PAS is a memory store. Its origin is a comment made to me by Conrad [he was always known just by his surname] over coffee at the APU [Applied Psychology Unit] in about 1964. Conrad was running experiments on how well we remember telephone numbers, and looked at the kind of errors that subjects made when numbers were presented visually and auditorily.
“You know, John, it’s funny, but I can always tell from the error patterns in an experiment whether the stimuli were presented visually or auditorily.” And he sketched them for me (figure 1).
The difference was that with auditory presentation there were scarcely any errors in the final position. With visual presentation, there were often 50% errors at the end of the list. So, where did the extra information come from? In the Logogen Model as it then was, categorisation took place in the logogen system [this being the place where context could supplement inadequacies in the stimulus in word perception – see the section on logogens]. Any differences between modalities had to be prior to that point in the information flow. My hypothesis, then, had to be that there was some source of information in the auditory system that could be accessed during recall to supplement the information in the later parts of the system. Since the advantage in recall for auditory over visual stimuli was only on the final one or two items, the capacity of that memory source was likely to be limited. If that was the case, then it should be possible to delete that information with another auditory input immediately following the stimulus. A crucial experiment followed. I showed that a final, irrelevant, auditory item at the end of the list, had a selective effect on recall of the last half of the list, just as I had predicted. Basically, the final item, now commonly known as a suffix, took away all the advantage of the auditory over the visual list.
Conrad had shown that a response prefix had a general effect on recall [his reason for doing the experiment was to look at the effect of the 0 which was about to be introduced as the first digit for all telephone calls outside the local area – a long time ago. The intuitive answer was that since you would always put the zero in whenever there was a non-local call, it would not have any effect on recall accuracy. Intuition, and the civil servants, were wrong.] The stimulus suffix, on the other hand, had this very specific effect at the end of the lists. I ran this experiment in 1967 and had a sabbatical that year. I was going to spend the year with Wendell Garner who I’d made friends with while he was at the APU that year. Wendell was at Johns Hopkins University in Baltimore, and that was where I was going to go. However, while he was in Cambridge, he got an offer from Yale that he could not refuse, and I joined him in New Haven instead. As luck would have it, Bob Crowder had just moved to Yale and was just about to have a paper published on the stimulus suffix. It was quite natural for us to get together. Bob came from a different tradition and did not believe my account of the suffix at all. “Why”, he said, “if that was the case, then an auditory suffix would not have a selective effect at all with a visually presented list.” “Exactly”, I replied. So we carried out the experiment. It was rather Heath Robinson since we used a tape recorder to control the experiment. On one channel there were pulses that advanced a film projector for the visually presented digits, and on the other channel we had the acoustic suffix. This meant that we had a tape loop about 10 feet long which had to be supported on wheels. Because of the make of tape recorder, the tape had to run horizontally. Quite a problem to keep the experiment going. Think how simple the experiment would have been to run with the smallest computer or one of our smart mobile phones or our mobile phones! Anyway, the data were absolutely clear, there was not a whisker of serial position effect in the data, as can be seen in Figure 2
We also established that the suffix effect worked with running memory span,[where the subject does not know when the stimulus list is going to end], and depended on presentation rate: the faster the digits were presented, the greater the effect. These results helped to establish the generality of the suffix effect.
On the basis of this, we put forward quite a profound theoretical proposition which included:
“… linguistic materials go through a somewhat stereotyped progression of perceptual stages (Posner, 1967). ‘What is learned’ by the S in an experiment may then be said to depend on how far into such perceptual processing the materials have passed at the termination of stimulus presentation.”
We saw PAS as a “primitive” level but felt “It seems necessary to distinguish at last two or three further levels of analysis which become involved when human Ss are given alphanumeric materials.” The inclusion of articulatory and semantic processing led to
“… a progressive and stratified account of linguistic perception, and hence of verbal learning.” P.372.
The paper had the usual problems getting published, being turned down by Psychological Review with a classic verbal learner’s review – they really hated things that were not stimulus or response. Once published, it took off and became a Citation Classic very quickly. Lots of experiments followed. My own showed that the suffix effect was attenuated when it was on a different channel, so with a monaural stimulus, the effect of a binaural suffix [which involves the same amount of stimulus in both ears] was the same as the effect of a contralateral stimulus. So it was not a question of what went into the ears. I also used the suffix effect to explore the nature of “speech-like.” We had found, by accident, that an artificial suffix did not always have an effect. The ones which had no effect sounded a bit artificial. So we varied the properties of the suffix, nearly always a single, simple vowel sound “ah”, to see how much variation there had to be before the effect came back. Our conclusion was that “speech-like” was part of a channel specification, just like location.
Other people found selective serial position effects with other kinds of stimuli and claimed that they were upsetting our position. But the effects were never as strong as the classic PAS effects – the non-suffix recency effects were never as big. Campbell and Dodds [date] showed that there was overlap between lip-read and acoustic materials. They thought that such data constituted a serious blow to PAS. At the worst, however, my position was that the effect came in the speech analysis channel, after the point of intersection of lip reading and acoustic information. As such, this data refined and strengthened the general position Crowder and I had taken. Very recently, Clive Frankish ran some experiments which after which his conclusion was that PAS was indeed precategorical, acoustic and storage (Frankish, 2008). Which is very nice to know. His experiment was very neat. He carried out the standard serial recall task using simultaneous visual and auditory presentation. The digits he presented were degraded in various ways so that if you just listened to them without the visual cues, they were not fully intelligible. However, if you present such stimuli together with the visual equivalent, the effect is that we hear the sound clearly, thanks to central integration of the information. The memory for terminal list items reflected the intelligibility of the speech signal rather than listeners’ perceptual experience. This fixes the effect early in the speech system where the noise is still affecting the neural signal. This interaction between modalities, with a simultaneous visual presentation affecting the quality auditory perception, is completely in line with logogen theory, and was the basis of a presentation I once made to the British Association of Science which led to a headline in the Times newspaper “Professor uses the Beatles to cure heart disease!” Don’t ask!