Upload
vunhi
View
226
Download
2
Embed Size (px)
Citation preview
Università degli Studi di Verona
Corso di Laurea Specialistica in Linguistica
Tesi di Laurea
TWO TOPICS IN THE SYNTAX AND
SEMANTICS OF ANAPHORA
Relatore:
Prof. Denis Delfitto
Laureando:
Gaetano Fiorin
Anno Accademico 2005-2006
2
3
Acknowledgments
I’ve spent hours trying to figure out the right words to thank Denis Delfitto and Giorgio
Graffi. Then, studying an article by Yosef Grodzinsky (Grodzinsky 2000b), I read a story
about Duke Ellington, who, late in his life, was interviewed during a TV show and I
thought it was perfect for my purposes. Grodzinsky reports it from Medini (1988). Here’s
the story: “[Ellington] was asked what he would like to say about the influences and
events that shaped his starry life. He said something like this: ‘Ya know what it’s like
runnin’ around life, like a maze, and ya comes to a corner and ya don’t know which was to
go, and some guy is standing there and ya asks him ‘which way?’ and he points and says,
‘That way,’ and ya run and run until you come to another fellow and ya asks him, and he
points and says ‘that way’ – what I want to do today…is to thank those guys who was
standin’ on these corners at the right time and pointed me off in the right way.’” Denis
Delfitto and Giorgio Graffi have been to me the guys around the corner and I thank them
for having pointed me in the right way.
During this period I have had the chance to meet a number of researchers, teachers, PhDs,
and fellow-students who made this experience extraordinary. I wish to thank them all:
Birgit Alber, Nicholas Asher, Camilla Bettoni, Roberto Bonato, Francesco Carbone, Frank
Drijkongingen, Paolo Frassi, Daniele Giaretta, Alessandra Giorgi, Jim Higginbotham,
Julie Hunter, Simona Marchesini, Chiara Melloni, Stella Merlin, Paola Paradisi, Laura
Pirani, Alan Prince, Stephan Rabanus, Eric Reuland, Jasper Roodenburg, Alessandra
Tomaselli, Alfredo Trovato, Maria Vender. I am grateful to Alessandra Giorgi and Eric
Reuland for their help and suggestions and to Francesco Carbone for having read and
4
commented the thesis and for helpful discussions. A very special thanks goes to
Guglielmo Bottari whose friendship is a precious gift. A heartfelt thanks goes to my
family, Valentina, Ornella and Angelo, to which I owe everything, and to Giorgio and
Alberto.
This thesis is dedicated to the memory of Attilio Boano.
Verona
October, 2006
5
Summary
1 A Theory of Anaphoric Relations Based on Economy Principles 10
1.0 Introduction 10
1.1 Canonical binding theory 17
1.1.1 Binding and binding conditions 17
1.1.2 Governing category 21
1.1.3 NPs as feature matrices 22
1.2 What’s wrong with CBT 23
1.2.1 Explanatory inadequacy 23
1.2.2 Binding conditions at the Conceptual-Intentional interface 25
1.2.3 Logophoric use of anaphors 27
1.2.4 Locally bound pronouns 32
1.2.5 Complementarity 32
1.3 A theory of anaphoric relations based on economy principles 33
1.3.1 Preliminary questions 33
1.3.2 Interpreting procedures and economy principles 35
1.3.3 Encoding linguistic dependencies 40
1.3.3.1 Syntactic encoding 40
1.3.3.1.1 Encoding a syntactic dependency 40
6
1.3.3.1.2 chain, Chain and CHAIN 42
1.3.3.1.3 Interpreting CHAINs 45
1.3.3.1.4 Theoretical remarks 52
1.3.3.2 Variable binding 52
1.3.3.3 Co-reference 53
1.3.4 Economy principles 55
1.3.4.1 Condition A 57
1.3.4.2 Condition B 58
1.3.4.3 Condition C 59
1.3.4.4 Locally bound pronouns 60
1.3.4.5 Logophoric and long distance anaphors 61
1.3.4.6 Complementarity 63
1.4 Discussion 63
1.4.1 Theoretical considerations 63
1.4.2 Rule BV 64
1.4.3 Rule I 68
1.5 Conclusions 71
2 Attitudes De Se 75
2.0 Introduction 75
2.1 Attitudes de dicto, de re, de se 78
2.1.1 Attitudes de dicto and attitudes de re 78
7
2.1.2 Attitudes de re and attitudes de se 79
2.2 Chierchia’s property-account of attitudes de se 81
2.2.1 Linguistic relevance of attitudes de se 81
2.2.2 Lewis (1979): a ‘self-locational’ view of beliefs 82
2.2.3 Chierchia (1989): properties versus propositions 83
2.2.4 The grammar of de se 85
2.2.5 Infinitives and gerunds: PRO as a λ-abstractor and self-ascription revised 86
2.2.6 Anaphora and attitudes de se 89
2.3 Reinhart (1990) 91
2.3.1 Empirical objections to Chierchia (1989) 91
2.3.1.1 Attitudes de se and variable-binding 91
2.3.1.2 Attitudes de se and long-distance anaphors 97
2.3.2 Reinhart’s alternative proposal 98
2.3.2.1 Self-representational relation 98
2.3.2.2 Argument reduction and θ-reduction 103
2.4 Higginbotham (1992) 104
2.4.1 Attributive versus propositional account of control 104
2.4.2 Higginbotham’s theory of control 107
2.4.3 Attitudes de se 110
2.4.4 Features of PRO 112
2.4.5 Appendix: a theory of control based on economy principles? 113
2.5 Conclusions 116
8
Works Cited 121
9
10
1 A THEORY OF ANAPHORIC RELATIONS BASED ON
ECONOMY PRINCIPLES
1.0 Introduction
Anaphoric relations characterize all natural languages. An anaphoric relation is an
asymmetric relation between two linguistic expressions occurring in the same
environment, according to which the syntactic and semantic value of one expression, the
anaphoric element, crucially depends on the syntactic and semantic value of the other, the
antecedent. For instance, in sentence (1), the reflexive pronoun himself can contribute to
the meaning of the whole sentence only in the case its anaphoric relation with Frank is
correctly taken into account.
(1) Frank loves himself
Sentence (1), is true if and only if Frank loves himself, for ‘Frank’ as value of ‘himself.’
Thus, Frank and himself are anaphorically linked, Frank being the antecedent of himself.
Anaphoric relations represent a central topic in the theory of natural language.
Such relations show up in different ways and with different functions. In sentence (1), we
see an anaphoric relation between two nominal phrases (NPs), but this is not the only way
natural languages make use of this tool: just to cite a couple of examples, natural
languages make extensive use of temporal anaphora, anaphora between NPs and clauses
11
or between NPs and full sentences. However, in this thesis I will concentrate on the
anaphoric relations between NPs.
The goal of a theory of language is to explain how the grammar encodes such
relations and represents them at the interface with the system of thoughts. In the
generative framework, initiated by the seminal work of Noam Chomsky, anaphoric
relations between NPs are traditionally accounted for by the module of grammar called
binding theory. The basic idea is that the core grammar is composed of different
subsystems, one of which is binding theory. Binding theory was first introduced by
Chomsky (1981), within what is known as Government and Binding (G&B) framework
and further developed in subsequent work, e.g. Chomsky (1986); I will refer to this
version of binding theory as Canonical Binding Theory (CBT).
The notion of binding, within CBT, is to be understood as a purely syntactic
notion, technically implemented as co-indexation under c-command. The idea underlying
this theory is that every NP must receive a numerical index, under well-specified
structural conditions, so that at the interface between the grammar and the system of
thoughts, i.e. at the level of Logical Form (LF), an anaphoric relation between two NPs
will be indicated by the fact that both NPs bear the same index. For instance, the LF of
sentence (1) would be something like (1′), where both NPs Frank and himself bear the
same numerical index 1.
(1′) [IP [NP Frank]1 [VP loves [NP himself]1]]
12
NPs are divided in three types: Referential expressions, pronouns, and anaphors
(i.e. reflexive pronouns and reciprocal). CBT consists of three conditions on binding, each
applying to an NP type:
(2) CBT
(Condition A) An anaphor is bound in its minimal domain
(Condition B) A pronoun is free (i.e. not bound) in its minimal domain
(Condition C) A Referential expression is free
CBT is not uncontroversial and different objections have been raised against it. In
this chapter, I will follow the line paved by the work of Tanya Reinhart and Eric Reuland1,
and I will raise two specific objections to CBT.
The first objection is theoretical. I will claim that CBT does not meet the condition
of explanatory adequacy and, particularly, the condition of explanatory adequacy as stated
in the Minimalist Program (Chomsky 1995). In the minimalist framework it is assumed
that: (i) a theory of language meets explanatory adequacy if it can characterize the initial
state of the language faculty in a principled way, rather than just describe a certain state of
the language faculty; (ii) the grammar of human language is a computational system that
meets the Inclusiveness Condition; this condition states that no new element can be add in
the course of the computation. Given these (minimalist) assumptions I will claim that
CBT, although descriptively adequate, does not attain explanatory adequacy since: (i) it
describes a certain state of the language faculty but it does not characterize the initial state;
1 E.g.: Reinhart (1976; 1983; 2000; 2001), Reinhart and Reuland (1991; 1993; 1995), Reuland (1996; 1998; 2000; 2001a; 2001b).
13
in other words, CBT does not tell us which basic properties of the human language faculty
underlie the binding conditions stated in (2); (ii) the mechanism of syntactic binding is not
consistent with our basic assumptions about the system of grammar; in fact, the
introduction of indices, which are syntactic objects visible at LF, in the course of the
computation does not respect the IC.
The second objection I will raise against CBT is empirical. Among other problems,
I will concentrate on what is known as the ‘logophoric’ use of anaphors. According to this
use, an anaphor can be used, in some circumstances, as un unbound element. Many
languages, including for example English and Italian, show relevant instances of such use.
In my presentation I will concentrate on examples from English and Icelandic. Recall that
Condition A of CBT entails that anaphors must always be bound; henceforth, the
logophoric use of anaphors violates this condition.
Several proposals have been made to account for these problems. Nevertheless, in
this thesis I will follow the path of a deep revision of the theory of anaphoric relations, a
revision based on the work that Reinhart and Reuland have done in the last twenty years.
In this thesis I will try to present their work as an organic and coherent theory of
anaphoric relations, alternative to CBT.
It will be first demonstrated that an anaphoric relation can be encoded at three
different levels: (i) it can be encoded in (narrow) syntax, and in this case it must be
understood as a relation resulting from the blind and automatic application of basic
computational operations to a pure morpho-syntactic lexicon, i.e. characterized in terms of
primitive features; (ii) it can be encoded at the semantic level, and in this case the relevant
operation is that of (logical) variable-binding by means of which a variable is bound by an
operator; (iii) it can be encoded at the discourse level and in this case the mechanism at
14
hand is that of co-reference, by means of which two linguistic expressions, syntactically
and semantically unrelated, are valuated, according to pragmatic principles, as referring to
the same entity in the discourse domain.
The syntactic relation that will be presented in this framework, which is different
from the notion of syntactic binding of CBT, prevents us from falling into the theoretical
problems that CBT had to deal with. In fact, this relation does not imply the use of indices,
hence IC is respected. Moreover, since such a relation results from the blind and automatic
application of computational operations to primitive lexical features, it attains explanatory
adequacy. In this case there is no postulation of a new primitive notion, such as syntactic
binding, rather the relation derives straightforwardly from the basic properties of the
computational system.
It is also worth noticing that both the syntactic and the semantic encoding are
subjected to specific structural conditions, such as c-command. However, it will be shown
that such conditions are also derived from more primitive properties of the computational
system, rather than postulated. More precisely, they derive from conditions on movement,
which is a primitive syntactic operation. It seems also correct to assume that co-reference
is subjected to pragmatic constrains, however I shall not consider this issue here since it
would require the introduction of a fully formalized theory of discourse, a task that I
cannot grant in this thesis.
It is then assumed that different costs are associated with the three levels. The
basic idea is that encoding an anaphoric relation between two linguistic expressions at the
level of syntax is cheaper than encoding a linguistic dependency between the same
expressions at the semantic level, and, similarly, encoding an anaphoric relation between
15
two linguistic expressions at the semantic level is cheaper than encoding a linguistic
dependency between the same expressions at the discourse level.
Note that, at this point, the system of language can rely on three different strategies
to achieve the same result. For instance, consider a structure like (3), where we want two
linguistic expressions α and β within σ to receive the same value:
(3) [σ …α…β…]
If the relevant structural conditions are met, the dependency between α and β can be
encoded within syntax and the encoding is associated with a certain cost. However, if the
structural conditions at the semantic level are also met, a linguistic dependency between α
and β can be encoded at this level as well, but the encoding would be more expensive.
Finally, α and β can receive the same value at the discourse level but this would be the
most expensive option.
The theory of anaphoric relations I will propose following Reuland (2001a) will be
based on an economy principle, that can be very informally stated as follows: ‘a certain
strategy is blocked whenever a cheaper strategy is available.’ For instance, if a linguistic
dependency can be encoded at the semantic level, then the encoding at the discourse level
will be blocked, and, if a linguistic dependency can be encoded at the syntactic level, then
the encoding at the semantic level will be blocked.
This theory of anaphora has been developed specifically to overcome the empirical
and theoretical problems of CBT, and it will be shown how the result is achieved.
However, the theory encounters other problems. One of such problems is that the theory
16
relies on mechanisms of global economy, as opposed to computational economy. Very
roughly, a principle of computational economy applies to a set of derivations obtained
using the same initial array of lexical items and determines the cheapest derivation. A
principle of global economy, on the other hand determines the cheapest derivation among
a set of derivations obtained by different arrays of lexical items. It will be shown that the
principles of the theory proposed in this thesis are principles of global economy.
In the first paragraph (1.1) I shall present CBT and I will give a brief sketch of
some developments of such theory.
In the second paragraph (1.2) I shall present a discussion of the objections raised
against CBT concentrating first on the theoretical problems (1.2.1 and 1.2.2) and then on
the empirical problems (1.2.3, 1.2.4 and 1.2.5). Particular attention will be given to the
logophoric use of anaphors and to some relevant literature on the subject.
In the third paragraph (1.3) I shall expose the theory of anaphoric relations based
on economy principles. In particular, in 1.3.1 I will introduce a working hypothesis on
which the development of the theory will be based and in 1.3.2 I will give a first account
of the structure of the theory. Then I will give a detailed account of the three interpretative
strategies at the three relevant levels: at the syntactic level (1.3.3.1), at the semantic level
(1.3.3.2), and at the discourse level (1.3.3.3). Finally I will introduce the economy
principles regulating the choice between the different interpretative strategies and I will
try to show the empirical power of the theory through a number of examples (1.3.4). I will
also show that the logophoric use of anaphors will be accounted straightforwardly by the
theory.
17
In a last paragraph (1.4) I shall present some personal considerations on the theory
and I shall try to consider some theoretical and empirical problems that the theory leaves
open.
1.1 Canonical binding theory
1.1.1 Binding and binding conditions
Binding theory is the module of Universal Grammar (UG) that regulates the interpretation
of Nominal Phrases (NPs). BT was introduced by Chomsky (1981) and then developed in
subsequent works published during the eighties (Chomsky 1982, 1986). I will refer to this
version of binding theory, according to the traditional usage, as Canonical Binding Theory
(CBT). Within the Government and Binding (G&B) framework, binding theory is
introduced as a subsystem of core grammar, along with other subsystems as bounding
theory, government theory, θ-theory, Case theory and control theory.
The central idea is that at the level of Logical Form (LF)-representation, all NPs,
including empty categories such as traces or PROs, must be indexed by a single index. BT
can then be seen as the collection of the principles that are responsible for the correct
indexing.
The notion of binding is, in CBT, a syntactic notion, technically implemented by
means of two more basic syntactic notions, namely, c-command and co-indexing. Two
18
NPs are co-indexed if they bear the same index2. C-command and binding are defined as
follows:
(1) C-command3
A node α c-commands a node β iff
(i) α does not dominate β;
(ii) β does not dominate α;
(iii) the first branching node dominating α also dominates β
(2) Binding
α binds β iff
(i) α c-commands β;
(ii) α and β are coindexed
A basic assumption for CBT is that NPs can be subdivided into three categories: (i)
anaphors, (ii) pronouns and (iii) R-expressions. Anaphors are reflexive pronouns (such as
himself, herself,…), reciprocals (such as each other,…) and NP-traces. Pronouns are all
personal and possessive pronouns such as he, him, his, her, etc.. Both anaphors and
pronouns do not independently select a referent from the universe of discourse. R-
expressions (i.e. Referential-expressions), on the other hand, are all nominal expressions
that do select a referent. CBT has one principle for each category. Formally:
2 In order for α and β to be co-indexed, the following condition must be satisfied: (i) α and β are non-distinct in features for person, number and gender. 3 The notion of c-command is due Reinhart (1976; 1983).
19
(3) Canonical Binding Theory
(Condition A) An anaphor is bound in its minimal domain
(Condition B) A pronoun is free (i.e. not bound) in its minimal domain
(Condition C) An R-expression is free
The notion of minimal domain is not uncontroversial. It will be better discussed in
the next paragraph. Actually, we might let us take the minimal domain of an NP as the
clause containing it. Now let me clarify the meaning of the conditions in (3) with some
examples.
Condition A: consider the following examples:
(4) Franki loves himselfi
(5) *Franki thinks [that Miriamj loves himselfi]
In sentence (4) the anaphor himself is bound by the R-expression Frank within the
anaphor’s minimal domain; Condition A is met and the sentence is grammatical. On the
contrary, in (5) the anaphor is bound by the R-expression Frank, which is outside the
anaphor’s minimal domain (indicated by square brackets); the sentence is then ruled out
by Condition A which requires the anaphor to be bound within its minimal domain.
Condition B:
(6) *Franki loves himi
(7) Franki thinks [that Miriamj loves himi]
20
Condition B states that a pronominal expression is free in its minimal domain. Then, (6) is
ruled out since the pronoun him is bound within its minimal domain by the R-expression
Frank, violating Condition B. Sentence (7), on the other hand, is well formed; the pronoun
is bound, but the binder is outside the pronoun’s minimal domain. Note that Condition B
states only the possibility for a pronoun to be bound, but it does not state that a pronoun
must be bound. Whilst, according to Condition A, an anaphor must be bound, otherwise it
cannot be interpreted, a pronoun needs not to be bound. How can a pronoun be interpreted
if unbound? It can be interpreted by means of pragmatic principles. In fact, the pronoun
him in sentence (7) could also receive an index distinct from Frank’s index and be
interpreted as someone else than Frank, let say John. Since Condition B works at syntax
level, it must just prevent the production of not-well-formed sentences, such as (6).
Finally, Condition C:
(8) *Hei loves Franki
(9) *Hei thinks [that Miriamj loves Franki]
In both cases, the R-expression Frank is bound by a pronoun, as the co-indexing shows,
but, since Condition C states that R-expressions are free (i.e. not bound), both sentences
are ruled out. In other words, Condition C simply states that R-expression cannot be
bound. In (8) and (9), Frank is bound by he, violating the condition.
