The syntax–semantics interface.pdf

(1135 KB) Pobierz
Available online at
www.sciencedirect.com
Lingua 130 (2013) 66--87
www.elsevier.com/locate/lingua
The syntax--semantics interface
Martin Hackl
*
Massachussetts Institute of Technology, USA
Received 26 October 2011; received in revised form 22 January 2013; accepted 22 January 2013
Available online 6 April 2013
Abstract
The study of the syntax--semantics interface is concerned with linguistic phenomena that are the product of interactions between
principles of syntactic organization and principles of semantic interpretation. Such interactions abound in natural language and can be
found in all subsystems of the grammar. This paper examines a particular subclass of such phenomena revolving around quantificational
expressions. The central concern of the discussion are the grammatical mechanisms that mediate between the syntactic position a
quantifier appears in and the semantic import it has on the sentence meaning. Of particular interest are cases where a quantifier is
interpreted in a position that is different from the position it seems to occupy in the syntax. A leading hypothesis to explain cases of this
sort, which exemplify a general property of natural language called displacement, is that they are the product of overt or covert movement
operations. Empirical support for this approach is presented in the form of correlations between three grammatical phenomena --
Quantifier Scope, Antecedent Contained Deletion, and Extraposition -- which receive a uniform account under the above hypothesis.
© 2013 Elsevier B.V. All rights reserved.
Keywords:
Syntax--semantics interface; Logical form; Quantification; ACD; Extraposition
1. Introduction
The study of the syntax--semantics interface is concerned with those aspects of syntactic structure that have an effect
on meaning composition and those aspects of meaning that have systematic effects on syntactic structure. Research in
this area has produced a wealth of empirical and theoretical knowledge that is impossible to survey comprehensively
within the limits of this paper, even if the discussion were confined to the main results or the foundational issues. However,
many of the key issues can be illustrated by focusing on a set of interrelated topics concerning the syntax and semantics of
quantificational expressions, which exemplify clearly and compactly what is at stake in the study of the syntax--semantics
interface.
Before getting started, a few general remarks about the nature of the research in this domain are in order. The empirical
goal of research on the syntax--semantics interface is to determine the extent to which the meaning of an expression
depends on its syntactic properties and the extent to which its syntactic properties are a function of its meaning. The
theoretical goal is to give a principled account of these interactions, rooted both in an understanding of the compositional
processes made available by the language faculty and in an understanding of how these processes interact with the rest
of the mind. At the current state of research, we do not have such an account nor do we have fail-safe criteria that identify
all and only those phenomena that are a product of interactions of syntax and semantics. There are, however, clear and
fairly uncontroversial cases that exemplify when syntax and semantics interact systematically and when they don’t. These
* Correspondence address: MIT Linguistics and Philosophy, 77 Massachusetts Avenue, 32-D966, Cambridge, MA 02139, USA.
Tel.: +1 6172539373.
E-mail address:
hackl@mit.edu.
0024-3841/$ -- see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.lingua.2013.01.010
M. Hackl / Lingua 130 (2013) 66--87
67
cases provide us with guidelines for how to structure the search for empirical evidence as well as how to construct theories
that explain the phenomena of interest.
To see how fundamental and ubiquitous the interplay between syntax and semantics is to natural language consider
simple examples like those in (1)a and b.
(1)
a.
b.
John kissed Mary.
Mary kissed John.
The two sentences in (1) are constructed from the same lexical material,
John, Mary, kiss,
and the past tense morpheme
--ed, yet they mean different things. In both cases,
kiss
takes two proper names,
John
and
Mary,
as arguments to form a
sentence describing a kissing event involving John and Mary. The difference in meaning concerns who is doing what to
whom. In (1)a,
John
is the agent or instigator of the kissing event and
Mary
is the patient or recipient while in (1)b the roles
are reversed. This difference in meaning corresponds to a difference in syntactic organization: in (1)a
John
is the syntactic
subject and appears to the left of the verb while
Mary
is the syntactic object and appears to the right of the verb; in (1)b the
opposite is the case. Thus, we observe a correlation between syntactic organization and meaning: the agent role of a verb
like
kiss
is assigned to the subject position while the patient role is assigned to the object position.
