Semantics with assignment variables*

Alex Silk ● [email protected]

Draft of July 2018

Abstract
This manuscript develops a framework for compositional semantics, and
begins illustrating its fruitfulness by applying it to certain core linguistic data.
The key move is to introduce variables for assignment functions into the syn-
tax; semantic values are treated systematically in terms of sets of assignments.
Assignments are theoretically interpreted as representing possibilities in the
model. The framework provides an alternative to traditional “context-index”-
style frameworks descending from Kamp/Kaplan/Lewis/Stalnaker. A prin-
cipal feature of the account is that it systematizes a range of seemingly dis-
parate linguistic shifting data, such as with quantifiers, intensionality, and
context-sensitivity under modals and attitude verbs. The treatment of the syn-
tax/semantics standardizes quantification across domains (individuals, worlds,
assignments) via a generalized (type-flexible, cross-categorial) binder-index
resulting from type-driven movement. The account affords a unified analysis
of the context-sensitivity of expressions such as pronouns, epistemic modals,
etc., in the spirit of contextualist theories, while compositionally deriving cer-
tain recalcitrant shifting/binding phenomena and providing a framework for
theorizing about differences in tendencies for local/global readings.
Extensions to questions, conditionals, and relative clauses are explored.
I show how certain independently motivated syntactic analyses can be imple-
mented in the foregoing assignment variable framework. Restrictive relative
clauses are treated as complements of the matrix determiner, which intro-
duces quantification over assignments. ‘If ’-clauses are treated as free rela-
tives/correlatives, interpreted as plural definite descriptions of possibilities.
* Draft available at goo.gl/kFVhw3. Thanks to the audience and commentators Dilip Ninan
and Brian Rabern at a 2018 Pacific APA symposium session, and to Ivano Caponigro, Salvatore
Florio, Daniel Rothschild, and Ede Zimmermann for discussion. Thanks to Evangelia Adamou and
consultants for Pomak (Slavic) data.
 Interrogative sentences denote a set of possible answers, conceived as sets of
possibilities. Additional shifting data are compositionally derived, e.g. con-
cerning “interrogative flip,” information-sensitivity, indexical shift, and don-
key anaphora. Speculative applications to weak crossover are briefly consid-
ered. The account affords a uniform compositional semantics for ‘if ’-clauses
in diverse conditional constructions (adjoined to NP/VP/IP/CP, with/without
a main clause modal or proform), and for individual correlative clauses ad-
joined to DP/IP/CP. The result is a unified approach to the syntax/semantics
of interrogatives, conditionals, and relativization. The semantics avoids intro-
ducing added interpretive principles or composition rules such as for quan-
tification, binding, movement (e.g. Predicate Abstraction, Predicate Modifi-
cation, Trace Conversion). The semantics is fully compositional.†

Contents (short)
1 Introduction 1
2 Basics 18
3 Syntax and semantics: Standardizing quantification 24
4 Examples 37
5 Recap. Next steps 60
6 Relative clauses. Quantification and binding with determiners 62
7 Conditionals 135
8 Interrogatives 176
9 Conclusion 208
Appendix Formal overview: Syntax, Semantics, Metasemantics 210
References 220

†Short version: A very long formal footnote to Stalnaker 1988.

Contents (detailed)

1 Introduction 1
1.1 Parameters and operators: A (Ptolemaic?) road not taken . . . . . . 8

2 Basics 18
2.1 Formal overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Preliminary derivation . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Syntax and semantics: Standardizing quantification 24

3.1 Preamble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Type-driven movement . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 World-binding: Complementizer . . . . . . . . . . . . . . . . . . . . 31
3.4 Assignment-binding: T, Modals, Attitude verbs . . . . . . . . . . . . 32
3.5 Generalized binder-index . . . . . . . . . . . . . . . . . . . . . . . . 34

4 Examples 37
4.1 Attitude ascription . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 De re/de dicto, Specific/non-specific: Global vs. local readings of
world arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Quantified modal attitude ascription . . . . . . . . . . . . . . . . . . 42
4.3.1 Standardizing quantification . . . . . . . . . . . . . . . . . . 44
4.3.2 Assignment modification . . . . . . . . . . . . . . . . . . . . 45
4.3.3 Epistemic modals: Locality and binding . . . . . . . . . . . 47
4.4 Pierre and friends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5 Recap. Next steps 60

6 Relative clauses. Quantification and binding with determiners 62

6.1 Relative clauses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.1.1 Syntax: Head raising + D-complement . . . . . . . . . . . . 63
6.1.2 Syntax/semantics: Trace Conversion? . . . . . . . . . . . . . 65
6.1.3 Semantics: Assignment-quantification with determiners . . 72
6.2 Donkey anaphora . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.2.1 Pronouns and copies . . . . . . . . . . . . . . . . . . . . . . 77
6.2.2 Asymmetric, universal, and existential readings . . . . . . . 88
6.2.3 Donkey pronouns in attitude ascriptions . . . . . . . . . . . 89
6.3 Extensions: Determiner quantifiers and assignment-quantification 95
 6.3.1 Quantifier phrases, specificity, and non-relative restrictive
modification . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3.2 Quantifier phrases and bound pronouns . . . . . . . . . . . 118
6.3.3 Donkey crossings: Weak crossover, inverse linking, genitive
binding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

7 Conditionals 135
7.1 Local and global readings . . . . . . . . . . . . . . . . . . . . . . . . 135
7.2 ‘If ’ and conditionality . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.2.1 Syntax: Free relatives + Movement . . . . . . . . . . . . . . 137
7.2.2 Semantics: ‘if ’-clauses as plural definite descriptions of as-
signments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.3 Adnominal conditionals . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.4 Sentence-final ‘if ’-clauses . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.5 Sentence-initial ‘if ’-clauses . . . . . . . . . . . . . . . . . . . . . . . . 149
7.6 Modalized conditionals: Restricting and shifting . . . . . . . . . . . 154
7.6.1 Direct restriction . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.6.2 “Double modal” and “information-sensitive” readings . . . 156
7.7 Correlatives and proforms: Individual and conditional . . . . . . . 161
7.7.1 Individual correlatives and proforms . . . . . . . . . . . . . 163
7.7.2 Conditional correlatives and ‘then’ . . . . . . . . . . . . . . 172

8 Interrogatives 176
8.1 Local and global readings in questions . . . . . . . . . . . . . . . . . 177
8.2 Syntax, semantics, metasemantics . . . . . . . . . . . . . . . . . . . . 178
8.3 wh questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.3.1 Recap: Choice functions and choice-function variables . . 190
8.3.2 Aside: Weak crossover revisited . . . . . . . . . . . . . . . . 193
8.4 “Interrogative flip” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
8.5 Conditional questions . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
8.5.1 Relevance conditional questions . . . . . . . . . . . . . . . . 199
8.5.2 Correlative questions . . . . . . . . . . . . . . . . . . . . . . 200
8.5.3 Hypothetical conditional questions . . . . . . . . . . . . . . 202
8.6 Recap: Standardizing quantification. Unifying relativization, con-
ditionals, questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

9 Conclusion 208
Appendix Formal overview: Syntax, Semantics, Metasemantics 210
A Sample lexical entries . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
B Metasemantics / Metalanguage . . . . . . . . . . . . . . . . . . . . . 211
C Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

References 220
 This manuscript develops a framework for compositional semantics, and begins
illustrating its fruitfulness by applying it to certain core linguistic data. The key
move is to posit variables for assignment functions in the syntax; semantic values
are treated systematically in terms of functions from assignments, now included in
the model. The core data involve a spectrum of linguistic “shifting” phenomena,
such as with quantifiers, intensionality, and context-sensitivity under modals and
attitude verbs. Extensions to questions, conditionals, relative clauses, and various
types of pronominal anaphora such as donkey anaphora are provided. The proposed
assignment-variable-based theory affords a standardization of quantification across
domains (individuals, worlds, assignments), and a unified approach to shifting phe-
nomena in natural language.

1 Introduction
An overview of the account’s theoretical context and key features is as follows.
Following Stalnaker (1970), Kamp (1971), Lewis (1980), and Kaplan (1989), a
standard move in formal semantics is to treat expressions as interpreted with respect
to two general parameters: to a first approximation, a context c which takes expres-
sions to intensions, and circumstance (index) i which takes intensions to extensions.
To handle quantification and free/bound pronouns, the context parameter may be
treated as an assignment function gc , mapping typed numerical indices ⟨n, τ ⟩ (ab-
breviated: nτ ) to items in the model.1 Nuances aside, the interpretation function
returns Truth for a sentence such as S = ‘It7 laughed’, JSKgc ,i = 1, iff such-and-such
individual relevant in c, gc (7e), laughed in the circumstance i, where 7 is an arbitrary
index for the free pronoun (type e for individuals):

(1.1) a. LF: [ It7 laughed ]

b. J(1.1a)Kgc ,i = 1 iff gc (7e) laughed in i

A trend in linguistic semantics has been to incorporate various parameters of in-

terpretation as elements of the model, and to posit linguistic reference to these el-
ements via covert variables.2 For instance, verbs may be treated as denoting sets of
events; modals may be treated as binding world pronouns; tenses may be treated as
1For discussion of differences among single-/double-indexing variants of the views, see Lewis
1980. Some authors distinguish the coordinates c and gc , reserving c for specific features of discourse
contexts such as speaker, addressee, world, etc. Given our purposes I simplify by identifying the
context coordinate with the contextually determined assignment.
2See e.g. Partee 1973, Cresswell 1990, Percus 2000, Schlenker 2003, Büring 2004, Hac-
quard 2006, 2010; for general discussion see Schlenker 2006.

1
pronouns referring to times. Compositional details aside, and bracketing tense and
aspect, the interpretation function, now J⋅Kgc , may return a semantic value for S as
in (1.2) (type s for worlds; world-variable left free).

(1.2) a. LF: [ It7 [ laughed w3 ]]

b. J(1.2a)Kgc = 1 iff gc (7e) laughed in gc (3s)

A natural question at this stage is whether there might also be object-language ref-
erence or variables with assignments. Diverse linguistic data involving quantifiers
have led some theorists to introduce a semantic type for assignments (Janssen 1997,
Sternefeld 1998, Kobele 2010, Rabern 2012, Kennedy 2014). The project pur-
sued here is to investigate the prospects for a linguistic framework that goes the
further step of positing variables for assignments in the syntax; semantic values are
treated systematically in terms of sets of assignments, also included in the model.
One way of understanding the assignment-variable-based approach is as for-
mally implementing Stalnaker’s (1988, 2014) seminal “multiple context” treatment
of attitude ascriptions (cf. Swanson 2011). On Stalnaker’s view, there are multiple
contexts “available to be exploited” (1988: 156) in describing individuals’ states of
mind — the “basic” (“global”) discourse context, and a “derived” (“local,” subordi-
nate) context representing the subject’s attitude state. In (1.3), whereas the discourse
context c1 supplies the interpretation of the embedded demonstrative ‘that woman’,
the derived context c2 representing (what is presupposed to be) Tom’s beliefs is
also available for interpreting the embedded change-of-state verb. In the case of
“Russell’s notorious yacht” (Stalnaker 1988: 158–159) in (1.4), the intuitive idea
is that whereas the second ‘it’ is interpreted in the discourse context, the first ‘it’ is
interpreted in the context representing (what is presupposed to be) Phoebe’s beliefs;
hence the belief being ascribed isn’t necessarily false. In some cases either the dis-
course context or derived context may be available for interpreting an expression, as
reflected informally in (1.5), where sc1 and sc2 represent the standard for counting as
rich accepted in the discourse and the standard accepted by Alice, respectively.

(1.3) [Context: It’s presupposed that Sue, the woman being demonstrated in the
discourse context c1 , never smoked.]
Tom believes that [that woman]c1 has quitc2 smoking.
a. ≈ Tom believes that Sue used to smoke and no longer smokes
(cf. Swanson 2011: ex. 31)
(1.4) Phoebe believes itc2 is longer than itc1 is. (cf. Stalnaker 1988: 158–159)
(1.5) [Talking about Bert:] Alice believes that hec1 is richc1 /c2 .

2
 a. ≈ Alice believes that Bert is sc1 /sc2 -wealthy
(cf. Silk 2016: 165–166, 2017: 1787–1788)

Stalnaker doesn’t offer specific technical implementations of these ideas (nor

does Swanson). One might attempt a pragmatic explanation of the shifts in in-
terpretation, perhaps drawing on general pragmatic accounts of local context (cf.
Schlenker 2009, 2010). One might treat attitude verbs as context-shifting op-
erators, and posit mechanisms for capturing the different ways shifting can occur
within a single clause (cf. Cumming 2008, Santorio 2010, 2012, Ninan 2012). A
natural alternative is to posit variables for the different “contexts.” Roughly put, in
(say) an attitude ascription, a context-sensitive expression receiving an intuitively
global reading would combine with a (context-)variable representing the discourse,
and a context-sensitive expression receiving an intuitively shifted reading would
combine with a (context-)variable representing the attitude state. The “multiple
contexts” with respect to which context-sensitive expressions may be interpreted
are represented via object-language variables for assignments.
Although there are precedents for introducing syntactic variables for items de-
termining shifted interpretations of referential expressions, e.g. under attitude verbs
(Percus & Sauerland 2003, Charlow & Sharvit 2014) or quantifiers (Chier-
chia 1992, Elbourne 2005, Johnson 2012), the project of developing an account
with variables for assignment functions — object-language variables for the sort of
item responsible for interpreting context-sensitive language generally — hasn’t been
pursued. Indeed, puzzles of referential expressions in attitude ascriptions can be
understood as instances of (what I call) local and global readings of context-sensitive
expressions more generally. Contextualism/relativism/expressivism debates have
focused on embedding contrasts between e.g. epistemic modals (predicates of per-
sonal taste, etc.), and paradigm context-sensitive expressions which receive their
interpretation from the discourse context: whereas ‘might’ in (1.7) characterizes
the subject’s information, ‘she’ in (1.6) is infelicitous if its gender implication isn’t
accepted in the discourse (though see Silk 2016, 2017 and Sudo 2012, respectively).

(1.6) #Ali thinks Bertj is a woman. Hei thinks shej is smart.

a. ≉Al thinks g(j) is a woman and is smart. (global reading obligatory)
(1.7) Alice thinks it mighti be raining.
a. ≉Alice thinks g(i) is compatible with its raining, where g(i) represents
the information accepted in the discourse context
(local reading obligatory)

3
As a counterpoint to indexicals, classically understood — expressions obligatorily
receiving their interpretation from the discourse context — common across various
language families are conventionalized anaphoric expressions (dedicated reflexive
and reciprocal forms aside). For instance, Upper Austrian German distinguishes
two forms for 3rd-person pronouns; in contrast to the morphologically more com-
plex forms, which cannot be used as bound variables, the reduced 3rd-person forms
cannot be used referentially, as reflected in (1.8). Similarly, the anaphoric demon-
strative ee and personal pronoun taman in Sinhala (Indo-Aryan) obligatorily re-
ceive their interpretation from the linguistic context ((1.9)). Only the specialized
anaphoric pronoun can serve as the main clause correlate in a correlative, as in
(1.10) for Bangla. (See also Gair et al. 1999, Bhat 2004 and references therein;
cf. Mortensen 2003 on copy reflexive languages.)3

(1.8) a. Gestan hot ea/∗ a gsunga.

Yesterday has he sung
‘Yesterday he sang.’
b. Neta do Honsi hot a Frog kriagt wos ai /∗ eai vastondn hot
Only det Hans has a question gotten which he understood has
‘Only Hansi got a question which hei understood.’
(Wiltschko 2016: exs. 35, 37; Upper Austrian)
(1.9) a. api meekә/∗ eekә kәrәmu.
‘Let’s do this.deict/*this.anaph.’
b. lamәyai tamanṭәi /eyaaṭәmәii dœn teerenәwa kiyәla
child self.anaph/he.anaph.emph ∅ now understand
kiwwa.
said
‘The childi said that hei understood now.’
(Gair 1998: 113, 131–132; Sinhala)
(1.10) [je mee-Tii okhane dãRie ache] Sei /∗ oi lOmba.
which girl there.deict stand be that.anaph/*that.deict tall
(lit. ‘Which girl is standing over there, that is tall’)
‘The girl who is standing over there is tall.’
(Bagchi 1994: ex. 3; Bangla)

Yet most context-sensitive expressions fall somewhere in the middle regarding their
3For readability, depending on the purposes at hand I often abbreviate the English glosses and
omit indications for phi features, classifiers, case, tense.

4
tendencies for local/global readings — e.g., with gradable adjectives and degree stan-
dards, additives and implications of alternatives, quantifiers and domain restric-
tions, perspectival expressions and perspectives, etc. (Silk 2014a, 2016).

(1.11) Bob: Pete thinks Al is at a local bar.

a. ≈ Pete thinks Al is at a bar local to Bob (global reading)
b. ≈ Pete thinks Al is at a bar local to Pete (local reading)
(1.12) Beth thinks Chip smokes too.
a. [Context: We’re talking about various of our colleagues C who smoke.
We know Beth is good friends with Chip; we know nothing about her
views on any of our other colleagues’ habits.]
≈ People other than Chip in C smoke and Beth thinks Chip smokes.
(global reading)
b. [Context: We’re talking about Beth; we know she is convinced that vari-
ous of our colleagues smoke. We know none of our colleagues smoke.]
≈ Beth thinks that Chip along with other of our colleagues smoke.
(local reading)
(1.13) Alice thinks everyone can vote.
a. [Context: We’re considering Alice’s beliefs about the legal status of certain
minority groups G, which we think are relevant in questions about voting
rights; we know nothing about Alice’s own moral/legal views.]
≈ Alice thinks everyone in G is legally permitted to vote.
(global reading)
b. [Context: We’re considering Alice’s moral/legal views; we know she is
aware that certain minority groups aren’t legally permitted to vote.]
≈ Alice thinks everyone in the groups she considers relevant to questions
about voting rights is legally permitted to vote.
(local reading)

Mechanisms for capturing the varieties of shifting phenomena have grown in-
creasingly complex in current “context-index”-style frameworks (see §1.1). Adding
assignment variables to the object-language is far from trivial. Yet I’ll show how an
account with assignment variables can compositionally derive a spectrum of linguis-
tic shifting phenomena using independently motivated syntactic/semantic resources
and a standardization of quantification across domains. Key features are as follows:
the proposed account

5
 • unifies seemingly disparate shifting phenomena — e.g., with bound readings
of pronouns, donkey (non-c-command) anaphora, indexical shift, intension-
ality, local/global readings — and captures them via familiar mechanisms of
quantifier movement and variable binding

• standardizes quantification, as over individuals, worlds, assignments

– defines a generalized (cross-categorial, type-flexible) binder-index to
combine with quantifiers, attitude verbs, modals, etc.
• provides compositional, non-syncategorematic treatments of quantification
and relativization
– posits no added parameters of interpretation or construction-specific
interpretation rules (e.g., Predicate Abstraction, Trace Conversion)
• offers a unified analysis of the context-sensitivity of epistemic modals, pro-
nouns, etc., in the spirit of contextualist theories; yet improves in compo-
sitionally deriving certain distinctive shifting/binding phenomena (e.g. with
epistemic modals), and providing a framework for theorizing about differ-
ences in tendencies for local/global readings

Though detailed theory comparison would be premature at this stage, features such
as these should make the project of interest to ongoing work on topics such as quan-
tification, modality, and context-sensitivity.
An overview of the paper is as follows: §2 introduces principal elements of the
basic syntax/semantics. §3 motives a more complex clausal architecture by examin-
ing how an assignment-variable-based theory can be integrated into independently
motivated treatments of the syntax-semantics interface. The resulting account cap-
tures intensionality and local/global interpretations of context-sensitive expressions
via general mechanisms of movement and variable binding, and it affords an ele-
gant standardization of quantification. A definition for a generalized binder index,
which attaches directly to moved expressions, is provided; the semantics is fully
compositional. §4 illustrates compositional derivations of various examples involv-
ing quantifiers, attitude ascriptions, and modals using the syntax/semantics from §3.
Speculative applications to classic puzzles of names in attitude ascriptions are briefly
considered. Working through these examples will afford opportunities for exam-
ining issues regarding quantification in the metalanguage, binding with pronouns
vs. traces, and (non-)conventionalized locality/globality principles for constraining
readings. An improved formalization of assignment modification captures binding

6
relations in derivations with long-distance binding. §5 recaps the main develop-
ments thus far.
§§6–8 explore how an assignment-variable-based framework might be extended
to other complex constructions — in particular, relative clauses (§6), conditionals
(§7), and interrogatives (§8). I examine how certain independently motivated syn-
tactic analyses can be implemented in the assignment-variable-based framework
and treatment of the syntax/semantics interface from §§3–4. Headed restrictive
relative clauses are treated as complements of the matrix determiner, which in-
troduces quantification over assignments (§6). ‘If ’-clauses are treated as free rel-
atives/correlatives, interpreted as plural definite descriptions of assignments (§7).
Interrogative sentences denote a set of possible answers, with answers conceived
as sets of assignments (possibilities) (§8). Derivations of various additional lin-
guistic shifting phenomena are provided, as concerning local readings in questions
and conditionals, information-sensitivity, and donkey anaphora. Applications to
binding phenomena with pronouns, including inverse linking, genitive binding, and
weak crossover effects, are also considered. Certain features of the treatments of
questions, conditionals, and relativization in these sections are of general interest,
independent of the particular assignment-variable-based implementation — e.g., an
analysis of various types of apparent binding out of DPs such as donkey anaphora; a
distinction between trace-binding and pronoun-binding, with applications to weak
crossover; a formal diagnosis of notions such as “specificity” (“Discourse-linking”)
in connection with several types of quantificational and non-quantificational DPs;
a formalization of “interrogative flip” in questions; with conditionals, derivations of
apparent single- vs. double-modal readings, indexical shift, conditional questions,
and shifting with adnominal conditionals and sentence-initial and sentence-final
‘if ’-clauses; a unified syntax/semantics for individual correlative clauses adjoined to
the main clause and adjoined directly to a correlate proform; compositional treat-
ments of relativization and interrogatives which avoid additional composition rules
or interpretive principles (Predicate Abstraction, Predicate Modification, Crosscat-
egorial Quantification Conversion, Trace Conversion); a unified analysis of wh-
words, relative determiners, and certain indefinites; and a unified approach to the
syntax/semantics of conditional, correlative, and interrogative clauses.
The specific account to be developed of course isn’t the only way of implement-
ing a syntax/semantics with object-language assignment variables. There will be
various choice points, many of them unforced, along the way. I leave additional
applications, development of alternative implementations, and comparisons with
existing frameworks for future research.
A preliminary methodological remark: The project in this paper isn’t to provide

7
a possibility proof; the question isn’t whether it is possible to construct a formal
syntax/semantics with object-language assignment-variables. I would be surprised
if it wasn’t. The project is to investigate the prospects for a specific type of syn-
tax/semantics of natural language so as to provide a concrete basis for overall theory
comparison. Some of the data will be new, though in many cases the aim will be
to provide a new take on old facts — to integrate data from diverse literatures in
new ways, and systematize phenomena independently familiar yet often not jointly
considered in theorizing. The goal is a more coherent theory of linguistic shifting
phenomena with improved empirical coverage and explanatory power. Sustained
investigation into a variety of linguistic phenomena is thus required. One must
ensure that particular choice points and analyses generalize across the spectrum
of examples, and can be systematized into an overall account that plausibly rivals
accounts in more traditional frameworks; the devil is in the details (lambdas, trees).
It will be critical to motivate particular syntactic implementations and apply the
account to diverse expressions and constructions. I ask for the reader’s patience
along the way. I hope the preliminary developments in this paper may illustrate
the fruitfulness of an assignment-variable-based framework for linguistic theorizing
and provide a more adequate basis for future theory comparison.
1.1 Parameters and operators: A (Ptolemaic?) road not taken
Before beginning our constructive project, this subsection briefly considers how cer-
tain of the core data with local/global readings might be captured in a less revisionary
operator-based semantics — roughly put, a semantics which analyzes modals, atti-
tude verbs, etc. as operators which shift the relevant contextual features, construed
as parameters of interpretation. To fix ideas I focus on adapting two prominent gen-
eral approaches to linguistic shifting phenomena from the literature — traditional
context-index-style frameworks, and approaches to indexical shift which introduce
assignment-quantification in the metalanguage. The aim isn’t to develop the ac-
counts in depth, or to show that no alternative (whether developed along the fol-
lowing lines or otherwise) can succeed. The aim is simply to highlight certain prima
facie costs so as to further motivate the paper’s central constructive project. Readers
satisfied with the motivations in §1 may wish to proceed to page 16.
The string in (1.14) permits a range of readings depending on context.⁴
⁴See Fodor 1970 on distinguishing the specific/non-specific distinction from the de re/de dicto
distinction (also von Fintel & Heim 2011; more in §§4.2, 6.2). See Stephenson 2007, Silk 2016,
2017 on obligatory local readings of epistemic modals (§1; more in §4.3.3). See Barker 2002, Silk
2016 on local/global readings of relative gradable adjectives (more in §8). See Silk 2014a, 2016 on
local/global readings of quantifier domain restrictions. Terminology varies among authors.

8
(1.14) Everyone thinks that a friend of mine might be rich.
• ‘a friend of mine’, Specific/Non-specific:
– Non-specific: the witness for the DP may vary across the quantifica-
tional subject’s belief-worlds
– Specific: the witness for the DP is constant across the quantificational
subject’s belief-worlds; the belief is about a particular individual
• ‘a friend of mine’, De re/De dicto:
– De dicto: the witness for the DP is a friend of mine in the local evaluation
world; the belief is about individuals conceived as being friends of mine
– De re: the witness for the DP is a friend of mine in the actual world; the
belief is about actual-world friends of mine
• ‘everyone’, ‘a friend of mine’, domain restriction:
– Global: the domain of quantification is determined by the discourse
– Local: the domain of quantification is determined by a local attitude/information
state (e.g., where the relevant subjects differ on which friends ought to
be considered in questions about richness (do my animal friends count?
do my toddler friends count?))
• positive form relative gradable adjective ‘rich’, standard of richness:
– Global: the standard for richness — how rich one must be to count as
rich — is determined by the discourse context (each subject’s attitude is
about whether such-and-such individuals are dc -wealthy, where dc is the
contextually presupposed standard for richness)
– Intermediate: the standard for richness is determined by the quantifica-
tional subject’s beliefs (each subject o’s attitude is about whether such-
and-such individuals are do -wealthy, where do is the standard for rich-
ness accepted by o)
– Local: the standard for richness is determined relative to the local epis-
temic possibility (the subjects may fail to accept particular standards
for richness, and each subject o’s attitude is about whether such-and-
such individuals are deo -wealthy, where deo is the standard for richness
regarded as live by o)
• epistemic ‘might’, body of information: the relevant body of information is
determined by the quantificational subject’s beliefs

9
For instance, suppose we are millionaires and we agree that one must be a millionaire
to count as rich. We are considering Alice’s, Bert’s, and Chip’s beliefs about Rita’s
income, where Rita is a friend of mine. I know that each of them thinks that Rita is
a millionaire. Though I may know nothing about Alice’s, Bert’s, or Chip’s standards
for richness, I can describe their states of mind using (1.14), as reflected in (1.15).
(I use the superscript/subscript indices informally to indicate certain of the intuitive
binder/bindee relations.)

(1.15) a friend-of-minei@ [everyone thinksm n

@ [mightm [ti rich@ ]]]
a. Intended reading: specific de re + local ‘might’ + global ‘rich’-standard
b. ≈ there is a particular individual o who is a friend of mine such that
everyone relevant in c thinks it’s possible that o is dc -wealthy

Alternatively, suppose we are considering Alice’s, Bert’s, and Chip’s beliefs about a
group of individuals who are actually friends of mine (say, the Sharks), and their
respective normative/economic views about how wealthy one must be to count as
rich. I know that each of them accepts a particular though different standard for
richness. The string in (1.14) can be used to describe their states of mind in this
alternative scenario as well, as reflected in (1.16).

(1.16) everyone thinksm n

@ [mightm [a friend-of-mine@ richm ]]
a. Intended reading: non-specific de re + local ‘might’ + intermediate ‘rich’-
standard
b. ≈ there is some group of individuals who are friends of mine — say, the
Sharks S — such that everyone o relevant in c thinks it’s possible that some
or other o′ ∈ S is do -wealthy

Consider (1.14) on the reading in (1.16). Non-specific de re readings (Fodor

1970) have provided one of the main motivations in the literature for positing object-
language world variables (cf. Percus 2000, von Stechow 2008, von Fintel &
Heim 2011). The quantifier in ‘a friend of mine’ must take narrow scope with respect
to the attitude verb in order to capture how the subject’s beliefs needn’t be about
a particular individual; yet the predicate’s world parameter must remain unshifted
so as to capture how the beliefs are about actual-world friends of mine. Treating
worlds as parameters of evaluation that are shifted by intensional operators obscures
how the DP’s predicate can be anchored to the actual world while scoping under
the modal/attitude verb (though see von Fintel & Heim 2011). Introducing world
variables in the syntax (cf. (1.2)), and treating modal quantifiers over worlds parallel

10
to determiner quantifiers over individuals, avoids the problem. Non-specific de
re readings are captured via long-distance binding of the embedded DP’s world
variable, as reflected in (1.17) and (1.18).⁵ (I use ‘⋋’ to distinguish the assumed
object-language binder index.)

(1.17) a. Al thinks a friend of mine won

b. LF: [ S ⋋2s Al [thinks w2 ] [ S′ ⋋4s [ S′′ [ DP a [[friend w2 ] of mine1 ]] won
w4 ]]]
(1.18) JDPKg = λPet . ∃xe ∶ P(x) ∧ x is a friend of g(1e) in g(4s)
JS′′ Kg = 1 iff ∃x∶ x won in g(4s) ∧ x is a friend of g(1e) in g(2s)
JS′ Kg = λws .JS′ Kg[w/4s] = λws . ∃x∶ x won in w ∧ x is a friend of g(1e) in g(2s)
JthinkKg = λws .λpst .λxe .∀w′ ∈ DOXx,w ∶ p(w′ ) = 1
S is true in c iff JSKgc (wc ) = 1
iff ∀w′ ∈ DOXAl,wc ∶ ∃x∶ x is a friend of gc (1e) in wc ∧ x won in w′

Turn to the shifted readings of epistemic ‘might’ and the standard associated
with ‘rich’. As noted in §1, many authors have observed that the body of information
relevant for interpreting epistemic modals under attitude verbs is generally shifted
to the subject’s information. In quantified epistemic attitude ascriptions such as
(1.14) the relevant information shifts with the quantificational subject. A promi-
nent (though certainly controversial) approach to capturing these data is to add
an informational coordinate to the index of evaluation, the parameter shiftable by
operators (Stephenson 2007, Yalcin 2007, Hacquard 2010, MacFarlane 2014;
cf. “nonindexicalism” in the sense of MacFarlane 2009). To a first approximation:⁶

(1.19) a. Al thinks a friend of mine might win.

b. [ S ⋋2s Al thinks-w2 ⋋4s [ S′ might ⋋7s [a friend-w2 -of-mine1 win-w7 ]]]
(1.20) Jmight ϕKg,s = 1 iff ∃w′ ∈ s∶ JϕKg,s (w′ ) = 1
⁵For present purposes I assume a simplified analysis of ‘friend of mine’, I leave open what
generates the world-binders, and I assume a standard Predicate Abstraction rule. The positive
account in the following sections derives the relevant binder indices from type-driven movement and
does away with non-compositional rules such as Predicate Abstraction. A more detailed analysis of
genitives is given in §6.3.3.
⁶An additional abstraction rule analogous to Heim & Kratzer’s (1998) Intensional Function
Application could be introduced for combining modals/attitude verbs with their complements
(cf. Santorio 2010, Kennedy 2014). For simplicity I treat the informational parameter as a set
of worlds, and I bracket details about the internal syntax/semantics of different flavors of modality.
A more precise semantics in our assignment-variable framework will be given in §3.4.

11
 ′
Jthink-wi ϕKg,s = λxe . ∀w′ ∈ s′ ∶ JϕKg,s (w′ ) = 1, where s′ = DOXx,g(is)
JS′ Kg,s = 1 iff ∃w′ ∈ s∶ ∃x∶ x is a friend of g(1e) in g(2s) and x wins in w′
JSKg,s = λws . ∀w′ ∈ DOXAl,w ∶ ∃w′′ ∈ DOXAl,w ∶ ∃x∶ x is a friend of g(1e) in w
and x wins in w′′
= λws . ∃w′′ ∈ DOXAl,w ∶ ∃x∶ x is a friend of g(1e) in w and x wins in w′′

(1.19) characterizes Al’s beliefs as being compatible with some actual friend of mine
being the winner, as derived in (1.20). The attitude verb obligatorily shifts the in-
formational coordinate in the index to the subject’s belief state.
The above sort of semantics may work well for expressions for which local read-
ings are conventionalized. Yet the approach is awkward for expressions permit-
ting both local and global readings — that is, the vast majority of context-sensitive
expressions.⁷ Take the standard-sensitivity associated with positive form relative
gradable adjectives. Since ‘rich’ can receive non-shifted readings in certain contexts,
it won’t do to treat e.g. ‘rich’ as directly dependent on a posited standard coordinate
d in the index, analogous to the semantics for ‘might’ in (1.20). So suppose we treat
‘rich’ as having an argument place filled either by a contextually provided standard-
pronoun, representing global readings, or an element stipulated as referring to the
standard coordinate of the index, representing local readings. A simplified seman-
tics is as follows, where DOX∗x,w is now a set of world-standard pairs compatible with
x’s belief state in w, and ‘o is d-wealthy’ abbreviates that o’s degree of wealth is at least
as great as d.⁸

(1.21) [S ⋋2s Al thinks-w2 ⋋4s

a. … [S′ Rita is rich-w4 -D]]
b. … [S′′ Rita is rich-w4 -d3 ]]
(1.22) JrichKg;s,d = λws .λdd .λxe . x is d-wealthy in w
JDKg;s,d = d
′ ′
Jthink-wi ϕKg;s,d = λxe . ∀⟨w′ , d′ ⟩ ∈ s∗ ∶ JϕKg;s ,d (w′ ) = 1,
where s∗ = DOX∗x,g(is) and s′ = {w ∣ ∃d′′ ∶ ⟨w, d′′ ⟩ ∈ DOX∗x,g(is) }
JS′ Kg;s,d = 1 iff Rita is d-wealthy in g(4s)
JS′′ Kg;s,d = 1 iff Rita is g(3d)-wealthy in g(4s)
⁷See Silk 2014a, 2016 on parallels among e.g. epistemic modals and paradigm context-sensitive
expressions in the discourse dynamics of unembedded uses.
⁸I ignore context-sensitivity regarding comparison classes, and bracket details about the internal
syntax/semantics of the positive form (Kennedy 2007).

12
 Local: (1.21a) is true in c iff (roughly) for all worlds w′ and standards d′
compatible with Al’s belief state in wc , Rita is d′ -wealthy in w′
Global: (1.21b) is true in c iff (roughly) for all worlds w′ and standards d′
compatible with Al’s belief state in wc , Rita is gc (3d)-wealthy in w′

Analogous moves can be made for other expressions optionally taking local or global
readings, as in (1.23)–(1.24) for quantifiers and domain variables (von Fintel 1994,
Stanley & Szabó 2000, Stanley 2002). (For simplicity I treat the added parameter
G in the index as a set of individuals, and I leave ‘can-vote’ unanalyzed.)

(1.23) a. (1.13a) LF (rough): [ ⋋2s Alice thinks-w2 ⋋4s [ [everyone P5 ] can-vote-

w4 ]]
b. (1.13b) LF (rough): [ ⋋2s Alice thinks-w2 ⋋4s [ [everyone G] can-vote-
w4 ]]
(1.24) JeveryoneKg;s,d,G = λPet .λQet . ∀x s.t. x is a person and P(x) = 1∶ Q(x) = 1
JGKg;s,d,G = G
′ ′ ′
Jthink-wi ϕKg;s,d,G = λxe . ∀⟨w′ , d′ , G′ ⟩ ∈ s∗ ∶ JϕKg;s ,d ,G (w′ ) = 1,
where s∗ = DOX∗x,g(is) , s′ = {w ∣ ∃d′′ ∃G′′ ∶ ⟨w, d′′ , G′′ ⟩ ∈ DOX∗x,g(is) }
Global: (1.23a) is true in c iff (roughly) for all worlds w′ , standards for rich-
ness d′ , and sets of individuals G′ relevant to voting rights compatible with
Alice’s belief state in wc , every person x ∈ G′ can-vote in w′
Local: (1.23b) is true in c iff (roughly) for all worlds w′ , standards for richness
d′ , and sets of individuals G′ relevant to voting rights compatible with Alice’s
belief state in wc , every person x ∈ g(5et) can-vote in w′

One might worry that capturing shifted readings of context-sensitive expres-

sions via coordinates in the index obscures one of the original motivations for dis-
tinguishing the index parameter in points of evaluation. Following Lewis 1980
the index consists of those contextual features that can be shifted by operators. If
nearly all contextual features can be shifted — even arguably features determining
the interpretation of paradigm indexicals such as ‘I’ (e.g., Santorio 2012) — one
might wonder if a simpler theoretical apparatus and treatment of shifted/non-shifted
readings is available.
On a more empirical level, it isn’t evident that the approach will generalize to
more complex examples. For instance, (1.23) represented the index coordinate for
shiftable quantifier domain restrictions simply as a set of individuals. This represen-
tation would need to be complicated to account for different restrictions associated

13
with different quantified expressions. The parameter might be represented instead
as a function from NP-meanings to domain restrictions, roughly as in (1.25), where
∗
JGKg;G ,... = G∗ is type ⟨et, et⟩ and cat/dog is the set of cats/dogs (ignoring intension-
ality). Intuitively, G∗ (cat) represents a set of individuals relevant when considering
cats, G∗ (dog) represents a set of individuals relevant when considering dogs, etc.

(1.25) Jevery dog G ⋋1 every cat G likes t1 Kg; G ,... = 1 iff for every y s.t. y is a dog
∗

and y ∈ G∗ (dog), for every x s.t. x is a cat and x ∈ G∗ (cat), x likes y

Yet examples such as (1.26) involving multiple occurrences of the same quantified
expression, where each occurrence receives a different shifted restriction, remain
problematic (cf. e.g. Stanley 2005).

(1.26) [Context: A panel survey is being carried out to discern sentiments about the
University’s practices in distributing a certain award. The panel members —
Alice, Bert, and Chloe — have different views on who should be allowed to
be nominated for the award, and who should be allowed to vote in deciding
the winner. Alice thinks that the award should be reserved for undergrad-
uates, and that only graduate students and faculty should be allowed to vote
for the winner; and she thinks that the award procedure in fact proceeds
accordingly. Bert thinks that the award should be open to graduate students
too, and that undergrads, grads, and faculty should all be allowed to vote;
and he thinks that the award procedure in fact proceeds accordingly. Chloe
thinks the award should be open to all members of the University, but she
thinks that faculty are (wrongfully) excluded from being nominated. When
asked about how well the University is doing regarding sentiments about the
fairness of the award practices, you report:]
Quite well. Most people think that everyone can vote for everyone.

One strategy for addressing these concerns is to abandon the notion of an in-
dex parameter, and treat expressions such as modals/attitude verbs as assignment-
shifters (cf. Cumming 2008, Santorio 2010, 2012, Ninan 2012). Previous accounts
have focused on indexical shift with individual pronouns. Suppose we generalize the
approach to shifting with other types of context-sensitive expressions. Just as the
attitudes and circumstances of interlocutors in a concrete discourse are assumed to
be representable by an abstract “contextually determined assignment” which deter-
mines values for pronouns, quantifier domain restrictions, degree standards, etc.
(e.g. Heim & Kratzer 1998, Heim 2008), so too, we can assume, for concrete at-
titude states more generally. Adapting the semantics in Santorio 2010, suppose

14
we relativize the interpretation function to an ordinary assignment g as well as an
assignment a shifted by certain modals/attitude verbs. The syntactic indices in a’s
domain can be distinguished accordingly, marked as [+a]. Expressions option-
ally receiving local/global readings may take variables of both types; other context-
sensitive expressions may lexically specify a particular type of variable. To a first
approximation, simplified derivations for the alternative readings of (1.21) and for
(1.26) might proceed as follows.⁹

(1.27) a. [S ⋋2s Al thinks-w2 ⋋4s [S′ Rita is rich-w4 -d3 /d3[+a] ]]

b. Jd3 Kg,a = g(3d)
Jd3[+a] Kg,a = a(3d)
JrichKg,a = λws .λdd .λxe . x is d-wealthy in w
′
Jthink-wi ϕKg,a = λxe . ∀⟨w′ , a′ ⟩ ∈ DOXx,g(is) ∶ JϕKg,a (w′ ) = 1
JS′ Kg,a = 1 iff Rita is g(3d)/a(3d)-wealthy in g(4s)
′
JSKg,a = λws . ∀⟨w′ , a′ ⟩ ∈ DOXAl,w ∶ JS′ Kg,a (w′ ) = 1
= λws . ∀⟨w′ , a′ ⟩ ∈ DOXAl,w ∶ Rita is g(3d)/a′ (3d)-wealthy in w′
(1.28) a. [S ⋋2s Most people-w2 P3 think-w2 ⋋4s [everyone-w4 P5[+a] ⋋1e
everyone-w4 P7[+a] can-vote-w4 for t1 ]]
b. JSKg,a = λws . for most z s.t. z is a person in w and z ∈ g(3et)∶ ∀⟨w′ , a′ ⟩ ∈ DOXz,w ∶
∀y s.t. y is a person in w′ ∧y ∈ a′ (5et)∶ ∀x s.t. x is a person in w′ ∧ x ∈ a′ (7et)∶
x can vote for y in w′

Such an approach may capture alternative readings where the relevant expres-
sions receive their interpretation from the discourse or embedding attitude (modal,
etc.). Yet neither operator-based account outline above captures intermediate read-
ings such as (1.16) where the interpretation of an expression is determined non-
locally but with respect to an environment distinct from the discourse context. (As-
sume the assignment-shifting implementation, and let si be a variable for a set of
world-assignment pairs.)
′
(1.29) Jmight-si[+a] ϕKg,a = 1 iff ∃⟨w′ , a′ ⟩ ∈ a(i)∶ JϕKg,a = 1
(1.30) [⋋2s everyone-P5 thinks-w2 ⋋4s [might-si[+a] ⋋7s [a friend-of-mine1 -w2
rich-w7 -d??? ]]]
⁹Metasemantic principles could be given regarding the relation between the assignments g and
a, what particular indices represent, and what it is for an assignment to be compatible with an attitude
state. For present purposes I ignore such complications; I return to such issues in §§3, 4.

15
On the intermediate reading in question, the supplied standard for how rich one
must be to count as rich is the standard accepted by the quantificational subject.
Yet identifying the richness-standard variable d??? under ‘might’ as di would rep-
resent the global reading, where the standard is supplied by the basic contextually
determined assignment g; and identifying the variable as di[+a] would represent the
local reading shifted by ‘might’, where the standard is supplied by the assignment
representing the witness epistemic possibility.
Such examples can be multiplied. Consider (1.31) on the intermediate reading
of ‘every adult’, where the relevant domain restriction represents the set of individ-
uals considered relevant by x in matters of voting rights, for each bigot x (ignoring
domain restrictions for ‘every bigot’ and ‘most people’).

(1.31) ‘Every bigot thinks most people think every adult can vote’
a. ≈ every bigot@ thinksk@ [most peoplek thinknk [every adultn Pk can-voten ]]
b. Intended reading: Al, Betty, and Chip are the bigots, who are prejudiced in
favor of their groups GA , GB , GC , respectively. Al thinks that most people
think that all the GA s (individuals in GA ) can vote; Betty thinks that most
people think that all the GB s can vote; Chip thinks that most people think
that all the GC s can vote.

This says that every bigot x is optimistic about most people’s beliefs about whether
anyone is improperly excluded (by x’s lights) from being legally permitted to vote.
To recap: Certain prominent phenomena that have led theorists to posit syntac-
tic world variables and object-language quantification over worlds can be observed
with features associated with various context-sensitive expressions. This raises a
challenge for any account providing distinct treatments of world- and individual-
quantification/shifting, on the one hand, and quantification/shifting with context-
sensitive expressions more generally. One response would be to further complicate
the syntax/semantics and attempt to capture the spectrum of local/non-local read-
ings within an operator-based framework (see von Fintel & Heim 2011 for general
discussion and possible strategies). I suggest that we put such epicycles to the side.
The present project pursues a fundamental reorientation in theorizing about
context and “shiftiness” in natural language. The traditional approach takes unshifted
readings for context-sensitive expressions as paradigmatic. Here is Kaplan: “What is
common to [indexical words] is that the referent is dependent on the context of use
and that the meaning of the word provides a rule which determines the referent
in terms of certain aspects of the context” (1989: 490; cf. Fillmore 1971/1997,
Kratzer 1998a, Rullmann 2004). Though such an approach might seem initially

16
plausible for English expressions such as (say) ‘I’ or ‘here’, it is puzzling from the
perspective of the broader spectrum of linguistic shifting phenomena — hence the
plethora of mechanisms for intensionality, quantification, and context-sensitivity,
and epicycles for capturing local readings across context-sensitive expressions. In-
deed Kaplan goes so far as to treat strings with referential readings and strings with
bound-variable readings as uses of homonyms (1989: 489–490). Lest one scoff, note
that Kratzer’s (1998a, 2009) “minimal pronoun” account of apparent bound-variable
(“fake indexical”) uses of 1st and 2nd person pronouns is a homonym account.1⁰
(I use ‘⋋’ to distinguish the assumed object-language binders.)

(1.32) a. Only I got a question that I understood.

b. [Only [1st]] ⋋1 got a question that ∅1 understood.
(1.33) J[1st]Kg,c = speaker(c)
J∅n Kg,c = g(n) (Kratzer 1998a: 94–95)

What classical theories ostensibly gain when it comes to prototypical global read-
ings, they lose when it comes to the spectrum of tendencies for local/global readings
across context-sensitive expressions. The notion of a crosslinguistic class of spe-
cialized anaphoric pronouns — pronouns conventionally excluded from receiving
their interpretation from the discourse context (§1) — is a borderline conceptual
impossibility from a classical perspective.
It is time to rethink the foundational assumptions about shifting and context-
sensitivity motivating the traditional formalism. Our understanding of the rich-
ness of contextual dependencies in natural language — e.g., indexicality, intension-
1⁰Further, whereas local fake indexicals are treated as minimal pronouns (mere indices) inter-
preted via an ordinary λ-binder ((i)), long-distance fake indexicals are treated as fully-specified
pronouns interpreted via distinctive context-shifting λ-binders ((ii)), syncategorematically defined
as in (iii). The “true” and “fake” indexical uses of the string ‘you’ in (i-a) are homonyms; the local
and long-distance fake indexical uses of ‘you’ in (i-a) and (ii-a), respectively, are homonyms; and the
binder indices in (i-b) and (ii-b) have distinct interpretation rules.
(i) a. Only you got a question that you understood.
b. Only [2nd] [⋋2 got a question that ∅2 understood]
(ii) a. You are the only one who knows somebody who understands your paper.
b. [2nd] … [⋋[2nd] know somebody who understands [2nd]’s paper]
(iii) J⋋n αKg,c = λx.JαKg ,c , where g′ is like g, except possibly that g′ (n) = x
′

J⋋[1st] αKg,c = λx.JαKg,c , where c′ is like c, except possibly that speaker(c′ ) = x

′

J⋋[2nd] αKg,c = λx.JαKg,c , where c′ is like c, except possibly that addressee(c′ ) = x

′

(from Kratzer 2009: exs. 53–54, 56–57; cf. 69–73)

17
ality, logophoricity, perspective, projection, local context, assignment shift — has
come a long way since e.g. Kaplan’s (in)famous ban on “monsters.” The project
in what follows is to develop a theory which takes the opposite approach. Shifta-
bility for context-sensitive expressions is the default; unshiftability and obligatory
shifting on the poles of the spectrum are what call for special explanation. The
proposed assignment-variable-based syntax/semantics provides unified analyses of
shifted/non-shifted readings among context-sensitive expressions, and a fully com-
positional standardization of quantification across domains (see §1). I encourage
the development of alternative overall theories with which the following assignment-
variable-based account may be compared.

2 Basics
2.1 Formal overview: Semantic values, models, domains, assignments, composition
I begin with core elements of the basic syntax/semantics.
Rather than having a traditional interpretation function (J⋅Kg )g∈G parameterized
by assignments (worlds, etc.), we have an unrelativized interpretation function J⋅K ,
which assigns expressions semantic values in terms of sets of assignments in the
model. (I’ll ignore tense/aspect and times/events.)

(2.1) Models M:
– E: set of entities
– T: set of truth-values, {0, 1}
– W: set of worlds
– G: set of assignments

Theoretically, I treat assignments as representing a possibility. This interpretation is

in keeping with common talk of contextually determined assignments representing
what world is actual, objects’ relative saliences, speakers’ intentions, attention, etc.
(Heim & Kratzer 1998, Schlenker 2003, Heim 2008). For instance, a syntactic
representation it7 and assignment mapping ⟨7, e⟩ to Fluffy might represent an in-
tention to refer to Fluffy with a token use of ‘it’ and a possibility in which Fluffy is
the center of attention.
It is common to identify basic semantic types with sets in the model. Since ex-
pressions’ semantic values will be treated as involving functions from assignments,
it will simplify our formalism to define semantic types in terms of such functions
(cf. Kobele 2010; contrast Sternefeld 1998). For instance, functions from assign-

18
ments to truth-values in T are type t, and functions from assignments to entities in
E are type e.
Care must be taken in our formalization of assignments and semantic types.
Including assignment variables and variables of arbitrary types has the potential
for paradoxes or non-wellfoundedness (cf. Groenendijk & Stokhof 1990, 1991;
for discussion see Chierchia 1994). For instance, one cannot have a case where
g(i) = g, for some assignment-index i, lest there be assignments g in their own
codomain. Likewise, we cannot allow ourselves to ask whether an assignment g is in
the value assigned to an index k for a set of assignments, i.e. whether g ∈ g(k): given
that functions are sets of ordered pairs, having a case where ⟨k, S⟩ ∈ g ∧ g ∈ S, for
some set of assignments S, would violate standard set-theoretic bans on ∈-chains.
Let the set of assignments G in the model be a set of ordinary assignments h —
functions from typed numerical indices ⟨n, τ ⟩, for any non-assignment type τ , to
elements of the model. For instance, h(⟨4, e⟩) returns an entity o ∈ E, say Fluffy;
h(⟨4, et⟩) returns (the characteristic function of) a set of entities in E, say {Fluffy, Fido};
and so on. Ordinary assignments h ∈ G are undefined for indices for assignments
or functions involving assignments. I let the domain of assignments Dg be a set
of assignments whose domain also includes indices ⟨n, a⟩ for assignments — i.e.,
where dom(g) = dom(h) ∪ {⟨n, a⟩ ∶ n ∈ N}, and range(g) = range(h) ∪ G. For
instance, for some g ∈ Dg , g(⟨2, a⟩) is an ordinary assignment h ∈ G; g(⟨2, e⟩)
is an entity o ∈ E, say Fido; and so on. (For purposes of the present exposition,
I refer to ordinary assignments h ∈ G as “assignmentsM ” (for assignments in the
Model), and to the richer assignments g ∈ Dg as “assignmentsD ” (for assignments in
the Domain). When the distinction is irrelevant I ignore the subscripting; context
should disambiguate. For readability I often abbreviate indices ⟨i, α⟩ with iα.)
Our system avoids the worries mentioned above regarding non-wellfoundedness
and ∈-chains. Since assignmentsD gg return elements in the model, there is no case
of an assignment being in its own codomain. For h ∈ G, h(ia) is undefined; and
for g ∈ Dg , g(ia) = hj ≠ g. Since there are no high type assignment indices in the
domain of assignments in Dg (or G), expressions such as h(iat), g(iat), g(ia)(iat),
etc. are undefined. Such a restriction in assignments’ domains is motivated by our
posited theoretical role for assignment-variables as an object-language mechanism
for tracking the interpretation of context-sensitive expressions; yet the assumption
that there are no pronouns for sets of assignments is ultimately an empirical one.
So, our semantic types are as follows, where the set of assignmentsM G and do-
main of assignmentsD Dg are defined as above — i.e. where the inputs of assignmentsM
h ∈ G are (non-assignment-)indices ⟨n, τ ⟩ for τ ≠ g, a, ⟨. . . g/a . . . ⟩, and the inputs
of assignmentsD g ∈ Dg are indices ⟨n, σ⟩ for σ ≠ g, ⟨. . . g/a . . . ⟩. (Although the for-

19
malization of assignments excludes pronouns for functions involving assignments,
there may be other expressions of type ⟨a, t⟩, etc.; more on this below.)11

(2.2) Domains / Semantic types:

– Dg = domain of assignmentsD
– De = EDg
– Dt = {0, 1}Dg
– Ds = WDg
– Da = GDg
– Dαβ = DDβ α

Unlike previous accounts introducing semantic types for assignments (§1), I

let the object-language include variables for assignments. A natural preliminary
idea would be to identify variable denotations with functions from assignments to
elements in the model — e.g., treating the denotation of an individual-variable Joi K
as λgg .g(ie), where g(ie) ∈ E.12 Such a move is unavailable in the system as devel-
oped thus far. For instance, an assignment-variable denotation (function a ∶ Dg → G)
couldn’t combine via (say) ordinary function application or function composition
with functions x ∶ Dg → E (individual-variable denotation), p ∶ Dg → 2W (proposition-
variable denotation), etc.
As one way of allowing for variables of complex types, I treat non-assignment
variables as having an initial argument of type a, and I treat semantic composition as
proceeding via function application. Variables viα for basic non-assignment types α
denote functions Jviα K ∈ Daα such that for any aa , gg , Jviα K(a)(g) = a(g)(iα) — e.g.,
Jo1 g1 K = Jo1 K(Jg1 K) = λgg .g(1a)(1e). Variables of complex types may be defined
via a metalanguage “down”-style operator ↓ which maps an element of a domain to
an item composed out of associated lowered elements of the model:

(2.3) For γ of (possibly basic) type σ = ⟨σn , ⟨. . . , σ0 ⟩ ⋯ ⟩, ↓ γ is defined by the

condition that, for any gg :
(↓ γ)(g) = the (possibly nullary) function f ∈ M s.t. for any γσnn . . . γσ1 1 ,
γ(γ n ) . . . (γ 1 )(g) = f((↓ γ n )(g)) . . . ((↓ γ 1 )(g))
11Thanks to Salvatore Florio, Dilip Ninan, and Ede Zimmermann for discussion.
12I use bold and single quotes for object-language expressions; I’ll continue to use bold also in
highlighting key points. I use gi for assignment variables, wi for world variables, and oi for individual
variables. I continue to use h, h′ , . . . for “ordinary” assignmentsM in G in the model, and g, g′ , . . . for
assignmentsD of type g in Dg .

20
For the degenerate case where γβ is of basic type β ∈ {e, s, t, a}, ↓ γ is a function
from an assignmentD gg to the item in the model that is the image under γ of g.
For instance, for x ∈ De , (↓ x)(g) is the individual (nullary function) o ∈ E s.t.
o = x(g), i.e. (↓ x)(g) = x(g). For P ∈ D⟨e,t⟩ , (↓ P)(g) is the function f ∶ E → T
such that for any xe , P(x)(g) = f((↓ x)(g)) = f(x(g)); and so on. The denotations
of viσ of complex types σ = ⟨σn , ⟨. . . , σ0 ⟩ ⋯ ⟩ can be defined accordingly as func-
tions Jviσ K ∈ Daσ such that, for any gg , aa , γσnn , . . . , γσ1 1 , Jviσ K(a)(γ n ) . . . (γ 1 )(g) =
a(g)(iσ)((↓ γ n )(g)) . . . ((↓ γ 1 )(g)). For instance, a pronoun for a set of worlds [p2 g1 ]
denotes a function Jp2 K(Jg1 K) ∈ Dst such that for any ws , gg , g(1a)(2st)((↓ w)(g)) =
g(1a)(2st)(w(g)); a choice-function pronoun [F1 g1 ] denotes a function JF1 K(Jg1 K) ∈
D⟨⟨e,t⟩,e⟩ such that for any P⟨e,t⟩ , gg , g(1a)(1ete)((↓ P)(g)) is a selected o ∈ E in (the
characteristic set of) (↓ P)(g), where (↓ P)(g) is the function f ∶ E → T such that
f(x(g)) = f(x)(g) for any xe ; and so on. I treat the semantic values of traces equiva-
lently yet lacking the initial type a argument — e.g., Jt1e K = λgg .g(1e). For complex
type σ = ⟨σn , ⟨. . . , σ0 ⟩ ⋯ ⟩, trace t, and pronoun-variable v (variables of basic types
could be understood degenerately where n = 0 and σ 0 ∈ {e, s, t, a}):13

(2.4) Jviσ K = λaa .λγσnn . . . λγσ1 1 .λgg . a(g)(iσ)((↓ γ n )(g)) . . . ((↓ γ 1 )(g))
a. For α ∈ {e, s, t}, Jviα K = λaa .λgg . a(g)(iα) b. Jgi K = λgg . g(ia)
(2.5) Jtiσ K = λγσnn . . . λγσ1 1 .λgg . g(iσ)((↓ γ n )(g)) . . . ((↓ γ 1 )(g))
13For presentational purposes I provide both interpretation rules in (2.4)–(2.5), though the
account isn’t committed to a distinction between traces and pronoun-variables. All variables may
be interpreted as in (2.4), and the definition of Quantifier Raising (May 1977, 1985) can be adapted
accordingly so that traces are sister to an identity function on assignments, e.g. as in (i) (say assuming
prior indexing; the superscript index is a binder-index feature (see below, §3.5)).
(i) QR (alternative): Move α⟨i,σ⟩ from XP. Leave via if σ = a, [viσ G] otherwise.
● JGK = λgg .g− , where g− = ιh ∈ G∶ ∀τ ≠ a∶ h(iτ ) = g(iτ )

a. [S . . . everyone⟨4,e⟩ . . . ] ⇒ everyone⟨4,e⟩ [S . . . o4e G . . . ]

For expository purposes I will continue to use traces interpreted as in (2.5) in representing displace-
ment (cf. Chomsky 1981). The framework is also compatible with a minimalist syntax which rejects
traces as theoretical primitives. For instance, remnants of movement may be understood as copies
of the moved expression, as in a copy theory of movement (Chomsky 1993, 1995, Sauerland 1998,
Fox 2000, 2002, Hornstein 2001, Nunes 2004; see Takahashi 2010 for comparative overview).
Unless indicated otherwise, one may assume that non-semantically-driven movement is undone at
LF, and any (higher) copies not necessary for semantic composition are uninterpreted. In cases
of movement for type reasons, the lower copies may be interpreted via a simple LF-interface rule
replacing the copy with a coindexed variable, e.g. as in (ii)–(iii) (irrelevant subscript indices deleted).
(ii) a. QR (alternative): Copy α⟨i,σ⟩ from A, and Merge with A.
b. Lower Copy Replacement: Replace α⟨i,σ⟩ with via if σ = a, [viσ G] otherwise.

21
 a. For β ∈ {e, s, t, a}, Jtiβ K = λgg . g(iβ)

(Note that variables g for assignments are type a (functions Dg → G); there are no
denotations of type g. In what follows I use ‘pronoun’/‘pro-form’ both in the present
technical sense for a pronoun-variable+assignment-variable [viσ gk ] complex, and
informally for pronounced forms such as ‘it’, ‘she’, etc.; context should disambiguate.)
2.2 Preliminary derivation: Pronouns, quantifiers, quantification
To get a feel for the basic system it will be instructive to consider a preliminary
derivation. I begin with the simple sentence in (2.6) with a free pronoun and subject-
position quantifier (see n. 12). (T is an assumed topmost assignment-binder, where
g− is, intuitively, the counterpart assignmentM h of g in G; see below. For space
purposes I leave intermediate calculations to the reader.)1⁴

(iii) a. [everyone] [⟨5,e⟩ ]

[everyone⟨5,e⟩ ]
⋮
[S Alice beat everyone⟨5,e⟩ ]
⋮
[S everyone⟨5,e⟩ [S Alice beat everyone⟨5,e⟩ ]]
b. LF ≈ [S everyone⟨5,e⟩ [S Alice beat [v5e G]]]
The precise formulation of such principles may be adapted for alternative views on the theoretical
status of indices. The general framework is compatible with treating chain relations as represented
in the narrow syntax or as captured in the interfaces (e.g., subscript indices could be understood as
explicitly represented in lexical item tokens in a lexical array (or a device for identifying complex
syntactic objects individuated in terms of such indices), or as representing the syntactic relations on
which (co)indexing in the interfaces supervenes; see also Collins & Stabler 2016, Larson 2016 for
discussion). (As we will see, the accounts of relativization and interrogatives will avoid introducing
additional rules for interpreting structures derived from movement, such as Predicate Abstraction
(Heim & Kratzer 1998) or Trace Conversion (Fox 2000, 2002, 2003; cf. Sauerland 2004, Johnson
2012, Moulton 2015).)
1⁴I ignore tense, aspect, voice. I often abbreviate “f(x) = 1” with “f(x).” To a first approximation,
g[i/n] is the unique assignment g′ that maps n to i and is otherwise identical to g; an improved
definition of metalanguage expressions of assignment-modification will be provided in §4.

