My question is, for a given AGREE relationship between X and Y, how could one determine which category hosts the interpretable feature and which one hosts the uninterpretable one?
AGREE is driven by interpretable/uninterpretable feature pairs e.g. uPhi/iPhi or uWH/iWH etc. For examples like phi checking between T and a subject, it is taken for granted that the subject has interpretable phi features and that T has uPhi features; but the subject has uT (or uCase) features while T has iT (or iCase).
One argument for this seems to be, in part, semantic: there is a semantic reality to, for example, number which make it plausible that number is interpretable on nouns, while uninterpretable on verbs. (Although, one must also concede that number on pluractional or reciprocal verbs is not all that far fetched, which undermines this type of argument). However, when looking at other constructions, or other types of syntactic interaction, it's not always clear that this type of argument works.
For example, in topicalization constructions such as (a,b,c), let's assume that the topicalized constituent moves to SpecTopicP to check a Topic feature. But is iTopic a feature on the moved constituent and uTopic on the head of TopicP? Or is it the other way around? Is this something that could be parameterized? More specifically, I'd like to know what *arguments* could be marshalled either way.
(a) Peter Florrick I could vote for.
The case of Topicalization seems to argue for the "reverse Agree" theory proposed by some people (including David Adger and Susanne Wurbrand). The topic feature would then always be uninterpretable in DPs, which would force DP movement to a "criterial position", as defined by Luigi Rizzi (Spec TopP in this case). In the more mainstream theory, in which an uninterpretable feature acts as a "probe", the topic feature in the C domain would have to be considered uninterpretable, but then it is not clear what that feature enters in Agree with (the D feature of the DP maybe?). I hope this is helpful.
Hi Marcello,
I think this is a good start. Sticking to mainstream theory for the moment, why exactly would the topic feature in C be considered uninterpretable? (Putting aside the fact that most people probably just stipulate it to be so). In other words, are there ways of ascertaining whether a feature is uF or iF? In other words, if I suddenly claimed that Topic was interpretable on C but uninterpretable on a DP, then would that claim be able to be subjected to linguistic tests to ascertain its truth?
I guess I'm wondering about the ontology of categories here: if uF is a "thing" then there should be evidence of it or ways of testing for it.
Yes Mark, I think you are absolutely correct, there is no testable or even logical way of deciding if a feature is i/u (as far as I can tell) and Chomsky had always know this and that it is a problem..., in fact, Chomsky seems to be moving away from that way of dealing with things (movement would then be a labeling-enabling mechanism, not a uninterpretable-feature-deletion operation or caused by some mysterious EPP feature). See his latest article in Lingua.
Thanks for the tip, Marcello. I'll read it (should have done this a while ago, but it was the kind of special edition I wanted to give a bit of quality time to... and in the end it was just swamped by urgent things to do...)
I think optimality theory gives a better answer to this problem. The hierarchical order of constraints solves the problem of interpretable versus un-interpretable. If the constraint is high than the item likeliness to appear in the output is very low and vice versa.
Dear Mark,
I'm really busy doing something highly discursive and historical and far from being scientific and technical ...But roughly speaking here is an example which belong to Ross's first type-adjunct islands
A-.you slept because you needed to do that
B-*What did you sleep because you needed to do __? - .
As you well know the Wh-movement fails in the b-sentences because the gap appears the adjunct clause.
Now what optimality theory can add if it is applied to the field of syntax is very simple in fact but very revealing. Ross presented us with a group of problems (7 types of islands) and left us with an implied duty to find an answer to these real problems.
OT says they are just constraints, like phonological constraints in every language. and we can deal with them by classifying the intensity and the strength of their constraining fields. In other words if the preverbal constraint for example is perennial in Chomsky's work why not integrate sub-constraints such as Ross's seven islands?
Ibrahim, sorry to put my nose in, but, in my understanding, that would explain nothing. Saying that islands are constraints and EPP is a contraint is as explicative as postulating that every language needs a Spec T subject. It just state the facts, rather than explain them. Phonology is a completely different manner since, in phonology, there seem to be real constraints, some universal like against syllables like "LBA", but allowing "BLA"; and some constraints that are language-particular (like not starting a syllable with "S" in Portuguese). There is nothing like that in syntax.
Dear Marcello,
You are fine. If you are looking for universals than your job will be indeed to eliminate constraints. Chomsky, as you well-know started with transformational grammar and u know where he is at now. But the problem is to my mind is very simple. What if there is nothing called pure universals? Or, in other words, what if the universal and the local coexist inseparably at the same n point in time?
So if the question is genuine enough to be asked than the corollary would be that any syntactical theory has to be universal and specific at the same time.
As to Optimality Theory I think it is a system of analysis rather than peculiar to any field. You can apply it to anything. For example, in classical mathematics we postulate that , two parallel lines never meet infinitely. Such a thing doers not exist in real life, but we need to pretend that it exists in order for example to easily measure angles if the two lines are intersected by a transversal. In other words, the absence of a real-life similarity does not exclude the applicability of a theory.