Syntax I

When you have an embedded clause, it could either be an embedded TP or an embedded CP. So, when do you have which?

First, if the embedded clause is finite, you have a CP, that much is guaranteed. Even if you can’t hear the C (since C for finite clauses can either be that or ∅). One example: Pat said [_CP (that) [_TP Chris left ] ].

If the embedded clause is non-finite, the answer is “it depends.” Sometimes you can hear the C, in which case, of course, there’s a CP. For example: Pat wants [_CP for [_TP Chris to leave ] ].

Another case where a non-finite clause is a CP is when its subject is PRO. If you have a PRO in an infinitive to-clause, there has to be a CP in order for PRO to get case. For example: Pat wants [_CP ∅_NULL [_TP PRO to leave ] ].

When the subject of the embedded non-finite clause gets case from the higher verb (ECM), then there is no CP, just a TP. For example: Pat wants [_TP me to leave ].

Those are, I think, the basic cases. So, it comes down to deciding whether there is a PRO in the embedded clause or not, to a large extent.

I’ve gotten some questions about a few things, so I’ll post some review notes about each. But don’t forget to read the relative extended review notes (and summary sheet) I wrote up for the last week of class. I think that they’ll help set the scene if there are things that you aren’t quite sure about. And don’t forget to read the blog, although I suppose there’s little point in my saying that here.

So, the first installment: What are islands and how do they work?

We have covered three kinds of islands, although Adger covers a fourth kind of island (subject islands) in the book as well. The three kinds of islands we have seen are adjunct islands, wh-islands, and DP islands (aka Complex Noun Phrase islands.

An island, somewhat pretheoretically, is a constituent that “traps” things from moving out of them. The idea is that something that is in (on) an island cannot escape, cannot be moved away.

So, adjuncts constitute one example of an island. If you have a complex clausal adjunct, such as if John buys a coffee, it will be adjoined to TP in a sentence like I will eat my hat if John buy a coffee. Suppose that you want to question what I’ll eat. So, you substitute what in for my hat, and use an interrogative complementizer, which will force what to move into SpecCP. The result is the perfectly well-formed question What will I eat if John buys a coffee?. (Or, if you started from If John buys a coffee I will eat my hat, with the clausal complement adjoined on the left, then turning it into a question results in the reasonably acceptable What if John buys a coffee will I eat?).

If you try to ask a question about something inside the clausal adjunct, however, the result is not well formed. So, suppose we want to ask a question about the thing John might buy that would result in a hat-eating by me. Replace a coffee with what, use an interrogative complementizer, and you wind up with *What will I eat my hat if John buys? (or *What if John buys will I eat my hat?).

So, the adjunct “traps” a wh-word from moving out of it. You can’t move a wh-word from a position inside an island to a position outside the island.

Another type of constituent that can trap a wh-phrase is a question (a wh-island). So, if you have a sentence with an embedded question, as in John wondered who brought chips to the party, you can’t then ask about something inside the embedded question, such as chips. If you replace chips with a wh-word, and use an interrogative complementizer, the result is *What did John wonder who brought to the party?—not so good.

The last type of constituent we looked at that can trap a wh-word is a definite DP. If you have a DP that is complex enough to contain something you can question, you can’t anyway (if it is definite). So, for example, John’s gift of granola to Mary. If you had a sentence with this DP (John’s gift of granola to Mary caused great offense), you couldn’t ask about the granola part: *What did John’s gift of to Mary cause great offense?. Other examples would be asking about poetry in John left his book of poetry on the T: *What did John leave his book of on the T?… or asking about cats in I watched the movie about cats yesterday: *What did you watch the movie about yesterday?.

There are a couple of auxiliary points that should be made about CNP islands. First, a DP is only an island if the DP is definite. Definite DPs are those that have the, or have something genitive in SpecDP (my book, the book, my destruction of the hard drive).

Another point about CNP islands is that they’ll often contain an adjunct anyway, such as when the CNP has a relative clause (the book (that) I gave Mary). This is basically a triple island—if you try to ask about Mary here by replacing it with who, the wh-phrase was already trapped inside the adjoined relative clause ((which/Op) (that) I gave who), whose SpecCP is full, and which is additionally inside the definite DP.