22
(2.6) Everything loves it.
S∶t

T⟨1,a⟩ ∶ ⟨t, t⟩ S1 ∶ t

everything ∶ ⟨et, t⟩ VP ∶ ⟨e, t⟩

⟨e, et⟩ e

loves ∶ ⟨s, ⟨e, et⟩⟩ s it1 ∶ ⟨a, e⟩ g1 ∶ a

w1 ∶ ⟨a, s⟩ g1 ∶ a

JlovesK = λws .λxe .λye .λgg . y(g) loves x(g) in w(g)

JeverythingK = λP⟨e,t⟩ .λgg .∀xe ∶ P(x)(g)
JT⟨1,a⟩ K = λTt .λgg . for a = λgg .g− , T(g[a(g)/1a]) (provisional)
● g− ∶= the h ∈ G s.t. for all non-assignment indices iτ , h(iτ ) = g(iτ )

Jw1 g1 K = Jw1 K(Jg1 K)

= [λaa .[λgg . a(g)(1s)]](λgg .g(1a))
= λgg .g(1a)(1s)
Jit1 g1 K = Jit1 K(Jg1 K)
= [λaa .[λgg . a(g)(1e)]](λgg .g(1a))
= λgg .g(1a)(1e)
JS1 K = JeverythingK(JlovesK(Jw1 K(Jg1 K))(Jit1 K(Jg1 K)))
= λgg .∀xe ∶ x(g) loves g(1a)(1e) in g(1a)(1s)
JSK = JT⟨1,a⟩ K(JS1 K)
= λgg .for aa = λgg .g− , ∀xe ∶ x(g[a(g)/1a]) loves g[a(g)/1a](1a)(1e) in g[a(g)/1a](1a)(1s)
= λgg .∀xe ∶ x(g[g− /1a]) loves g− (1e) in g− (1s)

23
 S is true in c iff JSK(gc ) = 1
iff ∀xe ∶ x(gc [g−c /1a]) loves g−c (1e) in g−c (1s)

First, pronouns are sister to assignment variables, which determine their interpreta-
tion. I assume that sentences have a topmost assignment-binder T⟨i,a⟩ , which effec-
tively maps variables sister to assignment variables coindexed with T⟨i,a⟩ to the values
provided by the input assignment. This anchors intuitively free pronouns to the
discourse context via the definition of truth-in-a-context (Percus 2000, von Fintel
& Heim 2011). In more complex examples, alternative local/global readings will
be reflected in different coindexings on assignment variables. A result of §3 will
be a derivation of binder expressions, like T⟨i,a⟩ , from an independently defined
generalized (cross-categorial, type-flexible) binder-index ⟨i,σ⟩ .
A more complex clausal architecture will be provided in §3. The world argu-
ment of a clause’s main predicate will be supplied by a world-trace, rather than
by a world+assignment-variable complex. For the moment, given the present sim-
pler syntax, one may assume that in the intended interpretation the first-positioned
world g(1s) represents the world of the possibility represented by g.
Importantly, the outputs of assignments are items in the model. What is loved
according to (2.6) isn’t a function y ∈ De but an individual o ∈ E, say Fluffy. Likewise,
although the metalanguage quantification is over functions x ∶ Dg → E, the items in
terms of which the condition is stated are images of the given assignment g under x,
i.e. individuals o ∈ E in the model. The universal quantification over xe includes
functions mapping g to object o1 ∈ E, functions mapping g to o2 ∈ E, etc. The
metalanguage quantificational condition ∀xe ∶ . . . in (2.6) is satisfied iff regardless
of which such function we look at, its value o ∈ E loves the contextually relevant
individual (=Fluffy): if there was an oi ∈ E that didn’t love Fluffy, then any function
xi ∈ De mapping gc to oi would be such that xi (gc ) doesn’t love gc (1e) (=Fluffy),
falsifying the condition; and if there was a function xj ∈ De whose value given gc
doesn’t love gc (1e), then there would be an oj ∈ De , namely xj (gc ), that doesn’t
love Fluffy (=gc (1e)). In this way the universal quantification over functions x ∈ De
makes a claim about every object o ∈ E in the set of entities (cf. Kobele 2010).
So, the semantics derives that S is true in c iff in the world of c everything loves
the individual o ∈ E represented by 1.

3 Syntax and semantics: Standardizing quantification

Adding assignment-variables to the object language raises non-trivial issues for the
syntax and lexical/compositional semantics. This section shows how a theory with

24
assignment-variables can be integrated into independently motivated treatments of
the syntax/semantics interface. The revised account captures phenomena of in-
tensionality and (non-)shifted interpretations of context-sensitive expressions via
general mechanisms of movement and variable binding, and affords an elegant stan-
dardization of quantification in the syntax and semantics (§1).
The particular treatments of the syntax and lexical/compositional semantics in
the remainder of the paper are of course not the only way of developing an assignment-
variable-based theory. I will spare the reader all my other failed attempts. I welcome
the development of alternatives with which the account may be compared.1⁵
3.1 Preamble
A worry with any framework positing object-language variables for worlds/times/etc.
is that they have the potential to overgenerate readings. The worry might seem espe-
cially pressing for a theory with assignment variables. Absent additional constraints,
nothing would seem to exclude a structure/interpretation such as (3.1), where the
embedded pronoun receives a local reading, being sister to an assignment variable
coindexed with ‘think’, and the embedded world variable receives a global reading,
being sister to an assignment variable coindexed with the topmost assignment-binder.

(3.1) Alice thinks it cried.

a.
T⟨1,a⟩

Alice

thinks⟨2,a⟩ w1 g1 it4 g2
⟨s, ⟨t, ⟨e, t⟩⟩⟩ cried w1 g1

1⁵I want to flag that the general assignment-variable framework, as from §2, doesn’t itself require a
project of standardizing quantification. One could build assignment-binding properties directly into
the lexical entries for modals, attitude verbs, etc., as in the Hintikka-style lexical entry in (i) (letting
Dox(o, u) be a set of assignments (possibilities) compatible with o’s beliefs in u).
(i) Jthink⟨i,a⟩ K = λws .λTt .λxe .λgg . ∀h ∈ Dox(x(g), w(g))∶ T(g[h/⟨i, a⟩])
Such a lexical entry contrasts with lexical entries for determiner quantifiers ((2.6)), where binding
of particular variables results from combining a binder-index with the quantifier, triggered (e.g.) by
movement. Methodologically, it is worth examining the prospects for an approach which unifies the
treatments of the various shifting phenomena. So I put options such as (i) aside.

25
 b. ≈ for every h representing a possibility compatible with Alice’s beliefs in
the actual world (=gc (1s)), the individual o ∈ E represented with 4 by h
(=h(4e)) cried in the actual world

A proliferation of constraints on readings seems in the offing.

It is important not to overstate the explanatory burdens particular to theories
positing object-language variables for worlds, times, assignments, etc. Take ‘it’. Sup-
pose for the sake of argument that ‘it’ cannot receive a shifted (local) reading under
e.g. modals/attitude verbs and that this constraint is conventionalized. Following
Kaplan and friends, such a constraint could be formally implemented by (say) an-
alyzing ‘it’ as a variable receiving its interpretation from a contextual parameter
on the interpretation function, and disallowing attitude verbs, modals, etc. from
shifting such a contextual parameter. An explanatory inquisition isn’t far behind.
What makes it the case that that formalism correctly represents the conventional
meaning and use of the string ‘i-t’ in such-and-such communities? Why would such-
and-such contextual parameter be unable to be shifted by attitude verbs, modals,
etc., though it can be shifted by other operators such as determiner quantifiers? If
‘it’ is analyzed as receiving its interpretation from a modally-unshiftable contex-
tual parameter, what explains the fact that other pronouns and context-sensitive
expressions can receive shifted readings in modal environments? — and by what
alternative mechanisms are the different readings compositionally derived? If there
is a constraint against local readings of ‘it’ in English, is the constraint universally
associated with analogous pronouns across languages? If so, what general aspects of
human cognition, sociality, conversation, etc. explain the crosslinguistic universal?
There is much one might say in response. For instance, with a first-person pro-
noun there is a natural relation between speaker and attitude subject that may ex-
plain the ready retrievability of shifted interpretations in attitude ascriptions, as are
indeed attested in various languages (arguably including English). No such general
relation seems available with ‘it’; addressees don’t generally know what the speaker
may be presupposing about what some attitude subject takes as relevantly salient.
Given the paucity of descriptive content of ‘it’, shifted readings under attitude verbs,
modals, etc. would seem generally unretrievable. For theories using unshiftable (or
selectively shiftable) context/assignment parameters on the interpretation function,
such stories may be understood at the “presemantic”/metasemantic level of what
formal objects correctly represent the shifting possibilities for a given string, or of
explaining what makes it the case that a given string corresponds to a lexical item
receiving its interpretation from such-and-such parameter; for a theory positing
object-language assignment-variables, at (say) the syntactic or lexical semantic level

26
of why there is a conventionalized locality/globality principle for a given expression.
All types of theories must ultimately provide an explanation of the contrasting ten-
dencies — and in some cases conventionalized constraints — for local/global read-
ings among expressions, both in English and crosslinguistically. Where one does is
a matter of bookkeeping (cf. Silk 2016, 2017).
Of course not all ways of carving up the explanatory terrain are empirically or
theoretically on a par — hence the present project. The classical approach takes
unshiftability for context-sensitive expressions as the default. In light of the spec-
trum of tendencies for local/global readings, I suggest we take the opposite tack.
Individual-, world-, and assignment-shifting are given a uniform general analysis.
Optionality with respect to local vs. global readings is the default; unshiftability and
obligatory shifting on the poles of the spectrum are what call for special explanation
(more on which in due course). We will see that proceeding in this way, and in-
troducing assignment variables into the syntax/semantics, affords diverse empirical
and theoretical advantages. In semantics as in tailoring (so I’m told), it is often easier
to start big and take in.
Preamble (=rant) over; the proof of the pudding is in the lambdas and whatnot,
so let us proceed.
Some limitations and patterns in available readings may be derived conversa-
tionally; not all LFs may be equally likely in representing speakers’ intentions in
uttering a given string across concrete discourses. Yet some constraints on readings
are certainly grammaticalized. In what follows I would like to focus primarily on
one constraint on readings, to begin reining in the system’s flexibility: the constraint
on embedded world-variables. (I revisit obligatory local/global readings of certain
pronouns in §4.3.) As Percus 2000 observes, the world argument of a clause’s main
predicate must be bound by the closest world-binder. In the present framework, the
aim is to derive that the main predicate’s world argument receives an obligatory local
reading, and to do so in a way that allows other embedded variables to receive global
readings linked to the discourse context. This section develops the preliminary §2-
account to capture these points. The revised account derives the binding of specific
variables from basic lexical entries and a generalized binder-index (§§1–2).

27
3.2 Type-driven movement
There is a familiar story about what generates binder/bindee relations with object-
position quantifiers over individuals: the quantifier moves because of a type mis-
match, and a binder-index attaches to the quantifier, leaving a coindexed trace:1⁶

(3.2) It loves everything.

S∶t

T⟨1,a⟩ ∶ ⟨t, t⟩ S′ ∶ t

everything⟨2,e⟩ ∶ ⟨t, t⟩ S′′ ∶ t

it1 g1
t2e
loves w1 g1
Preliminary type-specific denotation for the binder-index:
J⟨i,e⟩ K = λQ⟨et,t⟩ .λTt .λgg . Q(λxe .λg′′g .T(g[x(g′′ )/⟨i, e⟩]))(g)

Jeverything⟨2,e⟩ K = J⟨2,e⟩ K(JeverythingK)

= [λTt .[λgg .∀xe ∶ T(g[x(g)/2e])]]
Jt2e K = λgg .g(2e)
JS′′ K = JlovesK(Jw1 K(Jg1 K)) (Jt2e K) (Jit1 K(Jg1 K))
= λgg .g(1a)(1e) loves g(2e) in g(1a)(1s)

JS′ K = Jeverything⟨2,e⟩ K(JS′′ K)

= λgg .∀xe ∶ g[x(g)/2e](1a)(1e) loves x(g) in g[x(g)/2e](1a)(1s)
1⁶When labeling trees I sometimes use prime symbols A′ informally to distinguish different nodes
of category A, and sometimes formally for A/A-bar in the sense of X-bar theory. Context should
disambiguate. Note that the argument of the raised quantifier is type t (cf. Heim 1982, Kobele 2010,
Kennedy 2014), rather than property type (e.g. Heim & Kratzer 1998). The account maintains
the traditional view in syntax of representing indices as features on expressions: the binder-index
attaches directly to the moved expression, rather than occupying its own node and triggering a special
composition rule such as Predicate Abstraction (Heim & Kratzer 1998). More on this below.

28
 JSK = JT⟨1,a⟩ K(JS′ K)
= λgg .∀xe ∶ g− (1e) loves x(g[g− /1a]) in g− (1s)

Roughly, the binder-index combines with the quantifier so that the quantifier’s scope
argument becomes the set of individuals that make the proposition JS′′ K true when
returned as value for 2e (n. 16).
A natural hypothesis is to treat a parallel mechanism as at play with world- and
assignment-quantification. I suggest that we treat the relevant items as quantifiers,
and, like determiner quantifiers, as moving because of a type mismatch (cf. Hac-
quard 2006, 2010, von Stechow 2008). Specifically, I treat the complementizer
(e.g. ‘that’) as base-generated at the position of the main predicate’s world argument;
as a higher quantifier type over worlds, it moves, leaving a world-trace. I treat
modals and attitude verbs as base-generated at the position of a posited assignment
argument of the C head; as a higher quantifier type over assignments, it moves,
leaving an assignment-trace:1⁷

(3.3)
X⟨i,a⟩ CP/C
⟨j,s⟩
C0 ⋮

C tia V tjs

The complementizer introduces world-quantification/binding, i.e. intensionality;

modals, broadly construed (n. 17), introduce assignment-quantification/binding.
Both cases proceed parallel to the case of raised quantifiers: a binder-index attaches
to the quantificational expression (quantifier, complementizer, modal) due to type-
driven movement, leaving a coindexed (individual, world, assignment) trace.
There are precedents for treating the syntax/semantics interface in the proposed
way. Hacquard 2006, 2010 makes such a move for aspect, treating aspect as a
quantifier over events and moving from the verb’s event-argument position (see also
Moulton 2015). von Stechow 2008, following Heim 2001, treats attitude verbs
as quantifiers over worlds and moving from inside the complement (cf. von Fintel
& Iatridou 2009). Similar semantic effects of head movement are defended in
1⁷I often use ‘modal’ broadly for semantically modal elements of various categories (modal
verbs, attitude verbs, T), sometimes narrowly for modal verbs; context should disambiguate. X is
a placeholder for the category of the semantically modal item. I label the sister of T at LF as CP;
however, officially, I leave open its specific location or category, e.g. whether it raises to SpecCP or
heads some extended projection of the CP-layer such as ForceP (cf. Rizzi 1997).

29
Lechner 2006 and Szabolcsi 2011. As we will see in §6, a prominent approach
in syntax is to treat quantifiers as moving from within the nominal/clausal comple-
ment (Campbell 1996, Matthewson 1998, Ihsane & Puskás 2001, Giusti 2002,
Luján 2004). Heim 2012 treats the question operator as moving from an internal
argument of the interrogative complementizer as a quantifier over propositions (§8).
On a more general level, in light of crosslinguistic phenomena with nominal tense
Lecarme (1996, 1999b, 2004) offers parallel syntax for clauses and noun phrases
in terms of verbal/nominal T-chains: the main (verbal/nominal) predicate’s event
argument ei is assumed to be bound by a higher operator OPi in agreement with
T(ense)i (where OPi /Ti may themselves bear morphosyntactic relations with C/D),
as in (3.4).

(3.4) Clausal/Nominal T-chains OPi , Ti , V(ei )/N(ei )

a. [ CP OPi [ C C [ TP Ti [ … [ VP … ei V …]]]]]
b. [ DP OPi [ D D [ TP Ti [ … [ NP … ei N …]]]]]
(Lecarme 1999b, 2004; after Guéron & Hoekstra 1995, Campbell 1996)

Lecarme doesn’t address what generates the assumed operator or establishes the
(obligatory) binding relation between OPi and ei , what the operator’s specific seman-
tics is, or how the assumed agreement relations among OPi , Ti , ei figure in a com-
positional derivation of the (temporal, modal) interpretation of the verbal/nominal
predicate.1⁸ Though I will bracket projections such as for Tense, the proposed syn-
tax/semantics in this paper can be understood as deriving the relevant operators
and chain relations in a specific compositional semantics — e.g., linking the main
predicate, complementizer, and an embedding modal via type-driven movement, as
schematized in (3.3) (compositional details in due course). (§6 extends the movement-
based approach to the syntax/semantics interface to quantificational DPs, affording
parallel treatments of nominal and verbal predication, reference, quantification.)
The proposed treatment of the complementizer as the source of intensionality
could be refined accordingly in light of more detailed understandings of the clausal
1⁸Likewise for the other authors mentioned above, whose primary focus also isn’t on the specific
compositional semantics. Lecarme (1999b) writes: With clauses, the “relationship” between the
predicate’s variable ei and operator OPi in SpecCP is “mediated by the R(eference)-time” (e.g.,
S(peech) time); Opi “determines the value of C, which contains the R(eference)-time”; and the
[±past] tenses in Ti “tell us whether the time of an event E is before / after / simultaneous with a
given R-time. If R = S, tense locates the [event] before the ‘now’ of the utterance.” With nominals,
the “operator in Spec,DP binds the variable e-position. This accounts for the context-dependency of
[strong, presuppositional] determiners: DPs are temporally restricted by the context, a local process.”
I don’t know what exactly Lecarme has in mind here regarding the specific compositional semantics.

30
architecture. For instance, the complementizer could be analyzed as raising for
type reasons from an internal world-argument of (say) aspect, which, following
Hacquard, raises as a quantifier over events from the event argument of the main
predicate, as in (3.5), letting v be a type for events.1⁹

(3.5)
X⟨1,a⟩ CP

[C0 ]⟨1,s⟩ ⋮
AspP
C t1a
[Asp0 ]⟨1,v⟩ VP

Asp t1s rain t1v

What is important here is the move of capturing world-quantification/binding (in-
tensionality) via the syntax/semantics of the complementizer, which moves from
an internal world-argument position inside the clause; and capturing assignment-
quantification/binding via the syntax/semantics of modal elements, which move
from the complementizer’s assignment-argument position. For present purposes I
assume the simpler syntax/semantics which treats world-quantification/binding as
arising directly from the complementizer, which moves from the world-argument
position of the verb. (I consider extensions to functional projections with determin-
ers and other types of complementizers (in relative clauses, conditionals, and inter-
rogatives) in §§6–8. Additional independent motivations for the syntactic structures
proposed for particular constructions will follow in due course.)
3.3 World-binding: Complementizer
Modifying the clausal architecture as suggested in §3.2 has direct implications for the
lexical/compositional semantics. Start with the complementizer. The preliminary
§2-implementation adopted a metasemantic constraint that the first world-position
of gc , gc (1s), pick out the world of c. Even if we don’t require this, there will be some
position or other in an assignment that represents the world of the possibility that
the assignment represents (§2). Accordingly, it can be useful to define a metalan-
1⁹See also Hacquard 2006, 2010, Kratzer 2013, and Moulton 2015, which argue that ‘that’-
clauses supply propositional contents indirectly via “anchors” (events, situations), such as provided
by aspect (cf. Silk 2018 on applications to verbal mood). See Moulton 2015 and references therein
on additional possible complications regarding the syntactic derivation of CP complements.

31
guage function @ ∶ G → W that maps an assignment h to the world of the possibility
represented by h. I offer (3.6) as a lexical entry for (possibly unpronounced) ‘that’.

(3.6) JthatK = λaa .λp⟨s,t⟩ .λgg . ∀w s.t. w(g) = @(a(g)), p(w)(g)

(3.7) CP

[C0 ]⟨1,s⟩ IP

that tia rained t1s

Roughly put, (3.6) treats a CP ‘that S’ as relating the set of S-worlds to some further
modal domain; (3.7) says that it rained in the world identical to the world of the
given possibility, @(a(g)).2⁰ The modal element base-generated in the complemen-
tizer’s assignment-argument position will determine the modal domain. Note that
the verb’s world argument is now a trace, directly bound by the complementizer.
Although the complementizer’s condition in (3.6) picks out a unique world in the
model — namely, the world of the given possibility, @(a(g)) — the quantification
over w ∈ Ds is universal. There is a single world u ∈ W identical to the world of a
possibility, but there are many functions w ∈ Ds such that w(g) = u. The functions w
quantified over may differ in their values given other possibilities g′ , yet they agree
in mapping (say) g to u.
3.4 Assignment-binding: T, Modals, Attitude verbs
To fix ideas I focus on the top-level assignment-binder T, the attitude verb ‘think’,
and the modal verb ‘may’. Consider T. We need to ensure two things after it com-
bines with the binder-index: that it interprets the embedded CP with respect to a
modified assignment that maps coindexed assignment-variables to the (counterpart
of the) input assignment (g[g− /ia] above), and that the modal domain for evaluating
the embedded proposition is the actual world. Likewise for modals/attitude verbs,
except that the embedded CP is interpreted with respect to a modified assignment
mapping coindexed assignment-variables to the assignments being quantified over,
and the modal domain is the set of worlds compatible with the modality/attitude.
Our metalanguage function @ offers a way of unifying the modals’ lexical entries:
In each case the modal domain is λw.λg.w(g) = @(a(g)); what differs is the quan-
tification over a — for T it’s over the a such that a(g) = g− for any g; with ‘may’ it’s
2⁰Compare Wiltschko 2016 for a general syntactic framework which treats all functional heads
as predicating identity/non-identity relations between items of the relevant type (worlds, individuals,
etc.). Wiltschko also treats the Complementizer as determining world reference.

32
over a such that a(g) is in the set of accessible possibilities; and with ‘think’ it’s over
a such that a(g) is compatible with the subject’s state of mind:

(3.8) JTK = λA⟨a,t⟩ .λgg . for aa = λgg .g− , A(a)(g)

(3.9) JmayK = λws .λr⟨s,at⟩ .λA⟨a,t⟩ .λgg . for some aa s.t. r(w)(a)(g)∶ A(a)(g)
(3.10) JthinkK = λws .λA⟨a,t⟩ .λxe .λgg . for all aa s.t. a(g) is compatible with x(g)’s
state of mind in w(g)∶ A(a)(g)

The meaning for the modal verb in (3.9) can be understood as adapting a familiar
Kratzer-style semantics, treating modals as quantifying over a set of contextually
relevant possibilities (Kratzer 1977, 1981). For simplicity I use a basic accessibility
relation r (“modal background”), which maps the verb’s world argument to a set of
assignments.21 As usual, the meaning for ‘think’ in (3.10) proceeds analogously,
yet lexically specifying the set of possibilities being quantified over. As we will
see in the compositional derivations in §4, the set of worlds at which a clause is
evaluated is ultimately determined by the assignment-quantification introduced by
the modal. For example, with ‘think’, the complement is evaluated at a multiplicity
of worlds (assuming one isn’t maximally opinionated), i.e. the worlds w(g) identical
to worlds @(a(g)) of assignments representing possibilities compatible with the
subject’s state of mind. With T, the main clause is evaluated at a singleton set,
{@ (a(g))} = {@ (g− )}, i.e. ultimately the world of the assignment representing
the discourse context.
Recall that assignments are understood theoretically as representing possibili-
ties, where “possible ways things might be” may include what world is actual, foci of
attention, objects’ relative saliences, etc. It is a substantive question what it is for an
assignment, thus understood, to be “compatible with” (e.g.) a body of information
or a subject’s state of mind. The task of addressing this question is shared among
semantics which treat expressions such as modals/attitude verbs as quantifying over
assignments (e.g., Cumming 2008, Santorio 2010, 2012); it isn’t novel to a theory
with syntactic assignment-variables. For present purposes I simply note that the
issue is essentially the same as the issue, generally bracketed in formal semantics,
of what it is for an assignment to be the “assignment of the context” gc , or “de-
termine[d]” by the “physical and psychological circumstances” “of the utterance
situation” (Heim & Kratzer 1998: 243). Work on indexical shift and concept gen-
erators may be helpful here in providing further (gramatical/lexical/metasemantic)
21I consider an alternative argument structure for modals in §7.6.

33
resources for reining in the system’s flexibility (§3.1).22 (For instance, one might
require, say, that in the intended interpretation the first-positioned individual in an
assignment representing an epistemic possibility determined by gc be an epistemic
counterpart of the first-positioned individual in gc , who is the speaker of c — or,
generalizing, that the syntactic indexing determined by the concrete discourse be
such that, for any relevant index ⟨i, σ⟩, a(g)(iσ) represents an epistemic counterpart
of gc (iσ). More pedantically, one might require that if a concrete discourse c deter-
mines that a particular syntactic indexing ⟨i, σ⟩ and abstract assignment gc would
represent an intention to pick out such-and-such item s in the model with a token
use of an expression α in c, then for any assignment h representing an epistemic
possibility (possibility compatible with so-and-so’s state of mind, etc.), h(iσ) is an
epistemic (doxastic, etc.) counterpart of gc (iσ) = s.)
3.5 Generalized binder-index
Previous definitions of object language binder indices have been limited to DPs in
the binding of individual variables (e.g., Kobele 2010, Kennedy 2014). The above
treatments of the syntax/semantics afford a means of standardizing quantification
and defining a generalized binder-index, applying to quantificational expressions of
various types. I propose (3.11) — where χτ is a variable for the type of what is being
quantified over, σ is the type of the mother node (i.e. the result of combining the
binding expression with its scope argument), and γσ1 1 . . . γσnn are variables for any
intermediate arguments.

(3.11) Generalized binder-index feature

J⟨i,τ ⟩ K = λα⟨⟨τ,⟨1 σ1 ⋯σn , t⟩1 ⋯⟩n ⟩,σ⟩ . λβ⟨1 σ1 ⋯σn , t⟩1 ⋯⟩n .
α(λχ.λγ 1 ⋯λγ n .λg . β(γ 1 )⋯(γ n )(g[(↓ χ)(g)/iτ ]))
⟨i,e⟩
(3.12) J[everything] K = J⟨i,e⟩ K(JeverythingK)
= λTt .λgg . ∀xe ∶ T(g[x(g)/ie])
(3.13) J[that tja ]⟨i,s⟩ K = J⟨i,s⟩ K(Jthat tja K)
= λTt .λgg . ∀w s.t. w(g) = @(g(ja)), T(g[w(g)/is])
(3.14) J[think tjs ]⟨i,a⟩ K = J⟨i,a⟩ K(Jthink tjs K)
= λTt .λxe .λgg . for all a s.t. a(g) is compatible with x(g)’s state of mind
in g(js)∶ T(g[a(g)/ia])
22For discussion of metasemantic constraints on admissible assignments for the case of shifting
phenomena with individual-denoting expressions, see Santorio 2010, 2012, Ninan 2012; cf. Per-
cus & Sauerland 2003, Schlenker 2003, Anand & Nevins 2004.

34
The binder-index feature defined in (3.11) is cross-categorial and type-flexible — it
attaches to quantificational elements of any category in the binding of variables of
any type, simple or complex. Roughly put, the binder-index ⟨i,τ ⟩ takes an expression
α that quantifies over items of type τ (e.g. individuals/worlds/assignments), and it
lets α combine with its scope β by feeding α the set of τ -type items that verify β
when returned for ⟨i, τ ⟩.
The account in this section contrasts with prominent current theories of the
syntax and semantics of QR, which generally rely on syncategorematic treatments of
binding/quantification — including implementations in trace theories (e.g. Heim &
Kratzer 1998), copy theories (e.g. Fox 2002), and multidominant theories (e.g. John-
son 2012) (n. 24; see Rabern 2012, Kennedy 2014 for critical discussion). The ac-
count developed here avoids treating the movement in QR as introducing a distinct
node for a binder index inserted under the moved expression, and it avoids positing
specialized syntactic and semantic rules for deriving the interpretation of the re-
sulting post-movement structures, such as Predicate Abstraction or Trace Conver-
sion. The binder-index combines with the QR’d expression via function application,
rather than (say) occupying its own node and triggering a special composition rule.
Important advances in syntax have come from explaining movement operations
such as QR in terms of more fundamental grammatical principles — e.g., treating
QR as an instance of the general syntactic rule Move α, α any category (May 1985),
or in later minimalist syntax analyzing Move in terms of Merge (“remerge,” “copy
and merge”) (Chomsky 1993, 1995, Fox 2000, Collins & Stabler 2016). The
assignment-variable account developed here may be understood as following in the
spirit of such developments. On the semantic side, expressions undergoing QR com-
bine with a generalized binder-index via function application, and the QR’d expres-
sion combines with its sister via function application. The compositional semantics
of structures resulting from type-driven movement is an instance of ordinary func-
tion application. On the syntactic side, the account maintains the traditional view in
syntax of representing indices as features on expressions. The movement operations
in QR may proceed by ordinary applications of Merge: The expression α merges with
the binder index in its base position, yielding α⟨i,σ⟩ , remerged later in the deriva-
tion. Though our general framework needn’t be committed to a particular theory
of syntactic structure building or the interfaces, a natural way of understanding the
syntactic contribution of the generalized binder-index is as providing a “trigger”
feature (Collins & Stabler 2016), which would be checked by the application
of Merge in generating the post-movement structure.23 The lower copy may be
23In a model such as that in e.g. Groat & O’neil 1996, type-driven overt vs. covert movements

35
replaced with a coindexed variable viσ directly, as in a trace theory, or via a general
replacement rule at LF (see n. 13). The generalized binder index would thus be
what triggers QR syntactically, and allows for combination via function application
semantically. The result is a syntax and compositional semantics of type-driven
movement in terms of ordinary feature-driven Merge and function application.2⁴
The next section shows how the lexical entries and derived binder denotations in
(e.g., with attitude verbs vs. object-position determiner quantifiers) would be distinguished in terms
of whether phonological features are moved to the head of the chain.
2⁴Contrast the present sort of simplified derivation for a basic case of QR in (i) with the trace-
theoretic, copy-theoretic, and multidominant alternatives in (ii)–(iv) (cf. n. 13). (For concreteness in
(i) I treat the lower copy as being replaced with the coindexed variable via a general LF-interface
rule; see n. 13. Fox’s Trace Conversion in (iii) converts the lower copy of a quantificational DP
to a definite description by replacing the D head (here ‘every’) with ‘the’ and inserting an identity
predicate in the complement (here yielding [[boy][=o5 ]] from [boy]). I return to comparisons with
Trace Conversion in §6. Regarding (iv), in a multidominant syntax remerged syntactic objects are
literally in both positions — hence Johnson’s (2012) need to generate the quantifier (here ∀) in the
higher position, independent of the NP assumed to be its complement. (Johnson assumes that, by
some morphosyntactic principles, the syntactically displaced ∀ and ‘the’ get pronounced as ‘every’.)
I don’t know how such an approach would generalize across varieties of type-driven movement.)
(i) Merge + Function Application
a. [every boy] [⟨5,e⟩ ]
[[every boy]⟨5,e⟩ ] (by Merge)
⋮
[S Alice beat [every boy]⟨5,e⟩ ]
[S [every boy]⟨5,e⟩ [S Alice beat [every boy]⟨5,e⟩ ]] (by Merge)
b. LF: [S [every boy]⟨5,e⟩ [S Alice beat [v5e ]]] (by Copy Replacement)
● Replace α⟨i,σ⟩ with viσ .
c. ≈ for every boy o, Alice beat o (by FA)
(ii) Traces + QR + Predicate Abstraction (PA) (Heim & Kratzer 1998)
a. [every boy]
⋮
SS: [Alice beat [every boy]5 ] (by Merge)
b. LF: [[every boy] [5 [Alice beat t5 ]]] (by QR)
● QR: [. . . αi . . . ] ⇒ [αi [i [. . . ti . . . ]]]
c. ≈ for every boy o, Alice beat o (by PA, FA)
(iii) Copies + Trace Conversion (TC) + PA∗ (cf. Fox 2002, 2003)
a. [every boy]5
⋮
[Alice beat [every boy]5 ]
[[every boy]5 [Alice beat [every boy]5 ]] (by Merge)
b. LF: [[every boy]5 [Alice beat [DP the [NP boy] [=pro5 ]]]] (by TC)
c. ≈ for every boy o, Alice beat the o′ s.t. o′ is a boy ∧ o′ = o (by PA∗ , PM, FA)
● PA∗ : JDPi [. . . DPi . . . ]K = JDPi K (λx.JDPi Kg[x/i] )

36
this section capture the requisite binding relationships in sentences’ quantifications
over individuals/worlds/assignments.

4 Examples
4.1 Attitude Ascription: Intensionality, local/global readings
Start with a simple attitude ascription such as (4.1) with a “free” (global) reading
of an embedded pronoun. (I use ‘Cdec ’ for an unpronounced declarative comple-
mentizer, JCdec K = JthatK assumed to head the main clause (e.g. Chomsky 1995,
Lecarme 1999a, Franco 2013). I often use ‘S’ for root sentences and IPs; see
nn. 16, 17. For readability I abbreviate ‘o’s state of mind in u’ with ‘SOMo,u ’; I
suppress certain irrelevant assignment modifications, indicated with ‘≈’; I leave in-
termediate calculations to the reader (see notes).)2⁵

(iv) Multidominance + Decomposed scattered ‘every’ + Agree + PA∗ (cf. Johnson 2012)
a. [DP [the∗ pro5 ] boy]
⋮
[Alice beat [DP [the∗ pro5 ] [NP boy]]]
[Alice beat [DP [the∗ pro5 ] NP]] [QP ∀ NP] (by Merge)
● NP: identical token = [NP boy], sister to both Q0 and D0 (multidominance)
b. LF: [[Alice beat [DP [the∗ pro5 ] NP]] [QP5 ∀ NP]] (by Merge, Agreement∗ )
∗
● Stipulated Agreement : QPi must have the same index as the index sister to D0 in
[D0 D0 proi ] that is sister to the lower occurrence of the NP sister to Q
c. ≈ for every boy o, Alice beat o, provided o is a boy (by PA∗ , FA)
∗
● Jthe K = λxλP ∶ P(x) = 1 . x

2⁵Though the root declarative complementizer (here written Cdec ) is unpronounced in English,
in other languages the (declarative) complementizer in root sentences may — and in some cases
must — be overtly pronounced (e.g., Lecarme 1999a). Semantic values for certain relevant lower
spinal nodes are given in (i). (See (2.6), (3.2), (3.6)–(3.14) for relevant lexical entries and derived
denotations for the binding expressions.)
(i) JCP′ K = λgg . ∀w s.t. w(g) =, g[w(g)/2s](1a)(2e) cried in w(g)
JVPK = J[think t1s ]⟨2,a⟩ K(JCP′ K)
= [λxe .[λgg . for all a s.t. a(g) is compatible with SOMx(g), g(1s) ,
∀w s.t. w(g[a(g)/2a]) = @(a(g[a(g)/2a])),
g[a(g)/2a][w(g[a(g)/2a])/2s](1a)(2e) cried in w(g[a(g)/2a])]]
JCPK = J[Cdec t1a ]⟨1,s⟩ K(JVPK(λg′g .g′ (1a)(1e)))
= λgg . ∀w′ s.t. w′ (g) = @(g(1a)), JVPK(λg′g .g′ (1a)(1e))(g[w′ (g)/1s])

37
(4.1) He thinks that it cried.
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a
VP
he1 g1

⟨2,a⟩ CP′

thinks t1s
⟨2,s⟩ S′′

that t2a
it2 g1 cried t2s
JSK = JT⟨1,a⟩ K(JCPK) = λgg . for a′a = λg′′g .g′′− , JCPK(g[a′ (g)/1a])
≈ λgg . ∀w′ s.t. w′ (g) = @(g− ),
for all a s.t. a(g) is compatible with SOMg− (1e), w′ (g) ,
∀w s.t. w(g) = @(a(g)),
g− (2e) cried in w(g)

Roughly put: (4.1) is true in c, JSK(gc ), iff gc (2e) cried in the world of every possi-
bility compatible with gc (1e)’s state of mind in @(gc ). (For readability I often omit
the superscript in ‘g−c ’, though it should be understood (§2.2).)
Parallel to the movement of the quantifier in (3.2), movement of the comple-
mentizer from the embedded predicate’s world-argument position leaves a trace, t2s ,
⟨2,s⟩
and the binder-index attaches to the moved expression, [that t2a ] . This captures
Percus’s point (§3.1): Percus’s point is diagnosed as an obligatory local reading of
the predicate’s world argument; it is captured via general mechanisms of movement.
The embedded world argument is obligatorily shifted to the embedding pred-
icate ‘think’, being supplied directly by a trace left from movement of the clause’s
complementizer; however, the embedded pronoun can still receive a non-shifted
reading, receiving its interpretation from an assignment-variable. The intuitively

38
free, or global reading, of ‘it’ is reflected in its being sister to an assignment-variable
coindexed with the topmost assignment-binder, anchoring its interpretation to the
discourse context (via g[a′ (g)/1a](1a)(2e) = a′ (g)(2e) = g(2e)).2⁶
As with individual-quantification (§2.2), although the items quantified over by
the complementizer/modal are functions, the conditions concern worlds/assignments
in the model. This reflects a philosophical point from Stalnaker (1988, 2014), in his
emphasis on understanding shifted “contexts” as derived, in the sense of being deter-
mined by the discourse. Which features of the subject’s state of mind are relevant for
interpreting embedded material can depend on context. The formalism represents
this in treating the condition placed by the attitude verb as a condition on ways (a)
of mapping the discourse assignment (g) to an assignment (a(g)) representing the
subject’s state of mind.
Likewise, the “…cried in w(g)” in the last line shouldn’t mislead. The proposed
meanings for ‘think’ and ‘that’ restrict the quantification to functions w mapping g
to worlds @(a(g)) of the possibilities compatible with the subject’s state of mind.
The attitude ascription requires that, for any such w, the relevant individual cried in
w(g) ∈ W, a world compatible with the subject’s state of mind.
4.2 De re/de dicto, Specific/non-specific: Global vs. local readings of world arguments
(4.1) highlights a contrast between pronouns and traces in the system.2⁷ The trace
filling the world argument of a clause’s main predicate is coindexed with the nearest
c-commanding world-binder due to movement of the complementizer. This cap-
tures the obligatory local reading of the main predicate: (4.2) cannot receive the
interpretation in (4.3).

(4.2) Alice thinks a friend of mine won.

(4.3) (4.2) ≉
a. there is a winner that Alice thinks is a friend of mine
b. for all u′ compatible with Alice’s beliefs in u, some o who won in u is a
friend of mine in u′
2⁶The talk of intuitively free/bound readings of pronouns can be formalized more precisely. An
intuitively “free” reading of a pronoun, represented syntactically via a pronoun-+assignment-variable
complex [viσ gj ], is reflected in an LF where (i) the nearest c-commanding assignment-binder, if any,
is T⟨j,a⟩ , and (ii) there is no ⟨i,σ⟩ -binder c-commanded by the topmost world-/assignment-binders
that c-commands it.
2⁷I return to issues with pronominal anaphora, trace-/pronoun-binding, and weak crossover in
§§6.2–6.3.

39
Pronouns, in contrast, receive their interpretation from an assignment-variable (§2.1).
This predicts that world-pronoun arguments of embedded non-main predicates —
e.g. ‘a friend of mine’ in (4.2) — should receive optional local/global readings.2⁸

(4.4) JaK = λP⟨e,t⟩ .λQ⟨e,t⟩ .λgg . ∃xe ∶ P(x)(g) ∧ Q(x)(g)

JFoMK = λws .λxe . x(g) is a friend of mine in w(g)
Ja [FoM [wi gj ]]K = λQ⟨e,t⟩ .λgg . ∃xe ∶ x(g) is a friend of mine in g(ja)(is) ∧ Q(x)(g)

Binding configurations with world-pronouns afford a locus for capturing classic

contrasts between de re/de dicto and specific/non-specific readings.
Non-specific de dicto readings are captured via structures involving local binding
of the embedded world-pronoun, as in (4.5). Specific de re readings are captured via
structures where the DP is raised, hence local = global binding, as in (4.6):2⁹

(4.5) De dicto, Non-specific:

≈ Alice thinks there is some individual or other who is friend of mine that won
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a
Alice

⟨2,a⟩ CP′

thinks t1s
⟨2,s⟩ S′′

that t2a
DP
won t2s
a
FoM w2 g2

2⁸I bracket potential additional structure from quantifier domain variables (von Fintel 1994,
Stanley & Szabó 2000, Martí 2002, Stanley 2002, Büring 2004, Etxeberria 2005, Gillon
2013).
2⁹See §6 for discussion of alternative treatments of specific readings with indefinites.

40
(4.6) De re, Specific:
≈ there is some particular individual who is a friend of mine such that Alice
thinks (s)he won
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a

DP⟨1,e⟩

a Alice
FoM w1 g1 ⟨2,a⟩ CP′

thinks t1s ⟨2,s⟩ S′′

that t2a t1e

won t2s

As observed in Fodor 1970, DPs such as ‘a friend of mine’ in (4.2) can also have a
so-called non-specific de re reading — informally, a reading ascribing a belief that is
“de re” in the sense that it’s about actual-world friends-of-mine, yet “non-specific”
in the sense that it isn’t about any particular individual. Fodor’s non-specific de
re readings can be captured via structures involving long-distance binding of the
predicate’s world-pronoun:

41
(4.7) De re, Non-specific:
≈ there is some group of individuals who are friends of mine (say, the Sharks)
such that Alice thinks some or other of them won
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a
Alice

⟨2,a⟩ CP′

thinks t1s
⟨2,s⟩ S′′

that t2a
DP
won t2s
a
FoM
w1 g1

Like other accounts with covert world-variables, the present account has a resource
for capturing Fodor’s “third reading” in terms of long-distance/local binding. What
is particular to the present account is how the binding is implemented — via coin-
dexing involving the world-variable’s sister assignment-variable — and what gives
rise to the potential, or lack thereof, for the alternative readings with different pred-
icate positions. Intensionality is diagnosed as local context-sensitivity of embedded
world-variables, and it is captured via general mechanisms for capturing (possi-
bly obligatory) local interpretation. Distinctions among readings are diagnosed in
terms of movement and the general phenomenon of optional local/global readings
of pronouns, here world-pronouns. Constraints on possible readings are explained
in terms of the treatment of pronouns vs. traces.
4.3 Quantified modal attitude ascription
This section applies the proposed syntax/semantics to a more complex example such
as (4.8) with a quantified modal ascription. Working through this example will help

42
illustrate a range of features of the account, as concerning free/bound pronouns,
modality, and local/global readings with context-sensitive expressions. (Hereafter,
to improve readability I omit explicit reference to the quantification over w ∈ Ds
when the modal domain has been derived. As above I suppress certain irrelevant
assignment modifications, indicated with ‘≈’. Certain equivalences from assignment
modification are highlighted for comment below. I use ‘r’ for type ⟨s, at⟩.)3⁰

(4.8) Everything thinks that it might love something.

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a
[everything]⟨1,e⟩

t1e VP

⟨2,a⟩ CP′

thinks t1s

⟨2,s⟩ S′′

that t2a
⟨3,a⟩ CP′′

might t2s r1 g2
⟨3,s⟩ S′′′

Cdec t3a
[something]⟨2,e⟩

o1 g1 t2e
loves t3s

3⁰I consider an alternative argument structure for the modal in §7.6. For space purposes further
intermediate calculations must again be left to the reader:
JS′′′ K = λgg .∃x∶ g[x(g)/2e](1a)(1e) loves x(g) in g(3s)
JCP′′ K = J[that t3a ]⟨3,s⟩ K(JS′′′ K)
= λgg . ∃x∶ g[w(g)/3s][x(g)/2e](1a)(1e) loves x(g[w(g)/3s]) in @(g(3a))
J[r1 g2 ]K = Jr1 K(Jg2 K)
= [λa′a .[λws .[λaa .[λgg . a′ (g)(1r)(w(g))(a(g))]]]](λgg .g(2a))
= [λws .[λaa .[λgg . g(2a)(1r)(w(g))(a(g))]]]
JS′′ K = J[[might t2s ][r1 g2 ]]⟨3,a⟩ K(JCP′′ K)

43
 JSK ≈ λgg . for a′′ = λgg .g− ,
∀y∶ ∀aa s.t. a(g) is compatible with SOMy(g), @(a′′ (g)) ,
∃a′a s.t. a(g)(1r)(@(a(g)))(a′ (g)),
∃x∶ g[a′′ (g)/1a][y(g)/1e](1a)(1e) loves x(g) in @(a′ (g))
= λ gg . . . . ∃x∶ g[a′′ (g[y(g)/1e])/1a](1a)(1e) loves x(g) in @(a′ (g))
= λ gg . . . . ∃x∶ a′′ (g[y(g)/1e])(1e) loves x(g) in @(a′ (g))
= λ gg . . . . ∃x∶ g[y(g)/1e]− (1e) loves x(g) in @(a′ (g))
= λgg . ∀y∶ ∀aa s.t. a(g) is compatible with SOMy(g), @(g− ) ,
∃a′a s.t. a(g)(1r)(@(a(g)))(a′ (g)), ∃x∶ y(g) loves x(g) in @(a′ (g))
Roughly put: (4.8) is true iff for every individual o, for every possibility h compatible
with o’s beliefs, there is some possibility h′ accessible from h such that there is some
individual o′ whom o loves in the world of h′ .
4.3.1 Standardizing quantification. Binding with pronouns and traces
As discussed previously, bound readings of pronouns, intensionality, shifting under
modals, and context-sensitivity are captured via uniform syntactic/semantic mech-
anisms introducing quantification over individuals, worlds, and assignments.
Obligatory binding relationships may be established by (type-driven) movement.
Movement generates a cross-categorial binder-index attaching to moved expres-
sions, combining via function application. Notably, the complementizer moves from
the main predicate’s world-argument position, leaving a coindexed world-trace, fol-
lowed by the modal element’s movement from the complementizer’s assignment-
argument position, leaving a coindexed assignment-trace. This coindexing-via-movement
generates the local reading of each clause’s main predicate: the “loving” occurs in
worlds u′ (=@(a′ (g))) compatible with the relevant information; the relevant in-
formation is determined relative to the worlds u (=@(a(g))) compatible with the
≈ λgg . ∃aa s.t. g(2a)(1r)(g(2s))(a(g)), ∃x∶ g(1a)(1e) loves x(g) in @(a(g))
JVPK = J[think t1s ]⟨2,a⟩ K(JCP′ K)
≈ λye .λgg . ∀aa s.t. a(g) is compatible with SOMy(g), g(1s) ,
∃a′a s.t. a(g)(1r)(@(a(g)))(a′ (g)), ∃x s.t. g(1a)(1e) loves x(g) in @(a′ (g))
JS′ K = J[everything]⟨1,e⟩ K(JVPK(λgg .g(1e)))
≈ λgg . ∀y∶ ∀aa s.t. a(g) is compatible with SOMy(g), g(1s) ,
∃a′a s.t. a(g)(1r)(@(a(g)))(a′ (g)), ∃x∶ g[y(g)/1e](1a)(1e) loves x(g) in @(a′ (g))
JSK = JT⟨1,a⟩ K (J[that t1a ]⟨1,s⟩ K(JS′ K))

44
subject’s attitude state; the subject’s attitude state is assessed at the world u′′ (=@(g− ))
of the discourse context.
In contrast, the optional bound reading of ‘it’ is implemented via coindexing
configurations with the pronoun-complex [o1 g1 ]. The binder-index on everything⟨1,e⟩
says to interpret o1 with respect to the input assignment modified to take ⟨1, e⟩ to (in
this case) y(g). Although embedded under several assignment-shifters, the pronoun
can be linked to the main-clause quantifier via its assignment-variable g1 coindexed
with the topmost assignment-binder.
Analogous points hold with the modal’s epistemic modal-background pronoun
[r1 g2 ]. The variable r1 is interpreted with respect to (takes as argument) the as-
signment variable g2 coindexed with the assignment-binder attaching to the attitude
verb. Just as a concrete discourse may be represented via an abstract assignment that
assigns a set of epistemic possibilities to a certain syntactic index, say 1r, so too with
a subject’s concrete state of mind (§3.4). The shifted modal background a(g)(1r)
represents a doxastic counterpart of the (epistemic) modal background gc (1r) that
would be determined by the discourse, and the set of accessible possibilities is de-
termined by the possibilities a(g) compatible with the subject’s state of mind. The
generalized semantics for variables/traces (§2.1) derives how, although the lexical
entry for the modal specifies an argument of type ⟨s, at⟩ and the modal quantifies
over functions a′ ∶ Dg → G, the value returned by a(g)(1r) is a function from worlds
u ∈ W to (the characteristic function of) a set of assignments (possibilities) h′ ∈
G, and the quantified condition is a condition on a possibility h′ ∈ G in this set
a(g)(1r)(u) (see n. 30). (I return to issues about local/bound readings of modal
background variables below.)
4.3.2 Assignment modification and bound pronouns
Capturing the bound interpretation of ‘it’ raises interesting general issues about
assignment modification. The notion of a modified assignment is standardly intro-
duced by saying something to the effect that g[x/i] is the unique assignment which
is just like g except that i is mapped to x. A question rarely (if ever) addressed is how
to interpret expressions “g[...g.../i]” in our metalanguage, where the description
of what i gets mapped to uses the same letter as the letter used for the original
assignment (here ‘g’). This question becomes pressing in derivations involving re-
peated assignment modifications — in particular, when encountering assignment-
descriptions of the form “g[...g.../i][⋯].”
The final steps in (4.8) provide such a case when g[y(g)/1e] is to be modified
to g[a′′ (g)/1a][y(g)/1e]. Given the standard characterization of modified assign-
ments, g[y(g)/1e] is the assignment g′ that is just like g except that 1e is mapped

45
to y(g); this modified assignment is modified to the assignment g′′ that is just like
g′ except that 1a is mapped to a′′ (g). So, feeding 1a to g′′ would seem to return
a′′ (g). Since a′′ is (roughly) the identity function, a′′ (g) = g− , and so, it would seem,
g′′ (1a) = g− . The critical question is what this resulting assignment returns for 1e.
What we want isn’t whatever happens to be returned by the (counterpart in G of the)
original assignment g, i.e g− (1e), but what is returned by g as modified by the initial
modification, i.e. g[y(g)/1e](1e) = y(g). What we need — and as yet fail to have — is
a way of ensuring a sort of dynamic updating in repeated assignment modifications,
so that references to “g” in later modifications refer to the assignments as modified
in earlier steps.31
To capture this, I propose that we treat assignment modifiers as operators on
assignments, and repeated modifications as proceeding via function composition.

(4.9) Assignment modification

a. [z/iτ ] ∶= λgg . ιmg ∶ m(iτ ) = z ∧ m(jσ) = g(jσ), for all jσ ≠ iτ
b. [⋯][...] ∶= [⋯] ○ [...]
c. gg [⋯]1 . . . [⋯]n ∶= [⋯]1 ○ . . . ○ [⋯]n (g)

These definitions derive the crucial equivalence step in (4.8), reproduced in (4.10);
the modifier [a′′ (g)/1a] is correctly treated as mapping m = g[y(g)/1e] to an as-
signment m′ that is just like m except that it maps 1a to its image under a′′ .32

31The only place I’ve seen this issue addressed is Sternefeld 1998: 16–17. Sternefeld cheats in
the way mentioned in the main text.
32For readability I abbreviate the right conjunct in (4.9a) as m(jσ, ≠ iτ ) = g(jσ, ≠ iτ ); I use large
parentheses to enclose descriptions of assignments when prefixed to an argument, e.g. (ιg ∶ ...)(1s):
(i) [a′′ (g)/1a] = λgg . ιm∶ m(1a) = a′′ (g) ∧ m(jσ, ≠ 1a) = g(jσ, ≠ 1a)
[y(g)/1e] = λgg . ιm∶ m(1e) = y(g) ∧ m(jσ, ≠ 1e) = g(jσ, ≠ 1e)
g[a′′ (g)/1a][y(g)/1e]
= ([a′′ (g)/1a] ○ [y(g)/1e])(g)
= [λg′g . [a′′ (g)/1a]([y(g)/1e](g′ ))](g)
= [λg′g . [λg′′g . ιm∶ m(1a) = a′′ (g′′ ) ∧ m(jσ, ≠ 1a) = g′′ (jσ, ≠ 1a)]

(ιm′ ∶ m′ (1e) = y(g′ ) ∧ m′ (jσ, ≠ 1e) = g′ (jσ, ≠ 1e))](g)

= [λg′g . ιmg ∶ m(1a) = a′′ (ιm′ ∶ m′ (1e) = y(g′ ) ∧ m′ (jσ, ≠ 1e) = g′ (jσ, ≠ 1e))
∧ m(jσ, ≠ 1a) = (ιm′ ∶ m′ (1e) = y(g′ ) ∧ m′ (jσ, ≠ 1e) = g′ (jσ, ≠ 1e))(jσ, ≠ 1a)](g)
= ιmg ∶ m(1a) = a′′ (ιm′ ∶ m′ (1e) = y(g) ∧ m′ (jσ, ≠ 1e) = g(jσ, ≠ 1e))
∧ m(jσ, ≠ 1a) = (ιm′ ∶ m′ (1e) = y(g) ∧ m′ (jσ, ≠ 1e) = g(jσ, ≠ 1e))(jσ, ≠ 1a)

46
(4.10) g[a′′ (g)/1a][y(g)/1e](1a)(1e) = g[a′′ (g[y(g)/1e])/1a](1a)(1e)
g[...g...][⋯] = g[...g[⋯]...]

The remainder of the derivation proceeds straightforwardly: Given 1a, the resulting
assignment g[a′′ (g[y(g)/1e])/1a] returns a′′ (g[y(g)/1e]). Since a′′ = λgg .g− , this
reduces to g[y(g)/1e]− , which, given 1e, returns y(g), capturing the bound reading
of the pronoun by the quantifier, as desired.
4.3.3 Epistemic modals: Locality and binding
Paradigm context-sensitive expressions are at least optionally (if not obligatorily)
interpreted with respect to the context of utterance when embedded in attitude
ascriptions (§1). A principal challenge for contextualists has been to capture the con-
trasting behavior of epistemic modals, which seem obligatorily linked to the subject
(§1). Likewise with quantificational subjects, as in (4.8), seeming to reflect a kind of
binding. There is apparently no reading of (4.11) which ascribes to every contestant
o the belief that it’s compatible with Alice’s/Bert’s evidence that o is the winner.

(4.11) Alice: Bert thinks that [every contestant]i thinks shei mighti be the winner.
(cf. Stephenson 2007: ex. 5b)

Relativist/expressivist theories offer general shifting mechanisms to capture this (§1.1).

No rigorous compositional semantic account of the shifting and binding properties
of epistemic modals has been attempted by contextualists.
The syntax/semantics in this paper compositionally derives local and bound
readings of embedded epistemic modals, as in (4.8). Moreover it does so in a frame-
work which maintains the core contextualist idea of modeling the context-sensitivity
of recalcitrant expressions such as epistemic modals in the same kind of way as
the context-sensitivity of paradigm context-sensitive expressions such as pronouns,
namely via quantification/binding with assignment-variables.33 The account may
thus be of interest to theorists who are compelled by the thought that the inter-
pretation of (e.g.) epistemic modals depends, in some sense, on context, but have
reservations about innovations introduced by relativism/expressivism.
An assignment-variable-based account provides a framework for theorizing about
differences among expressions in tendencies for local vs. global readings (Silk 2016).
= [a′′ (ιh′ ∶ h′ (1e) = y(g) ∧ h′ (jσ, ≠ 1e) = g(jσ, ≠ 1e))/1a](g)
= [a′′ ([y(g)/1e](g))/1a](g)
= g[a′′ (g[y(g)/1e])/1a]
33There may of course be other differences among them (Tonhauser et al. 2013, Silk 2016).

47
For expressions permitting local and global readings, conversational explanations
may be given regarding the expressions’ tendencies for different readings. Such ex-
planations would be understood at the “presemantic” level of what LFs are (not) de-
termined by token utterances. For certain types of epistemic uses of modals a locality
principle might be given which excludes LFs in which the modal-background pro-
noun’s assignment-variable isn’t coindexed with the closest c-commanding assignment-
binder.3⁴ Such a principle would be no more ad hoc than a globality principle exclud-
ing LFs in which the relevant assignment-variable is bound by an element other than
T, such as for (say) English gendered pronouns or languages with conventionalized
referential pronouns, as in (4.12)–(4.13) (cf. Déchaine & Wiltschko’s (2002) class of
“pro-DPs”; see also §1).

(4.12) Zë’cy nnah Gye’eihlly nàiy me’s g-uhc-ëng.

thus say Mike yesterday teacher perf-be-3sg.prox
‘Mikei said yesterday that hek/∗ i was a teacher.’ (Lee 2003: ex. 13; Zapotec)
(4.13) Mékw’ ye swíyeqe kw’ákw’ets-et-es te stóles-s tú-tl’òlem.
every det.pl man looking det wife-3poss det-3pl
‘All meni are looking at theirk/∗ i wives.’
(Déchaine & Wiltschko 2002: ex. 10; Halkomelem (Salish))

Likewise, as discussed in §1, just as there may be expressions conventionally spec-

ified for local vs. global readings, common across various families of languages are
conventionalized anaphoric pronouns — pronouns which obligatorily receive their
interpretation from the linguistic context.
An elegant picture emerges regarding the landscape of (non-)conventionalized
readings: (i) classical indexicals with conventionalized global readings, lexically ex-
cluded from being bound by an element other than T; (ii) obligatorily shifty expres-
sions with conventionalized local readings, lexically required to be bound by the
nearest c-commanding assignment-binder; (iii) specialized anaphoric expressions
with conventionalized non-global readings, lexically excluded from being unbound
or bound by T;3⁵ and (iv) the majority of context-sensitive expressions, which are
conventionally compatible with both shifted and non-shifted readings. The pro-
3⁴See Silk 2016, 2017 on characterizing the relevant type of uses (dubbed “endorsing uses”) of
modals, and discussion of the putative conventionalization of the local readings.
3⁵Such a formulation would need to be refined for conventionalized anaphoric expressions
compatible with discourse anaphora; I continue to focus exclusively on intra-sentential binding
and anaphora. Conventionalized anaphoric pronouns requiring specifically local antecedents —
expressions obligatorily receiving local non-global readings — are possible as well, as with e.g. spe-
cialized anaphoric pronouns in correlatives (§1), or the Russian possessive svoj in (i).

48
posed assignment-variable framework affords a unified analysis of the context-sensitivity
of the different classes of expressions, along with a principled basis for distinguishing
and theoretically representing patterns of shifted/non-shifted readings.
4.4 Pierre and friends
The reader may have wondered how (/been grateful that) we have made it thus far
without mentioning Frege puzzles. A natural application of the assignment-variable
framework is to shifted interpretations of names in attitude ascriptions. I men-
tioned in §1 that ways of situating the present project are as formally implementing
Stalnaker’s (1988) informal “multiple context” approach to attitude ascriptions, and
as generalizing accounts which introduce syntactic variables for items determining
the interpretation of referential expressions. Swanson (2011) suggests extending
Stalnaker’s approach to the interpretation of the two occurrences of ‘London’ in
(4.14), where the proposition presupposed and not realized by “Puzzling Pierre”
isn’t necessarily true (Kripke 1979).

(4.14) Pierre doesn’t realize that Londonci is Londonck . (Swanson 2011: ex. 34)

Though Stalnaker and Swanson don’t offer specific formal implementations, recent
work provides various mechanisms, such as variables for “guises” or “concept gen-
erators,” for capturing how the interpretation of certain referential expressions can
shift in embedded clauses (cf. Percus & Sauerland 2003, Cumming 2008, San-
torio 2010, Ninan 2012, Charlow & Sharvit 2014). Such machinery could of
course be layered into the general assignment-variable framework developed thus
far. Since the terrain here is already well-trod, it may be more interesting to consider
an approach that hasn’t quite been considered as such. I leave it to the reader to make
relevant adjustments depending on their preferred broader views on the syntax and
semantics of names.
Consider a predicativist-style account in which a name ‘N’, qua lexical item,
denotes the property of being called N (cf. Geurts 1997, Elbourne 2005, Fara
2015).3⁶ In sentences such as (4.15) the name is in a predicate position, complement
to an overt determiner quantifier. In ordinary bare singular uses predicativists posit
that the name the complement of an implicit determiner, reflected in (4.16). In
(i) Vanjai znaet c̆to Voledjak ljubit svojuk/∗ i sestru.
‘Vanjai knows that Volodjak loves hisk/∗ i sister.’ (Rappaport 1986: ex. 13)
3⁶Cf. Beaver 2001, Cumming 2008 for analyses of names as definites or variables, respectively.
For treatments of names as constants, see Kripke 1980, Salmon 1986, Soames 2002. See Fara 2011
on the importance of the “called N” vs. “called ‘N’ ” distinction.

