tag:blogger.com,1999:blog-5275657281509261156.post7179924722955298252..comments2024-03-28T04:04:55.806-07:00Comments on Faculty of Language: The logic of the POS, one more timeNorberthttp://www.blogger.com/profile/15701059232144474269noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-5275657281509261156.post-67882806516638469752014-03-31T16:30:32.337-07:002014-03-31T16:30:32.337-07:00I think it is genuinely uncontroversial that there...I think it is genuinely uncontroversial that there are many behaviours in many species that are innate: spider webs, ungulates walking a few minutes after birth, some bird songs (but not all), nesting behaviours of some birds etc etc. and this can be verified by raising the animal in isolation and so on.<br />and it is I think pretty clear that there are learned behaviours that are very highly canalised by some innate structures --- in vision, navigation etc.<br />But all of these have some common factors -- they are evolutionarily very ancient (tens of millions of years), and they are clearly adaptive, and there is as a result no problem for Darwin.<br />Clearly there is a difference between these things and relative clause extraction.<br /><br />It's worth pointing out that Gallistel's notion of a domain is somewhat different from the way that you and I use it --- for example, he considers probabilistic learning a domain, whereas for me it would be a mechanism that could be used widely in many different domains, of which one might be language processing. This terminological difference might account for some of this disagreement.Alex Clarkhttps://www.blogger.com/profile/04634767958690153584noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-69048475077812550142014-03-31T16:13:59.464-07:002014-03-31T16:13:59.464-07:00Yes, I do mean 'covert', but also think I ...Yes, I do mean 'covert', but also think I mean 'unfortunately', because if the usual cases were more extreme, there would be more solid evidence for UG. Of course the rhetorical force of 'unfortunately' in English is interesting, I recall Howard Lasnik pointing out that it usually means 'unfortunately for the proponents of the idea I am attacking, but fortunately for me'.AveryAndrewshttps://www.blogger.com/profile/17701162517596420514noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-68000327576803689062014-03-31T08:21:40.887-07:002014-03-31T08:21:40.887-07:00I take it you meant 'covert' subjects? Al...I take it you meant 'covert' subjects? Also did you mean 'fortunately'? Norberthttps://www.blogger.com/profile/15701059232144474269noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-8238561163631016412014-03-31T08:20:48.426-07:002014-03-31T08:20:48.426-07:00There is an interesting presupposition in your ans...There is an interesting presupposition in your answer that I want to make explicit: you seem to assume that empiricist learning is a general cognitive mechanism that explains how cognition functions in other domains. In other words, it treats language as the outlier while other areas of mentation are easily described in empiricist terms (without much innate hardware necessary). So far as I can tell, (see Gallistel and my posts on him for discussion), nothing could be further from the truth. Given that this is so, the standard ML approaches, which make very weak assumptions about domain specific knowledge are probably wrong EVERYWHERE. So even if one is a partisan of Darwin, the specific approaches you seem to favor have very little if any Darwinian street cred. <br /><br />Seen in this light, it is not only my theories that are unlearnable (actually they are (or it is reasonable to think they are) given the right set up of the hypothesis space and the right priors) but almost every form of cognitive competence we are aware of. You really should read Gallistel's stuff on classical learning in rats. If he is right, and I believe he is, then classical learning theories of the empiricist (ML) variety are biologically hopeless. I take that to imply that Darwin would not favor them. Norberthttps://www.blogger.com/profile/15701059232144474269noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-76168540803270935652014-03-30T15:29:24.609-07:002014-03-30T15:29:24.609-07:00However, it's different if the contentious the...However, it's different if the contentious theoretical proposals go on to predict true facts for which clear evidence in the PLD is close to nonexistent, which seems to be the case for agreements with quirky and non-quirky case-marked convert subjects in Icelandic. Unfortunately, most of the cases people argue about aren't anywhere near so extreme.AveryAndrewshttps://www.blogger.com/profile/17701162517596420514noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-30401977043780554272014-03-30T03:30:12.496-07:002014-03-30T03:30:12.496-07:00As they say, one man's ponens is another man&#...As they say, one man's ponens is another man's tollens. If you have a theory of grammar and then on analysis it is clear that large chunks of it cannot be learned and must be innate, then there are two approaches. One is to say as you do, "Darwin be damned." The other is to question whether your theory might be false.<br /><br />Particularly when, as is uncontroversial, the grammars are severely undetermined by the linguistic evidence available [1], and Chomksyan linguists rely heavily on non-empirical assumptions (like the SMT, full interpretation etc etc ) in the process of theory construction. Given these facts, and the antipathy that many here have to theories of learning, it is unsurprising that the theories you come up with are not learnable. <br />One conclusion is that Darwin is wrong, the other is that these theories (the standard theory, the revised extended standard theory, P & P, etc etc. ) are wrong. <br /><br /><br />[1] e.g. " Choice of a descriptively adequate grammar for the language L is always much underdetermined (for the linguist, that is) by data from L."Alex Clarkhttps://www.blogger.com/profile/04634767958690153584noreply@blogger.com