So… here’s a bit of a thought experiment. Andrew is working on older people and the making/maintenance of meaning in contemporary art venues. We got talking about ‘meaning’; this got me thinking…
‘Meaning’ is an emotional response to an object (say), although the emotion is slightly more nebulous than joy or anger, more of an intuitive sense of the personal significance of an object, place or person. If it is an emotion, then it has to work like everything else really. Neurons (bless ’em!) work through summating inputs from synapses on their various dendrites. Summation occurs both through the spatial arrangement of the dendrites, the strength of the signal coming from each synapse and the synchrony of such signals. Computer modellers, such as Rogers and McLelland, have taken a neural approach to modelling object cognition in children using constitutive modelling techniques (aka ‘neural net models’). So, I thought, why not do that for ‘meaning’?
Each ‘thing’ that you have ever encountered is represented as a node, connected to other things / nodes through links. Let’s call the things, Thi and Thj.
Now, each time that you encounter an object, that increases it’s weight. So the weight of an object is called, Wi.
Each time a thing is associated with another thing then a link is created. So the total number of links to a given node is Nli.
The each time that a link between two nodes is referenced increases the strength of that link. Let’ call that, Slij.
In terms of the constitutive model these nodes and links comprise the hidden layer, where associations are forged through repeated encounters with the outside world. Of course, a model needs inputs and outputs.
The output is some analogue measure of the emotive content associated with the input, i.e. its meaning. There could be any number of inputs so lets say that the meaning is called, Mk.
That leaves the input. Of course, the input is the any (new) object that you encounter. We’ll call that Ok.
But there’s 2 things to say about the input:
(1) The input object must be assessed for likeness (or similitude) with the existing nodes, Simki. Likeness needs to be assessed with some sort of algorithm that considers the qualities of the input object and compares it with the existing nodes. Based on the comparison it needs to create an analogue ‘rating’ of similitude and if that rating exceeds a certain threshold then the algorithm indicates that the input value ‘evokes’ (i.e. reminds the model of) a node. In principal the qualities of an object could be limitless but in reality we do not attend to the world in all it’s detail so we probably based judgements on a finite set of qualities.
(2) Once the assessment of likeness has been carried out then the new object becomes a node in the model linked in the model both with items in it’s context and with the objects that it is like.
To go back slightly, then, meaningfulness can then be represented…
Mk = f(Ok, Simki, Wi, Nli, Slij)
Does this make sense? If so, can it be implemented? If so, how?
Having posted this, I realised that I was missing out a component: when we encounter an object we do so in a context and I haven’t captured the importance of the context very well. As a start, it might be possible to take into account the emotional state that we are in when we encounter an object, I quite like the idea of the emotional content of an object, so I’ll call that, ECk. Similarly, a memory of an object will have an associated emotional component to the memory. So, in this hypothetical model, every node should have an emotional content too, ECi.
That makes the formulation:
Mk = f(Ok, ECk, Simki, Wi, ECi, Nli, Slij)