49
languages such as Italian the determiner may be more generally pronounced, as in
(4.17)–(4.18); deleting the overt D-element and leaving the name in its NP position
renders the sentence ungrammatical.3⁷

(4.15) a. Some Alfreds are crazy; some are sane. (Burge 1973: 429)
b. There are at least two Tylers with philosophy degrees from Princeton.
(Fara 2015: ex. 8)
(4.16) Alfred is crazy.
a. [ DP ∅D [ NP Alfred]] is crazy
b. ≈ “the relevant individual called Alfred is crazy”
(4.17) a. La Callas ha cantato.
the Callas(fem) has sang
‘Callas sang.’
b. *Callas ha cantato.
Callas(fem) has sang (Longobardi 1994: ex. 25; Italian)
(4.18) a. Il mio/∗ Mio Gianni ha finalmente telefonato.
the my John has finally called
‘My John has finally called.’
b. [ DP the [ AgrP my [ NP John]]] has finally called
(Longobardi 1994: ex. 28, Bernstein 2001: exs. 11-12; Italian)

It is common among predicativists to treat the implicit determiner in examples

such as (4.16) as a demonstrative (e.g. Burge 1973) or definite article (e.g. Elbourne
2005, Fara 2015). Since such moves are familiar, let’s consider a variant approach in
which names in bare singular uses are sister to an implicit choice-function pronoun,
as reflected in (4.20). (I use ‘cf ’/‘F’ for indices/variables of type ⟨et, e⟩ that are choice
functions. Recall ↓ which “lowers” an item in a domain to an item composed out
of associated elements of the model (§2). I use (e.g.) ‘Alfredu ’ for the characteristic
function of the set of individuals o ∈ E such that o is called Alfred in u.)

(4.19) F ∶ [E → T] → E is a choice function iff for all non-empty S ⊆ E, F(χS) ∈ S

(where χS is the characteristic function of S)
(4.20) [ DP F1 g1 [ NP Alfred w1 g1 ]]
JAlfredK = λws .λxe .λgg . x(g) is called Alfred in w(g)
JF1 g1 K = λPet .λgg . g(1a)(1cf)((↓ P)(g))
3⁷Apt: Burge 1973 / Longobardi 1994.

50
 JDPK = λgg . g(1a)(1cf)([λgg .the function f s.t. ∀ye ∶ f(y(g)) iff y(g) is called
Alfred in g(1a)(1s)](g))
≈ λgg . g(1a)(1cf)(Alfredg(1a)(1s) )

The implicit choice-function pronoun [F1 g1 ] denotes a function Fcf (=g(1a)(1cf))

which selects a particular individual o ∈ E from the set of individuals (=Alfredg(1a)(1s) )
denoted by ‘Alfred’ in the world in question.
Appeals to choice functions and choice-function pronouns are common in se-
mantics for specific indefinites and functional readings of questions (e.g., Chier-
chia 1993, Dayal 1996, Kratzer 1998b).3⁸ The structure in the choice-function
predicativist implementation affords resources for capturing various types of shifted
and unshifted readings with names. First, like other predicativist and variabilist
accounts, we capture the context-sensitivity of unembedded uses, as schematically
in (4.21).

(4.21) [Context: from a UK advertising campaign in which the character Freddo

is standing next to a Freddo cookie]
a. Freddoi met Freddok .
b. [ S T⟨1,a⟩ … [ DP F1 g1 [ NP Freddo w1 g1 ]] met-ts [ DP F2 g1 [ NP Freddo w1 g1 ]]]
c. JSK ≈ λgg . g− (1cf)(Freddog− (1s) ) met g− (2cf)(Freddog− (1s) ) in @(g− )

The first use of ‘Freddo’ is interpreted with respect to a choice-function pronoun

[F1 g1 ] which denotes a choice function F1 (=gc (1cf)) that selects the Freddo char-
acter o1 ∈ E from the set of individuals called Freddo (=Freddog− (1s) ); the second use
of ‘Freddo’ is interpreted with respect to a choice-function pronoun [F2 g1 ] which
denotes a choice function F2 (=gc (2cf)) that selects the particular Freddo cookie in
the ad o2 ∈ E from the same set. The selection in (4.21) is from individuals that
are called Freddo in the actual (evaluation) world. Every account must adopt some
metasemantic assumptions — general or specific to particular discourses — about
the intended interpretations of different syntactic indices and about the relations
among values assigned to shifted assignments (§3.4). In what follows I will retain
our metalanguage @ operator while also assuming that the first-positioned world 1s
represents the world of the possibility represented by the assignment, i.e. h(1s) =
@(h), for any h ∈ G.
Turning to attitude ascriptions, first consider (4.22), adapted from an example
from Jennifer Saul (1998: 366), on a global reading of ‘Bob Dylan’.
3⁸Indeed we will consider broader applications of choice functions in the treatments of relativiza-
tion, certain indefinites, conditionals, and wh interrogatives in §§6–8.

51
(4.22) [Context: We’re talking about people’s views on Bob Dylan’s singing abilities.
I know that Glenda, one of his childhood friends, knows him only under the
name ‘Robert Zimmerman’. Since you know him only under the name ‘Bob
Dylan’, I say:]
Glenda thinks Bob Dylan has a beautiful voice. (Saul 1998: cf. ex. 7)

I can felicitously use ‘Bob Dylan’ to characterize Glenda’s belief, even though she
wouldn’t “put it that way,” since what matters for our purposes is Glenda’s belief
about the individual whom we associate with ‘Bob Dylan’ (cf. Silk 2016: §§4.2, 4.4;
2017: §3.3). A simplified LF and derivation for the specific global reading of ‘Bob
Dylan’ in (4.22) is as follows (treat ‘Bob-Dylan’ and ‘sings-beautifully’ as unanalyzed
predicates).

(4.23) S

T⟨1,a⟩ CP

⟨1,s⟩ S

Cdec t1a
VP
she1 g1

⟨2,a⟩ CP1

thinks t1s
⟨2,s⟩ S

Cdec t2a

F1 g1
BD w1 -g2 sings-beautifully t2s
JCP1 K ≈ λgg . g(1a)(1cf)(BDg(1a)(1s) ) sings-beautifully in @(g(2a))
JSK ≈ λgg . for all a s.t. a(g) is compatible with SOMg− (1e),@(g− ) ,
g− (1cf)(BDg− (1s) ) sings-beautifully in @(a(g))

The choice-function pronoun [F1cf g1 ] is linked to the topmost assignment-binder

T⟨1,a⟩ , anchoring its interpretation to the discourse assignment gc . The contextually
determined value gc (1cf) selects the particular relevant individual o ∈ E called Bob

52
Dylan in the world of the discourse (gc (1s) = @(gc )). The attitude ascription is
true iff for every possibility h compatible with Glenda’s (=gc (1e)’s) state of mind,
the selected individual o = gc (1cf)(BDgc (1s) ) — the individual with whom we in the
discourse context associate the name — has a beautiful voice in @(h). Whether o is
called Bob Dylan in @(h) is irrelevant.
As those with anti-Millian intuitions may be keen to point out, (4.22) may also
have a reading on which it seems false. To bring out such a reading, consider the
alternative context in (4.24).

(4.24) [Context: Gwen attends an afternoon reunion where she hears her child-
hood friend going by the name ‘Robert Zimmerman’ sing a dedication song.
Feeling nostalgic she says ‘I love Robert; he always sang so beautifully’. Later
that evening she attends a concert of a man named ‘Bob Dylan’. She can’t
stand big public events and says ‘This is awful; this Bob Dylan guy has a ter-
rible voice’. Since Bob Dylan is excellent at keeping his star status concealed
to his childhood friends, Gwen doesn’t realize that the man singing at the
concert is her childhood friend who also sang at the reunion. We’re talk-
ing about how factors such as kinship, environment, etc. can affect people’s
perceptions; you say:]
a. Gwen thinks Robert Zimmerman has a beautiful voice.
b. Gwen thinks Bob Dylan has a terrible voice.

In such a context your uses of (4.24a)–(4.24b) may both have readings on which
they seem true. One way of representing such uses is as follows (mutatis mutandis
for (4.24a)):

53
(4.25) S

T⟨1,a⟩ CP

⟨1,s⟩ S

Cdec t1a
VP
she2 g1

⟨2,a⟩ CP1

thinks t1s
⟨2,s⟩ S

Cdec t2a

F2 g1 sings-terribly t2s
BD w1 -g2

JSK ≈ λgg . for all a s.t. a(g) is compatible with SOMg− (2e),@(g− ) ,
g− (2cf)(BDa(g)(1s) ) sings-terribly in @(a(g))

Each of Gwen’s (=gc (2e)’s) belief-worlds includes two epistemic counterparts of the
actual individual BD/RZ — a childhood friend going by the name ‘Robert Zimmer-
man’ who sings beautifully, and a distinct celebrity going by the name ‘Bob Dylan’
who sings terribly. In the context in (4.24) the contextually determined value for the
choice-function pronoun, here gc (2cf), selects for any world u (=a(g)(1s) = @(a(g))
compatible with Gwen’s beliefs and set of individuals S = {o′ ∶ o′ is called Bob Dylan in u},
the individual o who is the singer-counterpart in u of BD/RZ. The attitude ascription
is true insofar as every such selected individual o (=gc (2cf)(BDa(g)(1s) )) sings terribly
in every such world u.
The above assignment-variable analysis captures how the embedded choice-function
pronoun receives its interpretation from the discourse context, while the world-
pronoun of the name-qua-predicate receives a local reading shifted by the attitude
verb: The discourse context determines the basis for selecting the relevant individual
o ∈ S in each of Gwen’s belief-worlds u, while each such set S is determined on the
basis of Gwen’s beliefs. Whereas the contextually relevant choice function in the
context in (4.22) selects BD/RZ from any set including him, as represented in (4.23),

54
the contextually relevant choice function in the context in (4.24) selects, from any
set of Bob Dylans in Gwen’s belief-worlds, a particular epistemic counterpart o of
BD/RZ, namely the singer-counterpart. In each possibility compatible with Gwen’s
beliefs the selected individual is called Bob Dylan. In the terminology from §4.2
the use of the name is represented as having a specific de dicto reading — “specific”
in the sense that every such o is identical to one another (or at least metaphysical
counterparts of one another (Lewis 1986)), and “de dicto” in the sense that the
selected terrible singer is conceptualized as being called Bob Dylan. Indeed (4.24b)
could be used to characterize Gwen’s state of mind even if the “Bob Dylan” craze was
an elaborate hoax and there is in fact no such individual. Ordinary uses of names
may thus constitute a systematic case of the specific de dicto. Santorio 2013 argues
that capturing such readings in the case of indefinite DPs requires overhauling the
semantics for descriptions; some have even expressed doubt about the possibility of
specific de dicto readings (e.g., von Fintel & Heim 2011). No such innovations or
skepticism need be required. The readings in contexts such as (4.24) follow from an
off-the-shelf predicativist semantics with choice functions, implemented in a general
assignment-variable-based framework.
Finally, recall the “Puzzling Pierre” example in (4.14), in which two uses of
‘London’ receive different interpretations under the attitude verb. Consider (4.26)
(assume a simple is-of-identity semantics for ‘is’).

55
(4.26) ‘He thinks London is London.’
S

T⟨1,a⟩ CP

⟨1,s⟩ S

Cdec t1a
VP
he3 g1

⟨2,a⟩ CP1

thinks t1s
⟨2,s⟩ S

Cdec t2a

F3 -g2 London-w1 -g2 is t2s

F3 -g1 London-w1 -g1
JSK ≈ λgg . for all a s.t. a(g) is compatible with SOMg− (3e),@(g− ) ,
a(g)(3cf)(Londona(g)(1s) ) = g− (3cf)(Londong− (1s) ) in @(a(g))

The LF in (4.26) represents the first occurrence of ‘London’ as receiving a global

reading, picking out the actual city of London — though perhaps not so-called in
Pierre’s belief-worlds — and the second occurrence as receiving a local reading, pick-
ing out the city called London as he conceives it. Pierre’s (=gc (3e)’s) state of mind
is such that each of his belief-worlds u = @(h) (=a(g)(1s)) includes a city named
‘Londres’ which is pretty, and another dreary city called London where he lives. The
contextually determined value for the choice-function pronoun, here F = gc (3cf),
selects (say) the present capital city of England from any set which includes it; hence
F(Londongc (1s) ) = o1 = London. The shifted value for the pronoun, F′ = h(3cf),
represents an epistemic counterpart of this function (§4.3); it selects (e.g.) the dreary
city o2 in u where he lives from any set of things called London in his belief-worlds
u. The attitude ascription in (4.26) is true iff for every possibility h compatible
with his state of mind, the city o1 = F(London@(gc ) ) is identical to the city o2 =
F′ (London@(h) ). Though F(London@(gc ) ) = F′ (London@(gc ) ) = London — the ac-
tual city of London is identical to the city so-called by Pierre — the belief ascribed in
(4.26) isn’t necessarily true. Indeed since the selected objects are distinct in Pierre’s

56
belief-worlds — o1 is named ‘Londres’ and o2 is named ‘London’, among other things —
(4.26) is correctly derived as false.
The assignment-variable account outlined in this section maintains features mo-
tivating various types of semantics for names. First, with predicativists, we give a
uniform semantics for names qua lexical items in predicative and bare singular uses.
In both environments the name ‘N’ denotes the property of being called N.
Second, like many non-Millian accounts, the account captures the context-sensitivity
of uses of names without treating names as systematically ambiguous. The uses of
the string ‘F-r-e-d-d-o’ in (4.21) are uses of the same lexical item. The different
interpretations are derived from the contextually determined values for the choice-
function pronouns; the first use of ‘Freddo’ is interpreted with respect to a choice
function selecting Freddo the character, the second use with respect to a choice
function selecting Freddo the cookie.
Third, the semantics is compatible with certain uses of names being “rigid desig-
nators,” in the sense (roughly) of being used to designate the same individual across
worlds. For instance, in a bare singular use of ‘N’ context may determine a value for
the choice-function pronoun, F, that is defined only for sets including a particular
individual o, and selects o given every such set:

(4.27) A choice-function F ∶ [E → T] → E is rigid iff for some o ∈ E and every

non-empty S ⊆ E∶ F(S) = o if o ∈ S, else undefined

However, rigidity isn’t encoded in the conventional meaning of names themselves.

This is for the better in light of descriptive uses such as (e.g.) (4.28) (cf. Cumming
2008).

(4.28) [Context: Bert has been receiving daily unsigned letters ostensibly from a
secret admirer. Bert doesn’t know who they’re from, and even wonders
whether they’re all from the same individual. We, who deliver the letters,
deliver them without the envelopes, which we see are in each case addressed
from ‘Ernie’. We see Bert walking excitedly to the mailbox; you say:]
Bert thinks Ernie sent him another letter.

Your use of ‘Ernie’ in (4.28) is appropriate even if (it’s presupposed that) it’s not the
case that there is some particular individual o ∈ E who sends Bert a letter across
Bert’s belief-worlds.
With variabilist accounts, we capture various types of “shifted” readings in em-
bedded contexts. Classic puzzles of names in attitude ascriptions are assimilated to
phenomena with local/global readings of context-sensitive expressions. The range

57
of readings are derived from independently motivated resources from the litera-
ture — choice-function pronouns and predicativism — and our general assignment-
variable syntax and semantics. Additional mechanisms specific to names or intu-
itively referential expressions (e.g., substitutional quantification, guises, etc.) aren’t
required.3⁹
Our assignment-variable-based framework formally implements the Stalnake-
rian (1988) idea that “multiple contexts” can be available for interpretation within
a single embedded clause (§1). Devices such as names can be used both to charac-
terize an individual’s “psychological semantic representation of a word” (Partee
1979: 11), and also “to pick out individuals in the basic context, and… express
propositions that alter the derived context” representing a subject’s state of mind
(Stalnaker 1988: 158–159). These dual functions, conspiring in classic puzzle
cases such as in (4.22)–(4.25), come together vividly in “Puzzling Pierre” examples
such as (4.14)/(4.26). It isn’t evident how to capture such examples in an operator-
based variabilist semantics, where names are analyzed as simple variables (e.g. Cum-
ming 2008), without treating the names as ambiguous. By contrast, the lexical item
‘London’ in (4.26) is treated as having the same semantic value in both occurrences.
The spectrum of shifted/unshifted readings are derived via interactions between how
individuals are selected (local/global readings of the choice-function pronoun) and
which individuals are candidates for selection (local/global readings of the name-
qua-predicate’s world pronoun).
Many variabilist and predicativist accounts of names are motivated by appeal-
ing to various types of non-purely-referential uses. Any overall account must also
be able to say something about the apparent differences in shiftability of names
in comparison with other context-sensitive expressions such as pronouns, definite
descriptions, etc., as in e.g. (4.29)–(4.30) (cf. Schlenker 2005).

(4.29) a. Every professor invited Alice to class.

≠ “every professor x invited the relevant individual x knows called Al-
ice to x’s class”
b. Every professor made the final exam as hard as possible.
≈ “every professor x made the final exam for x’s class as hard as possible”
c. The contestant answered every question before the question was finished.
≈ “the contestant answered every question x before x was finished”
d. Every guitarist thinks he plays the fastest.
3⁹Contrast Geurts’s (1997: 343) claim that “it is impossible to make substantial claims about the
semantics of names without making quite fundamental assumptions about the status and aims of
semantic theory.”

58
 ≈ “every musician x thinks x plays the fastest”
(4.30) a. Only Alfred did Alfred’s homework.
≠ “only Alfred = o is s.t. [λx.x did x’s homework](o) = 1”
b. Only you did your homework.
≈ “only you = o are s.t. [λx.x did x’s homework](o) = 1”

This isn’t the place to provide a general binding theory for names. For present
purposes suffice it to say that binding phenomena with names are far from straight-
forward. For instance, unlike languages such as English, copy reflexive languages
readily use names with bound-variable readings, as in (4.31)–(4.32) (see also Lasnik
1989, Mortensen 2003, Boeckx et al. 2007).

(4.31) Johni koonnuat khong Johni lae Peterk ko muankan.

John shave of John and Peter the same
‘Johni shaved himselfi , and Peterk did too (shave himselfk ).’ (Lee 2003;
Thai)
(4.32) R-ralloh Gye’eihllyi r-yu’lààa’z-ënn Gye’eihllyi chiru’ zë’cy cahgza’ Li’ebk .
think Mike like.1pl Mike also likewise Felipe
‘Mikei thinks we like himi , and so does Felipek (think we like himk ).’
(Lee 2003; SLQZ)

Even in English bound-variable readings may be possible in certain contexts. After

a surge of presidential apery one might say (4.33).

(4.33) ?Every Donald thinks Donald is the best.

≈ “every person x called Donald thinks x is the best”

Suppose then that in response to a wave of familial scorn ((4.34)), several Donalds
change their ways and begin telling others they are called Ronald. Characterizing
their families’ and partners’ states of mind, one might say (4.34), and later (4.35).

(4.34) ?Every family with a Donald thinks Donald should grow up.
(4.35) ?Every partner of a Donald thinks Donald is called Ronald.
≈ “every x s.t. x is the partner of some y called Donald thinks y is called
Ronald”

It isn’t immediately evident how such examples, if available, would be captured in

59
existing variabilist or predicativist accounts.⁴⁰ The extra structure afforded by the
choice-function implementation in this section would provide a natural account; a
first-pass LF for (4.33) may be as follows (assuming that in the intended interpreta-
tion the index 2cf represents roughly a “bestness” choice function):

(4.36) [ S T⟨1,a⟩ … [[ DP every Donald-w1 -g1 ] [[thinks t1s ]⟨2,a⟩ … [ IP [ DP F2 -g2 Donald-
w1 -g1/2 ] …]]]] (preliminary)

The bound shifted reading of the embedded ‘Donald’ in (4.36) is represented pre-
cisely analogously to the bound shifted reading of the embedded epistemic ‘might’
in (4.8). I leave further refinements in light of binding phenomena with names
crosslinguistically, and comparisons with alternative (assignment-variable-based)
variabilist and predicativist analyses, for future work.⁴1 (Hereafter I put the compli-
cations in this section with names aside; unless otherwise noted I will treat names
as constants.)

5 Recap. Next steps

Let’s take stock. Independent linguistic phenomena have led various theorists to
introduce assignments into the model, and to posit variables in the syntax for (e.g.)
worlds, times, and elements interpreting referential expressions. This paper begins
developing a linguistic theory which posits object-language variables for assignment
functions — variables for the sort of item responsible for interpreting quantifiers
and context-sensitive language generally — and treats compositional semantic val-
ues systematically in terms of sets of assignments. Principal features of the account
are that it standardizes quantification across domains (e.g. individuals, worlds, as-
signments); and it systematizes a range of linguistic shifting phenomena, as with
quantifiers, intensionality, and local/global readings of context-sensitive expressions.
A particular version of an assignment-variable-based account has been devel-
oped. The syntax and lexical/compositional semantics delineate the sources of in-
tensionality and assignment-shifting: world-quantification/binding arises from the
complementizer, which moves from the world-argument position of the clause’s
main predicate; assignment-quantification/binding arises from modals (broadly con-
⁴⁰Examples combining predicative and bare singular uses of names would raise challenges for
variabilist accounts; examples such as (4.35) would raise a distinctive challenge for predicativist
semantics analyzing names along the lines of “being called N.”
⁴1If the reader finds (4.34)–(4.35) to bear an uncanny resemblance to donkey sentences: Indeed.
A revised account is given below. I reconsider issues with pronoun binding and donkey anaphora in
§§6.2–6.3 (also §§7.7, 8.3).

60
strued), which move from the assignment-argument position of the C head. Bind-
ing with individuals/worlds/assignments is derived uniformly from a generalized
binder-index resulting from type-driven movement. This binder-index attaches di-
rectly to moved expressions. The account avoids quantification-specific compo-
sition rules or added parameters of interpretation. A distinction between trace-
binding and pronoun-binding — something arguably desirable for independent rea-
sons (Büring 2004, 2005) — falls out directly (more on which in §6.3). An improved
formalization of assignment modification was provided, which helps capture bind-
ing relations in examples with repeated modifications.
Philosophically, the account can be understood as providing a precise formal im-
plementation of Stalnaker’s “multiple context” approach to attitude ascriptions. The
syntax/semantics affords a unified analysis of the context-sensitivity of pronouns,
epistemic modals, etc., in the spirit of contextualist theories. Yet it improves in
compositionally deriving certain distinctive shifting phenomena (e.g. with epistemic
modals), and providing a framework for theorizing about expressions’ different ten-
dencies for local/global readings. Further (grammatical, lexical, metasemantic, con-
versational) constraints on readings call for more thorough investigation.
§§2–4 focused on applying the assignment-variable-based framework and par-
ticular treatment of the syntax/semantics interface to certain phenomena with quan-
tifiers, attitude verbs, and modal verbs. The remainder of the paper begins to exam-
ine how the account may be extended to other types of constructions. I focus on
phenomena with local/global readings in questions and conditionals.
Extending an assignment-variable-based account to a particular expression or
construction isn’t as straightforward as taking one’s favorite style of analysis and
adding assignment-variables to interpret any other variables or context-sensitive
elements. The treatments of intensionality and modals in §3 relied on particular
assumptions about the syntax and semantics to motivate a basis for introducing
the relevant world- and assignment-binders, traces, and variables. Whatever style
of analysis one assumes for a given further expression, one needs to ensure that
any binder indices and sources of shifting phenomena can be derived from fea-
tures of the syntax/semantics that are independently attested and continuous with
the theory developed thus far — e.g., base-generating complementizers in the world
argument position of the clause’s main predicate, and base-generating assignment-
shifters, such as certain semantically modal expressions, in a relevant assignment-
argument position. §6 draws on prominent head-raising and D-complement anal-
yses of relative clauses to develop an assignment-variable approach to relativization
and donkey anaphora, with determiner quantifiers now introducing quantification

61
over assignments. §7 turns to local/global readings in conditionals, drawing on
developments on ‘if ’-clauses as free relatives, now construed as definite descriptions
of possibilities, i.e. assignments. The approach to pronominal anaphora from §6
is generalized to proforms in correlatives and ‘then’ conditionals. §8 examines lo-
cal/global readings in interrogative sentences, drawing on developments from Heim
of an approach to questions as sets of possible answers, with answers now con-
strued as sets of assignments. The proposed syntax/semantics afford uniform analy-
ses of wh-words, indefinites, and relative determiners as choice-function pronouns,
and of interrogative sentences, ‘if ’-clauses, and non-modal correlative clauses; and
they capture a spectrum of shifting phenomena with ‘if ’-clauses in sentence-initial,
sentence-internal, and sentence-final positions (adjoined to NP/VP/IP/CP), with
individual correlative clauses adjoined to DP/IP/CP, and with conditionals involving
modalized/non-modalized and declarative/interrogative main clauses.

6 Relative clauses. Quantification and binding with determiners

The following sections examine applications of the assignment-variable framework
to various types of non-declarative clauses and linguistic anaphora. I begin with
relative clauses. §6.1 motivates a treatment of determiner quantifiers in headed rel-
ative clauses as quantifying over assignments. §6.2 applies the treatment of relative
clauses from §6.1 to several types of donkey anaphora. §6.3 considers several ways
of generalizing the syntax/semantics from §§6.1–6.2 to other types of quantifica-
tional DPs, restrictive modification, and pronoun binding. Features of the proposed
assignment-variable-based account include:

• The syntax/semantics of relative clauses derives individual- and assignment-

binders from independently motivated D-complement and raising analyses.
• The semantics is fully compositional: the account doesn’t require additional
composition rules such as Predicate Abstraction, Predicate Modification, or
Hamblin Function Application, and it avoids introducing independent prin-
ciples for interpreting reconstructed phrases or traces vs. pronouns; a limited
role for function composition is briefly considered.
• The semantics of determiner quantifiers and treatment of donkey pronouns as
copies of their linguistic antecedents allow for existential readings of donkey
sentences, capture specific and nonspecific readings of donkey pronouns in
intensional contexts, and avoid the proportion problem.
• The generalized syntax/semantics for quantificational DPs captures connec-
tions with notions such as “specificity” (“D(iscourse)-linking”); it yields a uni-

62
 fied treatment of several types of restrictive modification, with applications
to “free R” (modifier) and “inherent R” (argument) readings of prenominal
and postnominal genitives; and it extends the account of donkey anaphora to
apparent binding out of DPs with inverse linking and genitive binding, with
speculative applications to weak crossover.

I focus on headed restrictive relative clauses such as (6.1), in which a nominal (‘baby’)
is modified via a (possibly implicit) relative pronoun/determiner (‘which’) or com-
plementizer (‘that’). I return to free relatives and correlatives in §7.7.1.

(6.1) a. Every [baby which __ laughed] is cute.

b. Every [baby that Alice likes __] is cute.

(I use ‘relative clause’ for expressions such as the bracketed material, and ‘relative
phrase’ for the combination of the relative word and a nominal, e.g. ‘which baby’. My
usage doesn’t presuppose particular views on the syntactic category or semantic type
of the matrix determiner’s restrictor argument, the relation between relative words
and interrogative wh words, the semantic type of relative words such as ‘which’, or
the syntactic presence and semantic type of the intuitive “gap” position. We will
address these issues in due course.)
6.1 Relative clauses
6.1.1 Syntax: Head raising + D-complement
It is standard following Quine 1960 to treat restrictive relatives as supplying an
additional restriction to the domain of the matrix determiner — e.g., treating ‘baby
which laughed’ in (6.1a) as restricting the domain of ‘every’ to the set of babies o
such that o laughed. How to derive this intuitive interpretation in the syntax and
compositional semantics is controversial. A familiar idea is that the relative word
triggers Predicate Abstraction, and the head NP and relative clause CP combine by
Predicate Modification (Heim & Kratzer 1998):

(6.2) a. [NP baby [CPrel whichi ti laughed]]

b. JCPrel K ≈ {o ∶ o laughed}
JNPK ≈ {o ∶ o is a baby} ∩ {o ∶ o laughed} = {o ∶ o is a baby ∧ o laughed}

The syntax/semantics of relative clauses might necessitate additional composition

rules such as Predicate Abstraction. Yet assuming such rules would be theoretically
awkward at this stage, given the emphasis throughout the paper on avoiding syncat-
egorematic treatments of binding and quantification. I will suggest that alternative

63
head raising analyses provide a more attractive syntactic basis for developing an
assignment-variable-based account of relative clauses.
It isn’t uncommon in semantics to assume a “head external” syntax for relative
clauses in which the nominal head is generated external to the relative clause CP, as
e.g. in (6.3) (Montague 1970, Heim & Kratzer 1998).

(6.3) “Head-external” syntax

a. [ DP D [ NP NP [ CPrel reli [ C Crel [ IP … ti … ]]]]]
b. ⇒ [ DP every [ NP baby [ CPrel whichi [ C Crel [ IP ti laughed]]]]]

More common in contemporary syntax is to treat the head NP as having a represen-

tation internal to the relative clause. One compelling source of evidence comes from
languages with circumnominal relatives — relative constructions which overtly re-
alize the sort of LF proposed by raising analyses, where the head NP is pronounced
inside the relative clause (de Vries 2002, Hiraiwa 2017). Theories differ on what
syntactic mechanism triggers the movement of the head NP (e.g. case, agreement),
and what position the NP occupies at LF. What is important in what follows is simply
that the nominal head can be interpreted internal to the relative clause at LF. To fix
ideas I follow head-raising analyses in treating the head NP as base-generated inside
the relative clause CP, and I assume that the relative clause CP is the complement
of the matrix determiner (cf. Kayne 1994, Bianchi 1999, Bhatt 2002, de Vries
2002, Donati & Cecchetto 2015, Hiraiwa 2017), as reflected schematically in
(6.4). (More on the specific position of the relative phrase shortly.)⁴2

(6.4) “Head-raising” syntax

a. [ DP D [ CPrel … [ DPrel rel NP] … ]]
b. ⇒ [ DP every [babyi [ CPrel … [ DPrel which ti ] … ]]

Treating the head NP as interpreted internal to the relative clause raises prima fa-
cie challenges for the compositional semantics. Consider the DP ‘every baby which
⁴2An alternative “matching” analysis treats the head NP as having distinct representations internal
and external to the relative CP, unrelated by movement (see Sauerland 2003, Cinque 2013, 2015). I
ignore potential syntactic differences between postnominal relatives and left-headed circumnominal
relatives; and between D+CP analyses in which the head NP is in the specifier of the CP, and D+XP
analyses in which the head NP undergoes additional raising to a nominal projection which takes
the relative CP as its complement (see Bhatt 2002, de Vries 2002, Donati & Cecchetto 2015,
Hiraiwa 2017; see also n. 66, §6.3.1). That the relative CP is a complement rather than an adjunct is
accepted by theorists in head-external and head-internal camps (see also Partee 1975, Fabb 1990).
For discussion of the syntax of relative constructions crosslinguistically, see Chomsky 1977, Dayal
1996, de Vries 2002.

64
laughed’ in (6.1a). The predicate ‘laughed’ requires a sister of type e or ⟨et, t⟩, yet
‘baby’ is type ⟨e, t⟩; hence simply reconstructing the head NP to the gap position
in the IP would create a type mismatch. Reconstructing ‘which’ along with ‘baby’
could yield an argument of individual/generalized quantifier type to combine with
‘laughed’, given some suitable semantics for ‘which’. However, proceeding in this
way seems to predict that the relative clause is sentence type, although the matrix
determiner ‘every’ presumably requires an argument of type ⟨e, t⟩. The composi-
tional challenge is to capture both (i) that the IP-internal predicate (‘laughed’) can
combine with whatever fills the gap position, e.g. yielding a type t denotation for the
IP, and (ii) that the relative clause CP is predicate type so that it can combine with
the matrix determiner (‘every’).
6.1.2 Syntax/semantics: Trace Conversion?
The principle attempt to address this compositional semantic challenge for head-
internal analyses comes from Bhatt 2002, which resorts to non-compositional op-
erations for interpreting traces (qua copies of movement; n. 13) and effects of re-
construction (for parallel moves see Elbourne 2005, Moulton 2015). On Bhatt’s
analysis, the relative phrase ‘rel NP’ reconstructs to to its lowest position in the IP,
and a mechanism of Trace Conversion from Fox (2000, 2002) converts the relative
phrase to a definite description ≈ “the NP identical to xi .” The variable xi is bound
by a binder-index in the position of the highest copy of the relative phrase, where
the binder-index is assumed to remain above the IP even though the relative phrase,
whose movement triggered the insertion of the binder, is deleted. Formally, the
derivation proceeds from a full chain such as (6.5a) to (6.5b), in which deleting the
non-lowest copies is assumed to leave a binder index in the position of the highest
copy, and then from (6.5b) to (6.5c) by Trace Conversion, in which the relative
phrase is replaced by a variable-bound definite description.

(6.5) Bhatt 2002: Head-raising syntax with Trace Conversion

a. every [[rel baby] Alice thinks [[rel baby] that Bert likes [rel baby]]]
b. every λx [Alice thinks [that Bert likes [rel baby]]]
c. every λx [Alice thinks [that Bert likes [the baby identical to x]]]
(adapted from Bhatt 2002: exs. 35–38)

Bhatt’s notation conceals various non-compositional operations. In order to

generate the λ-binder above the IP Bhatt “assume[s] that when a copy is deleted,
the λ abstraction created as part of the movement is retained” (2002: 64). However,
such an assumption, consistently applied, would require retaining the λ-binder over

65
the deleted intermediate copy as well, reflected in (6.6). (I use the numerical indices
to highlight the assumption that the semantic λ-abstraction is created by an inde-
pendent Predicate Abstraction rule. Assume for the moment, following Heim &
Kratzer 1998, that the binder indices are introduced in the movement operation.
The strikethrough here indicates syntactic deletion. The steps in (6.6b)–(6.6c) indi-
cate the dual operations of Quantifier Replacement + Variable Insertion for deriving
the definite description in the gap position. The nodes for ‘baby’ and ‘=o1 ’ combine
by Predicate Modification.)

(6.6) a. every [[rel baby]1 1 Alice thinks [[rel baby]1 1 that Bert likes [rel baby]1 ]]
b. every [1 Alice thinks [1 that Bert likes [rel baby]1 ]]
c. every [1 Alice thinks [1 that Bert likes [the [[baby] [=o1 ]]]]]

The LF in (6.6c) is uninterpretable. The complement of ‘think’ is represented as

denoting (roughly) [λxe .Bert likes the baby identical to x]. This isn’t what Lewis
1979 had in mind in treating properties as the objects of belief. Even if the type-
mismatch were resolved, note that the highest binder-index doesn’t (syntactically
or semantically) bind the variable in the position of the lowest copy. The predicted
denotation of the relative clause (complement of ‘every’) is a constant function.
One way of avoiding these problems would be to stipulate different effects at LF
of deleting copies in different positions — say, that deleting a copy involves deleting
the binder-index created by its movement unless the copy is the highest copy of a
relative phrase. Such a principle would have to be specific to certain constructions,
such as relative clauses; it isn’t in general the case that retaining the binder-index
of the highest copy in a chain leads to an interpretable structure. As Bhatt himself
argues in Bhatt & Pancheva 2006, drawing on Iatridou 1991, certain sentence-
initial ‘if ’-clauses must reconstruct — e.g., in (6.7) to a position under the attitude
verb, and in (6.8) to a sentence-final position under the quantifier, hence the Con-
dition C violations in (6.7b)/(6.8b).

(6.7) [If it rains]i , Bert thinks ti the party should be canceled.

a. [If it rains]i Bert thinks [if it rains]i the party should be canceled.
b. *[If Bertk is sick]i hek thinks [if Bertk is sick]i the party should be canceled.
(6.8) [If hisk teacher is out sick]i , every boyk will be happy ti .
a. [If hisk teacher is out sick]i every boyk will be happy [if hisk teacher is
out sick]i .
b. *[If Alicej gives hisk teacher food poisoning]i every boyk will be grateful
to herj [if Alicej gives hisk teacher food poisoning]i

66
It would be unattractive to posit that ‘if ’ conditionals are ambiguous depending on
whether the ‘if ’-clause is interpreted in its base position.

(6.9) a. Chloe said that if hisk teacher is out sick, every boyk will be happy.
b. Chloe said that every boyk will be happy if hisk teacher is out sick.

An apparently simple Copy operation in the syntax of movement yields “copy and
delete along with any binder indices, unless at the base, in which case reinterpret via
Trace Conversion, and unless a relative phrase with no higher copies, in which case
delete but retain the binder index” at the syntax/semantics interface.
One way of alleviating at least some of these distinctions would be to treat the
relative phrase as only partially reconstructing (cf. Chomsky 1995, Moulton 2015).
Compare (6.11) in which the intermediate copies are erased at LF (indicated by the
strikethrough), the relative word is interpreted in the highest copy, the (pied-piped)
complement NP is interpreted in the gap position, and the lowest copy is converted
such that ‘the’ is inserted into the empty D position and a variable coindexed with
the relative DP is inserted in the identity predicate sister to the NP.

(6.10) Trace Conversion (modified)

a. Binder Insertion: DPi […DPi …] ↝ DP i […DPi …]
b. Determiner Insertion/Replacement: [ DP (D) NP]i ↝ [ DP the NP]i
c. Variable Insertion: [ DP the NP]i ↝ [ DP the [NP =oi ]]
(6.11) a. every [[rel baby]1 Alice thinks [[rel baby]1 that Bert likes [rel baby]1 ]]
b. every [[rel baby] 1 Alice thinks [[rel baby] 1 that Bert likes [rel baby]1 ]]
c. every [[DP rel] 1 Alice thinks [that Bert likes [DP baby]1 ]]
d. every [[DP rel] 1 Alice thinks [that Bert likes [DP the baby]1 ]]
e. every [rel 1 Alice thinks that Bert likes [DP the [baby =o1 ]]]

Following Heim & Kratzer 1998 the binder-index triggers Predicate Abstraction,
which binds the coindexed variable inside the gap position, and the relative word
can be treated as denoting the identity function λPet .P. The relative clause denotes
(roughly) [λx . Alice thinks Bert likes the baby identical to x].
The LF in (6.11e) yields an intuitively correct denotation for the relative clause.
Consider the needed array of assumptions: (a) Deleted copies are syntactically rep-
resented and semantically interpreted differently in different positions — e.g., the
highest copy of ‘baby’ and the intermediate copy of the relative phrase are erased at
LF, whereas the deleted lowest copy of ‘which’ is replaced at LF with an interpreted
definite determiner. (b) Inserted sister to the NP in the lowest copy is an identity

67
predicate ‘= o1 ’, where the variable ‘o1 ’ internal to the gap-position DP is coindexed
with that very DP — i.e., with the DP originally represented with the relative phrase
[rel baby]1 , the movement of which created a coindexed λ-binder. (c) The inserted
identity predicate (type ⟨e, t⟩) combines with the nominal ‘baby’ (type ⟨e, t⟩) via a
rule of Predicate Modification. (d) The λ-abstraction at the top of the relative clause
is derived syncategorematically via an added Predicate Abstraction rule; the node
consisting of the binder-index isn’t given a denotation. Further, note that there is
no constituent for the relative phrase ‘which baby’ in the final representation of the
relative clause; ‘which baby’ doesn’t itself receive a semantic value.
The above operations formalize what needs to be explained — roughly put, how
the relative clause CP ‘wh-NP Crel IP’ comes to have a property-type ⟨e, t⟩ denota-
tion, where the gap element of the IP is interpreted as coreferential with the NP-
individual intuitively abstracted over. Other things equal it would be preferable to

68
provide a semantics for the head-raising analysis without needing to invoke addi-
tional (non-compositional) syntactic and semantic mechanisms such as these.⁴3,⁴⁴
The goal is to compositionally derive an interpretation of the relative clause
⁴3The above discussion followed Bhatt 2002 (following Fox 2000, 2002) in treating operations
such as Trace Conversion are syntactic rules. The points carry over straightforwardly to implemen-
tations of Trace Conversion as a semantic rule. Adapting a suggestion in Fox 2003, one might treat
the narrow syntax as delivering (i-a), which is converted to (i-b) by copy deletion, and interpreted
straightway via the sort of semantic rule in (ii). The interpretation of the sister of ‘every’ in (i-b) may
be derived roughly as in (iii) (again assuming scattered copy deletion, and letting the relative ‘wh’
denote the identity function).
(i) a. every [CPrel [rel baby]1 Alice thinks [[rel baby]1 that Bert likes [rel baby]1 ]]
b. every [CPrel [DP rel]1 Alice thinks that Bert likes [DP baby]1 ]
(ii) Trace Conversion (semantic)
Let α be a structure of the form: DPi [β …DPi …]. JαK = JDPK(λx.Jβx′ /i K), where
a. βx′ /i is the structure derived from replacing the (possibly empty) head of every i-indexed
constituent in β with thex , where
b. Jthex K = λPet .JtheK(λy.y = x ∧ P(y) = 1) (cf. Fox 2003: ex. 52)
(iii) JCPrel K = (λQet .Q)(λx.J[. . . Bert likes [DP baby]1 ]′x/1 K)
= λx . J. . . Bert likes [thex baby]K
= λx . . . . Bert likes JtheK(λy.y = x ∧ JbabyK(y) = 1)
≈ λx . Alice thinks Bert likes the unique baby identical to x
The formalization of Trace Conversion in (ii) simply encodes the series of syntactic operations
considered in the main text in a compositional semantic rule. The effect is no more explanatory
for purposes of compositional semantics.
⁴⁴Compare Safir 1999, which analyzes certain lower copies like pronouns, and Sportiche
2006, which analyzes the lower copies as demonstratives. Safir appeals to Fiengo & May’s (1994)
mechanism of “vehicle change” to motivate treating copies of names and definite descriptions as
“evaluated as pronouns with respect to interpretive principles” (Safir 1999: 587); however, crucial
to Safir’s general account is that quantified expressions, which include relative clause heads, “do not
permit their variables to undergo vehicle change” (1999: 615). Given their purposes, neither Safir
nor Sportiche provide derivations for their posited interpretations in the compositional semantics.
Compare also Caponigro 2003 on semantics for various types of free relatives. Caponigro treats
relative determiners in free relatives as raising from the IP gap position to SpecCP, leaving a λ-binder
in C; the [C λi IP] node is interpreted via a non-compositional abstraction rule. Caponigro is careful
to restrict his analysis to free relatives. It isn’t immediately evident how it would apply to headed
relatives (or to wh-interrogatives, though Caponigro assumes that free relatives are introduced
with wh-words). I am not aware of other analyses of QR/wh-movement which treat the λ-binder
as occupying a head position, such as that for the complementizer C. That aside, an analogous
Caponigro-style treatment of headed relatives would need to assume that in headed relatives with
an overt complementizer (e.g. ‘baby that Bert likes’), the complementizer gets fully deleted prior
to wh-movement. Yet such an assumption would raise issues for (wh) interrogative clauses, which
require the presence of the substantive interrogative complementizer throughout the derivation (e.g.,
to trigger wh movement and satisfy selection requirements of question-embedding expressions (§8)).

69
from the lexical semantics of the relative determiner and relative complementizer,
function application, and binding relations arising from type-driven movement. An
attractive strategy is to treat the syntax/semantics of DPs in headed relative construc-
tions parallel to the syntax/semantics of assignment-quantifiers such as modals from
§3. Indeed, parallels between nominals and clauses provided key motivations for
the DP-hypothesis in X′ -theory, and are well attested in diverse areas of syntax and
semantics. As noted in §3.2, DP-internal operator-movement from the complement
of D has independent precedents in syntactic work on quantificational/definite DPs
(Campbell 1996, Matthewson 1998, Lecarme 1999b, Ihsane & Puskás 2001,
Giusti 2002, Luján 2004). So, we can try saying, just as modal quantifiers raise for
type reasons from inside their clausal complement, i.e. from an internal argument of
the declarative/interrogative complementizer, determiner quantifiers raise for type
reasons from inside the relative clause complement in headed restrictive relative
clauses, i.e. from an internal argument of the relative complementizer (cf. Luján
2004). An assignment-variable-based syntax/semantics for relativization along the
lines pursued here may thus provide yet another instance of the systematic linguistic
parallels between individual and modal domains.⁴⁵
There are various ways of implementing this approach in the compositional se-
mantics. Choice points include (i) the position of the relative phrase at LF; (ii) the in-
teractions among the relative complementizer, relative phrase, and (possibly gappy)
IP in deriving a suitable argument for the matrix determiner; and (iii) the relation
between relative words and interrogative wh words. To fix ideas: (i) I assume that
the relative phrase originates in Spec of Crel (e.g. Sternefeld 2001). (ii) Just as the
declarative complementizer raises for type reasons from a world-argument position
inside the complement IP, a natural hypothesis is to treat the relative complemen-
tizer Crel as raising for type reasons from the individual-type gap position in the IP.
(iii) I suggest a unified analysis of relative pronouns, wh-words, and (at least some)
indefinites as choice-function pronouns.
⁴⁵For discussion with further examples of DP-internal movement, see Longobardi 1994, 2001,
Campbell 1996, Matthewson 1998, Bernstein 2001, 2008, Boeckx 2001, Ihsane & Puskás
2001, Giusti 2002, Luján 2004, Hiraiwa 2005, Laenzlinger 2005, Ilkhanipour 2016. For general
discussion on syntactic parallels in nominal/verbal extended projections and formal/interpretive
parallels across domains (individual, spatial, temporal, modal), see also Partee 1984, Fukui &
Speas 1986, Abney 1987, Szabolcsi 1987, 1989, Guéron & Hoekstra 1995, Lecarme 1996,
1999b, 2004, 2008, Siloni 1997, Stone 1997, Kratzer 1998a, Cardinaletti & Starke 1999,
Pesetsky & Torrego 2001, Bhat 2004, Grimshaw 2005, Rijkhoff 2008, Wiltschko 2014 (though
cf. Bruening 2009); see also n. 2.

70
 The core components of the account of (headed restrictive) relative clauses are
as follows:

• Head raising: the head NP is interpreted inside in the relative clause at LF,
specifically as sister to the relative determiner in SpecCPrel
• D-complement: the relative clause is the complement of the matrix determiner
• Syntax/semantics interface:
– the relative complementizer raises for type reasons from its IP comple-
ment, specifically from the (individual) gap position;
– the matrix determiner raises for type reasons from its CPrel complement,
specifically from an internal argument of the relative complementizer
• Semantics: relative determiners denote choice-function pronouns

A simplified LF is as follows (ignoring worlds and treating names as constants (§4.4);

I address the question marks shortly). (I will use ‘wh’ in trees for relative words. We
will consider interrogative wh words in §8.)

(6.12) Headed relative DP: D CPrel (preliminary)

a. ‘every baby which Alice likes’
b. DP

every⟨i,?⟩ CPrel

C
baby
wh g?
⟨j,e⟩ IP

Crel ti? Alice

likes tje
In headed relatives the matrix determiner raises from an internal argument of the
complementizer to head the DP and the overt relative phrase restricts the deter-
miner’s restrictor argument. Free (non-headed) relatives with distributions of DPs —
roughly, tensed free relatives in argument positions — may be treated analogously.
Adapting Comp accounts of free relatives (Groos & van Riemsdijk 1981, Grosu
1996, Caponigro 2002), the relative phrase is in SpecCP and what raises from C0rel

71
is the implicit definite determiner picking out the unique (possibly plural) individ-
ual satisfying the property denoted by the relative clause, as reflected in (6.13) (for
simplicity assuming ‘what’ ≈ ‘which relevant thing(s)’; cf. e.g. Dayal 1996: 196.)⁴⁶

(6.13) Free relative as DP: D∅ CPrel (preliminary)

a. ‘(Bert likes) what Alice likes’
b. DP

OP⟨i,?⟩ CPrel

C
wh g? P g
⟨j,e⟩ IP

Crel ti? Alice

likes tje

6.1.3 Semantics: Assignment-quantification with determiners

Principal issues for the lexical/compositional semantics concern the nature of the
quantification introduced by the matrix determiner, and binding relations among
the matrix determiner, relative complementizer, and choice-function pronoun ‘which’.
To motivate answers to these questions, I take a brief detour to examine indefinites.
Indefinites in embedded contexts raise notorious challenges for compositional
semantics. What is relevant here is simply that the interpretation of indefinites can
vary with a quantificational subject. First, certain indefinites can exhibit apparent
intermediate readings in embedded contexts — readings “intermediate” between or-
dinary nonspecific readings, as in (6.14), and specific readings about a particular
individual, as in (6.15), where the indefinite is specific relative to an attitude sub-
ject, supposition, or quantificational subject (Abusch 1994, Kratzer 1998b), as in
⁴⁶See Caponigro 2003 on non-DP-like free relatives, and van Riemsdijk 2006 on the syntax
of free relatives. Caponigro (2003) also treats DP-like standard free relatives as denoting a maximal
entity, but treats the free relatives as CPs with the covert pluralization operator in an adjoined position
(as opposed to treating the free relatives as DPs with the operator as a covert D head). The approach
in (6.13) could be adapted accordingly. As we will see, the analyses of relative and wh words proposed
below and in §8, respectively, maintain a uniform semantics for relative and wh words, and are thus
compatible with Caponigro’s (2003) proposal to identify interrogative wh-words and the relative
words in free relatives. Caponigro catalogues various respects in which free relatives pattern with
wh-interrogatives in contrast to headed relatives crosslinguistically.

72
(6.16)–(6.18). The intermediate reading of (6.17) says that for every baby o there is a
specific toy of mine that scared o, though which toy did the scaring may vary across
babies (for Joe it was the clown, for Annie the jack-in-the-box, etc.).

(6.14) Alice thinks a friend of mine died in the fire.

a. Nonspecific reading: ≈ Alice thinks I had some friend or other who died
in the fire
(6.15) If a friend of mine from Texas had died in the fire, I would have inherited a
fortune. (Fodor & Sag 1982: ex. 60)
a. Specific reading: ≈ there is some particular friend of mine, say Tex, such
that I would have inherited a fortune if he had died in the fire
(6.16) Bert might think some stalker is out to get him.
a. Intermediate reading: ≈ it’s possible that there is some particular stalker
o such that Bert thinks o is out to get him (might > indef > think)
(6.17) Every baby cried because a (certain) toy of mine scared them.
(every > indef > because)
(6.18) Every professor rewarded every student who read a/some book he had rec-
ommended. (every prof > indef > every student)
(Abusch 1994: ex. 10; Kratzer 1998b: ex. 16)

Second, in donkey sentences the interpretation of the pronoun varies as a function

of the indefinite and supposed circumstance or quantificational subject:

(6.19) a. If a farmer owns a donkey, he beats it.

b. Most farmers who own a shovel use it.

(6.19b) isn’t true simply if most farmer-shovel pairs ⟨x, y⟩ are such that x uses y, or if
most shovel-owning farmers use some stolen shovel or other; the truth of (6.19b)
requires that most farmers x use some shovel owned by x. The quantificational
force and content of the donkey pronoun varies with the subject and value for the
indefinite in the quantifier’s restriction.
The interpretation of certain indefinites and expressions linguistically depen-
dent on them can shift, not only in “shifty” contexts such as conditionals and atti-
tude ascriptions, but also under ordinary quantifiers. A hypothesis is that just as
modal quantifiers can shift the interpretations of pronouns, so too, at least in some
cases, with determiner quantifiers. This section develops a syntax/semantics of de-
terminer quantifiers as involving quantification over assignments. §§6.2–6.3 apply

73
the account to several types of pronominal anaphora. I leave applications to specific
indefinites and other types of linguistic and discourse anaphora for future work.
I suggest that we treat determiner quantifiers in headed relative clauses as quan-
tifying over assignments, and raising for type reasons from an internal assignment-
argument of the relative complementizer. Intuitively put, a DP ‘D wh-NP VP’ such
as ‘every baby which Alice likes’ quantifies over those individuals which could be
chosen from among the NPs (babies) and would correctly answer the question of
what VPs (what Alice likes). I offer the following lexical entry for the relative com-
plementizer Crel :⁴⁷

(6.20) JCrel K = λaa .λPet .λye .λxe .λgg . x(g) = y(g) ∧ P(x)(g)

A relative clause, CPrel , denotes a singleton set of individuals — the singleton of

individuals x(g) ∈ E which have property ↓ P and are identical to a given individual.
The property is supplied by the IP complement of Crel ; the individual is supplied
by the relative phrase. Notably, in some languages the relativizing element may be
a specific indefinite determiner, as in the (circumnominal) relative clauses in (6.21)
(see also Williamson 1987, Culy 1990, Cinque 2013).

(6.21) a. [[Fo sẽ yã daw-ninga/∗ dawã zaamẽ] wã] kula me.

you C saw man-spec.indef/∗ man.the yesterday the went-home
‘The man who you saw yesterday went home.’
(Peterson 1974: 74; Mooré (Niger-Congo))
b. [[N ni puhl saan-so/∗ sana] la] tšaŋya.
∗
I C greeted stranger-spec.indef/ stranger.the the has-gone
⁴⁷Dayal (1996: 191–193) also proposes a substantive semantic value for the relative complemen-
tizer. Roughly speaking, Dayal treats the relative complementizer as type ⟨et, ⟨et, ⟨et, t⟩⟩⟩, returning
for a relative clause ‘wh-NP VP’ the set of properties of the unique maximal plural individual which is
NP and VP; the relative CP is itself a type-⟨et, t⟩ plural definite description. Though such a semantics
may be suitable for free relatives combining with a property-type or individual-taking argument,
it doesn’t apply to relatives combining with a determiner quantifier. Dayal is thus forced to treat
relative expressions as ambiguous. Cf. also Gajewski 2008 which captures the maximizing effects
via a syntactic definite feature on the relative complementizer. By contrast, the entry in (6.20) and
assignment-variable-based treatment of relative determiners generalizes to headed and free relative
clauses. The distinctive definiteness effects associated with free relatives are captured, not via a
distinct determiner-like relative complementizer, but via an implicit maximalizing operator (ID in
(6.13)), analogous to the matrix determiner in restrictive relatives. In both cases the relative CP
denotes a set of individuals. This clause is suitable to combine with a determiner quantifier, as in
headed relatives, or maximalizing operator, as in free relatives. Maximization/definiteness effects
are distinguished from relative clauses proper. (See also n. 20.)

74
 ‘The stranger who I greeted has gone.’
(Peterson 1974: 77; Dagbani (Niger-Congo))
c. [[Atia n da’ bua seka/∗ la da’a zaam] la] bɔi mɛ.
∗
Atia C bought goat spec.indef/ the market yesterday the lost
‘The goat that Atia bought at the market got lost.’
(Hiraiwa et al. 2017: 8; Gurenɛ (Niger-Congo))

A common approach is to treat specific indefinites as introducing a variable for

a choice function (Reinhart 1997, Kratzer 1998b). Drawing on data such as
(6.21) I suggest that we treat relative words as choice-function pronouns. A choice
function Fcf of individuals selects a particular individual o ∈ E from a non-empty set
of individuals.

(6.22) A function F ∶ [X → T] → X is a choice function iff ∀P ∶ X → T ≠ ∅∶ P(F(P)).

The external individual argument of Crel is supplied by the selected individual de-
noted by the relative phrase [ Drel whcf NP]. The assignment-quantification intro-
duced by the determiner shifts what individual is selected and thereby determines
the domain.
A first-pass entry for ‘every’ base-generated internal to a relative clause is in
(6.23). The derived semantic value of a simple sentence with a free pronoun and
headed relative follows in (6.24). (Here and in what follows I continue to assume that
in the intended interpretation h(1s) is the world of the possibility represented by h,
i.e. h(1s) = @(h) (§§3, 4.4). I will use ‘cf ’ and ‘Fcf ’ specifically for indices/variables of
type ⟨et, e⟩ that are choice functions. Recall ↓ which “lowers” an item in a domain to
an item composed out of associated elements of the model (§2). I use (e.g.) ‘babyu ’
for the characteristic function of the set of individuals o ∈ E such that o is a baby
in u. For space purposes I will omit many intermediate steps in derivations, and I
often continue to omit explicit reference to the quantification over w ∈ Ds when the
modal domain has been derived; further calculations are left to the reader.)⁴⁸

(6.23) JeveryK = λP+⟨a,et⟩ .λQ+⟨a,et⟩ .λgg . [∀xe ∃aa ∶ P+ (a)(x)(g)] Q+ (a)(x)(g)

⁴⁸In the general definition of the binder-index: with every, τ = a, σ = ⟨⟨a, et⟩, t⟩, σ1 = e; with
DP: D CPrel , τ = a, σ = t, σ1 = e. I assume that the relative word must have the same assignment-
variable as the local relative complementizer, as due to agreement. As with other determiners,
there may be reasons for incorporating domain variables to further restrict the domains of relative
determiners. In (6.24) the domain variable may be treated as sister to ‘baby’ or as an additional
argument of ‘which’, restricting the domain of the choice-function pronoun to a set of contextually
relevant babies; see n. 28. For simplicity I continue to bracket such additional structure.

75
(6.24) ‘Every baby which she likes laughed’
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a

DP⟨2,a⟩
laughed t1s

every⟨2,a⟩ CPrel

DP1 C

wh2 g2 baby ⟨2,e⟩

w1 g1
Crel t2a she1 g1
t2e
likes w1 g1
Jwh2 g2 K = λPet .λgg . g(2a)(2cf)((↓ P)(g))
JDP1 K = λgg . g(2a)(2cf)([λgg .the function f s.t. for all ye , f(y(g)) iff y(g) is
a baby in g(1a)(1s)](g))
= λgg . g(2a)(2cf)(ιf s.t. for all ye , f(y(g)) iff y(g) is a baby in g(1a)(1s))
≈ λgg . g(2a)(2cf)(babyg(1a)(1s) )
JCPrel K ≈ λxe .λgg . x(g) = g(2a)(2cf)(babyg(1a)(1s) ) ∧ g(1a)(1e) likes x(g) in g(1a)(1s)
JDP⟨2,a⟩ K ≈ λQet .λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(babyg(1a)(1s) )
∧ g(1a)(1e) likes x(g) in g(1a)(1s)] Q(x)(g[a(g)/2a])
JSK ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(babyg− (1s) )∧g− (1e) likes x(g) in g− (1s)]
x(g) laughed in @(g− )

76
Roughly put, the DP ‘every baby which she likes’ quantifies over those individuals
o ∈ E that are chosen by some choice function or other (a(g)(2cf)) from among
the babies (babyg− (1s) ) and are liked by the contextually relevant individual (g− (1e)).
The sentence (6.24) is true iff every such individual o laughed.
The above syntax/semantics compositionally derives the intuitive interpretation
delivered by familiar semantics with Predicate Abstraction and intersective mod-
ification, and does so without positing additional composition rules or principles
for interpreting reconstructed phrases (e.g. Trace Conversion). The syntax is co-
opted from prominent head-raising and D-complement analyses of headed rela-
tives. The compositional semantics parallels the treatment of type-driven bind-
ing/quantification with verbal quantifiers and the declarative complementizer: Just
as a declarative clause predicates a property of a world identical to a world in a
domain determined by an embedding modal, a relative clause predicates a property
of an individual identical to an individual in a domain determined by an embedding
determiner quantifier. Complementizers in matrix clauses raise from VP as quan-
tifiers over worlds, capturing the obligatory local reading of the main predicate’s
world argument; analogously the relative complementizer Crel raises from the gap
position in the relative clause, capturing the obligatory link between the gap and
the nominal head. Modal quantifiers raise from inside their complement clause as
quantifiers over assignments, determining the relevant modal domain; analogously
the determiner quantifier ‘every’ in (6.24) raises from inside its complement CPrel ,
determining the relevant domain of individuals. The syntax/semantics of quantifica-
tional DP: D CP provides a counterpart of the syntax/semantics of modal quantifiers
from §3 in the domain of individuals (§6.1.2).
It is standard in syntax to treat features on a head X as projecting to the XP.
Given the treatment of binder-indices as features on expressions, the assignment
binder-index on ‘every’ projects to the DP. In what follows I will suggest that this
assignment binder affords a resource for capturing shifted interpretations of pro-
nouns and various phenomena with donkey anaphora.
6.2 Donkey anaphora
6.2.1 Pronouns and copies
Consider (6.25), schematically represented in (6.25a) (ignoring worlds).

