Adrian Buck > Recent Status Updates

" [Aligning America with the independence movement] was less persuasive to others for whom self-government by Asians was not something to base a policy on and insignificant in comparision to the security of Europe."

Test

Test

"Biases affecting how often and when different learning strategies are employed or withheld can radically change what can be learned and how easy or difficult it is to learn it" - he presumably means biolological rather than intentional learning

"I have argued that the underlying source of each of the apparent paradoxes of language learning is a misleading assumption that learning is a one-dimensional process, in which a collection of individual memories is built up bit by bit, like adding items to a list, and in which general rules can only be derived by inductive generalizations.." - explains my struggle with Hungarian, but not multi-dimension learning

"I have argued that the underlying source of each of the apparent paradoxes of language learning is a misleading assumption that learning is a one-dimensional process, in which a collection of individual memories is built up bit by bit, like adding items to a list, and in which general rules can only be derived by inductive generalizations..." - explains my struggle with Hungarian, but not multi-dimension learning.

"I have argued that the underlying source of each of the apparent paradoxes of language learning is a misleading assumption that learning is a one-dimensional process, in which a collection of individual memories is built up bit by bit, like adding items to a list, and in which general rules can only be derived by inductive generalizations..." - explains my struggle with Hungarian, but not multi-dimension learning.

"I have argued that the underlying source of each of the apparent paradoxes of language learning is a misleading assumption that learning is a one-dimensional process, in which a collection of individual memories is built up bit by bit, like adding items to a list, and in which general rules can only be derived by inductive generalizations..." - explains my struggle with Hungarian, but not multi-dimension learning.

"I have argued that the underlying source of each of the apparent paradoxes of language learning is a misleading assumption that learning is a one-dimensional process, in which a collection of individual memories is built up bit by bit, like adding items to a list, and in which general rules can only be derived by inductive generalizations..." - explains my struggle with Hungarian, but not multi-dimension learning.

"Creolization occurs when are exposed to a pidgin as their first and only language and so reflects the ways that these young minds reinterpret a partial symbolic system as though it were complete" - given that language evolves, in what sense is it complete?

Feral children - "Their petsistent language limitations attest not to the turning off of a special language instinct but to the waning of a nonspecific language learning bias"

"Languages may thus be more difficult to learn later in life only because they evolved to be easier to learn when immature. The critical period for language learning may not be critical or time limited at all, but [an] incidental feature of maturation..." - what then of a second critical period that happens after first language acquisition between 12-16?

"How do symbolic systems evolve structures that are both capable of being learned and yet capable of being highly complex?"

"The initial stages of the symbolic shift in mnemonic strategies almost certainly would be more counterintuitive for a quick learner, who learns the details easily, than for a somewhat impaired learner, who gets the big picture but seems to lose track of the details" - these are not my definitons of quick, or impaired learning!

"Only by shifting attention away from the details of word-object relationships is one likely to notice the existence of superordinate patterns of combinatorial relationships between symbols..."

"Precisely because of children's learning constraints, the relevant large-scale logic of language "pops out" of a background of other details too variable for them to follow, and paradoxically gives them a biased head start. Children cannot tell the trees apart at first, but they can see the forest..."

"...it suggests that the structure of language and the way that it must be learned are linked. What may be essentially inaccessible under quite general learning conditions may become more accessible under more limited conditions."

"By structuring the learning process so that certain aspects of the problem could not be learned from the outset, he prevented the learning of relationships that are employed at different levels of language structure from interfering with one another" - is this similar or different to how we teach people langauges, etc.?

"Preventing neural nets from getting trapped in learning potholes can sometimes be accomplished by introducing "noise" into the net to disrupt convergence towards weakly predictive states, and forcing a wider "sampling" of possible solutions, or else by introducing biases against features that are common only to suboptimal solutions..."

"As soon as recursive structures were introduced (e.g. embedded clauses or structural transformations [🤔]) which had the effect of separating and inverting simple contiguity relationships between words as found in kernal sentences, no amount of training seemed to converge towards predictability." - i.e. the net couldn't learn natural language.

"If a net could be trained to [predict the next words in a sentence] as well as a person, it would indicate...(1) that the statistics of relationships between grammatical classes of words in the training strings contained sufficient structure from which to recover grammatical regularities; and (2) that these regularities were learnable in some form without explicit (rule-based) error correction" 🤔

"The training consisted in comparing predicted next "words" with actual next words and then modifying network connection strengths according to how each contributed to correct predictions." - Who compared? Who modified? And how is the contribution assessed?"

"In a neural network, the relationship that maps input to output is decomposed to minute facets of the whole that are distributed throughout the net and are literally embodied in its connectional logic. Since the input-output relationship is in effect computed by the whole network, it is called parallel distributed processing (PDP) of information."

"...the performance of a damaged neural net degrades incrementally as the extent of the damage is increased. This too is reminiscient of the way nervous systems appear to behave in response to damage, and demonstrate that the analogy is not entirely superficial." - this was written before ChatGPT

"These input-pattern to output-pattern relationships are thus mediated via the patterning of signals distributed through the web of interconnections that link output to input nodes by way of any individual node" - neural networks are as opaque to me as Chomsky's language acquisition device.

"Newport wondered instead whether children's learning limitations [memory consolidation, short working memory span] might actually be advantageous when it comes to learning language."

"The result is that layers of transformations become buried and implicit in the final product..." - so Deacon buys into this part of the Chomsky project.

"...there is no reason why Kanzi's brain should have evolved a language acquisition device.." - perhaps he means a 'human' language acquisition device, most humans struggle with languages after the critical period.