Linguistics 550
6. Subject movement

Over the past few weeks, we have been exploring the syntactic structure of sentences and their constituent phrases. In the course of our investigation, we have encountered evidence that subjects of sentences originate in the V projection and move from there to a structurally higher position in the I projection. But although we know that subjects move, we haven't yet raised the question of why they do so. These notes discuss two possible reasons for subject movement that have been proposed in the literature: case assignment and predication.

Prolegomena

In Notes 5, we argued that phrase structure nodes are not syntactic primitives, but combinations of syntactic features such as syntactic category and shell level. We began by treating projection level as a primitive syntactic feature, but then showed that a node's projection level can be read off of its syntactic environment.

In principle, then, any phrase structure node can be represented by two values for the features syntactic category and shell level, similar to a point in a Cartesian coordinate space. In practice, however, such representations are difficult to process, much like the representation in (4) of Notes 5. For expository reasons, therefore, we follow standard practice and continue to identify nodes using the conventional node labels that conflate syntactic category and shell level and that redundantly express projection level (cf. Chomsky 1995a).

What forces subjects to move?

Recall our analysis of sentences like (1) (= (27) of Notes 4).

(1) a.   All the children should find their toys.
b.   The children should all find their toys.

In (1b), the subject all the children originates in Spec(VP), as shown in (2a), and the DP the children moves to Spec(IP), stranding the quantifier all, as shown in (2b).

Note: For expository convenience, we assume a supercategory Q/DP, which can be filled by either QP or DP. In terms of the features introduced in Notes 5, this supercategory has category n, projection level 2 and shell level greater or equal to 1.

(2) a.       b.  

For the sake of uniformity, we assume that subject movement takes place in all sentences, even in ones in which there is no direct evidence for it. (1a), the pied-piped variant of (1b), then has a derivation analogous to (2). Both sentences in (1) share the derivational stage in (2a), but what moves in (1a) is the entire QP in Spec(VP), not just the DP. The resulting structure is shown in (3).

(3)    

Although we have argued on the strength of sentences like (1b) that subjects move, we have not so far raised the question of why they do so. Syntacticians have taken two approaches to answering this question, which we discuss in turn. The first relies on the concept of grammatical case, whereas the second appeals to a requirement that sentences have subjects.

Case theory

Consider the contrast between (4) and (5).

(4) a. ok They will help her.
b. ok She will help them.
(5) a. * Them will help she.
b. * Her will help they.

Why are the sentences in (5) ungrammatical? The answer is that subjects and objects in English are each required to appear in a specific case form: subjects in the so-called nominative and objects in the so-called oblique. The nominative forms relevant in (4) and (5) are she and they, whereas the oblique ones are her and them. The sentences in (5) are ungrammatical because both the subjects and objects are in the wrong form (the subject in the oblique, and the object in the nominative).

By cross-linguistic standards, English has a rather limited range of case forms: the nominative and oblique just discussed, and arguably also the possessive. Moreover, in English, the nominative and oblique are overtly distinguished only in pronouns, whereas in full noun phrases, the two cases are homonymous, as shown in (6).

(6) a. ok The neighbors will help my friend.
b. ok My friend will help the neighbors.

Nevertheless, for the sake of uniformity, full noun phrases are assumed to bear case as well, even though they don't wear it on their sleeve in English. In many other languages, however, case on full noun phrases is as visible as it is on pronouns. This is illustrated for German in (7), where the case form for both types of noun phrases depends on whether they are subjects or objects.

Note: The case in which these particular German objects appear is called the dative. German objects can also appear in the so-called accusative and (rarely) in the genitive. As in English, subjects appear in the nominative. English used to have a distinction between dative and accusative as well, but the distinction was lost during the Middle English period (1100-1300), leaving only a single case, the modern oblique.

(7) a. Die Nachbarn / Sie werden meinem Freund / ihm helfen.
the neighbors-NOM they-NOM will my friend-DAT him-DAT help
'The neighbors/they will help my friend/him.'
b. * Den Nachbarn / Ihnen werden mein Freund / er helfen.
the neighbors-DAT they-DAT will my friend-NOM him-NOM help
Intended meaning: as in (7a)

In English as well as in other languages, nominative and oblique case have been argued to be assigned under distinct syntactic conditions. As we will see, this has consequences for subject movement.

