Week 12 (April 16-20)

Terms and Concepts

The Case filter requires that every NP have a Case. The capital Cis used because we are not necessarily talking about the morphological case. E.g., even the NP the professor has an nominative Case in the English sentence The professor talked about Case.

Intransitive verbs are divided into unergative verbs and unaccusative verbs. The former class has a subject whereas the latter class only has an object. In general, this subject-object distinction is also correlated with the mannter in which an entity participates in the event described by the verb. The subject of an unergative verb is usally an active participant, but the object of an unaccusative verb is typically a passtive participant.

The Case Filter

The pair of examples below are regarded as evidence for the Case filter:
   1.    carefully writing an article
   2.    (the) careful writing of an article
The central question is why of is needed in (2). Given the fact that careful is used to modify writing in this example (in comparison with (1)), it is obvious that writing is a noun. If nouns somehow cannot assign a Case to their object, the data at hand can be explained easily. By hypothesis, an article need a Case in both examples. Used as a verb in (1), writing provides a Case (accusative Case); but the nominal counterpart in (2) can't provide the Case. So either the example crashes (cf. *the careful writing an article) or a means is called for to provide the Case. It is a fact that prepositions can provide a Case for their object, or it would be impossible to say about an article. Furthermore, if a P is used for this purpose in (2), it must be semantically empty, or the meaning of the expression would be altered. of precisely has both properties: it's a preposition and it doesn't mean anything at all. This analysis is neat, but it crucially depends on the claim that every NP needs a Case. Otherwise, there would be no reason to require of.

The same logic explains the presence of of for certain adjectives:
   3.    proud of her achievement
   4.    fond of the children.
For the same reason, Chinese uses prepositions like dui (and sometimes wei) to save NP objects:
   5.    xihuan nage haizi
          like      that  child   
          like that child
   6.    dui nage haizi hen manyi
          P    that  child very satisfied
          satisfied with that child
   7.    fanyi      naben xiaoshuo
          translate that    novel
          translate that novel
   8.    dui naben xiaoshuo de fanyi
          P    that    novel       DE translation
          (the) translation of that novel
Chinese verbs take an NP object directly as expected (cf. (5) and (7)). The adjective in (6), however, must rely on dui to introduce the object. When a verb is used as a noun (as marked out by the morpheme de (8)), again dui is required. In general, adjectives and nouns can't provide a Case for their object, whereas verbs and pre/postpositions can. But this is only a general statement. As we will see later, not every verb can provide the accusative Case.

Unergative vs. Unaccusative

What are called intransitive verbs in fact fall into two classes. In this section, we examine several phenomena across languages that demonstrate this division.

To begin with, consider Italian ne-cliticization. For the purpose of this discussion, let us ignore the obvious fact that the subject of an intransitive verb follows the verb in these examples.
   1.    Sono passate tre settimane.
          have  elapsed three weeks
   2.    Ne        sono passate tre.
          of-them have elapsed three
   3.    Hanno parlato tre persone.
          have  spoken three people
   4.    *Ne      hanno parlato tre.
          of-them have  spoken three
Romance languages use "light" pronouns which we call clitics. The ne in (2) and (4) is such an example. So one may translate (4), for instance, as meaning "three of them have spoken". The question is why ne is possible in (2) but not in (4).

Now let us make two hypotheses. First, that clitics move out of the NP where they are to be interpreted and land somewhere around the I node. The precise location need not concern us here (and is actually up to further investigation). Second, three weeks in (1) is in fact the object of the verb elapse , in contrast to three people in (3-4) which is the true subject of speak. Under this hypothesis, we have the following two structures (details ignored):

Obviously, if anything moves out of the object NP to I (or a location near I), the movement is "upward". In particular, I c-commands the initial site of movement, satisfying the movement condition. On the other hand, moving the same constituent out of the subject NP fails to meet the movement condition, precisely because I does not c-command the subject or anything inside it. Hence, as long as ne in Italian indeed involves movement, (4) is predicted to be impossible.

This distinction inside the group of intransitive verbs is structural in the sense that the NP is placed in different locations in a tree. But it also corresponds, rather neatly, to a semantic distinction. While people do need some degree of active participation in order to speak, time has no active role in its elapse. It simply happens. From this point on, intransitive verbs with a subject are called unergatives and those with an object are unaccusatives.

The second piece of data in support the disctinction can be seen ni examples (1-4) as well. Note that both sono and hanna are glossed as have. In reality, the former is a form of be and the latter is have. In Romance, it is typical that these two auxiliaries are used "selectively" depending on which intransitive verb to use. And quite consistently, the choice marks out the unaccusative/unergative distinction. 

The third construction to consider is English resutlatives:
   5.    Bill painted the barn green.
   6.    Beth watered the tulips flat.
What's characteristic of these examples is that the clause final adjective is "added" to an otherwise standalone sentence. And the added constituent describes the result of painting or watering. Relevant to us is the fact that this result adjective can only be about the final state of the object. For instance, while it is pragmatically possible for Bill to become green from his painting job, that's not what (5) describes. Only the barn became green as the result of painting. The observation is more clearly substantiated by (7):
   7.    Sam danced dizzy.
Again, it is certainly easy to imagine that Sam got dizzy from dancing, but the sentence doesn't mean that at all. This is precisely what we expect -- dizzy as a result cannot be about the subject.