(6.25) ‘Every baby which got a toy liked it’

a. [S …[[DP every⟨2,a⟩ [CPrel wh2 g2 baby [[Crel t2a ]⟨2,e⟩ [t2e got ???-toy]]]] liked ??? ]]

As is familiar, the donkey pronoun ‘it’ isn’t c-commanded by its apparent linguistic

77
antecedent, the indefinite ‘a toy’. Simply representing ‘it’ with some individual pro-
noun [it3 gi ] doesn’t capture the intuitive anaphoric⁴⁹ interpretation: coindexing
the assignment-variable with the topmost assignment-binder yields a claim about
gc (3e), and coindexing it with every⟨2,a⟩ yields a claim about gc (2a)(3e). Ap-
proaches to donkey anaphora are diverse. One might layer resources from one’s
preferred theory into an assignment-variable-based framework — e.g., adding unse-
lective binders and revising the treatment of indefinites (Kamp 1981, Heim 1982), or
treating donkey pronouns as E-type descriptions and introducing mechanisms for
recovering the relevant descriptive content (Heim 1990, Büring 2004, Elbourne
2005). However, it is worth exploring whether the assignment-quantification with
determiners could be exploited for developing an account of apparent non-c-com-
mand anaphora such as donkey anaphora.
A common approach in both dynamic and non-dynamic theories is to treat
indefinites as introducing a (new) variable, as for an individual (Kamp 1981, Heim
1982) or choice function (Reinhart 1997, Kratzer 1998b). Following our treat-
ment of relative words, suppose we represent the relevant uses of indefinites such
as ‘a’ likewise as choice-function pronouns.⁵⁰ To a first approximation, I suggest
representing donkey pronouns in sentences such as (6.25) as copies of their linguis-
tic antecedent. There are various ways of implementing this idea. One option —
the option I wish to pursue here — is to represent the anaphoric pronoun (n. 49)
as a copy of the interpreted material of the antecedent (refinements to follow). A
donkey pronoun such as ‘it’ in (6.25) is spelled out at LF with a coindexed choice-
function pronoun and elided nominal complement, as in (6.26). (Hereafter, unless
indicated otherwise I use the strikethrough for material that is interpreted at LF but
⁴⁹Here and throughout I use ‘anaphora’ and its kin informally for phenomena in which an
expression seems to receive its interpretation directly from the linguistic context. For familiarity I use
‘anaphor’/‘anaphoric pronoun’ broadly to cover uses of pronouns with intuitively bound readings,
though sometimes to the exclusion of anaphors proper; context should disambiguate. My usage
doesn’t presuppose that pronouns with “anaphoric” or “intuitively bound” readings are to be analyzed
as syntactically or semantically bound by their felt linguistic antecedents. I briefly consider anaphors
proper such as reflexives and return to ordinary bound readings of pronouns in §6.3. I continue to
focus exclusively on intra-sentential (non-discourse linguistic) anaphora.
⁵⁰I don’t assume that all indefinites are to be analyzed uniformly in this way. For instance,
various languages have distinct forms for specific and nonspecific indefinites. See e.g. Kratzer 1998b
for an ambiguity account positing choice-function/quantificational analyses of specific/nonspecific
indefinites, respectively; cf. Fodor & Sag 1982.

78
unpronounced. I often use ‘oi ’ as short for ‘oicf ’ in individual and choice-function
variables.)⁵1

⁵1Such an approach might be understood as an assignment-variable analogue of the NP-deletion

E-type theory in Elbourne 2001, 2005, 2013 (though see below); see Postal 1969, Wiltschko
1998, Radford 1997, Déchaine & Wiltschko 2002, Baltin 2012, Patel-Grosz & Grosz 2017 for
broader “X+elided-complement” analyses of certain pro-forms crosslinguistically. I focus exclusively
on instances of donkey anaphora where the pronoun has a linguistic antecedent. I leave open how
the account may be extended to “paycheck” pronouns, as in (i), where the pronoun’s covarying
interpretation is contextually determined.
(i) [Context: A new faculty member picks up her first paycheck from her mailbox. Waving it in the
air, she says to a colleague:]
Do most faculty members deposit it in the Credit Union? (Jacobson 2000: 89n.12)
I don’t assume that all pronouns or anaphoric expressions need be analyzed in this way. For instance,
depending on one’s views on the relation between referential and anaphoric/bound pronouns, free
readings could be represented with simplex type e pronoun variables [oie g], or with choice-function
pronouns coindexed with T and a (possibly trivial) property-type pronoun (≈“which (relevant)
thing”). For unified theories of pronouns, see e.g. Heim & Kratzer 1998, Elbourne 2005, 2013;
for contrasting approaches, see Aoun & Hornstein 1992, Noguchi 1997, Hornstein 2001, 2007,
Bhat 2004, Déchaine & Wiltschko 2002, 2017b, Kratzer 2009. Likewise I leave open the specific
syntactic category of the relevant English pronouns; the account needn’t be committed to treating
choice-function pronoun elements as uniformly in D positions (see Déchaine & Wiltschko 2002,
2015, 2017a,b, Baltin 2012, Patel-Grosz & Grosz 2017 for overviews on potential structural
differences among pronouns crosslinguistically; also n. 73). Given our purposes I ignore internal
projections such as for phi features. The analysis is compatible with there being differences in phi
features in certain examples, as in (iii). Such examples may be assimilated to independent phenomena
in which a pronoun’s phi features are apparently uninterpreted (e.g. due to Feature Deletion or
Feature Transmission), as with “fake indexicals” such as in (iv) (cf. von Stechow 2003, Rullmann
2003, Hornstein 2007, Heim 2008, Kratzer 2009, Sudo 2012, Déchaine & Wiltschko 2015,
Wiltschko 2016).
(ii) Every freshman who met a senior hoped that they would become friends.
(iii) a. I’m the only one around here who will admit that I could be wrong. (Partee 1989: fn. 3)
b. Only you remember our first appointment. (Kratzer 2009: ex. 87)

79
(6.26) ‘Every baby which got a toy liked it’
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a

DP⟨2,a⟩ VP

every⟨2,a⟩ CPrel
liked t1s
it3 g2
toy-w1 -g1
C
wh2 g2
baby-w1 -g1 ⟨2,e⟩

Crel t2a t2e

got-w1 -g1
a3 g2
toy-w1 -g1

JDP⟨2,a⟩ K ≈ λQet .λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(babyg(1a)(1s) )

∧ x(g) got a(g)(3cf)(toyg(1a)(1s) ) in g(1a)(1s)] Q(x)(g[a(g)/2a])
JSK ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(babyg− (1s) ) ∧ x(g) got a(g)(3cf)(toyg− (1s) ) in g− (1s)]
x(g) liked a(g)(3cf)(toyg− (1s) ) in @(g− )

CPrel denotes a singleton set of individuals that are selected by some particular choice
function and got a toy selected by some particular choice function. The determiner’s
assignment-binder binds the assignment-variables in the choice-function pronouns
representing the indefinite ‘a toy’ and relative phrase ‘which baby’. Roughly, the
DP ‘every baby which got a toy’ quantifies over those individuals o ∈ E s.t. there
are choice functions F, F′ s.t. F (=a(g)(2cf)) selects o from among the babies, F′
(=a(g)(3cf)) selects some o′ (=a(g)(3cf)(toyg(1s) )) from among the toys, and o got

80
o′ — babies o that got some toy or other o′ . The sentence is true iff for every such
o, o liked o′ . The apparent anaphoric connection between ‘a toy’ and ‘it’ is cap-
tured by (i) the syntactic identification of the donkey pronoun with its antecedent
[a3 -g2 toy-w1 -g1 ], and (ii) the assignment-binding introduced by the determiner.
The proposed treatment of certain anaphoric pronouns as “copies” of their lin-
guistic antecedents shouldn’t be conflated with copies in the sense of the copy theory
of movement (see n. 13). Copies of movement are treated as identical to the moved
expression (or at least identical with respect to phi features, [wh] features, Case, etc.;
cf. also Nunes 2004, van Koppen 2007). In contrast, the anaphoric expressions in
question, on the present proposal, need only be equivalent to their antecedents with
respect to features interpreted at LF (subject to a qualification below; see n. 51). The
pronouns needn’t be identical to their antecedents with respect to other features rel-
evant for non-semantic (purely syntactic, phonological) reasons. It is a substantive
question, for a given type of anaphoric expression, whether the anaphoric “copy,” in
the present usage, is the result of some syntactic Copy operation. (More on these
issues below and in §§6.3.3, 7.7, 8.3.)
If the LF representations of donkey pronouns were syntactically complex in the
proposed way, one might expect to find evidence for this complexity in other lan-
guages (n. 51). First, indirect evidence may come from independently attested struc-
ture in pronouns used as donkey pronouns. For instance, in languages with nom-
inal tense such as Somali, tense morphology may appear on any common noun,
as in (6.27) (Lecarme 1996, 2004, Nordlinger & Sadler 2004, Fenton 2010,
Adamou 2011). Interestingly, whereas (e.g.) indefinite nominals and even certain
syntactically marked definites are tenseless in Somali, individual pronouns may be
marked for tense — not only in referential uses, as in (6.28), but also in bound uses
and donkey sentences, as in (6.29). As with other nominals, the tense on a pronoun
can be independent of clausal tense.

(6.27) Nín-ka cáan-ka ah ee búug-gani qoray waa

man-def.nonpast fame-def be and book-def.dem.past wrote.past decl
Shákisbíir.
Shakespeare
‘The famous man who wrote this book is Shakespeare.’
(Lecarme 2008: ex. 3a)
(6.28) a. Isá-gíi baa hádal-kíi qaatáy, oo yiri:
he-emph.past decl talk-def.past take.past and said
‘He began to speak, and said:’ (Lecarme 2008: ex. 10a)

81
 b. Isa-ga waxaa arkay Axmed.
him-emph.nonpast decl saw.past Axmed
‘Axmed saw him.’ (Özyıldız & Ivan 2017: ex. 6)
(6.29) a. Qof walba oo hali leh iya-dai wuu garacaa.
person every rel she-camel has her-emph.nonpast decl beats
‘Everyone who has a cameli beats it i .’ (Özyıldız & Ivan 2017: ex. 18a)
b. Nini hád-díi uu seexdó oo sóo toosó, waa isá-gíii
man time-def.past he sleeps and dir wakes-up decl him-emph.past
ún.
only
‘If a mani goes to sleep and then wakes up, hei is only himself (i.e. the
same as he was before).’ (Lecarme 2008: ex. 19a)

Somali tense-related DPs have full TP complements (Lecarme 2008, 2012). Exam-
ples such as (6.29) suggest that such structure can apply to donkey pronouns, e.g.
roughly as in (6.30).

(6.30) Somali (pro-)DP: D+TP (Lecarme 2008, 2012)

[ DP D [ TP T [ … NP]]]
a. [ DP itcf [ TP Tnonpast [ … [ NP camel]]]]
b. [ DP hecf [ TP Tpast [ … [ NP man]]]]

Second, in some Romance and Scandinavian languages certain pronominal forms

freely allow overtly modified complements, even full restrictive relative clauses, as
in (6.31)–(6.32). The forms used in (6.31)–(6.32) are also used in donkey sentences,
as in (6.33).

(6.31) a. det som jeg spiste

it which I ate
b. han med rød hatt
he with red hat (Hestvik 1992: exs. 27, 29; Norwegian)
(6.32) a. lo bueno
it good
‘that which is good’
b. los buenos
they good
‘the good ones’

82
 c. lo que agrada
it that pleases
‘that which pleases’ (Luján 2004: exs. 2, 4; Spanish)
(6.33) a. Hvis en mann eier et esel, slår han det.
if a man owns a donkey beats he it
b. Si un hombre tiene un burro, lo trata mal.
if a man has a donkey it.acc treats badly

It is controversial how to analyze such constructions.⁵2 One hypothesis might be that

the pronouns are pro-NPs (pro-forms with the syntax of nouns), which can also take
overt modifiers, as in (6.34)–(6.35) with Japanese kare and English anaphoric ‘one’
(Postal 1969, Déchaine & Wiltschko 2002, Falco & Zamparelli 2016).

(6.34) a. [ DP kono [ NP kare]]

this he
‘this guy here’
b. tiisai kare
small he
‘he who is small’ (Noguchi 1997: 777; Japanese)
(6.35) a. this one
b. small one

However, such a hypothesis can be excluded. Unlike the Spanish/Norwegian pro-

forms, pro-NPs are generally incompatible with bound-variable readings (Déchaine
& Wiltschko 2002):⁵3
⁵2Notably, the Spanish third-person proclitic pronouns and Norwegian third-person gender-
neutral pronouns share a common form with (specific/definite, demonstrative) determiners. Such
synchronic and diachronic morphosyntactic connections among third-person pronouns, determin-
ers, and relative words are crosslinguistically common (Diessel 1999, Lyons 1999, Bhat 2004; also
e.g. Jespersen 1924, Postal 1969, Wiltschko 1998, Gair et al. 1999, Luján 2004, Trutkowsky
& Weiß 2016, Thurgood & LaPolla 2017). Luján (2004) analyses examples such as (6.32) as
pronoun-headed DPs, though the matter is controversial. With Luján, I assume as a working hy-
pothesis that the Spanish examples are like the Norwegian examples (Hestvik 1992) in being overtly
modified pronouns (though as discussed below I remain neutral on whether the pronouns are Ds;
n. 51). Proceeding under this assumption allows us take the overt complements at face value as e.g.
adjectives/relative clauses (as opposed to, say, being systematically reanalysed as nominalizations),
as shown below.
⁵3Though see Yashima 2015 on the specific binding-theoretic properties of Japanese kare.

83
(6.36) a. *Daremoi -ga karei -no hahaoya-o aisite-iru.
everyone he-gen mother love
‘Everyone loves his mother.’ (Noguchi 1997: 770; Japanese)
b. *Everybodyi thinks onei is better than average.
(6.37) a. Hver eneste gutti satte stolen bak hami .
every boy put chair.def behind him
‘Every boy put the chair behind him.’
b. Nadiei desea que el director lei dé un premio.
nobody wants that the director him give an award
‘Nobodyi wants the director to give himi an award.’

Following the typology in Déchaine & Wiltschko 2002, the pro-forms in (6.31)–(6.33)
thus have at least the structure of a φP, if not a DP (n. 52), which also take nom-
inal complements [ φP φ [ NP N]] (Patel-Grosz & Grosz 2017).⁵⁴ A diagnosis of
(6.31)–(6.32) follows straightway: The modifiers modify the implicit NP, e.g. roughly
as in (6.38) for (6.31) (using the English glosses). The position of the copied NP
posited in the representation of the donkey pronouns in (6.33) is the same position
as the position of the implicit NP overtly modified in (6.31)–(6.32), as in (6.39).

(6.38) a. [hancf /hecf [ NP NP [ PP with the red hat]]]

b. [det cf /itcf [ CPrel [which NP] … I ate t]]
(6.39) [det cf /itcf [ NP donkey]]

What distinguishes ‘it’ from han, det, etc. is whether it has a selectional property of
disallowing an overtly modified complement (e.g., CPrel or a multi-segmented NP).⁵⁵
More direct crosslinguistic support for the proposed analysis of donkey pro-
nouns as (internally complex) copies of their linguistic antecedents may come from
⁵⁴As is common I use φP as a cover term for an intermediate functional projection(s) in the
extended functional projection of the noun phrase.
⁵⁵Certain modified implicit complements of third-person plural ‘them’ may be possible to differ-
ent extents in some dialectics of American English, as in (ii).
(i) a. Them linguists are subversive.
b. [ φP them [ NP linguists]] (Déchaine & Wiltschko 2002: ex. 35a)
(ii) a. ?Them with the weird symbols are subversive.
[ φP them [ NP NP [ PP with the weird symbols]]]
b. ??Them who you used to hang with, what happened to ’em?
[ φP them [ CPrel [rel NP] … you used to hang with t]]
c. *them good
[ φP them [ NP good N]]

84
the morphosyntax of the donkey-anaphoric expressions themselves. Notably, in
certain Chinese and Hmong donkey sentences the anaphoric expression must be a
complete copy of its (non-c-commanding) antecedent, as in (6.40)–(6.41) with the
indefinites shei and (tug) twg.

(6.40) a. ni xihuan sheii wo piping sheii .

you like who I criticize who
‘If you like X, I will criticize X’ / ‘Whoever you like, I will criticize them.’
b. *ni xihuan sheii wo piping tai /∅i /na-ge-reni /shenme reni .
you like who I criticize him/her/∅/that-person/what-person
(Cheng & Huang 1996: exs. 14, 17; Mandarin Chinese)
(6.41) a. Tug twg i pum (proj ) los tug twg i yeej nyam (proj ).
clf which see top clf which always like
‘Whoever sees him will surely like him.’
b. *Tug twg i pum (proj ) los nwg i /proi yeej nyam (proj ).
clf which see top 3sg/∅ always like
(Mortensen 2003: ex. 35; Hmong)

Interestingly, overtly prenominal uses of English ‘it’ are even attested in toddler
speech, as in the naturally occurring example in (6.42a).

(6.42) a. Get it ladder! (McNeill 1970: 29; toddler English)

b. [ DP [ D it] [ NP ladder]] (Radford 1997: ex. 81)

Following Radford, uses such as (6.42) reflect a developmental stage where the child
hasn’t yet learned that the particular word ‘it’ in English selects for a phonetically
unrealized complement (unlike, say, ‘we’; see below).⁵⁶ One might imagine a pre-
cocious counterpart of McNeill’s toddler saying ‘Everyone who has a ladder likes it
ladder’, parallel to (6.40)–(6.41).
Second, although Déchaine & Wiltschko’s (2002) class of pro-DPs generally re-
sist bound readings, some are compatible with overtly realizing the morphosyntactically
complex DP, as in (6.43a) with the pronoun thú-tl’ò, in (6.43b) with English ‘we’, and
in (6.44) with the pronoun yi. Wiltschko 1998 argues on independent grounds
that German d-pronouns are also syntactically complex DPs with an elided nomi-
nal complement (see also Schwarz 2009, Patel-Grosz & Grosz 2017). Notably,
though d-pronouns are generally unacceptable with bound variable readings, they
⁵⁶Suggestively, Radford even glosses the toddler’s use of ‘it’ in (6.42) as analogous to an adult
speaker’s use of a specificity determiner such as ‘the’ (1997: 155; see n. 52).

85
may be used as donkey pronouns, as in (6.46).Class pronouns in Bùlì exhibit no
restrictions in bound or anaphoric readings, yet they also occur freely with full
nominals, as in (6.45) with wá (cf. also Postal 1969, Pesetsky 1978, Collins 1993,
Radford 1997).

(6.43) a. Tl’ó-cha-l-su qwemcíwe-t thú-tl’ò q’ami.

then.1sg hug d.fem-3sg girl
‘Then I’m going to hug her/that girl.’
(Déchaine & Wiltschko 2002: ex. 5; Halkomelem (Salish))
b. We students stick together.
(6.44) a. me na agbo yi.
I gave ram d.him
‘I gave him a ram.’
b. Kofi kpa ati-ɛ tati yi.
Kofi carved stick-def pestle d
‘Kofi carved the stick into a pestle.’
(from Collins 1993: 20–24, 32–33; Ewe (Niger-Congo))
(6.45) a. Siri lē dom wā.
bee prt bite d.cl1
‘A bee stung her.’
b. Nur ne wà doa ale jam bōro.
person with d.cl1 friend and past exist
‘There was a man and his friend.’ someone likes his mother-d
(Schwarz 2016: exs. 14, 21, Hiraiwa 2005: 200–201; Bùlì (Niger-Congo))
(6.46) Wenn ein Bauer einen Eseli hat, dann schlägt er deni .
if a farmer a donkey has then beats he d-pron
‘If a farmer owns a donkey, then he beats it.’
(Wiltschko 1998: ex. 57; German)

On the flip side, in Hindi donkey sentences the anaphoric expression can be
overtly realized as a copy of the antecedent nominal, as in (6.47) in which the bare
gadheKO ‘donkey’ is anaphorically linked to the non-c-commanding indefinite koii
gadhaa ‘some donkey’.

(6.47) har aadmii jis ke paas koii gadhaai hotaa hai gadheKOi maartaa hai.
every man REL with some donkey has donkey beats
‘Every man who has a donkey beats it.’ (Srivastav 1991: ex. 45a; Hindi)

86
Such examples can be represented parallel to their English counterparts at LF, as re-
flected schematically in (6.48) for (6.47) (ignoring worlds and using English glosses).

(6.48) Hindi donkey LF (schematic)

… [DP … [a3 -g2 donkey]]⟨2,a⟩ beats [F3 -g2 donkey]

In (6.47) it is the NP which is pronounced and the choice-function pronoun is im-

plicit. The representation of the Hindi donkey anaphor is the mirror image of the
representation of ‘it’ in (6.26).⁵⁷
A potentially contrasting hypothesis regarding (6.47) is that the definite expres-
sion includes a covert demonstrative determiner, and that it is this demonstrative
that captures the felt anaphoric connection. (6.47) would be relevantly analogous
not to (6.25), the thought goes, but to (6.49) (cf. Abbott 2002).

(6.49) Every child who got a toy will bring that toy for show-and-tell.

As noted above, many languages use the same format for demonstratives and pro-
nouns such as third-person pronouns (n. 52). Yet it won’t do to respond by simply
positing a covert demonstrative. Compare (6.47), in which the bare nominal can
function as the anaphoric expression, with (6.50), in which removing the demon-
strative renders the sentence ungrammatical.

(6.50) a. [jo laRkii khaRii hai]i vo laRkiii lambii hai.

REL girl standing is DEM girl tall is
lit. ‘Which girl is standing, that girl is tall.’
(=‘The girl who is standing is tall.’)
b. *[jo laRkii khaRii hai]i laRkiii lambii hai.
REL girl standing is girl tall is
(Srivastav 1991: exs. 13a, 17a)

Although correlative constructions such as in (6.50) generally require a demonstra-

tive correlate in the main clause, the ungrammaticality of (6.50b) cannot be due
to the lack of an overt demonstrative. (6.51), in which the correlate is implicit, is
acceptable (Srivastav 1991, Bhatt 2003; cf. Lipták 2012 on Hungarian).
⁵⁷It is perhaps interesting that the ostensible bound names in (4.34)–(4.35) may be represented
analogously to the Hindi donkey anaphor in (6.48), with an elided choice-function pronoun and
pronounced NP, roughly as in (i).
(i) [ … [[DP every⟨2,a⟩ [CPrel … a2 -g2 Donald]]⟨2,a⟩ [VP … F2 -g2 Donald … ]]] (cf. (4.33)–(4.35))

87
(6.51) [jo laRkii khaRii hai] lambii hai.
REL girl standing is tall is
If a covert demonstrative could capture the anaphoric connection in (6.48), one
might wonder why it couldn’t do as well in (6.50b).
The account of donkey anaphora in this section affords a unified syntax and
semantics for the English and Hindi donkey sentences in (6.25) and (6.47). In both
cases the anaphoric expression is spelled out via a copied choice-function pronoun
and NP complement. What distinguishes them is which material is pronounced.
(We will return to applications to anaphoric demonstratives in correlatives in §7.)
6.2.2 Asymmetric, universal, and existential readings
Although the entry for ‘every’ in (6.23) introduces quantification over assignments,
the determiner’s semantics is still a selective quantification that relates sets of indi-
viduals. The account avoids the proportion problem (Heim 1990) facing unselective
binder approaches, which fundamentally relate sets of assignments (Lewis 1975,
Kamp 1981, Heim 1982). Consider an asymmetric reading with ‘most’ in (6.52),
assuming an LF parallel to (6.26).

(6.52) ‘Most babies which got a toy liked it’

≈ λgg . [MOST xe ∶ ∃aa ∶ x(g) = a(g)(2cf)(babyg(1s) ) ∧ x(g) got
a(g)(3cf)(toyg(1s) ) in g(1s)] x(g) liked a(g)(3cf)(toyg(1s) ) in @(g)

(6.52) is correctly predicted false in a scenario where one baby b1 got four toys and
liked them, and two babies b2 , b3 didn’t like the unique toy they got: b2 /b3 satisfy
the restriction that there are choice functions which select them from among the
babies and select a toy they got; however, b2 /b3 fail to satisfy the scope condition
since they don’t like the selected toy, i.e. a(g)(3cf)(toyg(1s) ). The semantics avoids
giving symmetric construals of asymmetric readings.
The semantics for ‘every’ in (6.23) derives a so-called “universal reading” for
donkey sentences: (6.26) requires that for every baby o s.t. there is some choice
function F′ that selects a toy o got, o likes the toy selected by F′ . Although many
theories predict only universal readings for donkey sentences (Kamp 1981, Heim
1982, Groenendijk & Stockhof 1991, Elbourne 2005), it has been observed that
some donkey sentences have existential readings. (6.53b) is intuitively true as long
as every person who has a dime will put some dime that she has in the meter.

(6.53) a. Yesterday, every person who had a credit card paid his bill with it.
(R. Cooper)

88
 b. Every person who has a dime will put it in the meter.
(Pelletier and Schubert 1989)
c. Every person who submitted a paper had it rejected once.
(Chierchia 1995: ex. 3)

One way of capturing existential readings is to weaken the semantics of the deter-
miner. Consider the following alternative entry and semantic value for (6.26):

(6.54) JeveryK = λP+⟨a,et⟩ .λQ+⟨a,et⟩ .λgg . [∀xe ∃aa ∶ P+ (a)(x)(g)]

∃a′a ∶ P+ (a′ )(x)(g) ∧ Q+ (a′ )(x)(g)
(6.55) J(6.26)K ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(babyg(1s) ) ∧ x(g) got a(g)(3cf)(toyg(1s) ) in g(1s)]
∃a′a ∶ x(g) = a′ (g)(2cf)(babyg(1s) ) ∧ x(g) got a′ (g)(3cf)(toyg(1s) ) in g(1s)
∧ x(g) liked a′ (g)(3cf)(toyg(1s) ) in @(g)

The semantics in (6.54) requires that, for every individual satisfying the restriction
given some assignment a, there is some assignment — not necessarily a — that veri-
fies both the restriction and scope. The derived semantic value for (6.26) represents
the existential reading: (6.55) says, roughly, that for every o ∈ E s.t. [there are choice
functions F, F′ s.t. o = F(baby) and o got F′ (toy) = o′ ], there is a choice function
F′′ s.t. o got F′′ (toy) = o′′ and o liked o′′ . There is universal quantification over the
subject-babies o, but existential quantification in the scope over toys liked by o —
more precisely, over assignments which determine choice functions that select a toy
o liked from among the toys o received. The condition “P+ (a′ )(x)(g)” in (6.54)
ensures that the choice function F′′ relative to which the individuals o satisfy the
scope condition — that o likes F′′ (toy) — selects a toy received by o. (6.25) is still
correctly predicted false in a scenario where a baby b liked some toys or other but
hated all the toys it received.
It is contentious whether some donkey sentences conventionally have both ex-
istential and universal readings. I leave open whether all universal readings can be
derived conversationally, or positing some sort of ambiguity is ultimately necessary
(see Kanazawa 1994, Chierchia 1995, King 2004, Brasoveanu 2007). (Here-
after, unless indicated otherwise, I simplify by assuming “universal reading” entries
such as (6.23).)
6.2.3 Donkey pronouns in attitude ascriptions
Consider (6.56) where a donkey pronoun is embedded under an attitude verb.

89
(6.56) Every woman who has a secret admirer thinks he is stalking her.
(King 2004: ex. 18)

King 2004 argues that such examples are problematic for main approaches to don-
key pronouns such as DRT and dynamic theories. Insofar as these approaches treat
donkey pronouns as semantically bound variables, they predict a specific reading for
‘he’, i.e. a reading which attributes to each relevant woman a belief about a particular
individual. However, the attitude ascription can also receive a nonspecific reading
which “can be true even though the women in question don’t know who their secret
admirers are” (2004: 105).
The account in this section provides a unified analysis of specific and nonspecific
readings of donkey pronouns in intensional contexts. Consider the following LF and
derived semantic value for (6.57) on a nonspecific reading, where which mule did the
ruining may vary across the subject’s doxastic alternatives. (Imagine that a trickster
sent each zoo-owning epistemologist a letter with the latest delivery of animals that
some or other of the zebra-looking animals was a cleverly painted mule.)

90
 (6.57) ‘Every epistemologist who got a painted-mule thinks it ruined-epistemology.’

T⟨1,a⟩ CP

⟨1,s⟩

C∅ t1a

DP⟨2,a⟩

⟨3,a⟩

every⟨2,a⟩ CPrel thinks t1s

⟨1,s⟩

C C∅ t3a

wh2 -g2 epist-w1 -g1 ruined-ep t3s

⟨1,e⟩
it3 g3
Crel t2a t1e painted-mule-w1 -g1

got w1 -g1
a3 g3
painted-mule w1 -g1

JSK ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(epistg− (1s) )

∧ x(g) got a(g)(3cf)(painted-muleg− (1s) ) in g− (1s)]
∀a′a ∶ a′′ (g) is compatible with SOMx(g),@(g− ) ,
a′ (g)(3cf)(painted-muleg− (1s) ) ruined-epistemology in @(a′ (g))

The subject DP quantifies over individuals o ∈ E s.t. there are choice functions F, F′
s.t. F selects o from among the epistemologists and o got F′ (painted-mule). (6.57)
says that for every such o, for every possibility g′ compatible with o’s state of mind,
g′ (3cf)(painted-mule) ruined epistemology. Which choice function is assigned to
3cf, and hence which individual is selected, may vary across the subject’s doxastic
alternatives. This represents the nonspecific reading.
Coindexing the donkey pronoun’s assignment-variable with the subject DP’s
binder-index would represent a specific reading: the scope condition would require
that a particular individual selected by a(g)(3cf) ruined epistemology in every pos-

91
sibility compatible with the subject’s state of mind. Nonspecific readings such as
(6.57) may thus motivate allowing the anaphoric pronouns’ assignment-variables to
be locally bound by an intervening assignment-binder.
In contrast, the world-pronoun in the donkey pronoun’s elided NP is not lo-
cally bound. King (2004) characterizes the nonspecific reading of sentences such as
(6.56) as follows:

These sentences certainly appear to have readings on which they at-

tribute de dicto beliefs to the women in question. That is, they have
readings on which they attribute to the women in question general be-
liefs to the effect that they are being stalked by secret admirers. This is
why these sentences can be true even though the women in question
don’t know who their secret admirers are, and so have no beliefs about
particular persons stalking them. (King 2004: 105; underline added)

King’s gloss in the underlined portion is misleading. We saw in §4.2 that the de re/de
dicto distinction cannot be assimilated to the specific/general distinction. The fact
that the subject’s beliefs are nonspecific — that the beliefs aren’t about a particular
individual — doesn’t imply that the beliefs are de dicto in the sense that the subject
represent the stalkers in question as secret admirers. The long-distance binding of the
embedded world-pronoun in (6.57) captures this: The attitude ascription requires
that, for some group of actual painted mules, each epistemologist in question thinks
some or other of them ruined epistemology.
Examples such as (6.59) reinforce the de re reading of the donkey pronoun.

(6.58) Every boy who has a nickel in his pocket thinks it should go in the meter.
(6.59) [Context: The meter only takes quarters, and the boys know this. However,
the boys confuse their coins and think that the nickels they have are quar-
ters.]
Every boy who has a nickel in his pocket thinks it should go in the meter.
a. Nonspecific de re: ≈ every boy who has some coins o that are nickels thinks
that some o or other should go in the meter
b. ‘it’ ≈ “a nickel@ ”

(6.58) might be true because every nickel-owning boy correctly thinks that there are
nickels in his pocket, and thinks that some or other of them should go in the meter.
Yet as the context in (6.59) indicates, the quantified attitude ascription can be true
even if the boys don’t think that the coins in their pocket are nickels. What must be

92
the case, however, is that each boy thinks, of a certain collection of coins (=actual
nickels) he has, that some or other of them should go in the meter.
The above examples indicate that embedded donkey pronouns can have de re
readings — readings where the NP’s world-pronoun is coindexed with its antecedent
and is bound long-distance. The stronger claim that the pronouns must have de re
readings seems plausible as well. Contrast the infelicitous use of the donkey pronoun
‘it’ in (6.60) with the felicitous use of the overtly complex ‘a dime’, reflected in the
informal indexing in (6.61).

(6.60) [Context: Each of the boys has dimes and nickels in his pocket. They are all
confused about which coins are which: they think the dimes are nickels and
the nickels are dimes. They would each say ‘the meter only takes dimes’, but
since they are confused they would try to use a nickel.]
a. #Every boy who has a dime in his pocket thinks it should go in the meter
b. Every boy who has a dime in his pocket thinks a dime should go in the
meter
(6.61) a. #Every boy who has a dime@ in his pocket thinksi iti should go in the
meter.
b. Every boy who has a dime@ in his pocket thinksi a dimei should go in the
meter.

‘It’ is anomalous in (6.60) where the boys don’t think, of the actual dimes in their
pocket, that some or other of them should go in the meter.
Nonspecific readings of donkey pronouns in attitude ascriptions may thus con-
stitute a systematic case of the nonspecific de re. This point is arguably an in-
stance of a general idea about donkey pronouns: that “there is a pairing of indef-
inite antecedents with donkey-pronouns that is purely syntactic (as expected under
the unselective binding approach, but not available under the E-type approach)”
(von Fintel 1994: 176, drawing on Kratzer 1995; underline mine; cf. e.g. Heim
1990, Ward et al. 1991, Elbourne 2005, Patel-Grosz & Grosz 2010). Treat-
ing instances of donkey anaphora as copies of their linguistic antecedent (modulo
possible differences in the pronounced D-head’s assignment-variable) captures this.
King’s (2004) own “Context-Dependent Quantifier” (CDQ) account also treats
donkey pronouns as representing material from the linguistic context. Donkey pro-
nouns are analyzed as quantifiers which inherit their force and restriction from their
linguistic antecedent and other features of the sentence. Yet, first, absent additional
mechanisms for capturing de re readings with narrow scope quantifiers, simply say-

93
ing that the donkey pronoun “has the same quantificational force as its antecedent”
(2004: 105) fails to capture the above nonsepecific de re readings — readings where
the pronoun’s intuitive descriptive content is anchored to the world of the antecedent.
Whatever mechanisms are provided, a pressing challenge is to explain the contrasts
in available readings between donkey pronouns and explicit quantifier phrases, as
in (6.60)–(6.61) (see Büring 2004, Elbourne 2005).
Second, King diagnoses specific vs. nonspecific readings of donkey pronouns
in intensional contexts in terms of differences in scope: insofar as the pronoun
is quantificational, it may take varying scopes with respect to attitude verbs. This
diagnosis is problematic. Donkey pronouns can receive specific readings even when
embedded in scope islands such as ‘if ’-clauses, as in (6.62).

(6.62) Every star who has a/some secret admirer thinks that if he is a stalker, he is
evil.
a. Specific de re: ≈ every star who has a certain secret admirer o thinks that
if o is a stalker, o is evil

By contrast, the account in this section diagnoses the distinction between specific
vs. nonspecific readings of donkey pronouns in terms of coindexing on the choice-
function pronoun’s assignment-variable (see (6.57)).
Third, King must ensure that the donkey pronoun-quantifier’s restriction isn’t
identified with the restriction of its linguistic antecedent, but is rather “determined
by the predicative material in the sentence in which the antecedent occurs” (2004:
106). For instance, the restriction in (6.63a) must be recovered from ‘donkey’ and
some combination with ‘farmer’ and ‘owns’, so that ‘it’ quantifies over donkeys owned
by the quantificational subject; (6.63a) isn’t equivalent to (6.63b).

(6.63) a. Every farmer who owns a donkey beats it.

b. Every farmer who owns a donkey beats a donkey.

In other work applying the account to discourse anaphora King identifies the re-
striction of the anaphoric pronoun with “the intersection of the denotation of its an-
tecedent’s N-bar constituent, the denotation of the set term the antecedent attaches
to, and the denotation of any predicative material occurring in a sentence interven-
ing between S and the antecedent which contains a cdq [anaphoric pronoun] with
the same antecedent” (1994: 224). It isn’t immediately obvious how to implement
this idea in the case of donkey sentences. There is no syntactic constituent in (6.63a)
corresponding to the intuitive restriction ‘is owned by x’ or ‘is a donkey owned by x’.
The relation in more complex examples can be even more indirect:

94
(6.64) Every farmer who beats a donkey that loves a donkey hates it.

The syntax/semantics in this section avoids such issues. Instances of donkey anaphora
are represented syntactically as copies of their linguistic antecedents. The seman-
tic restriction in interpreting the donkey pronoun is captured by the assignment-
quantification introduced by the matrix determiner, which requires that the assign-
ment verifying the determiner’s scope argument verifies the restrictor argument.
6.3 Extensions: Determiner quantifiers and assignment-quantification
6.3.1 Quantifier phrases, specificity, and non-relative restrictive modification
The semantics in (6.23)/(6.54) treat determiners such as ‘every’ in headed relative
clauses as type ⟨⟨a, et⟩, ⟨⟨a, et⟩, t⟩⟩. The assignment binder-index resulting from
the movement from Crel and projecting to the DP helped capture the obligatory
link between the relative phrase and gap position in the relative clause, and certain
anaphoric dependencies between the determiner’s restrictor and scope. A pressing
question is how this syntax/semantics relates to their syntax/semantics when com-
bining with non-relative complements.
A conservative response would be to treat determiner quantifiers as systemati-
cally ambiguous between items taking arguments of type ⟨e, t⟩ vs. ⟨a, et⟩, or to intro-
duce a general lexical rule which converts the basic ⟨et, ⟨et, t⟩⟩ lexical entry into an
“assignment-lifted” entry. Such moves are familiar from type-shifting and flexible-
type approaches to quantifiers, connectives, etc. (Partee & Rooth 1983, Hendriks
1993, Jacobson 1999; cf. Heim & Kratzer 1998). LFs with the “wrong” homonym
or entry could be semantically excluded due to incurring a type-mismatch, as when
the determiner is generated inside a relative clause.
An alternative option is to analyze quantifier phrases generally as having internal
structure relevantly analogous to structure in a relative clause.

(6.65) [ DP every [ ZP … [ NP baby ] …]]

Headed relative clauses aren’t the only case of determiners taking CP complements
crosslinguistically (Caponigro 2002, de Vries 2002, Hiraiwa 2005, Hankamer &
Mikkelsen 2012). In some languages the determiners in headed relative construc-
tions can be used with factive clauses or root clauses, as in (6.66).

(6.66) a. Atim dɛ [[ CPrel Amoak ale da mango kuui diem] la]

Atim ate Amoak C bought mango rel yesterday dem
‘Atim ate the mango that Amoak bought yesterday.’

95
 b. [[ CP Atim ale dɛ mango-kú] la] tɛ Amoak po pienti.
Atim C ate mango-D dem gave Amoak stomach whiten
‘That Atim ate the mango pleased Amoak.’
c. [[ CP Atim nagi Amoak] la].
Atim hit Amoak dem
‘Atim hit Amoak (as I said).’
(from Hiraiwa 2005: 191, 31; Bùlì (Niger-Congo))

In light of crosslinguistic phenomena with nominal tense-marking, Lecarme (2008,

2012) and Chang (2012) explicitly treat DPs as taking tensed TP complements,
yielding a structure for DPs analogous to that for sentences (see §3.2); the time-
dependence of noun phrases is captured not in D itself but in DP-internal T(ense)
(see also Hiraiwa 2005, Kusumoto 2005, Ilkhanipour 2016). As we have seen,
the assignment(s) relevant for interpreting the clause’s main predicate can differ
from the assignment(s) for interpreting other predicates as well. Following Kayne
1994, Campbell 1996 and Luján 2004 offer generalized [ DP D CP] analyses, where
in non-relative/reduced-relative complements the CP is a small clause SC (i.e. lack-
ing Tense). Notably, on Campbell’s and Luján’s accounts, the main operator is even
analyzed as moving from within the complement CP/SC, as in (6.67) (see also Matthew-
son 1998, Ihsane & Puskás 2001, Giusti 2002, Laenzlinger 2005, Patel-Grosz
& Grosz 2017).

(6.67) a. [ DP Di [ SC ti NP]] (Luján 2004)

b. [ DP OPi [the [ SC ti NP]]] (Campbell 1996)

The focus in the above-cited accounts is on matters of syntax rather than com-
positional semantics. Though some of the authors gesture at interpretive and dis-
course effects of the posited DP-internal movements, semantic entries and deriva-
tions aren’t provided. The generalized assignment-quantificational syntax/semantics
for quantificational DPs pursued here can be understood as deriving the relevant
movement operations and DP-internal/-external binding relations in a specific com-
positional semantics via type-driven movement. The type-⟨aet, ⟨aet, t⟩⟩ implemen-
tation with determiner quantifiers is an instance of the general approach to the
syntax/semantics throughout the paper: Just as modals with clausal complements
are treated as verbal quantifiers raising from the assignment argument of (e.g.) Cdec ,
which raises from the world argument of the clause’s main predicate (§§3–4), deter-
miners such as ‘every’ with clausal complements are treated as nominal quantifiers
raising from the assignment argument of (e.g.) Crel , which raises from the individual
gap position (DPs with reduced complements to follow). A generalized assignment-

96
variable-based syntax/semantics for individual quantification/binding may thus be
understood as contributing to the long tradition of linguistic work on on parallels
between nominals and clauses, and individual and modal domains (see n. 45).
Treating determiner quantifiers as in general taking complements with struc-
ture analogous to that of a restrictive relative clause doesn’t require identifying the
complements as full CPs or representing all cases of restrictive modification as mod-
ification by a (possibly reduced) syntactic relative clause. For instance, in the D+CP
structure for ‘every baby’ in (6.68), the constituent [Fcf -g baby] needn’t be identified
as a relative (hence [+wh]) phrase, nor need the C+IP supplying the implicit domain
restriction (n. 28) be identified as Crel (again ignoring worlds; cf. Kayne 1994).

(6.68) Every baby laughed.

a. [ DP⟨2,a⟩ every⟨2,a⟩ [ CP F2cf -g2 baby [ C [ C0⟨i,e⟩ C t2a ] [ IP tie [P1et g1 ]]]]]
b. ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(baby) ∧ g− (1et)(x(g))] x(g) laughed

The complement of the quantifier needn’t even be treated as a CP/SC. Compare

(6.69) where the relevant structure is integrated in a nominal projection or a relevant
intermediate projection in the extended projection of the noun phrase (cf. Camp-
bell 1996, Matthewson 1998, Bhatt 1999, Bianchi 1999, Iatridou et al. 2001,
Laenzlinger 2005, Donati & Cecchetto 2015):

(6.69) a. [ DP every⟨i,a⟩ [ XP (YP) [ X X0 [ FP Fcf -tia baby]]]]

b. [ DP every⟨i,a⟩ [ FP Fcf -tia baby]]

In the LFs in (6.69) the quantifier raises for type reasons from the main nominal
constituent: Just as D raises from an internal argument of Crel when taking a com-
plement with a clausal structure (where the assignment-variable in Spec of C agrees
with the assignment-trace in C0 ), likewise D raises from an internal argument in
the nominal projection when taking a non-clausal complement (cf. (3.4), (6.67)).
The YP in the specifier of the nominal XP projection in (6.69a) may be understood
as a placeholder for relevant nominal modifiers in the extended NP layer (more on
which shortly; Cinque 1995, Giusti 2002). (Hereafter I often refer to analyses such
as those in (6.69) jointly as “D+XP” analyses.)⁵⁸
⁵⁸In (6.69a) I treat the complement of ‘every’ as XP, though certain modifiers may of course be
treated as themselves extended projections of N. In both (6.69a)–(6.69b), since the quantifier raises
directly from the nominal constituent, there is no need for an additional agreement mechanism
(§§6.1.3, 8.3). It would be interesting to examine potential parallels of the alternative D+XP/CP
treatments of the syntax/semantics interface in the verbal domain, such as with modals taking verbal
vs. clausal complements.

97
 In our examples with headed restrictive relative clauses there was an overt basis
for positing the constituent with the choice-function pronoun: the relative phrase.
Positing such a constituent in quantifier phrases with non-relative complements, as
in (6.68)–(6.69), may seem surprising. The remainder of the section examines how
a generalized assignment-quantificational syntax/semantics for determiner quanti-
fiers can help capture various further phenomena with complex quantifier phrases
and pronoun binding. Given our purposes we can bracket certain issues regarding
the specific internal syntax, as might be relevant for adjudicating among LFs such
as those in (6.68)–(6.69) (among others). But first it is worth pausing to consider
independent morphosyntactic/semantic evidence for the relevant structure — e.g.,
a choice-function pronoun or X/C head — in quantifier phrases more generally.
First, generalizing our syntax/semantics along the lines in (6.68)–(6.69) may give
precise expression to common ideas about the “specificity” (Enç 1991) of quantifi-
cational DPs. Here is Enç (emphases mine):⁵⁹

“[Q]uantifiers in natural languages quantify over contextually given

sets… For example, [(6.70) entails] only that [Sally] danced with every
contextually relevant man…
[(6.70)] Sally danced with every man.
If universal quantification is over contextually relevant sets of individ-
uals, it follows that [DPs] that quantify universally are specific. This
account also ensures that universally quantifying [DPs] presuppose ex-
istence… [T]he specificity requirement is not restricted to universally
quantifying [DPs]… All quantifiers are specific.” (Enç 1991: 11)

Lecarme draws on Enç’s notion of specificity in her accounts of nominal tense and
the distribution of nominal tense morphemes in Somali. What marks the distinction
between strong/presuppositional vs. weak/cardinal noun phrases (e.g. ‘every/most N’
vs. ‘many/three N’), according to Lecarme, is the “referentiality” of the Determiner
category. The inherent “definiteness” of D, following Déchaine & Wiltschko
2002, is treated as capturing differences in binding possibilities among pronouns
(emphases mine):

“Assuming D to be the locus of specificity, DPs… are directly con-

nected to the discourse, given their intrinsic deictic properties.”
“[T]ense is the relevant feature underlying the interpretive effects
⁵⁹Enç uses “NP” for our “DP.”

98
 that are subsumed under the notion of specificity.”
“[T]he tense morpheme in Somali nominals… links the reference
of DP to a discourse-identified set of possible referents. Specificity is
thus understood as ‘D(iscourse)-linking’ (Pesetsky 1987).”
(Lecarme 1999a: 301, 1999b: 16, 1996: 16)
“The cardinal vs. presuppositional distinction is tense-related: if true D
(i.e., D+T) relates to referentiality in some sense, an indefinite nonspecific
nominal phrase… is temporally indefinite… [N]onreferential nomi-
nals… are tenseless in Somali.” (Lecarme 2008: 204)
“[I]ndexicality is a necessary feature of… D-pronouns, and has a par-
ticular structural locus, namely the D-position.”
“D, the locus of indexicality, is responsible for assigning reference.”
“Consequently, fake indexicals are analyzed as instantiating φ rather
than D.” (Déchaine & Wiltschko 2015: 24; 2017b: 68, 80n.10)
“This view [locates] DP’s specific/anaphoric reference and discourse
function directly in its head Determiner… Accordingly, determiners
may be described as discourse-linking functions.” (Luján 2004: 130)

It isn’t always manifest to what extent authors take the above notions (index-
icality, presuppositionality, specificity, familiarity, definiteness, deixis, D-linking,
strength) to be relevantly related (cf. Matthewson 1998, Cardinaletti & Starke
1999). Plainly understood, the appeal to notions such as referentiality or definite-
ness can seem puzzling. One wouldn’t typically expect to find quantificational de-
terminers in a list of indexicals. On the face of it, quantifiers can also be bound, as
in (6.71). The weak/cardinal noun phrases in (6.72) would seem to be interpreted
with respect to “relevant sets of individuals” as much as ‘every man’ in Enç’s (6.70).
(6.72a) can be true even if Alice is satisfied with general cookie production; pointing
out in (6.72b) that lots of rival players were congratulating one another in other
games likely won’t earn one points with the tournament referee.

(6.71) In every music school, most guitar students wanted to play like the head
instructor. (cf. Heim 1991, von Fintel 1998)
(6.72) a. Alice gave Timmy a brownie because she doubted there were many cookies.
b. I regret to inform you that your teams didn’t meet the required standards
for sportsmanship: The referee didn’t observe three players congratulate
three players, and so you won’t be advancing to the next round.

99
It is a substantive question to what extent the properties being homed in on track
one another — e.g., being unbindable, triggering an existence/salience/familiarity
presupposition, being associated with implicit domain restriction, being compatible
with existential ‘there’ constructions, being potentially modally/temporally inde-
pendent of a matrix modal/tense. It is important not to conflate the observed syntac-
tic/semantic effects — domain restriction, specificity, discourse-linking, existence
presuppositions — with notional definiteness (familiarity, maximality, uniqueness)
or indexicality (see below; Lecarme 1996, 1999b, Giusti 1997, Cardinaletti &
Starke 1999, Ihsane & Puskás 2001, Matthewson 2013).
The syntax/semantics in this section affords a locus for the putative specificity,
in Enç’s (1991) sense, of “true” (Lecarme 2008: 204) quantificational DPs. DPs
such as ‘every baby’ in (6.68)–(6.69) “presuppose existence” (Enç 1991: 11) insofar
as the choice-function pronoun sister to the main NP is undefined for empty sets
of individuals. A lexical entry for C/Crel such as in (6.74a) would make the con-
nections with presuppositionality even more explicit (more on which in §8); if there
are no babies, the semantic value of the quantifier’s restrictor argument in (6.75) is
undefined for any value for y. An analogous preliminary entry for the X head is in
(6.74b), as compared to (6.73) (more on which shortly).

(6.73) JXK = λxe .λPet .λye .λgg . y(g) = x(g) ∧ P(y)(g)

(6.74) Presuppositional CP/XP (alternative)
a. JC/Crel K = λaa .λPet .λxe .λye .λgg ∶ x(g) = y(g) . P(y)(g)
b. JXK = λxe .λPet .λye .λgg ∶ x(g) = y(g) . P(y)(g)
(6.75) a. ‘every baby’
b. JCP/XPK = λye .λgg ∶ y(g) = g(ia)(kcf)(baby) . . . . y(g) . . .

Our analyses of certain anaphoric pronouns and relative words as choice-function

pronouns will also be extended to other anaphoric pro-forms/demonstratives and
interrogative wh phrases in §§7–8. Indeed Pesetsky’s (1987) paradigm case of D-link-
ing (discourse linking) — Enç’s “specificity” — is ‘which N’ phrases. Though includ-
ing a functional element such as a choice-function pronoun in the syntax of (non-
relativized) quantificational DPs might have seemed surprising, it has precedents in
work on D-linking. For instance, Holmberg 1993 and Rullmann & Beck 1998
distinguish D-linked NPs in the syntax by positing an unpronounced determiner/
demonstrative (cf. also Giusti 2002, Boeckx & Grohmann 2004, Jenks 2018).
Treating the specific determiner or determiner-like functional head as having an ad-
ditional internal argument (e.g. situation), as in (6.76), which figures in determining

100
the relevant domain, has precedents in various syntactic theories (cf. von Fintel
1994, Büring 2004, Elbourne 2005, 2013, Schwarz 2009, Patel-Grosz & Grosz
2017). As noted above, some accounts even treat the main operator as moving to its
interpreted position from within the complement of D (see (6.67), also §§3.2, 6.1).

(6.76) [ DP (D) [ FP [ F0 def/spec s] NP]]

(adapting Patel-Grosz & Grosz 2017: 261–262)

The choice-function pronoun posited in the extended NP layer in in [Fcf -g NP] —

and, on the D+XP analyses in (6.69), the movement from the posited internal (as-
signment) argument of Fcf — may be understood in parallel to such moves.
Although some of the accounts which posit an additional DP-internal element
identify it as a definite article or a deictic demonstrative, such identifications should
be regarded with caution.⁶⁰ Our identifying the choice-function pronoun in quan-
tificational DPs with (say) a definite determiner would be no more warranted for the
general case than our identifying it with an implicit relative/wh word on the basis of
our examples with headed relative clauses.
So, the choice-function-based analyses here and in the following sections may
provide a common basis for certain notions of “specificity” from the literature with
quantificational and non-quantificational DPs. Adopting the choice-function-based
analysis of names from §4.4, and extending the §6.2-analysis of donkey pronouns,
may help capture connections with treatments of definiteness/D-linking with names
and certain types of pronouns (cf. Boeckx & Grohmann 2004, Lecarme 2008,
Patel-Grosz & Grosz 2017). Alternative structural features or additional con-
straints may be associated with different types of expressions — e.g., a feature or
implicit functional head restricting to choice functions defined only for sets of in-
dividuals in the world of the discourse, or choice functions whose domain/range
is restricted to individuals included in a particular discourse-determined domain.
Such lexical or construction-specific differences may yield a more refined typology
of the various effects.⁶1
For purposes of illustration, consider an alternative D+XP syntax/semantics with
“weak” (non-presuppositional) quantifiers such as (6.77)–(6.79) (again ignoring worlds).
Sample entries for specifically cardinal readings are in (6.79). (I leave open what
category is projected in the movement of the weak quantifier. We will return to
⁶⁰One example: “Since both types of pronouns denote individuals, the most plausible assumption
is that both contain a definite determiner” (Patel-Grosz & Grosz 2017: 272). See below.
⁶1We will return to differences among (expressions represented here with) choice-function pro-
nouns below; see also §8.3.

101
plural objects (Link 1983) — objects with atomic individuals as (proper) parts — in
§7.2; an atom o is an individual with no proper parts, i.e. no o′ s.t. o′ < o, where
‘a ≤ b’ says that a is a possibly improper part of b.)

(6.77) Alice found no unicorns.

[no⟨2,a⟩ [unicorn [ X′ X [F2cf -t2a P1et -g1 ]]]]
(6.78) a. JnoK = λP⟨a,et⟩ .λQ⟨a,et⟩ .λgg . [¬∃xe ∃aa ∶ P(a)(x)(g)] Q(a)(x)(g)
JthreeK = λP⟨a,et⟩ .λQ⟨a,et⟩ .λgg . [3xe ∃aa ∶ P(a)(x)(g)] Q(a)(x)(g)
b. J(6.77)K ≈ λgg . [¬∃xe ∃aa ∶ x(g) = a(g)(2cf)(g− (1et)) ∧ x(g) is a unicorn]
Alice found x(g)
(6.79) Cardinal lexical entries
a. JδK = λP⟨a,et⟩ .λQ⟨a,et⟩ .λgg . #x(g)[∃aa ∶ P(a)(x)(g) ∧ Q(a)(x)(g)] Rδ nδ
● #o[ . . . o . . . ] ∶= the cardinality of the set of atoms o s.t. [ . . . o . . . ]

JnoK = λP⟨a,et⟩ .λQ⟨a,et⟩ .λgg . #x(g)[∃aa ∶ P(a)(x)(g) ∧ Q(a)(x)(g)] = 0

JthreeK = λP⟨a,et⟩ .λQ⟨a,et⟩ .λgg . #x(g)[∃aa ∶ P(a)(x)(g) ∧ Q(a)(x)(g)] ≥ 3
b. J(6.77)K ≈ λgg . #x(g)[∃aa ∶ x(g) = a(g)(2cf)(g− (1et)) ∧ x(g) is a unicorn
∧ Alice found x(g)] = 0
≈ the cardinality of the set of atoms x(g) ∈ E s.t. for some aa , x(g) =
a(g)(2cf)(g− (1et)) ∧ x(g) is a unicorn ∧ Alice found x(g), is 0

In (6.69) the strong quantifier ‘every’ is base-generated adjacent to the main NP,
and raises for type reasons to head the quantifier phrase; in (6.77), by contrast, the
weak quantifier is adjacent to ‘unicorn’ at LF, in the position of a modifier (see
(6.68)–(6.69); more on modifiers shortly). Distinguishing the LFs in this way co-
heres with common ideas in syntax about the predicative status of weak quantifiers
(cf. e.g. Giusti 1991, 1992, 1997, Cinque 1995, Giusti & Leko 1995, Dimitrova-
Vulchanova & Giusti 1996, Partee 1986). The semantics in (6.78)–(6.79) main-
tain the assignment-quantification of the determiner, as for capturing certain anaphoric
interpretations, without presupposing that the nominal predicate is satisfied in the
relevant domain. The semantic values derived for (6.77) require that there is no o
selectable among the relevant entities (=gc (1et)) s.t. o is a unicorn and Alice found o.
This condition can be satisfied even if (or because) there are no unicorns. The above
semantics also extends straightforwardly to weak quantifiers in predicative envi-
ronments, as in (6.80). In some languages the predicative uses are expressed with
distinct structural configurations or distinct lexical items, as in (6.81a) with the
use of cw7it ‘many’ without the (otherwise obligatory) additional determiner in

102
St’át’imcets (more on this below). Weak quantifiers can be systematically associated
with type ⟨et, et⟩ entries (alternatively, in some cases, ⟨e, t⟩) via a general lexical rule
or BE-style type shift (Partee 1986), as reflected in (6.82)–(6.83) with the cardinal
lexical entry in (6.79). (For simplicity I assume a non-assignment-quantificational
treatment of predicative uses in (6.82), and an off-the-shelf semantics for plurals
such as ‘girls’ in (6.83).)

(6.80) The winners are three (/many/∗ most) girls.

(6.81) a. [cw7it plísmen] [i=úxwal’=a]
many policemen det.pl=go.home=exis
‘The ones who went home were many policemen.’ (cardinal reading oblig-
atory)
b. [cw7it i=plísmen=a] úxwal’
many det.pl=policeman=exis go.home
‘Many policemen went home.’ (proportional reading obligatory)
(Matthewson 1998: 329, 137; St’át’imcets (Lillooet Salish))
(6.82) Predicative Weak Quantifier δ2
Jδ2 K = λPet .λze .λgg . JδK(λaa .P)(λaa .λye .λgg . y ≤ z)(g)
(6.83) Jthree2 K = λPet .λze .λgg . Jthree1 K(λaa .P)(λaa .λye .λgg . y(g) ≤ z(g))(g)
= λPet .λze .λgg . #x(g)[∃aa ∶ P(x)(g) ∧ x(g) ≤ z(g)] ≥ 3
= λPet .λze .λgg . #x(g)[P(x)(g) ∧ x(g) ≤ z(g)] ≥ 3
JgirlsK = λxe .λgg . every atomic part of x(g) is a girl
Jthree2 girlsK = Jthree2 K(JgirlsK)
= λze .λgg . #x(g)[every atomic part of x(g) is a girl ∧ x(g) ≤ z(g)] ≥ 3
J(6.80)K ≈ λgg . #x(g)[every atomic part of x(g) is a girl ∧ x(g) ≤ the-winners] ≥ 3
≈ the cardinality of the set of atoms o ∈ E s.t. [every atomic part of o is a
girl and o is a part of the-winners] is at least as great as 3
≈ the cardinality of the set of atoms o ∈ E s.t. [o is a girl and o is a part of
the-winners] is at least as great as 3
≈ ∣{o ∶ atom(o) ∧ girl(o) ∧ o ≤ the-winners}∣ ≥ 3

The particular extensions to weak quantifiers in (6.77)–(6.83) have been for pur-
poses of illustration. The aim has simply been to show how the proposed gen-
eral assignment-quantificational syntax/semantics in (6.68)–(6.69) can be adapted
to capture certain contrasts with weak quantifiers, given independently motivated

103
ideas in syntax about their predicative/modifier status. Hereafter I will put weak
quantifiers aside. I leave open what sort of reduced structure might be associated
with other types of noun phrases. Applications to contrasts regarding modal/temporal
independence, etc. may provide fruitful avenues for development.
The previous discussion examined how positing our choice-function-based con-
stituent in quantificational DPs can be motivated by independent work on notions
of specificity (broadly construed). There is also direct morphological and mor-
phosyntactic evidence for the extra structure. It is crosslinguistically common for
quantifier words to be diachronically related to or composed from a relative/wh
form or determiner/demonstrative — e.g., German jeder ‘every’ as je ‘ever’ + der
‘that’/‘the’/‘which’, Romanian fiecare ‘every’ as fie ‘be’ + care ‘which’. As noted above,
choice-function pronouns will be exploited in our analyses of relative/wh words and
certain demonstrative pro-forms; synchronic and diachronic connections between
definite articles and demonstratives, and among demonstrative, relative, and in-
terrogative forms, are crosslinguistically robust (Haspelmath 1995, Diessel 1999,
Lyons 1999, de Vries 2002, Bhat 2004, Thurgood & LaPolla 2017; see n. 52).
In many languages quantified expressions are even overtly realized via a quantifier
word and a distinct determiner phrase (Giusti 1991, 1992, 1997, Giusti & Leko
1995, Dimitrova-Vulchanova & Giusti 1996, Lecarme 1996, 2004, Matthew-
son 1998, 2008, 2013, Shimoyama 2001, Hanitramalala & Ileana 2012, Keenan
& Paperno 2012). For instance, the additional determiner is obligatory in the quan-
tified expressions in (6.84)–(6.86).⁶2
⁶2The evidence for the additional determiner is observed with weak and strong quantifiers alike
(cf. (6.77)–(6.79), (6.81)). As Matthewson shows, the overt determiner in St’át’imcets quantifier
phrases cannot be assimilated to a definite article on the model of English explicit partitives, e.g.
‘all of the women’. For instance, the relevant determiners can take a singular range, as in (i), and they
are unmarked for definiteness. St’át’imcets has a distinct structure corresponding to English-style
partitives, as in (ii) (see e.g. Matthewson 1998, 1999, 2013; cf. Gillon 2013 on Skwxwú7mesh).
(i) qvlqvl-ts-mín-lhkan [zí7zeg’ ta sqáycw-a áts’x-en-an].
badmouth.1sg each det man-exis see.1sg
‘Each man I saw, I swore at.’
(ii) kwán-lhkan [ku mulc] [lhél-ki zí7zeg’-a sk’úk’wm’it].
take.1sg det stick from-det.pl each-exis child
‘I took a stick from each of the children.’ (Matthewson 1998: 342; St’át’imcets)
Genitives in Somali can also be realized via a “construct state” construction, as in (iii).
(iii) kúlli ardáy-dii wáy gudbeen.
whole students-det+past decl.3pl succeeded
‘All (the) students succeeded.’ (Lecarme 1999b: ex. 35a; cf. 29–30; Somali)
Neither type of Somali quantifier (genitive) construction in (6.86)/(iii) involves a prepositional

104
(6.84) léxlex tákem i=smelhmúlhats=a
intelligent all det.pl=woman.pl=exis
‘All (the) women are intelligent.’
(Matthewson 2013: exs. 6–7; St’át’imcets (Lillooet Salish))
(6.85) Inona no andraikitry ny mpianatra tsirairay?
what foc responsibility.gen det student each
‘What is each student’s responsibility?’
(Keenan 2008: ex. 69; Malagasy (Austronesian))
(6.86) ardáy-dii kúlli-g-ood wáy gudbeen.
students-det+past whole-det+poss.3pl decl.3pl succeeded
(lit. ‘the students, their entirety’)
‘All (the) students succeeded.’ (Lecarme 1999b: ex. 35b; Somali)

In Somali each element of the more complex XP complement proposed in (6.69a)

is overt: the quantifier word, the choice-function pronoun via the determiner, the
X projection via the genitive, and the nominal predicate.⁶3 So the hypothesis goes,
what is explicit in the complex morphosyntax of quantifier phrases in (6.84)–(6.86),
and reflected in the morphology or diachronic sources of quantificational determin-
ers such as (e.g.) fiecare ‘every’ in Romanian, may be implicit in “strong” (presup-
positional, specific, D-linked) quantifier phrases in other languages.
I want to reiterate that these structural elements and configurations may be re-
alized in different ways in different constructions, both across languages and within
a particular language. We will return to the connections between the extended
nominal projections and genitives shortly. The posited choice-function elements
may also have different morphosyntactic properties. For instance, in the headed
relative clauses in §§6.1–6.2 the choice-function pronoun is realized via a [+wh]
relative determiner; in English quantifier phrases such as ‘every baby’ it is implicit;
in St’át’imcets quantifier phrases it is realized via an overt determiner; in Somali
quantifier phrases it is realized via a syntactically marked definite (see also §8.3).
Even in the case of Somali, syntactic Definite features needn’t correlate with se-
mantic/pragmatic definiteness, as Lecarme (1996, 1999b) is careful to point out.
Similarly, though Etxeberria & Giannakidou (2010) treat St’át’imcets determin-
element. Like Semitic construct state constructions, Somali quantified expressions of the (iii)-form
require adjacency without preposition or Case (Lecarme 1996, 1999b).
⁶3Like with the Somali genitive, the enclitic =a in St’át’imcets is obligatory in quantified expres-
sions. Since Matthewson (1998, 2001, 2013) treats the pro-clitic and enclitic portions as a circumfix
jointly constituting the Determiner, I leave open whether it should be regarded as specifying its own
projection such as X.