However, not all aspects of syntactic organization have semantic repercussions. To see this consider the sentences in
(2a) and (2b).
(2)
a.
b.
He kissed her.
She kissed him.
Here, the difference between subject position and object posistion is also marked using different forms of the third-person
pronouns:
he
and
she
are used for the subject position, while
him
and
her
are used for the object position. This is an
example of a phenomonon called structural case assignment, in which the subject position gets marked in one way, called
nominative case, while the object gets marked in another way, called accusative case.
1
The sentences in (1) and (2) might suggest that case marking is causally related to both the semantic role of the
nominals and their syntactic position. However, this is not the case. For instance, being marked nominative is not a
requirement for carrying the agent role of a verb like
kiss,
nor does nominative marking stand in the way of carrying the
patient role. This can be seen in the examples in (3).
(3)
a.
b.
c.
He kissed her.
She was kissed by him.
I expected him to kiss her.
(3)a shows, again, the basic pattern of case assignment in an active sentence. Passivizing the sentence, (3)b, results in
nominative case being assigned to the patient
she
2
while (3)c shows that in a non-finite clause an accusative form (him) can
carry the agent role of
kiss.
This suggests that nominative case marking is neither necessary nor sufficient for realizing a
specific thematic role of a verb and, thus, a syntactic process that is not (directly) causally related to meaning composition.
3
Semantic effects on syntax are just as ubiquitous as syntactic effects on meaning. For instance, the number of syntactic
arguments that need to accompany a verb to form a full sentence is a function of the lexical semantics of the verb:
smile
requires one,
kiss
requires two,
give
requires three. Nevertheless, not all aspects of meaning are syntactically visible. A
simple example of syntactic inertness of meaning can be seen when considering pairs of verbs like
frown
and
smile.
Clearly,
these two verbs have rather different lexical meanings yet there seem to be no syntactic processes that are sensitive to these
differences in meaning: both take one syntactic argument to form a complete sentence, both can at least marginally take a so
called cognate object argument (smile
a (big) smile, frown a(n ugly) frown),
yet neither seems to be able to undergo
passivization (*a
(big) smile was smiled, *a(n ugly) frown was frowned),
both can appear in the progressive form, etc.
These examples give but a first impression of the complexity that is characteristic of the interface between structure
and meaning. To probe how deep and how systematic the interactions are between these two components of the
grammar, we turn to the domain of quantifiers, which lend natural language much of its expressive power and which have
shaped much of the discussion and theoretical development in the study of the syntax--semantics interface.
In English, a morphologically impoverished language, case marking can be directly observed only on pronouns, though in other languages, it
can appear on all nominals.
2
The thematic subject, the bearer of the agent role, is realized in what is called the oblique by-phrase which, in English, requires the same form
as the accusative (him).
3
To show that nominative case is generally semantically inert requires a more extensive discussion, which would lead us too far away here.
1
68
M. Hackl / Lingua 130 (2013) 66--87
2. Quantifier Integration and Quantifier Scope
2.1. Quantifiers and the problem of quantifier integration
In the previous section we have seen that verbs introduce thematic roles such as agent and patient. We also saw that
proper names and pronouns can serve as the arguments of verbs, carrying these thematic roles in the composition of
sentence meaning. The class of expressions that can serve as verb arguments is not restricted to proper names and
pronouns, of course. Definite descriptions, e.g.
the boy,
as well as quantificational expressions such as
no one,
everybody, some girl, most boys, more than two students,
etc. can serve the same function, (4).
(4)
a.
b.
c.
The boy kissed every girl.
No one kissed more than two girls.
Some girl kissed the boy.
