Syntax I

Here’s what happened in class last time: We looked at some constraints on the allowable shapes of noun phrases and worked up to the conclusion that what we had been calling noun phrases are actually headed by D. So, they’re DPs. Given that we now have DPs, with NPs inside, we have some things to clarify. There were a number of properties that we assumed held of noun phrases, but now we need to figure out which of those properties are properties of D and which are properties of the N inside.

Many of the properties we thought were of N are actually of D. So: it is DPs that get θ-roles (not NPs), DPs that need case (not NPs), DPs that can be moved to SpecTP to satisfy the [uD*] (”EPP”) feature of T. One thing that still belongs to N are the φ-features (person, number, gender): It is the noun staplers that is intrinsically plural in the DP those staplers.

Since it is DPs that need case, the [ucase: ] features go on the D (not on the N).

Notice that it is quite possible for a DP that gets one case to have other DPs inside it that get other cases. Every DP needs its case checked, and everything that can check case (T, v, P, and the special V that appears in double-object constructions) can only check the case feature on a single DP.

To give an example, here’s what my dessert would look like in the sentence My dessert is burning. The subject is my dessert and so the whole DP gets nominative case. But the possessor my gets genitive case. I’ll draw the top of the tree, with the case features (I am leaving out the [uφ: ] features, among others):

                  TP
                /    \
             DP        T′
           /    \      |  \
         DP      D'    T   ...
    pronoun    /   \   ^----------[T, uD*, ucase:nom, …]
 [D, φ:1sg,   D      NP
 ucase:gen]   ∅poss  dessert [N]
             [D,
             ucase:gen,
             ucase:nom]

(Side note) Ok, so we end up with two [ucase: ] features on the ∅_poss. That’s perhaps not the best aspect of this theory, we need to stipulate something that says that the one that determines the pronunciation of case is the one that was valued (or perhaps the last one that was checked). This came about because ∅_poss both checks the case of something else (the genitive pronoun) and needs its own case checked. So, this only arises when we have a D that can check the case of something else, and there are only a small number of those (basically, just ∅_poss and maybe every in the king’s every whim).

To clarify how the tree above came about, let me build the DP up step by step:

            D′
          /   \
        D      NP
     ∅poss  dessert [N]
   [D, φ:1sg,
   ucase:gen,
   ucase: ]

Then, we put in the possessor, a pronoun with the features [D, φ:1sg, ucase: ], and its [ucase: ] feature is valued and checked by the [ucase:gen] feature of ∅_poss.

             DP
           /    \
         DP      D′
    pronoun    /   \
 [D, φ:1sg,   D      NP
 ucase:gen]   ∅poss  dessert [N]
             [D,
             ucase:gen,
             ucase: ]

Then, the derivation continues: We merge this in with V burn, merge the result with v (unaccusative), merge the result with T, at which point the [ucase:nom] feature of T can see the still-unvalued [ucase: ] feature on ∅_poss, and so it checks and values it, and the whole DP my dessert moves to SpecTP to check the strong [uD*] (”EPP”) feature of T. The top of the tree would then look like the one I drew above.

In class today, we officially met the DP for the first time, and its introduction brings up several issues. Here’s one (this isn’t a problem of any kind, but perhaps it’s noteworthy anyway): The φ-features, interpretable on N (still), might actually value and check two different [uφ: ] features (it depends on the sentence). One of these [uφ: ] features is on the D. If the DP is the subject as well, then the other [uφ: ] features are on T.

Consider: Citizens have voted. The subject, citizens, is a DP—subjects are always DPs. The DP is indefinite, and the noun is plural.

So, we take these…

D
[D, indef, uφ: , uN*, ucase: ]

N
citizens
[N, φ:3pl]

…and put them together… [Note: I have edited this since first posting it, I had some Ns where there should have been NPs]

             DP
           /   \
          D     NP
 [D, indef,     citizens
       uN*,     [N,
      uφ: ,      φ:3pl]
   ucase: ]

…check, value, check…(The φ-features of N value and check the [uφ: ] feature of D.)

