Teaching points

At reference 1, I was reading about image classification, getting the computer to say what an image was of. Elephant or octopus? Before that, at reference 2, I was reading about prediction in the context of language. What word is going to come next? What comes after ‘the cat sat on the’? While, more recently, I have been reading about sound clip classification, saying from what piece of music or from what performance it came. More in connection with the reference 1 problem, although there may be links to the reference 2 problem in that one can try to predict what will come next in music, as in language. From all of which the following elementary thoughts emerged. Teaching points in the jargon of the teachers’ trade when I was small.

In the beginning, an infant has neither song nor speech, but does have some sound capability. Infants can scream, moan and gurgle.

Luckily, the infant is also equipped with a trainable neural network, just like those used in the work described at references 1 and 2. This network can be trained both to produce and to hear sound – where by this last we mean to put some structure on the sound, to turn the sound from an incomprehensible photograph (as it were) to a comprehensible, information bearing diagram.

Which means, inter alia, that one important difference between sight and sound is that in the case of the latter we can, up to a point anyway, reproduce in speech the sounds we hear, certainly the sounds of our mother tongue, in a way that most of us cannot reproduce the sights we see. And, very importantly, we can hear and so, eventually, correct what we say.

Turning to the diagram above, we distinguish words, sounds – very roughly the individual letters and syllables of speech – and snatches – by which we mean small bits of tune or rhythm. The beginnings of song. Red for the ones in the outside world, blue for the ones stored away in memory. Starting in the middle, with sounds, and working out.

At the same time that a baby is learning to hear things, it is also learning to say things, to sing things. It is also learning to make other kinds of noises by banging things together or by throwing them about.

And babies seem to like repetition, so they learn to repeat sounds, both the sounds that someone or something else has made and the sounds that they have made for themselves. Part of this is discovering about control, that they can repeat something because they want to. So they want to do it again, at a time of their choosing. The beginning, perhaps, of consciousness. They are also training their motor apparatus to produce the complicated sequences of motor commands needed to do either thing and writing those sequences to memory. Put another way, all the necessary feedback mechanisms are in place and the neural networks in their brains are getting trained.

They also like exploration and variation. So having learned one sound they may well move onto another, related sound.

Gradually they learn about words and the fact the some words are the names of things which they can see. Gradually building the all-important three-way linkage between a word heard, a word said and the look of the thing referenced by the word; the look of the thing for real, on television or on paper. To know that a lion seen from the left is the same as the one seen from the right, from above, from below, from in front or from behind. And in some ways they end up being able to do this all better than the computer, in others they end up being worse. But they would hardly do it all without the word to help them along. So St. John definitely had a point at reference 3.

A point, but we must not get carried away. Dogs have likes and dislikes and horses know when they have been turned for home. Their brains must have ways of representing these things too, ways which can manage some of the things that we can do with words, with part of the reason that they do not have words being that they have the wrong sort of mouths – and I wonder now, as I type, whether this was something that was known at the time my father, who took a lively interest in such things, did his time as an oral surgeon during the second world war.

And on the other track, gradually, the computers will win. They will end up doing all this sort of stuff better than we can, at least in a narrow, geeky sense of the word ‘better’.

PS: Bing assures us that the word ‘apparatus’ has no plural. It is not that sort of word.

References

Reference 1: http://psmv3.blogspot.co.uk/2017/11/more-google.html.

Reference 2: http://psmv3.blogspot.co.uk/2017/11/reading-brain.html.

Reference 3: http://ebible.org/kjv/John.htm. The usual small prize for readers who uncover the significance of red.