Government

In traditional grammar, government refers to a syntactic head's ability to impose formal requirements on its complements, particularly what case form the complement must appear in. You may be familiar with this sense of the term from studying languages with numerous case forms, such as German, Latin or Russian. In these languages, different verbs and prepositions require their complements to appear in the dative, the accusative, or (in Latin) the ablative. Sometimes, general rules can help one learn which case goes with which verb or preposition, but mostly, each lexical item has idiosyncratic government requirements, which must simply be learned by heart.

Generative grammar has adopted the concept of government from traditional grammar, but redefined it, as might be expected, in structural terms. Several closely related definitions of government have been proposed in the syntactic literature. We adopt one according to which heads govern their complements, as spelled out in (8).

(8)   A governs B iff
i. A is a head,
ii. B is a maximal projection, and
iii. A and B are sisters (= mutually c-command each other).

Note: In contrast to mutual c-command, government is not a symmetric relationship.

In English, oblique case assignment is a proper subcase of government. This is because only V and P have the ability to assign oblique case in English. (In many other languages, oblique case can also be assigned by the remaining lexical categories, A and N.) It is useful to think of case assignment as the transmission of a piece of information from a case assigner to a case recipient. The relevant piece of information (an oblique case feature) starts out on a V or P. Since oblique case is assigned under government, the feature is transmitted to the case assigner's DP complement. The case feature is then passed down from the DP to the other members of the complement's spine (that is, to the D' and the D).

We will assume further that a D with a particular case feature is restricted to taking an NP complement with the same case feature. That is, the NP agrees in case with the D. As before, the case feature is then again passed down the spine of the NP, with the result that the head of NP, N, agrees in case with the head of the original DP complement, D. This assumption concerning case agreement is motivated by examples like (9a) from German, where accusative case is realized on both the determiner and the noun of the object; cf. the distinct nominative forms in (9b) and the ungrammatical mixed-case noun phrases in (9c,d).

(9) a.   Wir kennen dies-en Hün-en.
we know this-ACC giant-ACC
'We know this giant.'
   b.   Dies-er Hün-e ist riesig.
this-NOM giant-NOM is huge
'This giant is huge.'
   c. * Wir kennen dies-en Hün-e / dies-er Hün-en.
we know this-ACC giant-NOM this-NOM giant-ACC
Intended meaning: as in (9a)
   d. * Dies-er Hün-en / Dies-en Hün-e ist riesig.
this-NOM giant-ACC this-ACC giant-NOM is huge
Intended meaning: as in (9b)

Specifier-head agreement

In contrast to oblique case, nominative case in English is not assigned by V. We can see this by comparing case assignment in two semantically parallel subordinate clauses, one finite and the other nonfinite.

(10) a. We promised him [ that we would go ] .
b. * We promised him [ we to go ] .

Note: We discuss the grammatical equivalent of (10b), We promised him to go, later on in the course.

If nominative case were assigned by go, the verb of the subordinate clause, the contrast in (10) would remain mysterious. The logical conclusion is therefore that nominative case is assigned by some category other than the verb. The obvious choice of case assigner, which straightforwardly explains the contrast in (10), is I–specifically, finite I.

In addition to being assigned by a different category than oblique case, nominative case in English is also assigned in a different configuration: under so-called spec-head agreement (that is, by a head to its specifier).

Consequences for subject movement

The differences between nominative and oblique case assignment have subject movement as a straightforward consequence. To see this clearly, let us examine the familiar structure in (2a), repeated here as (11), which shows clausal structure before subject movement.

(11)    

In the structure in (11), the subject has no way of receiving case. It cannot receive case from the verb because it is not governed by it. It is also unable to receive case from Infl, because it is not in the required structural configuration. So the subject cannot receive case. Yet, for the sentence to be derived, the subject must receive case. Luckily, there is a solution to the paradox. Although the subject must receive case, it needn't receive case in Spec(VP). Rather, since Spec(IP) is conveniently empty, nothing stops the subject from moving into it and receiving nominative case there.