105
ers as definite, Matthewson (1999, 2013) argues that they can fail to exhibit no-
tional definiteness phenomena (uniqueness, maximality, familiarity); likewise for
the relevant determiners in Skwuwú7mesh (Squamish) Salish quantified expres-
sions, which Gillon (2013) argues are unmarked for definiteness/indefiniteness.
As we have seen, expressions represented with choice-function pronouns needn’t
be syntactically/semantically definite (§6.1.3) or even have intuitively specific read-
ings (§6.2.3). Likewise, for simplicity I have assumed that quantifier words project
a DP; yet the approach is compatible with treating them as occupying a distinctive
Quantifier category — say, analogous to how a modal might be treated as heading
ModP. Indeed, for all I have said, the complement of the Q(uantifier) could be
analyzed as a syntactic DP denoting an argument of type e. Treating the quantifier
as having an internal (implicit, possibly skolemized) domain restriction argument
(=P in (6.87)) may bring the parallels with modals even closer, as e.g. in the adapted
version of the LF from (6.69b) in (6.87) (cf. e.g. Giusti 1991, 1997, Giusti & Leko
1995, Matthewson 1998, 2001, 2013).

(6.87) QP: Q DP (alternative)

[ QP [ Q0⟨i,a⟩ every P-g] [ DP Fcf -tia baby]] laughed

Notably, the Q DP LF in (6.87) is essentially an assignment-quantificational ana-

logue of the analysis of St’át’imcets quantifier phrases proposed by Matthewson
(1999, 2001, 2013): Matthewson treats the quantifier words as categorially Q, taking
a DP complement of type e; and treats the nominal predicate as complement of
a determiner, represented as a choice-function pronoun. The convergence in our
accounts from such different directions is striking; it provides additional support for
a quantifier + choice-function pronoun analysis of quantifier phrases. Matthewson’s
specific account of the lexical/compositional semantics does differ in certain critical
respects. For instance, Matthewson assimilates the contribution of the determiner
qua choice-function pronoun with contextual domain restriction: quantifier words
are treated as quantifying over the atomic parts of a selected plural individual. The
comparison with modals and overt forms of restrictive modification provide in-
dependent bases for positing a distinct domain restriction argument. We will see
additional motivations for distinguishing the contribution of the choice-function
element from contextual domain restriction below in considering other forms of
restrictive modification. The account developed here is also compatible with al-
ternative analyses of the specific internal syntax of the relevant quantifier phrases.
The project of generalizing our assignment-quantificational account of quantifier

106
phrases with choice functions from §§6.1–6.2 can be motivated independently of a
specific Q DP syntax (more on the nominal XP projection shortly).⁶⁴
To recap, each element of the proposed generalized assignment-quantificational
account of quantifier phrases has independent precedents in syntax — notably, (i) the
additional internal structure in the complement of the quantifier, (ii) a specificity-
like element F in the extended functional projection of the NP; (iii) an internal
argument of the F head, which figures in determining the relevant domain; and
(iv) movement of the main operator from within the complement. Although ac-
counts differ in their particular implementations and theoretical interpretations of
the different elements, the convergence with such different empirical phenomena
in view is striking. Such precedents further motivate the generalized treatment of
quantifier phrases pursued here. On the flip side, the present assignment-variable-
based implementation may provide new applications and lines of inquiry on the
internal structure of nominals/quantifiers and interactions among the syntax, se-
mantics, and theories of discourse.
The previous discussion focused primarily on independent morphological, syn-
tactic, and semantic evidence for including a choice-function element in quanti-
fier phrases generally. I speculated that the overt forms of quantifier phrases in
other languages may provide evidence for additional internal structure. The use
of the genitive construction in Somali quantified expressions is particularly inter-
esting, as there are independent grounds for treating genitives as introducing a dis-
tinct nominal or functional projection. Building on these remarks, I will suggest
that connections with other forms of restrictive modification — notably, with gen-
itives — may further motivate positing the more complex complements such as in
(say) (6.68)–(6.69a) in contrast to (6.69b)/(6.87).
First, as suggested above, a generalized D CP/XP analysis affords a locus for
implicit domain restriction (n. 28), as reflected in (6.68) reproduced in (6.88). The
(possibly skolemized) domain variable representing the contextually relevant re-
striction in (6.88a) functions analogously to the restriction in a headed relative.

(6.88) Every baby laughed.

a. [ DP⟨2,a⟩ every⟨2,a⟩ [ CP F2cf -g2 baby … [ IP tie [P1et g1 ]]]]]
⁶⁴There may be good reasons for distinguishing the syntactic projection of quantifier words from
the projection of determiners such as those treated here as choice-function pronouns (e.g., certain
indefinites, relative words, the embedded determiners in (6.84)–(6.86); see also e.g. Giusti 1992,
1997, 2002). Yet for present purposes I will put aside Q DP analyses such as (6.87). I will continue
to speak simply of “determiner quantifiers” and “quantificational DPs,” though such talk may be
understood broadly to include approaches with distinct QP projection.

107
 b. ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(baby) ∧ g− (1et)(x(g))] x(g) laughed

Such an approach extends to overt restrictive NP modifiers: The nominal ‘girl’ in

(6.89) can be given its usual ⟨e, t⟩ meaning, without requiring a lifted type or compo-
sition rule such as Predicate Modification (Partee 1995a, Heim & Kratzer 1998).⁶⁵

(6.89) Every girl baby laughed.

a. [DP⟨2,a⟩ every⟨2,a⟩ [CP F2cf -g2 baby … [IP tie girl]]]] laughed
b. ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(baby) ∧ x(g) is a girl] x(g) laughed

Parallel moves are available for a D+XP implementation. The (possibly implicit)
restrictive modifier is represented in the specifier position of the nominal XP pro-
⁶⁵I will ignore any LF differences between overtly pre- and post-nominal restrictive modification.
There may of course be reasons for introducing an intersective modification rule — e.g., for DPs with
multiple modifiers, as in (i). Yet one way of maintaining the simple low-type+FA implementation
would be to follow e.g. Cinque 1995 and treat apparent modifier iteration as asyndetic coordination,
e.g. as in (ii). Support for such a move may come from languages in which the coordination with
multiple restrictive modifiers is overtly realized, as in (iii).
(i) Every freshman girl student aced the math test.
(ii) a. [ DP every⟨2,a⟩ [ CP Fcf -g2 student … [ IP tie [freshman [conj girl]]]]]
b. [ DP every⟨2,a⟩ [ XP [freshman [conj girl]] [ X X [Fcf -t2a student]]]]
(iii) a. Xaashí-díi yarayd ée caddayd
paper-def.past small.past and white.past
‘the small white sheet of paper’ (lit. ‘the sheet of paper small and white’)
b. búugga xisáabta ee ardayáddan
book.def mathematics.def and student.def.dem
‘this student’s mathematics book’ (lit. ‘the book of mathematics and of this student’)
c. búug yar oo Máryan lédahay
book small and Maryan has(it)
‘a small book of Maryan’ (lit. ‘a book small and Mary possesses it’)
(Lecarme 2008: ex. 6, 1996: ex. 18b, 1999b: ex. 30c; Somali)
In a D+XP implementation the effects of intersective modification could also be encoded in the syn-
tax/semantics of the nominal XP projection, whether by asyndetic coordination in a unique specifier
position (parallel to the suggestion above), or by an iteration of specifier positions. On the latter
approach, the iterated restrictive modifiers would be semantically interpreted via a correspondingly
revised type-flexible lexical entry for the X head, as in e.g. (iv).
(iv) Polymorphic X lexical entry (alternative)
JXK = λxe .λP1et . . . λPnet .λye .λgg . y(g) = x(g) ∧ P1 (y)(g) ∧ ⋯ ∧ Pn (y)(g)
Allowing iterated specifiers with lexical categories is independently attested in theories of phrase
structure (cf. Fukui & Speas 1986, Cinque 1999).

108
jection in (6.90a); the lexical entry for X from (6.73), reproduced in (6.90b), derives
the equivalent interpretations (see nn. 58, 65, 66).

(6.90) a. [ DP every⟨2,a⟩ [ XP girl/P1et -g1 [ X X [F2cf -t2a baby]]]]

b. JXK = λxe .λPet .λye .λgg . y(g) = x(g) ∧ P(y)(g)
JXPK = λye .λgg . y(g) = g(2a)(2cf)(baby) ∧ y(g) is a girl
JSK ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(baby) ∧ x(g) is a girl] x(g) laughed

Representing modifiers in specifier positions is well attested in various types of the-

ories of lexical and functional heads (e.g., Fukui & Speas 1986, Giusti 1991, 1992,
2002, Holmberg 1993, Longobardi 1994, Szabolcsi 1994, Cinque 1995, 1999,
Rizzi 1997).
Before moving on I would like to briefly consider one perhaps unexpected moti-
vation for a D+XP analysis, from genitives. Extending the syntax/semantics for rel-
ative clauses from §6.1 in this way may help capture connections between quantifi-
cational determiners, specificity, and genitive modification, often ignored in com-
positional semantic accounts.⁶⁶
⁶⁶For one additional motivation, a D+XP analysis may also be in a better position to capture
contrasts between relative and non-relative modification. For instance, Somali, which has a rich
overt nominal tense morphology, structurally distinguishes between inflected adjectival/nominal
modification and modification by a restrictive relative clause. Whereas modifiers must agree with
the head noun in features such as definiteness, gender, tense, the tense features in a relative clause
are independent and semantically interpreted (cf. Culy 1990 on independency properties with
circumnominal relatives):
(i) a. ardayád soomaalí ah/∗ ahayd [CPrel oo telefóon kúu soo dirtay]
student Somali is/∗ was.past and telephone you-to towards sent.past
‘A Somali student who phoned you’
b. ardayád-dii soomaalí-da ahayd [CPrel oo telefóon kúu soo dirtay]
student-def.past Somali-def was.past and telephone you-to towards sent.past
‘The Somali student who phoned you’
(Lecarme 2004: ex. 20; structure markings in original)
The tense domain of the head noun includes adjectival/nominal modifiers but not the relative clause
(Lecarme 1996, 1999b, 2004). Such properties of tense agreement, overtly marked in Somali,
arguably extend to worlds as well: A “mixed” de re/de dicto reading is perhaps more easily available
with the embedded restrictive relative in (iii) than with the modifier in (ii).
(ii) Bert thinks every freshman girl is smart.
a. De dicto ≈ there are no freshman girls but Bert thinks there are; he thinks all of them are
smart (e.g., Bert thinks the actual sophomore boys are the freshman girls)
b. De re ≈ let G be the set of freshmen and G′ be the set of girls; Bert thinks every o ∈ G ∩ G′ is
smart (compatible with Bert thinking some of the o’s are sophomores or boys)
c. #Mixed ≈ let G be the set of actual freshmen; Bert thinks every o′ ∈ G s.t. o′ is a girl is

109
 It is common following Partee 1983 to distinguish modifier readings vs. ar-
gument readings of genitives X’s Y, as in (6.91)–(6.92). Informally put, in argument
smart (compatible with some of the Gs being boys, and Bert thinking some of them are
sophomores)
(iii) [Context: Stan doesn’t realize that the Smiths are Russian spies. He thinks they live across the
street from him, but the house is actually a Russian safe house.]
??Mixed ≈ Stan thinksi every spy@ who livesi across the street from him is a loyal American.
The tense/world linking between the head noun and its (non-relative) modifiers might be captured
syntactically via agreement. Yet one way of deriving the linking in the compositional semantics would
be to incorporate intensionality into our D+XP syntax along the lines in (iv) (using ‘cf∗ ’ for the type
of an intensional choice function, mapping properties to individual concepts):
(iv) ‘Every (girl) baby laughed.’
a. [ DP every⟨2,a⟩ [ XP girl/P1set -g1 [ X [X w1 -g1 ] [F2cf∗ -t2a baby]]]]
b. JXK = λws .λx′se .λP′⟨s,et⟩ .λye .λgg . y(g) = x′ (w)(g) ∧ P′ (w)(y)(g)
JXPK = λye .λgg . y(g) = g(2a)(2cf∗ )(baby)(g(1a)(1s)) ∧ y(g) is a girl in g(1a)(1s)
JSK ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf∗ )(baby)(g− (1s)) ∧ x(g) is a girl in g− (1s)]
x(g) laughed in @(g− )
Merging a relative clause to the head nominal could yield an alternative structure for headed restric-
tive relatives, as in (v); note that the world of ‘likes’ in the relative clause is potentially independent
of the (obligatorily linked) worlds of the head nominal ‘baby’ and modifier ‘girl’.
(v) DP: D XP + individual-denoting CPrel (alternative)
a. ‘Every girl baby which Alice likes laughed.’
⟨2,e⟩
b. [ DP every⟨2,a⟩ [ XP girl [ X [X w1 -g1 ] [ CP [wh2cf∗ -t2a baby] Crel Alice likes-w1 -g1 t2e ]]]]
′ ′
c. JCrel K = λPet .λxse .λws .λgg . ιy(g) s.t. y(g) = x (w)(g) ∧ P(y)(g)
JSK ≈ λgg . [∀xe ∃aa ∶ x(g) is a girl in g− (1s) ∧ x(g) = ιy(g) s.t. Alice likes y(g) in g− (1s)
∧ y(g) = a(g)(2cf∗ )(baby)(g− (1s))] x(g) laughed in @(g− )
≈ λgg . [∀xe ∃aa ∶ x(g) is a girl in g− (1s) ∧ Alice likes x(g) in g− (1s)
∧ x(g) = a(g)(2cf∗ )(baby)(g− (1s))] x(g) laughed in @(g− )
Interestingly, the LF in (v-b) is nearly equivalent to the D+XP structure for headed restrictive relatives
proposed for independent syntactic reasons in Bhatt 2002 (see also Bianchi 2000). (Bhatt treats
the relative phrase as originating in the gap position, and the main NP in SpecXP as reconstructing
at LF; see §6.1.2 on Bhatt’s account of the LF transformations with Trace Conversion.)
(vi) [ DP D [ XP NP [ X X [ CP [ DP rel tNP ]i [ C C [ IP …ti …]]]]]] (Bhatt 2002: 84; adapted)
Treating the relative CP as denoting an individual (concept) would cohere with the prominent idea
in syntax that the role of functional D and C categories is to turn a predicate into an expression
that can serve as an argument (Higginbotham 1985, Szabolcsi 1987, 1989, Longobardi 1994,
Siloni 1997, Pesetsky & Torrego 2001; cf. Lecarme 1999a). Comparisons with a unified D+XP
syntax/semantics such as in (iv)–(v) warrant more careful investigation. It would be interesting to
compare the unified assignment-variable analyses of relativization, conditionals, and interrogatives
proposed in §§7–8 with alternatives drawing on the approach to the relative complementizer in (v).

110
(“inherent R”) readings the relevant relation for interpreting the genitive is linguisti-
cally supplied by the head noun, whereas in modifier (“free R”) readings the relevant
relation is contextually supplied (Partee 1978, 1983, Higginbotham 1983).⁶⁷

(6.91) “Free R” (restrictive modifier) reading

a. Alice’s cat
b. ≈ “the o s.t. o is a cat and o bears Rc (e.g. is a pet of ) to Alice”
(6.92) “Inherent R” (argument) reading
a. Alice’s mother
b. ≈ “the o s.t. o bears the is a mother of relation to Alice”

The approach to restrictive modification in (6.88)–(6.90) affords a natural struc-

tural basis for the apparent contrast in readings — namely, in whether the posses-
sor+relation (n. 67) is represented in the position suggested above for restrictive
modifiers, with the relevant relation supplied by an implicit pronoun, as in free R
readings ((6.93)); or in the position of the nominal head, with the relevant rela-
tion supplied by the overt nominal, as in inherent R readings ((6.94)). Suppose
first that we treat the genitive morpheme ’s as the head determiner and as having
the same assignment-quantificational semantics as ‘the’ (we will reconsider these
assumptions shortly):⁶⁸

(6.93) D+CP “free R” (modifier) genitive LF

a. Bert’s TV is broken.
b. [ DP⟨2,a⟩ ’s⟨2,a⟩ [ CP F2cf -g2 TV [[ C⟨1,e⟩ C t2a ] t1e R1eet -g1 Bert]]]
c. ≈ λgg . [ιxe (g)∃aa ∶ x(g) = a(g)(2cf)(TV) ∧ g− (1eet)(B)(x(g))] x(g) is broken
d. ≈ “the unique o s.t. there is a choice function that selects o from among
the TVs, and o bears the relevant relation R (=gc (1eet)) to Bert, is broken”
⁶⁷I will use ‘genitive DP’ for the complex phrase X’s Y; I use ‘nominal head’ informally for Y;
and I use ‘possessor DP’ for X, though the relevant relation needn’t be possession. I sometimes use
‘structural possessor’ or scare quotes around ‘possessor’ to highlight this point.
⁶⁸For concreteness I assume that the “possessor” DP and relation originate in the complement
of ’s. The possessor would raise to SpecDP for syntactic or morphological reasons, and reconstruct
at LF unless excluded for type reasons (more on which in §6.3.3); cf. n. 69. Uniform analyses of
free/inherent R readings are of course also possible — e.g., reducing the inherent R reading in (6.92) to
a free R reading in which the supplied relation is identified with the relation associated with ‘mother’,
or reducing the free R reading in (6.91) to an inherent R reading in which ‘cat’ is coerced to having
the relevant relational meaning (cf. Jensen & Vikner 1994). See Partee & Borschev 2003 for
critical discussion, in the context of different types of genitive constructions in different languages;
they conclude that a uniform approach of either type is incorrect for English.

111
(6.94) D+CP “inherent R” (argument) genitive LF
a. Bert’s hand is broken.
b. [ DP⟨2,a⟩ ’s⟨2,a⟩ [ CP F2cf -g2 [hand Bert] [[ C⟨1,e⟩ C t2a ] t1e P1et -g1 ]]]
c. ≈ λgg . [ιxe (g)∃aa ∶ x(g) = a(g)(2cf)(hand-of-Bert) ∧ g− (1et)(x(g))]
x(g) is broken
d. ≈ “the unique o s.t. there is a choice function that selects o from among the
hands of Bert, and o is in the relevant set of entities (=gc (1et)), is broken”

The domain restriction pronoun [P1et g1 ] captures how the inherent R reading can be
felicitous even if Bert has more than one hand. Assuming that the supplied value for
the pronoun restricts the domain to a set that includes only one of Bert’s hands, the
uniqueness condition is satisfied. Likewise for the felicitous use of ‘Alice’s brother’
in (6.95), even if Alice has older brothers; and for the free and inherent R readings
of ‘Bert’s hand’ in (6.96), even if Bert is ambidextrous and he and the other medical
students have been keeping individual inventories of their cadavers:

(6.95) [Context: Alice has two older brothers away at college, and one younger
brother. She wants to go to the movies with her friends, but since her parents
are away, she has to take her younger brother with them. Later one of their
other friends who couldn’t make it to the movie asks how it was. You say:]
Be glad you didn’t come. Alice’s brother was pestering us the whole time.
(6.96) [Context: we’re medical students practicing on cadavers]
a. Free R: Bert’s hand is broken. It’s obvious, but I bet he’ll miss it and get
the diagnosis wrong.
b. Inherent R: Bert’s hand is broken. There’s no way he’s going to be able to
pass the practice exam with his cast on.

The relevant value for the relation pronoun in (6.96a) maps Bert to the particular
(cadaver) hand he is currently practicing on.
The analyses in (6.93)–(6.94) maintain a uniform overall syntax for the genitive
construction, and a uniform semantics for the genitive morpheme ’s. The intuitive
contrast between free vs. inherent R readings of prenominal genitives needn’t require
positing a lexical ambiguity in the genitive morpheme, type-shifting, or coercion
(as in e.g. Partee 1983, Jensen & Vikner 1994, Vikner & Jensen 2002, Partee
& Borschev 2003). The distinction between genitives as arguments vs. restrictive
modifiers can be captured in terms of general features of the proposed structure for
quantificational DPs.
The D+CP implementation in (6.93)–(6.94) captures the specificity of prenom-

112
inal genitives in the semantics of the genitive morpheme, treated as the head deter-
miner. Such an analysis may be awkward for postnominal genitives where the Y of
X’s phrase is itself the argument of a determiner, as in (6.97).

(6.97) every/a/the/which cat/brother of Alice’s

A more promising approach would be to distinguish the contribution of the genitive

morpheme from specificity effects. Following Partee 1983, the prenominal genitive
may be treated as a postnominal genitive with an implicit definite determiner — i.e.,
A’s B ≈ (the) B of A’s. Our D+XP syntax affords a natural implementation: the genitive
morpheme can be understood as the overt realization of the nominal X head:

(6.98) [ DP D0 [ XP [ X0 ’s] ZP]]

A D+XP implementation along these lines yields an analogous diagnosis of free R

vs. inherent R readings: Parallel to (6.93)–(6.94), the free R reading in (6.100) is
represented with the possessor+relation in the position suggested in (6.90) for re-
strictive modifiers; and the inherent R reading in (6.101) is represented with the
possessor+relation in the position of the head nominal. Pronunciation differences
aside, the prenominal and postnominal genitives in (6.100a)–(6.101a) can be given
a common structure for purposes of interpretation. In both cases the genitive mor-
pheme is the head of the nominal XP complement of the (possibly implicit) deter-
miner.⁶⁹ An alternative semantics for ’s is in (6.99) (ignoring intensionality, def the
implicit determiner in prenominal genitives).

(6.99) J’sK = λxe .λPet .λye .λgg . y(g) = x(g) ∧ P(y)(g)

(6.100) D+XP free R prenominal/postnominal genitive LFs (alternative)
a. Alice’s cat / every cat of Alice’s (danced)
b. [ DP⟨2,a⟩ def/every⟨2,a⟩ [ XP [R1eet -g1 Alice] [ X ’s [ NP F2cf -t2a cat]]]]
c. JXPK ≈ λye .λgg . y(g) = g(2a)(2cf)(cat) ∧ g(1a)(1eet)(A)(y(g))
JSK ≈ λgg . [ιxe ∃aa ∶ x(g) = a(g)(2cf)(cat) ∧ g− (1eet)(A)(x(g))] x(g) danced
(6.101) D+XP inherent R prenominal/postnominal genitive LFs (alternative)
⁶⁹There are various ways in which the pronounced forms of inherent R readings (in English)
might be derived in the narrow syntax and morpho-phonolological components. The DP in the NP
complement could raise to SpecXP; the entire NP could raise, where the (string) adjacent DP and ’s
fuse (“brother (of) Alice’s”) and the N head reverts to its base position (“Alice’s brother”); ’s and the
DP could fuse via Local Dislocation, directly yielding [ DP [Alice+’s] brother]; etc. (see e.g. Fukui &
Speas 1986, Embick & Noyer 2001, Nunes 2004). We can put aside such morphosyntactic details.

113
 a. Alice’s brother / every brother of Alice’s (danced)
b. [ DP⟨2,a⟩ def/every⟨2,a⟩ [ XP [P1et g1 ] [ X ’s [ NP F2cf t2a [brother Alice]]]]]
c. JXPK ≈ λye .λgg . y(g) = g(2a)(2cf)(bro-of-A) ∧ g(1a)(1et)(y(g))
JSK ≈ λgg . [ιxe ∃aa ∶ x(g) = a(g)(2cf)(bro-of-A) ∧ g− (1et)(x(g))] x(g) danced

The semantic values in (6.100c)–(6.101c) derived for the prenominal genitives are
correctly equivalent to those in (6.93)–(6.94). No additional resources are required
for the semantics of postnominal genitives.
Observe that the semantics for the genitive morpheme in (6.99) is equivalent to
the semantics for X in (6.90) (cf. Partee 1983). An attractive hypothesis is that the
posited nominal XP projection just is the projection headed by the genitive mor-
pheme in genitive constructions (cf. Hiraiwa 2005). The genitive morpheme ’s, on
this line, is an overt (semantically equivalent, +EPP) analogue of the implicit X in
(6.69)/(6.90), licensing (indeed morphosyntactically requiring) the overt specifier
characteristic to genitives. Understanding the nominal XP projection in this way
affords an independent basis for the posited additional layer of structure in ordi-
nary (non-genitive, non-relative) quantificational DPs such as in (6.69)/(6.90). The
nominal X category in the DP might even be understood as analogous to “little” v
in a CP, furthering the parallels in the nominal and clausal domains (cf. Lecarme
2004, 2008, Hiraiwa 2005). As Partee & Borschev put it in the case of geni-
tives, “an argument genitive is most like a direct object, an ‘internal argument’…
A ‘possessor’ genitive, on the other hand, is most subject-like, agent-like, less like
an internal argument, more independent” (2003: 100). The LFs in (6.100)–(6.101)
reflect this: Whereas in inherent R readings the structural “possessor” originates as
the object argument of the relational noun in the main NP, in free R readings the
possessor originates in Spec of X. Though I have been ignoring projections such as
vP for introducing external arguments of verbs, perhaps a more refined syntax for
the verbal domain could afford a parallel assignment-variable-based treatment of
the LFs as well.
As in the verbal case, one may expect different features to be associated with
different X heads — syntactic features (genitive case, EPP properties), as well as se-
mantic features such as thematic roles. Unlike inherent R readings, free R readings
are often associated with a “default preference… for a genitive relation in the family
of ‘owns’, ‘possesses’, ‘controls’ ” (Partee & Borschev 2003: 70). In some languages,
modifier genitives obligatorily receive possession readings, as in Russian prenominal
possessives such as (6.102). Although the range of relations under the heading of
“possession” can be broad (Heine 1997), it is more constrained than the range of
relations potentially associated with genitives generally.

114
(6.102) Petin ubijca
Petja-poss-m.sg murderer-m.sg
a. #‘Petja’s murderer’ / ‘the murderer of Petja’
b. ‘a murderer Petja has hired’ (Partee & Borschev 2003: ex. 25b)

The LFs in (6.100)–(6.101) suggest a structural diagnosis: The “possessor” in free R

readings is directly generated in the specifier position of the genitive morpheme —
drawing on the analogy with little v, a position arguably analogous to the position of
a clausal subject. The argument projected by the genitive morpheme could thus be
understood as assigning a distinctive possessor θ-role — optionally in English, obli-
gatorily in certain genitive constructions in other languages. Though the structural
possessor in inherent R readings may receive its (genitive) case assignment from the
X head, its θ-role (if any) may be assigned in its base position by the relational noun.
Above we noted various diachronic and synchronic connections between quan-
tificational DPs and specificity/definiteness. The additional crosslinguistic connec-
tions with genitives may provide additional support for the D+XP implementation.
As we saw in (6.86), reproduced below, in languages such as Somali individual quan-
tification is overtly expressed via a genitive construction.

(6.103) ardáy-dii kúlli-g-ood wáy gudbeen.

students-def.past whole-def+poss.3pl decl.3pl succeeded
lit. ‘the students, their entirety’
‘All (the) students succeeded.’ (Lecarme 1999b: ex. 35b; Somali)

In light of her extensive work on nominal tense in genitives, Lecarme even concludes
that “genitive constructions universally involve a form of quantification” (2004: 17).⁷⁰
⁷⁰Lecarme notes that in genitives with kinship relations, body-part nouns, and quantifiers, tense
morphemes appear on the genitive phrase (e.g., ‘the students’ in (6.103)) rather than on the head
nominal as with other genitives, as in (i). An analogous pattern is observed in other languages with
nominal tense, as in (ii).
(i) ardáy-da macállin-k-ood-ii wúu baxay.
students-def teacher-def-poss.3p-past decl.3ms left
‘The students’ teacher left.’ (Lecarme 1999b: ex. 28b; Somali)
(ii) a. ’moj-et bu’bajko ’beše mih’tar
my-def.past father was mihtar
‘My father was mihtar (in charge of the community’s relations with outsiders).’
b. ka’de je tšy’ljak-an ti
where is man-def.dist 2sg.dat
‘Where is your husband (distal)?’ (Adamou 2011: ex. 14; Pomak (Slavic))
For genitives such as in (i) with non-relational nouns, Lecarme posits an independent “syntactic

115
rule that serves to delete the [+past] tense morpheme on a locally dependent DP” (1999b: 12). For
quantified phrases and kinship/body-part nouns Lecarme suggests distinct semantic explanations:
“Universal quantifiers in natural language quantify over contextually given sets of individuals… [T]he
genitive complement refers to the ‘familiar’ set… Such examples, then, provide further evidence that
nominal tense is linked to the notion of specificity (1999b: 15–16). Kinship/body-part nouns “are
not compatible with tense features, even in the syntactic contexts where they are normally required,”
on the ground that they “lack autonomous reference” and express “atemporal relations” (1996: 9).
One might wonder why relations such as being a part of, being a sister of, being a hand of, etc. would
necessarily be conceptualized as failing to hold at a time. Independent worries aside, such informal
semantic explanations are surprising given Lecarme’s emphasis on the “purely structural meaning”
(2004: 15) of genitive case.
Lecarme doesn’t distinguish free R (modifier) and inherent R (argument) readings of genitives. The
contrasting syntax in (6.100)–(6.101) may afford a more unified “purely structural” (Lecarme 2004)
account of the distribution of tense morphemes in genitives with quantificational/kinship/body-part
nouns. Very roughly, for purposes of illustration: Following Lecarme (1996, 1999b, 2004), suppose
that tense features are realized on a nominal originating in the NP, which, in Somali, raises in overt
syntax to adjoin to D. In free R readings such as (i) the “possessor” ‘the students’ originates as a
modifier external to the main NP; hence the element undergoing N-to-D raising (in Somali) and
thereby receiving the main tense feature is the head noun ‘teacher’, which projects its tense feature
to the DP. Given that interpretable tense may only be realized once per DP, the interpretable tense
feature on the possessor-modifier (if it has one) is deleted, as Lecarme suggests occurs in the general
case. (For instance, the tense feature on the possessor DP raising overtly to Spec could be treated
as deleted under m-command by the D-T-N complex, or via a rule deleting any interpretable tense
features within a DP distinct from the feature on the DP.) The “possessor” originates in the NP only
in inherent R readings, as the complement of the relational noun. Given that the “possessor” must
raise first for syntactic reasons — e.g., to check the relevant [+EPP] feature(s) or license the anaphoric
genitive pronoun — in (6.103) the tense feature in T can be deleted under c-command by the tensed
DP ‘the students’ when in its pronounced position in SpecDP, e.g. via Feature Deletion (von Stechow
2003). Hence there is no tense feature for the relational noun to combine with when it raises to D.
The relevant assumptions for deriving the contrasting distribution of tense features would thus be that
(a) N-to-D movement occurs before the movement of modifiers; (b) one interpretable tense feature
is generally allowed per DP; and (c) Feature Deletion. Such an approach would seem to predict that
body-part/kinship nouns in the relevant languages should be able to bear tense features in free R
readings such as (6.96a). This prediction appears to be borne out, as reflected in (iii) for Pomak.
(iii) a. ören’ciete ra’kota
student-def.past hand-def.past
‘the student’s hand (was broken)’ (free R reading, (6.96a): tense on head body-part noun)
b. ören’civata ’roka
student-def.past hand
‘the student’s hand (was broken)’ (inherent R reading, (6.96b): tense on “possessor”)
(Evangelia Adamou, p.c.; Pomak)
Examples such as (iii-a) provide prima facie support for a structural-ambiguity account such as
suggested above, as opposed to a broadly lexical account like Lecarme’s which treats the relevant
kinship/body-part nouns as essentially atemporal. Details aside, the proposed syntax for free R vs.

116
 Let’s recap. This section has examined several ways of extending the assignment-
quantificational treatment of determiner quantifiers from §§6.1–6.2 to quantifier
phrases with non-relative complements. I argued that a generalized D+CP/XP ac-
count is motivated by diachronic, morphological, syntactic, and semantic connec-
tions between individual quantifier phrases and expressions often associated with
notions of “specificity” (broadly construed); and helps capture a range of indepen-
dent linguistic phenomena, including implicit domain restriction, non-relative re-
strictive modification, the distinction between free R (modifier) and inherent R read-
ings of genitives. Although the assignment-quantificational semantics is compatible
with alternative analyses of the internal syntax, I speculated that a D+XP implemen-
tation for non-relative quantificational DPs may have advantages over a D+CP im-
plementation. The posited nominal XP projection is independently motivated by ac-
counts of genitive constructions and parallels in the verbal domain. The D+XP anal-
yses in (6.100)–(6.101) yield a common structure in the interpretation of prenomi-
nal and postnominal genitives, and capture the distinction between free R and inher-
ent R readings in terms of a more general treatment of restrictive modification, as in
(6.90) (also nn. 65, 66). Such syntactic and semantic features may be of broader the-
oretical interest independent of the particular assignment-variable implementation
developed here. It is worth examining to what extent the proposals might be adapted
to other types of quantificational, genitive, and specific/non-specific expressions,
and in English and in other languages. The issues regarding the internal syntax, and
comparisons among different notions associated with “specificity” warrant more
careful independent investigation. (We will return to genitives in connection with
donkey anaphora and weak crossover in §6.3.3.)
The syntactic speculations in this section, though not essential to the overall
account, haven’t been a mere detour; they highlight an important methodologi-
cal point (cf. §§1, 3.1, 5). It is critical for developing our syntax/semantics with
assignment variables that any binder indices and sources of linguistic shifting be
derived from independently motivated syntactic structures. Otherwise the pro-
posed compositional derivations will be little more than an ad hoc technical exercise.
The independent features of the D+CP and D+XP analyses provide further motiva-
tion for the LFs used in capturing the shifting data which constitute the project’s
primary focus. The remainder of the section applies our generalized assignment-
variable syntax/semantics of quantificational DPs to several additional phenomena
inherent R readings may afford a structural basis for the distribution of tense features in genitives
in the relevant languages, without requiring questionable semantic or metaphysical assumptions
about the (a)temporality of certain relations or fundamental notions of “alienable” vs. “inalienable”
possession (Lecarme 1996, 1999b, 2004).

117
with pronominal anaphora and binding. For uniformity with our accounts in §6.2
I assume the general D+CP implementation; the discussion may be adapted for the
alternative D+XP (or Q+DP) implementations.
6.3.2 Quantifier phrases and bound pronouns
Accounts of semantic binding following e.g. Heim & Kratzer 1998 require all bound
readings to be captured via A-movement such as QR (cf. May 1985, Kayne 2002).
Although the subject-position quantifier in examples such as (6.104) needn’t raise
for type reasons, it still must undergo QR in order to generate the binder index which
binds the pronoun.⁷1

(6.104) Pronoun-binding via A-movement (cf. Heim & Kratzer 1998: 262-264)
a. Every cat likes its owner.
LF ≈ [ S every cat [2 [ S t2 likes the-owner-of-o2 ]]]
b. Every cat thinks that it is smarter than its owner.
LF ≈ [ S every cat [2 [ S t2 thinks [ S it2 is smarter than the-owner-of-it2 ]]]]

I haven’t assumed that individual pronouns constitute a unified syntactic or se-

mantic class crosslinguistically, or that all instances of linguistic anaphora are to
be analyzed as copies of a linguistic antecedent (§§6.2.1, 8.3.1; nn. 51, 73). Yet the
assignment-binding introduced by the determiner may also capture other bound
readings of pronouns without needing to posit such non-type-driven QR of the
subject quantifier (cf. §4.3). The pronoun may be represented as a copy of the an-
tecedent [Fcf -g NP] internal to the quantificational subject DP, roughly as in (6.105)
for (6.104a) (again ignoring worlds).

(6.105) a. [[DP⟨2,a⟩ every⟨2,a⟩ [CP F2cf -g2 cat …]] [VP likes [the-owner-of-it2 -g2 cat]]]
b. JSK ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(cat) . . . ] x(g) likes the-owner-of-
a(g)(2cf)(cat)

Additional support for such an approach may come from languages in which a bound
pronoun can be overtly realized as a copy of its antecedent, as in (6.106).

(6.106) a. Aajarni khit waa puak rua chob aajarni .

teacher think that all we like teacher
‘The teacheri thinks we like himi .’ (Lee 2003; Thai)
⁷1Following Heim & Kratzer’s account of movement, the binder index occupies its own node.
They posit that the embedded subject pronoun also QRs; I ignore this additional movement. Assume
a simplified analysis of the possessive pronoun; more on this shortly.

118
 b. A w-nalààa’z bxuuhahzi g-uhcnèe Lia Paamm bxuuhahzi
already remembered priest helped Pam priest
‘The priesti remembered that Pam helped himi .’ (Lee 2003; Zapotec)

Interestingly, bound copies with morphologically complex antecedents are generally

excluded in such languages for independent morphosyntactic or phonological rea-
sons (Lee 2003, Mortensen 2003, Nunes 2004, Boeckx et al. 2007), as in (6.107),
and an alternative device must be used such as a (possibly covert) simple pronoun.

(6.107) a. *Thuk khoni bok waa rua chob thuk khoni .

every person say that we like every person
‘Everyonei said we like themi .’
b. *Yra’ta’ zhyàa’pi r-ralloh nsinnyi’cy yra’ta’ zhyàa’pi .
every girl think smart every girl
‘Every girli thinks shei is smart.’ (Lee 2003)

Indeed this restriction at PF is mirrored at LF on the present analysis: the bound

pronoun in examples such as (6.105)⁷2 is represented at LF as a reduced copy of the
c-commanding quantificational DP. The quantificational subject is represented with
⁷2I include the qualification to make room for alternative analyses of names and apparent binding
of pronouns by names. As discussed in §4.4 simple names pattern with examples such as (6.106) in
explicit copy languages.
(i) a. Povi yeej qhuas Povi .
Pao always praise Pao
‘Pao always praises himself.’
b. Povi has tas Maiv nyam Povi .
Pao say that May like Pao
‘Paoi said that May likes himi .’ (Mortensen 2003; Hmong)
In accounts of names as variables or constants, a pronoun bound by a name may be represented as an
identical complete copy at LF. In accounts of names as predicates or definites (as in §4.4), the pronoun
may be represented as a reduced copy parallel to the cases with quantificational DPs. I continue to
focus on examples with quantificational or indefinite antecedents.

119
the more richly articulated internal structure, in contrast to the reduced syntax of
the bound pronoun.⁷3
A treatment of bound pronouns along these lines may shed light on contrasts
⁷3There are of course independent differences in the degree of internal syntactic complexity
among different types of pronouns and (non-)quantificational DPs across languages (e.g. Giusti
1991, 1992, 2002, Cardinaletti & Starke 1999, Bernstein 2001, Matthewson 2001, 2013,
Déchaine & Wiltschko 2002, 2015, 2017b, Bhat 2004, Lecarme 2004, 2012, von Fintel &
Matthewson 2008, Gil et al. 2013, Gillon 2013, Grosz & Patel-Grosz 2016, Patel-Grosz
& Grosz 2017, Jenks 2018; see also n. 51). For instance, we have already observed as much in the
distinction between D CP quantifier phrases and indefinites which take an NP complement. Likewise
capturing the anaphoric potential of bound pronouns and donkey pronouns in the above ways is
compatible with there being other syntactic differences among them (§6.2.1). For instance, given
our purposes I have left open how exactly the non-movement copies (informally understood) enter
the syntactic derivation — e.g., whether they are the product of some sequence of syntactic operations
involving Copy, or the lexical material is duplicated in the numeration. Different types of anaphoric
pronouns and expressions may differ in this respect, both crosslinguistically and within a particular
language. On one approach (Nunes 2004, Bošković & Nunes 2007, Kandybowicz 2007), multiple
Copies (products of Copy) can be pronounced only if some Copy can be morpho-phonologically
fused with another element, thereby avoiding a violation of the Linear Correspondence Axiom
(Kayne 1994); roughly put, if multiple identical Copies are pronounced, the lexical items in the
numeration won’t be able to be consistently linearized, since some items are assigned to more than
one position (contrast e.g. Kandybowicz 2008, Nasu 2010 for non-fusion-based approaches). So,
insofar as complex elements are less susceptible to morpho-phonological reanalyzation, structures
with multiple pronounced Copies are generally excluded at PF. Interestingly, whereas morpholog-
ically complex bound pronoun copies are (to my knowledge) unattested (see also (6.106)–(6.107)
above), individual correlates in correlative constructions can be overtly realized as syntactically and
morphologically complex (§7.7); complex donkey-anaphoric expressions are also possible in some
languages ((6.41), below; cf. (6.46)).
(i) a. *Ob tug dlevi yeej tum ob tug dlevi .
two clf dog always bite two clf dog
‘The two dogs bite themselves/each other.’ (Mortensen 2003; Hmong complex reflexive)
b. Tug twg i pum (proj ) los tug twg i yeej nyam (proj ).
clf which see top clf which always like
‘Whoever sees him will surely like him.’ (Mortensen 2003; Hmong donkey sentence)
c. [jo (larkii) kharii hai] vo laRkii lambii hai.
REL (girl) standing is DEM girl tall is
‘Which girl is standing, that girl is tall.’ (Srivastav 1991; Hindi complex correlate)
One way of reflecting these contrasts might be to distinguish the representations of bound pronouns
(like reflexives) as products of a Copy operation. It would be interesting to examine whether
there may be crosslinguistic differences in this respect, and to what extent such differences may
come apart from differences such as in e.g. morphology and possibilities for morpho-phonological
reanalysis, zero anaphora, anaphoric devices, etc. (We will return to these issues and comparisons
with correlatives in §§7–8; for broader general discussion see also Hornstein 2001, Elbourne 2005,
Corver & Nunes 2007, Kratzer 2009, Leung 2009.)

120
with other types of pronouns. Consider reflexives. A prominent implementation
among movement theories of reflexivization is to analyze the reflexive itself as a pro-
nounced copy of movement (Lidz & Idsardi 1998, Hornstein 2001, 2007, Boeckx
et al. 2007, Bošković & Nunes 2007, Drummond et al. 2011; cf. Lebeaux 1983,
Chomsky 1993). A sentence such as (6.108a) is derived roughly as in (6.108b),
where ‘everyone’ moves from VP to its pronounced position (deleting the inter-
mediate copy in vP). In languages such as English the lower copy is converted to a
reflexive pronoun in the morphological and phonological components. In explicit
copy reflexive languages the lower copy can even be pronounced itself (Lasnik 1989,
Lee 2003, Mortensen 2003, Boeckx et al. 2007).

(6.108) a. Everyone likes herself.

b. [ IP everyonei [ vP everyonei [ VP likes everyonei -self]]]
(6.109) a. Povi yeej qhuas Povi
Pao always praise Pao
‘Pao always praises himself.’
b. Quas-dlevi pum quas-dlevi .
ind-dog see ind-dog
‘The/a dog sees itself.’ (Mortensen 2003; Hmong)

A straightforward way of implementing such an approach in the present frame-

work would be to represent reflexives in examples such as (6.108)–(6.109) — dedicated
local-domain forms/devices for expressing reflexive relations⁷⁴ — as identical (in
some cases reduced; n. 72) copies of their antecedent at LF. The account in this sec-
tion analyzes certain anaphoric pronouns as non-movement copies of an antecedent
choice-function pronoun and elided NP. Whereas certain internal differences may
be possible in the representation of a donkey pronoun or bound pronoun and its
antecedent (§6.2), no such differences would be possible with reflexives qua copies
of movement. The contrast between derivations of examples with QR and reflexives
might be reflected roughly as in (6.110)–(6.111). (For simplicity I ignore worlds,
additional structure in the quantifier phrases, intermediate copies, and any semantic
contribution of the ‘-self ’ morpheme.)⁷⁵

⁷⁴Hereafter I will omit the qualification limiting to dedicated local-domain forms/devices, but it
should be understood (see e.g. Zribi-Hertz 2008, Déchaine & Wiltschko 2017b).
⁷⁵The example in (6.110) assumes that the attachment of the binder-index feature is built into
the definition of QR. Alternatively, if the object DP enters the derivation with its individual binder-
index feature (as in (i) below), an LF where the lower copy is replaced by t2a could be filtered out
for type reasons, or which type of variable the lower copy is replaced by could be determined by

121
(6.110) ‘Everyone likes everyone.’
a. [everyone⟨1,a⟩ [likes everyone⟨2,a⟩ ]]
b. LF ≈ [everyone{⟨2,a⟩,⟨2,e⟩} [everyone⟨1,a⟩ [likes t2e ]]]
(6.111) ‘Everyone likes herself.’
a. [likes everyone⟨3,a⟩ ]
[everyone⟨3,a⟩ [likes everyone⟨3,a⟩ ]]
b. [[DP every⟨3,a⟩ [F2cf -g3 pers […]]]⟨3,a⟩ [likes [DP every⟨3,a⟩ [F2cf -g3 pers […]]]]]
c. LF ≈ [[DP every⟨3,a⟩ [F2cf -g3 pers …]] [likes [F2cf -g3 pers]]]
d. J(6.111c)K ≈ λgg . [∀xe ∃aa ∶ x(g) = a(g)(2cf)(pers) . . . ] x(g) likes a(g)(2cf)(pers)
≈ for all o selectable among the relevant persons, o likes o

In (6.111) the higher complete copy is pronounced and interpreted, and the lower
copy is reduced prior to LF (indicated by the strikethrough in (6.111b)).⁷⁶ The
reflexive pronoun in (6.111) is the overt realization of an identical (here reduced)
copy of movement in an A-chain (specifically, a chain spanning multiple θpositions).
One argument for movement analyses of reflexives is that they afford a means
of deriving Principle A, that dedicated reflexive forms have local antecedents; the
local relation with the antecedent is derived from the local nature of the kind of A-
movement. A live issue is whether Principle B — that bound pronouns have a non-
local antecedent, that bound pronouns be free in the domain in which reflexives are
bound — may also be explained in more fundamental terms.⁷⁷

(6.112) a. Percyi likes himselfi /∗ himi .

b. Percyi thinks Penny likes himi /∗ himselfi .
which binder-index feature is unchecked in that position, here determining replacement with the
individual-variable (see nn. 13, 24). We will return to object-position quantifiers in §6.3.3.
(i) a. [everyone⟨2,a⟩ ] [⟨2,e⟩ ]
[everyone{⟨2,a⟩,⟨2,e⟩} ]
⋮
[everyone⟨1,a⟩ likes everyone{⟨2,a⟩,⟨2,e⟩} ]
[everyone{⟨2,a⟩,⟨2,e⟩} [everyone⟨1,a⟩ [likes everyone{⟨2,a⟩,⟨2,e⟩} ]]]
b. LF ≈ [everyone{⟨2,a⟩,⟨2,e⟩} [everyone⟨1,a⟩ [likes v2e ]]]
⁷⁶Contrast e.g. Hornstein 2001, 2007, Boeckx et al. 2007, Bošković & Nunes 2007, in which
the chain link phonetically realized as the reflexive pronoun is treated as a complete copy.
⁷⁷See Reinhart & Reuland 1993, Gair et al. 1999, Zribi-Hertz 2008, Déchaine &
Wiltschko 2015, 2017b on non-complementary distributions of reflexives and bound pronouns
(long-distance reflexives, locally bound pronouns) crosslinguistically. See n. 74. Cf. Schlenker
2005 on Condition C.

122
As Hornstein puts it, “Why should anaphors, whose properties follow from the the-
ory of movement, block the the presence of pronouns, whose distribution and inter-
pretive properties follow from the theory of binding? The big fact [that bound pro-
nouns are in complementary distribution with reflexives] is rendered capricious, a
colossal grammatical accident” (2007: 354). In response Hornstein (2001, 2007) ar-
gues that what explains the complementarity of reflexives and bound pronouns is an
economy principle which favors the use of reflexives, qua copies of movement: “what
makes for economy in such cases is Movement. In particular, if a derivation can
converge without the use of pronouns, it must so converge. In short, Move trumps
Pronominalize… [Pronouns are] what the system uses when movement fails” (2007:
383). So, the thought goes, the use of a bound pronoun is available in (6.112) when,
and only when, ‘Percy’ cannot be linked to its lower copy via (A-)movement.
Movement-based theories of construal have provided diverse crosslinguistic ex-
amples of how the use of various types of pronouns — not simply bound pronouns —
anticorrelates with the possibility of movement rather than with the use of reflexives
as such. A critical issue for derivational economy accounts such as Hornstein’s
is to ensure that the core economy assumption — e.g., that syntactic derivations
resorting to movement are preferred as more economical than derivations resorting
to pronoun use — is itself sufficiently motivated to bear the relevant explanatory
weight. Prima facie, one might wonder why moving a lexical item or complex ex-
pression, even understood in terms copying, would be (universally) regarded by the
language faculty as relevantly preferred to using a pronoun. It is well established
that various types of even local A-/A-movements incur non-trivial processing costs
(Felser 2015). It may seem surprising that the relevant local movement operations
would be systematically less costly — or more “economical,” in some relevant sense
of economy — than inserting a bundle of (possibly uninterpreted) phi features.⁷⁸
Economy-based movement accounts aren’t silent on this issue (e.g. Hornstein
2001). This isn’t the place to critically assess movement analyses of reflexivization or
particular explanations of binding-theoretic principles such as Hornstein’s. Suffice
it to say that a generalized assignment-variable analysis, such as considered above,
may provide a basis for an economy-based account of the distribution of reflexives
and pronouns. Unlike bound pronouns, on the above line, reflexives are systemat-
ically represented from identical copies of movement. The relevant economy prin-
ciple needn’t be construed fundamentally as “Avoid Pronouns” (Chomsky 1981),
⁷⁸In order to treat the alternative versions of sentences such as in (6.112) as competing, Hornstein
must assume that the reflexive and bound pronouns are “semantically inert” and hence that their
phi features are uninterpreted — not only in “fake indexical” examples but in general, contrary to
standard theories (see n. 51).

123
or “Move (=Copy-and-Merge) trumps Pronominalize” (Hornstein 2001, 2007),
but, suggestively, “Copy-and-Merge trumps Find and (at least for the most part)
Repeat” — or, less syntactically put, “Using a form marked as a local identical copy of
movement trumps using a form variably interpreted as non-anaphoric or a possibly
non-identical copy of some possibly non-commanding local or non-local antecedent
or other.” An account along these lines might not treat the starred and unstarred
versions of (6.112) as strictly “competing” in the sense relevant in Chomsky 1993.
Yet one needn’t assume that all syntactic principles are fundamentally explained
by purely non-lexical properties (cf. Reinhart 1983a,b, 2006, Grimshaw 1997,
Barbosa et al. 1998, Fox 2000, Reuland 2001, Schlenker 2005, Zribi-Hertz
2008).⁷⁹
The extensions in this section to other types of pronominal anaphora have been
speculative. There are interesting crosslinguistic differences among bound pronouns
and reflexives — e.g., in their relative distributions, internal and external syntax, and
binding-theoretic properties (Déchaine & Wiltschko 2002, 2015, 2017b, Bhat
2004, Zribi-Hertz 2008 and references therein; see nn. 51, 73, also §8.3.1). In-
vestigation of other more articulated structures for certain types of DPs and pro-
nouns, and comparisons with movement analyses of other anaphoric expressions
may provide fruitful avenues to explore (e.g. resumptive pronouns (Pesetsky 1998,
Boeckx 2001, Rouveret 2011); cf. Kayne 2002 for a general movement theory of
pronouns). (We will return to these issues throughout the following sections.)
6.3.3 Donkey crossings: Weak crossover, inverse linking, genitive binding
Our discussion of headed relative clauses has focused on examples with subject-
position quantifiers. A critical issue for developing the account concerns examples
with QR’d quantifier phrases. In (6.113) the determiner ‘every’ raises from CPrel
⁷⁹Movement analyses provide one account of how certain reflexives (n. 74) get their status as spe-
cialized local-domain forms/devices, and thereby block the use of less specialized forms (Williams
1997); yet other explanations are of course possible. For instance, Déchaine & Wiltschko (2015,
2017b) treat the obligatory local binding with morphemes such as ‘self ’ as “forced by” the signal of
“inalienable possession” associated with body-part nouns (see also Vergnaud & Zubizarreta 1992,
Bhat 2004). Though it is true that reflexives are often formed using body-part nouns crosslinguis-
tically, a challenge is to explain why the “inalienable possession” couldn’t be associated with some
non-local subject, as indeed it is in some languages (e.g., Chinese zìji ‘self ’). Such accounts also leave
open what (possibly covert) features force the local binding in other types of reflexive forms (e.g.,
French se ‘3.refl’). See Reuland 2001 for a non-movement economy account of reflexivization as
blocking pronoun binding; see Drummond et al. 2011 for a critical overview.

124
as a quantifier over assignments to take its type ⟨e, t⟩ restrictor argument, and the
object-position DP QRs, binding an individual-trace (again ignoring worlds).⁸⁰

(6.113) Alice likes every baby which laughed.

S′

S′′
DP{⟨2,a⟩,⟨2,e⟩}

Alice likes t2e

⟨2,a⟩ CPrel
every

C
wh2 -g2 baby
⟨1,e⟩

t1e laughed
Crel t2a

In principle one could treat the determiner’s scope argument in such examples fun-
damentally as type ⟨e, t⟩. Yet such a move has an air of “multiplying senses be-
yond necessity.” One would also need to stipulate that the assignment binder-index
feature on the D head fails to project to the DP. A more attractive approach is to
maintain the uniform type ⟨aet, ⟨aet, t⟩⟩. As in the subject-position examples, the
assignment binder-index feature on D projects to the DP, and an individual binder-
index attaches to the QR’d DP, as reflected in the feature set on the DP in (6.113).
However, a prima facie worry may come from crossover.
Following Reinhart 1983a a prominent generalization of weak crossover is
that an expression β can bind a pronoun only if β is in an A-position (argument
position) that c-commands the pronoun at LF. This generalization allows trace-
binding but excludes pronoun-binding from A-positions (non-argument positions),
such as positions derived from QR or wh-movement, as in (6.114)–(6.115).

(6.114) a. Every cat i likes itsi owner.

b. *Itsi owner likes every cat i . (≠ ‘Every cat is liked by its owner’)

⁸⁰If Kennedy’s (2014) approach to Antecedent Contained Deletion is right, there may be reasons
for requiring coindexing between the individual binder features on the quantificational DP and its
relative complement CPrel , e.g. via agreement. I will ignore such a requirement in what follows.

125
(6.115) a. Whoi (ti ) likes heri child?
b. *Whoi does heri child like ti ?

Reinhart’s generalization has also been applied to donkey pronouns in cases of “don-
key crossover” (Haïk 1984, Reinhart 1987, Büring 2004), reflected in (6.116)–(6.117).

(6.116) a. Every farmer who beat a donkeyi killed itsi lawyer.

b. ≈ “every farmer x who beat a donkey killed the lawyer of the donkey
beaten by x”
(6.117) a. *Itsi lawyer sued every farmer who beat a donkeyi . (Büring 2004: ex. 8b)
b. ≉ “every farmer x who beat a donkey is such that the lawyer of the donkey
beaten by x sued x”

The object DP ‘every farmer who beat a donkey’ in (6.117) can bind its trace but,
unlike the subject DP in (6.116), it cannot provide an antecedent for the donkey
pronoun ‘it’.
Absent additional constraints, nothing would seem to exclude the rough LF and
semantic value for (6.117) in (6.118). This represents the unattested bound reading
where the interpretation of the donkey pronoun covaries with the raised DP. (I
will simplify by ignoring the internal structure in C0rel = [C0 Crel t2a ], in addition to
continuing to ignore worlds.)

(6.118) ‘Its lawyer sued every farmer who beat a donkey’

⟨1,e⟩
a. … [ S′ [ DP{⟨2,a⟩,⟨3,e⟩} every⟨2,a⟩ [ CP wh2cf -g2 -farmer C0rel [t1e beat a3 -g2 -
donkey]]] [ S′′ the-lawyer-of-it3 -g2 -donkey sued t3e ]]
b. ≈ λgg .[∀xe ∃aa ∶ x(g) = a(g)(2cf)(farmer) ∧ x(g) beat a(g)(3cf)(donkey)]
the-lawyer-of-a(g)(3cf)(donkey) sued x(g)

One could treat Reinhart’s generalization as a basic principle in the grammar.⁸1 LFs
such as (6.118a) would be excluded because they violate Reinhart’s generalization:
the assignment-variable in the pronoun [it3cf g2 ] (underlined in (6.118a)) is bound
by the assignment-binder from the A-position DP.
It would be theoretically preferable to derive generalizations such as Reinhart’s
from more basic features of the syntax/semantics. A more explanatory approach
might be to allow expressions to have multiple binder-indices, but include a con-
straint that no expression may bind distinct variables, in some relevant sense of
⁸1Cf. Safir’s (2004) Quantifier Dependency Condition, which essentially reformulates Reinhart’s
generalization that pronouns can only be bound from A-positions.

126
binding — call it “s-binding.” A first approximation is as follows. (I will often be
sloppy about distinguishing expressions/variables (qua types) from occurrences.)

(6.119) Variable Binding Constraint: An occurrence of an expression in a tree γ may

s-bind occurrences of at most one variable at LF.
a. An occurrence of an expression β s-binds an occurrence of a variable viσ
in γ iff the sister of β is the largest subtree of γ in which viσ is s-free.
b. viσ is s-free in γ iff there is no occurrence in γ of an expression with
binder-index feature ⟨i,σ⟩ that c-commands viσ .

An explanation of donkey crossover falls out directly: The raised object DP in (6.118a)
has the set of binder-index features {⟨2,a⟩,⟨3,e⟩} . Like all movement, QR leaves a coin-
dexed trace. On the unattested reading, the assignment binder-index binds the
assignment-variable in [it3 g2 ] and the individual binder-index binding the trace t3e .
This violates the constraint in (6.119). The donkey pronoun cannot be bound by the
raised DP because the DP’s binding capacities are exhausted from binding the trace
derived from QR. Hence a string such as in (6.118) can have an acceptable reading
but only with an intuitively free reading of ‘its’. No analogous constraint excludes
an intuitively bound reading of ‘its’ in (6.116) where the DP is in an A-position: the
subject DP doesn’t bind a trace and is free to bind the donkey pronoun.
Adopting a general type-⟨aet, ⟨aet, t⟩⟩ syntax/semantics of determiner quanti-
fiers provides a means of generalizing the above approach to weak crossover to ex-
amples with non-relative complements, as reflected schematically in (6.120). (I con-
tinue to assume the general D+CP syntax for quantificational DPs from §6.3.1.)

