Chapter 7: Transformations (II)


7.0. Introduction

In the preceding chapter we saw a number of sentence types which could not be generated by the PSG we proposed earlier. Instead of revising and further expanding our PSG, we saw that these sentences were in fact better accounted for, not by a further enriched PSG, but by the assumption that they are derived from some abstract underlying (deep) structures by means of transformations. The deep structures may themselves be generated directly by the rules of our PSG. Transformations then operate on the deep P-markers and successively turn the d-structures into appropriate s-structures.

Each transformation operates on an input structure by moving one or more elements within the structure (e.g., Affix Hopping and SAI), or by inserting some lexical material into the structure (e.g., Do Support). In addition to movement and insertion, a transformation may delete certain items from a P-marker, or it may substitute one item for another in a P-marker. These elementary operations--insertion, deletion, substitution, and movement--"affect" initial and intermediate underlying structures in such a way that they sometimes make the surface structures quite remote in appearance from the underlying structures from which they derive.

In this chapter, we shall illustrate two more transformational operations that involve movement, and turn to a few structures which call for the postulation of transformations that delete elements from their input structures.


7.1. Passive Constructions

7.1.1. Problems

Observe the following sentences:

(1) a. Bill was cheated (by his best friend).
b. Illegal aliens were greatly scared (by Governor Wilson).
c. She was bitten (by a dog barking in the Congress).
d. A book has been given to the baby.
e. The book was put on the table.

These sentences are commonly called "passive sentences". These sentences pose two important problems for a grammar that consists of a PSG of the sort we have assumed and a Lexicon where lexical items are specified for their idiosyncratic properties including their PAS, and s-selection and c-selection (subcategorization) requirements. For example, in order to account for the following contrasts:

(2) a. John cheated Bill.
b. *John cheated.
c. *cheated Bill.

(3) a. Wilson greatly scared illegal aliens.
b. *Wilson greatly scared.
c. *scared the aliens.
d. *scared.

(4) a. A dog bit her.
b. *A dog bit.
c. *bit her.
d. *bit.

we must assume that cheat, scare, and write are specified in the Lexicon transitive verbs with the PAS {1, 2}, the c-selection property +[____NP], and the s-selection property {Agent, Patient} (for cheat), {Cause, Experiencer} (for scare), and {Agent, Theme} (for write) respectively. The (b-d) sentences are out because they each violate the PAS requirements of their verbs. In each case the verb is a two-place predicate, and yet only one argument is found in each of the (b-c) sentences, and none in (d). In addition, the (b) and (d) sentences are also ruled out because they do not satisfy the c-selection requirement of the verbs (which require an object complement). Similarly, the following sentences are explained by the assumption that put, hand are three-place predicates each taking {Agent, Theme, Location/Goal} and subcategorizing for +[____NP PP]:

(5) a. John handed a toy to the baby
b. *John handed a boy.
c. *John handed to the baby.
d. *John handed.
e. *handed a toy to the baby.
f. *handed a toy.
g. *handed to the baby.
h. *handed.

(6) a. John put the book on the table
b. *John put the book.
c. *John put on the table.
d. *John put.
e. *put the book on the table.
f. *put the book.
g. *put on the table.
h. *put.

A problem posed by passive sentences like (1) is that all of these sentences actually violate the c-selection (and the PAS) requirements of their verbs, but contrary to our expectations, they are all perfectly well-formed. For example, the c-selection requirement of each verb provides that it take an NP as its complement (object), but in none of (1) is the verb followed by an object NP. Furthermore, although each verb is a two-place or three-place predicate, in (1) each can occur with one argument less--the Agent or Cause need not occur. In other words, it seems that passive sentences do not need to obey the c-selection and PAS requirements of their verbs.

The problem is in fact more serious. It's not just that the passive sentences do not seem to obey the c-selection requirements of their verbs, but also they seem to be required to disobey those requirements. For example, if one inserts an object after the verb in each of (1) (in order to meet the verb's c-selection requirement), ungrammaticality results:

(7) a. *Bill was cheated the boy (by his best friend).
b. *The aliens were greatly scared their relatives (by Governor Wilson).
c. *She was bitten a boy (by a dog barking in the Congress).
d. *A book has been given a toy to the baby.
e. *The book was put the pencil on the table.