Predication theory

An alternative proposal for why the subject moves relies on the idea, familiar since at least Aristotle, that sentences seem to fall naturally into two parts: a first part, which is being talked about, and a second part, in which something is said about the first part. In logic, these two parts of a sentence are called its subject and its predicate, and the predicate is said to be predicated of the subject. Borrowing from philosophy, it has been suggested that Universal Grammar requires all sentences to have a subject and a predicate, and that furthermore, the subject cannot be contained within the predicate, but that at least some part of it be the predicate's sister.

Considered from this perspective, the structure in (11) is unsatisfactory because the subject in Spec(VP) is contained within the predicate (= I'), not outside of it. Again, however, the difficulty is solved by subject movement, which makes the subject into a sister of the predicate.

It has proven difficult to adduce conclusive evidence in favor of either the case-based or the predication-based approach to subject movement. The standard analysis is the one based on on case theory.

Mediated case assignment

Nominative case in Arabic

In certain languages, case can be assigned in a way that combines the two simple forms of case assignment just discussed. Consider, for instance, the way that nominative case is assigned in Arabic. Just as in English, it can be assigned to the subject by finite I under spec-head agreement, as in (12a), which has the structure in (12b).

(12) a.   'il- 'awlaad laazim yi-laa-'u le9ab-hum.
the children should find-pl toys their
'The children should find their toys.'
b.

But in contrast to English, Arabic also allows the word order variant without subject movement in (13a), with the structure in (13b).

(13) a.   laazim 'il- 'awlaad yi-laa-'u le9ab-hum.
should the children find-pl toys their
'The children should find their toys.'
b.

Given our discussion of case assignment so far, the grammaticality of (13) is a puzzle, because the case assigner (finite I) and the case recipient (the subject in Spec(VP)) stand in neither of the two structural relationships discussed earlier for English.

In order to account for the grammaticality of (13), we must begin by assuming that in Arabic, finite I can transmit nominative case not only under spec-head agreement, as in English, but also under government. Let us further pursue the idea that when a case feature is transmitted to a node, it is shared by all the members of that node's spine. Both assumptions together mean that in (13), the nominative case feature is transmitted to the V projection. However, the feature is useless there, since case is a feature of noun phrases. The verb therefore passes on the nominative case feature to its specifier under spec-head agreement.

We will refer to the two-step process just described as mediated case assignment. In ordinary case agreement, a case feature is transmitted directly from a case assigner to that case assigner's own complement or specifier, under government or spec-head agreement, respectively. Mediated case assignment, on the other hand, requires an mediating head in addition to the original case-assigning head, and it involves both of the structural configurations introduced earlier: government and spec-head agreement. The case feature is first transmitted from the original case assigning head to the mediating head under government. It is then transmitted further to the mediating head's specifier under spec-head agreement.

Oblique case in English

Mediated case assignment is not restricted to Arabic, but is found in English as well. Consider (14a), which has the structure in (14b).

(14) a.   I was expecting her.
b.

Here, expecting assigns oblique case under government to its DP complement, where it is shared by all the members of the DP's spine.

Now consider (15a). Here, expect doesn't have quite the same meaning as in (14a). In (14a), it is a person (= an entity) that is expected, whereas in (15a), it is a state of affairs (= a proposition). This strongly suggests that the complement of expect in (15a) is not a noun phrase, but an IP, and that (15a) has the structure in (15b).

(15) a.   I was expecting her to dislike him.
b.

As before, expect transmits oblique case to its complement. But since the complement is an IP rather than a noun phrase, the oblique case must be passed on to the subordinate IP's specifier under spec-head agreement.

In addition to the semantic motivation for the IP complement structure in (15b), there is also a syntactic argument that favors it over conceivable alternatives. This argument is based on the distribution of so-called pleonastic or expletive there, which is restricted to subject position and unable to function as the object of a verb or preposition.

Note: The terms 'pleonastic' and 'expletive' come from Greek pleonazein 'to be excessive' and Latin ex-plere 'to fill out, complete'. The idea is that expletive subjects are superfluous from a semantic point of view and serve only to complete the syntactic structure. The common use of the term 'expletive', referring to obscene or profane language, is derived from the same Latin verb as the term in its grammatical use, but it is not exactly clear how the two meanings are related.

Expletive there differs from ordinary adverbial there in never bearing stress. It also can't be modified by over or right.