             DP
           /   \
          D     NP
 [D, indef,     citizens
       uN*,     [N,
    uφ:3pl,      φ:3pl]
   ucase: ]

Notice, though, that nothing has happened to the φ-features of N. They’re just like they were. They participated in valuing and checking the features on D, but the φ-features on N are interpretable so they don’t need to check anything, and nothing happens to them if they do. Point is: They’re still around as we continue building up the tree until we Merge in T:

              TP
            /    \
          T       vP
  [T, uφ: ,     /    \
  ucase:nom  DP        v′
  etc.]    /   \      / \
          D     NP  V+v  <VP>
 [D, indef,  citizens
       uN*,  [N,
    uφ:3pl,   φ:3pl]
   ucase: ]

Now, T can see N (T c-commands N), so the φ-features of N can come to the rescue again, now valuing the [uφ: ] feature of T. Meanwhile, the [nom] features of T and D match and check. There are other features on T that I haven’t bothered to list, such as the [uD*] (”EPP”) feature, so after this step, the subject will still move into SpecTP.

              TP
            /    \
          T       vP
[T, uφ:3pl,     /    \
  ucase:nom  DP        v′
  etc.]    /   \      / \
          D     NP  V+v  <VP>
 [D, indef,  citizens
       uN*,  [N,
    uφ:3pl,   φ:3pl]
   ucase:nom]

The point is just that the interpretable φ-features on N can be recruited more than once to check uninterpretable features. Note (added after the original posting): However, following Adger p. 262, we should assume that it is in fact the [uφ:3pl] feature of D that values the features on T. See this later blog post.

The DP we end up with here would sound like: “Citizens” (that is “∅ citizens”) because the indefinite plural article is ∅.

There is a sentence in which you need to draw a tree in which they and all appears. This sentence points out a mystery, however. Compare this to all the students.

(1)  All the students must have ordered sushi.
(2)  The students must have all ordered sushi.

We have a ready explanation for this: In (1), all the students moved to SpecTP together, while in (2), the students moved alone, leaving all behind.

However, if we try this with they instead of the students, we find that it isn’t quite the same.

(3) *All they must have ordered sushi.
(4)  They must have all ordered sushi.
(5)  They all must have ordered sushi.

What would make (3) ungrammatical?

It seems that the easiest way to approach this mystery is to suppose that in (3-5), they all starts out in that order, with all following they, and then either they all is moved as a unit, or they is moved, leaving all behind. This requires stipulating that when all has a pronoun complement, the complement comes on the left rather than the right (where complements are everywhere else generally after the head in English). Not great, but it does at least provide an easy explanation of why (3) is no good.

That solves this particular problem, but there are still some mysteries left. For example, the following seem to be possible as well:

(6)  They must all have ordered sushi.
(7)  The students must all have ordered sushi.
(8)  The students all must have ordered sushi.

What is happening here? How did all get between must and have? How did all come to follow the students in (8)? What does this suggest is happening?

Well, a stranded all is supposed to be evidence that the subject was once in that position, and then the complement of all moved away. If we believe this, then (6-7) seem to be telling us that the subject was once between must (M) and have (Perf). Although you won’t be responsible for this particular tweak to the system (you don’t need to consider it beyond the confines of this blog entry), it looks like we have evidence that the subject doesn’t just move to SpecTP—rather, it moves through the specifier of each auxiliary as well. That is, we would need to assume that Perf, Prog, and M each have a [uN*] feature just like T does, and the subject moves into each specifier in turn, on its way to the final destination in SpecTP, checking each of strong features along the way. In (6), they all moves to the specifier of Perf (have) and then they moves alone first to SpecMP and then to SpecTP. In (7), all the students follows the same path. In (8), all the students moves first to SpecPerfP, then to SpecMP, and finally the students is moved alone up to SpecTP.

