Sunday, 19 November 2023

Another instrument


A bit more than a year ago I noticed at reference 1 an optical instrument dating from the 1940s. Today it is the turn of the 1960s with something called a Model GB three-channel tachistoscope from the Scientific Prototyping Corporation of New York, an instrument widely used for vision experiments in the 1960s. A large contraption, weighing in at over 350lbs. Not the one snapped above, but very much along the same sort of lines, at least from the outside.

The core idea is to present a visual stimulus, prepared on something like a white index card, for some carefully specified time, for some number of milliseconds, perhaps 50, but ranging up to some seconds. In the present case the idea is to make the stimulus duration short enough that the subjects have trouble working out what it was.


This was all kicked off by reading about something called word superiority in Seth at reference 2. Word superiority seemingly being a well-attested phenomenon whereby people find it easier to recognise letters when they are embedded in a word than otherwise. Very much grist to the mill of whether this part of the brain works in a feed-forward, bottom-up mode or whether top-down also has a role to play, a matter which had already cropped up at reference 6. From there to reference 3 which I did not understand, possibly not helped by English not being the authors’ first language, from there to reference 4 which I did understand and from there to what Dehaene referred to as a classic experiment on the subject, reported on at reference 5, the paper of present interest. This turned out to be quite an old paper, from 1969, not quite in typescript but almost, and not coming with much in the way of helpful diagrams. On the other hand, I was not assaulted by pages of tricky statistics and tricky science in the way of reference 3 – and, as it happened, there were only six pages altogether.

The upshot was that it still took a while to get going and I thought that it might be an idea to fall back on the doctorial dissertation on which the paper was based. A dissertation which I found fast enough in something called the Deep Blue repository at the University of Michigan – but where it was not open access. Members only. Which was curious as I thought that most people who write such dissertations are only to pleased when someone want to read them, and similar repositories which I have come across in the past have been open access. Whatever the case, this one was not and the dissertation did not appear to have leaked out to anywhere else. So I pushed on with the original paper.

I did not attempt the arguments about vision processing structures which were active at the time, but concentrated on the experiment.
 

Where I rather missed the sort of flow diagram with pictures that you get in a paper today, would certainly get in a decent text book, and was reduced to Powerpoint, rather as I am sometimes reduced when whodunnits on television get too complicated for me. With the result snapped above.

The core of the experiment was a list of ‘216 four-letter words chosen such that each of the words could be changed by one letter to make up a new word. The letter which could be replaced (called the critical letter hereafter) to form a new word, as well as the letter substituted to form that new word, were the two response alternatives in the forced-choice procedure. For example, D and K were the alternatives for the word WORD, with D being the critical letter’. One could then present ‘WORD’ in a bad light – that is to say for a short duration – and ask the subject whether the terminal ‘D’ was a ‘D’ or a ‘K’. That was the type called ‘1W’ second from the right in the top row in the snap above. These responses could be compared with doing much the same thing with one letter  (1L), two letters (2L), two words (2W), one quadrigram (1Q) and, finally, two quadrigrams (2Q), where a quadigram (here anyway) is the nonsense word arrived at by rearranging the three non-critical letters in one of the real words.

As well as systematically varying the position of the critical letter in the four across and two down display used for the stimuli, this was also done for two conditions and three durations. In condition one cued the subject with the forced-choice to come for each observation, in the other one did not.


The results were summarised in the one and only figure, Figure 1, snapped above. Which with the help of the Powerpoint, now started to make sense. There were indeed a total of 36 data points. I had worked out that ‘2Q’ was not ‘2O’ and what it meant. I decided that ‘mean errors per s’ meant mean errors per subject, abbreviations of this sort counting for something in the days when scientists typed on typewriters with two fingers. I did not worry that I had not worked out what how this mean had been calculated. Nor about the p-values, also the subject of much argument these days, nor about the confidence with which the subjects made their decisions – beyond thinking that asking about this last must have made the experimental sessions last rather longer than they might otherwise have been.

Curiously, cueing clearly made things worse. But that apart, the subjects clearly did best when the letter to be identified appeared in a real four letter word (perhaps Mr. Reicher had a sense of humour) which appeared by itself, giving us the 1W lines at the bottom of both charts. The word superiority effect appeared to be real enough. What was not so clear, and what appears to remain unclear, is what the implication is for the brain’s processing arrangements. Bottom-up, top-down or both? Notwithstanding, present opinion seems to be firmly in favour of both – but I do not suppose that everyone is yet on-message.

A debate which Reicher expressed, more than half a century ago, in terms of serial processing versus parallel processing. And of letters versus chunks versus words.

In sum, an ingenious experiment which apart from corroborating the existence of word superiority has also served to highlight the difficulty of extrapolating from such an experiment to the arrangements of the visual processing pathways in the brain. And as a reminder that long lists of references at the end of papers are not just there to impress! With their value being much enhanced by the ease with which one can get hold of such stuff over the Internet these days; the need for well-stocked, specialist, bricks-and-mortar libraries is much reduced.

PS 1: along the way, trying to make more sense of reference 3 from its supplementary information, I got lost inside another data repository, this one belonging to the University of Radboud at Nijmegen in the Netherlands. I did manage to acquire access rights, but there was just too much information! See reference 7.


PS 2: while on exit, I was brought this snap by the Daily Express: 'A massive skyscraper, double the size of The Shard, stands abandoned in China after construction was halted by authorities in Beijing. In 2015, the construction of Goldin Finance 117 peaked at a height of 1,957 ft with a total of 128 floors. Building work began in 2008 during a period of frantic competition between Chinese cities to build the most impressive mega project'. What will the end game be? A hostel for migrant workers - thinking here of the chequered history of our own Centrepoint at Tottenham Court Road?

PS 3: from reference 8 it would appear the 'Goldin' is not a misprint. But also that the item is not really news at all. Which is normal for this sort of news from Microsoft.

References


Reference 2: Being You: A New Science of Consciousness – Anil Seth – 2021.

Reference 3: Word contexts enhance the neural representation of individual letters in early visual cortex – Heilbron M, Richter D, Ekman M and others – 2020.

Reference 4: Reading in the brain – Stanislas Dehaene – 2009. 

Reference 5: Perceptual recognition as a function of meaningfulness of stimulus material – Gerald M Reicher – 1969.


Reference 7: https://data.ru.nl/?1

No comments:

Post a Comment