The flexible nature of verb movement

Author: Olaf Koeneman
LOT Number: 37
ISBN:
Pages: 226
Year: 2000
€32.00
Download this book as a free Open Access fulltext PDF

This thesis looked at differences in verb placement across languages. On the basis of
the data presented in chapter 1, I set out to develop an account of verb movement
parametrization in declarative clauses. In this attempt, I tried to formulate hypotheses
that not only account for the correlations observed but are falsifiable as well. The
central claim is that cross-linguistic differences fall out from the interplay of universal
constraints and language-specific factors that are independently motivated. Hence, the
extent to which the present theory improves over standard analyses in this respect
largely determines its success. This final chapter summarizes the main results and
spells out the most important conceptual and empirical advantages.

The main hypothesis of this thesis is that verb movement is an operation that
takes place in order to project a particular feature of the verb. There is no longer a
prefabricated head position that attracts the verb. Rather, it is the verb itself that takes
the initiative. Output conditions will force the verb to move. They can demand that a
particular feature be visible in a particular structural position that does not correspond
to the position in which the verb is initially inserted. Hence, verb movement becomes
inevitable. Irrespective of one’s formulation of output conditions, this view on
functional structure has a number of conceptual advantages. Let me mention three.
First, since the presence of prefabricated head positions is no longer required
for providing a trigger for verb movement, it becomes possible to drastically reduce
feature redundancy in overt syntax. In standard approaches, a functional head has a
feature make-up that corresponds to similar features on the verb. This entails that these
features are represented twice in the structure. Under the alternative view, they are only
represented once, namely on the verb. This view becomes possible once we allow these
features to project after movement.
Second, violations of Chomsky’s (1995) extension condition are reduced. This
condition requires that movement always extends the tree structure. Movement of an
object to spec-IP, for instance, creates a specifier and thereby extends the root of the
clause. When heads are taken from the lexicon, they are merged with the top node of
the representation already built, never somewhere in between. Verb movement blatantly
violates this condition if analyzed as an operation that adjoins the verb to an empty
head, as standardly assumed. In the alternative conception, verb movement is just as
creating as XP-movement is: The operation extends the root of the clause just like
insertion of head does. Hence, verb movement no longer violates the extension
condition.
Third, the proposal allows a straightforward correspondence between
functional projections and affixes without falling into the same pitfalls as the
Morphological Head Hypothesis, the view according to which affixes are generated in a
VP-external position separate from the verb. In general, I take it that affixes are
generated on the verb in morphology. By hypothesis they can project after verb
movement if circumstances should require that. It is not unexpected then that we find
inflected verbs in both base and dislocated positions. This fact is not entirely expected
under a strong version of the MHH: Under the assumption that affixes are generated
VP-externally, a lowering rule becomes necessary to account for inflected verbs
occupying their base position. Such a rule is conceptually undesirable.2
I concluded from the empirical data presented in chapter 1 that there are two
verb movement operations for which the evidence is robust and for which it is clear that
parametrization is involved, V to I and V to C. With the alternative conception of
functional structure in mind, I designed two triggers for verb movement. Let me repeat
the essentials.
I proposed that V to I movement is an operation triggered by a condition on
rich agreement inflection. Rich inflection, defined in a particular way, is interpretable
and must be associated with the external theta role assigned by VP. For this reason, the
verb moves and projects Agr. The consequence is that Agr is in VP’s predicational
domain and can receive the external theta role, as required. This proposal was able to
account for four facts.
First and foremost, the analysis captures the pervasive correlation between rich
inflection and overt verb movement noted in the literature. Languages with rich
agreement move the verb to a VP-external position and differ in this respect from
languages with poor agreement. It accounts for the fact that this verb movement
1 The only operation that violates the extension condition is associate-to-expletive raising in chapter 4. This,
however, is an operation that takes place after spell out, where according to Chomsky (1995: 327) the condition
does not hold.
2 Recall from chapter 1 that Bobaljik (1995) offers an original and interesting way of accounting for this pattern
without adopting affix lowering. The analysis he offers, however, only works by introducing (i) additional
structure and (ii) readjustment rules. The present proposal works without these additional assumptions.