Our explanation of why an island “traps” wh-words like this has to do with the idea of phases. The sentence is built up in “chunks”, where each CP and each definite DP is a “chunk.” So, once a CP is complete, the only thing you can “see” in the CP from outside is whatever is in SpecCP. So, if a wh-word is going to move out of a CP and into a higher SpecCP, it needs to first get to the lower SpecCP so that it can be “seen” from outside.

A wh-island arises when the SpecCP in the middle is already full, either containing a wh-word that has moved there beforehand, or containing the invisible Op that one finds in SpecCP in yes/no questions. Since the wh-word that you’re trying to move out of the embedded question cannot be moved into SpecCP, it gets stuck. The CP phase completes, and the higher interrogative C can no longer “see” the wh-word because it’s inside a finished phrase.

The other islands work on essentially the same principle. In adjuncts, the idea is that you can’t see anything inside them, even something in SpecCP. So, whether the wh-word can make it to the edge of the adjunct or not, it’s still trapped once the adjunct has been built. For CNP islands, the idea is that it is not possible to move a wh-word into the specifier of a DP, and so if the DP is definite and therefore a phase, the wh-word will get stuck inside.

In preparation for the overview class on Tuesday, I’ve put together a short essay synthesizing things we’ve looked at over the semester. It’s linked to the syllabus page, or you can click here to get it.

I think it will be helpful to read over it at some point. My plan is to use it as a kind of guiding script on Tuesday, although we may skip over bits or get into more detail on other things. Actually, the most helpful way this could probably work is for you to read this before coming to class, and then help make sure I get to parts of it that didn’t make sense to you, or ask about parts that may not have made it into the document. There are many things we did over the semester that aren’t discussed there in any detail.

For your amusement, a step-by-step derivation of Carmela said AJ’s advisor was told that AJ might want to study event-planning and of The continuous drawing of complicated trees builds strong character.

It might be useful to read through these (although be warned that it’s just about 50 pages long). The PDF file is linked to the syllabus, or you can click here: Sample derivations.

You may remember on homework 5, we figured out that to was a modal, and in your analysis of the sentence I wanted Joss not to have been making movies, we had to say that the modal to is special in that it doesn’t raise to T.

I just noticed that on handout 10b, I’d accidentally drawn to moving to T. I’ve posted an updated handout on the syllabus page with that fixed.

Just to clarify, there are a couple of inconsistencies with respect to how I drew the trees with the infinitive marker to on some of the handouts. Specifically, there are slides here and there where to is drawn into the trees as if it were T—rather than an M that moves up to T. The official version of the theory we’re working with (and different from what is in the textbook) is that to is an M that moves up to T, just as in the homework problem from a few weeks ago. So, if you see somewhere to given as being a T, just think of it as a shorthand for a T that takes an MP complement, headed by the M to, which has moved up to T. I’ll try to catch this in future handouts, but I still might miss one here and there.

[Edit after original post: Part of the point of that homework was showing that to doesn’t in fact move up to T.]

The handout labeled “10b” has a couple of slides where the text is unreadable due to the background in the text boxes having been too dark. I’ve posted a new version of the handout on the syllabus page that fixes that problem, but if you don’t wish to print it out again, here’s what you missed:

First slide:

• Charlie seems to dislike bees
• First, does Charlie get a θ-role from seem?
• Well, no. Seem only assigns the one θ-role.
• So, unlike in Charlie tried [PRO to elude the bees], we have as many DPs as we have θ-roles.

Second slide:

• Charlie seems to dislike bees
• So, what θ-role does Charlie get?
• Still seems to be the Experiencer of dislike
• So, suppose that Charlie starts out in the same place, SpecvP.
• But now, after building vP, we add a nonfinite T…

We’ve just met quite a number of different kinds of embedded clauses, particularly embedded infinitive clauses, and I just want to take a moment to summarize what we have and suggest strategies for determining what we’re looking at in any given sentence.

Some things to keep in mind, in no particular order, are:

• DPs need case. Every single one.
• Each θ-role must be assigned to a distinct argument.
• Infinitive T (generally marked by the modal to) cannot check case.
• Finite clauses always have a C.
• T (whether finite or non-finite) requires there to be a DP in its specifier.
• PRO is a DP, and can only survive if it is assigned null case.
• The only thing that can assign/check null case (so far) is the silent C, ∅_NULL.
• If an embedded subject gets case from the higher clause, the embedded clause is a TP (not a CP).

