You seem to differentiate between words and concepts. And with words you mean words with a specific interpretation in a specific context. Not words everyone ‘knows’.
Take e.g. ‘correlation’ (the first candidate from your post). The laymans reading is roughly ‘some kind of relationship’. Being used in a LW context you may assume that the reader knows about the mathematical definition and its application to everyday use.
Now as long as you use ‘correlation’ as in your post and do not attach weight on the consequences of that use there is no risk.
But once you want to use consequences of that definition to support or weaken a point then you must consider what concepts your reader really knows.
As the other commenters noted the Sapir Whorf Hypothesis is mostly wrong. You can have concepts without words. But words clearly help.
The question also is: When do you really have a ‘word’ for a concept. ‘correlation’ alone is no concept. There are stochasitic correlation between stochastic variables or correlations between signals. Those that are discrete or continuous or multi dimensional. Pick your meaning. And arguing that the ‘concept’ is pure; the abstraction or generalization of these specific instances doesn’t really work because nobody really knows what the pure concept is. Thus you cannot communicate it but only appeal to it by reference to particular cases.
Stripping a word of its context (who and when and where it is uttered, onto which other words it is applied) leaves not a pure concept but only a sequence of letters.
Words give the hearer an indication of your indicated meaning but they are not the meaning. That is what EY means with
How words [...] are secretly a disguised form of Bayesian inference.
Words make a certain reading more plausible and (hopefully) let your mental representaticonverge locally toward that of the speaker (your brain will update its structures mostly bayesian, but it may fail given noise).
A large vocabulary then is a large set of ‘signals’ for which both the speaker and the heaer know the concepts signalled.
Because words are also used internally by conscious thought they can also be used to signal to yourself and thus a large vocabulary can be conducive to efficient reasoning.
Because private and public use of the vocabulary are often mixed one should reflect on what the other really knows.
You seem to differentiate between words and concepts. And with words you mean words with a specific interpretation in a specific context. Not words everyone ‘knows’.
Take e.g. ‘correlation’ (the first candidate from your post). The laymans reading is roughly ‘some kind of relationship’. Being used in a LW context you may assume that the reader knows about the mathematical definition and its application to everyday use. Now as long as you use ‘correlation’ as in your post and do not attach weight on the consequences of that use there is no risk. But once you want to use consequences of that definition to support or weaken a point then you must consider what concepts your reader really knows.
As the other commenters noted the Sapir Whorf Hypothesis is mostly wrong. You can have concepts without words. But words clearly help.
The question also is: When do you really have a ‘word’ for a concept. ‘correlation’ alone is no concept. There are stochasitic correlation between stochastic variables or correlations between signals. Those that are discrete or continuous or multi dimensional. Pick your meaning. And arguing that the ‘concept’ is pure; the abstraction or generalization of these specific instances doesn’t really work because nobody really knows what the pure concept is. Thus you cannot communicate it but only appeal to it by reference to particular cases.
Stripping a word of its context (who and when and where it is uttered, onto which other words it is applied) leaves not a pure concept but only a sequence of letters.
Words give the hearer an indication of your indicated meaning but they are not the meaning. That is what EY means with
Words make a certain reading more plausible and (hopefully) let your mental representaticonverge locally toward that of the speaker (your brain will update its structures mostly bayesian, but it may fail given noise).
A large vocabulary then is a large set of ‘signals’ for which both the speaker and the heaer know the concepts signalled.
Because words are also used internally by conscious thought they can also be used to signal to yourself and thus a large vocabulary can be conducive to efficient reasoning.
Because private and public use of the vocabulary are often mixed one should reflect on what the other really knows.
For more on this see http://lesswrong.com/lw/nu/taboo_your_words/ and http://lesswrong.com/lw/kg/expecting_short_inferential_distances/ .