In fact, if one uses a verb that does not c-select an object NP to make a passive, the sentence is also ungrammatical.

(8) a. *Bill was cried.
b. *John was lived.
c. *The book was come on the table (by Bill).

In other words, a passive requires the use of a transitive verb which subcategorizes for (c-selects) an NP, but the verb must not take an NP following it. That is, the 2nd problem is that a passive seems to be required to violate the c-selection property of its verb.

A related problem also arises with a given verb's s-selection property. For example, given (9) we know that the verb cheat s-selects an Experiencer as its object, which by definition must be animate:

(9) a. John cheated the woman.
b. *John cheated the blackboard.

Of course, this s-selection property is not met by the passives, because in the passive construction the verb is not followed by an NP at all, hence not by an Experiencer:

(10) The woman was cheated.

In brief, the passive sentences in (1) exhibit properties that appear to contradict the generalizations we have arrived at, based on the behavior of "active sentences," regarding PAS, c-selection and s-selection.


7.1.2. Passivization

How can one get out of this dilemma? A natural solution suggests itself if we assume a model of grammar that incorporates transformational rules. In particular, we can hypothesize, roughly speaking, that passive sentences are derived from their active counterparts by a rule of transformation which moves their object NPs to the subject position, in addition to making other modifications. Thus, the passive (11b) is derived from the active (11a), and (12b) is derived from (12a):

(11) a. John cheated Bill.
b. Bill was cheated (by John).

(12) a. The teacher put the glasses in the drawer.
b. The glasses were put in the drawer (by the teacher).

Let's see how this can be done by comparing the passives with their active "sources." Take (11a-b) for example and see what must be done to an active sentence to derive a passive from it. We can observe that passivization involves the following sequence of operations.

(13) a. The subject of the active sentence, John, is removed from the subject position, and is optionally placed in an adjunct by-phrase (a PP).
b. The object is moved into the vacated subject position.
c. The main verb is turned into an appropriate form of be followed by the verb in past participle form.

These changes are schematically depicted below:


Take a closer look at step 3. First, there is the additional verb was. Second, the Past Tense carried by the active cheated is now shifted to the copula be, resulting in was. Third, the main verb now appears in its past participle form, not its past-tense form. (We can see the last point with an example like The letter was written , but *The letter was wrote.) The shifting of Tense suggests that we should look at the d-structure underlying the active sentence, before Affix Hopping. The d-structure of (11a) is:


We propose that Step 3 of the Passive transformation consists of the operation described in (16c)

(16) Passive Transformation
a. The subject of the active sentence is removed from the subject position, and is optionally placed in an adjunct by-phrase (a PP).

b. The object is moved into the vacated subject position.

c. Insert the passive-voice morpheme, which consists of the sequence be -en, as the rightmost daughter of Aux, immediately before VP.

More formally, the rule can be formulated as:

(17) Passive Transformation


1 2 3 4

SC: 4 2 Pass 3 ¯ (by 1)

We further hypothesize that the Passive morpheme consists of the sequence be followed by the past participle suffix -en :

(18) Pass ----> be -en

Let us apply the Passive Transformation to (15). The result is the following string:

(19) Bill -ed be -en cheat by John.

This structure is subject to Affix Hopping, which turns (19) into (20):

(20) Bill be+ed cheat+en by John.

This will be turned into the surface structure (11b), Bill was cheated by John. Note that, given what we have assumed, it is not exactly correct to say that the passive sentence is derived from the active sentence. More precisely, we should say that passives are derived from the d-structures which underlie the actives, not from the s-structures of active sentences.

The passive (12b) can be derived in a similar way. The d-structure underlying the active (12a) is (21):


If Passivization does not apply, this structure will surface as the active (12a) after Affix Hopping applies. If Passivization does apply, the following intermediate structure will result:

(22) The glasses -ed be -en put on the table (by the teacher).