When you have a sentence embedded inside another one, the first question to ask yourself is whether the embedded sentence is finite or non-finite. The non-finite sentences generally have a to in them, which is a pretty good tip-off. Modals like should, may, etc., only appear in finite clauses.

If the embedded clause is finite, it is a CP (even if the C is silent). Finite clauses are always CPs.

If the embedded clause is non-finite, there are several possibilities. For one thing, it might be a CP or it might be a TP—it depends in large part on the verb that the embedded clause is the complement of. There’s also the question of whether (if no subject is visible in the embedded clause) there is a PRO or not.

The way to determine whether there is a PRO is to count the θ-roles. That’s pretty much the only reliable way to do it.

In a sentence like John wants Mary to leave, there are two verbs to consider: want and visit. The participants in want are the wanter (Experiencer, John in this instance), and the thing wanted (Proposition, the embedded clause Mary to leave in this instance). The sole participant in leave is the leaver (Agent, Mary here). So, nothing here is unaccounted for, all θ-roles are going to visible/audible DPs—there is no PRO here.

In a sentence like Mary seems to be leaving, if we run through that same procedure, we will again find that all θ-roles are accounted for. Seems has a single θ-role, the Proposition that appears to be true, and leave again has one θ-role, the Agent doing the leaving (Mary). However, in this case, there is no visible subject in the embedded clause, and moreover the logical subject (Mary) shows up as the subject of seems. There are two TPs here, one in the main clause, and one in the embedded clause. Each requires a DP to be in its specifier, but only the top one appears to have a DP in its specifier. There can’t be a PRO in the specifier of the lower TP because there aren’t enough θ-roles for there to be a PRO (if there were a PRO, it wouldn’t get a θ-role). The only conclusion is that the subject of the lower clause (Mary) was first in the specifier of the lower TP (checking the [uD*] feature of the lower T), and then moved up to the specifier of the higher TP.

Considerations of how case is checked also point to the same conclusions with respect to these two sentences. Every DP needs case. In John wants Mary to leave, the DP John gets case from the finite T in the main clause, and Mary gets case from wants (the non-finite T in the lower clause is incapable of checking case features, like all non-finite Ts). In Mary seems to be leaving, there is nothing in the lower clause that can check the case of Mary, so the main clause T values/checks its case as nominative.

In both of these two sentences (John wants Mary to leave and Mary seems to be leaving), the embedded clause is just a TP, not a CP. Since Mary in both cases, while originating in the lower clause, gets case from the higher clause, there can be no CP there. If it helps, you can think of CP as a kind of a “case shield”, you can’t check case across a CP.

In the sentence John wants to leave, if we run through the θ-role-counting procedure, we’ll find that there are more θ-roles than visible DPs. There’s the wanter (John) and the leaver. Since the visible DP John can’t get both θ-roles, there must be a PRO. And if there’s a PRO, there there must also be a null C (∅_NULL) to value/check null case on PRO.

Finally, consider the sentence John asked Mary to leave. Here, again, the procedure of counting θ-roles comes up short. We have the asker (John) and the askee (Mary), and the leaver. Since there are more θ-roles than visible DPs, there must be a PRO in the embedded clause (and therefore a ∅_NULL C as well).

To summarize in bracket notation we have:

• John wants [_TP Mary to leave] (ECM)
• Mary seems [_TP <Mary> to be leaving] (Raising)
• John wants [_CP ∅_NULL [_TP PRO to leave]] (Subject control)
• John asked Mary [_CP ∅_NULL [_TP PRO to leave]] (Object control)

Once we have these four kinds of verbs, it’s always possible to mix-and-match them (and it makes for good homework problems and test questions too). For example, John seems to want to ask Mary to leave:

John seems [_TP <John> to want [_CP ∅_NULL [_TP PRO to ask Mary [_CP ∅_NULL [_TP PRO to leave]]]]].

I_i seem [_TP t_i to have overlooked the fact that we never added into the UTAH the statement that a TP can be assigned a Proposition θ-role when it is the complement of verbs like seem.] I sort of assumed it without explicitly incorporating it into the system.