21
1.1.2 Governing category
The binding conditions of CBT imply a locality concept, namely that of ‘minimal
domain.’ We took the minimal domain of an NP to be the clause embedding the NP. This
definition of minimal domain is contradicted by examples like (10):
(10) Franki believes [himselfi to be a hero]
This is a so-called Exceptional Case Marking (ECM) construction. It is a perfectly well
formed sentence of English although it contradicts Condition A. Problems like this has
been solved assuming a more detailed definition of minimal domain which involves
reference to government.4 The minimal domain of an NP is defined as its Governing
Category (GC):
(11) Governing Category
The governing category of α is the minimal clause containing α, α’s governor and
a SUBJECT.5
4 Government and m-command are defined as follows (Chomsky 1986:8): (i) Government α governs β iff a. α is a governor; b. α m-commands β; c. no barrier intervenes between α and β. Maximal projection are barriers to government. Governors are heads. (ii) M-command α m-commands β iff α does not dominates β and the first maximal projection that dominates α also dominates β. 5 A more detailed definition of GC is (i):
22
The governor of himself, in example (10), is the verb believes. This means that the GC for
himself has to be the entire sentence. Then, defining the minimal domain in terms of GC
prevents this kind of counterexamples.
1.1.3 NPs as feature matrices
Chomsky (1982) states that NPs types such as anaphors, pronouns and R-expressions
cannot be taken as syntactic primitives. Therefore he proposes to replace these categories
with syntactic matrices. Two two-valued features, [±anaphor] and [±pronominal], are
assumed. NPs characterized by the feature [+anaphor] are subject to Condition A, whilst
NPs characterized by the feature [+pronominal] are subject to Condition B. NPs are
classified as in (12):
(12) Anaphors [+anaphor, -pronominal]
Pronouns [-anaphor, +pronominal]
(i) Governing Category The governing category of α is the minimal clause containing α, α’s governor and an accessible subject. The notion of accessible subject is defined as (ii): (ii) Accessible subject α is an accessible subject for β if the co-indexation of α and β does not violate any grammatical principle. The grammatical principle should be the i-within-i filter (Chomsky 1981:211-2), whose goal is to avoid circularity in reference. Given the purposes of this presentation, I do not intend to go deeper into this subject. For interesting discussions on the subject see, among others, Aoun (1986); Lasnik (1986); Giorgi (1987); Graffi (1988).
23
R-expression [-anaphor, -pronominal]6
Given the definition of GC and the classification of NP types just proposed, CBT can be
reformulated as in (13):
(13) Canonical Binding Theory
(Condition A) An NP [+anaphor] is bound in its GC.
(Condition B) An NP [+pronominal] is free in its GC.
There is no more need for stating Condition C since R-expressions are negatively
specified for both features, i.e. they are independently referential.
1.2 What’s wrong with CBT
1.2.1 Explanatory inadequacy
Since the early works of Chomsky (Chomsky 1957; 1965), it has been stated that a
linguistic theory must attain both descriptive and explanatory adequacy. A linguistic
theory is descriptively adequate if it can characterize a certain state attained by the
language faculty. A linguistic theory attains explanatory adequacy if it can characterize the
initial state of the language faculty and show how it maps experience to the state attained
(Chomsky 1995:3). Within a minimalist framework (Chomsky 1995; 2000) the claim that
6 Since the features just proposed are binary features, we would predict an NP positively specified for both features, i.e. [+anaphor, +pronominal]; this NP is PRO, the understood empty subject of non-finite clauses, whose distribution is governed by control theory.
24
a linguistic theory must attain explanatory adequacy is made even sharper. In fact one of
the prerequisites for attaining the goals of the Minimalist Program (MP) is to draw the
boundaries of syntax in a principled way. The properties of the human language faculty
should reflect the combinatorial properties of a morpho-syntactic vocabulary. The
morpho-syntactic vocabulary is explained in terms of primitive features. Every lexical
item (LI) contained in the lexicon is understood as a set of primitive features.
Given these assumptions, CBT is descriptively adequate although it has no
explanatory power. In fact, it can characterize a certain state of the language faculty but it
does not explain how experience is mapped to that state. If this is the case, CBT is not a
theory of the initial state of the language faculty, but just a description (not always correct)
of a certain state. As pointed out by Reuland (2001a:441), the binding conditions of CBT
are arbitrary as stated. They reflect no conceptual necessity in the sense that they follow
neither from any of the basic properties of the grammatical system nor from the
combinatorial properties of the morpho-syntactic vocabulary.
The linguistic distinction between anaphors and pronouns does not make a lot of
sense in such a framework. Anaphors and pronouns are not linguistic primitives, neither
the features [±anaphor] and [±pronominal]. In fact, these features do not explain what
anaphors and pronouns are and why they behave the way they do. It is also well known,
that many languages have much more complex systems of anaphoric expressions, which,
for instance, distinguish between complex anaphors (SELF anaphors like English himself)
and simplex anaphors (SE anaphors like Dutch zich). Given minimalist assumptions,
anaphoric expressions should be characterized in terms of primitive morpho-syntactic
features, and their behavior should follow from these features.
25
1.2.2 Binding conditions at the Conceptual-Intentional interface
In MP (Chomsky 1995; 2000) it is claimed that the computational system of the human
language (CHL) is a perfect system. A computational system is perfect when it meets the
Inclusiveness Condition (IC). IC states that any structure formed by the computation is
constituted of elements already present in the LIs selected. This means that no new
syntactic object can be added in the course of the derivation. Hence, the introduction of
indices in the course of the derivation violates IC7. Therefore, given a set of (empirical)
proofs8, Chomsky (1995) claims that binding conditions can apply only at the Conceptual-
Intentional (C-I) interface. In other words, since computation (narrow syntax) is a closed
system, which operates only with the material selected from the lexicon, binding
conditions must apply at the interface between the syntactic component and the
interpretative component (broad syntax).
The claim that binding conditions must apply at C-I interface leaves us with two
open questions: (i) since binding is defined as co-indexation under c-command and co-
indexation is not available anymore, how are we to define binding? (ii) Since binding
conditions imply locality conditions (namely, the notion of GC defined above), how is
locality to be defined at C-I interface? Both questions involve the same problem: how can
we define purely syntactic notions at the C-I interface?
Let us start asking how binding can be defined at C-I interface. At the level of C-I
interface the interpretative system is called for interpreting the logical forms computed by
7 Reinhart (2000) claims correctly that syntactic co-indexation is just a technical device with no psychological reality. Higginbotham (1983) notes that co-indexing is a mere typographical means of expressing the antecedence relation. 8 See Chomsky (1995:191-212).
26
CHL. At this level referential dependencies are encoded and binding can be
straightforwardly understood as ‘logical binding’ (Reuland 2001a:440). Logical binding,
as defined by Reinhart (2000), is the procedure of closing a property. A widely assumed
technical implementation of this procedure is obtained by binding a free variable to a λ-
operator. The definition of ‘logical binding’ is given in (1):
(1) Logical/variable Binding
α binds β iff α is the sister of a λ-predicate whose operator binds β9
The differences between logical and syntactic binding are clear. Logical binding is a
relation between operators and variables while syntactic binding is a relation between
indices (i.e. variables).10
Once binding at C-I interface is defined, binding conditions should apply at this
level. In line with Reuland (2001a), I will show that this is not possible since logical
binding is blind to locality conditions. Binding conditions, in fact, imply notions of
9 Reinhart (2000) calls the relation (1) A(rgument)-binding, to avoid confusion with just the standard logical binding. As pointed out by Reinhart, the term A-binding is not to be confused with the term used in syntactic binding theory, in which A-binding means binding between A-positions. Note that this definition of logical binding has interesting consequences: (i) it entails that α c-commands β; the welcome result in this case is that c-command follows straightforwardly from the definition, but has no independent role in defining binding; (ii) it eliminates the need to assume something like Condition C. In fact, we expect logical binding to respect the standard laws of logical syntax. Hence, only variable can be bound and just by one operator. Referential expressions are not variable. Wh-traces are already bound, since they result from movement. 10 Higginbotham (1983) points out that co-indexing is a problematic notion since it abstracts from the direction in which the anaphor-antecedent relation was assigned; in his article he proposes to represent the assignment of the anaphor-antecedent relation between two positions in a syntactic structure by linking those positions with a headed arrow, whose head points to the antecedent. Linking has the advantage of being an anti-symmetric relation. Note that the definition of binding just given has also the advantage of being an anti-symmetric relation, in fact it is a relation between an operator and a variable; it is then clear-cut the distinction between binder and bindee. Thus, this conception of binding solves also the problem posed by Higginbotham.
27
locality such as minimal domain or GC. These notions can be defined only in terms of
syntactic structures but not in logical terms.
That logical binding is blind to locality can be proved by the fact that it is not
sensitive to syntactic islands. This example should be sufficient:
(2) a. Miriam thought that John would tell Brian that the room Frank reserved
was supposed to be for her mother
b. Miriam λx (x thought that John would tell Brian that the room Frank
reserved was supposed to be for x’s mother)
The conclusion is that locality requirements, such as minimal domain or Governing
Category, find no natural expression in a logical syntax conception of binding since
binding at C-I interface can be understood only as a logical notion, not sensitive to
locality. Logical Forms, which are the objects taken by the interpretative system at C-I
interface, cannot express locality requirements.
1.2.3 Logophoric use of anaphors
A problem for CBT is the so-called ‘logophoric use of anaphors’. Recall that Condition A
of CBT states that an anaphor must always be bound in its local domain, i.e. it cannot
occur free and it cannot be bound by an antecedent outside its local domain. This is not
always true.
There are well known cases of logophoric use of anaphors. Very roughly, an
anaphor is used logophorically when, in particular contexts, it is used as a free variable
28
(Bonato 2006:37). The notion of logophoric pronoun is due to Hagège (1974) and
Clements (1975). Clements (1975:171) characterizes logophoric pronouns as elements that
satisfy conditions (i) to (iii):
(i) logophoric pronouns are restricted to reportive contexts transmitting the words or
thought of an individual or individual other than the speaker or narrator;
(ii) the antecedent does not occur in the same reportive context as the logophoric
pronoun;
(iii) the antecedent designates the individual or individuals whose words or thoughts
are transmitted in the reportive context in which the logophoric pronoun occurs.
Note that CBT does not prevent logophoric use of pronouns. In fact Condition B states
that a pronoun is free (i.e. not bound) in its Governing Category; this means that it can be
bound outside its local domain or it can correspond to a free variable. For instance,
sentence (3) can be mapped into two logical forms: (3a) and (3b):
(3) Frank thinks that Miriam loves him
a. Frank (λx (x thinks that Miriam loves x))
b. Frank (λx (x thinks that Miriam loves y))
In (3b) the pronoun can be used logophorically and, in this case, it is interpreted by means
of discourse factors.
29
Condition A, on the other hand, does not allow logophoric use of anaphors.
However, the are well known examples in the literature:
(4) Max saw a gun next to himself (Reinhart and Reuland 1993)
(5) Physicist like yourself are a godsend (Ross 1970)
(6) She gave both Brenda and myself a dirty look (Zribi-Hertz 1989)
In example (4) the anaphor himself clearly refers to Max, although it is not bound within
its governing category. In examples (5) and (6) an antecedent is lacking at all. Intuitively,
in (5) the anaphor refers to the hearer (yourself); in (6) it refers to the speaker (myself).
Other cases of logophoric use of anaphors, which I want to present in more details,
come from Icelandic. The following examples were first introduced by Thráinsson
(1976)11; in all these sentences sig, an anaphor, refers to Jón, which is outside the
anaphor’s GC. The main difference between the sentences in (7) and the sentences in (8)
is that in (7) sig is embedded in a subjunctive clause while in (8) it is embedded in a
infinitive clause:
(7) a. [Skoδun Jóns] er [aδ sig vanti hæfileika]
Opinion John’s is that SIG lacks talents
‘John’s opinion is that he (= John) lacks talents’
b. [Álit Jóns] virδist [vera [aδ ég hati sig]]
Belief John’s seems be that I hate SIG
11 See also Thráinsson (1990; 1991).
30
‘John’s belief seems to be that I hate him (= John)’
(8) a. Jón skipaδi mér [aδ PRO lemja sig]
‘John ordered me to hit SIG’
b. Jón skipaδi Pétri [aδ PRO raka sig á hverjum degi]
‘John ordered Peter to shave SIG every day’
Two groups of proposals have been developed in the literature to account for the
anaphoric use of sig. One group aims to give a unified syntactic account of sig.12 The
other group claims that the use of sig is ruled by discourse factors rather than syntactic
principles.13 However, both approaches treat the long distance uses of sig in subjunctive
and infinitive clauses as instances of the same phenomenon.
At this point, I do not intend to propose a solution; rather, I want to analyze the
differences between the sentences in (7) and those in (8), following Reuland and
Sigurjónsdóttir (1997). Let us start with subjunctives. In subjunctives, sig can take as its
antecedent a non c-commanding NP. However, the antecedent of sig must be the person,
distinct from the speaker, whose point of view is reported in the sentence.14 On the other
hand, sig shows a different behavior in infinitives clauses. In this case, sig can only take as 12 Along this first line different analysis has been proposed. Anderson (1986) postulates an ‘anaphoric domain’ in which sig has to be bound by a super-ordinate subject. Pica (1987) proposes that sig moves out of infinitive and subjunctive clause at LF level. 13 Along this second line are Thráinsson (1976, 1990, 1991); Mailing (1984; 1986); Sells (1987); Sigurδsson (1990) and Sigurjónsdóttir (1992). 14 This is proved by the agrammaticality of sentences like (i): (i) *[Vinur Jóns] telur [aδ ég hati sig] Friend John’s believes that I hate SIG ‘John’s friend believes that I hate him (=John)’ Compare it with (7). Other examples and relevant discussion can be found in Mailing (1984:222) and in Sigurδsson (1990:335).
31
antecedent the subject of the matrix sentence.15 Moreover, sig in infinitives cannot take a
non-c-commanding NP as antecedent, even if it is a possible perspective holder. The
antecedent of sig in infinitives is subjected to structural conditions and its interpretation is
not ruled by discourse factors. In the case of infinitives, the anomaly is that the antecedent
is outside the GC. Hence, it is an instance of long-distance binding16, rather than
logophoric use. All in all, it seems reasonable to think that sig is ruled by discourse factors
when it occurs in subjunctives and by syntactic principles when it occurs in infinitives.
CBT cannot account for the phenomena just presented. In particular, all instances
of logophoric use of anaphors and long distance binding contrast with Condition A’s
claim that anaphors must be bound within their local domain. In the case of logophoric
use, an anaphor is not syntactically bound by an antecedent. In the case of long distance
binding, an anaphor is bound outside its local domain. Are these phenomena to be
considered exceptions to the theory or do we need to propose a different theory of
anaphoric relations that can explain in terms of general principles these phenomena? In the
following section I will expose how Reinhart and Reuland has pursued this second line in
their works. Before that, I want to point out some other phenomena that CBT leaves
unexplained.
15 Compare (i) with (8): (i) *Ég hótaδi Jóni [aδ PRO lemja sig] ‘I threatened John to hit SIG (=John)’ 16 For an introduction to long-distance binding and relevant discussion on the subject see, among others, Giorgi (1983/1984; 2004).
32
1.2.4 Locally bound pronouns
Condition B of CBT states that a pronoun must be free, i.e. not bound, in its Governing
Category. However, also in this case there are counterexamples. The first kind of
counterexample was first introduced in the literature by Reinhart (1983:169):
(9) I know what Bill and Mary have in common. Mary adores Bill and Bill adores him
too.
In sentence (9), the second occurrence of him must admit Bill as its value. This means that
him’s antecedent is within him’s GC, contravening Condition B.
Another kind of counterexample comes from Frisian. This language allows
pronouns instead of anaphors in local binding relations. In fact, in Frisian, sentence (10)
means ‘Willem shames himself,’ although the Frisian him is a pronoun.
(10) Willem skammet him
The same sentence in Dutch would require a SE anaphor (zich), but, since Frisian does not
have SE anaphors, it allows a pronoun (him), violating again Condition B.
1.2.5 Complementarity
Condition A and Condition B of CBT predict strict complementarity in the distribution of
pronouns and anaphors. It means that the same interpretation cannot be obtained using
33
either an anaphor or a pronoun. However, there are well known contexts where
complementarity breaks down. In the following examples both an anaphor and a pronoun
can occur, producing grammatical sentences with same meanings:
(11) a. Frank saw a gun near himself/him
b. Miriam counted three tourist apart from herself/her
(12) a. Miriam saw a picture of herself/her
b. Frank likes jokes about himself/him17
These examples must show a violation of at least one of the canonical binding conditions.
1.3 A theory of anaphoric relations based on economy
principles
1.3.1 Preliminary questions
As we have seen above, some languages allow logophoric use of anaphors. An anaphor is
used logophorically when it is not bound by an antecedent and it is interpreted by means
of discourse factors. Logophoricity is a rather puzzling phenomenon for CBT. In fact,
within CBT the logophoric use remains an exception to be explained.
17 Note also that the anaphors used in these examples represent instances of logophoric use of such elements.
34
The problem raised by logophoricity can be understood in two different ways. A
first line is to assume statement (1), known as the ‘deficiency thesis,’ and then ask
question (2).
(1) In order for an argument α to be interpreted, it must have a full specification for φ-
features. (Bouchard 1984)
(2) Why, under certain conditions, is logophoric interpretation of anaphors allowed?
Statement (1)’s consequence is that referentially defective arguments, as anaphors, are
forced to take a linguistic antecedent to be interpreted. If (1) is assumed to hold,
logophoricity must be understood as an exceptional phenomenon which, under certain
conditions, can be allowed.
A second approach is to assume (3) and ask (4).
(3) If α has fewer φ-features than β, there are fewer constraints on the interpretation of
α than on the interpretation of β.
(4) Why, under certain conditions, is logophoric interpretation of anaphors blocked?
Statement (3) entails that there is no intrinsic property of anaphors that prohibits an
unbound interpretation. As a consequence a logophoric interpretation is always possible
but blocked under certain conditions. The goal of the theory is then to describe these
circumstances.
35
Questions similar to (4) can be addressed, given the other counterexamples to CBT
raised in the preceding section. Take for instance the Frisian sentence (5), interpreted as
(5a):
(5) Willem skammet him
a. Willem λx (x skammet x)
Sentence (5), interpreted as (5a), is a grammatical sentence of Frisian. Again, instead of
asking why under certain circumstances local binding of a pronoun is allowed, we can
assume that there is no intrinsic property of pronouns that prohibits a bound interpretation
and ask (6).
(6) Why, under certain circumstances, is local binding of a pronoun blocked?
Let me sum up questions (4) and (6) in just one more basic question:
(7) Why, under certain circumstances, a certain interpretative procedure is blocked?
The theory I’m going to expose provides an answer to such a question.
1.3.2 Interpreting procedures and economy principles
How is a pronominal element, which lacks lexical features, to be interpreted? Consider the
tree structure σ in (8).