(6.120) a. Every dog i likes itsi owner.

[[ DP⟨2,a⟩ every⟨2,a⟩ [ CP F2cf -g2 -dog …]] [ VP likes the-owner-of-it2 -g2 -dog]
b. *Itsi owner likes every dog i .
[ S′ [ DP{⟨2,a⟩,⟨3,e⟩} every⟨2,a⟩ [ CP F2cf -g2 -dog …]] [ S′′ the-owner-of-it2 -g2 -dog
likes t3e ]]

The anaphoric pronouns are represented as copies of their linguistic antecedents

(§6.3.2). In (6.120b) the trace t3e must be bound by the individual binder-index
on the QR’d DP. This excludes the assignment-variable in the representation of the
pronoun ‘its’ from being bound by the assignment binder-index projecting from D
to the DP.
The foregoing approach carries over to other recalcitrant cases of apparent bind-
ing out of DPs (see esp. Büring 2004). The remainder of the section considers two
such cases: genitive binding and inverse linking.

127
 Consider donkey-style anaphora and crossover with genitives:

(6.121) a. Every boyi ’s cat likes himi .

b. ≈ “for every boy b, the cat owned by b likes b”
(6.122) a. Every boyi ’s sisterj ’s cat likes himi /herj
b. ≈ “for every boy b, the cat of the s s.t. s is the sister of b likes b/s”
(6.123) a. *Hisi dog likes every boyi ’s cat.
b. ≉ “for every boy b, the cat owned by b is liked by the dog owned by b”

In (6.121) the embedded DP ‘every boy’ can provide an antecedent for ‘him’ even
though it doesn’t c-command the pronoun at LF, being in a specifier position of the
main DP. In (6.122) the pronoun can be linked either to the embedded DP ‘every
boy’s sister’ or to the DP ‘every boy’ embedded in it. Yet the interpretation of the
pronoun cannot vary when the main genitive DP QRs in (6.123).
For present purposes we can put aside many controversial issues about the syn-
tax of different types of genitive constructions, in English and crosslinguistically.
For concreteness let’s assume the D+CP implementation of “free R” readings of
prenominal genitives X’s Y from §6.3.1 — where the relevant relation is supplied
by an implicit pronoun R (Partee 1978, 1983, Higginbotham 1983), the “pos-
sessor” X+relation R originate as modifiers in the complement, and the genitive
morpheme ’s is the head determiner, J’sK = JtheK , as reflected in (6.124) (n. 67; see
§6.3.1 for an alternative syntax/semantics which analyzes ’s as heading the nominal
XP projection complement to an implicit definite determiner).

(6.124) a. ‘Alice’s cat’

b. [ DP⟨2,a⟩ ’s⟨2,a⟩ [ CP F2cf -g2 cat [ C [ C⟨1,e⟩ C t2a ] t1e R1eet -g1 Alice]]]

In (6.121) the quantificational possessor ‘every boy’ must be interpreted for type
reasons in its pronounced position as a specifier of the main genitive DP; ‘every
boy’ raises to SpecDP and binds its trace in D (cf. e.g. May 1985). Thus far our
compositional semantics has proceeded solely via function application; however,
adapting the treatment of inverse linking in Kobele 2010, suppose that we allow
a role for function composition in deriving the semantic values of certain adjunc-
tion structures, such as complex DPs formed from DP-internal QR/movement.⁸2
⁸2Kobele (2010) also introduces assignments into the model and type system. He doesn’t go the
further step of incorporating assignment-variables into the syntax, or consider assignments or binder
indices for variables of types other than e. I consider additional applications of function composition
with ‘if ’-clauses in §7; more on inverse linking shortly.

128
Combining D of type ⟨et, t⟩ and the raised DP of type ⟨t, t⟩ by function composition
yields the complex subject DP of generalized quantifier type ⟨et, t⟩, as shown in
(6.125). The assignment binder-index feature projected to the DP binds the donkey
pronoun, again represented as a copy of the antecedent choice-function pronoun
and complement NP (here internal to the complex quantificational DP∗∗ ).⁸3

(6.125) Every boyi ’s cat likes himi .

S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a

DP{...⟨2,a⟩... } ∶ ⟨et, t⟩ VP

likes t1s
DP∗∗{⟨2,e⟩,⟨2,a⟩} ∶ ⟨t, t⟩ D⟨3,a⟩ ∶ ⟨et, t⟩
him2 -g2 boy-w1 -g1

’s⟨3,a⟩ CP
every⟨2,a⟩ CP

C C

F2cf -g2 boy-w1 -g1 …t2a … F3cf -g3 cat-w1 -g1 ⟨1,e⟩

t1e
C t3a R1 g1 t2e

JDPK = JDP∗∗ K ○ JDK = λPet .JDP∗∗ K(JDK(P))

≈ λPet .[λTt .[λgg .∀xe ∃aa ∶ x(g) = a(g)(2cf)(boyg(1a)(1s) ), T(g[x(g)/2e][a(g)/2a])]]
(λgg .ιye ∃a′a ∶ y(g) = a′ (g)(3cf)(catg(1a)(1s)) ) ∧ g(1a)(1eet)(g(2e))(y(g)), P(y)(g[a′ (g)/3a]))
≈ λPet .λgg . ∀xe ∃aa ∶ x(g) = a(g)(2cf)(boyg(1a)(1s) ),
ιye (g)∃a′a ∶ y(g) = a′ (g)(3cf)(catg(1a)(1s) ) ∧ g(1a)(1eet)(x(g))(y(g)),
P(y)(g[x(g)/2e][a(g)/2a][a′ (g)/3a])
JSK ≈ λgg . ∀xe ∃aa ∶ x(g) = a(g)(2cf)(boyg− (1s) ),
⁸3I assume that, like [wh] features, binder-index features on/c-commanding a head X in an XP
project/percolate to the XP. For present purposes I ignore intensionality (an internal world-argument)
in the genitive relation.

129
 ιye (g)∃a′a ∶ y(g) = a′ (g)(3cf)(catg− (1s) ) ∧ g(1eet)(x(g))(y(g)),
y(g) likes a(g)(2cf)(boyg− (1s) ) in @(g− )

This says, roughly, that for every o ∈ E s.t. there is a choice function that selects o
from among the boys, the unique o′ ∈ E s.t. o′ is selected by some choice function
from among the cats and bears the relevant relation R (“is owned by,” “is a pet of ”)
o, is such that o′ likes o — i.e., for every o in the set of boys, the unique o′ among the
cats which bears R to o likes o.
The above account of weak crossover/“donkey crossover” carries over to “geni-
tive crossover” violations such as (6.123). (I use ’s both for the overt morpheme and
for the silent morpheme combining with the individual pronoun in the spell-out of
the possessive pronoun.)

(6.126) *His dog i likes every boyi ’s cat.

S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a

DP{⟨4,e⟩,⟨2,a⟩... } S′′

DP∗∗{⟨2,e⟩,⟨2,a⟩} D∗⟨3,a⟩ DP
likes t1s t4e
⟨3,a⟩ ’s⟨4,a⟩ CP
’s CP
every⟨2,a⟩ CP

C C

F2cf -g2 boy-w1 -g1 …t2a … F3cf -g3 cat-w1 -g1 ⟨1,e⟩ F4cf -g4 dog-w1 -g1
t1e . . . t4a . . .
C t3a R1 g1 t2e
. . . he2 -g2 boy-w1 -g1 . . .

The object DP ‘every boy’s cat’ has a set of binder-index features including an indi-
vidual binder-index resulting from QR, and an assignment binder-index projecting
from DP∗∗ . The individual binder-index binds the DP’s coindexed trace t4e . In order
for the interpretation of the pronoun to covary with the DP, the assignment binder-

130
index must bind the assignment-variable in [he2 g2 ]. This violates the binding con-
straint in (6.119) which excludes a single expression from binding distinct variables.
Inverse linking readings, as in (6.127), raise well-known challenges for compo-
sitional semantics. The embedded DP ‘every parent’ in (6.127) is a constituent of
the subject DP (e.g. May 1985, Heim & Kratzer 1998). So, first, ‘every parent’
must be able to combine with its sister ‘some child of t’ to form the subject DP.
Second, the interpretation of ‘her’ covaries with the quantificational subject even
though the pronoun isn’t c-commanded by ‘every parent’. Third, as with donkey
pronouns, pronouns in inverse linking examples exhibit weak crossover effects:

(6.127) a. Some child of every parent loves her.

≈ “for every parent o, some child o′ of o loves o”
b. *Her book is loved by some child of every parent.
≉ “for every parent o, there is some child o′ of o such that o’s book is
loved by o′ ”

The raised ‘every parent’ can bind its trace in (6.127a)–(6.127b); but it can provide a
linguistic antecedent for the pronoun only in (6.127a) when embedded in the subject
DP, and not in (6.127b) when embedded in the QR’d object DP.
The compositional semantics proceeds analogously to examples with complex
genitives, as reflected in (6.128). DP∗∗ and D∗ of types of types ⟨t, t⟩ and ⟨et, t⟩,
respectively, combine by function composition to yield the main DP of generalized
quantifier type ⟨et, t⟩. The assignment binder-index feature projected from DP∗∗ to
the DP binds the donkey pronoun. (As previously I assume the generalized D+CP
implementation from §6.3.1.)

131
(6.128) Some child of every parent loves her.
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a

DP {...⟨2,a⟩... } ∶ ⟨et, t⟩ VP

loves t1s
∗∗{⟨2,e⟩,⟨2,a⟩}
DP ∶ ⟨t, t⟩ DP∗⟨3,a⟩ ∶ ⟨et, t⟩ her2 -g2 parent-w1 -g1

some⟨3,a⟩ CP
every⟨2,a⟩ CP

C C

F2cf -g2 parent-w1 -g1 …t2a … F3cf g3 …t3a …

t2e
child-w1 -g1
JDPK = JDP∗∗ K ○ JDP∗ K
≈ λPet .λgg . ∀xe ∃aa ∶ x(g) = a(g)(2cf)(parentg(1a)(1s) ),
∃ye ∃a′a ∶ y(g) = a′ (g)(3cf)(child-of-x(g)g(1a)(1s) ),
P(y)(g[x(g)/2e][a(g)/2a][a′ (g)/3a])
JSK ≈ λgg . ∀xe ∃aa ∶ x(g) = a(g)(2cf)(parentg− (1s) ),
∃ye ∃a′a ∶ y(g) = a′ (g)(3cf)(child-of-x(g)g− (1s) ),
y(g) loves a(g)(2cf)(parentg− (1s) ) in @(g− )

The derived semantic value captures the inverse linking reading: (6.128) says that
for every o ∈ E s.t. there is a choice function that selects o from among the parents,
there is an o′ ∈ E selected by some choice function from among o’s children s.t. o′
loves o — i.e., for every parent o, there is an o′ among the children of o s.t. o′ loves o.
The above account of weak crossover carries over to “inverse crossover” exam-
ples such as (6.127b) or (6.129):

132
 (6.129) *Her cat i likes some friend of every childi .
≉ “for every child o, there is some friend of o that is liked by o’s cat”
S

T⟨1,a⟩ CP

⟨1,s⟩ S′

Cdec t1a

DP{⟨4,e⟩,⟨2,a⟩... } S′′

DP
DP∗∗{⟨2,e⟩,⟨2,a⟩} DP∗⟨3,a⟩
t4e
likes t1s
’s⟨4,a⟩ CP
every⟨2,a⟩ CP some⟨3,a⟩ CP

C
C
F2cf -g2 child-w1 -g1 …t2a … F4cf -g4 cat-w1 -g1
F3cf g3 …t3a …
t2e . . . t4a . . .
friend-w1 -g1 . . . she2 -g2 child-w1 -g1 . . .

The QR’d object DP binds its coindexed trace t4e . Given the constraint that an
expression cannot semantically bind distinct variables, the assignment-variable g2
in the representation of ‘her’ cannot be bound by the DP, and the intuitively bound
reading is excluded.
Let’s recap. This section has speculated about how the assignment-quantificational
approach to donkey anaphora from §6.2 might be extended to other recalcitrant
phenomena with pronominal anaphora — e.g., regarding weak crossover effects and
apparent binding out of DPs with inverse linking and genitive binding. The D+CP-
style syntax for inverse linking and genitives captures how expressions such as ‘some
child of every parent’ and ‘every boy’s cat’ form constituent DPs (contrast Horn-
stein 1995). The intuitively bound readings of pronouns in the DPs’ scope fol-
low from (i) the treatment of anaphoric pronouns as copies of their linguistic an-
tecedents, (ii) the generalized assignment-quantificational syntax/semantics for de-
terminer quantifiers, and (iii) the use of function composition in the compositional
semantics of (e.g.) complex DPs formed from DP-internal movement. I suggested
potentially explaining weak crossover data such as Reinhart’s generalization via a
general principle that no expression can bind distinct variables. Since a moved ex-

133
pression must bind its trace, the assignment-variable in a subject pronoun cannot be
bound by a QR’d object DP — in examples with simple DPs ((6.120b)), donkey sen-
tences ((6.117)–(6.118)), inverse linking ((6.127b)/(6.129)), and genitive binding
((6.123)/(6.126)) alike. There is no analogous obstacle to bound readings when the
DP is in subject position — hence Reinhart’s generalization that bound pronouns,
unlike traces, must be bound from A-positions.⁸⁴
A worry with many accounts of (donkey) crossover is that they end up for-
malizing what needs to be explained — the distinction between trace-binding and
pronoun-binding. Contrasts such as in (6.116)–(6.117) may be captured via stipu-
lations on admissible indexings (e.g. Haïk 1984, Safir 1984, Reinhart 1987), or ad
hoc syntactic/semantic distinctions between traces and pronouns (e.g. Büring 2004,
Safir 2004, Elbourne 2005). For instance, Büring 2004 — arguably the most ex-
tensively developed account — introduces distinct syntactic categories for traces and
pronouns (including, crucially, individual-variables in E-type representations), dis-
tinct binding operators and domains for assignments corresponding to the two cat-
egories, and distinct principles regarding admissible LFs for traces/pronouns and
their respective binding operators. At minimum the present assignment-variable-
based framework offers independent grounds for formally distinguishing pronouns
and traces (§§2, 4.3). Pronouns [viτ gj ] include an assignment-variable from which
the element receives its interpretation; no assignment-variable is included in the
representation of a trace tiτ since the binder-index attaching to the moved expres-
sion binds the variable directly. This independent distinction may be exploited in
an account of weak crossover. No additional binder operators or constraints on
admissible indexings are required. Assignment functions and the generalized binder
index make no distinction between traces and pronouns as such.
I hope the preliminary discussion in this section may provide a basis for a more
explanatory treatment of weak crossover. The prospects for the specific assignment-
variable-based approaches in this section to determiner quantification, relativiza-
tion, and trace-/pronoun-binding remain to be seen. Ellipsis, “sloppy identity” read-
ings, and discourse anaphora may offer fruitful additional applications to explore.
⁸⁴Yes, the last inference was too quick. The above discussion applied specifically to DPs with
assignment binder-indices. Nothing has been said about weak crossover effects with wh interrog-
atives, as in (6.115). There are independently attested contrasts between wh-chains and QR-chains
regarding the interpretation of traces/pronouns (e.g. Safir 1999). How to extend the approach to
weak crossover in this section will depend on one’s views on the syntactic/semantic status of wh
movement, reconstruction, pied piping, etc. A more general binding theory is certainly needed. I
return to wh interrogatives in §8.3.

134
7 Conditionals
7.1 Local and global readings
Conditionals provide diverse sources of local and global readings of variables and
context-sensitive expressions. For instance, there are (i) global readings in both ‘if ’-
clause and consequent clause, as in (7.1); (ii) local readings in the ‘if ’-clause, as in
(7.2); and (iii) local readings in the main clause, as in (7.3).

(7.1) If iti breaks, hej will cry.

● ≈ If gc (i) breaks, gc (j) will cry. (global readings, both clauses)
(7.2) a. [Giving to Charity X wouldn’t maximize overall utility.]
If we shouldi give to Charity X, classical utilitarianism must be incorrect.
● ≈ If the correct norms are i and i requires us to give to Charity X, then
classical utilitarianism must be incorrect.
b. If it’s rainingi , we should bring an umbrella.
c. [We know how much everyone’s income is. Rita is getting a tax break.
We haven’t settled ourselves about what income should count as “rich.”]
If Rita is richi , then Miguel is rich. So Miguel should get the tax break too.
(local reading, ‘if ’-clause)
(7.3) a. [If John wasn’t invited,]i everyonei will come.
● ≈ [If John wasn’t invited,]i everyone who is relevant given i will come.
b. [If it rained,]i John broughti an umbrella.
c. [If the intersection is clear,]i we shouldi speed through.
(local reading, main clause)

Semantic work on conditionals has focused nearly exclusively on examples where

the ‘if ’-clause combines with a complete clause. Yet there are also conditionals where
the ‘if ’-clause appears sentence-finally and modifies the verb, or appears sentence-
internally and modifies a nominal subject, as in “adnominal” conditionals. Syntactic
tests confirm that the ‘if ’-clauses in such examples cannot be construed as scoping
over the rest of the sentence, but rather combine with the VP/NP, as reflected in the
VP-ellipsis and Condition C evidence in (7.4) and coordination evidence in (7.5)
(Iatridou 1991, Lasersohn 1996, Bhatt & Pancheva 2006).

(7.4) Sentence-final ‘if ’-clause. VP-adjunction

a. I will leave if you do and John will [VP leave if you do] too.

135
 b. *Shei yells at Bill if Maryi is hungry.
(Bhatt & Pancheva 2006: exs. 19a, 21a)
(7.5) Adnominal ‘if ’-clause. NP-adjunction
[[The [ NP location if it rains]] and [the [ NP location if it doesn’t rain]]] are
within five miles of each other. (Lasersohn 1996: ex. 10)

Shifted readings can be observed in each of these positions, as in (7.7)–(7.8) as well.

For instance, Santorio 2012 argues that examples such as (7.6) involve a shifted
(i.e. local) reading of ‘I’, referring to epistemic counterparts of the speaker (see 2012:
363n.2 for precedents). Shifted readings of ‘I’ are no less possible when the ‘if ’-clause
combines with the VP, as in (7.7d)–(7.7e); likewise for the shifted readings of ‘my’
in (7.8b) when the ‘if ’-clause combines with an NP.

(7.6) [Context: Lingens and Lauben know they are kidnapped amnesiacs. They
are informed that they will be anesthetized, and a coin will be flipped: if it
lands tails, Lingens will be released in the Stanford library and Lauben will
be killed; if it lands heads, Lauben will be released in the Harvard library
and Lingens will be killed. After the experiment, one of them wakes up and
says:]
If the coin landed heads, I am in Widener Library, Harvard.
(Santorio 2012: ex. 6)
(7.7) a. John broughti an umbrella [if it rained]i (…and Alice did [bringj an umbrella
[if it rained]j ] too).
b. Everyonei will come [if John wasn’t invited]i .
c. We shouldi block Shaft A [if the miners are in Shaft A]i .
(cf. Kolodny & MacFarlane 2010)
d. Ii [VP am in Widmore Library [if the coin landed heads]i ].
e. Bert is in Widmore Library if the coin landed heads, and Ii am in Widmore
Library [if the coin landed heads]i too).
(7.8) a. The [ NP righti thing [if utilitarianism is correct]i ] is to donate.
b. [[ DP Myi location [if the coin landed heads]i ] and [ DP myj location [if the
coin landed tails]j ]] are on opposite coasts.

Most existing semantics for (sentence-initial) conditionals are inapplicable as

they stand to VP-/NP-adjunction examples. For instance, on a standard restrictor
semantics (Kratzer 1981, 1991), the ‘if ’-clause combines with a (possibly implicit)
operator/modal and restricts its modal base, or domain of quantification. It isn’t

136
immediately obvious how such an analysis might be generalized to compositionally
derive the relevant range of shifted/non-shifted interpretations in examples such as
(7.4)–(7.5) or (7.7)–(7.8) where the ‘if ’-clause combines with a VP or subject NP.
Historically, work on the semantics of conditionals has proceeded largely in
isolation from syntax literatures on ‘if ’-clauses and other types of conditional con-
structions. To some extent this is to be expected. Many questions regarding the
semantics of conditional can be addressed without taking a stand on, say, whether
‘if ’ is in C or SpecCP in different languages, possibilities for reconstruction, etc.
But if our project is to provide a genuine compositional semantics, we must ensure
that whatever type of semantics we go in for will generalize across the spectrum of
environments in which ‘if ’-clauses (and other markers of conditionality) appear.
Whatever syntactic/semantic package we take on board must (i) derive local
readings in the consequent and in the antecedent — shifting in the scope of a sup-
position as well as in the supposition itself, as in (7.2)–(7.3); (ii) must do so in a way
that still allows for global readings in both clauses, as in (7.1); and (iii) must capture
how ‘if ’-clauses can combine directly with a VP/NP and shift the interpretation
of material within the VP/NP they modify, as in e.g. (7.7a)/(7.8a), as well as of
external material in the sentence, as in e.g. (7.7d). The type of the ‘if ’-clause must be
suitable to combine with and modify expressions of multiple types. The remainder
of this section shows how we can satisfy these desiderata, drawing on independent
syntactic work on ‘if ’-clauses as free relatives/correlatives. The compositional se-
mantics affords a uniform analysis of ‘if ’-clauses in diverse types of conditionals —
conditionals with sentence-initial/-internal/-final ‘if ’-clauses, and in conditionals
with/without an overt modal or a proform such as ‘then’. As we will see further in
§8, the account affords uniform treatments of conditional and individual proforms,
and of conditional, correlative, and interrogative clauses.
7.2 ‘If ’ and conditionality
7.2.1 Syntax: Free relatives + Clause-internal operator movement
A standard story following Lewis (1975) and Kratzer (1981, 1991) is that ‘if ’-
clauses restrict the domain of a modal or other operator. Yet as von Fintel notes,
“it is very probable… that tripartite structures are merely a convenient meta-level
notation” (1994: 77), and that movement operations don’t literally generate tripartite
LFs for conditionals like (7.9) parallel to determiner quantifiers (cf. Partee 1995b).

(7.9) [S [MODi [if-clause]j ] [CP tj [. . . ti . . . ]]]

So I assume that our best syntactic story for conditionals will involve some other way

137
of capturing the idea that ‘if ’-clauses function to modify a domain. In §6 we drew on
D-complement and head-raising analyses to motivate treatments of individual- and
assignment-quantification in headed relative clauses — a relative complementizer
Crel , and a determiner which moves for type reasons from an internal assignment
argument. I suggest that prominent syntactic theories of ‘if ’-clauses as free relatives
(Iatridou 1991, Lycan 2001, von Fintel 1994, Bittner 2001, Schlenker 2004,
Bhatt & Pancheva 2006) provide analogous motivations for sources of world-
and assignment-quantification in conditionals. Like other free relatives (Jacobson
1995, Dayal 1996, Caponigro 2012), ‘if ’-clauses are interpreted as definite descrip-
tions. Indeed in some languages conditionals are overtly realized as such, as we saw
in §6.2.1 with Somali, where the conditional clause is introduced by a modal-past
(remoteness) marked definite description.

(7.10) a. Nin hád-díi uu seexdó oo sóo toosó, waa isá-gíi

man time-def.past 3 sleeps and dir wakes-up decl him-emph.past
ún.
only
‘If (lit. the time) a man goes to sleep and then wakes up, he is only himself
(i.e. the same as he was before).’
b. Hád-díi aad rabtó na ráac!
time-def.past you want us follow.imp
‘Come along with us, if (lit. the time) you want!’ (Lecarme 2008: ex. 19)

However, rather than treating the variable relativized over as a variable over worlds
or events (as in Schein 2003, Schlenker 2004, Bhatt & Pancheva 2006, Haege-
man 2010), we can treat it as a variable over assignments: ‘if ’-clauses denote definite
descriptions of possibilities, represented via assignments (cf. Bittner 2001).
Given the variable means of expressing conditionality, both in English and crosslin-
guistically, it is common to treat conditional interpretations as arising independent
of particular complementizers such as ‘if ’ (e.g. Bhatt & Pancheva 2006, Rawlins
2008). A prominent approach is to treat the conditional element as moving to a
specifier position from within the clause (Bhatt & Pancheva 2006, Haegeman
2010; cf. Larson 1985, Iatridou et al. 2001). Movement analyses of conditional
clauses provide independently motivated resources for implementing an analysis
of ‘if ’-clauses into the §3-framework — notably, a complementizer such as ‘if ’, and
clause-internal movement of an operator responsible for conditional interpretations:
Just as determiner quantifiers raise from the internal assignment argument of the
relative complementizer in relatives of individuals (§6), the conditional operator

138
raises from an internal assignment argument of complementizers such as ‘if ’.⁸⁵ A
schematic LF is as follows, where PD is the implicit operator (for Plural Description),
parallel to the definiteness operator in certain free relatives of individuals.

(7.11) CP

PD⟨i,a⟩
C

⟨j,s⟩ IP

if tia . . . tjs . . .

7.2.2 Semantics: ‘if ’-clauses as plural definite descriptions of assignments

Consider the following preliminary lexical entries for ‘if ’ and the general condition-
alization/definiteness operator PD (to be revised; cf. n. 20).

(7.12) JifK = λaa .λpst .λws .λgg . @(a(g)) ≤ w(g) ∧ ∀w′ s.t. w′ (g) = @(a(g)), p(w′ )(g)
(to be revised)
(7.13) JPDK = λAat .λA′at .λgg . [ιa(g)∗ ∶ A(a)(g)] A′ (a)(g)
= λAat .λA′at .λgg . [∃aa ∶ A(a)(g) ∧ ∀a′′ s.t. A(a′′ )(g), a′′ (g) ≤ a(g)] A′ (a)(g)
= λAat .λA′at .λgg . [ιa(g)∗ ∶ a(g) is the unique maximal h∗ ∈ G s.t. (↓ A)(g)(h)] A′ (a)(g)

The lexical entry for ‘if ’ in (7.12) is a rough intensional analogue of the entry for the
relative complementizer Crel . The world argument in the specifier position deter-
mines the relevant restricted domain of possibilities. PD converts the set of possi-
bilities into a definite description of the maximal plural possibility in the set. Let’s
take these components in turn.
There is typically more than one p-possibility for a given ‘if ’-clause if p. First,
following general treatments of free relatives as plural definite descriptions, the as-
signment described by the ‘if ’-clause may be a plural assignment representing a plu-
ral possibility (cf. Jacobson 1995, Dayal 1996, Grosu & Landman 1998, Schein
2003, Bhatt & Pancheva 2006). The components of the model E, W, G are struc-
⁸⁵I assume that ‘if ’ is in C0 . Although Bhatt & Pancheva 2006 don’t specify the generation
site of the conditional operator, Haegeman 2010 argues against treating it as moving from within
the VP. Haegeman treats the operator as semantically modal and as sharing properties specifically
with syntactically high (broadly epistemic) modals. Such a view may provide further support for the
present treatment of the operator as moving from above the VP.

139
tured to include plural objects — objects which contain atomic objects as parts (Link
1983, Schwarzschild 1996). The metalanguage ι operator in (7.13) can be un-
derstood along the lines of Link’s (1983) σ operator, which returns the maximal
entity of a set — the element of the set that isn’t a proper part of any other element
of the set (formally, the m in the set S such that ∀m′ ∈ S∶ m′ ≤ m). The uniqueness
implication applies to the items in the model. Just as there may be many functions
x ∶ G → E mapping g to the unique individual o ∈ E intuitively picked out by a definite
description, there may be many functions a′ ∶ G → G mapping g to the relevant
unique possibility h ∈ G described by an ‘if ’-clause. (I often use an asterisk ∗ when
introducing a variable to flag that its value may be a plurality (“ιo∗ ∶ o…”). I won’t
distinguish between singular vs. plural predicates; my saying that a property holds
of an object leaves open whether the predication is of an atomic object, a plural
object, or every (atomic) part of a plurality.)
Even after incorporating plurality, our analysis should reflect some way of re-
stricting the quantification to relevant (salient/live/remote) assignments satisfying
the antecedent. The external argument of ‘if ’ supplies a relevant modal possibility
which further restricts the domain of the definiteness operator PD, as reflected in
(7.14).⁸⁶ (⊔S is the maximal element of a set S — e.g., if S = {a, b, c}, ⊔S = a ⊔ b ⊔ c.
For readability I continue to omit explicit reference to the quantification over w ∈ Ds
when the modal domain has been derived.)

(7.14) CP⟨2,a⟩ ∶ ⟨t, t⟩

PD⟨2,a⟩ C′′ ∶ t

C′ ∶ ⟨s, t⟩
⊔ p -g
1st 1
⟨2,s⟩ IP ∶ t

if t2a rained t2s

JC′ K = λws .λgg . @(g(2a)) ≤ w(g) ∧ ∀w′ s.t. w′ (g) = @(g(2a)), it rained in w(g)
JCP⟨2,a⟩ K ≈ λTt .λgg . [ιaa (g)∗ ∶ @(a(g)) ≤ (⊔g(1a)(1st))
∧ it rained in @(a(g))] T(g[a(g)/2a])
≈ λT.λgg . [∃aa ∶ a(g)∗ is the unique maximal h∗ ∈ G s.t. @(h) ≤ (⊔g(1a)(1st))
∧ it rained in @(h)] T(g[a(g)/2a])
⁸⁶In the definition of the binder-index (§3.5): with if, τ = s; σ = ⟨s, t⟩; with PD, τ = a; σ = ⟨at, t⟩.

140
Like the declarative complementizer, the complementizer ‘if ’ is base-generated in
the world argument position of the clause’s main predicate, and moves for type
reasons as a quantifier over worlds. This captures how the main predicate in the ‘if ’-
clause receives an obligatory local reading under the supposition. Like determiner
quantifiers in relative constructions and modal quantifiers, the conditionalization
element PD is base-generated in the complementizer’s assignment argument posi-
tion, and moves for type reasons (here clause-internally), introducing an assignment
binder. Intuitively put, the specifier argument in the conditional clause supplies a
topical possibility which provides the backdrop for the supposition. The conditional
clause picks out the sum of relevant possibilities in which it rained. (Recall that I’m
not distinguishing between singular/plural predicates; so, saying e.g. that it rained
in @(h∗ ) can be understood as saying that it rained in the world of every possibility
that is a part of h∗ .)
The preliminary structure in (7.14) treats the external argument of ‘if ’ as a rele-
vant plurality of worlds, and the semantics for ‘if ’ requires the world of the internal
assignment-argument to be a part of that plurality. The analysis of relative clauses in
§6.1, in contrast, treated the external argument of Crel as a selected individual o ∈ E
supplied by the relative phrase, and the semantics required the individual in the gap
position to be identical to o. Given the proposal that ‘if ’-clauses are free relatives of
possibilities, it is worth exploring how we might unify our analyses of relative clauses
and ‘if ’-clauses.
I offer the following revised lexical entry for ‘if ’ and structure for an ‘if ’-clause.
The analysis of (7.16) parallels the analysis of free relatives of individuals from (6.13),
reproduced below.

(7.15) JifK = λaa .λpst .λws .λgg . @(a(g)) = w(g) ∧ ∀w′ s.t. w′ (g) = @(a(g)), p(w′ )(g)
(7.16) CP⟨2,a⟩ ∶ ⟨t, t⟩

PD⟨2,a⟩ C′′ ∶ t

C′ ∶ ⟨s, t⟩

wh2sf g2 p1st g1 ⟨2,s⟩ IP

if t2a rained t2s

JC′′ K = λgg . @(g(2a)) = g(2a)(2sf)(g(1a)(1st))

141
 ∧ ∀w s.t. w(g) = @(g(2a)), it rained in w(g)
⟨2,a⟩
JCP K ≈ λTt .λgg . [ιaa (g)∗ ∶ @(a(g)) = a(g)(2sf)(g(1a)(1st))
∧ it rained in @(a(g))] T(g[a(g)/2a])
(6.13) a. ‘(Bert likes) what Alice likes’
b. [DEF⟨i,a⟩ [ CPrel wh-gi P-g [ C [Crel tia ]⟨j,e⟩ [ IP Alice likes tje ]]]]

In an individual relative clause (see §6.1): The relative phrase ‘what’ supplies the
external individual argument of Crel and is analyzed intuitively as “which relevant
thing(s).” The relative choice-function pronoun is (obligatorily) coindexed with
the relative complementizer’s assignment argument, e.g. due to agreement, and the
property-type argument supplies a set of contextually relevant individuals. The assignment-
quantification introduced by the raised determiner then binds the choice-function
pronoun, and the determiner quantifies over individuals that satisfy the property
denoted by the relative clause C and are identical to some or other relevant individual
o ∈ gc (1et) ∈ E. The compositional semantics derives that the free relative (6.13)
‘what Alice likes’ picks out the sum of relevant things that Alice likes.
Analogously: Just as the relative phrase ‘what’ in (6.13) is analyzed intuitively
as “which relevant thing(s),” the external argument of ‘if ’ is analyzed intuitively as
“which relevant world(s).” Indeed the specifier-position argument could be under-
stood as an implicit ‘whether’ analogous to the relative phrase in SpecCPrel (see e.g.
Larson 1985, Kayne 1991 on ‘whether’ being in SpecCP). Roughly put, the relative
complementizer Crel (e.g. ‘that’) is to ‘if ’, as an individual relative phrase (e.g. ‘what’)
is to ‘whether’, as reflected in (7.17).

(7.17) a. Every baby {which, that, ∗ which that} laughed is happy.

b. Alice knows {whether, if, ∗ whether if} it rained.

In §6 I suggested analyzing relative pronouns as choice-function pronouns. A natu-

ral move is to analyze the ‘whether’-like element in an ‘if ’-clause likewise as a choice
function pronoun of worlds — a function s ∈ D⟨st,t⟩ s.t. p(s(p)) for any pst ((6.22)).
(For clarity I distinguish ‘sf ’ for choice-function indices/variables of type ⟨st, s⟩; I
continue to use ‘cf ’ with choice functions of type ⟨et, e⟩.) As above, the pronoun
is (obligatorily) coindexed with the conditional complementizer’s assignment ar-
gument, and the external ‘whether’-like argument supplies a particular world from
among the set of contextually relevant worlds (=h∗ (2sf)(gc (1st))). The assignment-
quantification introduced by the raised pluralization operator PD⟨2,a⟩ binds the choice-
function pronoun, here [wh2sf g2 ], and quantifies over possibilities in the restricted
set that satisfy the property denoted by the ‘if ’-clause C. The compositional seman-

142
tics derives that the conditional free relative (7.16) ‘if it rained’ picks out (roughly)
the sum of relevant possibilities in which it rained — the unique maximal possibility
h∗ ∈ G s.t. it rained in @(h), where @(h) is identical to some or other selected
relevant u ∈ gc (1st) ∈ W.⁸⁷
The revised syntax/semantics for ‘if ’-clauses affords a unified assignment-variable-
based analysis of relativization. Free relative ‘if ’-clauses are analyzed as intensional
analogues of free relatives of individuals. (I return to the extent of the similarities in
§§7.7, 8.)
Note that the semantic type of the ‘if ’-clause CP is derived to be type ⟨t, t⟩ — the
same type of the adjoined DPs in our examples with inverse linking and genitive
binding from §6.3.3 (n. 83). We will see that function composition and the ⟨t, t⟩-
type of the ‘if ’-clause can similarly be exploited to help satisfy our desiderata from
§7.1 — e.g., capturing the spectrum of (non-)shifted readings with sentence-initial
‘if ’-clauses as well as with sentence-final ‘if ’-clauses adjoined to VP and with NP-
adjoined ‘if ’-clauses in adnominal conditionals.
7.3 Adnominal conditionals
Though rarely considered in semantics literatures on conditionals, let’s start with
adnominal conditionals such as (7.18), where the ‘if ’-clause adjoins to the subject
NP (Lasersohn 1996). (For simplicity assume that ‘the’ has a standard ⟨et, ⟨et, t⟩⟩
type, and treat ‘expensive’ as a simple predicate.)

⁸⁷Strict conditional and material conditional interpretations could be understood as cases where
the semantic value for [p g] in the specifier argument is ultimately identified with W vs. @(gc ), re-
spectively. Extensions to aspect, mood, tense may afford resources for an assignment-variable-based
treatment of counterfactuals. It would be interesting to examine whether our parallel treatments of
the syntax for individual relative clauses and conditional clauses — in particular the extra structure
afforded in the external argument of the complementizers — might provide a locus for capturing
parallel crosslinguistic phenomena of modal (non-temporal) interpretations of past morphology
in DPs (in languages with nominal tense) and counterfactuals (Iatridou 2000, Palmer 2001,
Nordlinger & Sadler 2004, Lecarme 2008, Adamou 2011, Romero 2014).

143
(7.18) Adnominal conditional (NP-adjunction)
‘The fine if you drive is expensive.’
S∶t

T⟨1,a⟩ CP

⟨1,s⟩

Cdec t1a
DP
expensive t1s
the NP ∶ ⟨e, t⟩

⟨2,a⟩
NP1 ∶ ⟨e, t⟩ CP1 ∶ ⟨t, t⟩
fine w1 g2
PD⟨2,a⟩

⟨2,s⟩
wh2sf -g2 p1st -g1
if t2a you2 -g1 drive-t2s
⟨2,a⟩
JCP1 K ≈ λTt .λg . [ιa(g)∗ ∶ g(1a)(2e) drives in @(a(g))
∧ @(a(g)) = a(g)(2sf)(g(1a)(1st))] T(g[a(g)/2a])
⟨2,a⟩
JNPK = JCP1 K ○ JNP1 K
⟨2,a⟩
= λxe .JCP1 K(JNP1 K(x))
≈ λxe .λgg . [ιa(g)∗ ∶ g(1a)(2e) drives in @(a(g))
∧ @(a(g)) = a(g)(2sf)(g(1a)(1st))] x(g[a(g)/2a]) is a fine in a(g)(1s)
JSK ≈ λgg . [ιx(g)∶ [ιa(g)∗ ∶ g− (2e) drives in @(a(g))
∧ @(a(g)) = a(g)(2sf)(g− (1st))] x(g) is a fine in a(g)(1s)]
x(g) is expensive in @(g− )

144
This says, roughly, that the fine in the maximal relevant possibility h∗ where you
drive is expensive in the actual world @(gc ). The modified subject DP picks out
the unique o ∈ E s.t. o is the fine in the world (=h(1s)) of the maximal relevant
possibility where you gc (2e) drive — i.e., the maximal possibility in which you drive
and which is identical to some or other possibility h(2sf)(gc (1st)) in the contextually
relevant domain gc (1st). The sentence is true iff o is expensive in the world of the
discourse @(gc ).
As in simple sentences, the main clause complementizer raises from the world-
argument position of ‘expensive’; and the topmost assignment-binder raises from
the assignment-argument position, linking the modal domain for the main clause
to the world of the discourse via the definition of truth-in-a-context. The world
argument of the subject predicate ‘fine’ is supplied by a world pronoun. However,
the assignment binder-index on the ‘if ’-clause shifts the coindexed world pronoun
[w1 g2 ] to the world of the assignment representing the hypothetical possibility,
i.e. the maximal relevant possibility a(g)∗ where you drive. The ‘if ’-clause CP is
type ⟨t, t⟩, the predicate is type ⟨e, t⟩, and they combine via function composition to
yield the modified subject NP of type ⟨e, t⟩.
The §7.2-semantics compositionally derives the local reading of the subject NP’s
world pronoun in (7.18). Yet nothing said thus far excludes an LF where the world
pronoun is coindexed with the topmost assignment-binder, e.g. [NP1 fine w1 g1 ], rep-
resenting an unattested reading implying that the fine in the actual world is ex-
pensive. Although the complex subject NP needn’t receive a local reading in an
embedded adnominal conditional, as reflected in (7.19), the interpretation of the
world pronoun must still shift, it seems, with the adjoined ‘if ’-clause.

(7.19) Alice thinks that the venue if it rains is going to close before the party.
a. de dicto ≈ Alice thinks there is some venue or other that will be the venue
in the possibility where it rains, and it, whatever it is, is going to close
before the party
b. de re ≈ the venue in the possibility where it rains is such that Alice thinks
it is going to close before the party

It would be implausible to derive the (apparent) obligatory shifted reading by

treating the ‘if ’-clause as moving from the NP’s world argument, since the ‘if ’-clause
is base-generated in an adjoined position. A more promising approach may be to
understand LFs representing global readings as generally excluded for conversa-
tional reasons, failing to represent cooperative uses. For instance, in an LF for (7.18)
where NP1 ’s world pronoun is bound by T⟨1,a⟩ , the ‘if ’-clause would be trivial in the

145
interpretation of the complex subject and the sentence. An interpretive principle
against such semantically trivial modifications might be represented roughly as fol-
lows:⁸⁸

(7.20) Non-triviality principle:

Let i be a set of binder-index features {⟨i1 , τ1 ⟩, ⟨i2 , τ2 ⟩, . . . }, and γ be a branch-
ing node in the domain of J⋅K whose daughters are α and β i , where β i oc-
cupies an A-position. For any binder-index feature k ∈ i that is semantically
necessary in γ , there are g′ ≈k g′′ such that JαK(⋯)(g′ ) ≠ JαK(⋯)(g′′ ).
a. h ≈k h′ iff h and h′ are otherwise identical except that h(k) ≠ h′ (k)
b. k is semantically necessary in γ iff, for γ ∗ a branching node whose daugh-
ters are α and β i/{k} , γ ∗ is not in the domain of J⋅K .
⁸⁸An interpretive principle along the lines in (7.20) needn’t be an exceptionless grammatical
constraint. For instance, there are wh-ever adjuncts such as in (i) which don’t antecede a pronoun.
(i) Whatever Bert did/does, Chip was/is/will be happy.
Though the adjuncts in (i) arguably might still modify some other operator (temporal, generic,
modal) in the main clause, there are likely other exceptions. Indeed relevance conditionals — like
“frame-setting” topics such as (ii) (Reinhart 1981) — may provide precisely such a case. Though
it is contentious how relevance conditionals ought to be analyzed, a natural approach in the present
framework might be as follows:
(ii) Frame-setting topic
As for cats, I like Fluffy.
(iii) Biscuit conditional (CP-adjunction)
a. ‘If you’re hungry, I baked’
b. [S T⟨1,a⟩ [CP [PD⟨2,a⟩ [wh2sf -g2 p1st -g1 [[if t2a ]⟨2,s⟩ you2 -g1 hungry-t2s ]]]⟨2,a⟩
[CP [Cdec t1a ]⟨1,s⟩ I1 -g1 baked-t1s ]]]
c. JSK ≈ λgg . [ιa(g)∗ ∶ g− (2e) is hungry in @(a(g)) ∧ @(a(g)) = a(g)(2sf)(1st)]
g− (1e) baked in @(g− )
≈ S is true in c iff the maximal relevant possibility where you g−c (2e) are hungry — the
maximal possibility where you’re hungry and which is included in the contextually relevant
domain g−c (1st) — is s.t. I g−c (1e) baked in the actual world @(g−c )
Like in simple sentences the topmost assignment-binder raises from the assignment-argument posi-
tion of the main clause complementizer, linking the modal domain for the main clause to the world
of the discourse via the definition of truth-in-a-context. The LF and semantic value in (iii) thus
derive common intuitions about relevance conditionals, such as that they involve an assertion of
the consequent, and independence of the antecedent and consequent: The ‘if ’-clause introduces a
modal topic — the possibility that you gc (2e) are hungry — but then fails to comment on it in the
main clause; S implies that I gc (1e) actually baked. For present purposes I simply assume that (7.20)
provides a sufficiently robust interpretive principle as to capture the modifying function of ‘if ’-clauses
in adnominal conditionals.

146
Intuitively put, this says that  binder-indices necessary for interpretability in a given
step in a derivation must do non-trivial semantic work in that step in the derivation.
(7.20) is of course directly satisfied in cases of QR or wh-movement, where the
moved expression necessarily binds its trace. With base-generated adjuncts such
as ‘if ’-clauses, (7.20) amounts to a norm that the modifier genuinely modify the
expression to which it is adjoined.⁸⁹ In an adnominal conditional such as (7.18)
the ‘if ’-clause’s assignment-binder is “semantically necessary” in the sense of (7.20);
without it, CP and NP1 would be types ⟨at, t⟩ and ⟨e, t⟩, respectively, and hence the
complex NP node would be uninterpretable. So, lest the non-triviality principle be
violated, the world-pronoun must be coindexed with the ‘if ’-clause’s assignment-
binder, representing the shifted reading.
7.4 Sentence-final ‘if ’-clauses
The account of ‘if ’-clauses combining with subject predicates carries over to con-
ditionals with sentence-final ‘if ’-clauses that adjoin to verbal predicates. First, note
that treating the world argument of the main predicate as supplied by a world trace
coindexed with the higher complementizer would fail to capture the role of the ‘if ’-
clause in shifting the modal domain:

(7.21) a. [S T⟨1,a⟩ [CP [Cdec t1a ]⟨1,s⟩ [ Alice [VP [VP1 won t1s ] [CP1 if-clause]⟨2,a⟩ ]]]]
b. JVP1 K = λxe .λgg . x(g) won in g(1s)
JVPK = λxe .λgg . [ιa(g)∗ ∶ . . . ] x(g[a(g)/2a]) won in g[a(g)/2a](1s)
JCPK ≈ λgg . ∀w s.t. w(g) = @(g(1a)), [ιa(g)∗ ∶ . . . ] Alice won in w(g)
JSK ≈ λgg . [ιa(g)∗ ∶ . . . ] Alice won in @(g− )

A crucial move in §3 was to treat complementizers such as ‘that’ as base-generated

in the world-argument position of the clause’s main predicate, and as raising for
type reasons as quantifiers over worlds. The approach to the syntax/semantics in-
terface helped capture (among other things) the apparent obligatory local reading
of the predicate’s world argument. Yet we needn’t assume that all occurrences of
complementizers necessarily undergo raising from their base-generated position.
The relevant modal domain for interpreting the main predicate in examples with
VP-modifiers isn’t a domain directly determined by the embedding assignment-
quantifier — e.g., T⟨1,a⟩ , determining @(gc ) — but rather a domain as modified by
the adjunct.
Accordingly, I suggest that the main predicate’s world argument in such ex-
⁸⁹See below for related discussion regarding the proform binding requirement in correlatives.

147
 amples may be supplied by a world-pronoun, rather than a world-trace left from
movement of the complementizer. Coindexing between the world pronoun and the
assignment-binder on an adjoined ‘if ’-clause captures the role of the ‘if ’-clause in
modifying the VP and shifting the modal domain, as derived below. Modal ele-
ments, including the topmost assignment-binder, can still be treated as raising from
an assignment-argument position; however, unlike the examples considered thus
far, the complementizer won’t be what provides the source of intensionality. The
embedding C0 can be treated as denoting the identity function and base-generated
in its position at LF. (We will see additional applications and motivations for the
vacuous complementizer shortly.)
I offer the following LF for (7.7d), assuming the context from Santorio 2012 in
(7.6) where the pronoun ‘I’ receives a local reading. (Note that the vacuous comple-
mentizer, written C∅ , is distinct from the unpronounced declarative complemen-
tizer Cdec , which is still given its substantive lexical entry.)

(7.22) Sentence-final ‘if ’-clause. Shifted indexical. (VP-adjunction)

‘I am in Widmore Library if it landed heads.’
S∶t

T⟨1,a⟩ CP

IP
C∅ t1a

I1 g2 VP ∶ ⟨e, t⟩

⟨2,a⟩
VP1 ∶ ⟨e, t⟩ CP1 ∶ ⟨t, t⟩

in-WL w1 g2 PD⟨2,a⟩

⟨2,s⟩
wh2sf -g2 p1st -g1
if t2a it2 g1 heads t2s

148
 JIPK = (JCP1 K ○ JVP1 K)(λgg .g(2a)(1e))
≈ [λxe .[λTt .[λgg .[ιa(g)∗ ∶ g(1a)(2e) landed heads in @(a(g))
∧ @(a(g)) = a(g)(2sf)(g(1a)(1st))] T(g[a(g)/2a])]]
(λgg .x(g) is in WL in g(2a)(1s))] (λgg .g(2a)(1e))
≈ λgg . [ιa(g)∗ ∶ g(1a)(2e) landed heads in @(a(g)) ∧ @(a(g)) = a(g)(2sf)(g(1a)(1st))]
a(g)(1e) is in WL in a(g)(1s)
JC∅ K = λaa .λTt .λgg . T
JSK ≈ λgg . [ιa(g)∗ ∶ g(2e) landed heads in @(a(g)) ∧ @(a(g)) = a(g)(2sf)(g− (1st))]
a(g)(1e) is in WL in a(g)(1s)

The conditional is true iff the maximal relevant possibility h∗ ∈ G where the contex-
tually relevant object gc (2e) landed heads, @(h) = h(2sf)(gc (1st)), is s.t. the shifted
counterpart individual h(1e) is in WL in the world of that possibility h(1s).
The pronoun ‘it’ in the ‘if ’-clause is interpreted with respect to the discourse
assignment, while the predicate ‘landed-heads’ receives an obligatory local reading
derived via movement of the complementizer ‘if ’. As with the adnominal conditional
in (7.18), the ‘if ’-clause combines with the adjoined VP1 via function composition,
yielding the complex VP of type ⟨e, t⟩. The world argument of the predicate ‘be-in-
WL’ in VP1 is supplied by the world pronoun [w1 g2 ]. Given the sort of interpretive
principle in (7.20), LFs where the world pronoun isn’t coindexed with the adjoined
‘if ’-clause’s assignment binder will be generally excluded. The ‘if ’-clause thus shifts
the interpretation of the adjacent world-pronoun, capturing the local reading. Cru-
cially, although the ‘if ’-clause doesn’t combine at the level of the main clause, the
subject pronoun ‘I’ can still receive a local reading under the supposition. The shifted
reading of the subject is compositionally derived from the general analysis of ‘if ’-
clauses and function composition. (As in our examples with modals, one might
invoke additional principles specifying that the syntactic indexing determined by
the concrete discourse be such that a(g)(1e) represents an epistemic counterpart of
gc (1e), who is the speaker of c (§§3–4).)
7.5 Sentence-initial ‘if ’-clauses
Conditionals with sentence-initial ‘if ’-clauses may be treated as adjoined to IP or CP
(cf. Iatridou 1991, von Fintel 1994, Izvorski 1996, Bhatt & Pancheva 2006).
Consider the following alternative LFs, again on the relevant local reading of ‘I’ from
Santorio 2012. (The ‘if ’-clause CP in (7.24) is as in (7.23).)

149
(7.23) Sentence-initial ‘if ’-clause (CP-adjunction)
‘If it landed heads, I am in Widmore Library.’

S∶t

T⟨1,a⟩ CP

CP ∶ t
C∅ t1a

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ CP2 ∶ t

PD⟨2,a⟩ ⟨1,s⟩

I1 -g2 in-WL-t1s
Cdec g2
⟨2,s⟩
wh2sf -g2 p1st -g1
if t2a it2 -g1 heads-t2s

JCP2 K = λgg . ∀w s.t. w(g) = @(g(2a)), g(2a)(1e) is in WL in w(g)

JSK ≈ λgg . [ιa(g)∗ ∶ g− (2e) landed heads in @(a(g))∧@(a(g)) = a(g)(2sf)(g− (1st))]
∀w s.t. w(g) = @(a(g)), a(g)(1e) is in WL in w(g)
(7.24) Sentence-initial ‘if ’-clause (IP-adjunction)
a. [ S T⟨1,a⟩ [ CP [C∅ t1a ] [ IP [ CP1 . . . ]⟨2,a⟩ [ IP I2 -g2 in-WL-w1 -g2 ]]]]
b. JSK = J(7.22)K ≈ λgg . [ιa(g)∗ ∶ . . . ] a(g)(1e) is in WL in a(g)(1s)

Roughly, in both (7.23)–(7.24), the conditional is true iff the maximal relevant pos-
sibility h∗ ∈ G where gc (2e) landed heads, @(h∗ ) = h∗ (2sf)(gc (1st)), is such that
the first-positioned individual h∗ (1e) is in WL in the world of that possibility.
The semantic value in (7.24) where the ‘if ’-clause adjoins to IP is identical to
the semantic value in (7.22): The world argument is again supplied by the world
pronoun [w1 g2 ] which receives its interpretation from the adjoined ‘if ’-clause; and
the derived modal domain is a(g)(1s), which, given our metasemantic assumptions,
represents the world of the possibility represented by a(g), i.e. @(a(g)). By contrast,

150
in (7.23) the modal domain for interpreting the main clause is explicitly identified
with @(a(g)). This represents an important point about the LFs.
In (7.23) the main predicate’s world argument is, as usual, supplied by a world-
trace left from movement of the main clause complementizer. Yet unlike the exam-
ples considered thus far, the complementizer’s assignment argument is supplied by
an assignment-variable coindexed with the ‘if ’-clause. The topmost assignment-
binder raises from the assignment argument position of C∅ , as in (7.22)/(7.24).
However, in (7.23), where the ‘if ’-clause adjoins to a full CP, there is an indepen-
dently motivated mechanism for generating the vacuous complementizer embed-
ding the conditional: CP-recursion (Iatridou & Kroch 1993; see e.g. Vikner 1995,
Browning 1996, Nyvad et al. 2017 for general discussion). Conditionals with
sentence-initial ‘if ’-clauses — and correlative constructions generally (de Vries 2002) —
follow the complementizer in attitude ascriptions.

(7.25) Alice thinks [ CP1 that [ CP2 [if it snows] [ CP3 (then) school will close]]]

Treating the complementizer embedding the conditional as semantically vacuous

satisfies the requirement for CP-recursion in Iatridou & Kroch 1993 that any
content of the higher complementizer be recoverable from the content of the lower
complementizer. C∅ may be pronounced, as in (7.25), where what raises from the
assignment argument position in C0 is the embedding attitude verb, or implicit, as
in (7.23), where what raises is the topmost assignment-binder T.
Hypothetical conditionals (in the sense of Iatridou 1991) can be understood as
a type of correlative construction, where the free relative clause obligatorily binds a
correlate/proform in the main clause (Geis 1970, Iatridou 1991, von Fintel 1994,
Bittner 2001, Schlenker 2004, Bhatt & Pancheva 2006, Arsenijević 2009).

(7.26) Correlatives / Proform binding

[IP/CP [CP free relative clause]i [IP/CP … correlatei …]]

Syntactic work on correlatives indicates that fronted correlative proforms must be

structurally adjacent to the sentence-initial relative clause (Bhatt & Pancheva 2006,
den Dikken 2009, Lipták 2012).⁹⁰ This requirement helps capture the apparent
⁹⁰One way of formalizing the relevant notion of adjacency might be that β is structurally adjacent
to α iff (i) α c-commands β , and (ii) there is a non-branching path from β which terminates at
the sister of α. (One could say that α and β are structurally adjacent iff α is structurally adjacent
to β or β is structurally adjacent to α.) For broader discussion of locality effects in correlatives
crosslinguistically, see Dayal 1996, Bhatt 2003, Davison 2009a, Leung 2009, Lipták & Rebuschi
2009, Lipták 2012, Beshears 2017.

151
obligatory local reading of the main clause’s modal domain in a hypothetical ‘if…’
conditional, even if the ‘if ’-clause is a base-generated adjunct. For instance, the
correlative ‘if ’-clause in (7.23) binds the assignment-variable in the main clause
complementizer, shifting the modal domain to that of the topical antecedent pos-
sibility.
It is standard in analyses of ‘if ’-clauses as free relatives to treat conditional ‘then’
as the correlate proform in correlative constructions.

(7.27) [If it snowed,]i theni it was cold.

Not all sentence-initial ‘if…’ conditionals are compatible with ‘then’, e.g.:

(7.28) a. If John is dead or alive, (#then) Bill will find him.

b. Even if John is drunk, (#then) Bill will vote for him.
(Iatridou 1994: exs. 6,10)
(7.29) a. What does John think that if his mother comes (∗ then) the guests will
eat?
b. How/where did Mary say that if her mother visits (∗ then) the car will be
fixed?
(Iatridou & Kroch 1993: exs. 58–59)

Such data lead Bhatt & Pancheva 2006 to deny that hypothetical ‘if…’ condition-
als without ‘then’ are correlatives. This strikes me a prima facie cost. Like other
correlatives, hypothetical ‘if…’ conditionals without ‘then’ consist of a sentence-
initial free relative clause which shifts the interpretation of an element in the main
clause. Unlike relevance conditionals, which intuitively predicate the consequent
of the actual world (cf. n. 88), hypothetical conditionals shift the modal domain
with or without ‘then’. The main clause correlate needn’t be overtly expressed —
in conditional correlatives, or in certain correlatives of individuals such as (7.31b)
(§6.2; see also Iatridou 2013 on extensions to conditional correlatives). The re-
lation between relevance conditionals and hypothetical conditionals with/without
‘then’ is analogous in this respect to the relation between “frame-setting” topics
and individual correlatives (“aboutness” topics) with/without a proform (Reinhart
1981; cf. Lipták 2012).

(7.30) Relevance conditionals vs. Hypothetical ‘if…(then)’ conditionals

a. If you’re cold, I have a jacket.
b. If you’re cold, (then) you should shut the window.

152
(7.31) Frame-setting topics vs. Individual correlatives
a. As for music, I like jazz.
b. [jo laRkii khaRii hai] (vo) lambii hai
REL girl standing is (DEM) tall is
‘The girl who is standing is tall.’ (Srivastav 1991, Bhatt 2003; Hindi)

The analysis in this section captures the connection between hypothetical ‘if…’
conditionals and correlatives: the free relative ‘if ’-clause is coindexed with the assignment-
variable determining the modal domain relevant for interpreting the main clause.
Treating hypothetical conditionals generally as correlatives is compatible with ac-
knowledging a contrast between conditionals with and without ‘then’. Overtly ex-
pressing the conditional proform requires raising it from its base position and topi-
calizing it (cf. Izvorski 1996, Arsenijević 2009, Lipták 2012), as reflected in (7.32)
where ‘then’ occupies its own projection in SpecCP (Iatridou 1991, Iatridou &
Kroch 1993, Collins 1998).

(7.32) [ S … [ CP [ CP if-clause]⟨i,a⟩ [ CP [ XP then-Xi ]k [ C . . . tk . . . ]]]]

Explicitly topicalizing the proform can have distinctive interpretive and syntactic
effects. It is often claimed that ‘If p, then q’ conditionals carry an implication to the
effect that some/all ¬p-possibilities are ¬q-possibilities (cf. Iatridou 1991, 1994,
von Fintel 1994).

(7.33) If the weather is good, Alice will win

↝ In some possibility where the weather isn’t good, Alice won’t win

Izvorski 1996 shows that this exhaustiveness implication associated with condi-
tional ‘then’ is shared among correlative proforms generally. In correlatives of in-
dividuals the proform is also interpreted at LF in a topicalized position in the left
periphery of the main clause (also Lipták 2012). It isn’t implausible that focusing
the moved proforms in this way is what leads to the apparent exhaustiveness impli-
cations — hence the exclusion of ‘then’ in (7.28) where the antecedent is already ex-
haustive. Topicalizing the proform can also have syntactic implications, as in (7.29).
A-raising the proform to SpecCP excludes further argument or adjunct extractions
out of the main clause (Iatridou 1991, Iatridou & Kroch 1993, Collins 1998,
Bhatt & Pancheva 2006, Lipták 2012; cf. Dayal 1996, Izvorski 1996).
So, the assignment-variable-based syntax/semantics for hypothetical condition-
als in this section unifies ‘if…(then)’ with correlatives in the sense that there is a
left-adjoined free relative clause which binds a correlate in the main clause. In con-

153
ditionals with overt ‘then’ the proform is topicalized and raises to SpecCP of the
main clause. We can capture the idea that ‘if ’-clauses function notionally as topics
(Haiman 1978, Bittner 2001) without analyzing all conditional constructions as
involving movement via topicalization. (I return to analyses of ‘then’ and correlative
proforms in §7.7.)
7.6 Modalized conditionals: Restricting and shifting
So far we have focused on “bare” conditionals — conditionals without an overt op-
erator in the main clause. Although there may be evidence for a covert operator
in some bare conditionals (epistemic, generic, frequency; Lewis 1975, Kratzer
1991), the account in this section captures the function of ‘if ’-clauses in shifting
a modal domain without needing to posit an additional operator. Let’s turn now
to modalized conditionals and consider the role of ‘if ’-clauses in modifying the
interpretation of a modal. (I focus on hypothetical conditionals with sentence-initial
‘if ’-clauses.)
7.6.1 Direct restriction
The traditional Kratzerian line is that the ‘if ’-clause in a modalized conditional such
as (7.34) restricts the domain of the modal.

(7.34) If it rains, he might cry.

≈ for some epistemic possibility in which it rains, he cries

A prominent approach to quantifier domain restriction is to analyze quantifier phrases

as interpreted with respect to a contextually supplied resource domain (n. 28). Adapt-
ing von Fintel 1994, one option is thus to treat modal verbs as taking a resource do-
main argument in addition to the modal background representing the reading of the
modal, i.e. epistemic, deontic, etc. A revised lexical entry for ‘may’ is in (7.35). (The
structure of the modal’s restrictor argument is modified as well to give the modal
a “world-invariant” entry analogous to determiner quantifiers (cf. von Fintel &
Heim 2011). As previously I use ‘r’ in indices for type ⟨s, at⟩.)

