majority view at the time. Maybe now we are a bit more careful, knowing that it easy enough to dream up models which fit all manner of facts and graphs.
One hot sub-subject was the question of whether anticipation learning was better than recall learning, with the weight of academic opinion shifting towards the latter. According to the graph, it also happened to be better. The former was perhaps originally preferred because it seemed closer to the business of learning sequences, pervasive in the life of complicated animals. The difference is suggested in Figure 6 above, in which the red boxes are what the subject does and the red stars signify repetition.
 |
| Figure 7 |
I note in passing, that learning series or sequences can be thought of a special case of learning associations – with the special bit being that the current element is also the cue for the next element, the stimulus for the next response.
Another hot sub-subject was question of whether the association between a stimulus and its response was either just there or not there or whether this association had a strength, perhaps a percentage or a number between zero and one. In the first case, if the association was there, the stimulus would nearly always elicit the right response. In the second case, if the association was stronger than some threshold, perhaps 66%, there would be noise in the system, but the stimulus would usually elicit the right response.
Rock had the cheek to come up with a neat experiment – worked with filing cards rather than the computer screens which would be used today – which was suggestive of on or off rather than strength, very much against the prevailing wisdom of the day, and to publish it without taking proper advice. In this experiment, there were the two groups, the experimental group and the control group. Both groups had to work at a short list of stimulus-response pairs until they got them all right. But in the experimental group, every time the subject got a response wrong, for the next trial, that pair was randomly replaced with new pair, a proceeding which one might have thought would make the task harder. Instead of strengthening the links between existing stimuli and responses, they kept having to start over. But it didn’t; there was no significant difference between the performance of the two groups. Suggestive, that in any one trial, the subjects got an association or didn’t. No building up over trials.
 |
| Figure 8 |
The snap above, taken from the reference 4 mentioned above, is intended to give the flavour of experiments of this sort in the middle of the last century. A tachistoscope is a sort of specialised slide projector.
A difficulty which Rock recognised and his attackers locked onto, was the possibility that in the experimental group the hard associations were getting weeded out, making the test as a whole easier. A possibility called in the business, a selection artifact. This despite the fact that in one of his experiments the stimuli were things like ‘A’ or ‘HH’, just 50 different ones altogether (‘I’ having been excluded) and the responses were small numbers. Not obvious that one pairing of such was going to be any easier or harder than another, and Rock, in his original paper, set this difficulty aside.
 |
| Figure 9 |
Notwithstanding, the evidence is rather mixed. Miller’s paper at reference 6, for example, claims to show that Rock’s procedure was indeed tainted by selection artifact and that his conclusion that associative memory was all or nothing failed. A paper which is even shorter than the original Rock paper and which I have attempted to summarise in the snap above. Groups I and II replicate Rock’s first experiment. Group III is pairs which preliminary work with Group A suggests are easy and Group IV is pairs which preliminary work with Group B suggests are hard. The figures bottom left look clear enough, but I am not enough of a statistician to know what they really tell us, if anything.
Hogan’s thesis at reference 4, already mentioned above, does something of the same sort.
Roediger and Arnold rehearse the whole rather unedifying story of the reception of this paper, which the academic establishment seemed unwilling or unable to take on its merits. They just saw it as an attack to be fought off.
And all in all, an entertaining read; a civilised account of a rather unpleasant academic spat.
From which my first takeaway is that the academic life might be life in an ivory tower, but there are still plenty of stresses and strains. Not that most of this is not needed; the competitive rough and tumble is the best way known for teasing out the truth – but it is not for the faint hearted. A far cry from the quiet, secluded and fairly idle life which I attribute to Oxbridge dons of the middle of the nineteenth century. All tea on the lawn in the summer, toast and crumpets by the fire in the winter.
My second, is that the business of making inferences from experiments of this sort about processing going on in the brain is not for the faint hearted either. Experiments which look straightforward enough can easily gloss over all kinds of problems. Not test what you think you are testing at all. Notwithstanding, while such an experiment may not help much with developing a theory, when you do have a theory, it should explain the results of the experiment: the experiment do provide a useful test of the theory.