36
(8)
We will focus on an argument β in δ. If α has lexical features it will be directly interpreted
as a suitable element in the domain of discourse. If β lacks of lexical features it functions
as a semantic variable and it can be interpreted, according to Chierchia (1995), by means
of two strategies. It can be taken as a free variable or as a bound variable. In the first case,
α takes as its value an element in the discourse domain. In the second case, α is bound by
an operator and it receives its value from that operator. The difference between the two
strategies can be seen in example (9):
(9) Miriam thinks that she is smart
a. Miriam (λx (x thinks that y is smart))
b. Miriam (λx (x thinks that x is smart))
The pronominal expression she is interpreted as a free variable in (9a) (it is not bound by
the λ-operator) and as a bound variable in (9b) (it is bound by the λ-operator). In the
second case, when the open λ-expression receives Miriam as its argument all the
occurrences of the bound variable receive Miriam as their value.
37
Now, suppose expressions α and β, in σ, have to be assigned the same value. This
goal may be reach, according to Reuland (2001b:353), by means of three different
operations:
(10) a. α and β may be members of the same syntactic chain;
b. α and β may be identical variables bound by the same operator;
c. α and β may independently be assigned the same object from the domain of
discourse.
If α and β are members of the same syntactic chain, they are interpreted as occurrences of
the same variable and the chain represents one semantic object. If α and β are bound by
the same operator, they receive the same value; this is what happens in example (9b). If α
and β are different variables nothing prevents them form receiving the same value; this is
what can happen in (9c) if y receives Miriam has its value18.
Still according to Reuland (2001a; 2001b), I will assume that the three operations
in (10) pertain to three different cognitive components or different modules of human
communicative competence. Rule (10a) pertains to the syntactic component (i.e.
Chomsky’s CHL, see Chomsky 1995). Rule (10b) pertains to the semantic component.
Rule (10c) operates at the discourse level19. The central point in the theory I am going to
18 Formally: (i) Miriam (λx (x thinks that y is smart)) & y=Miriam 19 We are actually claiming that there exists a separate semantic component among our cognitive faculties. Note that, in the works of both Reinhart and Reuland, the semantic component is taken to be part of the computational system of human language in its broad sense (in Chomsky’s terms: broad syntax). Hence, we assume a first opposition between computational system (broad syntax) and discourse level, and a second opposition, within broad syntax, between the proper syntactic
38
present is that the choice between these three rules is governed by economy principles.
The idea can be understand considering scheme (11):
(11)
a. Discourse level (objects) A
Semantic level (varables) x1
Syntactic level (chains) C1 (Op. 3a) C1
Basic expressions α … β
b. Discourse level (objects) A
Semantic level (varables) x1 (Op. 3b) x1
Syntactic level (chains) C1 C2
Basic expressions α … β
component (narrow syntax) and the semantic component. The first opposition can be understood in terms of psychological modularity (as, for instance, in Fodor 1983; for the purposes of the present discussion, let us take a module as an encapsulated computational system, relatively domain-specific and characterized by a monotonic-logic, opposed to the central systems characterized by the opposite features): thus, broad syntax pertains to the cognitive module of language, while discourse level operations rely on general cognitive resources. The second opposition can be characterized by the fact that only narrow syntax obey to IC (Inclusiveness Condition); according to Chomsky (1995:219) a language L is a generative procedure that construct pairs (π; λ), where π provides the instruction for the sensorimotor system and λ the instruction for the system of thoughts; Reuland (2001a:471) assumes that CHL (narrow syntax) generates pairs (π′; λ′) which obey IC, and that the C-I interface contains a translation procedure that maps λ′ into λ, where λ differs from λ′ in that it does not obey IC. (It is naturally assumed that something similar must happen at A-P (articulatory-perceptual) interface). Some of the rules of the translation procedure, actually just those concerning the present discussion, will be illustrated in (22).
39
c. Discourse level (objects) A (Op. 3c) a
Semantic level (varables) x1 x2
Syntactic level (chains) C1 C2
Basic expressions α … β
The down-top arrows represent cross-modular operations, i.e. operations that map objects
relevant at a certain level into object relevant at the higher level. For instance, a cross-
modular operation is the operation that translates a syntactic chain into a semantic
variable. Each cross-modular operation is associated with a certain cost. This means that
different interpretative strategies would have different costs. This is represented in (11).
Let us start with (11a); the goal is to assign the two basics expressions α and β the same
value; α and β are linked in the same chain at the syntactic level; therefore, the chain is
translated in one semantic variable (first cross-modular operation), which receives as a
value one object of the discourse domain (second cross-modular operation). The result is
achieved: α and β have received the same value. The rule used was (10a) and it applied at
the syntactic level. The total cost was of two cross-modular operations (as indicated by the
down-top arrows). Consider now (11b); α and β are translated as variable bound by the
same operator (operation 10b). The result is achieved again, but the cost was of three
cross-modular operations. Finally, in (11c), α and β represent two syntactic object mapped
into two variables which receive the same value at discourse level. This procedure
required four cross-modular operations.
40
The cheapest way to assign a value to some element α is to do it at the level of
syntax; the costliest way is to do it at discourse-level. There is only one last thing left
unexplained: which principle governs the choice between the interpretative procedures
available. I give in (12) a very informal version of such a principle:
(12) An interpretative procedure is blocked whenever a cheaper procedure that leads to
the same interpretation is available.
We have seen above that the question this theory should answer is why, under certain
conditions, an interpretative procedure is blocked. (12) answers such a question.
At this stage, our theory is composed by: (i) three possible operations for encoding
linguistic dependencies, operating at distinct levels, and (ii) a principle of economy. In the
following discussion I will analyze in more technical details these notions.
1.3.3 Encoding linguistic dependencies
In the present framework, linguistic dependencies can be encoded at three distinct levels:
(i) syntactic level; (ii) semantic level; (iii) discourse level. Each level encodes linguistic
dependencies by means of different operations: (i′) syntactic chain; (ii′) variable binding;
(iii′) co-reference. In this section I shall give a detailed account of these operations.
1.3.3.1 Syntactic encoding
1.3.3.1.1 Encoding a syntactic dependency
41
So far, we have considered syntactic encoding of a linguistic dependency as chain
formation. A chain is formed by the operation Move, which moves a LI or some of its
features to a new position in the syntactic structure. The material moved in the new
position and the material left in the base position (a copy or a trace) form a chain. A chain
is one syntactic object which is mapped into one semantic object, i.e. one argument.
This kind of syntactic dependency is not the one we are looking for. In fact the
chain configuration, as just exposed, cannot establish a dependency between two lexical
items. Rather it establishes a relation between a moved lexical item and its trace or copy.
The kind of syntactic configuration we are looking for must be able to relate two lexical
items, since the goal of the operation is to assign the same value to two expressions. Given
two expressions α and β, we need a certain syntactic relation R such that α R β forms one
syntactic object. This relation is formalized by Reuland (2001a) and it is provided by the
interaction between two basic operations of CHL, namely Move/Attract and Check. In MP,
these two operations are defined as follows:
(13) Move/Attract
K attracts F if F is the closest feature that can enter into a checking relation with a
sublabel of K. (Chomsky 1995:297)
(14) a. A checked feature is deleted when possible.
b. A deleted feature is erased when possible. (Chomsky 1995:280)
42
(14) means that checking takes place as soon as a checking configuration is established.
The difference between deletion and erasure is that deleted features can be recovered
while erased features cannot. The expression ‘when possible’ is to be understood as ‘when
it does not contradict the principles of UG.’ The principle of UG that must be respected in
this case is the Principle of Recoverability of Deletion (PRD), which requires that no
information be lost by the operation (see Chomsky 1965:182 and Chomsky 1995:44). The
consequences of the PRD are, according to Chomsky (1995:280-1), that (i) only
uninterpretable features can be erased and (ii) interpretable features cannot be deleted.
One of Reuland’s main claims (Reuland 2001a) is that the second consequence is not
correct. These are his considerations: (i) the core syntactic dependency is that between two
occurrences of a feature, one being deleted by the other (let us take ‘occurrences’ of a
feature two copies of the same feature within a checking configuration); this is Chomsky’s
(1995, 2000) basic idea behind Move/Attract, i.e. movement is triggered by the need to
eliminate (erase) redundant features; (ii) there is features’ redundancy every time two
occurrences of a feature enter a checking configuration; (iii) suppose an interpretable
feature F1 enters a checking relation with an occurrence of the same feature, call it F1′, by
Move/Attract; at this point, Checking applies blindly, as all the other operations within
CHL; F1′ deletes F1, and this is possible because deleted features can be recovered. This is
Reuland main assumption: interpretable features can be deleted and then recovered
without contradicting the PRD, and it is exactly this process of deleting/recovering that
expresses the presence of a linguistic dependency.
1.3.3.1.2 chain, Chain and CHAIN
43
Let me explain, with an example, how the idea just exposed works. Consider the Dutch
sentence in (15):
(15) Oscar voelde zich wegglijden
Oscar felt SE slide-away
Then consider the stage of the derivation exposed in (16):
(16) [Oscar [[I,I voelde I] [tOscar [tv [zich I [wegglijden]]]]]]
This structure is derived by first adjoining voelde to I and then moving Oscar to [spec, IP].
In covert syntax, i.e. after Spell-out, zich’s Formal Features (FFzich) must move covertly
for Case checking of the verb. The only place for FFzich to move is the matrix V/I complex
giving rise to structure (17)20:
(17) [Oscar [[I,I FFzich [I,I voelde I]] [tOscar [tv [zich I [wegglijden]]]]]]
FFzich enters a checking configuration with FFOscar. FFzich are just D and 3rd person and
they are both interpretable. Moreover, both features are also included in FFOscar. Given our
assumptions on Checking and the PRD, these features are occurrences of the same feature,
i.e. they are copies of the same feature within a checking configuration; hence, FFzich will
20 As noted by Chomsky (1995:304), the trace of the moved V is not a possible target for FFzich. The reason is that the verb undergoes pied-piping, since it moves before Spell-out, and it does not leave functional material in its base position. Functional material could constitute a possible target for FFzich. This problem is still subject of debate.
44
be deleted and then recovered under identity, since they are interpretable features. As we
have seen, the process of deleting/recovering expresses the presence of a linguistic
dependency, and this is exactly the case. Let us call the relation between FFzich and Oscar
‘checking chain,’ abbreviated as ‘Chain’ and formalized as follows:
(18) Chain
(α, β) forms a Chain iff
a. β’s features have been (deleted by and)21 recovered from α, and
b. (α, β) meets standard conditions on chains such us uniformity, c-command,
and locality.
It is worth noticing that this definition has no theoretical cost. The Chain relation is just
the result of the blind/automatic application of the rules of CHL. In other words, (18) does
not introduce a new operation within CHL, rather it describes a certain configuration that
can be obtained given certain LIs and the basic rules of computation.
Structure (17) shows two syntactic dependencies. The first dependency is a chain
relation between FFzich and zich; this relation is the result of Move/Attract. The second
relation is a Chain relation between Oscar and FFzich; this relation is the result of
Checking. Given that these two relations can undergoes composition, in a way similar to
the chain linking proposed by Chomsky and Lasnik (1993), we can allow linking between
Chains and chains. Call the resulting relation CHAIN:
21 Since there is no recovery without previous deletion, the ‘deleted by’ part of the definition can be omitted.
45
(19) CHAIN
If (α1, α2) is a Chain and (β1, β2) is a chain and α2=β1, then (α1, β2) is a CHAIN.
Remember what we are looking for: we need a certain syntactic relation R such that α R β
forms one syntactic object. CHAIN is that relation. Oscar and zich enters a CHAIN
relation. As noted in the case of (18), also definition (19) has no theoretical cost, for the
same reasons exposed above for Chain.
1.3.3.1.3 Interpreting CHAINs
So far, two questions remain open: (i) which elements can enter a CHAIN relation and (ii)
how a CHAIN is interpreted by the semantic component.
Let us start with the first question: which elements can enter a CHAIN relation?
Only SE anaphors. To show that the answer is true, two proofs are in need: (i) a proof that
pronouns and referential (and quantified) DPs cannot enter a CHAIN relation; (ii) a proof
that SELF anaphors cannot enter a CHAIN relation. Let us analyze pronouns in terms of
formal features. The formal features of a pronoun are category, person, gender and
number. The formal features of a SE anaphor are just category and person. Category,
person and gender are inherent features. Number, on the other hand, as a nominal feature,
is both interpretable and optional (i.e. it is not inherent). Number is to be considered as a
lexical element; hence different occurrences of it in the numeration cannot be taken as just
copies of each other. Every occurrence of the number feature will give a context
dependent contribution to interpretation. Therefore, an occurrence of a number feature
cannot be deleted and recovered on the basis of another occurrence of the same feature
46
and, consequently, a Chain relation cannot be established. If a Chain relation cannot be
established, also a CHAIN relation cannot. The same point can be made for referential
(and quantified) DPs.
What about SELF anaphors? We start distinguishing two types of SELF anaphors
on the basis of their structure:
(20) a.
b.
(20a) is the case of the Dutch zichzelf, which is composed of a SE anaphor and a self
component. (20b) is the case of the English himself or the Italian lui stesso, which are
composed of a pronoun and a self component.
Just to clarify some terminology I shall use, with the term SELF I intend anaphors
with structures as either (20a) or (20b), as opposed to SE anaphors such as Dutch zich;
with the term self I intend the self component in the NP embedded in the SELF anaphor,
as in structures (20a) and (20b). Henceforth, a SELF anaphor can have a structure
47
[pronoun[self]], as in English, or a structure [SE[self]], as in Dutch; what both structures
have in common is the self component.
Structure (20a) shows a SE anaphor embedded in a DP; nothing prevents the SE
anaphor from entering a CHAIN relation when the necessary structural conditions are met.
In other words, the SELF anaphor cannot enter a CHAIN relation, but the SE anaphor
embedded in it can. This not the case of (20b) where the pronoun bears a number feature.
We can turn to the second question: how is a CHAIN interpreted? Consider a
structure like (21a), where DP heads a CHAIN tailed by SE and P is a 1-place predicate:
(21) a. DP P′ SE
b. DP (λx (x P))
c. DP (λx (x P x))
According to Reuland (2001a:477) a structure like (21a) can be translated into a semantic
object by means of two strategies. According to the first strategy, a CHAIN can be fully
reduced to a chain and, consequently, the chain 〈DP, SE〉 is translated as one semantic
object. As I understand it, this procedure produces a LF such as (21b). According to the
second strategy, a CHAIN represents a syntactic dependency which is translated as
indicated by the translation rules listed in (22).
(22) a. An expression DP[…t…] is translated as DP (λx (…x…)); vblDP is the
variable of the λ-expression derived by translating DP;
48
b. A SE anaphor is translated as a variable; vblSE is the variable that the
translation assigns to SE;
c. If XP is CHAIN-linked to DP, their φ-features are copies. If so, vblXP =
vblDP;
d. A pronoun is translated as a variable.22
These rules produce the LF (21c). This second strategy is the one chosen by Reuland.
Given that P′ is a 1-place predicate, (21c) is correct only under the assumption that the two
occurrences of x in the expression P(x, x) are interpreted as one argument.23 Consider
sentence (23):
(23) a. Oscar schaamt zich
b. Oscar (λx (x schaamt x))
22 As explained in a preceding note, these rules are among the rules pertaining to the C-I interface that translate λ′ (an LF representation which obeys IC) into λ (an LF representation which does not obey IC). 23 According to Reuland, (i) can be reduced to (ii): (i) DP (λx (x P x)) (ii) DP (λx (x P)) In other words, Reuland assumes that (i) and (ii) are identical. It is not clear to me, if the wanted result is (ii), why Reuland chooses the second strategy, while, using the first strategy, (ii) could have been obtained straightforwardly without assuming that (i) and (ii) are identical. In the present discussion I will stick to Reuland’s assumption, leaving the possibility to implement the alternative strategy to further research. The assumption that (i) can be reduced to (ii) will have critical consequences on what follows, and giving it up would require a deep revision of the whole theory. Just for the sake of speculation, note that under the assumption that (i) can be reduced to (ii), a LF like (iii) is problematic; in fact, the two occurrences of x should be treated as copies, as they are in (i), and they should constitute one argument. (iii) Frank (λx (x thinks that x is stupid)) Reuland solves the problem assuming that the semantic component operates compositionally. The compositional mechanism is illustrated in the discussion of example (24).
49
Since Oscar and zich stand in a CHAIN relation, given rules (2), (23a) is translated into
(23b). According to the preceding assumption, the two occurrences of x are copies and,
therefore, they form one argument. Schamen is a 1-place predicate, hence the arity of the
predicate is respected and the sentence is correct.
What about sentence (15), repeated here as (24a)?
(24) a. Oscar voelde zich wegglijden
b. Oscar voelde [τ zich wegglijden]
The verb voelde is a two-place predicate. However, we have seen that zich enters into a
CHAIN relation. The solution proposed by Reuland is exposed in (24b). The idea is that
the two arguments of voelde are Oscar and τ. This is possible only assuming a
compositional interpretative procedure. In fact, Oscar and zich forms a CHAIN, hence a
syntactic object and a semantic argument. However, zich must saturate wegglijden and
then, in CHAIN with Oscar, it must saturate one of voelde’s arguments. Very roughly, it
seems that zich is implied, one way or another, with three arguments at the same time. It is
the argument of wegglijden; it is, in CHAIN with Oscar, the first argument of voelde; it is,
together with wegglijden within τ, the second argument of voelde. Reuland guarantees that
a compositional procedure solve the problem. First, zich is combined with wegglijden,
forming τ; hence, wegglijden’s arity is respected. Then, τ combines with voelde saturating
one of its arguments. Finally, Oscar, in CHAIN with zich, saturates voelde’s second
50
argument (and it is at the moment of the CHAIN formation that zich is translated
according to Rule (22c)).24
Finally, what about reflexivized 2-place predicates? Within the proposal I’m
presenting 2-place predicates can occur only with SELF anaphors. Let us see why by
means of some examples:
(25) a. *Willem haat zich
‘William hates SE’
b. *Willem (λx (x haat x))
(26) a. Willem haat zichzelf
‘William hates SELF’
b. Willem haat [τ zich [N zelf ]]
c. Willem (λx (x haat f(x))
(25a) is ill formed since its semantic counterpart, (25b), shows an arity violation. In fact,
the two occurrence of x are copies and they form one argument. On the other hand, (26a)
is well formed. Structure (26b) is very similar to (24b), in the sense that zich enters a
CHAIN relation with Oscar but the predicate sees as its arguments Willem and τ. Reuland
(2001) proposes that zelf functions as a protective element whose role is to preserve the
arity of the predicate. The LF of (26a) is then (26c). The predicate takes two arguments,
one identical to x, the other to the value of some function of x.
24 In line with this argument, Fox (1993) proposes that chains may share their members with different predicates. According to Reuland (2001a:478 note 40) the present approach derives rather than stipulates this.
51
This proposal works for SELF anaphors embedding SE anaphors. However in
some languages, as English and Italian, SELF anaphors embed pronominal forms. In this
case self has a ‘reflexivizing’ rather than protective function. Self is a relational noun
interpreted as an identity relation that produces LFs such as (27a) or (27b):
(27) John hates himself
a. John (λx (hates (x, him) & him = x))
b. John (λx (him λy (hates (x, y) & y = x)))
The function of self is to put two arguments of the same predicate in a relation of identity.