Expressions that can serve as arguments of verbs are canonically referred to as determiner phrases (DPs) and
semantically fall into two main groups: DPs that pick out individuals in the world (referring DPs) and DPs that don’t
(quantifiers). The first class contains proper names, pronouns, definite descriptions, among others. For instance, the
proper name
Mary
can be analyzed semantically as referring to the individual known as Mary, while
the woman in the
bright yellow hat
can be analyzed as referring to the most salient woman in a bright yellow hat in the conversational
context. If that woman is Mary, then
Mary
and
the woman in the bright yellow hat
will have the same referent.
Quantifiers, by contrast, cannot be said to refer to individuals. Which individuals could we take to be the referents of
expressions like
no girl, most boys,
or
more than two students?
Thus, rather than analyzing them as referring
expressions, quantifiers are canonically analyzed as second order predicates (Frege,
1879; Barwise and Cooper, 1981;
etc.). Whereas expressions that denote first order predicates take DPs as their arguments, second order predicates take
other predicates as their arguments to form sentences. For instance,
no one
is true of a predicate it takes as argument as
long as there is no individual that satisfies that predicate,
every boy
is true of a predicate if all the boys satisfy it,
more than
two professors
is true of a predicate if the number of individuals that are both professors and satisfy the predicate is three
or higher, etc.
The difference in meaning between referring DPs and quantificational DPs has a profound impact on sentence
meaning and, as we will see, raises deep questions about the syntax--semantics interface. A first indication of this can be
seen in the fact that sentences with quantifiers can be ambiguous when sentences employing a referring expression can’t.
Consider the example in (5).
(5)
John hasn’t read more than two novels by Tolstoy.
The sentence in (5) is ambiguous. It can be understood to say that the number of novels written by Tolstoy that John has
read is two or smaller but it can also be understood to say that there are more than two novels by Tolstoy that John hasn’t
read. Under the first reading, the sentence puts an upper limit on the Tolstoy novels John has read, i.e. the sentence would
be false in a situation where John has read three or more. The second reading, by contrast, is consistent with such a
situation. For instance, John could have read six novels by Tolstoy and the sentence would still be true as long as there are
(at least) three he hasn’t read, e.g. The Cossacks, Sebastopol, and The Kreutzer Sonata.
We can isolate the two readings of (5) by using more cumbersome variants, such as those in (6). (6)a unambiguously
expresses the first meaning of (5) and (6)b unambiguously expresses the second meaning of (5).
(6)
a.
b.
It is not the case that John has read more than two novels by Tolstoy.
More than two novels by Tolstoy are such that John hasn’t read them.
These variants also give us a clue how to think about the origin of the ambiguity. The first reading arises when the negative
operator
not
negates the sentence
John has read more than two novels by Tolstoy.
The second reading arises when the
quantifier
more than two novels by Tolstoy
takes a negated predicate describing the things John has not read as argument.
Modeling predicates as sets, an approximation which will do for our purpose, we can represent this idea a bit more formally.
For instance, the predicate
‘having
been read by John’ is a predicate of individuals and so can be thought of as denoting the
set of things John has read (in set notation {x: John has read x}). The second order predicate
more than two novels by
Tolstoy,
by contrast, is a predicate of predicates, rather than of individuals and so can be modeled as {P:
j{x:
x is a novel by
Tolstoy}
\
Pj > 2}, or
‘the
collection of sets such that at least two of the entities in any set of the collection are also novels
written by Tolstoy’. Thus, it could be true of predicates like {x: Mary has read x}, {x: John has not read x}, {x: x is written in
M. Hackl / Lingua 130 (2013) 66--87
69
Russian}, etc. Putting these assumptions together using a loose mixture of predicate logic notation and English, we can
represent the composition of the two readings of (5) as in (7).
(7)
a.
a.
0
b.
b.