And along the way I also get to know what RTT is all about, which is where I started out. It seems that if you mug up for something like one of Rock’s associative memory tests and are then tested twice, the second time more or less straight after the first, with no extra revision between, you tend to do much worse than the first time. A result which was deemed a victory for the all or nothing camp.
Subsequently, in the ensuing bruhaha, the RTT paradigm morphed into the STT paradigm, with ‘S’ for study. Work on this paradigm has suggested that, if you replace the associative memory test by a free recall test, that is to say rather than being cued by the first element of a pair, the subjects just try to recall of members of a previously presented list, subjects do better second time around than first time around, with new remembering more than outweighing more forgetting. In some way, the testing had prompted remembering. An interaction between the business of storing memories and retrieving memories, an interaction which seems to be much in people’s minds just presently, fifty years later. Memory is a now dynamic system, shaped by use as well as by formation. With this particular phenomenon being named ‘hypermnesia’ and making its way to Wikipedia at reference 7.
Eventually the whole business died down and psychology moved onto to pastures new, without ever having properly sorted this one out.
Rock went on to become well known for his work on visual perception.
PS: for some reason a lot of this work on associative memory is hidden away behind the paywall operated by the American Psychological Society at reference 10. Unusual these days, with so much stuff being more or less open access. Perhaps the authors of this old material are fairly old themselves, and so no longer so anxious to promote their work that they push free copies out onto Internet.
LWS-N digression
The LWS-N frames of consciousness (of reference 11) involved in these experiments look fairly straightforward, with frames corresponding to the steps outlined at Figure 6 above.
The focus of these frames will be a list of associated pairs, an associated pair or a cue with a space for a response. If the experiments are being carried out on a computer screen, there will probably be some instructional text and there will probably be other frames organising the experiment as a whole. There will also be some peripheral stuff, depending on how much attention the subject is giving the experiment.
It is not clear how much of the modest amount of information involved here needs to be in these frames for the subject to be able to work through the experiment. If the subject is poked, is asked to say something about what he is looking at, he will be able to say. But that is not the same as saying that that information was already in consciousness.
Perhaps one can look at a number, take it in and be able to play it back if asked – but without actually articulating it, being actively conscious of it at all? While in the case of a list, one certainly can’t vocalise, even sub-vocalise, more than one number at once, if that is what it takes for that number to be truly conscious.
Despite the apparent simplicity, clearly something which needs more thought.
References
Reference 1: Memorability of words in arbitrary verbal associations modulates memory retrieval in the anterior temporal lobe – Weizhen Xie, Wilma A. Bainbridge, Sara K. Inati, Chris I. Baker, Kareem A. Zaghloul – 2020.
Reference 2: The One-Trial Learning Controversy and Its Aftermath: Remembering Rock (1957) – Henry L. Roediger III and Kathleen M. Arnold – 2012. Inter alia, my source for Figure 5.
Reference 3: The role of repetition in associative learning – Irvin Rock – 1957.
Reference 4: The Role of Repetition in Associative Learning – Michael Wogan – 1958. Lifted from the webs site of the George A Smathers Libraries of the University of Florida.
Reference 5: https://www.rutgers.edu/.
Reference 6: Supplementary report: A selection artifact in Rock’s study of the role of repetition – Williams JR – 1961.
Reference 7: https://en.wikipedia.org/wiki/Spontaneous_recovery.
Reference 8: https://en.wikipedia.org/wiki/Mnemonist.
Reference 9: https://en.wikipedia.org/wiki/Jonas_von_Essen.
Reference 10: https://content.apa.org/home.
Reference 11: http://psmv3.blogspot.co.uk/2018/01/an-introduction-to-lws-n.html. A little out of date, but hopefully it will serve here.
No comments:
Post a Comment