LanguidSlog 3 outlined the items needed in a sentence structure. It was rather long but not too difficult to follow, I hope. This one is much shorter but nonetheless may require some thought. It looks more closely at how junctions are formed in real time, concluding that the story is a bit implausible.
LS3 showed the structure of John kissed Lucy being formed thus:
Remember, the word-items are part of the hearer’s permanent language knowledge. It is the junction-items – c, e and f – that must be created to form a structure for the sentence. The issue is about how relevant parts of permanent words get into transient junctions.
The way the diagram is drawn implies that ‘semantic’ material is copied from permanent to transient items, and then perhaps from one transient to another.
Exactly how the meaning of a word is held in the mental architecture is not known. However it must be distributed – enabling parts to be reused for other purposes – rather than localised. Otherwise the subtlety with which humans can conceptualise could only be achieved in an absurdly inefficient way.
If that’s not clear, consider a taxonomy – say, ANIMAL / MAMMAL / DOG / LABRADOODLE. Each of these concepts has all the properties of the one to its left, plus some more specific properties. If, for example, the LABRADOODLE concept were freestanding and actually included all the properties of DOG, those properties would need to have been copied from DOG – which would itself contain all the properties of MAMMAL, ditto ANIMAL. And a further copy would be needed for each and every other type of pooch.
It’s safe to assume that ‘specific properties’ of a concept at one level are somehow picked up by all subordinate concepts without copying. That’s why ‘meaning…must be distributed’.
But ‘distributed’ means that copying from one item in the structure to another would be absurdly inefficient.
In contrast, with computer architecture, a software designer could simply load a transient item with pointers leading back to the appropriate permanent items and avoid any need to copy.
For LS3, apparently reasonable assumptions were made about how structure must be held for participation in sentence processing. However the questionable efficiency of that processing suggests there is something wrong. Can that be sorted when we think about mental architecture in the next piece?