operation gets lost once the agreement paradigm is significantly eroded.
Second, the definition of rich agreement that determines whether V to I must
take place allows a natural description of the null subject parameter. I hypothesized that
rich agreement comes in two kinds, pronominal and anaphoric, adopting Rizzi’s (1982)
terminology. The distinction is as follows. Anaphoric Agr is rich enough to be
interpretable (and trigger movement) but not rich enough to be interpretable as a
subject on its own. Therefore, a DP must occupy the specifier position of the moved
verb in order to supply missing feature values. This accounts for the fact that, although
Icelandic and Italian both have V to I movement, only the latter has pro drop.
Pronominal Agr can be interpreted as a subject on its own and no overt DP-specifier
need be present.
Third, the idea that in languages with anaphoric agreement Agr must be
specified by a DP further accounts for the ban on expletives in spec-AgrP in languages
as Yiddish, Icelandic and German. Under the assumption that the associate covertly
moves to the expletive, it will be unable to specify Agr in the adjoined position since the
expletive counts as the specifier. As a consequence, Agr fails to obtain missing feature
values and the associate remains without an interpretation.
Fourth, the idea that rich Agr counts as the grammatical subject provides an
explanation for the fact that NP-raising is syntactically optional in languages with
anaphoric or pronominal agreement. It was observed that in OV languages NP-raising
is optional as well, even if they have poor agreement such as Dutch and Afrikaans.
Since languages with rich Agr and OV languages with rich or poor agreement do not
form a natural class, it was argued that a syntactic explanation for the optionality of
NP-raising seems unlikely. Once rich Agr is analyzed as a subject, however, a
straightforward generalization can be formulated: NP-raising is optional in languages
where the 'logical subject' linearly precedes the verb. Williams’ (1997) 'general pattern
of anaphoric dependence' (GPAD) was used to explain why NP-raising is optional in
languages with the aforementioned properties. Irrespective of this explanation, however, it
should be pointed out that the generalization can only be formulated under the assumption
that rich Agr counts as a grammatical subject, just as the theory of V to I hypothesizes.
Let us next consider V to C movement. In chapter 3 I postulated the Tense
condition, requiring that the Tense features anchoring the event expressed by the
proposition in time should have COMMAND over the subject and the predicate. It was
argued that V to C movement is just one way of satisfying this constraint. In verb
second languages the verb moves up to a position higher than the subject and the
predicate and projects Tense features. These then satisfy the Tense condition under ccommand.
I argued that XP-fronting is an operation that takes place for an independent
reason, namely to assign a value to the variable introduced by the implied specifier.
Together, the verb movement and XP-fronting lead to a verb second effect. Given this
analysis of verb second, the question becomes why Romance and English are different.