Affix Hopping will turn this into (23):

(23) The glasses be+en put+en on the table (by the teacher).

This will be morphologically realized as the passive (12b).

What about a passive sentence with no by-phrase, like (1d-e)? We can say that these sentences have active underlying sources whose Agent subjects have no specific reference, i.e., NPs like somebody, something, etc. Thus (1d) has the d-structure (24):


Passivization and Affix Hopping will successively turn it into (25a) and (25b) respectively, resulting in (1d) if the by-phrase is omitted:

(25) a. A toy -ed be -en give to the baby (by somebody).
b. A toy be+ed give+en to the baby (by somebody).
c. A toy was given to the baby (by somebody).

Now you should be able to derive the appropriate passive sentences from more complicated sentences like the following:

(26) a. John has been cheating Bill.
b. Bill must have been explaining the problem to her.
c. Jim has made the claim that Bill should put the idea behind him.

If you start with the correct d-structure P-marker and apply the transformational rules properly, the following correct passive sentences will result:

(27) a. Bill has been being cheated (by John).
b. The problem must have been being explained to her (by Bill).
c. The claim that the idea should be put behind him (by Bill) has been made (by Jim).

In fact, given the rules we have so far proposed, we can also derive yes/no questions from passive sentences:

(28) a. Was John cheated?
b. Were the cars neatly parked in the lot (by Bill)?
c. Has he been convinced that he should try the dish?

As you can verify for yourself, each of these sentences can be derived from its d-structure by applying Passivization, SAI, and Affix Hopping.


7.1.3. Why the Passive Transformation

Let us take stock briefly and ask why we posited the Passivization rule as in (16) or (17). What does this rule buy us? First, it gets us out of the dilemma described above with respect to the lexical properties of verbs, e.g., generalizations about their PAS, s-selection and c-selection properties. Since passive sentences are derived from d-structures underlying their active counterparts, the problems we encountered above do not arrive at all. Recall that the idiosyncratic requirements of a verb are relevant to the process of Lexical Insertion, which inserts appropriate lexical items at the bottom of P-markers which match the PAS, s-selection, and c-selection properties of given heads (e.g., verbs). If the d-structure of Bill was cheated by John is John -ed cheat Bill, under our assumption, the problems of PAS, s-selection and c-selection do not arise. At the level of D-Structure, Bill appears after the verb cheat, readily satisfying the latter's PAS, s-selection, and c-selection requirements. Then Passivization applies to prepose Bill to the subject position. After this, of course, the lexical requirements of cheat are no longer satisfied. Under the transformational hypothesis, then, there is no dilemma. All we need to do is clarify that the PAS, s-selection and c-selection requirements of a lexical item are to be satisfied at D-Structure, by the process of Lexical Insertion, but not necessarily at S-Structure.

Second, the postulation of Passivization enables us to express a significant generalization illustrated by (9) and its corresponding passive forms:

(9) a. John cheated the woman.
b. *John cheated the blackboard.

(29) a. The woman was cheated by John.
b. *The blackboard was cheated by John.

The generalization is that any restriction that holds between the verb and its object in the active sentence, holds between the verb and its subject in the corresponding passive sentence. This generalization very naturally follows because the passive is derived from the active by moving the object to the subject position. If an NP does not qualify as a legitimate object in the underlying structure, it will not be allowed in the object position at d-structure. Since there is no felicitous underlying source of the form represented by (9b), there is also no felicitous surface structure like (29b). The transformational hypothesis captures the generalization that the same contrast holds between (29a, b) as it does between (9a, b).

Third, our hypothesis further enables us to account for a fact about "idiom chunks" in a straightforward way. Consider the following examples:

(30) a. Tabs were kept on Jane Fonda by the FBI.
b. Heed should be paid to this problem.
c. Advantage was taken of John recently.

