Yeah, I think the first chair we observe is the template for recognizing other chairs later. Surely that 'seed' chair will undergo substantial revision in time

Yeah, I think the first chair we observe is the template for recognizing other chairs later. Surely that 'seed' chair will undergo substantial revision in time as we become aware that it wasn't the ideal basis on which to base our modeling of all chairs to come, and we unconsciously 'normalize' the model for chair in our head (and adjust that memory too). It stands to reason that this happens to all of the baseline models for the referents we learn (nouns, verbs, adjectives, etc) that subsequent learning surely reshapes. This is likely comparable to the revision of 'concept' clusters that goes on in all unsupervised learning NN models, especially in humans, those that arise pre-linguistically (since words surely provide landmarks to memories enabling future retrieval). Perhaps we rewrite our early word2vec tables, making early memories less retrievable.