After all, the Tense condition is taken to apply universally. I argued that in both cases
the lack of verb second is related to a language-specific property that is independently
motivated. English has an empty head that selects a VP. This head, which is marked for
Tense, is motivated by the do-support paradigm. Once it enters into a dependency
relation with Tense features on the verb, the Tense condition is satisfied under mcommand,
making any verb movement unnecessary. In Italian and French one verb
movement projecting Tense feature suffices to bring Agr into VP’s predicational
domain and to satisfy the Tense condition under c-command. The absence of an
independent AgrP follows from the assumption that pronominal Agr can appear as a
subject on its own: No spec-head relation has to be established with a DP-specifier.
Since pronominal Agr can assign a value to the variable introduced by the implied
specifier, XP-fronting is not a necessary operation. Hence, these languages lack
generalized verb second.
Looking at both the Agr and Tense parameter, I conclude that especially the former
is a successful one in the sense that setting it will give the child knowledge about at least four
properties of his/her language. Although no such cluster has been established for the Tense
parameter, it has been shown that in combination with a positive setting for the V to I
parameter the setting of the Tense parameter provides knowledge of two more properties of
the language.
First, the two settings together determine the extent of verb movement taking place
in embedded clauses. It was argued that Icelandic and Yiddish satisfy the Tense condition
under c-command. The verb moves up higher than the subject and predicate. XP-fronting is
triggered for independent reasons. That verb second effects also show up in embedded
clauses is caused by the fact that these languages have rich, anaphoric, agreement and
therefore an independent AgrP. The consequence was that the complementizer cannot be
involved in satisfaction of the Tense condition and verb movement takes place as a last resort
operation. In Mainland Scandinavian, on the other hand, no AgrP has to be projected, given
the poor nature of Agr, and the complementizer is able to satisfy the Tense condition by
entering into a dependency relation with the Tense features on the verb. For this reason,
generalized verb second is absent in embedded clauses.
Second, if in a language AgrP must be generated and the Tense condition is
satisfied under c-command, two projections will dominate VP in declarative clauses. The
consequence of this is that transitive expletive constructions can be generated. This option is
excluded if either of these parameters is set differently. An interesting consequence of the
analysis is that it singles out Dutch as a unique case. In no way does the distribution of er fit
in with the generalizations that can be made about expletives in Germanic. It was shown that
there are independent reasons for not considering this element to be an LF place-holder of
the logical subject. This strengthens the analysis of er as an expletive (i.e. semantically
empty) adverb rather than an expletive subject, a claim already defended in Bennis (1986).
In the introduction I remarked that the view on functional structure adopted in
this thesis and the specific triggers I argued for are logically distinct. The theory of verb
movement could be right but the triggers wrong or vice versa. It should be noted,
however, that it is the interaction of the two that makes a couple of explanations
straightforward. The explanation for the lack of verb second in English as well as the
explanation for the distribution of TECs are directly related to the view on functional
projections that I adopt. In a more standard view holding that every functional
projection is headed by an empty head, it becomes impossible to directly relate the lack
of verb second in English to the presence of an empty head in this language, since
empty heads appear much more unrestrictedly. Likewise, the grammaticality of a TEC
hinges on the number of functional projections available. In the present proposal verb
movement is directly indicative of available structure, so that Vikner’s generalization is
derived in a natural way. Since in standard approaches empty heads can in principle be
postulated without an overt head moving to it, it becomes hard to see what blocks
generation of a TEC. One could of course claim that a projection only becomes active
once a verb has been moved to it: Only where two projections are activated through
movement is generation of a TEC allowed. Note, however, that in English a subject can
occupy the specifier position of a projection with an empty head position. Why this
possibility should not be available in general is therefore not immediately obvious. In
general, one can always introduce extra assumptions but the point is that the
explanation for the lack of verb second in English as well as the explanation for the
distribution of TECs follow without such extra assumptions from the theory of overt verb
movement in this thesis and the constraint it puts on the occurrence of empty heads.
If the triggers for V to I and V to C are correct, the nature of the two verb
movement operations can be successfully unified. Recall from the introduction that the
MHH and AHH both had trouble explaining why V to I is related to a morphological
effect (namely richness of inflection) whereas no such effect can be observed with V to
C. The dichotomy between the category C and I is eliminated by the present analysis
since C is replaced by T. Hence, V to C and V to I movement are related to morphology
to the same extent: Both operations take place to project some feature of the verb and in
both cases this feature correlates with existing morphology.

This thesis looked at differences in verb placement across languages. On the basis of
the data presented in chapter 1, I set out to develop an account of verb movement
parametrization in declarative clauses. In this attempt, I tried to formulate hypotheses
that not only account for the correlations observed but are falsifiable as well. The
central claim is that cross-linguistic differences fall out from the interplay of universal
constraints and language-specific factors that are independently motivated. Hence, the
extent to which the present theory improves over standard analyses in this respect
largely determines its success. This final chapter summarizes the main results and
spells out the most important conceptual and empirical advantages.