In English, the expressions keep tabs on, pay heed to, and take advantage of, are idiomatic in the sense that their meanings cannot be calculated on the basis of the meanings of their parts. Furthermore, these expressions are frozen expressions: if a word is replaced, the expression becomes ill-formed or loses its idiomatic meaning: *maintain tabs on, *direct heed to, *obtain advantage of. Therefore, we must list these expressions each as a whole expression, in the frozen form, in the Lexicon, with an appropriate annotation of its specialized meaning. The problem with the passive sentences in (30) is that we do not see these expressions in the form they are listed in the lexicon, and as such we should expect these sentences to be ungrammatical, or grammatical but without the idiomatic meaning that they have. However, if we assume that these sentences are derived from their active counterparts, the problem disappears. At D-Structure the elements tabs, heed, and advantage appear immediately after keep, pay, and take respectively, as part of an idiomatic expression listed in the Lexicon. That is, they can each be inserted from the Lexicon as part of an idiomatic expression. Passivization may apply and move the idiom chunks away, but this does not alter the meaning of the expressions (which were inserted from the Lexicon as idioms).

A fourth and quite important argument for the Passive rule is that it enables us to capture the notion of synonymy in an explicit and insightful way. As speakers we know that passives and their active counterparts mean essentially the same:

(31) a. John cheated Bill.
b. Bill was cheated by John.

(32) a. The FBI kept tabs on Jane Fonda.
b. Tabs were kept on Jane Fonda by the FBI.

In the model of grammar we are assuming, the active and passive sentences are derived from a common d-structure source; but depending on whether the source undergoes Passivization or not, we get very different s-structures.


That is, we characterize the notion of syntactic synonymy as follows: 2 s-structures are synonymous if they are derived from the same d-structure. Our model captures speakers' intuition that passive sentences mean "the same thing" as their active counterparts by saying that they come from the same underlying source. If the passives and actives were to be derived independently by PSRs without a common underlying source, this fact of synonymy would not be captured insightfully.

Note that this notion of d-structure as the common "source" of superficially very different s-structures has a parallel in phonology and morphology. In phonology, the distinction between a phoneme and its allophones is a distinction between abstract representations that correspond to what we think, and concrete realizations that correspond to what we say or hear. For example, the letter p is pronounced in three different ways in the words pie, spy, adapter, and if you have a trained ear you can tell the difference. However, we feel in our minds that they are all "the same" sound. We capture this mental-physical distinction by saying that these are three allophones of the same phoneme. Similarly, even though the ending of the following words are pronounced differently: cats, boys, kisses, we know that at some level they are the "same thing". We capture this "double" property of being both the same and different, by saying that the different endings are three different allomorphs of the same morpheme.

According to our analysis of passives, then, D-Structure determines the conceptual meaning of a sentence, and S-Structure determines its form. Transformations like Passivization, Affix Hopping, Do Support serve to change the form of a P-marker generated at D-Structure, but they do not change their meaning. The hypothesis that transformations preserve meaning was first explicitly put forth by Katz and Postal (1962) and has come to be known as the Katz-Postal Hypothesis. The Katz-Postal Hypothesis requires us to make a small adjustment of our analysis of yes/no questions. We have assumed that yes/no questions are derived from their declarative counterparts by the application of SAI, followed by other rules (Do Support, Affix Hopping). However, we cannot simply say that declaratives are the d-structures of interrogatives, because declaratives and interrogative are not synonymous. In other words, SAI does not preserve meaning. To solve this problem, Katz and Postal proposed that the d-structure of a yes/no question is not identical to that of its declarative, but contains an abstract Q-morpheme which is not present in the d-structure of a declarative. Following this hypothesis, let's assume that the Q morpheme is dominated by the node C(omplementizer). A normal declarative sentence is then an S, but a question is an S' whose Comp contains the Q-morpheme. According to this hypothesis, the d-structure of Did Clinton swallow his own tongue? is something like (34):


The presence of the Q morpheme indicates that this is the deep structure of a question, not that of a declarative sentence, or a complement clause with the complementizer that. Because it is a question (given the presence of Q), SAI has to apply, followed by Do Support and Affix Hopping, resulting in the desired yes-no question.