(7.35) JmayK = λaa .λAat .λA′at .λgg . ∃a′a s.t. a′ (g) ≤ a(g) ∧ A(a′ )(g)∶ A′ (a′ )(g)
(7.36)
may g r g ts

In modalized conditional correlatives, the correlative binding requirement can be

satisfied by coindexing the resource domain with the ‘if ’-clause, as reflected in (7.37).

154
 (Ignore any contextual restriction in the specifier argument of ‘if ’, and assume W
represents the domain of worlds.)

(7.37) Modalized conditional (IP-adjunction)

‘If it rains, he may cry.’
S∶t

T⟨1,a⟩ CP ∶ t

⟨1,s⟩ IP ∶ t

Cdec t1a

CP⟨2,a⟩ ∶ ⟨t, t⟩ IP ∶ t

PD⟨2,a⟩ ⟨3,a⟩

⟨3,s⟩
⟨2,s⟩
wh2sf -g2 W he3 -g1 cry-t3s
rains t1s may g2 r1 g1 t1s Cdec t3a
if t2a

JIPK ≈ λgg . ∃a′ s.t. a′ (g) ≤ g(2a) ∧ g(1a)(1r)(g(1s))(a′ (g))∶

g(1a)(3e) cries in @(a′ (g))
JSK ≈ λgg . [ιa(g)∗ ∶ it rained in @(a(g)) ∧ @(a(g)) = a(g)(2sf)(W)]
∃a′ s.t. a′ (g) ≤ a(g) ∧ g− (1r)(@(g− ))(a′ (g))∶ g− (3e) cries in @(a′ (g))

This treats the modalized ‘may’ conditional as saying that the maximal relevant pos-
sibility h∗ in which the antecedent is verified is such that some accessible possibility
that is a part of h∗ verifies the consequent. Roughly put, (7.37) is true iff the maximal
relevant possibility where it rains h∗ is such that the contextually relevant individual
gc (3e) cries in some possibility h′ that is a part of h∗ and is included in the set of ac-
cessible possibilities, as determined by the contextually relevant modal background
gc (1r) at the actual world (gc ).
The modal ‘may’ in the consequent is interpreted with respect to the same con-

155
textually relevant domain of possibilities as would figure in the interpretation of
an unembedded ‘may’ sentence: the relevant reading is supplied by the discourse
assignment, and the modal background is indexed to the actual world. The role
of the ‘if ’-clause is simply to restrict this domain to possibilities where it rained.
As von Fintel notes, the relation between the ‘if ’-clause and the modal’s restrictor
argument in such structures is apparently stronger than an optional binding relation;
von Fintel suggests analyzing it as an A-chain (1994: 88–89; cf. Dayal 1996). For
instance, in nested conditionals such as (7.38) the modal must receive the restriction
of the closest ‘if ’-clause (cf. Bhatt & Pancheva 2006).

(7.38) If you get back in time for the show, then [if Timmy isn’t tired]i we’ll have toi
take him.

The close connection between the topical possibility described by the ‘if ’-clause
and the modal’s resource domain can be captured by the locality requirement on
correlative correlates discussed above. This general requirement suggests a new way
of understanding the correlate in modalized hypothetical ‘if…’ conditionals: The
structurally adjacent element bound by the ‘if ’-clause in (7.37) is [may g2 ] (n. 90).
The binder-index on the ‘if ’-clause is coindexed with the assignment-variable sup-
plying the resource domain argument of the modal.
7.6.2 Indirect restriction/modification: “Double modal” and “information-sensitive”
readings
The above analysis captures both the correlative binding requirement and the re-
stricting function of the ‘if ’-clause via the modal’s resource domain. In (7.37) the
modal’s resource domain variable supplies the correlate, the correlative ‘if ’-clause
combines with IP, and (as in §§3–4) the embedding complementizer raises from
the modal’s world-argument position. One might wonder what predictions would
follow if the complementizer raised locally under the ‘if ’-clause, as in (7.39).

156
(7.39) Modalized conditional (CP-adjunction)
S∶t

T⟨1,a⟩ CP ∶ t

CP ∶ t
C∅ t1a

CP⟨2,a⟩ ∶ ⟨t, t⟩ CP ∶ t

…if…
⟨3,s⟩ IP

Cdec g2 ⟨3,a⟩ CP

may g t3s . . . t3a . . .

r g
Just as with the bare conditional in (7.23), the complementizer adjacent to the ‘if ’-
clause supplies the correlate, and the modal domain for the (here modalized) main
clause is determined by the ‘if ’-clause. I suggest that such LFs can represent so-called
“double modal” and “information-sensitive” readings of modalized conditionals.
First, examples such as (7.40) have led some theorists to posit that at least some
modalized conditionals have a covert modal in addition to overt modal in the conse-
quent (Frank 1996, Geurts 2004, von Fintel & Iatridou 2005, Swanson 2010).

(7.40) If marijuana is illegal here, we have to report Alice. (cf. Geurts 2004: ex. 1)
≈ for every epistemically accessible world w′ where marijuana is illegal,
every world w′′ conforming to the law in w′ is s.t. we report Alice in w′′

In a Kratzerian restrictor approach, the ‘if ’-clause restricts the domain of the covert
(epistemic) necessity modal, and the overt deontic modal is evaluated at each world
in the covert modal’s restricted domain. The account in this section captures pu-
tative “double modal” readings without positing a covert modal, as reflected in the
simplified LF and derivation in (7.41). (Assume that in the intended interpretation

157
 r2 represents a modal background for the relevant laws. For simplicity in what
follows I will often ignore the modal’s potential resource domain argument.)

(7.41) “Double modal” reading

S∶t

T⟨1,a⟩ CP ∶ t

CP ∶ t
C∅ t1a

CP⟨2,a⟩ ∶ ⟨t, t⟩ CP1 ∶ t

PD⟨2,a⟩ ⟨3,s⟩ IP

Cdec g2
⟨2,s⟩
wh2sf -g2 W
⟨3,a⟩
if t2a pot illegal-t1s
⟨4,s⟩
have-to
t3s
r2 g1 Cdec t3a
we1 -g1 report-t4s Alice

JCP1 K ≈ λgg . ∀a′ s.t. g(1a)(2r)(@(g(2a)))(a′ (g))∶ g(1a)(1e) reports Alice in @(a′ (g))
JSK ≈ λgg . [ιa(g)∗ ∶ marijuana is illegal in @(a(g))∧@(a(g)) = a(g)(2sf)(W)]
∀a′ s.t. g− (2r)(@(a(g)))(a′ (g))∶ g− (1e) reports Alice in @(a′ (g))

This says, roughly, that the maximal possibility where marijuana is illegal h∗ is such
that for every possibility h′ compatible with the law in @(h∗ ) — i.e. every h′ ∈ gc (2r)(@(h∗ )) —
we report Alice in @(h′ ).
As in (7.23), the correlative ‘if ’-clause is coindexed with the assignment-variable
g2 in the adjacent complementizer, which determines the modal domain for evalu-
ating the main clause. The felt “double modal” interpretation follows from the fol-
lowing combination of features: (i) the modal’s modal-background pronoun [r2 g1 ]
receives its interpretation from the discourse assignment gc , as in (7.23); however,
(ii) the world applied to the modal background is the world of the antecedent-verifying
possibility @(h∗ ). The discourse context supplies the relevant function ⟨s, at⟩ from

158
worlds to sets of possibilities, gc (2r), which represents the law and determines the
intended deontic reading of the modal. What the law provides may vary across
worlds. In contrast to (7.23), the specific content of the law relevant for evaluating
the modalized main clause is the law in the possibility h∗ in which marijuana is
illegal. The ‘if ’-clause introduces a topical possibility, and the modal’s deontic modal
background is indexed to the world of that possibility.
The informal intuition in the literature on information-sensitivity is that the
deontic modal’s modal background seems to be updated, in some sense to be ex-
plained, in light of the information in the antecedent, as reflected informally in (7.42)
(e.g. Kolodny & MacFarlane 2010, Dowell 2012, Charlow 2013b, Cariani
et al. 2013, Silk 2014b, 2016).⁹1

(7.42) [Context: Ten miners are trapped in shaft A or shaft B, but we don’t know
which, and floodwaters are threatening. All ten miners will be saved if we
block the shaft they’re in, but all ten will drown if we block the wrong shaft.
One miner will drown if we block neither shaft.]
If the miners are in shaft A, we have to block shaft A.
≈ If the miners are in shaft A, then, given that information, the deontically
preferred worlds are worlds where we block shaft A. (Though, given our
actual information, the deontically preferred worlds are worlds where we
block neither shaft.)

The “shifty” interpretation in examples such as (7.40)–(7.41) was derived by apply-

ing the shifted worlds of the possibility introduced by the antecedent to the deontic
modal background supplied by the global context. I suggest that LFs in which the
modal-background variable receives its interpretation from an assignment-variable
bound by the ‘if ’-clause can represent information-sensitive readings. The putative
non-restricting function of the conditional clause is diagnosed as a local readings of
the modal’s modal-background pronoun, as reflected in (7.43).

⁹1On related phenomena with modalized conditionals generally, see Gillies 2010, Yalcin 2012,
Charlow 2013a. Given our purposes I ignore any differences between weak/strong necessity modals
in information-sensitivity, though see Silk 2013, Charlow 2013b.

159
 (7.43) “Information-sensitive” reading
S∶t

T⟨1,a⟩ CP ∶ t

CP ∶ t
C∅ t1a

CP⟨2,a⟩ ∶ ⟨t, t⟩ CP ∶ t

⟨3,s⟩
PD⟨2,a⟩ IP

Cdec g2
⟨2,s⟩ ⟨3,a⟩
wh2sf -g2 p1st -g1
miners in-A-t1s ⟨4,s⟩
if t2a
have-to
t3s
r3 g2 Cdec t3a we1 -g1 block-A-t4s

JSK ≈ λgg . [ιa(g)∗ ∶ the miners are in A in @(a(g)) ∧ @(a(g)) = a(g)(2sf)(g− (1st))]
∀a′ s.t. a(g)(3r)(@(a(g)))(a′ (g))∶ g− (1e) blocks A in @(a′ (g))

This says, roughly, that the maximal relevant possibility where the miners are in
shaft A, h∗ , is such that, for every possibility h′ compatible with the deontic ideal
determined by h∗ — i.e. every h′ ∈ h∗ (3r)(@(h∗ )) — we block shaft A in @(h′ ).
As in (7.41) the modal domain for evaluating the modalized main clause is set
by the assignment-variable g2 in the complementizer coindexed with the correla-
tive ‘if ’-clause. However, the deontic modal-background pronoun — here [r3 g2 ] —
receives a local reading as well. The deontic modal background for evaluating the
modal’s prejacent is determined by the topical antecedent possibility that the miners
are in shaft A. Crucially, this possibility may be a plural object. The plural possibility
a(gc )∗ represents a state throughout which the miners are in shaft A. Which deontic
ideal, set of accessible worlds, etc. is determined by a possibility may be due to a
feature of the possibility qua plurality. The discourse context may be such that gc (3r)
implies that we block neither shaft; yet its image under a, such that the miners are

160
in shaft A in the world of every part of the plurality a(gc )∗ , may determine a deontic
ideal a(gc )∗ (3r) implying that we block shaft A.
Some theorists have appealed to information-sensitive readings to motivate re-
vising traditional semantics for modals and conditionals. For instance, modals’
domains of quantification may be treated as determined relative to an additional
parameter such as an information state (e.g. Kolodny & MacFarlane 2010, Silk
2014b). Plural domains may provide the structure to capture intuitions behind cer-
tain revisionary approaches to information-sensitivity while still relativizing modal
backgrounds simply to worlds, as in the traditional semantics. An analysis of con-
ditionals as plural definite descriptions may thus be of general interest, independent
of the assignment-variable implementation developed here.
Let’s recap. This section has examined how an assignment-variable-based ap-
proach to local/global readings and §§3–6-accounts of modals and relative clauses
may be extended to conditionals. The syntax/semantics in §7.2 yields a uniform
analysis of ‘if ’-clauses in various positions and conditional structures — in sentence-
final positions adjoined to VP, in sentence-initial positions adjoined to IP/CP, and
in adnominal conditionals when adjoined to NP; in conditionals with/without a
proform such as ‘then’; and in conditionals with/without a main clause modal. ‘If ’-
clauses are analyzed syntactically as free relatives and interpreted as plural definite
descriptions of possibilities, represented by assignments. The account composition-
ally derives various types of local readings and ways ‘if ’-clauses may shift/modify
the interpretation of modals and other expressions, while still allowing for global
readings in the ‘if ’-clause and the rest of the sentence. In correlative constructions
the sentence-initial ‘if ’-clause binds a correlate in the main clause. The correlate may
be overtly expressed and topicalized such as in hypothetical ‘if…then’ conditionals.
7.7 Correlatives and proforms: Individual and conditional
The account in §§7.5–7.6 exploited connections between hypothetical ‘if…(then)’
conditionals and correlative constructions. Before moving on I would like to briefly
consider how the proposed assignment-variable-based syntax/semantics for ‘if ’-clauses
might be extended to a general treatment of correlatives and correlative proforms.
Crosslinguistic work demonstrates robust links among conditionals, interrog-
atives, and correlatives of individuals (see Dayal 1996, Bhatt & Pancheva 2006,
Lipták 2009b and references therein; cf. Caponigro 2003 on crosslinguistic con-
nections between individual free relatives and interrogatives). For instance, there
are systematic structural parallels between conditional and interrogative clauses,
and many languages (e.g. Bulgarian) use the same type of complementizer in ex-

161
pressing conditionals and questions. Although English doesn’t have correlatives
of individuals, many languages where correlativization is more productive use the
same type of marker in both conditionals and individual correlatives for introducing
the relative clause, and for the main clause proforms. In some languages the same
construction can be ambiguous between a conditional and individual correlative in-
terpretation, as in (7.44)–(7.45) (cf. Cable 2009 on Lhasa Tibetan, Davison 2009b
on Sanskrit).

(7.44) [Ako je (ko) već (ko) ustao] (taj) (onda) neka (taj) i izade.
if aux who already who raised that then let that and go.out
a. ‘[If anyone already stood up]i (theni ) let him also go out.’
b. ‘Whoeveri stood up, let himi also go out.’
(Arsenijević 2009: ex. 10; Serbo-Croatian)
(7.45) [Maliki-rli kaji-ngki yarlki-rni nyuntu] ngula-ju kapi-rna luwa-rni
dog same.top bite you DEM-top fut shoot
ngajulu-rlu.
me
a. ‘[If a dog bites you], then I’ll shoot it.’
b. ‘As for the dog that bites you, I’ll shoot it.’
(Bittner 2001: ex. 7; Walpiri)

Suppose as a working hypothesis that we give a uniform analysis of correlative

clauses in conditionals and individual correlatives. (We will return to interrogative
clauses in §8.) I suggest that what distinguishes conditional and non-conditional
interpretations of correlative clauses (cf. (7.44)) is the nature of the complementizer’s
external world argument and its relation to the main clause’s modal domain. In
conditional correlatives, the ‘if ’-clause introduces a topical modal possibility, and
the correlative binding requirement is satisfied by binding a correlate determining
the main clause’s modal domain. In individual correlatives the topical possibility
is still the possibility described by the correlative clause. However, rather than in-
troducing a modal possibility, the correlative clause introduces an actual possibility
about certain topical individuals. The modal domain for both clauses is the world
of the discourse: the external argument of the correlative clause complementizer is
identified with the actual world, and the main clause’s modal domain is determined
by the embedding modal/assignment-binder. The correlative binding requirement
is satisfied by binding an individual correlate in the main clause. Extending the

162
treatments of linguistic anaphora from §6, conditional and individual proforms alike
may be represented as copies of their antecedent relative/wh word.
7.7.1 Individual correlatives and proforms
An example with a multiple correlative of individuals is in (7.46).⁹2 (I use ‘Cif ’ for the
complementizer, where JCif K = JifK . The external argument could be represented
with [whsf -g pst -g], as in conditionals, where the semantic value for [p g] is ulti-
mately identified with {gc (1s)} = {@(gc )}. However, for clarity I represent the
external argument in individual correlative interpretations with a world-pronoun.)

(7.46) Correlative of individuals

[jis laRkii-ne jis laRke-ke saath khelaa] us-ne us-ko haraayaa.
REL girl REL boy with played DEM DEM liked
‘Which girl played with which boy, she defeated him.’ (Dayal 1996: 197)
⁹2I treat the main clause as an IP, but it could also be a CP. As in §§6, 7.2 with the relative
complementizer and ‘if ’, I assume that the relative word must have the same assignment-variable
as Cif (cf. Davison 2009a; more in §8). As previously I use abbreviations such as ‘babyu ’ for the
characteristic function of the set of individuals o ∈ E such that o is a baby in u.

163
 S∶t

T⟨1,a⟩ CP ∶ t

⟨1,s⟩ IP ∶ t

Cdec t1a

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ IP1 ∶ t

PD⟨2,a⟩

w1 g1 she2 -g2 girl-w1 -g1

defeated t1s
⟨2,s⟩ him3 -g2 boy-w1 -g1

Cif t2a

wh2 -g2 girl-w1 -g1

played-with t2s
wh3 -g2 boy-w1 -g1
JIP1 K ≈ λgg . g(2a)(2cf)(girlg(1a)(1s) ) defeated g(2a)(3cf)(boyg(1a)(1s) ) in g(1s)
⟨2,a⟩
JCP1 K ≈ λTt .λgg . [ιa(g)∗ ∶ a(g)(2cf)(girlg(1a)(1s) ) played with a(g)(3cf)(boyg(1a)(1s) )
in @(a(g)) ∧ @(a(g)) = g(1a)(1s)] T(g[a(g)/2a])
JSK ≈ λgg . [ιa(g)∗ ∶ a(g)(2cf)(girlg− (1s) ) played with a(g)(3cf)(boyg− (1s) ) in @(a(g))
∧ @(a(g)) = g− (1s)] a(g)(2cf)(girlg− (1s) ) defeated a(g)(3cf)(boyg− (1s) ) in @(g− )

Informally, this treats the correlative as saying that the assignment providing a true
answer to the question “which girli played with which boyj ?” verifies the matrix
clause statement “that girli defeated that boyj ”. The correlative clause describes the
unique possibility in which the selected girl o (=a(g)(2cf)(girlg(1s) )) played with the
selected boy o′ (=a(g)(3cf)(boyg(1s) )), where the world u of that possibility is s.t.
u = gc (1s). The sentence is true iff o defeated o′ in @(gc ).
First, the individual correlative clause in (7.46) is given the same structural rep-
resentation as an ‘if ’-clause. In (7.46) the world of the possibility introduced in the

164
correlative clause is specified as being identical to, gc (1s) — given our metasemantic
assumptions, the world identical to the world of the discourse @(gc ). The topical
possibility is the unique actual possibility in which a certain girl played with a certain
boy. The main clause says that that same girl defeated that same boy. The correlative
binding requirement is captured by (i) the syntactic representation of the proform as
a copy of its antecedent relative word and here elided NP,⁹3 and (ii) the assignment-
quantification introduced by the operator PD in the correlative clause.
In a slogan: Individual correlatives are material relevance conditionals with
donkey pronouns (nn. 87, 88) — (i) “material” insofar as the modal domain for the
correlative clause is restricted to the actual world gc (1s): the correlative clause de-
scribes a possibility in the actual world; (ii) “relevance” insofar as the modal domain
for the main clause is identified directly with @(gc ) via the syntax/semantics inter-
face: the main clause comments on the actual world; and (iii) “donkey” insofar as the
main clause proforms are anaphorically identified with their non-c-commanding
antecedents [wh-g2 P], via the correlative clause assignment-binder from PD⟨2,a⟩ .
The relevant similarity between correlative proforms and donkey pronouns is
that both are represented (roughly) as copies of a non-c-commanding linguistic
antecedent. Comparing them in this way isn’t to assimilate correlative constructions
to donkey sentences (see also n. 73; more on this below).⁹⁴ For instance, as discussed
⁹3Recall that the anaphoric expression need only be a copy with respect to interpreted material at
LF (§6.2, subject to the qualification in §6.2.3; see below). Possible differences such as with respect to
Case raise no issues (see Bhatt & Lipták 2009). I assume that [wh]-features are present for syntactic
and phonological reasons; correlate proforms needn’t be [+wh] or spelled-out as relative/wh-phrases.
The analysis is also compatible with there being differences in phi features in certain examples, as in (i)
where the plural proform is linked to both singular relative phrases. As in §6.2 for the case of donkey
pronouns, such examples may be understood analogously to other phenomena where pronouns’ phi
features are uninterpreted, such as with “fake indexicals”; see n. 51.
(i) [jo laRkiii jis laRkej se baat kar rahii hai] ve{i,j} ek saath sinemaa jaayeNge
which girl which boy with talk do-prog-pres they together movie go
‘Which girl is talking to which boy, they will go to the movies together.’ (Dayal 1996: 204)
⁹⁴Other types of non-c-command anaphora with conventionally anaphoric expressions, as in (i)
with Sinhala ee ‘that.anaph’, may be analyzed similarly. Though conventionally anaphoric pro-forms
are used as the obligatory correlates in correlatives (in languages that have them, as in (1.10) with
Bangla), they may also be used in environments not requiring a matching anaphoric expression, such
as donkey configurations, as in (i) (contemporary Sinhala lacks correlatives; Gair 1998: 306n.5).
(i) [Gunәpaalәgei pawulә gœnә wiswaasәyә nisaa] eyaai karәdәree wœṭuna.
Gunapala.gen wife about belief because that.anaph trouble fell
‘Because of Gunapala’si faith in his wife, hei fell into trouble.’ (Gair 1998: 134)

165
in §6.2, in Hindi a bare nominal can serve as a donkey pronoun but not the correlate
in a correlative.

(7.47) a. har aadmii jis ke paas koii gadhaai hotaa hai gadheKOi
every man REL with some donkey has donkey
maartaa hai.
beats
‘Every man who has a donkey beats it.’
b. *[jo aadmii aayaa]i aadmiii acchaa hai.
REL man came man nice is
‘Which man came, man is nice.’ (Srivastav 1991: ex. 45; Hindi)

Individual correlates are also often overtly realized as morphologically complex (see
§§6.2.1, 6.3.2). In some cases the overt NP in the correlate may differ from the
antecedent NP, typically denoting a contextual superset (e.g. Dayal 1996, Davi-
son 2009a, Beshears 2017) — hence the qualification above that the proforms are
analyzed as copies of the antecedent relative/wh word.
That said, there are interesting crosslinguistic similarities between donkey sen-
tences and correlatives. For instance, though Cable 2009 avoids giving “a ‘simple’
conditional analysis” (219n.7; 204–205) of Tibetan correlatives, he notes that the
correlatives are glossed literally as conditional donkey sentences:

(7.48) [Khyodra-s gyag gare nyos yod na] nga-s de bsad pa yin.
you-erg yak what buy aux if I-erg that kill perf aux
‘I killed whatever yak you bought.’
lit. ‘If you bought what/a yak, I killed that/it.’ (Cable 2009: 195; Tibetan)

Indeed correlative proforms are typically formed from a demonstrative — or, in lan-
guages which have them, a specialized anaphoric pronoun.⁹⁵ Individual correlates
also pattern with donkey pronouns, in contrast to bound-variable anaphors, in being
replaceable with an epithet, as reflected in (7.49) vs. (7.50)/(7.51).

⁹⁵It may be interesting to compare the proposed representation of correlative proforms with
Sportiche’s (2006) informal analysis of copies reconstructed from preposed/A-bar positions as
demonstratives, as in (i).
(i) a. ‘Je lui ai donné la photo que le vieux peintre m’a demandée.’
b. ‘I gave him the picture the old painter asked for yesterday.’
c. [The picture the old painter asked for]i I gave him this picturei
(Sportiche 2006: ex. 132; indexing added)

166
(7.49) Every donkey sued its/*the damn thing’s owner.
(7.50) Every farmer who had a donkey took it/the damn thing to court.
(7.51) [jis aadmii-se tum bahut pyaar-se baat kar rahe the]
REL man you much love talk do prog be
us aadmii-ne/haraamii-ne mujh-pe muqadma Thonk rakh-aa hai.
DEM man/bastard me-on court.case hammer.in keep be
‘Which man you were talking with so nicely, that man/bastard is suing me.’
(Potts et al. 2009: exs. 27, 29)

Finally, drawing on diagnostics from Sells 1987, Dayal observes that correlative
proforms contrast with resumptive pronouns (roughly, pronominally realized links
in a wh chain) in being compatible with non-specific readings (Srivastav 1991,
Dayal 1996). Whereas the resumptive pronoun ota ‘her’ in (7.52) can only be inter-
preted as referring to a particular individual who is the object of Dani’s seeking. the
correlative in (7.53) is compatible with a reading in which there isn’t any particular
thing needed by the speaker.

(7.52) dani yimca ’et ha’iša še hu mexapes ota

Dani will-find the woman that he seeks her
(Sells 1987: ex. 46, from Doron 1982; Hebrew)
(7.53) [jo ciiz mujhe caahiye thii] vo usKO mil gayii
REL thing I need was DEM he got
‘Which thing I needed, he got it.’ (Srivastav 1991: ex. 48a)

As we saw in §6.2.3, donkey pronouns are also compatible with non-specific read-
ings, as in (6.56)–(6.57) reproduced below.

(7.54) Every woman who has a secret admirer thinks he is stalking her.
(King 2004)
(7.55) Every epistemologist who got a painted mule thinks it ruined epistemology.

In light of examples such as (7.54) –(7.55) §6.2.3 argued that the choice-function
assignment-variable in the representation of a donkey pronoun needn’t be coindexed
with that in the antecedent. Examples such as (7.53) motivate a similar qualification
for correlative proforms. Unlike copies of movement, anaphoric proforms needn’t
be interpretively equivalent to their linguistic antecedents.
Although correlative clauses characteristically appear sentence-initially adjoined

167
to the matrix clause, in some languages certain correlative clauses can be base-generated
adjoined directly to the correlate DP, as reflected overtly in (7.56).

(7.56) Ram-ne [DP [jo laRkaa tumhaare piichhe hai]i us laRke-koi ] [DP [jo
Ram REL boy your behind is DEM boy REL
kitaab Shantiniketan-ne chhaapii thii]k vo kitaabk ] dii
book Shantiniketan print was DEM book give
‘Ram gave the book that Shantiniketan had published to the boy behind you.’
(lit. ‘Ram gave [[which book Shantiniketan had published] that book] to
[[which boy is behind you] that boy]’)
(Bhatt 2003: ex. 32; Hindi)

A key feature of the account of ‘if ’ conditionals in this section is that it yields a unified
semantics for ‘if ’-clauses in various structural positions, such as in conditionals
involving adjunction to NP, VP, IP, CP. This feature carries over to correlatives,
as reflected in the DP-adjunction structure in (7.57).⁹⁶

(7.57) Individual correlative (DP-adjunction)

[[jo laRkii khaRii hai] vo] lambii hai.
REL girl standing be DEM tall be
⁹⁶Dayal 1996 argues that such examples actually involve IP adjunction, but for illustrative
purposes I follow her in assuming an LF where the correlative clause forms a constituent with the
correlate. Bhatt 2003 argues that Hindi single correlatives are in general base-generated adjoined to
DP and optionally move via scrambling to adjoin to IP/CP in overt syntax. See also Beshears 2017.
For present purposes I treat ‘tall’ as a simple predicate and ignore any context-sensitivity associated
with the positive form; we will return to this in §8.

168
 ‘Which girl is standing, she is tall.’
S∶t

T⟨1,a⟩ CP ∶ t

⟨1,s⟩ IP ∶ t

Cdec t1a

DP ∶ ⟨et, t⟩ VP ∶ ⟨e, t⟩

tall t1s

⟨2,a⟩ DP1 ∶ ⟨et, t⟩

CP1 ∶ ⟨t, t⟩

PD⟨2,a⟩ she2 -g2 girl-w1 -g1

w1 g1
⟨2,s⟩

Cif t2a
standing t2s
wh2 -g2 girl-w1 -g1
JDP1 K = λPet .λgg . P(λgg .g(2a)(2cf)(girlg(1a)(1s) ))(g)
⟨2,a⟩ ⟨2,a⟩
JDPK = JCP1 K ○ JDP1 K = λPet .JCP1 K(JDP1 K(P))
≈ λPet .λgg . [ιa(g)∗ ∶ a(g)(2cf)(girlg(1a)(1s) ) is standing in @(a(g))
∧ @(a(g)) = g(1a)(1s)] P(λgg .g(2a)(2cf)(girlg(1a)(1s) ))(g[a(g)/2a])
JSK ≈ λgg . [ιa(g)∗ ∶ a(g)(2cf)(girlg− (1s) ) is standing in @(a(g))
∧ @(a(g)) = g− (1s)] a(g)(2cf)(girlg− (1s) ) is tall in @(g− )
⟨2,a⟩
The correlative clause CP1 describes the unique possibility in which the selected
girl o ∈ E (=a(g)(2cf)(girl1s )) is standing, where the world of that possibility is
identical to the world of the discourse g(1s) = @(gc ). The sentence is true iff that
selected girl o is tall in @(gc ).

169
 The type e individual proform can be lifted to type ⟨et, t⟩ in the usual way. Analo-
gous to our examples with NP-/VP-adjoined ‘if ’-clauses in §§7.3–7.4, the type ⟨et, t⟩
proform DP1 combines via function composition with the adjoined type ⟨t, t⟩ cor-
⟨2,a⟩
relative clause CP1 , yielding the main subject DP of type ⟨et, t⟩. The complex
subject denotes, roughly, the set of properties P characterizing the selected girl o =
a(g)(2cf)(girlg(1s) ) in the relevant actual possibility in which o is standing. As in
the clausal adjunction example in (7.46), the identification of the topical individual
picked out by the relative phrase with the individual picked out by the correlate
is captured by the representation of the anaphoric expression as a copy of its an-
tecedent, here [F2cf -g2 girl-w1 -g1 ], and the assignment-quantification/binder intro-
duced by the definite pluralization operator in the correlative clause.
To my knowledge the only compositional semantics for correlatives to explicitly
address DP-adjunction correlatives is Dayal 1996. Dayal gives the correlative clause
the semantics of a generalized individual quantifier (type ⟨et, t⟩), which, in clausal
adjunction examples, binds the proform — represented as a variable — in the main
clause. For DP-adjunction examples Dayal briefly suggests invoking a mechanism of
crosscategorial quantification (1996: 207). Very roughly, the proform variable DP1
is first lifted to type ⟨et, t⟩, yielding λPet .P(i); this is then converted to type ⟨e, t⟩
by adding a property variable Q and abstracting over the proform variable, yielding
λx.[λP.P(x)](Q) = λx.Z(Q). This combines with the type ⟨et, t⟩ correlative clause
to yield an object of type t for the subject DP, which is then converted to type ⟨et, t⟩
by abstracting over the previously introduced property variable Q, yielding roughly
λZet .Z(ιx∶ girl(x) ∧ standing(x)) for a DP such as in (7.57).
This is not compositional. Dayal’s account invokes independent abstraction
rules and principles for introducing property variables into the semantics in order to
derive the semantic value for the modified subject DP — notably, in converting the
proform to predicative type ⟨e, t⟩ to combine with the correlative clause, and then in
converting the result of type t back to generalized quantifier type ⟨et, t⟩ in order to
combine with the clause’s main predicate. Note that in combining the nodes of type
⟨et, t⟩ it must be stipulated which node is to be converted to predicative type ⟨e, t⟩
(namely, the node for the proform).
The account in this section avoids needing to introduce additional semantic
mechanisms such as these. Correlative clauses and proforms in examples involv-
ing IP/CP adjunction and DP adjunction are given a uniform syntax/semantics: In
both cases the correlative clause denotes an object of type ⟨t, t⟩, and the proform is
represented as a copy of its linguistic antecedent. The shifted interpretation of the
proform is derived from the general analysis of the relevant anaphoric expressions

170
(following §6.1), and the derived assignment binder projecting to the correlative
clause. As the reader can verify, the semantic value derived in (7.57) is equivalent
to the semantic value derived for the alternative structure in which the correlative
clause adjoins at the clausal level.

(7.58) Individual correlative (IP-adjunction)

S∶t

T⟨1,a⟩ CP ∶ t

⟨1,s⟩ IP ∶ t

Cdec t1a
⟨2,a⟩ IP1 ∶ t
CP1 ∶ ⟨t, t⟩

...
tall t1s
she2 -g2 girl-w1 -g1
JSK = J(7.57)K ≈ λgg . [ιa(g)∗ ∶ a(g)(2cf)(girlg− (1s) ) is standing in @(a(g))
∧@(a(g)) = g− (1s)] a(g)(2cf)(girlg− (1s) ) is tall in @(g− )

Finally, it is common in semantics for correlatives to give roughly uniform in-

terpretations of correlative clauses and ordinary DP-like free relatives, say as plural
definite descriptions (e.g. Srivastav 1991, Dayal 1996, Grosu & Landman 1998).
Yet it is interesting that individual correlative clauses pattern with interrogative and
conditional clauses with respect to various syntactic/semantic properties, in contrast
to non-correlative free relatives. For instance, they also pattern with interrogative
clauses in the availability of multiple wh-words, reconstruction effects, and exhaus-
tivity (Dayal 1996, Cable 2009, Citko 2009).

(7.59) a. Which boy read which book? (cf. (7.46))

b. Everyone got what she/*who wants.

The account in this section assimilates individual correlative clauses and conditional
clauses: they include the same definite operator of assignments, which raises from
an equivalent complementizer. Ordinary non-correlative free relatives may be ana-
lyzed analogously to restrictive relatives, as reflected in (6.13), reproduced in (7.60)

171
(ignoring worlds and any contextual restriction associated with ‘what’; though see
n. 46).

(7.60) ‘(Bert likes) what Alice likes’

DP⟨2,a⟩

DEF⟨2,a⟩ CPrel

C
wh3 -g2 E
⟨1,e⟩

Crel t2a Alice likes t1e

JDEFK = λP+⟨a,et⟩ .λQ+⟨a,et⟩ .λgg . [ιx(g)∗ ∃a∶ P+ (a)(x)(g)] Q+ (a)(x)(g)

JDP⟨2,a⟩ K ≈ λQet .λgg . [ιx(g)∗ ∃a∶ x(g) = a(g)(3cf)(E) ∧ Alice likes x(g)]
Q(x)(g[a(g)/2a])

As in the analysis of headed relatives from §6, a determiner quantifier of individuals

raises from the relative complementizer Crel . The implicit quantifier in the free rela-
tive can be given the same sort of assignment-quantificational semantics as the ma-
trix determiners in §6: The plural definite operator DEF picks out a unique maximal
individual. In contrast, the plural definite operator PD in a correlative/conditional
picks out a unique maximal possibility.
7.7.2 Conditional correlatives and ‘then’
Despite extensive syntactic and typological research on correlatives, there has been
surprisingly little work on the compositional semantics of conditional proforms
such as ‘then’. Pioneering developments came with Iatridou 1991, 1994, who showed
that conditional ‘then’ isn’t semantically inert, and with Izvorski 1996, who ob-
served that the exhaustivity implications associated with ‘then’ are shared among
correlative proforms generally, as discussed in §7.5. Yet specific derivations of these
apparent interpretive effects aren’t provided. A prominent approach, following es-
pecially Schlenker 2004, is to analyze ‘then’ as a world pronoun (see also Bhatt &
Pancheva 2006, Arsenijević 2009). Such a move is certainly compatible with the
present framework; ‘then’ could be analyzed as a topicalized expression of the main
predicate’s world argument, as reflected schematically in (7.61).

172
(7.61) [ S . . . [ CP [ CP if-clause ]⟨k,a⟩ [ CP [ XP then1s gk ]⟨1,s⟩ [ C . . . t1s . . . ]]]]

Yet I think there are reasons to be dissatisfied with approaches along these lines,
reasons both theory-internal and independent of an assignment variable framework.
Analyzing ‘then’ as a simple pronoun leaves various linguistic phenomena with
conditional proforms unexplained. First, crosslinguistic work on correlatives shows
that correlative proforms undergo movement to the left periphery of the main clause
(esp. Izvorski 1996, also Bhatt 2003, Arsenijević 2009, Lipták 2012). Interest-
ingly, whether this movement occurs overtly or covertly in a given language often
patterns with certain A-movements such as wh-movement. For instance, in English,
an overt wh-movement language, ‘then’ also raises overtly, as in (7.62). An individ-
ual correlative analogue of (7.62a) is in (7.64).

(7.62) a. If it rains then I think that we should stay at home.

b. *If it rains I think that then we should stay at home.
(Izvorski 1996: ex. 29)
(7.63) Whoi did you persuade ti to come?
(7.64) [Akik korán jönnek]i azokat i Péter hallotta, hogy ingyen beengedik ti .
REL early come those-acc Peter heard that free admit
lit. ‘Who comes early, Peter heard that they will be admitted for free.’
(Lipták 2012: ex. 51)

With individual proforms, one might say that the movement patterns with wh move-
ment given their association with the relative/wh phrases in the correlative clause.
One might wonder what would provide the basis for the movement patterns with a
conditional proform like ‘then’, if the correlative construction simply involves world
pronoun anaphoric to a pluralization operator. Even if the proform isn’t itself a
wh-phrase and the raising may be triggered for reasons of topicalization, analyzing
(e.g.) ‘then’ as a simple variable leaves it obscure why the overt/covert status of the
movement would correlate with that of wh-words.
Second, if ‘then’ is a simple world pronoun, one might wonder why the pronoun
would in many cases be lexicalized by an expression specified as being anaphoric to
an ‘if ’-clause. For instance, although English has various proforms which prima
facie serve as propositional anaphors, such as ‘that’, ‘it’, or ‘so’, only ‘then’ can be
anaphoric to an ‘if ’-clause (or other supposition). Likewise, in German, although
proforms such as es can be anaphoric to a proposition, only dann ‘then’ can correlate
with a conditional wenn ‘if ’ clause:

173
(7.65) Q: Unter welcher Bedingung bedauert Max, [dass Lea singt]i ?
under which condition regrets Max, that Lea sings
‘Under which condition does Max regret that Lea sings?’
A: Max bedauert esi DANNk , [wenn Lea nicht geÜBT hat]k .
Max regrets it then if Lea not practiced has
‘Max regrets it then if Lea has not practiced.’
(Schwabe 2016: ex. 42; German)

It isn’t the case that correlative proforms in general lexically specify a particular type
of antecedent. Crosslinguistically, the proform is typically a demonstrative.
Third, in individual correlatives the correlate may consist of a bare demonstra-
tive, as in (7.66a), or a demonstrative with an explicit nominal, as in (7.66b).

(7.66) a. jo laRkii khaRii hai vo lambii hai

REL girl standing is DEM tall is
lit. Which girl is standing, that is tall
b. [jo khaRii hai] vo laRkii lambii hai
REL standing is DEM girl tall is
lit. Who is standing, that girl is tall
‘The girl who is standing is tall.’ (from Srivastav 1991: ex. 13a; Hindi)

One might wonder whether additional structure is similarly involved in conditional

proforms, especially given the point, noted above, that many languages use the same
type of marker for the proforms in conditionals and individual correlatives.
Our revised syntax/semantics for ‘if ’-clauses in §7.2 affords a straightforward
way of capturing the above data and of unifying our analyses of correlative proforms:
Proforms are topicalized copies of their antecedent relative/wh word, bound by the
correlative clause. The proform in an individual correlative is a copy of the choice-
function pronoun of individuals and (elided) NP ((7.59), §7.7.1). In conditional
correlatives the antecedent is the (possibly implicit) ‘whether’-like element in the
specifier argument of ‘if ’. ‘Then’ is a copy of the choice-function pronoun of worlds
and (elided) proposition-type argument.
‘Then’ may still be analyzed as a topicalization of the main predicate’s world

174
argument. However, it is represented as anaphoric to the implicit ‘whether’-like
element in SpecCP of the correlative conditional clause, as reflected in (7.67).⁹⁷

(7.67) ‘If…then’
‘If it landed heads, then I am in Widmore Library’
S∶t

T⟨1,a⟩ CP

CP ∶ t
C∅ t1a

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ CP2 ∶ t

PD⟨2,a⟩ XP⟨1,s⟩ ∶ ⟨t, t⟩ C∶t

⟨2,s⟩ then2sf -g2 p1st -g1 e IP

wh2sf -g2 p1st -g1
if t2a it2 -g1 heads-t2s
I1 -g2 in-WL t1s

JXP⟨i,s⟩ K = λTt . [λpst .[λgg .p(λgg .g(2a)(2sf)(g(1a)(1st)))(g)]](λws .λgg .T(g[w(g)/is]))

= λTt .λgg . T(g[g(2a)(2sf)(g(1a)(1st))])/is])
JSK ≈ λgg . [ιa(g)∗ ∶ @(a(g)) = a(g)(2sf)(g− (1st)) ∧ . . . ]
a(g)(1e) is in WL in a(g)(2sf)(g− (1st))
⁹⁷I leave ‘then’ in its fronted position, though I leave open whether it might instead be interpreted
in its trace position (cf. Bhatt 2003, Sportiche 2006). The LF in (7.67) leaves the main clause
complementizer position empty. This choice of representation isn’t forced on us. For instance, one
could represent the structure of the ‘if…then’ conditional as in (7.23), and analyze ‘then’ as topi-
calizing the unpronounced declarative complementizer’s assignment-variable; the derived semantic
value would be equivalent. Yet one motivation for (7.67) may come from comparisons with V2 (verb-
second) languages such as German, where ‘then’ (e.g. dann) is followed directly by the verb in the
complementizer position (see Iatridou 1991, Iatridou & Kroch 1993 for discussion in the context
of CP-recursion). ‘Then’ could still be analyzed as in (7.67), though with the verb in C. I return to
these issues in §8.5 when considering conditional questions.

175
The type s world argument is lifted to type ⟨st, t⟩ in the usual way, and after at-
tachment of the binder-index becomes type ⟨t, t⟩. The ‘then’-phrase combines via
function application with the consequent, shifting the relevant modal domain to that
determined by the ‘if ’-clause. Parallel to the individual correlative in (7.46), the cor-
relative binding requirement is captured by (i) the representation of the proform as a
copy of the implicit ‘whether’-element in the correlative clause, here [wh2sf -g2 p1st -g1 ],
and (ii) the assignment-quantification introduced by the operator PD. Roughly put:
‘then’ is to ‘whether’ ((7.67)) as DEM/‘that’ is to REL/‘which’ ((7.59)/(7.66)).
To recap: The accounts of relative clauses and conditionals in §§6–7 have devel-
oped a unified assignment-variable-based syntax/semantics for relativization. This
subsection suggested one way of extending the comparisons to correlative clauses
and proforms in individual and conditional correlatives. Alternative interpreta-
tions are derived from differences in the (individual, modal) relative phrases in
the correlative clause and their relation to corresponding (individual, modal) ele-
ments in the main clause. For instance, individual correlative interpretations arise
when (among other things) the correlative clause is such that the external argument
for the complementizer is identified with the actual world, the relative (individual)
choice-function pronoun [whcf gi ] is bound by PD⟨i,a⟩ , and the world pronoun in
the relative-phrase NP is bound long-distance and anchored to the world of the
discourse. Correlative proforms are analyzed uniformly as topicalized copies of
their relative/wh antecedent. I am not aware of other analyses of ‘then’ that capture
the syntactic/semantic connection between conditional proforms and proforms in
individual correlatives.
Our discussion has been speculative. It is worth examining more thoroughly
how the accounts of relativization and conditionality in these sections might help
capture other similarities/differences among free relatives, correlatives, condition-
als, and interrogatives — e.g., regarding the connections between exhaustivity and
uniqueness interpretations and topicalization/wh-movement. §8 examines how the
assignment-variable-based treatments of conditionals and correlatives can be ex-
tended to interrogatives.

8 Interrogatives
This section begins investigating how the assignment-variable framework may be
applied to non-declarative sentences, focusing on interrogatives. Key features of the
proposed account are as follows:
• it affords a unified syntax/semantics of interrogative, conditional, and cor-
relative clauses (§8.2)

176
 • it unifies the treatment of wh-words with the §6-treatments of relative pro-
nouns and (certain) indefinites as choice-function pronouns (§8.3)
• it provides a precise diagnosis of “interrogative flip” in terms of local readings
under the question operator (§8.4)
• it integrates with the §7-accounts of ‘if ’-clauses and correlatives, providing
compositional derivations for various types of conditional questions (§8.5

The resulting assignment-variable account affords a unified approach to condition-

ality, relativization, and questions.
8.1 Local and global readings in questions
A principal aim of the assignment-variable framework has been to capture various
types of local/global readings of variables and context-sensitive expressions. Con-
sider (8.1)–(8.5) with interrogatives. (Following common practice I typically use
‘interrogative’ for the clause/sentence and ‘question’ for the semantic object.)

(8.1) Did youi feed itj ? (global reading)

a. ≈ Given that gc (i) = o1 and gc (j) = o2 , did o1 feed o2 ?
(8.2) Is it rainingi ? (local reading)
a. ≈ Is there a relevant raining event in the actual world, whatever it is?
(8.3) [Context: S isn’t sure who the killer is, and wants to see if A has a better idea:]
Mighti the gardener have done it? (local reading)
a. ≈ What is the relevant evidence like (=value for g(i))? Is our information,
whatever it is, compatible with the gardener’s being the killer?
(8.4) [Context: We all accept classical utilitarianism. S isn’t sure about Charity X’s
reliability in getting aid to the people who need it most:]
Shouldi we give to Charity X? (global reading)
a. ≈ Does our giving to Charity X follow from what our norms gc (i) enjoin?
Would our giving to Charity X maximize overall happiness?
(8.5) Is Rita richi ?
a. [Context: S knows approximately how much money Rita earns (say, $X/yr),
and S thinks that A does too. Hoping to ascertain A’s views on whether
such a salary counts as rich, S utters (8.5).]
≈ What is the standard for richness like (=value for the degree standard
g(i))? Is it, whatever it is, greater than around $X/yr? (local reading)

177
 b. [Context: We’re millionaires and we agree that one must be a millionaire
to count as rich. Hoping to ascertain Rita’s income, S utters (8.5).]
≈ Is Rita’s income enough to make her a millionaire? (global reading)

Informally put, an expression receiving a “local reading” in an interrogative sen-

tence is one whose value or interpretation is being questioned. S’s question in (8.5a)
targets the relevant standard associated with ‘rich’ — how rich one must be to count
as rich. If A gives a yes-answer S can infer that A assumes that the standard for
richness g(i) is no greater than $X/yr. S’s question in (8.4), by contrast, isn’t asking
A what substantive normative view to accept; it’s presupposed that classical utili-
tarianism is correct. S is inquiring about the implications of these norms given the
empirical facts. Likewise, whereas the individual-pronouns in (8.1) receive their
interpretation from the discourse context, the value for the world-variable in (8.2)
is targeted by the question. Possible answers to (8.2) aren’t propositions (roughly)
like {w ∶ it is/isn’t raining in u} which take as given a particular world, say g(i) = u.
Data such as (8.1)–(8.5) will constrain how we formalize interrogative elements’
lexical entries and implement world- and assignment-binding in the syntax/semantics.
(i) Like with declarative sentences, the interpretation of certain variables in an inter-
rogative must be provided by the discourse assignment, to capture global readings
of expressions whose values aren’t being questioned. Yet (ii) the question operator
should allow certain variables to be targeted by the question, so that the values for
expressions receiving local readings vary across possible answers. In particular, (iii)
unlike with the declarative complementizer and T, the world argument of the main
predicate shouldn’t be linked to the discourse but should receive a local reading.
8.2 Syntax, semantics, metasemantics
It is standard to distinguish at least two elements in the interpretation of interrog-
ative sentences: (i) an interrogative complementizer (call it C? ), which may e.g.
trigger interrogative movement in languages such as English, and (ii) a question
operator (call it Q), which provides the source of the question semantics (e.g. Dayal
1996, Cable 2010, Kotek 2014). Heim 2012 suggests a syntax in which the question
operator is base-generated as an argument of the interrogative complementizer and
raises as a quantifier over propositions. Heim’s account provides precisely the sort
of precedent for an interrogative analogue to our §3-treatment of declaratives. Our
syntax/semantics for declaratives treated assignment-quantifiers as raising for type
reasons from an internal assignment argument position of the declarative comple-
mentizer (e.g. ‘that’). Drawing on Heim’s suggestion, we can try treating Q as raising
for type reasons from an assignment argument position of C? . Whereas Heim in-

178
cludes an independent node for the binder-index to trigger a non-compositional
Predicate Abstraction rule, we can treat Q as combining via function application
with the generalized binder-index, as reflected in (8.6).

(8.6) S

Q⟨i,a⟩
[C0 ]⟨j,s⟩ IP

C? tia …tjs …
The semantic interactions among the question nucleus, interrogative complemen-
tizer, and question operator are fully compositional.
In §7 we noted systematic crosslinguistic links among conditional, correlative,
and interrogative clauses. §§7.5, 7.7 applied an analysis of ‘if ’-clauses as free relatives
of assignments (possibilities) to correlative ‘if…(then)’ conditionals; §7.7 consid-
ered how the syntax/semantics might be extended to correlative clauses generally.
In light of the crosslinguistic data an attractive idea is that (at some relevant level of
analysis) interrogative clauses share a common linguistic representation with con-
ditional/correlative clauses. So, let’s suppose as a working hypothesis that C? has a
roughly parallel argument structure and semantics as ‘if ’/Cif .
I offer (8.7)–(8.8) as lexical entries for the interrogative complementizer C? and
question operator Q — where ‘g ≈ g′ ’ says, intuitively, that g and g′ are equivalent in
the values assigned across indices, yet potentially different in what world constitutes
the world of the possibility represented by the assignment.

(8.7) JC? K = λaa .λpst .λws .λgg ∶ @(a(g)) = w(g) . ∀w′ s.t. w′ (g) = @(a(g)), p(w′ )(g)
(8.8) JQK = λAat .λTt .λgg . for aa = λgg .g− , ∃g′′g ∶ T = [λg′ ∶ g′ ≈ g′′ . A(a)(g′ ) = A(a)(g′′ )]
a. g ≈ g′ ∶= ∀iτ [¬(τ = s ∧ g(iτ ) = g′ (iτ ) = @(g) = @(g′ )) → g(iτ ) = g′ (iτ )]
(8.9) JQ⟨j,a⟩ K = JQK(J⟨j,a⟩ K) = λT′t .λTt .λgg . for aa = λgg .g− , ∃g′′g ∶
T = [λg′ ≈ g′′ . T′ (g′ [a(g′ )/ja]) = T′ (g′′ [a(g′′ )/ja])]

To fix ideas I assume an approach to interrogatives which treats question meanings

as a set of possible answers (Hamblin 1973, Karttunen 1977, Heim 1994). The
interrogative complementizer C? in effect forms an equivalence class with respect
to the embedded proposition (cf. Groenendijk & Stockhof 1984). Q constructs
a set of possible answers, where answers are represented by sets of assignments; the
semantic value of an interrogative sentence is type ⟨t, t⟩. Following our generalized

179
theoretical interpretation of assignments as representing possibilities (§2), the pos-
sibilities constituting a question’s answers may fix facts including facts determining
values for context-sensitive expressions. Local readings are captured by shifting
assignment-variables coindexed with Q⟨j,a⟩ to the assignment in the given answer.
The values for items targeted by the question may vary across the possible answers;
yet the constraint g′ ≈ g′′ requires the assignments in each answer to agree on the
particular value assigned.
To clarify, let’s start with a simple yes/no question:⁹⁸

(8.10) Simple yes/no question

‘Did you laugh?’
S ∶ ⟨t, t⟩

Q⟨2,a⟩ CP ∶ t

w1 g2 C

⟨1,s⟩ IP

C? t2a
you2 g1 laugh t1s
JCPK ≈ λgg ∶ @(g(2a)) = g(2a)(1s) . g(1a)(2e) laughed in @(g(2a))
≈ λgg . g(1a)(2e) laughed in @(g(2a))
JSK ≈ λTt .λgg . ∃g′′ ∶ T = [λg′ ≈ g′′ . g′ (1a)(2e) laughed in @(g′− ) = g′− (1s)
iff g′′ (1a)(2e) laughed in @(g′′− ) = g′′− (1s)]

This treats the semantic value of the interrogative sentence in (8.10) as a set of
propositions T, where each such proposition is a set of assignments g′ that return
the same truth value for the proposition that a certain individual — the individual
assigned to 2e by the assignment g′ happens to assign to 1a — laughed.
First, note the world pronoun [w1 g2 ] supplying the external argument of the
complementizer. With complementizers such as Crel and ‘if ’/Cif in subordinating
clauses there was a natural story about what would supply the external (individual,
⁹⁸Officially, as with T, I leave open whether Q raises to SpecCP or projects its own phrase such as
ForceP (n. 17). In the general definition of the binder-index (§3.5): with C? , τ = s, σ = ⟨s, t⟩; with Q,
τ = a; σ = ⟨t, t⟩.

180
world) argument: In restrictive relatives the quantifier’s restrictor argument targets
a specified subset of individuals supplied by the relative phrase. In individual cor-
relatives the correlative clause targets topical individuals in the (actual) world. In
conditionals the ‘if ’-clause targets a topical subdomain of the background modal
possibility relevant for the supposition supplied by the ‘whether’-like element. For
interrogatives a natural thought is that the argument determines the possibility tar-
geted by the question. In simple unembedded questions such as (8.10) the external
argument of C? supplies a world argument bound by Q;⁹⁹ the question is about what
is the case in the actual world. In (8.10) the argument is semantically trivial, given
our general metasemantic assumption that, for any h ∈ G, h(1s) represents the world
of h (§§2, 6.1). We will see more substantive roles for the argument in more complex
examples.
The semantic value in (8.10) captures the local reading of the main predicate’s
world argument: the question nucleus places a constraint on the worlds of the as-
signments g′ in each possible answer T. Which world represents the world of the
assignments may vary. The assignments in each answer are otherwise equivalent in
what values are assigned across indices, via the constraint g′ ≈ g′′ ; this constraint
will help capture local readings of context-sensitive expressions (below).
The derived semantic value in (8.10) obviously fails to represent the intended
global reading of ‘you’. The semantic value represents the question as targeting what
assignment to associate with the index 1a, rather than as targeting a property of the
particular individual gc (2e) ∈ E determined by the discourse assignment gc . The
assignments in each T may not even agree on the identity of the individual o ∈ E
that did/did not do the laughing.
One response would be to capture global readings by complicating the semantics
of Q. An alternative is to maintain the simpler lexical entry and semantic value for
interrogative clauses, and capture global readings in the metasemantics of answers.
Our metasemantics for declaratives from §2 treated a declarative sentence S as true
in a context c iff the derived semantic value JSK was true with respect to the assign-
ment gc representing the context. Correspondingly, I treat a proposition T ∈ JS?K in
the semantic value of an interrogative sentence S? as an answer to S? in c iff every
assignment g′ ∈ T assigns g−c to all assignment-indices.

(8.11) Metasemantics
⁹⁹Parallel to the case of correlative clauses discussed in §7.7, the world argument might instead be
represented with a complex [whsf g2 pst -g], where the semantic value for [p g] is ultimately identified
with (e.g.) {g(2a)(1s)} = {g′− (1s)}. However, to clarify the intended interpretation I represent the
world argument with a simple world-pronoun.

181
 a. A declarative sentence S is true in c iff JSK(gc ) = 1
b. Tt is an answer to an interrogative sentence S? in c iff (i) T ∈ JS?K , and
(ii) for all g′ ∈ T and assignment-indices ia, g′ (ia) = g−c .

Applied to our example (8.10) above, for any T ∈ JSK , T is an answer to S in c iff
∀ia, g′ ∈ T∶ g′ (ia) = g−c . Since g′ (1a) = g−c , answers will be sets of assignments g′
s.t. g−c (2e) (=g′ (1a)(2e)) did/did not laugh in @(g′− ). The assignments in each an-
swer constitute an equivalence class with respect to the possible-worlds proposition
that g−c (2e) individual laughed, as desired. (Hereafter I continue often to omit the
superscripts in ‘g′− ’, ‘g−c ’, etc., though they should be understood when relevant.)
In effect, what T does in the semantics of declarative sentences, (8.11b) does in
the metasemantics of interrogative sentences. Yet we shouldn’t overstate the con-
trast: In both cases a bridge principle is required to link a sentence’s compositional
semantic value to an interpretation and function in discourse.1⁰⁰ The semantics of
T links a certain assignment-index, say 1a, to the assignments g in the sentence’s
semantic value JSK ; and the metasemantics of declaratives treats the sentence as true
in c iff the assignment representing the context, gc , is in that set of assignments JSK .
Analogously, the metasemantics of interrogatives treats a set of assignments T as
constituting an answer in c only if every assignment in T assigns the assignment
representing the context, gc , to that same index — the index that would be targeted
by T in a use of a declarative. In this sense, for declarative sentences and interroga-
tive sentences alike, global readings are ultimately captured in the metasemantics.
Capturing global readings in interrogative sentences in the metasemantics has
the advantage of maintaining a uniform theoretical interpretation of assignment-
variables/quantification. If a concrete discourse determines that the assignment
index 1a represents the context, it does so regardless of whether a sentence being
used is declarative or interrogative. Just as assignment-variables coindexed with
‘wonder’ in (8.13) represent readings of expressions whose values are objects of Al-
ice’s wondering, assignment-variables coindexed with Q in (8.12) represent readings
whose values are being questioned.

(8.12) a. Is Rita richi ?

b. ≈ Q⟨i,a⟩ … Rita rich-gi
(8.13) a. Alice wondered whether Rita is rich.
b. ≈ …Alice wondered⟨i,a⟩ … Rita rich-gi
1⁰⁰Cf. e.g. Stalnaker 1978, 2014, Lewis 1980, Dummett 1973, Stanley 1997, MacFarlane
2014, Yalcin 2014.

182
Coindexing between assignment variables and assignment binders represents read-
ings of expressions whose interpretation is targeted by the assignment quantifier.
8.3 wh questions
There is a divide in approaches to interrogative wh-words about whether they are re-
lated to relative pronouns, typically construed as λ-binders (Groenendijk & Stock-
hof 1982), or related to indefinites (Karttunen 1977). For instance, on the one
hand, in many languages the same forms can be used for interrogative pronouns
and indefinites (e.g., various Sino-Tibetan and Australian languages; Li 1992, Gair
et al. 1999, Lecarme 1999a, Kratzer & Shimoyama 2002, Mortensen 2003,
Bhat 2004, Thurgood & LaPolla 2017). On the other hand, it is crosslinguis-
tically common for interrogative words to be morphologically related to relative
words. Relative words also have a [+wh] feature, like interrogative wh words.1⁰1
In some languages the relative clauses in correlatives are even introduced with in-
terrogative wh words (e.g. Polish; Citko 2009); on the flip side, in languages such
as Sanskrit (Davison 2009a) and Somali (Lecarme 1999a), complement questions
are expressed by substituting the interrogative with a relative construction, and, in
the case of Sanskrit, using a correlative.
§§6, 7.2, 7.7 developed analyses of certain indefinites, relative pronouns, and
several types of anaphoric pronouns as choice-function pronouns. A natural move
is to extend the account to interrogatives and treat wh-words similarly as choice-
function pronouns (see Yatsushiro 2001, Cable 2010 for precedents).1⁰2 Indeed
relative forms that are morphologically related to demonstratives are also common
(e.g., German, Dutch, Tuvaluan (Polynesian); de Vries 2002, Bhat 2004). Some
languages even use the same pro-forms for each of these functions, as in Lyele (Niger-
Congo) (Showalter 1986; see Bhat 2004, esp. §8.3). Giving the expressions a
common semantic contribution, at some level of abstraction, reflects this (more on
this shortly). In the case of [+wh] choice-function pronouns, just as relative words
are targeted by a determiner quantifier so that the domain is determined from the
1⁰1For instance, relative clauses are also subject to complex NP and wh island constraints, as in (i).
The presence of the [+wh] feature across languages is independent of whether a wh format is used.
(i) a. *every cat (which, that) Alice heard the rumor that I like __
b. *every cat (which, that) I wondered who likes __
1⁰2I ignore differences between interrogative ‘which’ phrases and simplex interrogative pronouns,
e.g. with respect to presuppositionality/D-linking (§6.3.1). Note that the account is neutral on Cable’s
“Q-particle”-based syntax for wh movement and pied piping. Cable’s Q-particle — also a variable over
choice functions — is posited as an element independent of wh words and functional interrogative
elements (complementizer, question operator).

183
set of NP-individuals, interrogative words are targeted by the question operator so
that the values for wh-phrases vary across answers.
An example with a simple wh question is in (8.14). (For purposes of illustration
I show how the variation in values for the wh-phrase can be derived while keeping
the wh-word in situ below C; more on this below. As previously ignore any con-
textual restriction associated with ‘what’, and assume E represents the domain of
individuals. Hereafter when describing the answers to a question I often omit the
presuppositional constraint when trivial.)