The main hypothesis of this thesis is that verb movement is an operation that
takes place in order to project a particular feature of the verb. There is no longer a
prefabricated head position that attracts the verb. Rather, it is the verb itself that takes
the initiative. Output conditions will force the verb to move. They can demand that a
particular feature be visible in a particular structural position that does not correspond
to the position in which the verb is initially inserted. Hence, verb movement becomes
inevitable. Irrespective of one’s formulation of output conditions, this view on
functional structure has a number of conceptual advantages. Let me mention three.
First, since the presence of prefabricated head positions is no longer required
for providing a trigger for verb movement, it becomes possible to drastically reduce
feature redundancy in overt syntax. In standard approaches, a functional head has a
feature make-up that corresponds to similar features on the verb. This entails that these
features are represented twice in the structure. Under the alternative view, they are only
represented once, namely on the verb. This view becomes possible once we allow these
features to project after movement.
Second, violations of Chomsky’s (1995) extension condition are reduced. This
condition requires that movement always extends the tree structure. Movement of an
object to spec-IP, for instance, creates a specifier and thereby extends the root of the
clause. When heads are taken from the lexicon, they are merged with the top node of
the representation already built, never somewhere in between. Verb movement blatantly
violates this condition if analyzed as an operation that adjoins the verb to an empty
head, as standardly assumed. In the alternative conception, verb movement is just as
creating as XP-movement is: The operation extends the root of the clause just like
insertion of head does. Hence, verb movement no longer violates the extension
condition.
Third, the proposal allows a straightforward correspondence between
functional projections and affixes without falling into the same pitfalls as the
Morphological Head Hypothesis, the view according to which affixes are generated in a
VP-external position separate from the verb. In general, I take it that affixes are
generated on the verb in morphology. By hypothesis they can project after verb
movement if circumstances should require that. It is not unexpected then that we find
inflected verbs in both base and dislocated positions. This fact is not entirely expected
under a strong version of the MHH: Under the assumption that affixes are generated
VP-externally, a lowering rule becomes necessary to account for inflected verbs
occupying their base position. Such a rule is conceptually undesirable.2
I concluded from the empirical data presented in chapter 1 that there are two
verb movement operations for which the evidence is robust and for which it is clear that
parametrization is involved, V to I and V to C. With the alternative conception of
functional structure in mind, I designed two triggers for verb movement. Let me repeat
the essentials.
I proposed that V to I movement is an operation triggered by a condition on
rich agreement inflection. Rich inflection, defined in a particular way, is interpretable
and must be associated with the external theta role assigned by VP. For this reason, the
verb moves and projects Agr. The consequence is that Agr is in VP’s predicational
domain and can receive the external theta role, as required. This proposal was able to
account for four facts.
First and foremost, the analysis captures the pervasive correlation between rich
inflection and overt verb movement noted in the literature. Languages with rich
agreement move the verb to a VP-external position and differ in this respect from
languages with poor agreement. It accounts for the fact that this verb movement
1 The only operation that violates the extension condition is associate-to-expletive raising in chapter 4. This,
however, is an operation that takes place after spell out, where according to Chomsky (1995: 327) the condition
does not hold.
2 Recall from chapter 1 that Bobaljik (1995) offers an original and interesting way of accounting for this pattern
without adopting affix lowering. The analysis he offers, however, only works by introducing (i) additional
structure and (ii) readjustment rules. The present proposal works without these additional assumptions.

operation gets lost once the agreement paradigm is significantly eroded.
Second, the definition of rich agreement that determines whether V to I must
take place allows a natural description of the null subject parameter. I hypothesized that
rich agreement comes in two kinds, pronominal and anaphoric, adopting Rizzi’s (1982)
terminology. The distinction is as follows. Anaphoric Agr is rich enough to be
interpretable (and trigger movement) but not rich enough to be interpretable as a
subject on its own. Therefore, a DP must occupy the specifier position of the moved
verb in order to supply missing feature values. This accounts for the fact that, although
Icelandic and Italian both have V to I movement, only the latter has pro drop.
Pronominal Agr can be interpreted as a subject on its own and no overt DP-specifier
need be present.
Third, the idea that in languages with anaphoric agreement Agr must be
specified by a DP further accounts for the ban on expletives in spec-AgrP in languages
as Yiddish, Icelandic and German. Under the assumption that the associate covertly
moves to the expletive, it will be unable to specify Agr in the adjoined position since the
expletive counts as the specifier. As a consequence, Agr fails to obtain missing feature
values and the associate remains without an interpretation.
Fourth, the idea that rich Agr counts as the grammatical subject provides an
explanation for the fact that NP-raising is syntactically optional in languages with
anaphoric or pronominal agreement. It was observed that in OV languages NP-raising
is optional as well, even if they have poor agreement such as Dutch and Afrikaans.
Since languages with rich Agr and OV languages with rich or poor agreement do not
form a natural class, it was argued that a syntactic explanation for the optionality of
NP-raising seems unlikely. Once rich Agr is analyzed as a subject, however, a
straightforward generalization can be formulated: NP-raising is optional in languages
where the 'logical subject' linearly precedes the verb. Williams’ (1997) 'general pattern
of anaphoric dependence' (GPAD) was used to explain why NP-raising is optional in
languages with the aforementioned properties. Irrespective of this explanation, however, it
should be pointed out that the generalization can only be formulated under the assumption
that rich Agr counts as a grammatical subject, just as the theory of V to I hypothesizes.
Let us next consider V to C movement. In chapter 3 I postulated the Tense
condition, requiring that the Tense features anchoring the event expressed by the
proposition in time should have COMMAND over the subject and the predicate. It was
argued that V to C movement is just one way of satisfying this constraint. In verb
second languages the verb moves up to a position higher than the subject and the
predicate and projects Tense features. These then satisfy the Tense condition under ccommand.
I argued that XP-fronting is an operation that takes place for an independent
reason, namely to assign a value to the variable introduced by the implied specifier.
Together, the verb movement and XP-fronting lead to a verb second effect. Given this
analysis of verb second, the question becomes why Romance and English are different.