The postulation of the Q morpheme as a possible instance of Comp requires us to add it to the inventory of Comp. In Chapter 4, we had the following inventory for the category Comp:

(35) Comp ----> that, whether, if, for, which, why, . . . .

We need to add "Q" as a possible member of this list:

(36) Comp ----> that, whether, if, for, which, why, Q, . . .

According to the Katz-Postal hypothesis, we should also assume that the d-structure of a negative sentence is not exactly that of its affirmative counterpart, but something like (37) or (38). (Again a further modification of the PSR1 is required to allow for the generation of Neg.)



A rule of Negative Placement would place the negative element after Tns and another element of Aux, if any, as in (37), or simply after Tns (if there is no more Aux element) as in (38), resulting in the following:

(39) Clinton -ed shall not swallow his own tongue.

(40) Clinton -ed not swallow his own tongue.

(40) is subject to Do Support, because otherwise the affix -ed would be stranded. When both sentences have undergone Affix Hopping, the following sentences are derived:

(41) Clinton should not swallow his own tongue.

(42) Clinton did not swallow his own tongue.


7.2. Raising Constructions

We have seen that there is good reason to assume a rule of Passivization to relate passive sentences and their active counterparts. In this section we shall see that each of the (a) sentences below should be related to its (b) counterpart by a rule of transformation also.

(43) a. John seems to be unhappy.
b. It seems that John is unhappy.

(44) a. The man appears to have witnessed the accident.
b. It appears that the man has witnessed the accident.

(45) a. The Anteaters are likely to lose again.
b. It is likely that the Anteaters will lose again.

(46) a. They happen to have been invited to the same party.
b. It happens that they have been invited to the same party.

(47) a. The answer turns out to be very simple.
b. It turns out that the answer is very simple.

In particular, we shall assume that both members of each pair are derived from a common d-structure. The common d-structure of (43a, b), for example, is (48):


If no transformation other than Affix Hopping applies to this structure, we get sentence (43b). Before Affix Hopping, however, a process of Subject Raising can apply, which does the following 3 things:

(49) Subject Raising
a. Change the Tns (plus Modal, if any) to the tenseless to.
b. Delete the complementizer that.
c. Raise the embedded subject to the matrix subject position, replacing the expletive subject it.

This is schematically shown in the following tree:


The result of applying Subject Raising is (51):

(51) John -es seem to be unhappy.

After Affix Hopping applies, we have the sentence (43a). The other sentences are similarly derived.

Note that the embedded subject that undergoes Subject Raising may originate as an embedded subject as in this case, but it may also have been brought to the embedded subject position by another rule, say Passivization. This is the case of (46a-b), whose common d-structure is (52):

(52) It -es happen that someone -es have -en invite them to the same party.

First, Passivization applies to the embedded clause of (52), giving (53):

(53) It -es happen that they -es have -en be -en invite to the party (by someone).

Subject Raising may now apply, giving:

(54) They -es happen to have -en be-en invite to the same party (by someone).

After Affix Hopping and the optional deletion of by someone, the intermediate structure (54) is turned into the surface sentence (46a). On the other hand, if neither Passivization nor Subject Raising applies to the d-structure (52), Affix Hopping will turn it into the sentence (46b). Finally, it is also possible for Subject Raising to directly apply to the d-structure (52) without first applying Passivization. In this case the embedded subject someone would be raised, giving rise to the following:

(55) Someone seems to have invited them to the same party.

Both the rules Passivization and Subject Raising may be applied more than once. Multiple applications of these rules account for sentences like the following, among others:

(56) John seems to have turned out to be likely to be seen.

The d-structure of this sentence is:

(57) It seems that it -es have -en turn out that it -es be likely that someone -es will see John.

To derive the surface (56), Passivization first applies to the most deeply embedded S, followed by successive-cyclic application of Subject Raising. You can examine how the surface structure can be correctly derived in step-by-step fashion. Since both Passivization and Subject Raising are optional rules (but Affix Hopping is obligatory), the d-structure (57) may also give rise to any of the following synonymous sentences:

(58) a. It seems that it has turned out that it is likely that someone will see John.
b. It seems that it has turned out that someone is likely to see John.
c. It seems that someone has turned out to be likely to see John.
d. Someone seems to have turned out to be likely to see John.
e. It seems that it has turned out that it is likely that John will be seen (by someone).
f. It seems that it has turned out that John is likely to be seen (by someone).
g. It seems that John has turned out to be likely to be seen (by someone).
h. John seems to have turned out to be likely to be seen (by someone).