The idea is that, given the structure (28a), self undergoes covert movement and adjoins to
the predicate, as shown in (28b):
(28) a. DP … [P] [pronoun [self]]
b. DP … [self P] [pronoun [e]]
According to Reuland, the fact that self moves onto the predicate head explains locality. In
fact, the relation is possible only between arguments of the same predicate. A predicate
with more than one argument is then necessary. The result of this operation is that the
arity of the predicate is not violated.25
25 In Reinhart and Reuland (1993) it is assumed that only SELF anaphors can reflexive mark a predicate (of course, just in the case the predicate is not inherently/lexically reflexive marked). The account just given of SELF anaphors (both for SELF anaphors embedding a SE anaphor and for SELF anaphors embedding a pronoun) gives an explanation of their ability to license reflexivity. In fact their role is to avoid arity violation, a role that cannot be covered by SE anaphors.
52
1.3.3.1.4 Theoretical remarks
The syntactic relation CHAIN is derived from the basic properties of CHL. In fact it is
based on the blind application of basic computational operation (Move/Attract and
Checking) and on a purely morpho-syntactic lexicon. Nothing as indices is implied and IC
is thus respected. However, we had to consider a slightly modified version of the PRD.
1.3.3.2 Variable binding
In the preceding sections we have already introduced the notions of logical binding (and
its difference from syntactic binding) and co-reference. Here, I will just summarize the
basic concepts.
The distinction between variable binding and co-reference is due to Reinhart
(1983). Consider sentence (29):
(29) Frank thinks that he is clever
a. Frank (λx (x thinks that x is clever))
b. Frank (λx (x thinks that y is clever))
Pronouns, according to the translation rule (22d), are translated into variables. So is he in
example (29). In (29a) the variable is bound by a λ-operator while in (29b) it is not. In this
second case the variable must receive its value from the discourse storage for the sentence
to receive a truth-value.
53
Variable binding is a semantic operation while co-reference applies at discourse
level. We have already seen in the previous section how variable binding is defined; I
repeat the definition as (30).
(30) Variable Binding
α binds β iff α is the sister of a λ-predicate whose operator binds β
Variable binding is just the procedure of closing a property. It is a relation between an
operator and a variable.
1.3.3.3 Co-reference
Two expressions co-refer if they refer to the same entity of the discourse domain.
Consider, for instance, a sentence like (31a) in the context (31b):
(31) a. Lili thinks she’s got the flu
b. Lucie didn’t show up today. Lili thinks she’s got the flu
c. Binding Lili (λx (x thinks x has got the flu))
d. Co-reference Lili (λx (x thinks y has got the flu)) & y = Lucie
As it is clear, she in (31b) refers to Lucie rather than Lili. The right LF in this case is (31d)
where only x is bound by the λ-operator. The other variable, y, which cannot be bound by
Lucie lacking the right configuration, will receive its value at the discourse level, picking
up a female singular entity which has been introduced, one way or another, in the
54
discourse storage. In the case of (31b) the entity Lucie has been introduced by another
sentence ‘Lucie didn’t show up today,’ and it has become the focus of the conversation.
Co-reference is possible also in out-of-context sentences, for instance in (32):
(32) Most of her friends adore Lucie
Here, her is interpreted as Lucie, although it is not c-commanded. Since the structural
requirements are lacking, variable binding is not an option, and her must receive its value
at the discourse level. Then, nothing prevents her to take Lucie as its value by mean of co-
reference.
Co-reference is defined as follows:
(33) Co-reference
α and β co-refer iff neither binds the other and they refer to the same discourse
entity
An interesting case is presented in (34):
(34) a. Frank thinks that he is clever
b. Frank (λx (x thinks that x is clever))
c. Frank (λx (x thinks that z is clever)) & z = Frank
d. Frank thinks that he is clever and Bill does too.
e. Frank (λx (x thinks that x is clever)) and Bill (λy (y thinks that y is clever))
55
f. Frank (λx (x thinks that z is clever)) and Bill (λy (y thinks that z is clever))
& z = Frank
Consider (34a): the reading in which he is taken as Frank can be obtained either by
variable binding (34b) or through co-reference (34c). What is the difference between these
two LFs, if there is any? According to Grodzinsky and Reinhart (1993), in the case of
variable binding, it is the property of thinking of oneself as clever which is attributed to
Frank, while, in the case of co-reference it is attributed to Frank the property of thinking
of Frank as a clever person. This difference, which is invisible in the case of (34a)26,
becomes visible in the case of elliptic VPs. Consider sentence (34d). It can have two
readings. The first reading, the so-called sloppy reading, is obtained copying (34b) and it
means that ‘Frank thinks that Frank is clever and Bill thinks that Bill is clever.’ The
second reading, the identity reading, is obtained copying (34c) and it means that ‘Frank
thinks that Frank is clever and Bill thinks that Frank is clever.’
1.3.4 Economy principles
So far I have analyzed in technical terms three possible tools for encoding linguistic
dependencies: CHAIN, which operates at syntactic level; variable binding, which operates
at semantic level; and co-reference, which operates at discourse level. The theory I am
26 As we will see later, the LF (33c) is ruled out by Rule I, at least if we take Rule I as stated in (36a); however, it is a question why (33c), although it is ruled out, is still available for copy in the case of (36f); this problem as been faced by Reinhart (2000) which proposes a modified version of Rule I. I shall present an account of this problem in the next section.
56
presenting requires also an economy principle, which was informally stated in (12),
repeated as (35).
(35) An interpretative procedure is blocked whenever a cheaper procedure that leads to
the same interpretation is available.
Three principles have been proposed in the literature. They are shown in (36):
(36) a. Rule I (Interface Rule)
(Reinhart 1983; Grodzinsky and Reinhart 1993)
α cannot co-refer with β if replacing α with γ, γ a variable bound by β,
yields an indistinguishable interpretation.
b. Rule BV (Bound variable Rule)
(Reuland 2001a)
α cannot be bound by β if replacing α with γ, γ such that β heads a CHAIN
tailed by γ, yields an indistinguishable interpretation.
c. Rule L (Logophoric interpretation Rule)
(Reuland 2001a)
α cannot be used logophorically if there is a β such that a CHAIN 〈β, α〉
can be formed.
Very roughly: Rule I states that co-reference is blocked if variable binding is available;
Rule BV states that variable binding is blocked if a CHAIN relation is available; Rule L
57
states that logophoric use is blocked if a CHAIN relation is available. Before considering
how these rules work through some examples, let me point out a couple of remarks. The
technical implementation of these rules and their psychological reality will be subject of
further discussion; however, it is clear by now that these rules are to be considered
economy conditions in the sense that they deal with a notion of global economy rather
than computational economy. I will leave these problems to further discussion.
I want to show now how the theory can account for Condition A, Condition B and
Condition C of CBT, and also for logophoric and long-distance anaphors, locally bound
pronouns, lack of complementarity in the distribution of anaphors and pronouns.
1.3.4.1 Condition A
Consider the following examples:
(37) a. Frank hates himself
b. Frank (λx (hates (x, him) & him = x))
c. Frank (λx (him λy (hates (x, y) & y = x)))
d. Frank haat zichzelf
e. Frank (λx (x haat f(x)))
(38) a. *Frank thinks that Miriam loves himself
b. Frank (λx (x thinks that Miriam (λy (y loves (x, him) & him = y))))
Cases like (37a) and (37d) have already been discussed in a previous section. The
interpretation of (37a) follows straightforwardly from the syntax and semantic of self. As
58
we have seen above, self covertly to moves to V. The result is an identity relation between
him and the other predicate’s argument, namely Frank, as shown by the LFs (37b and c).
In languages like Dutch, where SELF anaphors are composed by a SE anaphor and a self
element (see (20) for a detailed representation), the dependency is still encoded within
syntax but in a different manner. Zich, in fact, tails a CHAIN headed by Frank. The
dependency is mapped into the semantic component according to the translation procedure
(22), producing (37e).
The agrammaticality of (38) is easily explained considering that self moves to the
predicate head. Since it is an argument of the predicate love, it moves to this predicate
producing an identity relation 〈him, Miriam〉, whose members do not agree in gender.
Similar considerations could account the Dutch counterpart of (38) with the only
difference that instead of self moving to V, we would have zich tailing a CHAIN headed
by Miriam. In this case, since zich does not have gender features, the sentence would be
correct but it would mean that ‘Frank thinks that Miriam love herself.’
1.3.4.2 Condition B
Consider the following examples:
(39) a. Frank hates him
b. Frank (λx (x loves x))
c. Frank (λx (x loves y)) & y = Frank
(40) a. Frank thinks that Miriam loves him
b. Frank (λx (x thinks that Miriam λy (y loves x)))
59
c. Frank (λx (x thinks that Miriam λy (y loves z))) z = Frank
d. Frank (λx (x thinks that Miriam λy (y loves y)))
Consider example (39). We want to block the reading in which him refers to Frank. This
reading can be obtained by co-reference, as in (39c). Co-reference is ruled out by Rule I.
In fact, (39b), the bound variable option, and (39c) lead to undistinguishable
interpretations. Finally, (39b) is ruled out since it violates the arity of the predicate, being
love a 2-place predicate. In the case of English Rule BV, does not rule out (39b), since
English does not have SE anaphors in its lexicon, as Frisian.
On the other hand, in (40) him can refer to Frank. However, Rule I rules out (40c)
the co-reference option, since (40b) and (40c) leads to indistinguishable interpretations.
The bound variable option is not ruled out since arity is respected. Just to see Rule BV in
action, imagine that English, like Dutch, has SE anaphors in its lexicon. Bound variable
would not be ruled by rule BV, because the interpretation we would get with a CHAIN
would be (40d), which is not undistinguishable from (40c).
1.3.4.3 Condition C
Consider the following examples:
(41) He loves Frank
(42) He thinks that Miriam loves Frank
60
Condition C is derived directly from the definition of logical binding. An R-expression is
not translated into a variable then it can not be bound by an operator. However, it is
interesting noticing that Rule I does not block co-reference, which, as we have seen in the
preceding sections, is an option to consider. Take for instance a sentence like ‘he is John,’
where, clearly, he refers to John.
1.3.4.4 Locally bound pronouns
Consider the following example:
(43) a. Willem skammet him
b. Willem (λx (x skammet x))
As we have seen in the preceding section, Frisian shows cases of local binding of
pronouns. Consider that skammen is a 1-place predicate (in fact the Dutch verb schaamen
requires zich and CHAIN formation). However, Frisian, like English, has no SE anaphors
at disposal. In the case of (34) Rule BV checks the lexicon for an element that can enter a
CHAIN relation, but it does not finds such elements since the Frisian lexicon does no have
SE anaphors. Then Rule BV does not rule out (43b) since an alternative with CHAIN
formation is not available. (43b) respects the arity of the predicate, since, as we have
assumed, two occurrences of the same variable are just copies, hence they form one
argument.
61
1.3.4.5 Logophoric and long distance anaphors
Consider the following examples:
(44) a. [Skoδun Jóns] er [aδ sig vanti hæfileika]
Opinion John’s is that SIG lacks talents
‘John’s opinion is that he (= John) lacks talents’
b. [Álit Jóns] virδist [vera [aδ ég hati sig]]
Belief John’s seems be that I hate SIG
‘John’s belief seems to be that I hate him (= John)’
In the previous section I illustrated, in line with Reuland and Sigurjónsdóttir (1997), that
sig in subjunctives is interpreted logophorically. This follows straightforwardly once we
have seen that in subjunctives sig cannot tail a CHAIN. We assume, according to Manzini
(1993), that subjunctives must be licensed by an operator and this operator attracts the V/I
complex. Since the licensing relation must be syntactically encoded, the V/I complex must
move to the licenser, as shown in (45), where Op is the relevant operator and δ is the
starting position for the V/I movement:
(45) (a) Op [IP Oscar [[I,I FFsig [I,I V subj I]] [tOscar [tv [sig…]]]]]
(b) [Op,Op [I,I FFsig [I,I FFsubj I]] Op] [IP Oscarφ… [[δ] Vsubj]…sigφ…]
62
This operation moves FFsig from the position where they can enter in Chain relation with
Oscar. Then a CHAIN cannot be formed. Given Rule L, nothing prevents a logophoric use
of sig in subjunctives since it cannot enter a CHAIN relation.
Other cases of logophoric use of anaphors were raised in the previous section.
They all dealt with SELF anaphors. It has been noticed (Reinhart and Reuland 1991; 1993
and Pollard and Sag 1992) that there is a systematic pattern in the distribution of
logophoric anaphors in English:
(46) a. Physicist like yourself are a godsend
b. *A physicist has looked for yourself
(47) a. She gave both Brenda and myself a dirty look
b. *She gave myself a dirty look
(48) a. Max boasted that the queen invited Lucie and himself for a drink
b. *Max boasted that the queen invited himself for a drink
The correct generalization is that SELF anaphors cannot occur logophorically when they
are an argument of the predicate (see the (b) cases). They can be interpreted
logophorically when they are embedded in an argument, but they do not constitute an
argument by their own. This behavior can be explained given the semantic of self as a
relational noun. In the (b) cases self try to produce an identity with a predicate’s argument.
Take (46b): the predicate is look which takes two arguments, a physicist and yourself; self
produces an identity relation 〈you, a physicist〉; the relation is then ruled out for reasons of
person’s agreement. In the (a) cases, the SELF anaphor is not a predicate’s argument. It
63
seems plausible that in such cases self cannot construe a relation since it takes no part in a
predicate-arguments relation. The interpretation then must be carried by means of
discourse factors. Rule I does not rule out co-reference since, lacking c-command, variable
binding is not an option.
1.3.4.6 Complementarity
Consider the following examples:
(49) Physicist like you/yourself are a godsend
(50) She gave both Brenda and me/myself a dirty look
(51) Max boasted that the queen invited Lucie and him/himself for a drink
We have just seen that in (49-50), anaphors are used logophorically. The same argument
explains why also pronouns can occur in the same environments. Co-reference is not
blocked since variable binding is not available.
1.4 Discussion
1.4.1 Theoretical considerations
Independently from its empirical adequacy, the theory of anaphoric relation based on
economy principles meets explanatory adequacy. It respects IC since it does not use
indices. The syntactic operations (in particular the CHAIN formation), result form the
blind and automatic application of computational operations and reflect the combinatorial
64
properties of a purely morpho-syntactic vocabulary. The morpho-syntactic vocabulary is
explained in terms of primitive features. Furthermore, the theory does not rely on locality
conditions, as CBT did; ‘c-command is relevant for the syntactic conditions under which
λ-predicates can be formed (compositionally), but has no independent role in defining
binding’ Reinhart (2000:7).
However, the theory relies on some non-trivial assumptions. One assumption,
which is directly related with the explanatory adequacy of the theory, is that the basic
economy principles (Rule I, Rule BV and Rule L) are primitives, in the sense that they are
not reducible to more basic properties of the human language’s cognitive system. A
possible reduction is the object of further investigations. Nevertheless, I would focus on a
couple of points.
1.4.2 Rule BV
Rule BV states that variable binding is blocked if, substituting a lexical item with a new
lexical item that can enter a CHAIN relation, leads to the same interpretation. Take an
example from Dutch:
(4) *Willem schaamt hem
Rule BV explains why hem cannot be interpreted as a variable bound by Willem.
According to this rule, we check the lexicon for an element that can enter a CHAIN
relation leading to the same interpretation. The Dutch lexicon has such an element, namely
zich. Variable binding is then blocked. But, if we move to Frisian things work differently.
65
In fact, the Frisian counterpart of (4) would be correct, because Frisian does not have in its
lexicon something like zich, i.e. it has no lexical item that can tail a CHAIN.
The logic of Rule BV and Rule L are absolutely different. If a SE anaphor is
introduced in the numeration and at a certain stage of the derivation it enters a CHAIN,
this is just a matter of computational principles and not of economy principles. If it does
not enter a CHAIN then it must interpreted one way or another, maybe as a bound
variable, maybe as a free variable, i.e. logophorically.
Rule BV is a principle of global economy, as opposed to computational economy.
To appreciate the comparison, let me quickly explain what computational economy is. We
start with a numeration, i.e. a set of LIs for the computation. The computation takes these
LIs one by one until the derivation is concluded. If the derivation does not converge (i.e. it
cannot be interpreted by the interfaces) it crashes. However, it is plausible to think that
given the same numeration different converging/interpretable derivations can be
computed. In this case only the cheapest derivation would survive. The cheapest
derivation is the derivation that requires the minor number of computational steps. What
matters, then, is that economy considerations can be done on the basis of the same
numeration. The numeration is the reference-set for economy conditions to apply. Given
two derivations, we can establish which one is the cheapest only if they are computed
from the same numeration.27
This is not the case for Rule BV. Rule BV must basically confront two derivations
based on different numerations. Consider again example (4). Rule BV must compare a
derivation based on a numeration including hem with a derivation based on a numeration
27 For relevant discussion on the subject see Chomsky (2000).
66
in which hem as been substituted with zich. As guaranteed by Reuland, there is no risk of
a computational blow-up. This is Reuland’s point (2001a:475),:
The features of a SE anaphor are a subset of the features of the corresponding pronominal.
This entails that the comparison does not require selecting a different lexical item, but that
the SE anaphor involved in the comparison can be construed from information already
available in the pronoun itself.
I argue that there is something wrong in Reuland’s statement. Specifically, that the SE
anaphor can be construed from information already available in the pronoun itself.
However the computation must also know if a SE anaphor is available in the lexicon. Let
me explain this point with an example. Suppose Reuland’s statement is correct; what
happens in the case of (4)? At a certain point in the derivation the process checks whether
a subset of hem’s features can enter a CHAIN relation with Willem. The answer is ‘yes,’ a
subset hem’s features can effectively enter a CHAIN relation with Willem. As a
consequence the bound variable interpretation is blocked. Now take the Frisian (5), which,
remember, means ‘Willem shames himself.’
(5) Willem skammet him
Rule BV works the same way. At a certain point in the derivation the process checks
whether a subset of him’s features can enter a CHAIN relation with Willem. The answer
will be ‘yes,’ since a subset of him’s features can effectively tail CHAIN. The difference
between Dutch and Frisian is that in Dutch the subset of hem’s features corresponds to a
67
LI listed in the Dutch lexicon while in Frisian the subset of him’s features does not
correspond to a LI listed in the Frisian lexicon. If we want the correct result, Rule BV
must check the lexicon and see if it contains a SE anaphor. Then the question would be:
does the lexicon contain a SE anaphor? In the case of Dutch (4) the answer would be ‘yes’
and variable binding would be blocked; in the case of Frisian the answer would be ‘no’
and variable binding would be allowed. Then, to account for the Frisian facts, Rule BV
should state that ‘α cannot be bound by β if replacing α with γ, a γ such that β heads a
CHAIN tailed by γ and such that it correspond to an actual lexical item, yields an
indistinguishable interpretation.
However, as noted by Delfitto (p.c.), the fact that Rule BV requires a ‘look into the
lexicon’ does not lead to a computational blow-up. It is possible to implement
computationally the rule without troubles. Just think of an algorithm that, in the case of a
pronoun in a certain syntactic configuration, calls for a look into the lexicon.