After all, the Tense condition is taken to apply universally. I argued that in both cases
the lack of verb second is related to a language-specific property that is independently
motivated. English has an empty head that selects a VP. This head, which is marked for
Tense, is motivated by the do-support paradigm. Once it enters into a dependency
relation with Tense features on the verb, the Tense condition is satisfied under mcommand,
making any verb movement unnecessary. In Italian and French one verb
movement projecting Tense feature suffices to bring Agr into VP’s predicational
domain and to satisfy the Tense condition under c-command. The absence of an
independent AgrP follows from the assumption that pronominal Agr can appear as a
subject on its own: No spec-head relation has to be established with a DP-specifier.
Since pronominal Agr can assign a value to the variable introduced by the implied
specifier, XP-fronting is not a necessary operation. Hence, these languages lack
generalized verb second.
Looking at both the Agr and Tense parameter, I conclude that especially the former
is a successful one in the sense that setting it will give the child knowledge about at least four
properties of his/her language. Although no such cluster has been established for the Tense
parameter, it has been shown that in combination with a positive setting for the V to I
parameter the setting of the Tense parameter provides knowledge of two more properties of
the language.
First, the two settings together determine the extent of verb movement taking place
in embedded clauses. It was argued that Icelandic and Yiddish satisfy the Tense condition
under c-command. The verb moves up higher than the subject and predicate. XP-fronting is
triggered for independent reasons. That verb second effects also show up in embedded
clauses is caused by the fact that these languages have rich, anaphoric, agreement and
therefore an independent AgrP. The consequence was that the complementizer cannot be
involved in satisfaction of the Tense condition and verb movement takes place as a last resort
operation. In Mainland Scandinavian, on the other hand, no AgrP has to be projected, given
the poor nature of Agr, and the complementizer is able to satisfy the Tense condition by
entering into a dependency relation with the Tense features on the verb. For this reason,
generalized verb second is absent in embedded clauses.
Second, if in a language AgrP must be generated and the Tense condition is
satisfied under c-command, two projections will dominate VP in declarative clauses. The
consequence of this is that transitive expletive constructions can be generated. This option is
excluded if either of these parameters is set differently. An interesting consequence of the
analysis is that it singles out Dutch as a unique case. In no way does the distribution of er fit
in with the generalizations that can be made about expletives in Germanic. It was shown that
there are independent reasons for not considering this element to be an LF place-holder of
the logical subject. This strengthens the analysis of er as an expletive (i.e. semantically
empty) adverb rather than an expletive subject, a claim already defended in Bennis (1986).
In the introduction I remarked that the view on functional structure adopted in
this thesis and the specific triggers I argued for are logically distinct. The theory of verb
movement could be right but the triggers wrong or vice versa. It should be noted,
however, that it is the interaction of the two that makes a couple of explanations
straightforward. The explanation for the lack of verb second in English as well as the
explanation for the distribution of TECs are directly related to the view on functional
projections that I adopt. In a more standard view holding that every functional
projection is headed by an empty head, it becomes impossible to directly relate the lack
of verb second in English to the presence of an empty head in this language, since
empty heads appear much more unrestrictedly. Likewise, the grammaticality of a TEC
hinges on the number of functional projections available. In the present proposal verb
movement is directly indicative of available structure, so that Vikner’s generalization is
derived in a natural way. Since in standard approaches empty heads can in principle be
postulated without an overt head moving to it, it becomes hard to see what blocks
generation of a TEC. One could of course claim that a projection only becomes active
once a verb has been moved to it: Only where two projections are activated through
movement is generation of a TEC allowed. Note, however, that in English a subject can
occupy the specifier position of a projection with an empty head position. Why this
possibility should not be available in general is therefore not immediately obvious. In
general, one can always introduce extra assumptions but the point is that the
explanation for the lack of verb second in English as well as the explanation for the
distribution of TECs follow without such extra assumptions from the theory of overt verb
movement in this thesis and the constraint it puts on the occurrence of empty heads.
If the triggers for V to I and V to C are correct, the nature of the two verb
movement operations can be successfully unified. Recall from the introduction that the
MHH and AHH both had trouble explaining why V to I is related to a morphological
effect (namely richness of inflection) whereas no such effect can be observed with V to
C. The dichotomy between the category C and I is eliminated by the present analysis
since C is replaced by T. Hence, V to C and V to I movement are related to morphology
to the same extent: Both operations take place to project some feature of the verb and in
both cases this feature correlates with existing morphology.

Categories