0
:
[[{P:
j{x:
x is a novel by Tolstoy}
\
Pj > 2}] ({x: John has read x})]
:
[>2x[x is a novel by Tolstoy & John has read x]]
‘it
is not the case that there are more than two things that are both novels by Tolstoy and read by John’
[{P:
j{x:
x is a novel by Tolstoy}
\
Pj > 2}] ({x:
:[John
has read x]})
>2x[x is a novel by Tolstoy & John has not read x]]
‘there
are more than two things that are both novels by Tolstoy and have not been read by John’
(7)a, which corresponds to (6)a, is a negative statement. It states that the second order predicate denoted by
more than
two novels by Tolstoy
is not true of the predicate {x: John has read x}. Simplifying the formula by folding {x: John has read
x} into the predicative variable slot, P, of {P:
j{x:
x is a novel by Tolstoy}
\
Pj > 2} we get the negative statement in (7)a
0
: it
can’t be that John has read 3 or more novels by Tolstoy if the sentence is true.
4
In (7)b, which corresponds to(6)b, negation
is part of the predicate that
more than two novels by Tolstoy
takes as argument, {x: John has not read x}. Thus, the
statement is true if the number of things that are both novels of Tolstoy and not read by John is three or higher.
The difference between the relative positions of the quantifier and negation in the composition of the sentence
meanings is know as a difference in
scope,
and the ambiguity of sentences like (5) as a
scope ambiguity.
5
What we see,
then, is that quantifiers can give rise to scope ambiguities. Referring expressions, by contrast, cannot. For instance, when
we replace the quantifier
more than two novels by Tolstoy
with a referential expression like
Anna Karenina
or
Tolstoy’s
seventh novel,
the ambiguity disappears, (8). The sentence simply states that John hasn’t read Anna Karenina/the
seventh novel written by Tolstoy -- irrespective of whether we use the basic sentence, (8)a, or one of its more cumbersome
variants, (8)b and c.
6
(8)
a.
b.
c.
John hasn’t read Anna Karenina/Tolstoy’s seventh novel.
It’s not the case that John has read Anna Karenina/Tolstoy’s seventh novel.
Anna Karenina/Tolstoy’s seventh novel is such that John hasn’t read it.
Analyzing quantifiers as second order predicates can, thus, explain the semantic import quantifiers have on sentence
meanings. It also helps us appreciate a fundamental puzzle associated with quantifiers.
7
If quantifiers are second order
predicates and, thus, require predicates (sets of individuals) to form sentence meanings, how are they able to realize
argument positions of verbs, which require individual denoting expressions to form sentence meanings? Take the verb
kiss,
again. It denotes a relation between individuals, the agent and patient of a kissing event. We expect that forming a sentence
with
kiss
requires two referring expressions, such as
John
and
Mary,
which denote individuals. What we, prima facie, don’t
expect is that we can form a sentence with quantificational expressions realizing the argument positions of
kiss;
the meaning
of
kiss
is not that of relation between quantifiers. In other words it is puzzling that (9)b,c are as acceptable as (9)a.
(9)
a.
b.
c.
John kissed Mary.
No boy kissed Mary.
Mary kissed every boy.
I will not introduce a rigorous meta-language to describe how meanings are composed. Instead, I will rely on a notation that reflects the main
idea of how complex meanings arise as function of their parts in an intuitive way. In (7), for instance, negation is represented as a sentence level
propositional operator,
:,
and the idea that the quantifier
more than two novels by Tolstoy
combines with a predicate with the help of brackets and
parenthesis, [a](b), which is reminiscent of a widely employed notation to represent a function,
a,
applying to an argument,
b.
See e.g.
Heim and
Kratzer (1998)
for an introduction to the formal techniques underlying the discussion presented in this paper.
5
In Sections
2.2 and 2.3
we will see how this semantic ambiguity can be seen as a product of a covert syntactic ambiguity of (5), which is not
reflected in the word order of the sentence.
6
As before, we can take to (8)b and c to correspond to two different ways in which negation can combine with the rest of the sentence, (8)b
0
,c
0
,
respectively, however, the resultant meaning will be the same.
(8) b.
0
:
[[{x: John has read x}] (Anna Karenina)]
c.