You can verify for yourself that (58a) is the result of applying neither Passivization nor Subject Raising. (58b-d) are the results of applying Subject Raising once, twice, and 3 times, respectively to the most deeply embedded subject someone, but not Passivization. (58e) is derived from applying Passivization to the most deeply embedded clause, but not Subject Raising. (58f-h) are derived from applying Passivization, and Subject Raising to the subject of the embedded passive sentence John once, twice, and 3 times, respectively.

Note further that Subject Raising can affect both the most deeply embedded subject and any intermediate subject, including the expletive subject it:

(59) a. It seems to have turned out that it is likely that someone will see John.
b. It seems that it has turned out to be likely that someone will see John.
c. It seems to have turned out to be likely that someone will see John.

Before we leave the topic of Subject Raising, let us ask what motivations we have for positing this transformation. There are a number of important motivations we can provide. First, together with other transformations, it enables us to explicitly characterize a host of synonymous sentences (e.g., (56), (58a-h) and (59a-c)) as being derived from the same underlying d-structure. Secondly, it enables us to capture certain linguistically significant generalizations in a simple and straightforward way. One such generalization has to do with the distribution of idiom chunks. The following sentences are sentential idioms in that the meaning of the entire sentence is not the sum of its parts:

(60) The cat was out of the bag.

(61) Great. The shit really hit the fan this time.

Note that these sentences have the idiomatic meaning only when they appear in the form they do above. In particular, be out of the bag must take the cat as its own subject so that the whole sentence can have the meaning that "the secret was revealed." Similarly, hit the fan must take the shit Ýas its subject for the whole sentence to mean that we now have a mess. Even pronoun cannot be used in place of the required subject. Thus, the following sentences are either unacceptable or have only the literal, non-idiomatic meaning:

(62) The cat was afraid that it might be out of the bag.

(63) The shit is so. . . . that it hit the fan.

This generalization about sentential idiom chunks seems to be contradicted by the following examples, however:

(64) a. The cat seems to have been out of the bag.
b. The cat seems to be likely to be out of the bag.
c. The cat seems to be likely to turn out to be out of the bag.

(65) a. The shit seems to hit the fan (every time we plan to....).
b. The shit seems to be likely to hit the fan.
c. The shit is likely to turn out to hit the fan.

In each of these sentences, the cat and the shit occur as the subject of the matrix verb seem or be likely; neither appears as the subject of the predicates be out of the bag or hit the fan. However, all of these sentences have an idiomatic meaning, contradicting the generalization just made about (62)-(63). We are now in a dilemma.

The dilemma disappears, however, under the hypothesis that the sentences in (64) and (65) are derived by Subject Raising from d-structures underlying the following:

(66) a. It seems that the cat has been out of the bag.
b. It seems that it is likely that the cat is out of the bag.
c. It seems that it is likely that it will turn out that the cat is out of the bag.

(67) a. It seems that the shit has hit the fan.
b. It seems that it is likely that the shit will hit the fan.
c. It is likely that it will turn out that the shit will hit the fan.

That is, the cat and the shit originate as the subject of be out of the bag and hit the fan respectively at d-structure, satisfying the lexical requirements of the idiom chunks as specified in the Lexicon. Then Subject Raising applies to displace the subject in various ways, giving the false appearance that the generalization based on (62)-(63) may not be valid.

Another linguistically significant generalization has to do with the range of possible interpretations of a reflexive pronoun. In the following sentences, notice that a reflexive pronoun in the object position of a clause must have as its antecedent an NP within the same clause:

(68) a. John thinks that Mary admires herself.
b. *John thinks that Mary admires himself.
c. They told me that I should nominate myself for the position.
d. *They told me that I should nominate themselves for the position.