To repeat: You don’t need to draw such movement in your trees, just stick with the one movement to SpecTP. That’s because we’re going to disregard this evidence just to keep the trees simpler. But if you take this evidence seriously (and we probably should, except that the book doesn’t get into this), we seem to have evidence that the subject “stops off” in the specifier of each auxiliary and modal on the way up. So, in the “real” theory in the end, we probably want to include these intermediate steps.

Incidentally, if we accept that the subject stops off in the specifier of each auxiliary along the way, then we could replace the previous stipulation (about they all starting as they all and not all they) with one that says that if all has a pronoun complement and it in SpecMP, then only the pronoun can move further to SpecTP.

You know, I assigned this problem last year, and I thought I’d dodged the problem this time around, but I hadn’t. There is an issue that comes up when you try to draw the tree for ditransitive verbs now that we have case to worry about.

Ditransitive verbs come in two flavors. One has (an Agent,) a Theme and a Goal, and sounds like Chris gave books to me. The other has (an Agent), a Theme and a Possessee and sounds like Chris gave me books.

In either case, we have three NPs that need to have their [ucase:] feature valued and checked (since all NPs have such a feature). So far, we have discussed in class how T can value and check such features (by virtue of its [ucase:nom] feature) and how v can (by virtue of its [ucase:acc] feature).

In the first kind of ditransitive verb, the Goal consists of a preposition (to) and an NP. And, in fact, in general, we have NPs as complements to P. The conclusion, then, is that prepositions can also value and check case features. That is, P comes with a [ucase:acc] feature, just like v does. If you assume that, there’s no real problem anymore with ditransitives of the first type.

Ditransitives of the second type, however, do not have such a preposition, so we’re still in need of a third head in the structure to carry a feature that can check the case on one of the objects. Our options are pretty limited, but here’s what we can assume: In the type of ditransitive that has a Possessee (the type that does not have the preposition to), the V itself carries a [ucase:acc] feature. In this way, the V will check the case of the Possessee (valuing it as accusative), the v will check the case of the Theme (valuing it as accusative), and the T will check the case of the Agent (valuing it as nominative).

Something i just wrote in the preceding entry reminds me of something that I wanted to say about the terminology of Merge, Move, head-adjunction, etc. As I was grading the midterm, it became clear to me that there was sometimes some confusion about what is meant when I say things like Merge, Merging, Merged, etc. Let me say a couple of words about that.

Generally, I will reserve the word Merge (and its derivatives) for use in referring to the operation Merge, the one that takes two syntactic objects and puts them together into one object whose features are projected from one of the component objects. The one that only happens when it has to, either to check a strong uninterpretable feature or to satisfy the Hierarchy of Projections. So, if I ever talk/write about Merging, that’s what I mean.

When you move a head up to another head—for example, moving V up to v or moving Perf up to T—the head that is moving is adjoined to the other head. There is a sense in which, in the way the word merge is used out in the world beyond this class, one might think that V moves up to v and “merges” with it—but because we have defined the operation Merge in a specific way, you should not ever think of head-adjunction as a “merging” process, because that will only cause confusion. It’s movement, it’s adjunction, it creates a complex head that is perhaps in some sense a fusion of the two component heads, but it is not a Merging of those two heads.

I hope that’s now at least a little bit clearer than it was before. In sum: Merge means Merge. Head-movement is not Merging.

I was just reminded over email about a slide in the last handout (where we covered case checking) where, in the tree for the sentence “She likes them”, the verb likes was drawn with just a [V] feature indicating that it is a verb, but without the [uN*] feature that usually accompanies it on transitive verbs.

Take no notice of the fact that the [uN*] feature is missing. In general, I may sometimes omit features that aren’t directly relevant to the point at hand, but you shouldn’t take this as an affirmative claim that in that sentence the verb lacks a [uN*] feature (to c-select the Theme), it’s just that I only wrote down the features that seemed relevant at the time. If written in full detail, it would have had the [uN*] feature as usual, since that was the reason that the Theme NP was Merged with the verb in the first place.

