Infants actually have a phenomenal memory, literally photographic in some testable ways. Their deficiency is not in working memory, but in focus; they lack both the experience and the commensurate neural structures to identify what they “should” and “should not” find important. This is why babies are often obsessed with things that adults find trivial. Synaptic pruning (among other things) reinforces neural pathways that are used frequently while eliminating pathways that are rarely used. This severely diminishes general learning ability and broad spectrum working memory, while substantially boosting the ability to focus on already-developed areas. Learning decreases, but ability to perform increases.

Synaptic pruning occurs in a variety of ways at a variety of ages. Major linguistic pruning is completed at age 5, which is why humans who haven’t learned any language past that age generally can’t learn a language. (The brain is also optimized for linguistic acquisition at younger ages, which is why deaf children who don’t learn ASL until age 3 are already at a substantial cognitive deficit and will almost always be outperformed by earlier acquirers.)

And as one would imagine based on this data, it’s less about “how many” languages you know (which isn’t neurologically well-defined anyway) but rather at what age you learn them. The success of immersion before 42 months, for example, depends really only on the age and only negligibly on whether or not a first language had already been acquired. The time-frame of linguistic pruning provides a partial explanation for this: before pruning, the new linguistic information is absorbed “more directly” (whatever that means). After pruning, the ability to learn broadly is culled, but the ability to use what one has already learned is increased, so that one is more apt to learn a second language in terms of existing known languages. It’s nearly universal for adults who learn a new language to process “bonjour means hello”, rather than to directly understand the meaning of the word.

An interesting correlate of this is that a psychologist with an MRI can’t tell you how many languages you know (again, not neurologically well-defined), but he can tell you (very roughly) when you learned them.

Questions such as these are very interesting, but because of the highly ad-hoc nature of the brain, it’s virtually impossible to predict the answer, or even the general behavior that motivates the question, a priori.

And Oxford has some great podcasts, though the post frequency is lower: http://www.philosophy.ox.ac.uk/podcasts.

I attended a Cog Sci talk at Rutgers roughly 15 years ago presenting evidence that (if I remember correctly) those who learned two ‘first’ languages simultaneously (both were often spoken in the home) surprisingly had deficiencies in one of the two languages, deficiencies that would not be noticed by native speakers but nonetheless were measurable. I could easily misremember the talk, so it would be interesting to hear from those in the field who know more.

Which raises the pretty fantasy that there might be a way to create just the right distractions to enable us to learn a second language in the same way that we learned the first, by preventing us from focusing on it well. But I’ve not heard of anyone trying to do that.

I was looking at some kind of comparative notion, maybe one that can be turned into a degree concept if the comparisons work well.

It is possible, indeed desirable, that there could be a small set of fundamentally E-natural properties, and relative E-naturalness could be derived in terms of them, but I don’t assume that.

What I am assuming, really, is that there is something like evidential probability. Whether the magical evidential probability function is something whose nature can be derived from a few key posits, or whether it is much more complicated than that, is of course far from clear.

I’m glad you liked the paper!

My view for now is that the subsentential stuff works the same way as the sentential level. It would be unreasonable to assume that someone else is using a signalling convention with an unnatural syntax. It would be even more unreasonable to expect other people to know you were doing that, without any explicit notification. By the kind of ought-to-is magic that the principle of charity allows, that makes it impossible to use a signalling convention with unnatural syntax without explicit marking that you are using such an unnatural syntax.

So I think we can get the result that sentences don’t get magically different meanings when there is a 200th occurence of “cauliflower” in them, and more generally that the parts relate to the whole in reasonably predictable ways, out of the pre-1983 theory, updated only to include that naturalness impacts rationality.

I was wondering: what’s your view, in the end, of what Lewis should say about the determinacy of subsentential content? And similarly, of the determinacy of sentential content for sentences where there’s no regularity of use (so no convention by his lights)—-maybe those too long for us to comprehend, or closer to home: for garden path sentences and those which we can’t deal with because our parser can’t cope?

My picture before hearing you and wo on this stuff was that, on the headfirst Lewisian strategy, we first appeal to naturalness to secure fairly determinate coarse-grained mental content. But then, because sentence-proposition pairs for sentences actually used radically underdetermine Lewisian “grammars”, we’d have to appeal to naturalness all over again to get a fix on that extra stuff—-subsentential stuff and sentences that are “beyond use”. I thought that fitted nicely with the discussion of Kripkenstein in Meaning Without Use. It also fits naturally with the passages in the “Language and Languages” where Lewis raises underdetermination worries, and appeals to something extra—-simplicity in the that case—-to resolve it. Do you now think you can get away without that second appeal to naturalness? I wasn’t sure whether the final sections of the paper were speaking to that or not…

In general, I am very much like the idea that the role of naturalness in determining mental content comes in via induction in something like the way you suggest. I hadn’t thought of things that way before hearing you on this stuff. But I don’t really buy the claim that this gives us all the determinacy of content we need (if you like, it’s not fundamentally a mind vs. language thing; it’s coarse-grained vs. fine-grained content determination).

Another question was just about these E-natural properties. When reading your paper I guess I started by thinking of them as some early modern list of macro-detectable primary and secondary qualities; but at times it looks like the test was more for a notion of “degree of E-naturalness” measured by something like the amount of evidence required to warrant a generalization (or something)—-extrinsic properties like “being such that all emeralds are green” might be E-natural in the latter sense but not in the former. Just for starters: are you thinking of what you’re appealing to being an all-or-nothing distinction (from which other notions can be built, maybe by looking at lengths of definitions in an E-natural language…), or a comparative notion of P being more E-natural than Q, or a notion of degree of E-naturalness?