But (8-9) appear like a problem:
   8.    The river froze solid. 
   9.    The cup broke into pieces.
In both examples, the adjective or PP describe the result of freezing and breaking, and they are obviously about the subject of each sentence. The problem disappears if freeze and break are in fact unaccusative verbs which take objects. So at least at some stage of derivation, the river and the cup are in fact in the object position. Since dance is clearly an unergative semantically, it in fact makes perfect sense that resutlatives are possible in (8-9) but not in (7).

This brings up a new question: if the cup in (9) is the object, why does it show up in the subject position of the clause? The current explanation is that unaccusative verbs somehow are incapable of providing the accusative Case to their object. Under the Case filter, then, either the object stays dead or it goes elsewhere to find a Case. As it happens, unaccusative verbs by definition have no subject, so the subject position starts empty. This creates the perfect solution: the object has no Case while the subject position is available. So the object moves to the subject position of the clause to get the nominative Case. In this theory, the more accurate structure of (9) should be (10):
   (10)    The cup broke t into pieces.
where t stands for the trace after the cup moves away. In a way, into pieces really describes the resulting state of t. Since t is intrinsically related to the cup, we get the interpretation that the cup became pieces.

Noun Incorporation (Again)

Now let's see how our knowledge of Case and the unaccusative/unergative distinction affects our understanding of NI.

To begin with, NI provides further evidence for the distinction:
   Southern Tiwa
   1.    I-k'uru-k'euwe-m.
          agr-dipper-old-pres
          The dipper is old.
   2.    We-fan-lur-mi.
          neg-snow-fall-pres
          The snow isn't falling.
   3.   *0-khwien-teurawe-we.
          agr-dog-fun-pres
          Intended reading: The dog is running.
   Onondaga
   4.    *H-ate-tsi?kti-?se:-?.
          agr-refl-louse-drag-asp
          Intended reading: The louse crawls.
Examples in (3) and (4) confirm what we already know: that NI demonstrates the subject-object asymmetry. Since dog and louse are the subject of their respective verbs, NI is expected to be impossible. But then what about (1-2)? Again, we see that these apparent exceptions in fact involve unaccusative verbs, judging from the semantic relations between the nouns and the verbs.

On the one hand, this seems to further confirm the syntactic analysis of NI. Recall that NI is said to result from moving the head N of the NP to the verb. Since unaccusative verbs are shown to have an object, the fact that the head N of the object NP can merge with the verb follows straightforwardly from syntax. However, we also saw from last week's file that syntactic word-formation should be generally prohibited. In particular, we saw that causativization had better be a lexical process. Now let's see what the theory composed of the LCA and the MSM would say about NI.

First of all, there is clearly a semantic relation between N and V of an NI compound (cf. snow-fall). If that relation were expanded in syntax, it would be as follows:

Note that snow-fall is a lexically formed verb (under the LCA). So it cannot be inserted under N. But if it is inserted under V, then it causes the same problem that de-adjectival causatives do. the compound can't stay in V because N would then be created for nothing; nor can it move to N to make the latter "useful" because downward movement fails to satisfy the movement condition. Either way, the structure violates some general requirement of syntax. So according to the MSM, the semantic relation betrween snow and fall in the compond cannot be expanded in syntax, forcing the word be treated like an indecomposable verb. In sum, an NI compound in Mohawk and Southern Tiwa simply must behave like a verb that takes no syntactic object. Needless to say, this is against Baker's syntactic theory of word-formation.

Next we examine the N-Ving compound in English. As noted before, we want such compounds to be lexically formed. But it's equally obvious that there is a semantic relation between N and V here as well. So the MSM requires that we try to expand this relation and let syntax determine whether the expansion is legitimate or not. Using cake-eating as an example, we have the structure below:

Since both cake and eating are nouns, the compound is also a noun. This means that the compound can be inserted in either of the N positions in the tree. If it is inserted in the lower N position, subsequent raising to the upper N is permitted by the movement condition. Does this mean that cake-eating is after all expandable in syntax? Not really. This is so because the MSM requires the structure to satisfy all the syntactic requirements, not just the movement condition or "usefulness" requirement for each node. We also have other UG principles such as the Case filter to worry about. According the Case filter, the NP object in the tree must have a Case. We also saw that nouns cannot assign a Case to its object. This means that in syntax, the upper N cannot give the NP object a Case and the tree is ruled ungrammatical for failing to satisfy the Case filter. Hence, no syntactic expansion of the semantic relation is possible for cake and eating, and the compound must be treated like a single noun, as we want it to be.

It must be pointed out that this conclusion, while keeping the overall account of word-formation consistent, does leave unanswered the question of why there is subject-object asymmetry in word-formation. For Baker, the answer lies in syntax for NI. But we noted that that is not obviously an advantage because we still have English N-Ving compounds unexplained (in addition to the fact that the syntactic theory of word-formation cannot handle Arabic de-adjectival causatives). For us, syntax has nothing to do with the asymmetry when it comes to compounding. Whatever linguistic mechanism is responsible for it in English compounding is responsible for NI in Amerindian languages. While this is no explanation at all, it at least keeps the theory "clean" and simple.