《转换生成语法导论》10
Chapter 10: Movement Theory
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Move α
In the course of this book we have identified a number of transformations responsible for the derivation of various kinds of construction. So far, the list includes Topicalisation, Wh-movement, DP-movement, Extraposition, Heavy NP-shift, Quantifier Raising, I-lowering, V-raising, I-raising, Do-support, as well as the transformation which raises the uppercase verb to v in VP-shell structures.
Apart from Do-support, all these transformations have in common the property of moving a category from one position to another in the phrase marker. This fact suggests the possibility of reducing all these transformations to a single, general process which performs all the operations previously performed by individual transformations. The general process is called Move α, where α is a variable which ranges over all categories.
1. Move α
Move any category anywhere.
To accommodate Do-support, as well as some deletion operations we will discuss in this chapter, Lasnik and Saito (1984) have suggested an even more general version of (1) called Affect α, where ‘affect’ ranges over the operations of movement, insertion and deletion.
An important fact we (are supposed to) have learned from our discussions in the previous chapters is that generally only certain categories move from certain positions to certain other positions over a certain distance.
Thus, we need to impose appropriate conditions on Move α to prevent it from overgenerating. The task involved here is similar in principle to the task carried out in Chapter 6 in relation to the replacement of PS rules with X-bar schema.
This move is in keeping with the general attempt to replace construction-specific rules with general principles and conditions on representations.
For example, in Chapter 6 we saw that the Structure Preserving Hypothesis (SPH) has the effect of forcing head categories to move to head positions and maximal projections to maximal (Spec) positions. The consequence is that (5) is excluded, while (4) is allowed:
4a. Which car will John fix?
5a. *Will which car John fix?
The Spec-Head Agreement requirement (Chapter 6) also imposes stringent conditions on the positions which can be targeted by wh-movement. The latter can only target the Spec position of a CP marked with the feature [+Q]. This has the effect of excluding (7) while allowing (6):
6a. John wonders who Mary saw.
7a. *John believes who Mary saw.
Wh-phrases move to Spec, CP in simple wh-questions to satisfy the requirement we called the [+Q]-CP Principle. On the other hand, quantified phrases and wh-phrases in-situ move at LF for scope reasons.
It is possible that some of these constraints belong to the Movement theory module itself, and function as conditions on the application of Move α or on the representations derived by Move α. In this chapter we will discuss one major principle called Subjacency which functions as a condition on Move α and another major condition called the Empty Category Principle (ECP) which serves as a constraint on representations derived by Move α.
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Bounding theory: Subjacency
However, the similarities between the conditions on wh-movement discussed are sufficiently strong to motivate an attempt to reduce them all to a single underlying condition. Chomsky (1973) argues that they are indeed reducible to a single condition which he calls Subjacency. The definition of Subjacency we will adopt here is stated in (12):
12. Subjacency
Movement cannot cross more than one bounding node in a single step, where bounding nodes are IP and DP
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Empty Category Principle (ECP)
(18) and (19) illustrates the general fact that objects are much easier to extract than subjects, at least in languages such as English. This phenomenon is known as subject-object asymmetry with respect to extraction.
(20) shows that the object can be successfully extracted out of the embedded clause irrespective of whether the complementizer that is present or missing. (21), on the other hand, shows that a subject can only be successfully extracted out of the embedded clause if the complementizer that is missing. The phenomenon illustrated in (21c) is known as the that-trace effect, referring to the sequence that-t at the beginning of the embedded clause.
One such major difference relates to the fact that objects of verbs are usually governed by a lexical category, namely the verb, whereas subject of finite clauses are governed by a non-lexical category, namely I. Let us assume that government by a lexical catgory is a ‘stronger’ form of government. Let us then call the ‘stronger’ form of government proper government, and define it as in (22):
22. Proper Government
α properly governs β iff α governs β, and α is a lexical category.
23. Empty Category Principle (ECP)
Non-pronominal empty categories must be properly governed.
28. Antecedent-government
α antecedent-governs β iff:
i) α and β are co-indexed
ii) α c-command β
iii) α is not separated from β by a barrier.
