tag:blogger.com,1999:blog-5275657281509261156.post1436076763988673998..comments2024-03-18T21:50:10.131-07:00Comments on Faculty of Language: A derivation "towards LF"? Hardly. (Lessons from the Definiteness Effect.)Norberthttp://www.blogger.com/profile/15701059232144474269noreply@blogger.comBlogger13125tag:blogger.com,1999:blog-5275657281509261156.post-91782477900968022017-04-13T10:37:40.645-07:002017-04-13T10:37:40.645-07:00@Dan: Yes, there is an alternative theory, based o...@Dan: Yes, there is an alternative theory, based on the (old) idea that certain operations are syncategorematically triggered by the merger of certain heads, obligatorily and irrespective of the representational outcome. This is the alternative, SMT-free theory.<br /><br />Is it better than the SMT-based theory? Well, it is if your criterion is capturing the principles that underpin natural language. (What the handout shows is that the SMT-based theory, given certain fairly reasonable premises, cannot do this; for the alternative, positive proposal, you'll have to read the book.) So, no, it is not the case that "they [are all] equally good." At the same time, I readily admit that the theory I'm defending has a harder time, e.g., from the standpoint of "Darwin's Problem" – since it obviously enriches the amount of language-specific machinery.<br /><br />The way I understand minimalism – and the only way in which I can count myself as a minimalist, actually – is if the goal is to craft a theory that better meets these goals (the Beyond-Explanatory-Adequacy goals) while not forfeiting the ability to account for how natural language actually works. You can see the work we've been discussing as a demonstration of how a strict SMT-compliant theory has forfeited that ability.<br /><br />So now we (and I do mean both of us) have to choose: is accounting for natural language less important than "Darwin's Problem"-compliance, or more important? This is a methodological choice, not a principled one, since having both does not seem to be an option. But since our understanding of the lay of the land concerning "Darwin's Problem" is, comparably, a fuzzy heap of speculations (and Chomsky would be the first to tell you this), I am going to place my intellectual bet on the theory grounded in the domain we know something about. So, yes: a better theory.<br /><br />Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-52097140511002949452017-04-13T07:43:54.714-07:002017-04-13T07:43:54.714-07:00I'll take one more run at this.
Suppose we acc...I'll take one more run at this.<br />Suppose we accept your arguments here, in the handout, and in your monograph. What's next? Do we take the conjunction of all your premises as our new theory? What are the consequences? Is it any better than a theory based on SMT? <br /><br />Refuting a theory isn't that tough. Building a better theory is. As philosopher of science Imre Lakatos once wrote "All theories ... are born refuted and die refuted. But are they [all] equally good?"Anonymoushttps://www.blogger.com/profile/07202099199620783128noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-85305435780028758502017-04-11T18:56:02.090-07:002017-04-11T18:56:02.090-07:00Oh, and as for two logical structures of arguments...Oh, and as for two logical structures of arguments you noted:<br /><br />(1) things like "X&Y can account for D" are seldom evaluable without some further premises concerning what it means to "account for D", and so, are susceptible to the same modus tolens maneuver <br /><br />(2) if W logically entails X and Y, then yes; but otherwise, what it means for W to "subsume" X and Y is laden with premises (cf. the debates on the Movement Theory of Control)<br /><br />So maybe there are premise-free arguments in linguistics, but you still haven't shown me one.<br />Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-73946165539422972822017-04-11T18:43:12.906-07:002017-04-11T18:43:12.906-07:00Interesting pair of examples. (1) It is my impress...Interesting pair of examples. (1) It is my impression that the Movement Theory of Control is empirically less successful than most of its competitors. (2) The unification of Merge and Move is nice, but doesn't work nearly as well as advertised (problems like, e.g., why lower copies/occurrences don't count as interveners, which ultimately require a reification of 'chain', which undoes much of the oomph of the unification in the first place).<br /><br />I suspect you and I have very different definitions of "explanatory force." I think a theory that cannot, given reasonable premises, account for very robust data, has no explanatory force at all. A theory like that becomes an exercise in philosophy, not linguistics. (I know Norbert differs with on this, and you may too. That's fine – I'm just stating my position.)<br /><br />If you have an argument against one of the relevant premises on its own terms, let's hear it. (Note: that it forces you to reject a conclusion you are fond of – such as interface-driven-syntax – is not an argument, in this sense.) Pending that, the onus is not on my to come up with "explanatory force"; the competitor has none.<br /><br />Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-59771782110509919142017-04-11T18:32:44.280-07:002017-04-11T18:32:44.280-07:00@Omer: I think "every argument in the history...@Omer: I think "every argument in the history of generative linguistics" is a bit strong. I can think of two other forms of arguments off the top of my head:<br /><br />(1) Our theory says X. X doesn't account for data-point D. If we add Y to our theory, X&Y can account for D. What's more, X&Y accounts for data-point E. Therefore, let's add Y to our theory<br /><br />(2) Our theory says X&Y. W subsumes both X and Y. Therefore let's replace X&Y with W.<br /><br />If memory serves, (1) shows up in arguments in favour of the movement theory of control, and (2) is how Merge and Move were unified.<br /><br />The point of my response was that the antecedent reasons for rejecting LF-as-telos are on relatively shaky ground.<br /><br />The point of my original comment is in its last paragraph. Suppose we accept your rejection of LF-as-telos and interface-driven syntax. What next? What other parts of the theory are we forced to to reject? What does the new theory look like? Does it have any explanatory force? If generative linguistics is a science, it should progress, and I don't see where we would go from rejecting LF-as-telos.