The situation is in fact more complex, but for the cases that we shall consider this "clause-mate" condition is entirely valid: reflexive pronouns and their antecedents must be clause-mates. Note that this generalization runs into apparent difficulty when we consider sentences like the following:

(69) a. John seems [ to have admired himself].
b. They are likely [ to nominate themselves for the position].

In (69a), himself takes John as its antecedent, but John is not a clausemate of himself; it is the subject of the matrix verb seem. Similarly, in (69b) we have the reflexive themselves taking the subject of the higher clause as its antecedent. The generalization that we observed with (68) then seems to be invalidated by (69).

Notice, however, that this difficulty disappears once we hypothesize that sentences like (69a-b) involve Subject Raising (whereas those in (68) do not). In particular, under our assumption, (69a-b) are related to the ones in (70):

(70) a. It seems that John has admired himself.
b. It is likely that they will nominates themselves for the position.

In these sentences, the reflexives and their antecedents are clausemates, in accordance with the generalization that we established on the basis of (68).

In other words, the Subject Raising analysis allows us to explain all the facts we have observed so far, with no contradiction whatsoever. What needs to be spelled out explicitly is that the "clausemate condition" represents a valid generalization that applies at D-Structure. The sentences in (69) only appear to have violated this condition because the relevant antecedents have been moved out of their clauses by the Subject Raising rule.


7.3. Equi-NP Deletion

The two transformational rules, Passivization and Subject Raising, are movement transformations. We shall see that other syntactic phenomena in English call for the postulation of transformations that delete elements from the input structures. Compare the sentences in each of the following pairs:

(71) a. John seemed to be happy.
b. John tried to be happy.

(72) a. Mary is likely to win.
b. Mary is eager to win.

The two sentences in each pair look almost alike, the only difference being in the choice of the verb. One might be tempted to conclude, then, that try and eager behave similarly to seem and likely, respectively. This conclusion is not warranted, however, once we realize that their similarly almost stops right here. If we look at the following sentences, it is clear that these verbs behave very differently in other respects:

(73) a. It seems that John is happy.
b. The book seems to be expensive.
c. The shit seems to have hit the fan.

(74) a. *It tries that John is happy.
b. *The book tries to be expensive.
c. *The shit tires to hit the fan.

That is, whereas seem can take an expletive pronoun (i.e., it), an inanimate NP, or an idiom chunk as its subject, try cannot take any such subject. Instead, it requires an animate subject like John, as in (71). The differences between seem and try also show up between likely and eager.

(75) a. It is likely that Mary will win.
b. The book is likely to get expensive.
c. The shit is likely to hit the fan.

(76) a. *It is eager that Mary will win.
b. *The book is eager to get expensive.
c. *The shit is eager to hit the fan.

The verbs in each pair under consideration are, therefore, more different from each other than they are similar. The question is, how do we account for their different behaviors, which seem clearly systematic? An answer suggests itself if we consider what it means for a raising predicate like seem and likely to be able to take an expletive subject. This means that these verbs do not s-select a subject as an argument to which it will assign a thematic role. Such verbs are one-place predicates each with one single argument (a Theme or Event) which occurs in the form of a complement clause. The subject position, not being occupied by an argument, is then filled in by this expletive element, which is there to "hold the place" (until something is moved into it, replacing it, say by Subject Raising). On the other hand, we have seen that predicates like try and eager cannot take an expletive subject; their subjects must always be filled with a lexical subject with substantive content. Given this observation, we can say that predicates like try and eager are two-place predicates, with the PAS {1, 2}, each s-selecting an Agent or Experiencer as its subject and an Action (e.g. to win or to be happy, etc.) as its complement. This immediately explains the differences they display from the raising predicates. An expletive pronoun (without no conceptual content) is not a true argument, so it cannot serve as the subject of try or eager. Since the book is inanimate, it cannot be selected to play the role of an Agent or Experiencer. Finally, because an idiom chunk has a meaning only an part of an idiom phrase, it cannot function meaningfully on its own to receive a thematic role from the verb. Hence an idiom chunk is like an expletive and it cannot fulfill the role an independent thematic-role. (A sentence like The cat is eager to be out of the bag is well-formed only under the literal, non-idiomatic meaning that refers to a real cat.)