29. Proper Government
α properly governs β iff:
i) α governs β and α is a lexical category OR
ii) α antecedent-governs β.
This contrast is sometimes said to show that adjunct-extraction does not give rise to the that-trace effect.
The trace involved is an adjunct-trace and therefore does not have to satisfy ECP until LF.
Lasnik and Saito suggest that at the LF level the complementizer that deletes (Affect α) on the grounds that it is semantically empty and does not play a role in interpretation.
39. θ-Government
α θ-governs β iff α is an X0 category that θ-marks β.
40. Proper Government
α properly governs β iff:
i) α θ-governs β OR
ii) α antecedent-governs β.
Let us assume as a working hypothesis that an antecedent is ‘too far away’ if it is separated from its trace by another wh-phrase.
Thus, raising constructions also involve a process of CP-reduction affecting the embedded non-finite clause, as with ECM constructions (Chapter 8).
46a. Who saw what?
47a. *What did who see?
The contrast between the two examples is a further illustration of the subject-object asymmetry with repect to extraction, this time holding at the LF level. An object wh-phrase can move to a Spec, CP already filled with another wh-phrase (46), whereas a subject wh-phrase cannot (47). This phenomenon is called superiority (effects).
48a. Why did you say what?
49a. *What did you say why?
53a. *Who disappeared why?
54a. *Why did who disappear?
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Unifying Subjacency and government: Barriers framework
Subjacency and ECP were said above to be two separate conditions. The former is a condition on movement, and the latter a condition on the representation of traces. The difference between them is reflected in their respective definitions. The definition of Subjacency does not make reference to government, whereas the definition of ECP relies crucially on the notion of government: proper government is a ‘stronger’ form of government.
It is perhaps desirable to have a unified approach to government (ECP) and movement (Subjacency), such that Subjacency would make reference to the same structural relations as government. One such approach is outlined in Chomsky (1986b) and is often called the Barriers framework.
57. L-marking
α L-marks β if α θ-governs β.
Let us now assume further that maximal projections which are L-marked are never barriers.
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Unifying antecedent-government and binding
N/A
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Move α
In the course of this book we have identified a number of transformations responsible for the derivation of various kinds of construction. So far, the list includes Topicalisation, Wh-movement, DP-movement, Extraposition, Heavy NP-shift, Quantifier Raising, I-lowering, V-raising, I-raising, Do-support, as well as the transformation which raises the uppercase verb to v in VP-shell structures.
Apart from Do-support, all these transformations have in common the property of moving a category from one position to another in the phrase marker. This fact suggests the possibility of reducing all these transformations to a single, general process which performs all the operations previously performed by individual transformations. The general process is called Move α, where α is a variable which ranges over all categories.
1. Move α
Move any category anywhere.
To accommodate Do-support, as well as some deletion operations we will discuss in this chapter, Lasnik and Saito (1984) have suggested an even more general version of (1) called Affect α, where ‘affect’ ranges over the operations of movement, insertion and deletion.
An important fact we (are supposed to) have learned from our discussions in the previous chapters is that generally only certain categories move from certain positions to certain other positions over a certain distance.
Thus, we need to impose appropriate conditions on Move α to prevent it from overgenerating. The task involved here is similar in principle to the task carried out in Chapter 6 in relation to the replacement of PS rules with X-bar schema.
This move is in keeping with the general attempt to replace construction-specific rules with general principles and conditions on representations.
For example, in Chapter 6 we saw that the Structure Preserving Hypothesis (SPH) has the effect of forcing head categories to move to head positions and maximal projections to maximal (Spec) positions. The consequence is that (5) is excluded, while (4) is allowed:
4a. Which car will John fix?
5a. *Will which car John fix?
The Spec-Head Agreement requirement (Chapter 6) also imposes stringent conditions on the positions which can be targeted by wh-movement. The latter can only target the Spec position of a CP marked with the feature [+Q]. This has the effect of excluding (7) while allowing (6):
6a. John wonders who Mary saw.
7a. *John believes who Mary saw.