The Boston University Conference on Language Development is almost upon us, so I wanted to outline the extra credit opportunity that is available, and make some suggestions about talks/posters that might relate to things we’ve been talking about in the class.

Extra credit. If you go to the BUCLD (which you can do for free if you volunteer a small amount of your time), write a short (1 page) summary each for two different talks, outlining what you understood the talk to be about (what they proposed, what data they looked at, what conclusions they drew). I’ll take that and substitute it for your second-lowest homework grade (your lowest one will be dropped anyway).

Recommendations. There are a number of talks that relate to the acquisition of syntax, although in several of them, they will be assuming a bit more background knowledge than we’ll have gotten by then. The talks below are the ones that I think are most likely to be both mostly-comprehensible given what we’ve talked about and about syntax. I’ll try to say a couple of words in class about these right before the conference, too (to tell you briefly what “root infinitives” are, what some of the issues are with question formation, passives, phases, SLI, etc.).

Friday, Nov. 4

• 9:30am. Brun, Babyonyshev. Aspectual properties of root infinitive verbs in child Russian
• 10:00am. Rus, Chandra. Child language imperatives: Questioning the “imperative as an RI-analogue” hypothesis
• 11:45am. Viau. Give = CAUSE + HAVE/GO: Evidence for early lexical decomposition of dative verbs in English child corpora

Saturday Nov 5

• 11:15am. Grebenyova. Multiple interrogatives in child language
• 2:30pm. Babyonyshev, Hart, Gigorenko. The acquisition of passives by Russian-speaking children with SLI
• 3:00pm. Perovic, Wexler. New data on passives in Williams syndrome: Evidence for a grammatical delay
• 4:30pm. Gavarró, Torrens. Participle agreement in Catalan and Spanish and some of its implications
• 5:00pm. Hyams, Snyder. Reflexive clitics and the Universal Phase Requirement

Sunday, Nov 6.

• 9:30am. Hirsch, Wexler. By the way, children don’t know “by”
• 11:00am. Schulz. Evidence for wh-scope-marking in advanced Japanese-English interlanguage grammars

I’ve put the midterm scores on the Courseinfo site. They are out of 50 points. Although these are numbers that will be fed into formulas at the end of the semester, if I had to assign grades based on these numbers alone right now, I would assign them like this: 45-50 A, 41-44 A−, 36-40 B+, 31-35 B, 26-30 B−, 21-25 C.

Overall, it went pretty well I think. There were a couple of questions on the test that were a bit confusing in what was being asked. For example, the one about what “ensures satisfaction of Principle A.” I compensated by making that question worth only a point. Also, on the last question, I awarded a lot of partial credit.

You’ll get your tests back with a key tomorrow.

The grader for this class just stopped by and gave me the homeworks 5, so if you happen to read this soon and want to stop by my office (4th floor, 718 Comm Ave) to pick it up, you’re welcome to. I should be here for a while longer today, and I should be back by a bit before 10am tomorrow.

A concise summary of this part of the discussion in the review class on Thursday:

Unergatives and unaccusatives are intransitive verbs, that is, verbs that require only one argument. They differ in which θ-role they assign to their one argument.

Unaccusatives are verbs that have only a Theme. Examples are melt, sink, fall.

Unergatives are verbs that have only an Agent. Examples are dance, hop, jog.

The tree structure for an unergative looks a bit funny due to the fact that a verb with no arguments except Agent will be both a head and an VP in the tree (simultaneously). The V will be a head (thus eligible to head-move up to v) because we just took it out of the lexicon. But since it is “finished” (having no strong uninterpretable features to check), it will also be a VP. A sentence like I jog with an unergative vP would look like this:

      TP
     /  \
    NP   T′
    I   /  \
       T    vP
           /  \
         <NP>  v′
              /  \
             v    <VP>
            / \
           V   v
         jog

It’s a little bit weird-looking because the V jog moves as a head to head-adjoin to v, but since it doesn’t project (in the complement of v), it is also a VP, and in order to keep the notation consistent elsewhere, complements are written as XPs.