The Noisy Hypernym
A word is a commonality at the intersection of multiple meanings.
A hypernym can be thought of as a superset of any given word. For instance, ‘animal’ would be a hypernym of cat, dog, walrus, and of course, dung beetle.
Examples of hypernyms from wordnet:
bay: {‘hypernyms’: [‘body_of_water’, ‘water’, ‘cry’, ‘laurel’, ‘compartment’, ‘recess’, ‘niche’, ‘horse’, ‘Equus_caballus’, ‘talk’, ‘speak’, ‘utter’, ‘mouth’, ‘verbalize’, ‘verbalise’, ‘bark’]}
What is horse doing in there? Well, it’s a kind of horse.
What are all those talk, speak, utter words doing in there, along with bark? Well, “I had rather be a dog and bay the moon, than such a Roman” said Shakespeare. It means to howl at the moon.
What is laurel doing in there? I have no idea. Is ‘cry’ a superset of bay? Maybe if you consider bay to mean a kind of sad moan, then cry is a superset of it, and a sad moan, taken to the extreme, becomes a cry.
So ‘in this sense’…
Let’s look at another one.
pawns: {‘hypernyms’: [‘pledge’, ‘assistant’, ‘helper’, ‘help’, ‘supporter’, ‘chessman’, ‘chess_piece’, ‘borrowing’, ‘consign’, ‘charge’]}
Pledge? Maybe in the sense of ‘I’m pawning my watch,’ but it’s a stretch. I assume that ‘borrowing’, ‘consign’, and ‘charge’ also have to do with a pawnshop. Assistant/help/supporter? Well I guess, in the sense that a pawn on a chessboard is a helper. Chessman and chess piece make sense.
You’re unlikely to see a hippopotamus described as a hypernym of a mouse, because it isn’t. But along the size dimension (in this sense) hypernym is to hyponym as hippopotamus is to mouse. The former in each pair are ‘larger,’ which is similar to ‘more inclusive,’ and the latter are ‘smaller’ (less inclusive). So by this blend of dimensional interweaving, we glide seamlessly over a meadow of commonalities and similarities. And the structure of language reflects the hierarchical structure of the world which we inhabit.
But everything is noisy. Everything, including each word and its close cousins, like blades of grass in the wind. The only question is, along what dimension, and at what distance. And true artificial general intelligence will require a way to build in this generality, to know what is signal and what is noise.