Now consider a language like English, which does not have SE anaphors. In
English CHAIN formation is not an option. If we claim that Rule BV is a linguistic
universal, it is difficult to see why English should use such a rule. Moreover, as far as
English is concerned, the theory predicts some wrong judgments. Consider the following
examples:
(6) a. John behaves himself
b. *John (λx (x behaves him) & him = x)
(7) a. *John behaves him
b. John (λx (x behaves x)
68
(6a) is a grammatical sentence of English. However, the LF (6b) produced by the
reflexivizing self shows an arity violation. In fact, behave is a 1-place predicate, inherently
reflexive. (7b), on the other hand, is an ungrammatical sentence of English although the
theory would predict its grammaticality. English, like Frisian, has no SE anaphors, so no
chance to build a CHAIN. Rule BV then should not block the bound variable reading (7b);
moreover, the bound variable option produces a LF that respects the arity of predicate,
given that the two occurrences of x form one argument. One possible move to rule out (7)
is to assume again a version of Rule BV in which it is not necessary to check the lexicon.
If the rule just checks the possibility for a subset of him’s features to tail a CHAIN,
without asking if the subset corresponds to a actual LI, then the bound variable
interpretation is ruled out. Yet, this is not enough for explaining the grammaticality of (6)
and, moreover, we fall back to the Frisian case which, conversely, needs the ‘check into
the lexicon.’
1.4.3 Rule I
Also Rule I deals with global economy since it does not operate within one specific
linguistic module, rather it chooses between the derivations produced by different
modules. Its computational implementation is not an easy task. We must deal with two
problems. A first problem is that there are contexts in which Rule I is systematically
violated. Take example (8):
(8) a. Frank thinks that he is clever
69
b. Frank (λx (x thinks that x is clever))
c. *Frank (λx (x thinks that z is clever)) & z = Frank
d. Frank thinks that he is clever and Bill does too.
e. Frank (λx (x thinks that x is clever)) and Bill (λy (y thinks that y is clever))
f. Frank (λx (x thinks that z is clever)) and Bill (λy (y thinks that z is clever))
& z = Frank
Rule I does not allow (8c) as a possible interpretation for (8a), however in the ellipsis
contexts both the bound variable interpretation and the co-reference interpretation are
available for copy. In fact (8d) is ambiguous between (8e) and (8f).
The second problem is how to computationally implement Rule I. This problem
has also been noted by Reinhart (2000); she points out that it is not clear whether the
human processor is indeed sensitive to the type of economy considerations involved in the
processing of Rule I. She concluded that this economy view has not been on the right
track.
An alternative, proposed by Reinhart, is to turn the rule the other way around. Rule
I basically aims to formalize the intuition that co-reference is blocked whenever an
equivalent variable binding reading is possible. The alternative is that co-reference is
excluded whenever an equivalent variable binding interpretation is impossible. It is
formalized as follows; I will call this rule Rule I*28:
28 Reinhart (2000) calls this rule just Rule I. Her definition differs from the one given in (9) in that she uses ‘co-valuation’ instead of ‘co-reference.’ I leave the discussion of these two concepts apart, but see Heim (1993) and Reinhart (2000).
70
(9) Rule I*
α and β cannot co-refer iff:
a. α is in a configuration to bind β;
b. α cannot bind β;
c. the co-reference interpretation is indistinguishable from what would be
obtained if α binds β.
If just one of the three clauses in (9) is violated, then the co-reference option is possible.
Take (8): (i) Frank is in a configuration such that it can bind he, then clause (9a) is
satisfied; (ii) Frank can bind he, then clause (9b) is not satisfied. The co-reference option
is hence available. Note that we did not need to check clause (9c). This solves the problem
of (8) but still we have to face the problem of the computational implementation of clause
(9c). Consider example (10):
(10) She said we should invite Lucie
a. She (λx (x said we should invite Lucie)) & she = Lucie
b. She (λx (x said we should invite x)) & she = Lucie
We want to block co-reference between she and Lucie. Condition (9a) and (9b) are
satisfied: she is in a configuration to bind Lucie and she cannot bind Lucie since Lucie is
not a variable. Then we need to check also clause (9c), and for this we need to construct a
reference-set including the current derivation and the derivation with the variable binding
71
interpretation. If the two members are indistinguishable, the co-reference option is
blocked. The procedure is specified as follows (Reinhart 2000):
(11) To check clause (9c) construct a comparison-representation by replacing β with a
variable bound by α.
In the case of (10) also condition (9c) is satisfied, then the coreference option is rule out.
We should expect that such a complex strategy, as the one exposed in (11), would
correspond to a visible processing cost. The data from applied linguistics seem to confirm
this expectation29.
1.5 Conclusions
CBT encounters both theoretical and empirical problems. The theoretical problem is that
CBT does not attain explanatory adequacy and violates the IC. Among the empirical
problems, particular attention was given to the phenomenon of logophoric use of
anaphors. To overcome such problems I proposed to follow the road paved by Reinhart
and Reuland, and then to pursue a deep revision of the theory of anaphoric relations.
To overcome the theoretical problems we substituted the notion of syntactic
binding, with three different mechanisms for encoding a linguistic dependency between
two expressions:
29 For relevant discussion and references on the subject see, among others, Avrutin (1999, 2004), Baauw (2000), Baauw and Delfitto (1999),Grodzinsky (2000a), Grodzinsky and Reinhart (1993), Manzini and Wexler (1987), Piñango (2002), Piñango and Burkhard (2001), Wexler and Chien (1991).
72
(1) Syntactic encoding (i.e. CHAIN formation)
(2) Variable-binding
(3) Co-reference
The syntactic mechanism proposed does not violates the IC since it does not entail the use
of indices. Moreover, it results from the blind and automatic application of primitive
operations of the computational system, i.e. movement and checking. The structural
conditions on CHAIN are straightforwardly derived from the structural conditions on
these primitive operations.
The mechanism of variable-binding is a relation between operators and variables
and it applies at the C-I interface. As is expected at this level the mechanisms is blind to
locality, hence not sensitive to syntactic island.
As far as the logophoric use of anaphors is concerned, we have seen that this
phenomenon turns out to be a problem under the assumption that the deficiency thesis of
Bouchard holds. According to this thesis, anaphors are referentially deficient expressions
since they lack a full specification of φ-features, hence they must be bound by an
antecedent to be interpreted. If we take this thesis as correct, logophoricity turns out to be
an unexplained exception. Henceforth, according to Reuland (2001b) we assumed that the
deficiency thesis is incorrect: the fact that an expression lacks a full specification of φ-
features simply means that there are fewer constraints on its interpretation. Under this
assumption, the logophoric interpretation is the ‘default’ option and the question we must
answer is why under certain conditions logophoric interpretation is blocked. This question
73
finds its answer in the basic economy principles characterizing the theory. Such economy
principles state that a certain interpretative procedure is blocked whenever the same
interpretation can be achieved by means of a cheaper interpretative procedure. Therefore,
the logophoric interpretation is blocked whenever the same interpretation can be obtained
by means of variable-binding or CHAIN formation.
The economy principles, namely Rule I and Rule BV, are the primitive principles
at the basis of the theory. They are principles of global economy since they work
comparing two derivations based on two distinct numerations. Economy conditions that
does not take as their reference set the initial numeration, i.e. the initial array of lexical
items, are seen as dangerous since they could raise the risk of a computational blow up.
This problem has been considered by both Reinhart (2000) and Reuland (2001a). It seems
to me that Reinhart’s proposal is on the right track. Her modified Rule I, which I dubbed
Rule I* (see example (9) in the previous paragraph), calls for the substitution of a lexical
item only in few cases. Moreover, the formalization of a comparison procedure, as in
example (11) in the previous paragraph, provides a precise instruction for the
computational system.
The interesting result of this theory is that different anaphoric relations are
associated with different computational costs. It is an intriguing subject for further
research, to check the data from applied and clinical linguistic and see if different
computational costs are correspond to different processing costs in production,
comprehension and acquisition. Moreover, since different operations are said to pertain to
different linguistic levels, we should expect subjects with specific linguistic impairment to
fail in specific tasks and succeed in others. For instance, a well-founded hypothesis is that
Broca’s patients are limited in their abilty to construct syntactic dependencies in real-
74
time30, hence we expect them to fail in the comprehension of those anaphoric relations that
requires computation at the syntactic level (for example, in the case of Rule BV).
30 This is known as the ‘Slow Syntax Hypothesis,’ see for instance Piñango (2002) and Avrutin (2006) and references cited.
75
2 ATTITUDES DE SE
2.0 Introduction
It has been noted in the philosophical literature (Castañeda 1966; Kaplan 1977; Lewis
1979) that sentences like (1) are ambiguous:
(1) John believes that he is a musical genius
Consider the interpretation in which he is anaphoric of John. Such a sentence is
ambiguous since it can mean that John believes that he is a genius and he is aware of the
fact that the belief he is entertaining is about himself, but it can also mean that John
believes that he is a genius although he is not aware of the fact that the thing he is having a
belief about is actually himself. To appreciate this second interpretation consider the
following context: John reads an article on a newspaper and he believes that the journalist
who has written that article is a genius; however, John is not aware of the fact that he
himself wrote the article; in fact, he has written that article years before but now he has no
recollection of that. From a philosophical point of view, the difference between the two
interpretations stems from the different epistemic access that John has to himself. In both
cases John is having a belief about a certain res, namely John, but just in one case he
76
recognize that res as himself. Yet, in spite of this difference, both situations can be
reported by sentence (1).
It is an interesting question, and, I believe, an intriguing challenge for a theory of
anaphora, whether this ambiguity has some linguistic relevance. If we assume that the
ambiguity has linguistic relevance we must show that the two distinct interpretations
correspond to two distinct logical forms. In other words we must show how the ambiguity
is encoded in grammar. This possibility has not received much attention in the linguistic
literature, except from the texts that will be discussed here and few others31. However, I
believe it can be an interesting ground on which a theory of anaphora can be tested.
In this chapter I will not propose an original theory or an enrichment of an already
existing theory. Rather, I will try to provide a detailed account of the phenomenon and of
the relevant literature on the subject. The problem of the ambiguity of sentences like (1)
has been first brought to the attention of the linguistic community by Gennaro Chierchia
in his article Anaphora and Attitudes De Se (1989). In this article, he analyzes some
instances of the phenomenon under consideration and proposes a theory based on an
attributive account of embedded clauses. Very roughly, he proposes that in the case of a de
se belief, i.e. a belief in which the subject is aware of the fact that the res he is having a
belief about is himself, what the subject does is to attribute to himself a property. For
instance, John, in the de se interpretation of sentence (1), attributes to himself the property
of being a genius. This idea was first introduced by Lewis (1979), who demonstrated by
means of some thought experiments that propositions have not enough structure to
31 Reinhart (1990) and Higginbotham (1992) are discussed in the present chapter. Relevant contributions to the study of the subject, although not discussed here, are Schlenker (2003) and Higginbotham (2003).
77
represent essential indexicality. It means that a proposition P(ε) is true whether or not ε is
aware of having the property P. For instance, the proposition ‘John is a musical genius’
does not necessarily entail the proposition ‘I am a musical genius’ uttered by John. Hence,
Lewis proposed that in the case of de se beliefs what is at hand is a property rather than a
proposition. Chierchia (1989), capitalizing on Lewis’s idea, proposes a linguistic theory
according to which the two interpretations of sentences like (1) correspond to two distinct
logical forms. Chierchia’s theory makes two very interesting predictions: (i) that
controlled structures, like infinitivals, are always interpreted de se and (ii) that logophors
are always associated with a de se interpretation.
Objections have been raised against Chierchia’s theory. In this chapter I will
consider two articles: Reinhart (1990) and Higginbotham (1992). Reinhart raises two
empirical problems. A first problem is that Chierchia’s theory predicts that the sloppy
reading of a deleted VP is always interpreted de se. As shown by Reinhart, this is a wrong
prediction since also deleted VPs can be interpreted de re even when they are assigned a
sloppy reading. A second problem is that it is not true that logophors are always associated
with a de se interpretation. Reinhart considers some examples, taken from written
narrative, of logophors occurring in structures that Chierchia’s theory cannot account for.
Henceforth, Reinhart proposes an alternative theory, which is based on the work of
Kaplan.
Objections of a slightly different kind are raised by Higginbotham (1992). He
shows that (i) an attributive account of embedded clauses and, particularly, of controlled
structures runs against serious empirical problems and (ii) that attitudes de se simply
results from a different strategy of interpretation of pronouns. It must be noted that
78
Higginbotham has serious doubts not only about the linguistic relevance of attitudes de se,
but also about their philosophical relevance. In my presentation of Higginbotham’s article
I will devote some attention to the theory of controlled structures that the author proposes
since it conctitutes the necessary background for understanding his objections and his
proposals.
The chapter is organized as follows: in the first paragraph I will introduce and
explain with some examples the distinction between attitudes de dicto and attitudes de re
(2.1.1) and the distinction between attitudes de re and attitudes de se (2.1.2). In the second
paragraph I will introduce Chierchia (1989)’s theory. In the third paragraph I will discuss
Reinhart (1990)’s objections to Chierchia (2.3.1) and I will present her alternative
proposal (2.3.2). In the fourth paragraph I will discuss Higginbotham (1992)’s article with
particular attention to his theory of control structures; in a final appendix to the paragraph
I will propose some lines of further investigation in the theory of control. Finally, in the
concluding notes I will try to summarize the empirical evidence against Chierchia’s theory
of attitudes de se, introducing some original counterexamples to that theory.
2.1 Attitudes de dicto, de re, de se
2.1.1 Attitudes de dicto and attitudes de re
It has been noticed in the philosophical literature (Castañeda 1966; Kaplan 1977; Lewis
1979) that there maybe a significant difference between a thought one is entertaining
about a certain res and a thought one is entertaining without referring to a certain res. To
appreciate this difference consider the following example (Chierchia 1989):
79
(1) Pavarotti believes that the one who can sing ‘O che gelide manine’ without
mistakes is a musical genius
Sentence (1) is ambiguous since it can be true in two distinct sets of circumstances. In the
first case, Pavarotti’s belief is purely conceptual; he does not know if there is a singer that
can sing ‘O che gelide manine’ without mistakes but he does know that, if there is one,
that singer is a musical genius. In the second case, Pavarotti knows that Domingo is the
only singer that can sing ‘O che gelide manine’ without mistakes; then, what Pavarotti
means with sentence (1) is that he believes that Domingo is a musical genius.
In the first case, Pavarotti’s belief is called de dicto. In the second case, Pavarotti’s
belief is called de re.
2.1.2 Attitudes de re and attitudes de se
It has also been noticed that, within de re beliefs, two different kinds of belief may be
further distinguished. Suppose that the res one is having a belief about is her/himself; two
cases can then be distinguished: on one hand the subject can just know the res she/he is
having a belief about; on the other hand the subject can know the res and also recognize
that res as her/himself. In the first case the belief would be de re, in the second case it
would be de se.
Note that in both cases the belief is about a certain res; then, a de se belief entails a
de re belief; in fact, only if we are having a belief about a certain res we can recognize
that res as ourselves. On the contrary, the opposite entailment does not hold, i.e. a de re
80
belief does not entail a de se belief; the fact that we are entertaining a belief about a
certain res does not entail that we recognize that res as ourselves.
The most famous example of ambiguity between a de re and a de se reading is due
to Kaplan (1977). Consider this scenario: John is looking in the mirror and sees someone
whose pants are on fire; the person he sees is actually himself. Then, sentence (2) can be
true in two different sets of circumstances:
(2) John believes that his pants are on fire
On one hand, John is aware that his belief concerns himself. In this case John’s belief
would be something like the direct speech assertion ‘my pants are on fire.’ On the other
hand, John may not be aware of the fact that the man he sees is himself. In this second
case his belief would be ‘his pants are on fire.’
In both cases John’s belief is de re. In fact, there exists a specific entity whom John
has a belief about. However, he may still not be aware of the fact that that specific entity is
himself. Only in the case John recognizes that the entity he is having a belief about is
himself, we have, according to Lewis (1979), a de se belief.
Another example of the same ambiguity, although with a more realistic setting, is
due to Reinhart (1990). Suppose Lucie is a broadcast manager and is looking for the right
female voice for a radio commercial. She listens to some samples of female voices.
Unbeknown to Lucie, the sound technician records her too and dubs her sample as number
18. While listening to sample 18, Lucie does not recognize her own voice and rules out
sample 18 because she thinks the woman she has listened to in the sample sounds too
aggressive. In this scenario, sentence (3) cannot be true de se:
81
(3) Lucie believes that she sounds too aggressive
Sentence (3) is true de re, since Lucie knows the res whom she is having a belief about,
but she does not recognizes that entity as herself, hence it is not true de se.
Sentence (3) has the same ambiguity shown in (2). It can present a belief that Lucie
has about a certain entity that she does not recognize as herself, and this belief is de re; but
it can also present a belief that Lucie has about an entity that she fully recognizes as
herself, and this belief is de se.
To sum up, de re attitudes are attitudes that a subject has about a certain entity; de
se attitudes are attitudes that a subject has about a certain entity that she/he recognizes as
her/himself.
2.2 Chierchia’s property-account of attitudes de se
2.2.1 Linguistic relevance of attitudes de se
The de se/de re ambiguity suggests that there is a qualitatively relevant difference between
accessing a certain res and accessing ourselves. Take sentence (1), for instance:
(1) Lucie believes that she sounds too aggressive
In both readings Lucie has a certain belief about a certain res but only in one reading, the
de se reading, she recognizes that entity as herself. A de se attitude can then be seen as a
82
special case of the de re attitude since the epistemic access to oneself necessarily requires
the access to a certain res: the de se reading entails the de re reading.
It is an intriguing question whether the de re/de se ambiguity has some linguistic
relevance and more specifically whether it can be resolved at some level in the grammar.
In effect, two lines of research have been pursued: the first line takes the ambiguity to be
connected to a more general epistemic problem regarding the way one can access a certain
res; since an object can be accessed by different perspectives (even leading to the
attribution of contradictory properties to the same object), the problem of de se attitudes
can be reduced to a more general problem of de re attitudes.32
A second approach to the problem is to assume that the ambiguity has actual
linguistic significance and that it can be resolved at some level in the grammar assigning
distinct logical forms to sentences like (1). Chierchia (1989) pursues this second line: his
aim is to demonstrate that: (i) de se readings are associated with specific linguistic
structures and (ii) the compositional semantics of such structures resorts to properties
rather than propositions.
2.2.2 Lewis (1979): a ‘self-locational’ view of beliefs
The idea that a de se attitude is a relation between an agent and a property, rather than a
proposition, is due to Lewis (1979). Lewis claims that it is in principle possible to know
the truth-value of all the propositions in a given world and still not know the relation
between oneself and these propositions. For any entity ε in a world W, we might know
32 Indeed, this is the position the position taken by Higginbotham (1992) who does not believe that the distinction between de se and de re attitudes has neither philosophical nor truth-conditional relevance; further details will be given below.
83
whether ε has the property P without knowing if we are ε. This is the case of John, who
knows that a certain res has the property of having pants on fire, without knowing that that
res is himself. It is also the case of Lucie, who knows that a certain res sounds too
aggressive without knowing that she is the res in question.