<br /><br />I'll admit that might be too much to expect from a blog post and a handout. I guess I should pick up your monograph. Anonymoushttps://www.blogger.com/profile/07202099199620783128noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-14118251031266882762017-04-11T17:27:16.216-07:002017-04-11T17:27:16.216-07:00@Dan: What you're saying is true for every arg...@Dan: What you're saying is true for every argument in the history of generative linguistics, as far as I can tell. If someone shows that X, Y, and Z together entail W, you can cling to W and assert that therefore, one of X, Y, and Z must be false. You can question the premises of any argument – actually, let me correct myself: you can question the premises of any argument that is explicit enough about its premises – so I'm not quite sure what your point is.Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-58132814995066510632017-04-11T17:17:54.306-07:002017-04-11T17:17:54.306-07:00@Omer: I can't speak to the contents of your m...@Omer: I can't speak to the contents of your monograph, but the handout looks like it suffers from the same issue. It purports to be arguing against a theoretical statement (i.e. Syntax is driven by interface conditions) but in order to make the argument it brings in additional premises (both explicit (P1 and P2) and implicit (e.g. what is a primitive for a given module)) and shows that the data is incompatible with interface-driven-syntax+P1+P2+implicit-premises and conclude that syntax cannot be (solely) interface driven. <br />This seems pretty sound, but the same data could probably be used to argue against any one or combination of your premises if we assume syntax is interface driven. Anonymoushttps://www.blogger.com/profile/07202099199620783128noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-69999006816400684702017-04-11T15:48:21.994-07:002017-04-11T15:48:21.994-07:00Great. Thanks. I would be curious to see how the...Great. Thanks. I would be curious to see how the information structure people deal with this data.Anonymoushttps://www.blogger.com/profile/07765373355210375175noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-25280222368393273742017-04-11T09:39:45.519-07:002017-04-11T09:39:45.519-07:00@David: Fair enough. I've added some more data...@David: Fair enough. I've added some more data to the post, following your comments (as well as Ethan Poole's). I hope these address your concerns.Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-50711084346172840302017-04-11T08:52:32.385-07:002017-04-11T08:52:32.385-07:00I am unfamiliar with the syntactic analysis given ...I am unfamiliar with the syntactic analysis given by S as well as the Icelandic data so I am confused about some aspects of the first set of Icelandic data. The ungrammatical example (14c), which I assume to be, in English, 'the child was felt to be troublesome' does not behave like English at all, since it is marked as ungrammatical. But (14a), in which the aux 'be' seems to be missing in the upstairs clause, allows 'the child' as a subject, and it gets some sort of impersonal reading (with 'people'). So what is happening in the difference between (14a) and (14c)? Also, depending on where you want to have the base position for the subject, the example in (15a) would have would have the subject in a derived position (specifier of some TP) if we take that the subject starts out lower, in some sort of small clause complement to 'be'. Again, if this represents a derived position for the infinitival subject this is not possible in English--'*There seems a dog to be in the garden.' I'm not sure a Milsarkian based theory would have anything to say about these grammaticality contrasts.Anonymoushttps://www.blogger.com/profile/07765373355210375175noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-27068395107299306082017-04-11T07:41:39.953-07:002017-04-11T07:41:39.953-07:00@Dan: Yes, I agree with your characterization of t...@Dan: Yes, I agree with your characterization of the logical structure of the problem. But I have some antecedent reasons (that go unstated in this post) to opt for rejecting (what you call) S. For example, the interpretable/uninterpretable distinction, and the idea that all syntactic operations are "free" and their occurrence enforceable via "interface conditions" only, is untenable, for reasons having nothing to do with semantic interpretation. (See my 2014 monograph, and for a handy – but partial – cheat-sheet, see <a href="http://omer.lingsite.org/files/Preminger---The-heartbreak-of-interface-driven-syntax.pdf" rel="nofollow">this handout</a>.) So your suggestion that we reject (what you call) L is, from where I sit, an attempt to salvage something that doesn't work in the first place.Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-2869594151543858692017-04-11T07:32:53.721-07:002017-04-11T07:32:53.721-07:00There's an unstated assumption/axiom in this a...There's an unstated assumption/axiom in this argumentation that I think might be leading you to the wrong conclusion. You're assuming that our current theory of LF/interpretation/semantics (call it Theory L) is (more or less) correct. So what the data is testing is not the LF-as-telos theory (call it Theory S), but the conjunction of LF-as-telos and Theory L. Assuming the Definiteness Effect data is, in fact, incompatible with S&L, then we have two possible conclusions:<br />1) Reject Theory S (your choice)<br />2) Reject Theory L<br /><br />Personally, I think Theory L is on much shakier ground than Theory S and it deserves some rethinking. <br /><br />Obviously if we reject L we need to replace it with a new theory, just as rejecting S necessitates a new syntactic theory. Either choice opens up a whole new research programme with all sorts of interesting questions. For instance, if we assume syntax is "doing its own thing" we have to ask what that thing is and how it came about. Where did the features that drive syntax come from? Why are some interpretable? Is there a difference between semantic and non-semantic features? Are there linguistic expressions that are purely sound-meaning pairs, or do they all have some residue of syntax? And so on...Anonymoushttps://www.blogger.com/profile/07202099199620783128noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-80098519780843515972017-04-10T17:06:23.431-07:002017-04-10T17:06:23.431-07:00Some interesting discussions of this over on faceb...Some interesting discussions of this over <a href="https://www.facebook.com/omer.preminger/posts/441394319544201" rel="nofollow">on facebook</a>. (I've made the post public, but you might still need a facebook account to see it...)<br /><br />See in particular the data from Vangsnes (2002) that Ethan Poole notes!Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.com