Wh-phrases move to Spec, CP in simple wh-questions to satisfy the requirement we called the [+Q]-CP Principle. On the other hand, quantified phrases and wh-phrases in-situ move at LF for scope reasons.
It is possible that some of these constraints belong to the Movement theory module itself, and function as conditions on the application of Move α or on the representations derived by Move α. In this chapter we will discuss one major principle called Subjacency which functions as a condition on Move α and another major condition called the Empty Category Principle (ECP) which serves as a constraint on representations derived by Move α.
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Bounding theory: Subjacency
However, the similarities between the conditions on wh-movement discussed are sufficiently strong to motivate an attempt to reduce them all to a single underlying condition. Chomsky (1973) argues that they are indeed reducible to a single condition which he calls Subjacency. The definition of Subjacency we will adopt here is stated in (12):
12. Subjacency
Movement cannot cross more than one bounding node in a single step, where bounding nodes are IP and DP
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Empty Category Principle (ECP)
(18) and (19) illustrates the general fact that objects are much easier to extract than subjects, at least in languages such as English. This phenomenon is known as subject-object asymmetry with respect to extraction.
(20) shows that the object can be successfully extracted out of the embedded clause irrespective of whether the complementizer that is present or missing. (21), on the other hand, shows that a subject can only be successfully extracted out of the embedded clause if the complementizer that is missing. The phenomenon illustrated in (21c) is known as the that-trace effect, referring to the sequence that-t at the beginning of the embedded clause.
One such major difference relates to the fact that objects of verbs are usually governed by a lexical category, namely the verb, whereas subject of finite clauses are governed by a non-lexical category, namely I. Let us assume that government by a lexical catgory is a ‘stronger’ form of government. Let us then call the ‘stronger’ form of government proper government, and define it as in (22):
22. Proper Government
α properly governs β iff α governs β, and α is a lexical category.
23. Empty Category Principle (ECP)
Non-pronominal empty categories must be properly governed.
28. Antecedent-government
α antecedent-governs β iff:
i) α and β are co-indexed
ii) α c-command β
iii) α is not separated from β by a barrier.
29. Proper Government
α properly governs β iff:
i) α governs β and α is a lexical category OR
ii) α antecedent-governs β.
This contrast is sometimes said to show that adjunct-extraction does not give rise to the that-trace effect.
The trace involved is an adjunct-trace and therefore does not have to satisfy ECP until LF.
Lasnik and Saito suggest that at the LF level the complementizer that deletes (Affect α) on the grounds that it is semantically empty and does not play a role in interpretation.
39. θ-Government
α θ-governs β iff α is an X0 category that θ-marks β.
40. Proper Government
α properly governs β iff:
i) α θ-governs β OR
ii) α antecedent-governs β.
Let us assume as a working hypothesis that an antecedent is ‘too far away’ if it is separated from its trace by another wh-phrase.
Thus, raising constructions also involve a process of CP-reduction affecting the embedded non-finite clause, as with ECM constructions (Chapter 8).
46a. Who saw what?
47a. *What did who see?
The contrast between the two examples is a further illustration of the subject-object asymmetry with repect to extraction, this time holding at the LF level. An object wh-phrase can move to a Spec, CP already filled with another wh-phrase (46), whereas a subject wh-phrase cannot (47). This phenomenon is called superiority (effects).
48a. Why did you say what?
49a. *What did you say why?
53a. *Who disappeared why?
54a. *Why did who disappear?
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Unifying Subjacency and government: Barriers framework
Subjacency and ECP were said above to be two separate conditions. The former is a condition on movement, and the latter a condition on the representation of traces. The difference between them is reflected in their respective definitions. The definition of Subjacency does not make reference to government, whereas the definition of ECP relies crucially on the notion of government: proper government is a ‘stronger’ form of government.
It is perhaps desirable to have a unified approach to government (ECP) and movement (Subjacency), such that Subjacency would make reference to the same structural relations as government. One such approach is outlined in Chomsky (1986b) and is often called the Barriers framework.
57. L-marking
α L-marks β if α θ-governs β.
Let us now assume further that maximal projections which are L-marked are never barriers.
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
Unifying antecedent-government and binding
N/A
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