Propositions, according to Lewis, do not have the structure needed to represent
essential indexicality: a proposition P(ε) does not entail that ε is aware of being P. In fact,
for P(ε) to be true, ε must have the property P; but the fact that ε is or is not aware that he
is ε does not change the truth-value of the proposition P(ε). For instance, the truth of the
proposition ‘Lucie sounds too aggressive’ does not necessarily entail the truth of the
proposition ‘I am too aggressive’ uttered by Lucie.
Lewis proposes that the mechanism underlying de se beliefs is that of ‘self-
ascription’ of a property. A property is an open formula, an unsaturated expression which
cannot receive a truth-value unless it is saturated by an argument. A property must contain
an open position, i.e. a variable. If ε believes de se that ψ, then ε is attributing to
her/himself the property of being in a world W where ψ is true. Take, for instance, the de
se reading of (1), in which Lucie is aware of the fact that the person sounding too
aggressive is herself: according to Lewis, Lucie attributes to herself the property of
sounding too rude.
2.2.3 Chierchia (1989): properties versus propositions
Lewis’s proposal forces us to interpret that-clauses as properties, a move that, somehow,
runs against our pre-theoretic intuitions. Chierchia’s proposal (1989:6) is to take an
intermediate position: that-clauses normally denote propositions, but in certain cases, i.e.
84
when de se reports are involved, they may denote properties. In formalized terms, the two
readings of (1), de se and de re, correspond to the logical forms in (2):
(2) a. Lucie λx ( believe (x, x sounds too aggressive))
b. believe (Lucie, λx (x sounds too aggressive))
(2a) represents the de re reading of (1). Here the predicate believe is a relation between an
agent and a proposition, which turns out to be about Lucie. On the other hand, (2b)
represents the de se reading, where believe is a self-ascriptive relation, in the sense of
Lewis (1979), between Lucie and a property, i.e. the property of being too aggressive.
Note that a relation towards self entails a relation towards a certain res, and not
vice versa. In fact, the truth of a de se belief entails the truth of a de re belief but not vice
versa. A self-ascriptive relation is then defined as follows (Chierchia 1989:7):
(3) A relation of x to a property P is self-ascriptive iff it entails a de re relation to the
proposition P(x) without being entailed by it.
According to Chierchia, although this definition will be assumed in what follows, it
should be regarded as provisional.33
33 Chierchia (1989:7-9) suggests different strategies to implement a definition of self-ascriptive relation such that it yields as a consequence that the truth of a de se belief entails the truth of a de re belief. One way (not far from Cresswell (1985)’s speaker-centered worlds discourse semantics) is to adopt (i) and (ii): (i) x stands in a self-ascriptive relation with the property P iff x believes de re that P and furthermore K(x, x), where K is the cognitive access we have to ourselves. (ii) K(x, x) =df x is disposed to describe the relevant belief by referring to x by means of the first person pronoun.
85
2.2.4 The grammar of de se
Chierchia assumes that semantic entities must result from a non-trivial relation with
syntactic structures. That is, the logical forms (2a) and (2b) must be mapped from
different syntactic representations. It is very important to note that the possibility of a de
se reading is crucially linked to the presence of a pronominal element. Since pronominals
are translated into variables, a sentence containing a pronominal expression can act as an
open/unsaturated formula or, in other words, a property. As we know, variables can get
their values from the context or being bound by an operator. Chierchia’s proposal is that
pronouns can also be bound by a property-abstractor. Technically, the two logical forms
(2a) and (2b), here repeated as (4b) and (5b), are mapped from two different syntactic
structures, respectively (4a) and (5a):
(4) a. Lucie believes [CP that [IP she sounds too aggressive]
b. Lucie λx ( believe (x, x sounds too aggressive))
(5) a. Lucie believes [CP O that [IP she sounds too aggressive]
b. believe (Lucie, λx (x sounds too aggressive))
A possible objection to this definition is that cognitive access to self is taken as a primitive, as in (ii), which could lead to think that creatures with limited symbolic capacity are incapable of de se beliefs. Another way is to consider de re beliefs relative to a certain perspective, as argued by Quine and Kaplan, and de se beliefs as ‘absolute.’ This line would lead to a definition like (iii): (iii) x stands in a self-ascriptive relation R with the property Q iff R(x, Q(x)) (iii) presupposes that the only thing we can access from an absolute point of you is ourselves. However, crucial for Chierchia’s proposal is that attitudes de se are construed from relation between agents and properties. Then, definition (3) would suffice.
86
O, in (5a), is an operator adjoined to CP that binds the pronoun and is in turn controlled by
Lucie. In the case of (4a) the pronoun is directly bound by the c-commanding NP Lucie,
while in the case of (5a) it is bound indirectly via the operator. The operator O is
understood as a λ-abstractor hence (5a) is translated as (5b), which is the desired result.
2.2.5 Infinitives and gerunds: PRO as a λ-abstractor and self-ascription
revised
Chierchia (1984) argues that infinitives and gerunds denote properties. The correctness of
this statement is proved by the following inferences:
(6) a. Pavarotti tried/practiced/began everything that Domingo
tried/practiced/began
b. Domingo tried/practiced/began singing Rigoletto
c. Pavarotti tried/practiced/began singing Rigoletto
(7) a. The cat wanted to eat the cheese
b. The mouse got what the cat wanted
c. The mouse got to eat the cheese (Fodor 1977:142ff)
The validity of the arguments in (6) and (7) follows straightforwardly if we assume that
gerunds and infinitives denote properties. If we assume, alternatively, that they denote
genuine propositions, the inferences in (6) and (7) become difficult. Suppose in fact that
the arguments of tried in (6b) are Domingo and the proposition P = sing(Domingo,
87
Rigoletto). Premise (6a) states that Pavarotti is going to try everything that Domingo tried.
Since what Domingo tried was the proposition P, Pavarotti is going to try the proposition
P as well. The unwanted result is that what Pavarotti tries is P = sing(Domingo, Rigoletto).
(7) represents a similar problem: if want takes a proposition as its complement, the result
of premises (7a) and (7b) would be that what the mouse got was the proposition ‘the cat
eats the cheese.’
If infinitives and gerunds denote properties, as it seems correct, we expect them to
be interpreted unambiguously de se. Chierchia’s proposal is that PRO headed sentences
must be regarded as unsaturated structures and PRO must be interpreted as a λ-abstractor.
This would explain (i) the fact that infinitives and gerunds denote properties and (ii) the
fact that gerunds and infinitives are unambiguously interpreted de se. According to this
proposal, a sentence like (7a) is mapped from the logical form (8b), which in turn is
mapped from the syntactic structure (8a):
(8) a. the cati wants [PROi to eat the cheese]34
b. the cat wants (λx (x eats the cheese))
Sentences like (7a) are unambiguously interpreted de se. The cat wants PRO to eat the
cheese means that the cat wants to bring about a situation where he (and only he) eats the
cheese. John wants PRO to be fired means that John wants to bring about a situation
where he (and only he) is fired. What happens in all these sentences is that a property is
34 Here co-indexing means control. Alternatively, Chierchia proposes that the interpretation of PRO as a λ-abstractor is mediated by an operator at LF, as in (i). (i) the cati wants Oi [PROi to eat the cheese]
88
attributed to an agent. However, it is clear that the relation between the agent and the
property is not of self-ascription, since it is not the subject-agent who does the self-
ascription. For instance, in John persuaded Mary PRO to be fired, it is not Mary who
attribute to herself the property of being fired. Chierchia proposes to keep the label of
‘self-ascription,’ generalizing the definition to every relevant relation:
(9) An n-place relation R is self-ascriptive iff it (asymmetrically) licenses the
following entailment:
R(…x…Q…) → R(…x…Q(x)…)
Just to see an example, the definition in (9) claims that (10a) asymmetrically entails (10b),
i.e. if (10a) is true then (10b) is also true but not vice versa:
(10) a. John wants to be fired
b. John wants to bring about a situation where he is fired
(11) a. John persuaded Mary to be fired
b. John persuaded Mary to bring about a situation where she is fired
Self-ascription is then the semantic translation of syntactic control.
As noted by Chierchia, it is a curious circumstance that English verbs of attitudes,
as believe, do not subcategorize for infinitives. Since these verbs admit properties as one
of their relata we should expect them to allow as arguments those syntactic structures that
89
denote unambiguously properties. However, this is what actually happens in languages as
Italian, where sentences (12) and (13) are unambiguously interpreted de se.
(12) Gianni crede di essere intelligente
‘John believes to be intelligent’
(13) Gianni sa di essere intelligente
‘John knows to be intelligent’
2.2.6 Anaphora and attitudes de se
Attitudes de se are thus always associated with properties rather than propositions.
Henceforth, attitudes de se are also always associated with the presence of a pronoun in
the embedded clause. A sentence like (14), in fact, is not ambiguous between a de se and a
de re reading.
(14) Pavarotti believes that Domingo is a genius
If attitudes de se are associated with properties, rather than propositions, and properties are
unsaturated expressions then the only way to obtain a property is through a bound
variable. As we know variables are the semantic translation of pronouns.
90
The theory of de se attitudes proposed has an interesting connection with the
interpretation of long distance reflexives since they seem to be always associated with a de
se reading. Consider the following examples from Italian35:
(15) a. Pavarotti crede che i suoi pantaloni siano in fiamme. Ma non si è accorto
che i pantaloni sono i propri.
‘Pavarotti believes that his pants are on fire. But he hasn’t realized that the
pants are his own.’
b. # Pavarotti crede che i propri pantaloni siano in fiamme. Ma non si è
accorto che i pantaloni sono i propri.
‘Pavarotti believes that self pants are on fire. But he hasn’t realized that the
pants are his own.’
c. # Pavarotti crede di avere i pantaloni in fiamme. Ma non si è accorto che i
pantaloni sono i propri.
‘Pavarotti believes [PRO to have pants on fire. But he hasn’t realized that
the pants are his own.’
The Italian proprio is a long-distance reflexive, whose interpretation is self-oriented. The
contradictoriness of (15b) confirms that proprio can be associated only with a de se
reading. Moreover, the contradictoriness of (15b) parallels the contradictoriness of (15c).
Thus proprio can be said to patterns like PRO and, perhaps, it can be viewed as the
35 Examples (15a) and (15b) were personally communicated to Chierchia by Bonomi (Chierchia 1989:24).
91
phonologically counterpart of PRO. Chierchia proposes that (16a) has that logical form
represented in (16b), which is mapped from the syntactic-structure (16c).
(16) a. Pavarotti crede che i propri pantaloni siano in fiamme
b. crede (Pavarotti, λx (i pantaloni di x sono in fiamme))
c. Pavarottii crede [CP Oi che [IP i proprii pantaloni siano in fiamme]]
(16b) is a self-ascriptive relation between Pavarotti and the property of having pants on
fire; as a consequence the belief is de se.
Concluding, one of Chierchia’s claims (Chierchia 1989:24) is that self-oriented
long-distance reflexives are a case of pronominal elements associated with a de se
interpretation.
2.3 Reinhart (1990)
2.3.1 Empirical objections to Chierchia (1989)
2.3.1.1 Attitudes de se and variable-binding
Let me briefly summarize Chierchia’s proposal about the mechanism underlying attitudes
de se. Following Lewis, Chierchia takes attitudes de se as relations between agents and
properties, rather than between agents and propositions. As demonstrated by Lewis
(1979), we can in principle know all the true propositions in a given world without
knowing whether we are the agent of those propositions. Then, what happens with de se
92
readings, and only in such cases, is that the agent attributes to himself/herself a certain
property. The sentence in (1a) can thus be interpreted by means of two distinct LFs.
(1) a. Lucie believes that she sounds too aggressive
b. λx (believe (x, x sounds too aggressive)) (Lucie)
c. believe (Lucie, λx (x sounds too aggressive))
The LF (1b) represents the de re reading, where the predicate believe is a relation between
an agent, namely Lucie, and a proposition, which turns out to be about Lucie. The LF (1c)
represents the de se reading, where believe is a self-ascriptive relation between an agent
and a property: in this case Lucie attributes to herself the property of sounding too
aggressive. The relation of self-ascription, which is taken as the linguistic primitive
underlying attitudes de se, is defined as follows:
(2) An n-place relation R is self-ascriptive iff it (asymmetrically) licenses the
following entailment:
R(…x…Q…) → R(…x…Q(x)…)
The conditional guarantees that the de se reading entails the de re reading.
It follows straightforwardly from the theory that attitudes de se are always
associated with the mechanism of variable binding. First, recall that properties are
unsaturated expressions which, once they are assigned an argument, turn into proposition.
It is then clear that properties require an open position; an open position is a variable,
93
which is the semantic translation of a pronominal expression bound by an operator,
technically implemented as a λ-abstractor. This is what we see in (1c), where the pronoun
she is translated into a variable bound by a λ-abstractor.
This analysis has interesting consequences in the case of coordinate structures with
VP deletion. Consider example (3):
(3) Lucie believes that she sounds too aggressive and Lili believes it too
(4) a. Lili believes Lucie sounds too aggressive
b. it = Lucie sounds too aggressive
(5) a. λx (believe (x, x sounds too aggressive)) (Lucie)
b. believe (Lucie, λx (x sounds too aggressive))
c. believe (Lucie, λx (x sounds too aggressive)) & believe (Lili, λy (y sounds
too aggressive))
The question is what Lili believes, a question that can be reduced to the interpretation of
the pronoun it in Lili believes it too. As we know by now, (3) is ambiguous between a
strict and a sloppy reading. Under the strict reading, Lili believes that Lucie sounds too
aggressive. In this case it denotes a proposition, as shown in (4b). Under the sloppy
reading, conversely what Lili believes is that she, i.e. Lili, sounds too aggressive. How is
this interpretation obtained? Suppose the first clause of (3) is analyzed as (5a); the
embedded clause ‘x sounds too aggressive’ cannot serve as a denotation for it since it is
94
just an open proposition whose variable would not be bound by any operator. The only
available alternative is to take (5b) as the LF of the first clause of (3). Here we have a
property ‘λx (x sounds too aggressive)’ that can be taken as the denotation of it. The result
is (5c). Comparing (1c), (5b) and (5c), the conclusion is that the sloppy reading of (3) is
possible only if the first clause of (3) is interpreted de se. As a consequence also the
second clause of (3) must be interpreted de se.
If this line of reasoning is correct we expect that sloppy readings of sentences like
(3) may only be understood de se. In other words, sentence (3) can mean that Lucie
believes that she, Lucie, sounds too aggressive and Lili believes that she, Lili, is too
aggressive but such interpretation is possible only if Lucie’s belief is de se and Lili’s
belief is de se as well. Reinhart (1990) correctly points out that this is mistaken. Suppose
for instance that both Lucie’s and Lili’s voices are recorded. Then suppose that Lucie,
listening to her own voice, believes that it is too aggressive although she does not
recognize that it is her own voice. Suppose also that Lili listens to her own voice and, still
not recognizing her own voice, believes that it is too aggressive. This state of affairs can
be reported by sentence (3) and both Lucie’s and Lili’s beliefs would be de re. In other
words, sentence (3), in its sloppy reading, is ambiguous between a de re and a de se
interpretation in the same way as (1), although Chierchia’s theory predicts only the de se
interpretation. Reinhart (1990) concludes that the anaphoric process illustrated in these
contexts is independent of the de se issue.
The conclusion that the mechanism underlying de se readings must be distinct
from that of variable binding is supported by other facts. Consider the following example:
95
(6) Frank strongly believes that he will be remembered as a war hero
Sentence (6), as the other examples we have faced so far, is ambiguous between a de re
and a de se reading. To appreciate the ambiguity, imagine a context a la Castañeda (1966):
Frank has been a famous war hero but now is a very old man who suffers from amnesia.
In fact, he does not remember anything of his time in the army. One day Frank reads a
book whose subject is Frank himself and his military enterprises, but he does not
recognize the subject of the book as himself. Frank, however, believes that the man who
he has read about will be remembered as war hero and this situation can be reported by the
sentence in (6); Frank’s belief would be de re. The de se reading can be obtained just if
Frank has a strong feeling that he will be remember as a war hero and he is aware of the
fact that the belief is about himself. Recall again that, given Chierchia’s theory, the
presence of a pronominal element in the embedded clause is crucial for expressing a de se
belief. However, if we just turn (6) into a passive sentence, as in (7), a pronominal element
in the embedded clause would be lacking:
(7) That Frank will be remembered as a war hero is strongly believed by him
Reinhart’s example, reported in (8), presents a two-folded problem.
(8) That the president will be remembered as a war hero is strongly believed by him
Things here get more complicated. (8) can of course be ambiguous in the way (6) and (7)
are, but there is more. Consider the context proposed by Reinhart (1990:6): suppose Frank
96
was appointed as president of a veterans’ association as a result of his military enterprises.
He now suffers from temporary amnesia and he remembers only his time in the army but
he is not aware of being president of anything. He strongly believes, given his vivid
memories of the war, that he will be remembered as a war hero. This situation could be
reported with a sentence like (8) and the belief would be de se, since Frank is aware of the
fact that the belief is about himself. However, he is not aware that he is the president. In
other words, the president knows that he is entertaining a certain belief about himself
whether or not he is disposed to refer to himself as the president. The fact that (8) can be
true either in a context in which Frank would refer to himself as ‘the president’ and in a
context in which he would not, simply correspond to the ambiguity de se/de re.36
However, the mechanism of variable binding cannot be responsible of such ambiguity.
The conclusion drawn by Reinhart is that there is no sufficient evidence to assume that de
se readings are allowed only for pronominal elements and the procedure translating belief
complements into properties cannot depend on the presence of a pronominal element.
Henceforth, if the presence of a pronominal element is not linked to the availability of a de
se interpretation, also the mechanism of variable binding turn out to be distinct from the
mechanism underlying de se interpretations.37
36 Note that the context presented guarantees that we are not talking about a de dicto belief. 37 In Chierchia’s theory the fact that the mechanism underlying attitudes de se is that of variable binding explains also why constructions with infinitives and gerunds are always interpreted de se. In fact, PRO is always translated as a variable bound by a λ-operator. The issue is not discussed in Reinhart’s paper but the behavior of such constructions and their property status, represent a central topic in Higginbotham’s (1992) discussion of Chierchia’s article. I will present Higginbotham’s paper in the next section.
97
2.3.1.2 Attitudes de se and long-distance anaphors
Another important claim made by Chierchia (1989), and criticized by Reinhart, is that
logophoric anaphors are always associated with a de se interpretation. Chierchia’s basic
idea, which is drawn from Sells (1987), is that self-oriented long-distance reflexives are
mediated by a certain discourse role; this role is associated with the argument of a relation
whose mental state or attitude is described. In Chierchia’s terms, it means that self-
oriented long-distance reflexives can be only associated with a de se interpretation.
However, there are examples, taken from written narrative, that contradict Chierchia’s
claim. Consider the following examples:
(9) He [Sapp] sat down at the desk and opened the drawers. In the top right-hand one
there was an envelope addressed to himself.