But then how do we account for the similarity observed above in connection with (71) and (72)? In particular, we have argued that (71a) and (72a) are derived via the rule of Subject Raising, but given that try and eager each have their own subject arguments, we cannot derive (71b) and (72b) by Subject Raising. How do we obtain these sentences then? We propose that the deep structures underlying these sentences are as follows:

(77): John -ed try [John to be happy]

(78) Mary -es be eager [Mary to win]

Each predicate here s-selects an Action as its complement and c-selects a clause to express that complement. (77) has the meaning that John tried to bring about the event "John is happy" where the two instances of John have "equivalent" reference. (78) means that Mary is eager for herself, Mary, to win. We assume that, in a structure like these, each with two identical NPs, the second NP is erased by a transformation, called Equi-NP Deletion. The result of applying this process to (77) and (78), followed by Affix Hopping, will be the sentences (71b) and (72b), as desired.

The pairs of sentences we saw in (71) and (72) therefore involve two types of predicates: raising predicates and "Equi" predicates. Raising predicates are one-place predicates without a subject argument, while Equi predicates are two-place predicates with an animate subject. From here a number of differences between them follow.


7.4. Summary

In this chapter we have seen that a number of interesting syntactic phenomena receive a natural explanation under the hypothesis that our Grammar includes the transformational processes Passivization, Subject Raising, and Equi-NP Deletion. Again, as we saw in the last chapter, the derivation of a sentence may involve a combination of transformations. Given the assumptions so far, we can now explain another every interesting contrast in English syntax:

(79) a. John seems to have been examined by the doctor.
b. The doctor seems to have examined John.

(80) a. John tried to be examined by the doctor.
b. The doctor tried to examine John.

The interesting contrast is that whereas (79a) and (79b) are synonymous, (80a) and (80b) are not. A transformational grammar incorporating the three rules we introduced in this chapter provides an insightful way to explain this contrast.



Homework 7


(1) For each of the following sentences, indicate what its d-structure is (with a P-marker), and list the transformations that are responsible for turning it into the desired s-structure. Give the transformations in the order they apply and indicate the effect of each transformation by giving its output intermediate (or surface) structure. You can use arrows to show its effects as well.

a. John must have been writing letters.
b. Should Bill be sleeping now?
c. Bill was being cheated.
d. Will the toy have been given to the baby (by Bill)?
e. Did you say that the letter should have been sent to her office?
f. Does John seem to like his new toy?
g. Is he likely to have been cheated by Bill?
h. Was Bill criticized?

(2) (a) Give the d-structure for the sentence It appears that it has happened that Bill has been telling lies. (b) Then, list at least 5 surface sentences that can be derived from the d-structure, using any combination of transformations we have discussed in class.

(3) Turn the following affirmative sentences into negative form (by inserting not at the appropriate place:

a. John will be cheated.
b. John has been cheated.
c. John will have been cheated.
d. John will have been being cheated.
e. John is being cheated.
f. John was cheated.

(b) Consider the two sentences:

a. John was not cheated.
b. Was John cheated?

Consider how they may each be derived from their respective d-structures by transformations. What can these tell us about how the rule Passivization should be ordered with respect to Negative Placement and with respect to SAI?

(4) Can you explain the contrast alluded to at the end of the text? Remember that, in our model of grammar, synonymy is characterized as two different surface structures having derived from the same deep structure. Sentences that are not synonymous do not share the same deep structure. In other words, the sentences in (79) should have the same deep structure, but those in (80) should not. Can you show how each sentence is derived?

(5) All of the following verbs can appear in the form of "NP verb to VP" (John appears to be happy, etc.):

appear, want, sure, willing, hope, prefer, happen, afraid

Using the reasoning used in the text, can you decide, for each item above, whether it is a raising predicate or an Equi predicate?