(16) a. ok There is a fly in my soup.
b. * I dislike there in my soup.
c. * I dislike flies in my there.

It was noticed early on in generative grammar that there is a contrast in examples like (17).

(17) a. * I was expecting there.
b. ok I was expecting there to be trouble.

The contrast in (17) would be mysterious if there were a complement of expecting in both (a) and (b). The mystery dissolves if there is a subject in (17b), just as it is in the paraphrase in (18).

(18)     I was expecting that there would be trouble.

The contrast in (17), then, provides further evidence for the structure in (15b).

Predication revisited

In this section, we return to the topic of predication. The structures expressing predication that we have considered so far in this course, whether finite, as in (19), or non-finite, as in (20), have been projections of I.
Note: (19a), while perhaps unusual in American English, is perfectly acceptable in British English.

(19) a.   I considered that she was a friend.
b. I expect that he will be off the ship by midnight.
c. They found that he was a jerk.
d. She imagines that she is a helpful person.
(20) a.   I considered her to be a friend.
b. I expect him to be off the ship by midnight.
c. They found him to be a jerk.
d. She imagines herself to be a helpful person.

It is also possible, however, for predication structures to be projections of categories other than I, as shown in (21). The relevant subjects of predication are in boldface, and the relevant predicates are italicized.

(21) a.   I considered her a friend.
b. I expect him off the ship by midnight.
c. They found him a jerk.
d. She imagines herself a helpful person.

Minimal predication structures of the type in (21) are known as small clauses. That is, small clauses are predication structures without an I projection. As the data in (22)-(24) show, predicates that subcategorize for small clauses do not necessarily subcategorize for ordinary clauses, and vice versa.

(22) a. They made him { (into) a star, sing the National Anthem }.
b. * They made him to { be (into) a star, sing the National Anthem }.
c. * They made that he { was (into) a star, sang the National Anthem } .
(23) a. * She realized herself a helpful person.
b. ?* She realized herself to be a helpful person.
c. She realized that she was a helpful person.
(24) a. * They think him a jerk.
b. * They think him to be a jerk.
c. They think that he is a jerk.

A number of syntactic structures are conceivable for small clauses. One hypothesis that immediately comes to mind, for instance, is that small clauses are structurally parallel to full clauses, but that they differ from them in containing silent heads, as illustrated in (25).

(25) a.  
b.

But the idea that small clauses are projections of silent I and V heads is inconsistent with contrasts as in (26) and (27).

(26) a. * I expect him like the National Anthem.
b. * I expect him considerate.
c. * I expect him a jerk.
d. I expect him off the ship by midnight.
(27) a. I made him sing the National Anthem.
b. His work with sick people has made him considerate.
c. His lamentable upbringing has made him a jerk.
d. * I'll make him off the ship by midnight.

For if the small clause complements of expect and make were IP complements headed by a silent I, as in (25b), then the ungrammatical examples in (26) and (27) should be fine. Contrasts as in (26) and (27) have therefore led syntacticians to conclude that small clauses are projections of the syntactic category of the small clause's predicate.

At first glance, it would then seem that the structure of small clauses like (21a), repeated here as (28a), must be as in (28b), with the subject of the small clause occupying a specifier position.

(28) a.   They considered her a friend.
b.

However, this cannot be the structure of sentences like (29a), where the specifier position of the small clause predicate must be occupied by a noun phrase distinct from the small clause subject, and which is sometimes represented as in (29b).

(29) a.   They considered her John's friend.
b.

It is worth noting that the representation in (29b) does not satisfy the schema of classical X' theory (see (5) of Notes 3, which does not provide for a 'super-specifier' position. Related to this is the structural asymmetry in the way the subject is expressed: in small clauses like (29b), the subject must be the sister of a maximal projection, whereas in ordinary clauses, it is the sister of an intermediate projection. But given that bar levels are not syntactic primitives (recall the discussion in Notes 5), both of these apparent conceptual drawbacks are illusory. Instead of assuming a fixed trinity of types of syntactic slots (specifier, complement, adjunct), we will take the view that all phrase structure nodes must be licensed by primitive linguistic relations, which in addition to argumenthood and modification include predication. All subjects of predication will then simply share the property of being licensed by predication, no matter what their configurational relation to the head of their projection.