(Zribi-Hertz 1989, quoted from Lodge)
(10) Suddenly he said aloud: ‘Possessiveness is the devil.’ Maggie looked at him. Did
he mean herself and the baby?’ (Zribi-Hertz 1989, quoted from Y. Woolf)
(11) That no one can cook apart form himself is one of Alfred’s recent beliefs
In examples (9) and (10) there is neither a predicate of belief nor a believer argument. The
anaphors refer to a narrative character rather than a believer: in (9) himself refers to Sapp,
which is not entertaining any belief about himself, and, similarly, in (10) herself refers to
Maggie. In example (11) there is a belief that Alfred is entertaining about himself but this
belief is not a complement of a belief predicate.
98
Obviously, the theory proposed by Chierchia cannot deal with such examples. In
particular, its compositional procedure is inapplicable in these cases. It is possible to
translate the clauses embedding the anaphors as properties, but the result would be
uninterpretable. In fact, according to the theory, the property must be a co-argument of the
subject of a belief predicate. For instance, the second sentence of (9) can be translated as a
property, as in (9′), but (9′) alone would be uninterpretable.
(9′) λx (in the top right shelf there was an envelope addressed to x)
Reinhart concludes that to account for these cases a substantial enrichment of the
mechanism would be needed.
2.3.2 Reinhart’s alternative proposal
2.3.2.1 Self-representational relation
The analysis developed by Reinhart (1990) points out several empirical problems for
Chierchia’s theory of attitudes de se. The analysis demonstrates that a property-account of
de se readings lacks independent motivations. Reinhart proposes a theory of attitudes de se
alternative to Chierchia’s theory, based on a unified propositional account of embedded
clauses, an account that would enable us to maintain a unified analysis of tensed clauses.
She then pursues the analysis initiated by Kaplan (1971).
Let us start with attitudes de re. According to Kaplan, a belief sentence reports a
belief of a certain person, that we will call the center (C) of the belief. The belief C is
99
entertaining is about a certain entity (call it ε), independently of the name C would use to
refer to ε. Then Kaplan proposes that a belief report is true de re iff there is a name or
expression by which C refers to ε and C believes the proposition expressed in the sentence
using this name. The idea is captured through a 3-place relation of representation R,
defined as follows:
(12) R (k, ε, C) iff
a. k denotes ε;
b. k is a name of ε for C;
c. C (vividly) associates k with ε.
The meaning of (12) is that a name or expression k represents an entity ε for the center C
if and only if clauses (12a-c) hold. To appreciate this idea, consider sentence (1a),
repeated below as (13a), under its de re reading. Recall the context proposed above: Lucie
listen to sample number 18 and has a belief that the person recorded sounds too
aggressive; however, Lucie is not aware of the fact that the voice she is listening to is her
own voice. Hence, Lucie is both the center of the belief and the entity Lucie is having a
belief about, independently of the name Lucie would use to refer to the entity Lucie. Then
the sentence is true de re if there is a name or expression k by which C (= Lucie) refers to
ε (= Lucie) and C believes that k sounds too aggressive. For instance, Lucie could refer to
the entity Lucie with the expression ‘sample 18.’ Hence, (13a) would be true de re if
Lucie believes that ‘sample 18’ sounds too aggressive. It must be noted that in examples
such (13a) two arguments of the representational relation R happen to be identical,
100
namely, the center (Lucie) and the entity (Lucie). However, nothing changes in the
interpretation of the relation since it holds as long as there is a name by which C refers to
ε, independently of the fact that C and ε are identical. Imagine a context in which every
manager believes de re that she sounds too aggressive. Such a situation could be depicted
by sentence (14a), where (14b) represents the de re interpretation. These examples show
that the identity between C and ε and the interpretation of such identity (by means of
variable-binding or co-reference) is irrelevant for the representational relation.
(13) a. Lucie believes that she sounds too aggressive
b. ∃k (R (k, Lucie, Lucie) & Lucie believes |k sounds too aggressive|)
(14) a. Every manager believes that she sounds too aggressive
b. ∀x (manager (x) (∃k (R (k, x, x) & x believes |k sounds too aggressive|)))
The representational relation R accounts for de re readings only. The identity of two
arguments of R is clearly insufficient for representing de se readings. (13b), for instance,
says that there is a certain name by which Lucie refers to a certain entity. The fact that that
entity turns out to be Lucie herself does not entail that Lucie is aware of such identity. For
representing de se readings, Reinhart introduces a new primitive relation that she calls
self-representational relation. Self-representation (abbreviated as SELF-R) is a 2-place
relation between a center C and a name k that self-represents for a person C. The de se
reading of (13a) is represented in (15).
101
(15) ∃k (SELF-R (k, Lucie) & Lucie believes |k sounds too aggressive|)
Sentence (13a) is true de se, iff there is a name self-representing for Lucie such that she
believes the sentence expressed in a sentence using this name. A self-representing name
for Lucie may be ‘I,’ hence, for the sentence to be true de se, Lucie must believe the
proposition expressed by ‘I sound too aggressive.’
Three considerations are necessary. (i): Reinhart’s proposal parallels Chierchia’s
one, in that it relies on the assumption of a primitive notion. Chierchia, following the path
of Lewis, assumes the self-ascriptive relation, by means of which an agent ascribes to
him/herself a certain property. Reinhart assumes the self-representational relation, by
means of which the center of a certain belief accept to use a certain name to represents
him/herself. (ii): Reinhart’s self-representational relation, as illustrated so far, does not
entail that there must be a referential identity between the denoter and the denoted of the
name k. This entailment is necessary since we must guarantee that the de se reading
entails the de re reading, as Chierchia’s theory did. Hence, following Chierchia’s line,
Reinhart assumes the meaning postulate (16):
(16) ∀x∀k (SELF-R (k, x) → R (k, x, x))
(16) guarantees that a self-representational relation always entails a basic representational
relation, as defined in (12), where the center C and the entity ε are identical. (iii) Also in
the case of SELF-R, the choice of the name k is free as long as k self-represents for C.
102
Reinhart’s proposal can easily be applied to examples like (13), repeated here as
(17a):
(17) a. That no one can cook apart from himself is one of Alfred’s recent beliefs
b. ∃k (SELF-R (k, Alfred) & |no one can cook apart from k| is one of Alfred’s
recent beliefs
c. ∃k (R (k, Alfred, Alfred) & |no one can cook apart from k| is one of
Alfred’s recent beliefs
If, for instance, a name self-representing for Alfred is ‘I’, then the proposition expressed
by ‘No one can cook apart from me’ must be one of Alfred’s recent belief, for the
sentence to be true in its de se reading. Moreover, given the meaning postulate (16), the de
se reading entails the de re reading, i.e. (17b) entails (17c).
Although this line seems to be suitable, at least in principle, for cases like (9) and
(10) repeated as (18) and (19), where no belief-predicate is involved, Reinhart fails to
account for the precise interpretation of these cases (unless postulating some abstract
belief-predicate).
(18) He [Sapp] sat down at the desk and opened the drawers. In the top right-hand one
there was an envelope addressed to himself.
(19) Suddenly he said aloud: ‘Possessiveness is the devil.’ Maggie looked at him. Did
he mean herself and the baby?’
103
The ambiguity of (8), repeated as (20), can be captured assuming that the definite
description ‘the president’ can be translated as a variable existentially bound.
(20) That the president will be remembered as a war hero is strongly believed by him
The role of logophors, in this theory, is understood as a self-representational
relation to the center, rather than a variable-binding relation. In other words, a logophoric
anaphor has the role of self-representing the center.
2.3.2.2 Argument reduction and θ-reduction
According to Reinhart’s proposal, the primitive notion underlying attitudes de se is the
self-representational relation SELF-R. In her paper (Reinhart 1990:9 note 7), she explores
the possibility to derive the self-representational relation from the representational relation
R by means of a more basic operation. The operation reduces two identical arguments to
one, as illustrated in (21):
(21) R (k, x, x) ⇒ SELF-R (k, x)
This operation is very similar to the lexical operation that turns a transitive verb into a
reflexive one. For instance, the verb wash can be reflexivized both in syntax, as in
example (22), and in the lexicon, as in example (23):
(22) a. William washed himself
104
b. wash (θ1, θ2)
(23) a. William washed
b. wash (θ1)
(23) is obtain by means of a lexical operation that reduces two arguments of the verb’s θ-
grid to one.38
Hence, a possibility to explore is that the same operation derives the self-
representational relation from the representational relation. However, according to
Reinhart, it is not yet possible to fully spell out this proposal. The problem is that it is not
clear whether the semantic relation between a de re and a de se reading is equivalent to the
semantic relation between, for instance, (22) and (23). In fact, while it is clear the
difference between a de se and de re reading, it is not clear the semantic difference
between (22) and (23).
2.4 Higginbotham (1992)
2.4.1 Attributive versus propositional account of control
The theory that Chierchia (1989) proposes for attitudes de se has significant consequences
on the area of grammar known as theory of control. In Chierchia’s account, controlled
structures, i.e. structures whose subject is PRO, are interpreted as unsaturated structures,
i.e. properties. The semantic interpretation of controlled structures is obtained by
38 According to Reinhart (2000:5) the operation can apply to a pair of roles one of which is the external role. For an introduction to argument-structure and relevant discussion on the subject see Williams (1981) and Grimshaw (1990).
105
translating PRO into a λ-abstractor or, alternatively, by associating it with an operator
which is translated as a λ-abstractor. Apart from technical details, what matters is that the
result of the translation of a controlled structure is a property rather than a proposition.
However, Chierchia correctly observes that controlled structures are understood as having
a subject. For example, sentence (1) means that John wants to bring about a situation
where he, John and not someone else, wants to win.
(1) John expects PRO to win
How is then that the property of being an x such that x wins is attributed to John
and not to someone else? According to Chierchia’s theory, the solution is that the relation
between the agent and the property is that of self-ascription. This meaning postulate
prevents the attribution of the property to someone distinct from the agent. For instance, in
the case of (1), the self-ascriptive relation assures that the property of winning is attributed
to John. Such an account gives us a clear-cut explanation of two phenomena: (i) that
controlled structures are unambiguously interpreted de se and (ii) that PRO, in contexts of
VP-ellipsis, allows sloppy reading only. These two phenomena are illustrated in the
following examples:
(2) a. Frank expects that he will turn out to be a hero
b. Frank expects PRO to turn out to be a hero
(3) a. Frank expects that he will win and Bill does too
106
b. Frank expects PRO to win and Bill does too
The first phenomenon is illustrated by example (2): while (2a) is ambiguous between a de
re and a de se reading, (2b) is unambiguously interpreted de se. The second phenomenon
is detected in examples such as (3): while (3a) can allow both a sloppy and a strict
reading, (3b) allows the sloppy reading only. It should be clear that, whatever account of
controlled structures turns out to be correct, it must face and explain these two
phenomena.
In Chierchia’s framework, the fact that controlled structure are always interpreted
de se is a direct consequence of the self-ascriptive relation they entail. The sloppy
interpretation of PRO follows from the property status of control structures.
The attributive account of control structures, an account that takes them as
properties, is not uncontroversial. Chierchia himself finds such an account somewhat
contrary to our pre-theoretical intuitions. Higginbotham (1992) rejects the attributive view
and sustains a propositional account of control, an account that takes controlled structures
as propositions. Higginbotham strongest objection to the attributive account of control
stems from examples such as (4):
(4) John and Mary expect PRO to learn the same language
Suppose the following context: John and Mary are studying in a foreign country; they
speak different languages and they do not communicate to each other; however, they are
so simpatico to each other that they independently think they will end up learning the
same language. Under the attributive account the logical form of the complement of
107
expect can be represented as in (5), assuming, correctly, that same is construed with
respect to the embedded subject:
(5) ||λx (x learn the same language)||
The problem is that (5) cannot truly apply to John or Mary since same is construed within
(5) and the predicate expect is plural. To solve the trick we could assume that expect is not
distributive and it applies to John and Mary as one entity, which is implausible, or that
each of John and Mary has a de se expectation expressed by a predicate that does not
apply to either of them. Both solutions are clearly implausible. Henceforth, Higginbotham
claims that the attributive account of control structures is not correct and that a
propositional account is to be preferred.
Naturally, we expect that this alternative will be also able to explain the
phenomena we noticed. In the following sections I will expose Higginbotham’s theory of
control and I will explain, given the theory, why PRO is subject to the interpretative
constrains we have seen above.
2.4.2 Higginbotham’s theory of control
Consider example (1), repeated as (4):
(4) John expects to win
108
(4) shows an instance of the phenomenon that Higginbotham dubs as ‘understood
reference’ and defines as follows: ‘a reference to a thing x is said to be understood with
respect to a given position of a given head if there is no expression in that position
referring to x, but one takes it that the position is appropriately related to x, either through
another independent argument or position in the sentence or discourse in question, or as
pragmatically supplied’ (Higginbotham 1992:79)39. In sentence (1), for instance, the
argument of the head win is understood. In fact, although there is no expression in the
argument position we understand such argument as John.
In generative grammar it is assumed that understood elements are present in the
syntactic structure, although they have no phonological content, hence are not visible at
the level of Phonetic Form (PF). The syntactic structure of (4) is roughly represented in
(4′), where PRO represents the understood element:
(4′) John expects [PRO to win]
Semantically, understood elements are arguments. However, since they are not
independently referential, they are anaphorically related to an antecedent (assuming also
that the anaphoric relation is constrained by structural factors). For the moment, and apart
from technical details, I shall represent the anaphoric relation the way Higginbotham does,
that is by means of co-indexing, as in (4′′):
39 Raising and NP movement in passive sentences are not to be considered instances of understood reference; the difference between these cases and understood reference is accounted for in Higginbotham (1989).
109
(4′′) Johni expects [PROi to win]
Understood elements, like all anaphoric elements, inherit their value from the value of
their antecedent. Hence, the complement of expect is a proposition, represented in (5), and
that John must be the value of PRO follows from co-indexing.
(5) ||win(x)||
This account of control structures crucially differs from Chierchia’s. In Chierchia
(1989), PRO is interpreted as a λ-abstractor and consequently the control structure is
interpreted as a property, as shown in (6a); the postulated self-ascriptive relation
guarantees that the property is attributed to the agent and not to someone else. In
Higginbotham (1992), PRO is interpreted as an argument, although phonologically silent,
and the control structures as a proposition. The understood element receives its value
through the anaphoric relation. This account is represented in (6b).
(6) a. John expects (λx (x to win))
b. Johni expects ||win(xi)||
We might ask now, how the propositional account explains that control structures
are subjected to certain interpretative constrains: they are unambiguously interpreted de se
and they allow sloppy reading only. However, before doing this, let me introduce
110
Higginbotham’s account of attitudes de se since it will be crucial for the following
discussion.
2.4.3 Attitudes de se
The discussion of the theory of control is obviously connected with the discussion of
attitudes de se which, nevertheless, is the central topic of the present chapter. It should be
clear by now that a theory of control must entail an explanation of the fact that control
structures can only be interpreted de se and, conversely, a theory of attitude de se must be
implemented in a way such that it predicts the unambiguous interpretation of control
structures. The two things are henceforth deeply connected to each other.
Higginbotham’s account of attitudes de se begins with the analysis of Castañeda
(1966) and his examples. Consider sentence (7):
(7) The unfortunate man believes that he is a hero
As other examples we have analyzed so far, also (7) can be true in two different situations.
In the first case, the unfortunate man has a belief about himself being aware of the fact
that the res he is having a belief about is himself. This is the de se reading. However,
suppose the unfortunate man suffers from amnesia; he has been a war hero but he has no
memory about that; suppose then that this unfortunate person reads a book about the war
he was in and he finds in the book a detailed account of his own exploits. Then what he
believes is that the man he has read about is a hero, unaware of the fact that that man is
actually himself.
111
Both Castañeda and Lewis claim that a belief about oneself pertains to a category
distinct from that of beliefs about a certain res. In fact, Lewis decides to dub this special
category of beliefs as ‘de se’. Higginbotham claims that there is no reason to introduce a
new category since both beliefs are just beliefs de re. The distinction unearthed by
Castañeda and Lewis, according to Higginbotham, can be simply reduced to two different
strategies of interpretation of the pronoun and de se beliefs can be reduced to de re beliefs.
Consider again example (7). Suppose that the pronoun he in the embedded clause can be
interpreted by means of two strategies. According to the first strategy, the pronoun simply
takes the subject as its antecedent. According to the second strategy, the pronoun,
although taking the subject as its antecedent, goes proxy for something more complex in
which the subject himself perhaps figures. Using Higginbotham’s terminology, let us call
this more complex thing a ‘concept of’ the thing to which the pronouns refers. For
instance, in the case of (7) the referent of he could be something like (8):
(8) ||the person whom x is reading about||
Hence, the ambiguity of (7) can be represented by the two logical forms (9a) and (9b),
where co-indexing simply indicates the presence of an anaphoric relation:
(9) a. [The unfortunate man]i believes that ||xi is a hero||
b. [The unfortunate man]i believes that ||the person whom xi is reading about
is a hero||
112
We will say that in the case of (9a) the pronoun is interpreted ‘directly’ and what is
attributed to the agent is a ‘direct’ belief while in the case of (9b) the pronoun is
interpreted ‘indirectly’ and what is attributed to the agent is an ‘indirect’ belief. Both
beliefs, however, are about a certain res which, in the indirect case, namely (9b), turn out
to be a concept of that res. Higginbotham’s conclusion is that ‘de se belief amounts to de
re belief, and that no evidence has been produce that justifies the introduction of the new
category’ (Higginbotham 1992:86).
2.4.4 Features of PRO
Higginbotham’s account of attitudes de se does not solve the problems left open in the
preceding section but it casts a different light on them. Consider examples (10) and (11):
(10) The unfortunate man expects that he will turn out to be a hero
(11) The unfortunate man expects PRO to turn out to be a hero
While in sentence (10) he can be interpreted both directly and indirectly, hence both a
direct and an indirect thought can be attributed to the agent, in sentence (11) PRO can only
be interpreted directly, hence only a direct thought can be attributed to the agent.
Therefore, what must be explained is why PRO can only be interpreted directly.
The second question left open is why controlled structures allow sloppy
interpretation only, as shown in (3). According to Higginbotham, both properties of
controlled PRO belong to properties of understood reference in general. In particular
understood elements are featured by two properties: (i) they are purely referential, i.e. they
113
can never stand alone and can never have any sense attached to them; (ii) they are local
anaphors, i.e. they must always find a local antecedent.
The first property explains why they are always interpreted directly. Differently
from pronominals, silent elements can never stand-alone and cannot take a reference on
their own; henceforth they must rely completely on the presence of an antecedent from
which they receive a value. The second property, explains why understood elements
allows sloppy interpretation only. In fact, they must find a local antecedent under any
process that involves copying.
2.4.5 Appendix: a theory of control based on economy principles?
Let me summarize the main properties of PRO, and of understood reference in general,
according to Higginbotham’s account; I will refer to those to properties as (PRO1) and
(PRO2):
(PRO1) Referentiality:
An understood element cannot be autonomously referential, rather it must
receive its value from an antecedent.
(PRO2) Locality:
The antecedent of an understood element must be found locally.