0
[{x:
:[John
has not read x]}] (Anna Karenina)
This puzzle does not depend on analyzing quantifiers as 2nd order predicates. Rather, it stems from the fact that quantifiers do not refer. Any
model of quantifier meanings that recognizes this elementary fact will also run into the issue how quantifiers can serve as arguments. What is
specific to the analysis of quantifiers as second order predicates is that no combinatorial puzzle arises for quantifiers in the subjects position of
verbs as shown in (10). See
Johnson and Tomioka (1998)
for discussion.
7
4
70
M. Hackl / Lingua 130 (2013) 66--87
To solve this puzzle, we need to find a way of systematically constructing suitable predicates for the quantifiers to
combine with from the remainder of the sentence. In (9)b the quantifier,
no boy,
is in subject position and the remainder is
kissed Mary.
We can construct the predicate {x: x kissed Mary} by assuming that
kiss
and
Mary
form a constituent,
traditionally labeled V
0
. The meaning of this constituent is a result of
Mary
filling the inner argument position of
kiss,
thereby
carrying the patient role in the kissing event described by the sentence, (10)a.
No boy,
which realizes the second
(external) argument of
kiss,
the subject position carrying the agent role, can combine with this predicate. Thus, the
syntactic requirements of the sentence are met:
no boy
can fill the subject argument position of the verb
kiss
(in the same
way that a referring DP would), while Mary can fill the object argument position. The semantic requirements are also met:
no boy
can take the predicate
kissed Mary
as its argument. The result is a verb phrase, VP, which is a constituent that has
all argument positions introduced by the verb
kiss
filled, (10)b.
8
(10)
a.
b.
V′
V
kiss
DP
Mary
DP
no boy
VP
V′
V
kiss
DP
Mary
Thus, the reason that
no boy
can serve as external argument of kiss in (9)b is not that the predicate
kiss Mary
might have
the denotation of
no boy
in its extension. Rather, it is because the second order predicate denoted by
no boy
might have
the predicate denoted by
kiss Mary
in its extension.
The situation is quite different in (9)c. Here the quantifier
every boy
is the object of
kiss
and so carries the patient role of
the kissing event described by the sentence. The remainder of the sentence
Mary kissed
is, however, not a constituent.
This means that it cannot be directly composed to form a predicate (the set of things that Mary has kissed). This suggests,
at first glance, that this sentence should be uninterpretable:
kiss
cannot directly combine with its object,
every boy
since it
is not referring to an entity that could be the patient in a kissing event, nor can it directly combine with its subject,
Mary.
How, then, do these two expressions form the predicate {x: Mary kissed x} which the quantifier
every boy
needs to
combine with to generate the desired meaning?
One possible solution would be to argue that in (9)c, the VP has a different structure so that
Mary
is the first expression
kiss
combines with, (11)a, and
every boy
the second. However, this is simply moving the puzzle around, since we no
longer understand how Mary can carry the agent role in the kissing event described by the sentence and
every boy
gets to
carry the patient role, (11)b.
9
(11)
a.
b.
V′
DP
Mary
V
kiss
DP
Mary
V′
VP
DP
every boy
V
kiss
What we need to solve the puzzle of how the object DP
every boy
is integrated into the sentence is a method that
delivers a suitable predicate, {x: Mary kissed x}, from the remainder of the VP,
Mary
and
kiss,
fulfilling the semantic
requirements of
every boy,
while maintaining that
Mary
serves as the external (subject) argument of
kiss
and
every boy
serves as the internal (patient) argument.
I will ignore the past tense morpheme -ed here. It is a realization of a tense operator which is structurally above the verb and its argument
positions and locates the event described by the VP in the past.
9
Attributing the reversal in thematic role assignment to a difference in linear order by claiming that the patient role is assigned to the second
argument when a quantifier is to the right of
kiss
simply turns the puzzle into the question why linear order matters with quantifiers but not with
referring expressions.
8

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika:

Zgłoś jeśli naruszono regulamin