So:

• CP sister of V: Proposition
• TP sister of V: Proposition

Notice that even though there are two ways to get a Proposition θ-role, it’s still possible to tell, based on where the TP or CP is in the tree, what &theta-role is assigned. The relation between structural positions and θ-roles is not one-to-one. In fact, this observation is not really new—for example, we already had two places where Agents could be found as well (SpecnP and SpecvP).

I’ve been meaning to write up some of the things that were covered during the review class on Tuesday, but I haven’t had a spare moment. However, since I was asked something over email that relates obliquely to this, let me run through one of those things here.

The only place that the genitive case ever shows up is in the specifier of DP, and more specifically, it shows up in the specifier of a DP whose head is ∅_GEN. The D ∅_GEN has the features [D, ucase:gen*, ucase: ] (that is, it is a D, it assigns genitive case to its specifier, and the DP it heads must itself get case).

So, if you see something in the genitive case (something like his, her, their, senator’s or Pat’s), you know that it is in SpecDP. In fact, that is the basic function of Case, to tell you where something is in the tree.

(Likewise, if you see something that is in nominative case, you know that it is in the specifier of TP.)

Meanwhile, if we consider the function of the UTAH, the UTAH tells us the θ-role that goes with a DP (or, in the case of Goal, a PP) is based on the position in which it first appeared in the tree (as it was built from the bottom up using Merge). A DP in the specifier of vP (the DP daughter of vP) or a DP in the specifier of nP (the DP daughter of nP) will be an Agent (or, sometimes, an Experiencer). The DP daughter of NP and the DP daughter of VP will be a Theme. A PP sister of N and a PP sister of V will be a Goal.

Thinking first about verbal clauses (TPs), notice that the θ-role of the nominative argument (the subject) can be one of several different possibilities:

(1) I jumped.
(2) I fell.
(3) I was pushed.
(4) I was given a book.
(5) I was introduced to Mary.

In each case above, I is of course nominative, it is the subject, it is in the specifier of TP. But it started in different places in different sentences, even though in each case it moved into the specifier of TP. In (1), I is an Agent, so it started in SpecvP. In (2-5), I is a Theme, so it started in SpecVP. In (2), we have an unaccusative verb, in (3) we have a passive transitive verb, in (4) we have the passive of a ditransitive verb (of the “have” type, it was caused to be that I had a book), and in (5) we have the passive of a ditransitive verb (of the “go” type, it was caused that an introduction of me went to Mary).

The point here is that: Case tells you where a DP ends up, and the θ-role tells you where that DP started out.

The same is true in DPs. So, we have several possibilities:

(6) [My hat] was ugly.
(7) [My buying of a hat] was a mistake.
(8) [My falling] was painful.

In each case (6-8), my is genitive, meaning that it is in SpecDP. In (6), my is the possessor of the hat (it has a Possessor θ-role). In (7), my is the agent of the buying event (it has an Agent θ-role). In (8), my is the theme of the (unaccusative) falling event (it has a Theme θ-role). So, my started out in a different place in each of the three examples in (6-8).

Starting first with (7), we know that because my is getting the Agent θ-role, it must have started in SpecnP. In (8), we know that my started as the DP daughter of NP (in this case the sister of N), since it is getting the Theme θ-role.

In (6), because my is a Possessor, we know that it started in the specifier of a PossP. Because it is not Agent or Theme, it couldn’t have started as DP daughter of nP or DP daughter of NP. Because there is a Possessor θ-role assigned, there must have been a PossP in whose specifier my was first Merged.

The Hierarchy of Projections within the DP goes like this: D > (Poss >) n > N. So, the PossP was between n and D.

This also means, incidentally, that the only time you need a PossP is when you have something that gets a Possessor θ-role (the Poss is optional in the HoP for the DP, just like Pass, Perf, Prog, Neg, and M are in the HoP for the TP).

Anyway, the strategy for deciding what the tree should look like when you are drawing a DP is to first determine the θ-role of the genitive DP. If it is a Possessor, you need a PossP, and the Possessor starts in SpecPossP, then moves to SpecDP. If it is an Agent, you don’t need a PossP, and the DP starts in SpecnP, then moves to SpecDP. If it is a Theme, you don’t need a PossP, and the DP starts as the DP daughter of NP, and moves to SpecDP.