It should be clear that the present approach to understood reference would
encounter theoretical and empirical problems similar to those raised against CBT. In fact,
114
(PRO1) seems to be strictly connected to something similar to the deficiency hypothesis
of Bouchard (1984), repeated below as (12):
(12) In order for an argument α to be interpreted, it must have a full specification for φ-
features. (Bouchard 1984)
According to (12), since understood elements are referentially defective arguments, they
are forced to take a linguistic antecedent. We could easily assume, according to
Higginbotham’s view, that understood elements are not fully specified for φ-features,
hence they must take an antecedent, i.e. that (PRO1) follows from (12).
Recall that the main reason to give up (12) was represented by the logophoric use
of anaphors. In fact, under such an approach, logophoricity was an exception to explain.
The open question was: why, under certain conditions, logophoric interpretation of
anaphors is allowed? What is very interesting in the present discussion is that, as well as
for anaphors, there are cases of ‘logophoric interpretation of PRO.’ Bach (1977) presents
examples of what he calls ‘pragmatic control:’
(13) Here’s a book [PRO to read to each other]
In example (13) PRO has no antecedent; a possible solution, considered also by
Higginbotham (1992), is to assume an understood benefactive. However, it seems to me
that such a solution would be based on a mere stipulation that would contradict the
minimalist assumptions I decided to stick to.
115
This becomes clear if we look at (PRO2). Locality, as in the case of canonical
binding principles, is just a stipulation since it does not follow from the combinatorial
properties of a morpho-syntactic lexicon. In other words, (PRO1) and (PRO2) are just a
descriptive statements, if correct: they explain how understood elements behave but they
do not explain why.
Given these premises, it seems to me that an interesting plan for further research
would be the following: (i) assume (14) instead of (12).
(14) If α has fewer φ-features than β, there are fewer constraints on the interpretation of
α than on the interpretation of β.
It follows from (14) that no intrinsic property of understood reference would block its
‘pragmatic’ interpretation; (ii) describe understood reference in terms of morpho-syntactic
features; (iii) describe the combinatorial properties of understood reference on the basis of
its morpho-syntactic features and the basic combinatorial operation of CHL; (iv) answer
question (15):
(15) Why under certain interpretation is ‘pragmatic’ interpretation of control blocked?
The result would be a theory of control similar to the theory of anaphoric relations I
presented in the previous chapter, a theory that attains explanatory adequacy and satisfies
of minimalist assumptions.
116
2.5 Conclusions
I would like to dedicate these concluding remarks to three main points: (i) give a brief
account of the theory of attitudes de se proposed by Chierchia (1989); (ii) give a summary
of the empirical problems raised by Reinhart (1990) and Higginbotham (1992); (iii) add
some further evidence against the attributive account of attitudes de se.
Chierchia’s theory of attitudes is based on Lewis’s idea that propositions lack the
structure necessary for entailing essential indexicality. Henceforth, attitudes de se result
from the agent’s self-attribution of a property, while attitudes de re are predicative
relations between agents and propositions. The primitive notion underlying attitudes de se
is that of self-ascription and a meaning postulate guarantees that attitudes de se entail
attitudes de re. The distinct interpretations, de re and de se, correspond to distinct LFs,
and the distinct LFs result from distinct syntactic structures. In Chierchia’s theory,
controlled structure are interpreted as properties, since PRO is translated into a λ-
abstractor, and this explains why they are unambiguously interpreted de se. Moreover,
according to Chierchia’s attributive account, logophors are always associated with a de se
interpretation.
Reinhart’s objections are the following:
(i) Chierchia’s theory predicts that the sloppy reading of deleted VPs is always
interpreted de se. This can be seen in example (1). The sloppy interpretation of (1a) is
obtained through the LF (1b) but (1b) is the LF of de se readings.
(1) a. Lucie believes that she sounds too aggressive and Lili believes it too
117
b. believes (Lucie, λx (x sounds too aggressive)) and believes (Lili, λy (y
sounds too aggressive))
This prediction is incorrect since sentences like (1a) can be interpreted both de re and de
se.
(ii) According to Chierchia’s theory, the mechanism underlying attitudes de se is
that of variable-binding; this is contradicted by sentence (2), where the believer can or
cannot be aware of the fact that he is the president. Hence, the expression ‘the president’ is
ambiguous between a de re and a de se interpretation, although the mechanism of
variable-binding cannot be responsible for this ambiguity.
(2) That the president will be remembered as a war hero is strongly believed by him
(iii) According to Chierchia, logophors are always associated with a de se
interpretation. However, Reinhart collects some examples from written narrative that
contradict this claim. Just to recall one of such examples, consider (3). Chierchia’s theory
wouldn’t be able to account for (3) since, according to the theory, the second sentence of
the example would be translated as a property, as in (3b); however, since there is no agent
to which ascribe the property, (3b) remains an open proposition that cannot receive a
truth-value.
(3) a. He [Sapp] sat down at the desk and opened the drawers. In the top right-
hand one there was an envelope addressed to himself.
118
(Zribi-Hertz 1989, quoted from Lodge)
b. (λx (in the top right-hand one there was an envelope addressed to x))
Higginbotham’s objections are the following:
(i) Chierchia’s attributive account of controlled structures runs against problems of
the kind presented by sentence (4):
(4) a. John and Mary expect PRO to learn the same language
b. expect (John and Mary, λx (x learn the same language))
Under the attributive account, (4a) is problematic since ‘expect’ is distributive and ‘same’
is construed with respect to the embedded subject. Hence, (4b) is not well-formed.
(ii) ‘de se belief amounts to de re belief, and […] no evidence has been produce
that justifies the introduction of the new category’ (Higginbotham 1992:86). Henceforth,
according to Higginbotham, the distinction between de re and de se is nothing more than
the result of two strategies of interpretation of pronouns.
An important feature of Chierchia’s theory of attitudes de se is that it requires the
presence of a predicate of attitude, a predicate that takes at least two arguments, one of
which is syntactically realized as an embedded clause. Then, the presence of an embedded
clause is crucial for the theory to work correctly. This can be seen in example (3): in fact,
the second sentence of (3a) does not contain an embedded clause, hence, as shown by
Reinhart, Chierchia’s theory cannot work correctly. In other words, Chierchia’s theory
119
predicts that the ambiguity de re/de se can show up only in certain syntactic contexts, i.e.
where a verb takes an embedded clause as one of its complements.
I believe that this prediction is incorrect since the ambiguity de se/de re can be
seen also in much simpler syntactic context. I will present two examples, one mine and
one from Delfitto (p.c.). The first example is the following:
(5) Narcissus loves himself
Sentence (5) can be true in two different sets of circumstances. In the first case, Narcissus
loves himself, and he is perfectly aware of the fact that the res he loves is himself; in fact
he would say ‘I love myself.’ In the second case, Narcissus is in love with the man he sees
in the water, unaware of the fact that that man is his own reflection. This situation can also
be reported using sentence (1); however, Narcissus would not say ‘I love myself;’ he
would rather say ‘I love the man below the water.’ I believe this ambiguity is just the
ambiguity de re/de se. However, the structure of (1) (roughly [NP [VP [NP ]]]), which is
a simple proposition, does not allow us to use Chierchia’s theory.
The second example is much more interesting:
(6) Oedipus loves his mother
The sentence is ambiguous between a de dicto, a de re and a de se interpretation. Let us
consider the three cases. (i) Oedipus does not know his mother since she abandoned him
right after he was born; however, he forgives his mother and says: ‘I love my mother,
whoever she is.’ This situation can be reported by sentence (6). (ii) Oedipus loves Jocasta,
120
unbeknownst of the fact that Jocasta is his mother. Also this situation can be reported by
sentence (6). (iii) Oedipus discovers that Jocasta is his mother and, hence, he discovers
that the person he loves is actually his mother. Again, this situation can be reported by
sentence (6).
The conclusion I draw from these examples and the examples from Reinhart and
Higginbotham is that Chierchia’s theory is wrong. I believe that if the problem of attitudes
de se has a linguistic relevance, the line of research to pursue is that of a deeper analysis of
anaphoric relations. In particular, it seems to me that what shouldd be investigated in
much more details is the amount of information that an anaphoric relation can bear and the
role this information has in assigning a meaning to an anaphoric expression.
121
WORKS CITED
Anderson, S. R. (1986) The typology of anaphoric dependencies: Icelandic (and other) reflexives.
In L. Hellan and K. K. Christensen (eds.) Topics in Scandinavian Syntax, 65-88.
Dordrecht: Reidel.
Aoun, Y. (1986) Generalized Binding, the Syntax and Logical Form of Wh-interrogatives.
Dordrecht: Foris.
Avrutin, S. (1999) Development of the Syntax-Discourse Interface. Boston: Kluwer.
Avrutin, S. (2004) Optionality in child and aphasic speech. Lingue e Linguaggio 1:95-115.
Avrutin, S. (2006) Weak syntax. In Y. Grodzinsky and K. Amunts (eds.) Broca’s Region. Oxford:
Oxford University Press.
Baauw, S. (2000) Grammatical features and the acquisition of reference: A comparative study of
Dutch and Spanish. Dissertation, Utrecht Institute of Linguistic OTS. LOT International
Series 39. Landelijke Onderzoekschool Taalwetenschap, Utrecht.
Baauw, S., and D. Delfitto (1999) Coreference and language acquisition. In Recherches de
Linguistique Français te Romane d’Utrecht XVIII, 23-35. Utrecht Institute of Linguistic
OTS.
Bach, E. (1977) Comments on the paper by Chomsky. In P. Culicover et al. Formal Syntax, 133-
155. New York: Academic Press.
Bonato, R. (2006) An integrated computational approach to binding theory. Doctoral dissertation,
Università degli Studi di Verona.
122
Bouchard, D. (1984) On the content of empty categories. Dordrecht: Foris.
Castañeda, H.-N. (1966) ‘He:’ A study in the logic of self-consciousness. Ratio 7:130-57.
Chierchia, G. (1984) Topics in the Syntax and Semantics of Infinitives and Gerunds. Doctoral
dissertation, University of Massachussets, Amherst.
Chierchia, G. (1989) Anaphora and attitudes de se. In R. Bartsch, J. van Benthem and P. van Emde
Boas (eds.) Semantics and Contextual Expression. Dordrecht: Foris.
Chierchia, G. (1995) Dynamics of Meaning. Chicago: University of Chicago Press.
Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris.
Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding.
Cambridge, MA: MIT Press.
Chomsky, N. (1986) Barriers. Cambridge, MA: MIT Press.
Chomsky, N. (1995) The Minimalist Program. Cambridge, MA: MIT Press.
Chomsky, N. (2000) Minimalist inquiries: The framework. In R. Martin, D. Michaels, and J.
Uriagereka (eds.) Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik,
89-155. Cambridge, MA: MIT Press.
Chomsky, N., and H. Lasnik (1993) The theory of principles and parameters. In J. Jacobs, A. von
Stechow, W. Sternefeld, and T. Vennemann (eds.) Syntax: An International Handbook of
Contemporary Research. Berlin: de Gruyter.
Clements, G. N. (1975) The logophoric pronoun in Ewe: Its role in discourse. Journal of West
African Languages 10:141-177.
Cresswell, M. J. (1985) Structured Meanings. Boston: MIT Press.
Fodor, J. (1977) The Language of Thought. Hassocks: Harvester Press.
123
Fodor, J. (1983) The Modularity of Mind. Cambridge, MA: MIT Press.
Fox, D. (1993) Chain and binding: A modification of Reinhart and Reuland’s ‘Reflexivity.’ Ms.,
Cambridge, MA: MIT.
Giorgi, A. (1983/1984) Toward a Theory of Long Distance Anaphors: a GB Approach. The
Linguistic Review 3:307-361.
Giorgi, A. (1987) The notion of complete functional complex: some evidence from Italian. LI
18:511-8.
Giorgi, A. (2004) Long Distance Anaphors and Temporal Relations. In Proceedings of West Coast
Conference on Formal Linguistics 23:223-236. Davis, CA: UC Davis.
Graffi, G. (1988) Structural subject and thematic subject. Linguisticae Investigationes XII:397-
414. Amsterdam: John Benjamins.
Grimshaw, J. (1990) Argument Structure. Cambridge, MA: MIT Press.
Grodzinsky, Y. (2000a) The neurology of syntax: Language use with broca’s area. Behavioral and
Brian Sciences 23:1-71.
Grodzinsky, Y. (2000b) Overarching agrammatism. In Y. Grodzinsky, L. Shapiro, and D. Swinney
(eds.) Language and the Brain: Representation and Processing – Studies Presented to
Edgar Zurif on His 60th Birthday, 73-86. San Diego: Academic Press.
Grodzinsky, Y., and T. Reinhart (1993) The innateness of binding and coreference. LI 24:69-101.
Hagège, C. (1974) Les pronoms logophorique. Bulletin de la Société de Linguistique de Paris
69:287-310.
Heim, I. (1993) Anaphora and semantic interpretation: A reinterpretation of Reinhart’s approach.
SfS report 07-93. University of Tübingen. Reprinted in MIT Working Papers in
124
Linguistics 25: The Interpretative Tract, 205-46. MITWPL. Cambridge, MA: MIT,
Department of Linguistic and Philosophy.
Higginbotham, J. (1983) Logical Form, binding, and nominals. LI 14:395-420.
Higginbotham, J. (1989) Elucidations of meaning. Linguistics and Philosophy 12:465-517.
Higginbotham, J. (1992) Reference and control. In R. K. Larson, S. Iatridou, U. Lahiri, and J.
Higginbotham (eds.) Control and Grammar. Dordrecht: Kluwer.
Higginbotham, J. (2003) Remembering, imagining, and the first person. In A. Barber (ed.)
Epistemology of Language. Oxford: Clarendon Press.
Lasnik, H. (1986) On accessibility. LI 17:126-9.
Lewis, D. (1979) Attitudes de dicto and de se. The Philosophical Review 88:513-43.
Kaplan, D. (1971) Quantifying in: In L. Linsky (ed.) Reference and Modality, Oxford: Oxford
University Press.
Kaplan, D. (1977/1989) Demonstratives. In Almog, Perry, and Wettstein (eds.) Themes for
Kaplan. Oxford: Oxford University Press
Mailing, J. (1984) Non-clause-bound reflexives in modern Icelandic. Linguistic and Philosophy
7:211-41.
Mailing, J. (1986) Clause-bound reflexives in modern Icelandic. In L. Hellan and K. K.
Christensen (eds.) Topics in Scandinavian Syntax, 53-63. Dordrecht: Reidel.
Manzini, M. R. (1993) The subjunctive. Paris 8 Working Papers 1.
Manzini, M. R., and K. Wexler (1987) Prameters, binding, and learnability. LI 18:413-44.
Medini, G. (1988) Clinical psychology in Israel: identity and directions. Israeli Journal of
Psychiatry 25,2:157-66.
125
Pica, P. (1987) On the nature of the reflexivization cycle. NELS 17:483-99. GLSA, University of
Massachusetts, Amherts.
Piñango, M. (2002) Cortical reflections of two pronominal relations. In H. J. Simon and H. Wiese
(eds.) Pronouns – Grammar and representation, 233-52. Amsterdam: John Benjamins.
Piñango, M., and Burkhardt, P. (2001) Pronominals in Broca’s aphasia comprehension: The
consequences of syntactic delay. Brain and Language 79:167-68.
Pollard, C., and I. Sag (1992) Anaphors in English and the scope of binding theory. LI 23:261-305.
Reinhart, T. (1976) The syntactic domain of anaphora. Doctoral dissertation, MIT.
Reinhart, T. (1983) Anaphora and Semantic Interpretation. London: Croom Helm.
Reinhart, T. (1990) Self-representation. Lecture delivered at Princeton Conference on Anaphora,
October 1990.
Reinhart, T. (2000) Strategies of anaphora resolution. In H. Bennis, M. Everaert, and E. Reuland
(eds.) Interface Strategies, 295-325. Amsterdam: Royal Academy of Arts and Sciences.
Reinhart, T. (2001) The theta system: Syntactic realization of verbal concepts. Uil OTS working
paper: Utrecht University.
Reinhart, T., and E. Reuland (1991) Anaphors and logophors: An argument structure perspective.
In J. Koster, and E. Reuland (eds.) Long-distance Anaphora, 283-321. Cambridge:
Cambridge University Press.
Reinhart, T., and E. Reuland (1993) Reflexivity. LI 24:657-720.
Reinhart, T., and E. Reuland (1995) Pronouns, anaphors, and Case. In H. Haider, S. Olsen, and S.
Vikner (eds.) Studies in Comparative Germanic Syntax, 241-69. Dordrecht: Kluwer.
Reuland, E. (1996) Pronouns and features. In NELS 26:319-33. GLSA, University of
Massachusetts, Amherts.
126
Reuland, E. (1998) Structural conditions on chains and binding. In NELS 28:341-56. GLSA,
University of Massachusetts, Amherts.
Reuland, E. (2000) The fine structure of grammar: Anaphoric relations. In Z. Frajzyngier and T.
Curl (eds.) Reflexives: Forms and Functions, 1-40. Amsterdam: John Benjamins.
Reuland, E. (2001a) Primitives of binding. LI 32:439-92
Reuland, E. (2001b) Anaphors, logophors, and binding. In P. Cole, G. Hermon, and C.-T. James
Huang (eds.) Long-distance Reflexives, 343-70. Syntax and Semantics 33. San Diego, CA:
Academic Press.
Reuland E., and S. Sigurjónsdóttir (1997) Long distance ‘binding’ in Icelandic: Syntax or
discourse? In H. Bennis, P. Pica, and J. Rooryck, 323-340. Dordrecht: Foris.
Ross, J. R. (1970) On declarative sentences. In R. A. Jacobs and P. S. Rosenbaum (eds.) Readings
in English Trasformational Grammar. Walham, MA: Ginn.
Schlenker, P. (2003) A plea for monsters. Linguistic and Philosophy 26:29-120
Sells, P. (1987) Aspects of logophoricity. LI 18:445-79.
Sigurδsson, H. Á. (1990) Long distance reflexives and moods in Icelandic. In L. Hellan and K. K.
Christensen (eds.) Topics in Scandinavian Syntax, 309-346. Dordrecht: Reidel.
Sigurjónsdóttir, S. (1992) Binding in Icelandic: Evidence from Language Acquisition. Doctoral
dissertation, UCLA.
Thráinsson, H. (1976) Reflexives and subjunctives in Icelandic. NELS 6:225-39. GLSA,
University of Massachusetts, Amherts.
Thráinsson, H. (1990) A semantic reflexives in Icelandic. In J. Mailing and A. Zaenen (eds.)
Modern Icelandic Syntax, 289-307. New York: Academic Press.
127
Thráinsson, H. (1991) Long distance reflexives and the typology of NPs. J. Koster, and E. Reuland
(eds.) Long-distance Anaphora, 49-75. Cambridge: Cambridge University Press.
Wexler, K., and Y.-C. Chien (1991) Children’s knowledge of locality conditions on binding as
evidence for the modularity of syntax and pragmatics. Language Acquisition 1:225-295.
Williams, E. (1981) Argument structure and morphology. The Linguistic Review 1:81-114
Zribi-Hertz, A. (1989) A-type binding and narrative point of view. Language 65:695-727.