Lecture 17: decentralized models of language (part 2)

what’s a decentralized model?

a model that does not have a centralized representation of one thing (ex: dog representation isn’t a single representation, it’s distributed across al your experiences with dogs)

current models propose that structure is learned by abstraction at [learning/retrieval]

• learning

• there are some proposal that it can be constructed with abstraction at the retrieval

define “relativist theory of language”

language rises from social communication patterns → culture defines language

define “relativity theory of language”

langage evolved through common traits, it’s something you are born with

true or false: language is learnable, the differences are due to culture

true

what are the types of memory traces encoded through BEAGLE coding scheme? (2)

• order: sequence of words that was processed

• context: context in which the word was heard

what is context used for?

to cue order: you give words to a model and it needs to put the words in a correct order so that it respects the syntax

true or false: BEALGE uses circular convolution

true: it will form a unique third variable that will represent the association of two things

how does retrieval work?

• you have a probe and it activates traces in memory

• a context vector is used as a cue

• similarity is taken between context probe and all the context vectors stored in memory

how does production work?

• take the words you want to produce and test each possible combination of this word

• then compare to the echo content vector and produce what’s most similar to the echo content

how do exemplar models organize behaviour?

they use similarity structure of stimuli

what actually contributes to language learning?

the redundancy in language (people say the same thing but differently): it shows you how you should communicate