(8.14) wh question
‘What hit you?’
S ∶ ⟨t, t⟩

Q⟨2,a⟩ CP ∶ t

w1 g2
⟨1,s⟩

C? t2a
wh1 -g2 E
hit-t1s you2 -g1
Jwh1 g2 K = λPet .λgg . g(2a)(1cf)((↓ P)(g))
JCPK ≈ λgg ∶ @(g(2a)) = g(2a)(1s) . g(2a)(1cf)(E) hit g(1a)(2e) in @(g(2a))
A proposition Tt ∈ JSK is an answer to S in c iff
∃g′′g ∶ T = [λg′g ≈ g′′ . g′− (1cf)(E) hit g−c (2e) in @(g′− )
iff g′′− (1cf)(E) hit g−c (2e) in @(g′′− )]

The possible answers to the wh question are propositions T, where each such propo-
sition is a set of assignments g′ that return the same truth value for the proposition
that a selected individual g′ (1cf)(E) ∈ E hit gc (2e).
First, as in (8.10), the global reading of ‘you’ is captured via the metasemantic
condition on answerhood: A set of assignments T in the sentence’s semantic value
constitutes an answer only if g′ (1a) = gc . Given the constraint that g′ ≈ g′′ , the
assignments in each answer constitute an equivalence class with respect to a proposi-
tion about the contextually relevant individual g′ (1a)(2e) = g′′ (1a)(2e) = gc (2e) ∈ E.
Second, the “shifting” of wh-words in the interpretation of interrogatives is as-

184
similated to the general phenomenon of local readings of context-sensitive expres-
sions. The local reading of ‘what’ is captured in the coindexing between the question
operator and the choice-function pronoun [wh1 g2 ] representing the wh-phrase.
Just as with relative pronouns and the relative complementizer, I assume that wh-
words, having the feature [wh], must have the same assignment-variable as the
interrogative complementizer, e.g. due to agreement (cf. Kratzer & Shimoyama
2002, Cable 2010, Johnson 2012). Relative and wh words thus receive obligatory
local readings under the local assignment-binder — e.g., a matrix determiner in a
restrictive relative (§6), PD in a conditional or individual correlative (§7), or the
question operator in an interrogative. Given the constraint that g′ ≈ g′′ , the assign-
ments g′ in each answer T assign the same value to the choice-function index and
hence select the same individual. Yet the particular value assigned and individual
selected may vary across answers: If E = {o1 , o2 }, the possible answers are the
propositions that o1 hit gc (2e), and that o2 hit gc (2e). The wh-phrase is targeted
by the question, and the possible answers are about different individuals.
Third, although the assignments g′ in each answer assign the same choice func-
tion to the index 1cf and agree about whether the selected individual hit gc (2e), they
may differ regarding other worldly facts. The possible answers T aren’t singleton sets.
Questions with multiple wh-phrases raise notorious syntactic/semantic chal-
lenges — e.g., regarding movement and reconstruction, single-/pair-list readings and
answers, and exhaustiveness and uniqueness, such as the apparent implication of
(8.15) that each baby liked a unique toy (e.g. Dayal 2006, 2016 and references therein).

(8.15) Which baby liked which toy?

a. ≈ Timmy liked the doll, Clio liked the blocks, …

For present purposes I simply note two features of the assignment-variable account
in this section. It is common in diverse approaches to the semantics of questions
to require all wh-phrases to be interpreted in a specifier position of the interroga-
tive complementizer (e.g. Karttunen 1977, Pesetsky 2000, Cable 2010, George
2011).1⁰3 A critical issue in such accounts is what in general requires the movement
and how the order at LF of wh-phrases in multiple wh-interrogatives is derived,
given the crosslinguistic variation in pronunciation rules for wh-words (e.g., re-
quiring all wh-words to be fronted (Bulgarian), requiring all wh-words to be in
situ (Japanese), or allowing variation in movement (English)). Even for languages
1⁰3Contrast Chomsky 1995, Kratzer & Shimoyama 2002. The movement may be triggered for
semantic reasons (Kotek 2014), or syntactic reasons, e.g. that any XP headed by an item with [wh]-
features move to SpecCP.

185
such as English with overt wh movement, Rullmann & Beck (1998) have argued
that D-linked wh-phrases (‘which N’) are obligatorily interpreted in situ (also e.g.
Boeckx & Grohmann 2004). The account in this section is compatible with al-
ternative approaches to the syntax/phonology of wh-movement and reconstruction
in different languages. The account captures the local reading of wh-words and the
contribution of wh-words to the semantics of the question without requiring or for-
bidding wh-movement to SpecCP at LF. Further, the semantics extends to multiple
wh questions without requiring innovations such as (e.g.) a type-flexible question
operator or tuple types, which track the number of wh-words and their relative
positions (as in e.g. Higginbotham & May 1981, Groenendijk & Stockhof 1984,
George 2011).
A derivation for a multiple wh question such as (8.15) is as follows (n. 102; I
continue to use abbreviations such as ‘babyu ’ for the characteristic function of the
set of individuals o ∈ E such that o is a baby in u (§§2, 6.1.3).)

(8.16) Multiple wh question

S ∶ ⟨t, t⟩

Q⟨2,a⟩ CP ∶ t

w1 g2

⟨2,s⟩

C? t2a

wh2 -g2 baby-w1 -g1 liked t2s

wh1 -g2 toy-w1 -g1
A proposition Tt ∈ JSK is an answer to S in c iff
∃g′′g ∶ T = [λg′g ≈ g′′ . g′− (2cf)(babyg−c (1s) ) liked g′− (1cf)(toyg−c (1s) ) in @(g′− )
iff g′′− (2cf)(babyg−c (1s) ) liked g′′− (1cf)(toyg−c (1s) ) in @(g′′− )

This says, roughly, that given sets of actual babies and toys — say, b1 , b2 , b3 ∈ babygc (1s)
and t1 , t2 , t3 ∈ toygc (1s) — the set of possible answers is the set of propositions that bi
liked tj . Given the metasemantic condition on answerhood, the world-pronouns in

186
the noun phrases ‘baby’ and ‘toy’ receive global readings anchored to gc (1s), which
is taken to represent the world @(gc ) of the discourse (§6.1.3). However, as in
(8.14), the world-argument of the main predicate ‘liked’ and the wh-words receive
(obligatory) local readings, targeted by the question. For each possible answer T,
the set of assignments g′ ∈ T constitute an equivalence class with respect to what
actual baby bi = g′ (2cf)(babygc (1s) ) ∈ E and what actual toy tj = g′ (1cf)(toygc (1s) ) ∈ E
are selected, and whether the selected baby bi liked the selected toy tj in the world
@(g′ ) of the possibility represented by g′ . The account derives how each answer in
the multiple wh question is about a particular pair of individuals.
The proposed syntax/semantics of wh interrogatives may afford an improved
treatment of examples with intuitively bound readings of pronouns such as (8.17).
The value for the wh-phrase ‘which baby’ varies across possible answers; and the
interpretations of the subject-internal pronoun ‘his’ and the object pronoun ‘him’
are linked to the value for the wh-phrase, though ‘him’ isn’t c-commanded by the
specifier position for the possessor in the genitive DP.

(8.17) Which baby’s picture of his grandmother made him cry?

a. {Timmy’s picture of Timmy’s grandmother made Timmy cry, Clio’s pic-
ture of Clio’s grandmother made Clio cry, …}

Consider how we might capture such a reading in a traditional framework and Karttunen-
style (1977) semantics for wh interrogatives. Wh-words such as ‘which’, on Kart-
tunen’s account, are given the same quantificational lexical entry as indefinites; what
distinguishes their contribution to the meanings of questions is a syntactic require-
ment that (all and only) [+wh] phrases be in SpecCP at LF, as reflected in (8.18)–(8.19).
(I ignore compositional details about how the question meaning is derived from the
interrogative complementizer and question operator; see Heim 2012.)

(8.18) JwhichK = JsomeK = λPet .λQet .∃x∶ P(x), Q(x)

(8.19) Which studenti (ti ) left?
a. [CP … which student 1 [C C0 [IP t1 left]]]
b. J(8.19a)Kw = λpst . ∃x∶ x is a student inw, p = [λw′ .x left in w′ ]

The wh existential quantifier scopes over the interrogative complementizer identi-

fying the form of the answers, and the set of answers includes one proposition per
student quantified over.
In a complex genitive such as ‘which baby’s NP’ in (8.17), the [wh]-feature of

187
the wh-phrase projects/percolates to the DP, triggering wh-movement of the entire
phrase to Spec of C:

(8.20) overt wh-movement to SpecCP:

[CP … [DP [+wh] [which baby] ’s NP] 1 [C C0 [IP t1 …]]]

The challenge is to derive the intuitively bound readings of both pronouns from
operations on this (possibly intermediate) syntactic structure. One might posit that
for some syntactic or semantic reason there is subsequent covert movement of the
wh-phrase, and that — unlike the examples of DP-internal movement from §6.3.3 —
the wh-phrase raises above the entire DP, as in (8.21). (Assume a simplified analysis
of the possessive ‘his grandmother’ as ‘the-grandmother-of-him’.)

(8.21) a. [CP … [which baby [2 [[DP t2 [D ’s pic-of-the-GM-of-him2 ]] [1

[C C0 [IP t1 made him2 cry]]]]]]]
b. J(8.21a)Kw ≈ λpst . for some x s.t. x is a baby in w, the y s.t. y is a picture-
of-the-GM-of-x in w is s.t. p = [λw′ .y made x cry in w′ ]

Such an analysis derives how the wh-phrase binds both pronouns, but at the cost of
requiring an unlikely movement operation (cf. May 1985, Heim & Kratzer 1998)
and denying that ‘which baby’s picture of his grandmother’ forms a constituent. Al-
ternatively, one might treat the wh-phrase as raising (for some syntactic or semantic
reason) internal to the DP, and posit a mechanism of partial syntactic reconstruction
by which the non-wh portion of the DP is interpreted inside the IP at LF. By some
reconstruction operations to be specified, the wh-phrase and λ-binder generated
from the DP-internal movement remain in Spec of C; hence the wh-phrase existen-
tial quantifier scopes over the interrogative complementizer, which identifies the
form of the answers, and binds the pronouns, roughly as in (8.22).

(8.22) a. covert movement of ‘which baby’:

[CP … [[DP which baby [2 [t2 [D ’s pic-of-the-GM-of-him2 ]]]] [1 [C … t1 …]]]]
b. post-reconstruction LF:
[CP … [which baby [2 [C C0 [IP [DP t2 [D ’s pic-of-the-GM-of-him2 ]] made
him2 cry]]]]]
c. J(8.22b)Kw ≈ λpst . for some x s.t. x is a baby in w, p = [λw′ .the y s.t. y is a
picture-of-the-GM-of-x in w′ made x cry in w′ ]

Details of how the requisite (partial) reconstruction is realized, and whether re-
construction is to be understood in syntactic terms at all, are controversial (cf. e.g.
Chomsky 1995, Heycock 1995, Sternefeld 1998, Fox 2000, Fox & Nissenbaum

188
2004, Sportiche 2006). Note that the fronted DP ‘which baby’s picture of his grand-
mother’ still fails to receive a semantic value.
The assignment-variable-based syntax/semantics for wh interrogatives in this
section captures the intuitively bound reading of (8.17) without requiring particular
commitments on issues regarding the movement of wh-phrases, pied-piping, or
reconstruction. (For present purposes I continue to assume the simplified repre-
sentation of the possessive pronoun (see §6.3.3).)

(8.23) Which baby’s picture of his grandmother made him cry?

S

Q⟨2,a⟩ CP

w1 g2
⟨2,s⟩

C? t2a
DP VP

⟨2,e⟩ D scared t2s

he2 -g2 -baby-w1 -g1

wh2 -g2 baby-w1 -g1 ’s⟨3,a⟩

⟨1,e⟩

t1e
F3cf g3 C t3a t2e
R1 g1

picture-w1 -g1 of the-grandmother-of-he2 -g2 -baby-w1 -g1

A proposition Tt ∈ JSK is an answer to S in c iff (roughly) ∃g′′g ∶ T = [λg′g ≈ g′′ .

[ιxe (g′ )∃a′a ∶ g−c (1eet)(g′− (2cf)(babyg−c (1s) ))(x(g′ )) ∧
x(g′ ) = a′ (g′ )(3cf)(picture-of-the-grandmother-of-g′− (2cf)(babyg−c (1s) )-in-g−c (1s))]
x(g′ ) scared g′− (2cf)(babyg−c (1s) ) in @(g′− )
iff

189
 [ιxe (g′′ )∃a′a ∶ g−c (1eet)(g′′− (2cf)(babyg−c (1s) ))(x(g′′ )) ∧
x(g′′ ) = a′ (g′′ )(3cf)(picture-of-the-grandmother-of-g′′− (2cf)(babyg−c (1s) )-in-g−c (1s))]
x(g′′ ) scared g′′− (2cf)(babyg−c (1s) ) in @(g′′− )]

Roughly put, each possible answer T is a set of assignments g′ that return the same
truth value for the proposition that [the o′ s.t. o′ bears the contextually relevant re-
lation gc (1eet) to a selected baby o = g′ (2cf)(babygc (1s) ) ∈ E and o′ is a picture of the
grandmother of o (=g′ (2cf)(babygc (1s) ))] is such that o′ scared o (=g′ (2cf)(babygc (1s) ))
in @(g′ ). For each answer T, the set of assignments g′ ∈ T constitute an equivalence
class with respect to what actual baby oi = g′ (2cf)(babygc (1s) ) ∈ E is selected, and
whether oi ’s picture of oi ’s grandmother scared oi in the world @(g′ ) of the possibility
represented by g′ .
The apparent anaphoric connections between the pronouns ‘his’/‘he’ and the wh-
phrase ‘which baby’ are captured derivatively via (i) the local reading of the wh-
phrase under the question operator, and (ii) the representation of the pronouns
as copies of the antecedent wh-phrase (see n. 93). The variation in values for the
wh-phrase across answers and the intuitive bound readings of the pronouns can be
captured while leaving the wh-phrase in situ and giving the complex subject DP a
particular semantic value. Controversial assumptions about pied piping and wh-
movement of the complex genitive DP, subsequent movement of the DP-internal
wh-phrase, and syntactic reconstruction aren’t required.
8.3.1 Recap: Choice functions and choice-function variables
At various points we have appealed to variables for choice functions in the repre-
sentations of various types of expressions — e.g., relative words (§6.1), certain in-
definites (§6.1), certain anaphoric pro-forms (§§6.2, 6.3, 7.7), quantificational DPs
(§6.3.1), and wh words (§§7.2, 8.3). We have seen the empirical fruits of these anal-
yses in developing assignment-variable-based compositional semantics for a wide
range of constructions, including relative clauses, donkey sentences, conditionals,
correlatives, and interrogatives, among others. It is also worth also briefly recapping
certain of the independent motivations for the unified approaches, and relevant
comparisons among the different types of expressions.
As discussed in the previous sections, researchers in various subdisciplines have
demonstrated crosslinguistically robust morphosyntactic parallels among the ex-
pressions. For instance, we noted, first, that relative words generally have the same
format as interrogative wh-words or d-words, and that in many languages wh-words
are also morphologically related to indefinites (Gair et al. 1999, de Vries 2002,
Bhat 2004, Thurgood & LaPolla 2017). Second, in many languages (strong)

190
quantifiers are diachronically related to or transparently composed with a relative/wh
or definite/demonstrative form (Haspelmath 1995). Third in some languages bound-
variable pronouns and/or donkey pronouns can be expressed as morphologically
complex or with bare nominals (Srivastav 1991, Mortensen 2003). Although
German d-words take an elided complement, in other languages pronominal ele-
ments can be used with an overt nominal (Déchaine & Wiltschko 2002). Fourth,
the proforms in correlatives are generally formed from a demonstrative element —
indeed the correlative binding requirement is typically referred to as the “demon-
strative requirement” (Dayal 1996, Bhatt 2003, Lipták 2009b). Interestingly, the
A-movement of the demonstrative correlates in a language generally patterns with
wh-movement (Izvorski 1996). Uniqueness and exhaustivity implications asso-
ciated with wh interrogatives can also be observed with correlatives (Dayal 1996,
Izvorski 1996). Some languages even use the same pronouns for each of these func-
tions — as demonstratives, indefinites, interrogatives, and relatives (Bhat 2004).
The proposed unified semantics gives semantic expression to such intra- and
inter-linguistic relations among the expressions. This isn’t to say that there are no
differences among them — not only morpho-phonologically, but also syntactically,
semantically, distributionally (see nn. 51, 73). For instance, though I have been
ignoring internal projections for features such as phi features, differences in the
internal structure and pronunciation rules for pronominal elements are to be ex-
pected (Déchaine & Wiltschko 2002, 2017b, Kratzer 2009, Grosz & Patel-
Grosz 2016, Patel-Grosz & Grosz 2017). Ordinary simplex pronouns are in
effect a conventionalized system of demonstratives prioritizing the expression of
certain packages of phi features1⁰⁴ — over (say) spatial features or the overt real-
ization of a nominal complement — for retrieving the relevant selectional domain.
(Indeed in some theories the class of pronouns is defined in terms of lacking an overt
complement (cf. Baltin 2012, Patel-Grosz & Grosz 2017).) It isn’t uncommon,
1⁰⁴And in some languages features such as tense, aspect, mood, as in (i)–(ii) (also Demirdache
1996, Bhat 2004, Nordlinger & Sadler 2004, Fenton 2010, Adamou 2011; see §6.2):
(i) Isá-gíi baa hádal-kíi qaatáy, oo yiri:
he.emph.past decl talk-def.past take.past and said
‘He began to speak, and said:’ (Lecarme 2008: ex. 10a; Somali)
(ii) Ma taha na fyè e!
you.nondecl follow my.nondecl footprints in
‘Follow me (lit. in my tracks), please!’ (Carlson 1994: 522; Supyire (Niger-Congo))
As Lecarme (2008) puts it regarding Somali, “Since pronouns do not contain nouns that can be
predicated of a stage,” the nominal tense in examples such as (i) is used to “[locate the] individual
temporally, at a past time which overlaps (or coincides with) the time of the event.”

191
in various language families, for the same format to be used for certain types of
pronouns and demonstratives, such as in various Chinese dialects with third-person
pronouns and distal demonstratives (cf. Jespersen 1924, Postal 1969, Gair et al.
1999, Bhat 2004, Luján 2004, Thurgood & LaPolla 2017). Conversely, recall
our examples with nominal tense in Somali from §6.2.1. Like individual pronouns,
Somali demonstratives are compatible with syntactic definiteness marking; in both
cases the co-occurrence has an intensifying effect, as we saw with the emphatic
pronouns in (6.28)–(6.29).

(8.24) a. ani-ga waxaa i arkay Axmed.

I-emph.nonpast decl 1s.obj.cl saw.past Axmed
‘Axmed saw me.’ (Özyıldız & Ivan 2017: ex. 10a)
b. Tuulá-dóo horé éeg!
village-dem.emph.nonpast before look
‘Look at that village in front!’ (Lecarme 2008: ex. 5a)

However, unlike pronouns, demonstratives are obligatorily deictic in Somali (Lecarme

2004, 2008, 2012). Even if (say) both types of pronominal elements were represented
via choice-functions, the choice-function-qua-individual-pronoun plays a distinc-
tive syntactic/semantic role in the expressive potential of the pronominal system.
Other differences may be derived from differences in [wh] features — notably,
distinguishing relative and wh words, on the one hand, and indefinites and anaphoric
pronouns, on the other. For instance, in languages such as Chinese the relation
between wh words and indefinites is morphologically transparent: wh words just
are, in some cases, indefinites with a phonetically unrealized [+wh] feature. In
other languages the [+wh] feature may be realized via an interrogative morpheme,
or indefinites and wh words may be derived from a more basic underlying form,
as in languages with systems of “indeterminate” pronouns such as Japanese. Sec-
ond, that there should be a specifically demonstrative requirement with correla-
tives is in one sense unsurprising. A “demonstrative correlate,” on the analysis in
§7.7.1, just is a copy of the interpreted material in a relative/wh phrase — at some
level of abstraction, REL/WH = [+wh]-DEM. Put the other way, a relative/wh word
in a correlative clause is a spelled-out [+wh] demonstrative in agreement with a
[+wh] Cif (cf. Davison 2009a). The “matching requirement” between correlates
and antecedents is then, in effect, a form of topic-comment congruence.
Leaving matters here would of course be much too simplistic, even at an infor-
mal impressionistic level. For instance, not all correlatives are syntactically real-
ized across languages via a particular type of topicalization structure (Lipták 2012).

192
Characterizing a crosslinguistically applicable matching requirement (what is uni-
versal, and what is parameterized), and deriving the relevant locality effects in par-
ticular languages has proven far from straightforward (Bhatt 2003, Davison 2009a,
Leung 2009, Lipták 2012). Likewise for analyses of syntactic/semantic relations
between indefinites and interrogatives, the internal and external syntax of different
classes of pronouns, interactions among the syntax and the phonological and mor-
phological components in the realization of pronominal and anaphoric expressions,
etc. The aim in these sections has been simply to provide an account which captures
a common semantic core among the expressions in question, and thereby reflects
various relations among them crosslinguistically. For present purposes what rele-
vantly unifies e.g. certain indefinites, relative and wh words, d-words, and certain
anaphoric expressions is that they are analyzed in terms of choice-function pro-
nouns. We have seen how integrating these common analyses in broader assignment-
variable-based syntax/semantics can compositionally derive a wide range of linguis-
tic shifting phenomena in various types of relativization, conditional, and interrog-
ative structures. I hope the proposed semantics may provide a fruitful basis for
delineating further dimensions of difference.
8.3.2 Aside: Weak crossover revisited
§6.3.3 suggested an account of weak crossover in various types of examples with
QR’d DPs. The account was deliberately neutral regarding weak crossover with wh
interrogatives in light of independent contrasts between QR-chains and wh-chains
(e.g., Safir 1999), and contentious issues regarding the data and the syntax/semantics
of wh-movement (n. 84). For instance, there are cases where wh weak crossover
configurations are acceptable, both in English and crosslinguistically. In scrambling
languages such as German the sentence corresponding to (8.25b) can be grammat-
ical, as in (8.27). In English, wh weak crossover examples improve in certain focus
constructions, as in (8.28) with ‘only’.

(8.25) a. Whoi (ti ) likes heri child?

b. *Whoi does heri child like (ti )?
(8.26) a. [Which boyi ’s friend]k (tk ) accidentally introduced hisi girlfriend to hisi
mother?
b. *[Which boyi ’s mother]k did hisi friend accidentally introduce hisi girl-
friend to (tk )?
(8.27) Weni mag ihri Kind?
who-acc likes her child

193
 ‘Whose child likes her?’ / ‘Who is liked by her child?’
(8.28) ?Which babyi did only itsi mother say (ti ) was cute?

How to diagnose such examples is controversial (see Safir 2017 for an overview).
Yet given such intralinguistic and crosslinguistic variations, it isn’t implausible that
wh weak crossover effects shouldn’t follow simply from general binding principles,
such as (6.119) in §6.3.3, but from properties of wh chains within the language. I
won’t pretend to provide a general account of wh weak crossover here; as emphasized
above, the proposed semantics for wh interrogatives is compatible with different
views on the syntactic/semantic status of wh movement. But now that the treatment
of wh words is on the table, it may be worth briefly speculating on binding configu-
rations with pronouns.
To a rough (inadequate) first approximation, consider the constraint in (8.29)
on relations among copies of wh-phrases.1⁰⁵

(8.29) Let αi be an expression with feature [wh] such that there is no coindexed γi
higher than αi ; and let C = (a1i , . . . , ani ) be the sequence of (lower) copies
j
with the same subscript index as αi . Then for every ai ∈ C, there is some aik ∈
j
C and [+wh]-β such that (β, ai ) is a chain and β ’s [wh]-feature percolates
from aki .

Copies of movement (traces) are typically required to be identical to the moved

item; in theories relaxing this requirement the lower copies must at least have the
same phi features (person, number, gender), [wh]-features, agreement features, etc.
(cf. Nunes 2004, van Koppen 2007; see n. 13). Non-movement copies, in contrast,
need only be equivalent with respect to elements relevant to satisfying requirements
for coindexing; that is, they must have the same phi features (e.g. Kratzer 2009).
The principle in (8.25a) thus permits examples such as (8.25a): the copies of whoi =
[[+wh] wh-g person-w-g]i are the trace ti and bi = [she-g person-w-g]i in the repre-
sentation of the possessive pronoun; ti = [[+wh] wh-g person-w-g]i is [+wh]; and (ti )
and (ti , bi ) are chains (the former trivially so). Examples such as (8.25b), however,
are excluded: there is no [+wh]-β , whose [wh]-feature arises from some aki ∈ C,
1⁰⁵The superscripts in (8.29) are included simply to delineate the copies; they have no theoretical
import. I continue to use ‘copy’ broadly both for traces (n. 13) as well as for copies not derived via
movement (§6.2). Likewise for ‘chain’: the chains needn’t be movement chains (though one may
assume other familiar conditions such as uniformity and c-command); the principle is compatible
with treating chain relations as represented in the narrow syntax or the interfaces. The principle is
formulated to apply at a level at which wh movement has taken place, though it could be reformulated
in other terms.

194
such that (β, bi ) is a chain; the pronoun bi lacks a [wh]-feature (n. 93), and the
[+wh] trace ti doesn’t c-command bi . The treatment of “secondary weak crossover”
effects, as in (8.26), proceeds likewise. Unlike (8.26a), (8.26b) is excluded since
the only potential [+wh]-β is tk = [[+wh] Which boyi ’s mother]k , and tk c-commands
neither copy in the representations of the possessive pronouns.
Note that the principle in (8.29) carries over to weak crossover effects in relative
clauses — even given the analysis in §6 in which the relative phrase originates in its
position in Spec of Crel — as reflected schematically in (8.30)–(8.31).

(8.30) a. Every cat which likes its owner…

⟨i,e⟩
b. every [CPrel wh cat C0i ti likes the-owner-of-oi ]
c. “every cat x s.t. x likes the owner of x…”
(8.31) a. Every cat which its owner likes…
⟨i,e⟩
b. *every [CPrel wh cat C0i the-owner-of-oi likes ti ]
c. ≉ every cat x s.t. x is liked by the owner of x…

In (8.30)–(8.31) the IP in the relative clause includes two copies of the [+wh]-C0rel —
the [+wh] trace ti , and the coindexed variable oi in the representation of the pronoun
‘its’. (Representing the pronoun with a copy of the relative phrase would violate
Principle B.) The configuration in (8.30) is permitted since (ti ) and (ti , oi ) are both
chains. (8.31) is excluded since the higher copy oi lacks a [+wh] feature; hence there
is no available [+wh]-β such that (β, oi ) is a chain.1⁰⁶
The general binding principle in (6.119) (§6.3.3) essentially functions to con-
strain movement for purposes of repairing a type-mismatch from introducing new
kinds of binding relations; the interpretive effects arising from QR’ing an expres-
sion α are restricted to those associated with α’s semantic role in its base posi-
tion. Whereas QR is a semantically driven mechanism applying for reasons of in-
terpretability, wh movement and associated reconstruction effects depend on prin-
ciples and representations at various grammatical levels. Hence it might not be
surprising to find greater crosslinguistic variation with respect to weak crossover
with wh-interrogatives, as we in fact do. Crosslinguistic variations in weak crossover
1⁰⁶Though relative pronouns are often morphologically related to interrogative pronouns, forms
morphologically related to demonstratives are also common (e.g., in German). Crosslinguistic
syntactic work has shown that relative pronouns bear a [+wh] feature which is checked by the relative
complementizer, and that this holds regardless of whether a wh format is used (Chomsky 1977,
de Vries 2002). For instance, relative clauses are subject to complex NP and wh island constraints:
(i) a. *every cat (which, that) Alice heard the rumor that I like __
b. *every cat (which, that) I wondered who likes __

195
with wh-interrogatives might be diagnosed in terms of differences in the status and
precise formalization of a principle such as (8.29) along with independent grammat-
ical differences regarding wh movement — e.g., differences with respect to param-
eters such as the relevant command relations in (8.29), at what syntactic level the
principle is checked, the syntactic/semantic status of wh movement, reconstruction,
pied-piping etc. I leave further developments and investigation of relations among
different types of weak crossover configurations for future research.
8.4 “Interrogative flip”
The assignment-variable account in §8.2 affords a precise diagnosis of so-called “in-
terrogative flip” (Speas & Tenny 2003): cases of “interrogative flip” can be assim-
ilated to (possibly conventionalized) local readings (§§3.1, 4.3). Consider (8.32),
on the salient reading which questions the relevant evidence and the doings of the
contextually relevant individual. (I bracket any resource domain argument associ-
ated with the modal (§7.6), and I treat ‘be the killer’ as an unanalyzed predicate. As
previously I use ‘r’ in indices for type ⟨s, at⟩.)

(8.32) Might he be the killer?

S

Q⟨2,a⟩ CP

w1 g2

⟨1,s⟩

C? t2a
⟨3,a⟩

⟨2,s⟩
might
t1s
r1 g2 Cdec t3a
he3 -g1 be-the-killer-t2s

JSK ≈ λTt .λgg . ∃g′′g ∶ T = [λg′g ≈ g′′ . ∃aa s.t. g′− (1r)(@(g′− ))(a(g′ ))∶
g′ (1a)(3e) is the killer in @(a(g′ ))
iff ∃aa s.t. g′′− (1r)(@(g′′− ))(a(g′′ ))∶ g′′ (1a)(3e) is the killer in @(a(g′′ ))]

196
 A proposition Tt ∈ JSK is an answer to S in c iff
∃g′′g ∶ T = [λg′g ≈ g′′ . ∃aa s.t. g′− (1r)(@(g′− ))(a(g′ ))∶ g−c (3e) is the killer in @(a(g′ ))
iff ∃aa s.t. g′′− (1r)(@(g′′− ))(a(g′′ ))∶ g−c (3e) is the killer in @(a(g′′ ))]

Each possible answer T is a set of assignments g′ that determine the same modal
background R (=g′ (1r), via g′ ≈ g′′ ), and return the same truth value for the propo-
sition that the contextually relevant individual gc (3e) (=g′ (1a)(3e)) is the killer in
the world of some epistemic possibility R(@(g′ )), as determined by g′ . Roughly,
answers to the epistemic modal question are propositions of the form that there is/is
not an epistemic possibility (=g′ (1r)(@(g′ ))) where gc (3e) is the killer.
Local readings under assignment-quantifiers are represented uniformly in terms
of coindexing between assignment-variables and binder indices (§8.2). The compo-
sitional semantics of the interrogative elements C? and Q derive the obligatory local
reading of the modal’s world argument t1s . The question nucleus places a constraint
on the worlds of the assignments g′ in each possible answer, i.e. that gc (3e) is the
killer in some world accessible from @(g′ ). Likewise the obligatory local reading of
the embedded predicate’s world argument t2s is captured via the coindexing between
the embedding (declarative) complementizer and the modal’s binder index. The
local reading of the epistemic modal is captured by allowing the value for the modal-
background pronoun [r1 g2 ] to vary across the propositions comprising the possible
answers. The assignments in each answer assign the same value to 1r; however, the
particular epistemic modal background assigned may differ across answers. The
relevant evidence is targeted by the question.
In §4.3 we discussed how there may be reasons for positing conventionalized
locality/globality principles for certain context-sensitive expressions — e.g., a glob-
ality principle for gendered pronouns, or a locality principle for certain types of
epistemic uses of modals. We noted that a locality principle requiring certain modal-
background pronouns to be coindexed with the closest c-commanding assignment-
binder may help capture the linking of epistemic modals to the subject in attitude
ascriptions. Such a principle would apply to epistemic modals in questions as well:
An epistemic modal question such as (8.32) would typically be interpreted as being
about what the evidence is — represented by what value R ∈ D⟨s,at⟩ to assign to
the modal-background index 1r — rather than about what the logical implications
are of a contextually agreed-upon body of evidence gc (1r) at the evaluation world
(contrast (8.4)). For present purposes I leave open whether conventionalizing such
an interpretation is appropriate in the general case (cf. Silk 2016, 2017).

197
8.5 Conditional questions
Informally, whereas the possible answers in a non-conditional question partition the
relevant space of possibilities, the possible answers to a conditional question such
as (8.33) partition the subdomain of possibilities that verify the antecedent.

(8.33) If it snows, will school be canceled?

Capturing this idea has proven a persistent challenge for traditional approaches to
questions and quantificational analyses of conditionals (see Isaacs & Rawlins 2008,
Groenendijk & Roelofsen 2009, 2010, Starr 2010; cf. Charlow 2010, 2011 on
challenges with conditional imperatives). An additional challenge is to capture the
various types of local and global readings in both the ‘if ’-clause and interrogative
main clause. Consider the following conditional-question analogues of the shifted-
indexical example from Santorio 2012 in §7:

(8.34) [See (7.6). After the experiment, one of the amnesiacs wakes up. Feeling
fuzzy about the experimental protocol, he asks himself:]
If it landed heads, might I be in Widener?
(8.35) [The awoken amnesiac also can’t remember if funds were going to be left in
one of the libraries, or what the standards for richness are wherever he is.]
(Qk ) [If iti landed heads]j am Ij in Widener and richk ? (cf. Santorio 2012)

In (8.35), for instance, we need to capture (among other things) (i) the global reading
of ‘it’ in the ‘if ’-clause; (ii) the local reading of ‘I’ in the consequent, interpreted with
respect to the assignment representing the conditional supposition; and (iii) the local
reading of the world argument and standard of richness associated with ‘rich’ in the
consequent, targeted by the question operator and varying across possible answers.
The conditional question in (8.35) is in part a question about how rich one must be
to count as rich, on the supposition that the coin landed heads.
In light of crosslinguistic links among conditionals, correlatives, and interrog-
atives, §8.2 suggested as a working hypothesis that we treat the interrogative com-
plementizer C? as having an analogous argument structure and semantics as ‘if ’/Cif .
As a way of developing the approach I suggested interpreting the external argument
of C? as determining a relevant possibility targeted by the question. In unembedded
questions the argument has been a world argument coindexed with Q (n. 99); the
questions are about what is the case in the actual world. We will see that this ar-
gument, though semantically trivial in the examples thus far, can play a non-trivial
role in the interpretation of complex sentences such as conditional questions.

198
8.5.1 Relevance conditional questions
Before turning to hypothetical conditional questions such as those in (8.34)–(8.35),
let’s start with a “non-shifty” relevance conditional question such as (8.37).

(8.36) If Alice is hungry, there are biscuits in the kitchen.

(8.37) If Alice is hungry, are there biscuits in the kitchen?

In an ordinary relevance conditional such as (8.36), the ‘if ’-clause introduces a modal
topic and the main clause declarative is used to assert something about the actual
world: that there are biscuits. Likewise in the relevance conditional question in
(8.37), the ‘if ’-clause introduces a modal topic and the main clause interrogative
is used to question something what is actually the case: whether there are biscuits.
Relevance conditional questions raise no new complications. The sentence-initial
‘if ’-clause can be analyzed as in §7.5, and the interrogative main clause can be ana-
lyzed as in our examples with unembedded interrogatives, as in (8.38) (cf. n. 88).

(8.38) Relevance conditional question

S ∶ ⟨t, t⟩

Q⟨3,a⟩ CP ∶ t

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ CP2 ∶ t

PD⟨2,a⟩ w1 -g3
⟨1,s⟩

⟨2,s⟩ C? t3a
wh2sf -g2 p1st -g1 there are biscuits t1s
if t2a Alice is hungry t2s
JCPK ≈ λgg . [ιa(g)∗ ∶ Alice is hungry in @(a(g)) ∧ @(a(g)) = a(g)(2sf)(g(1a)(1st))]
there are biscuits in @(g(3a)) = g(3a)(1s)
A proposition Tt ∈ JSK is an answer to S in c iff ∃g′′g ∶ T = [λg′g ≈ g′′ .
[ιa(g′ )∗ ∶ Alice is hungry in @(a(g′ ))∧@(a(g′ )) = a(g′ )(2sf)(g−c (1st))]

199
 there are biscuits in @(g′− ) = g′− (1s) iff
[ιa(g′′ )∗ ∶ Alice is hungry in @(a(g′′ )) ∧ @(a(g′′ )) = a(g′′ )(2sf)(g−c (1st))]
there are biscuits in @(g′′− ) = g′′− (1s)]

Roughly put, each possible answer T is a set of assignments g′ that return the same
truth value for the proposition that the contextually relevant possibility where Alice
is hungry is such that there are biscuits in @(g′ ).
First, the ‘if ’-clause introduces the maximal relevant possibility h∗ where Alice is
hungry. The global reading of the pronoun [p1st g1 ] which supplies the contextually
relevant background domain is captured in the metasemantic answerhood condi-
tion: for each answer T, every g′ ∈ T is s.t. g′ (1a)(1st) = gc (1st).
Second, just as external argument of ‘if ’ determines the possibility targeted by
the supposition, the external argument of C? determines the possibility targeted by
the question. In relevance conditional questions this possibility is the actual world:
Intuitively, (8.37) isn’t about whether there are biscuits in some modal subdomain
where Alice is hungry; it’s about whether there are actually any biscuits. So, the
external argument of C? is again supplied by a world-pronoun coindexed with the
question operator. The assignments g′ in each answer T constitute an equivalence
class with respect to whether there are biscuits in the world @(g′ ) of the possibility
represented by g′ .
A remark on interpreting the metalanguage quantification over assignments is
in order. The argument of a in the denotation of the ‘if ’-clause varies across the
assignments g′ , g′′ in the set of possible answers; however, as discussed in §§2.2, 7.2,
the items in terms of which the uniqueness condition is stated are images of the
assignments under a, i.e. items h ∈ G in the model. The metalanguage expres-
sion “ιa(g)∗ ∶ . . . ” abbreviates the quantificational condition that for some function
a ∈ Da , its value constitutes the unique maximal plurality h∗ ∈ G such that… (§7.2).
Varying the argument of a has no effect on which h∗ ∈ G constitutes the unique such
maximal plurality in the model.
8.5.2 Correlative questions
Correlatives with main clause interrogatives are rarely if ever considered in seman-
tics for interrogatives or correlatives. It is instructive to compare the treatment of
relevance conditional questions in (8.38) with the predicted analysis of an individ-
ual correlative question such as (8.39). (Assume a local reading targeted by the
question for ‘tall’, and assume a toy context-sensitive semantics for positive form
relative gradable adjectives, where d is a variable for degrees d ∈ Dd and ‘o is s-tall’
abbreviates that o’s height is at least as great as the degree-standard s for counting as

200
tall. As in §7.7, to highlight the intended reading I represent the external arguments
of the complementizers with a simple world-pronoun.)

(8.39) Individual correlative question

jo laRkii khaRii hai vo lambii hai?
REL girl standing is DEM tall is
‘Which girl is standing, is she tall?’ (cf. Srivastav 1991: ex. 3a)
S∶t

Q⟨3,a⟩ CP ∶ t

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ CP2 ∶ t

PD⟨2,a⟩
w1 g3
w1 g1 ⟨1,s⟩

⟨2,s⟩
C? t3a
Cif t2a tall t1s d1 g3
standing t2s she2 -g2 girl-w1 -g1
wh2 -g2 girl-w1 -g1
JtallK = λws .λdd .λxe .λgg . x(g) is d(g)-tall in w(g)
JCPK ≈ λgg . [ιa(g)∗ ∶ a(g)(2cf)(girlg(1a)(1s) ) is standing in @(a(g))
∧ @(a(g)) = g(1a)(1s)]
a(g)(2cf)(girlg(1a)(1s) ) is g(3a)(1d)-tall in @(g(3a)) = g(3a)(1s)
A proposition Tt ∈ JSK is an answer to S in c iff ∃g′′g ∶ T = [λg′g ≈ g′′ .
[ιa(g′ )∗ ∶ a(g′ )(2cf)(girlg−c (1s) ) is standing in @(a(g′ )) = g−c (1s)]
a(g′ )(2cf)(girlg−c (1s) ) is g′− (1d)-tall in @(g′− ) = g′− (1s) iff
[ιa(g′′ )∗ ∶ a(g′′ )(2cf)(girlg−c (1s) ) is standing in @(a(g′′ )) = g−c (1s)]
a(g′′ )(2cf)(girlg−c (1s) ) is g′′− (1d)-tall in @(g′′− ) = g′′− (1s)]

Like with the relative conditional question in (8.38), the modal domain for the main
clause interrogative is targeted directly by the question operator. But whereas the
‘if ’-clause in (8.38) introduces a topical modal possibility, the correlative clause

201
introduces an actual possibility about a topical individual. As in (7.46), the world
of the possibility introduced in the correlative clause is identified with the world of
the discourse, gc (1s). Though the correlative clause in (8.39) doesn’t shift a modal
domain, it shifts the interpretation of the correlative proform. The correlative clause
introduces a topical girl (=a(g)(2cf)(girlgc (1s) )) who is actually standing, and the
interrogative main clause comments on that same girl, asking whether she is tall.
Each possible answer T is, roughly, a set of assignments g′ which determine the
same standard s for counting as tall (=g′ (1d)), and return the same truth value for
the proposition that the selected girl (=a(g)(2cf)(girlgc (1s) )) who is standing in the
actual world @(gc ) is s-tall in @(g′ ).
So, our analyses of correlatives, proforms, and interrogatives compositionally
derive the relevant inventory of readings: (i) the global reading of ‘girl’ in the cor-
relative clause, (ii) the local reading of the proform in the main clause bound by the
correlative clause, and (iii) the local reading of ‘tall’ in the main clause, targeted by
the question. The global reading of ‘girl’ and correlative clause domain is captured
via the metasemantic condition on answerwood. The local reading of the correlate
follows from the representation of the proform as a copy of its antecedent relative
expression, here [she2 -g2 girl-w1 g1 ], and the correlative clause assignment-binder
from PD⟨2,a⟩ . The proform in the interrogative clause correctly receives a local read-
ing under the correlative clause without being targeted by the question. On the flip
side, the world-argument and standard-pronoun [d1 g3 ] associated with ‘tall’ receive
local readings under the question operator without being shifted by the correlative
clause. (8.39) is in part a question about how tall one must be to count as tall. The
assignments g′ in each answer determine the same standard of tallness g′ (1d) (via
g′ ≈ g′′ ), though which standard is determined varies across possible answers. So, the
correlative interrogative is correctly derived as a question about whether a certain
girl who is actually standing is tall.
8.5.3 Hypothetical conditional questions
Finally, let’s turn to hypothetical conditional questions such as (8.33)–(8.35). The
§8.2-syntax/semantics of interrogatives integrates straightforwardly with the account
of correlative conditionals in §§7.5–7.7. The predicted LF and interpretation for
(8.35) are as follows, on the shifted reading of ‘I’ under the conditional supposition,
and local reading targeted by the question of the relevant standard for richness as-
sociated with ‘rich’. (As with ‘tall’ in (8.39), assume a simplified context-sensitive
semantics for ‘rich’, where ‘o is s-rich’ abbreviates that o’s degree of wealth is at least
as great as the degree-standard s for counting as rich.)

202
 (8.40) Hypothetical conditional question
‘If it landed heads, am I rich?’
S ∶ ⟨t, t⟩

Q⟨3,a⟩ CP ∶ t

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ CP2 ∶ t

PD⟨2,a⟩
w1 g2
⟨1,s⟩

⟨2,s⟩
wh2sf -g2 p1st -g1 C? t3a I1 g2
it2 -g1 heads-t2s rich t1s d1 g3
if t2a

JCP2 K ≈ λgg ∶ @(g(3a)) = g(2a)(1s) . g(2a)(1e) is g(3a)(1d)-rich in @(g(3a))

JCPK ≈ λgg . [ιa(g)∗ ∶ g(1a)(2e) landed heads in @(a(g)) ∧ @(a(g)) = a(g)(2sf)(g(1a)(1st))]
a(g)(1e) is g(3a)(1d)-rich in @(g(3a)), provided @(g(3a)) = a(g)(1s)
A proposition Tt ∈ JSK is an answer to S in c iff ∃g′′g ∶ T = [λg′g ≈ g′′ .
[ιa(g′ )∗ ∶ g−c (2e) landed heads in @(a(g′ )) ∧ @(a(g′ )) = a(g′ )(2sf)(g−c (1st))]
a(g′ )(1e) is g′− (1d)-rich in @(g′− ), provided @(g′− ) = a(g′ )(1s) iff
[ιa(g′′ )∗ ∶ g−c (2e) landed heads in @(a(g′′ )) ∧ @(a(g′′ )) = a(g′′ )(2sf)(g−c (1st))]
a(g′′ )(1e) is g′′− (1d)-rich in @(g′′− ), provided @(g′′− ) = a(g′′ )(1s)]

Roughly, each possible answer T is a set of assignments g′ which (i) determine the
same standard s for counting as rich (=g′ (1d)), and (ii) return the same truth value
for the proposition that the relevant possibility h∗ where the coin gc (2e) landed
heads, @(h∗ ) = h∗ (2sf)(gc (1st)), is s.t. the shifted individual h∗ (1e) is s-rich, pro-
vided (iii) that the world of g′ is identical to the world of h∗ .
The account derives the range of local/global readings for conditional questions
discussed above and in §8.1 — notably, global readings (e.g. ‘it’), local readings un-
der the conditional supposition (e.g. ‘I’), and local readings in the interrogative con-
sequent targeted by the question (e.g. ‘rich’). The standard associated with ‘rich’

203
correctly receives a local reading under the question operator. (8.40) is in part a
question about what standard of richness to accept; the question is about the na-
ture of a topical possibility, not simply about the world narrowly construed. The
assignments in each answer agree in what standard of richness g′ (1d) they deter-
mine (via g′ ≈ g′′ ), though which standard is determined varies across the possible
answers. By contrast, the interpretations of the pronouns ‘it’ and ‘I’ are constant
across answers. Just as in each possible answer the topical modal possibility h∗ is
about the same contextually relevant object gc (2e), likewise each possible answer is a
proposition about the individual h(1e) representing (say) the epistemic counterpart
of the speaker in that possibility (§3.4). ‘I’ in the interrogative consequent receives
a local reading under the conditional supposition without necessarily receiving a
local reading targeted by the question. The global readings of ‘it’ and specifier ar-
gument of ‘if ’ are again captured via the metasemantic condition on answerhood,
anchoring their interpretation to the discourse context. I am not aware of other
accounts of conditionals, questions, or conditional questions which similarly derive
the observed range of local and global readings.
The semantics captures the intuition that the possible answers in a hypotheti-
cal conditional question are specifically about the possibility described by the an-
tecedent. There are two critical moving parts: the presupposition associated with
the interrogative complementizer, and the correlative binding requirement.
First, §§6–7 pursued a uniform approach to relativization in our semantics for
relative complementizers — Crel in relativization of individuals (§6.1), and ‘if ’/Cif
in relativization of possibilities (§§7.2, 7.7). Drawing on crosslinguistic links be-
tween conditional and interrogative clauses, §8.2 followed suit in proposing a par-
allel treatment for the interrogative complementizer C? . The relevant contrast in
the entry for C? in (8.7) is that the condition on the external (world, individual)
argument is a presupposition. Intuitively, a no-answer to a conditional question
such as (8.33) shouldn’t end up including possibilities where school is canceled (and
perhaps even where it snows) but fail to verify the interrogative due to being outside
the contextually relevant set of snow-possibilities. The possibilities in each answer
should count as relevantly equivalent due to being equivalent with respect to whether
school is canceled. Implementing the condition as a presupposition captures this.
The assignments in each possible answer T in (8.40) are restricted to assignments
representing the topical modal possibility described by the ‘if ’-clause — assignments
g′ s.t. @(g′− ) = h∗ (1s) = @(h∗ ).
By contrast, there isn’t an analogous no-answer relevant in the semantics of
relative clauses or ‘if ’-clauses. For purposes of deriving the embedding operator’s
restrictor argument, what matters is simply the items that do satisfy both of the

204
conditions in question — e.g., in ‘every toy which broke’, the broken things which
are identical to some toy or other, or in ‘if it snows’, the snow-possibilities which are
part of the relevant domain. That said, it would be interesting to compare presup-
positional variants of the lexical entries for Crel /‘if ’ in (6.20)/(7.15) analogous to the
entry for C? in (8.7), as e.g. in (8.41) where JifK = JCif K = JC? K .

(8.41) JifK = JCif K = JC? K = λaa .λpst .λw′s .λgg ∶ @(a(g)) = w′ (g) . ∀w s.t. w(g) =
@(a(g)), p(w)(g)

Developing the account in this way could be understood as providing a general

presuppositional approach to quantifier domain restriction. The connections with
notions of “specificity” (Enç 1991) and “discourse linking” (Pesetsky 1987) from
§6.3.1 may provide additional avenues to explore.
Second, as part of the unified approach to relative, conditional, and interrog-
ative clauses, I suggested understanding the external world argument in a condi-
tional/interrogative clause as determining the relevant possibility to be targeted by
the supposition/question. The questions in our previous examples were about what
is the case in the actual world. Accordingly, the external argument was a world pro-
noun receiving a local reading under Q, and the presuppositional constraint was the
trivial constraint that @(g′− ) = g′− (1s) (given our general metasemantic assump-
tion). In contrast, in hypothetical conditional questions the possibility targeted by
the interrogative clause is the possibility introduced in the antecedent. The external
argument of C? receives a local reading under the antecedent supposition rather
than the question operator, represented via the world-pronoun [w1 g2 ] coindexed
with the ‘if ’-clause (cf. n. 99). The derived presuppositional constraint is the non-
trivial constraint @(g′ ) = @(h∗ ) that the assignments g′ in each possible answer
represent the world of the topical possibility h∗ . The possible answers T are correctly
diagnosed as being propositions specifically about the relevant possibility where the
coin landed heads.
There is an important contrast between the treatments of the correlates/proforms
in hypothetical conditionals with main clause interrogative vs. declarative CPs. In
§§7.5–7.7, the predicted correlate of the ‘if ’-clause was directly linked to the main
predicate’s world argument. The assignment-binder T — or an embedding modal, as
in (7.25) reproduced in (8.42) — was treated as raising from a semantically vacuous
complementizer generated by CP-recursion above the conditional.

(8.42) Alice thinks [ CP1 that [ CP2 [if it snows] [ CP3 (then) school will close]]]

In contrast, in correlative conditional questions the correlate cannot be directly

205
linked to the main predicate’s world argument; the (non-vacuous) embedded in-
terrogative complementizer must be generated (e.g.) to trigger wh movement, sat-
isfy the selection requirements of the question-embedding operator, and derive the
intended question meaning. The predicted correlate is instead the adjacent world
argument in SpecCP — an adjacent world-pronoun, as in (8.40), or a topicalized
proform such as ‘then’, as in (8.43) (n. 99).

(8.43) ‘If…then’ conditional question

‘If it landed heads, then am I rich?’
S ∶ ⟨t, t⟩

Q⟨3,a⟩ CP ∶ t

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ CP2 ∶ t

PD⟨2,a⟩ XP⟨3,s⟩ ∶ ⟨t, t⟩

t3s
⟨2,s⟩ then2sf -g2 p1st -g1 ⟨1,s⟩
wh2sf -g2 p1st -g1
if t2a it2 -g1 heads-t2s C? t3a I1 -g2 rich-t1s d1 -g3

A proposition Tt ∈ JSK is an answer to S in c iff ∃g′′g ∶ T = [λg′g ≈ g′′ .

[ιa(g′ )∗ ∶ gc− (2e) landed heads in @(a(g′ )) ∧ @(a(g′ )) = a(g′ )(2sf)(g−c (1st))]
a(g′ )(1e) is g′− (1d)-rich in @(g′− ), provided @(g′− ) = a(g′ )(2sf)(g−c (1st)) iff
[ιa(g′′ )∗ ∶ g−c (2e) landed heads in @(a(g′′ )) ∧ @(a(g′′ )) = a(g′′ )(2sf)(g−c (1st))]
a(g′′ )(1e) is g′′− (1d)-rich in @(g′′− ), provided @(g′′ ) = a(g′′ )(2sf)(g−c (1st))]

Just like in our examples with unembedded interrogatives, the question operator
raises from the assignment argument position of the embedded interrogative com-
plementizer. No mechanism of CP-recursion is predicted. This is a critical feature of
the account, since CP-recursion has been argued to be degraded or impossible under
interrogative complementizers (e.g. Iatridou & Kroch 1993). Unlike (7.25), the
examples in (8.44)–(8.45) are unacceptable with ‘then’.

206
(8.44) Every boy wonders if his mother comes (∗ then) what he will eat.
(Iatridou & Kroch 1993: exs. 54a–55a)
(8.45) Every boy wonders whether if his toy breaks (∗ then) he will get a new one.

The A-movement from topicalizing the proform, as in (8.43), degrades the further
wh movement in (8.44)–(8.45) (cf. §7.5). The contrast between embedded ‘then’
conditionals with main clause declaratives vs. interrogatives falls out of our gen-
eral account of conditional proforms and syntax/semantics for interrogative (con-
ditional, relative) clauses.1⁰⁷
8.6 Recap: Standardizing quantification. Unifying relativization, conditionals,
questions
The compositional semantics of interrogatives in this section parallels our previous
treatments of type-driven movement with other quantifiers and complementizers:
• Interrogative, conditional, and (non-vacuous) declarative complementizers
raise from V as quantifiers over worlds, capturing the obligatory local reading
of the main predicate’s world argument. Analogously, the relative comple-
mentizer raises from the gap position in an individual relative clause, captur-
ing the obligatory link between the gap and nominal head.
• Modal quantifiers — including the topmost assignment-binders T and Q, modal
and attitude verbs, and PD — raise from inside their complement clause as
quantifiers over assignments, determining the relevant modal domain. Anal-
ogously, determiner quantifiers in relative constructions raise from inside
their complement relative clause, determining the relevant domain of indi-
viduals.
• Relative/wh pronouns are analyzed as choice-function pronouns — type ⟨et, e⟩
in relativization of individuals (e.g. ‘which’), type ⟨st, s⟩ in relativization of
possibilities (e.g. ‘whether’). Conditional and individual proforms are repre-
sented as copies of their antecedent relative/wh expression.
• Relative, conditional, correlative, and interrogative complementizers — Crel
in relativization of individuals, ‘if ’/Cif in relativization of possibilities, C? in
1⁰⁷It is interesting that conditional ‘then’ questions such as (8.43) (where ‘then’ is fronted) are
ungrammatical in V2 languages; unlike in (7.67), there is no empty complementizer position for
the verb to move to. This of course isn’t a general account of the licensing of embedded ‘then’
conditionals. CP-recursion is degraded in various environments, not simply in questions (Iatridou
1991, Iatridou & Kroch 1993, Bhatt & Pancheva 2006).

207
 interrogatives — relate a pair of individuals/worlds with respect to the prop-
erty of individuals/worlds expressed by the complement.
– In individual relatives the relative clause targets a topical subset of indi-
viduals supplied by the relative phrase. The set of individuals may then
be commented on by a determiner quantifier’s scope argument.
– In conditional interpretations with ‘if ’, the ‘if ’-clause targets a topical
subdomain of a relevant modal possibility. This topical possibility may
then be commented on by an adjoined phrase (NP/VP/IP/CP).
– In individual-correlative interpretations with Cif the correlative clause
targets a possibility describing topical individuals in the actual world.
This topical possibility is commented on by the adjoined main clause.
– In interrogatives the interrogative clause targets the possibility to be
targeted by the question.
The proposed assignment-variable-based account affords a unified approach to con-
ditionality, relativization, and questions, and an elegant generalized treatment of
quantification in various types of sentences and clausal structures (cf. §§1, 3.2, 6.1,
6.3.1, 7.7).1⁰⁸

9 Conclusion
This paper (?) has initiated a project of developing a linguistic theory which posits
variables for assignment functions in the syntax, and treats semantic values sys-
tematically in terms of sets of assignments in the model. Principal features of the
account are that it standardizes quantification across domains, and it systematizes
various seemingly diverse linguistic shifting phenomena. Such phenomena include
data with quantifiers and scope, intensionality, and local/global readings in vari-
ous embedded environments, such as with modals, attitude verbs, questions, and
conditionals. The semantics affords a unified analysis of the context-sensitivity of
pronouns, epistemic modals, etc., in the spirit of contextualist theories. Yet it im-
proves in compositionally deriving certain distinctive shifting/binding phenomena
(e.g. with epistemic modals), and in providing a framework for theorizing about
expressions’ different tendencies for local/global readings. Resources for capturing
these phenomena have grown increasingly complex in current theories.
The syntax and lexical/compositional semantics delineate the sources of inten-
sionality and assignment-shifting in the clausal architecture. Binding with indi-
viduals/worlds/assignments is derived uniformly from a generalized binder-index,
1⁰⁸If only.

208
which attaches directly to expressions. The account avoids introducing added pa-
rameters of interpretation, quantification-specific composition rules, or interpre-
tive principles (such as for reconstructed phrases, pronouns vs. traces, donkey pro-
nouns). The semantics is fully compositional.
Along the way I appealed to independently motivated resources from the syntax
and semantics literatures in motivating analyses of particular constructions. Cer-
tain features of the account are of general interest, independent of the particular
assignment-variable-based implementation. These include the formalization of as-
signment modification; the definition of the generalized binder-index; the approach
to the syntax/semantics interface with determiner quantifiers, modal quantifiers,
and various types of complementizers; the treatments of locality/globality constraints
(e.g., with pronouns, epistemic modals); applications to classic puzzles with names
in attitude ascriptions, including apparent non-specific, de dicto, and bound uses;
a unified analysis of certain indefinites, wh-words, relative words, and anaphoric
demonstrative pro-forms; a formal diagnosis of notions such as “specificity” (defi-
niteness, D(iscourse)-linking) in connection with several types of quantificational
and non-quantificational DPs; a unified treatment of various types of restrictive
modification in quantificational DPs, with applications to “free R” (modifier) and
“inherent R” (argument) readings of English genitive constructions (prenominal
and postnominal); a distinction between trace- and pronoun-binding, with applica-
tions to weak crossover; a speculative general treatment of apparent binding out of
DPs and weak crossover effects with donkey pronouns, inverse linking, and genitive
binding; uniform compositional semantics for ‘if ’-clauses when adjoined to NP, VP,
IP, and CP, and for individual correlative clauses adjoined to the main clause or
directly to the correlate DP; a unified syntax/semantics for ‘if ’-clauses and individual
correlative clauses, and unified analysis of conditional and individual proforms; a
unified approach to relative, conditional, and interrogative complementizers; and
a fully compositional semantics for quantifier raising, wh interrogatives, and rela-
tivization.
The initial focus of the paper was on applying the assignment-variable-based
framework to local/global readings with quantifiers, attitude verbs, and modal verbs;
however, we have seen that the potential applications of the account extend more
broadly. Diverse types of shifting phenomena were examined across a range of lin-
guistic environments, such as relativization structures, conditionals, and interrog-
atives. Although I emphasized certain formal similarities among context-sensitive
expressions, there are of course differences among them. Issues regarding further
(grammatical, lexical, metasemantic, conversational) constraints on readings, to help
rein in the flexibility of the system, call for more thorough investigation. The spec-

209
ulative discussions of assignment-quantification/binding with determiner quanti-
fiers, names, different types of pronoun binding, and comparisons among condi-
tionals, individual correlatives, and interrogatives raise difficult questions about bind-
ing principles, traces vs. pronouns, (non-)linguistic anaphora, and wh movement
and topicalization, among many others. Applications to other types of conventional
meanings, such as conventional implicature, presupposition, and expressives; cate-
gories such as tense, aspect, and mood; anaphora such as discourse anaphora; and
expressions and constructions such as resumptive pronouns, control constructions,
imperatives, counterfactuals, modal adjectives/adverbs, and ellipsis, may provide
interesting avenues to explore. I hope the preliminary developments here may il-
lustrate the fruitfulness of an assignment-variable-based approach to investigating
these and additional linguistic phenomena.

Appendix Formal overview: Syntax, Semantics, Metasemantics

A Sample lexical entries
(2.6) JlovesK = λws .λxe .λye .λgg . y(g) loves x(g) in w(g)
(2.6) JeverythingK = λP⟨e,t⟩ .λgg .∀xe ∶ P(x)(g) (preliminary)
(2.4) Jviσ K = λaa .λγσnn . . . λγσ1 1 .λgg . a(g)(iσ)((↓ γ n )(g)) . . . ((↓ γ 1 )(g))
a. For α ∈ {e, s, t}, Jviα K = λaa .λgg . a(g)(iα) b. Jgi K = λgg . g(ia)
(2.5) Jtiσ K = λγσnn . . . λγσ1 1 .λgg . g(iσ)((↓ γ n )(g)) . . . ((↓ γ 1 )(g))
a. For β ∈ {e, s, t, a}, Jtiβ K = λgg . g(iβ)
(3.11) Generalized binder-index
J⟨i,τ ⟩ K = λα⟨⟨τ,⟨1 σ1 ⋯σn , t⟩1 ⋯⟩n ⟩,σ⟩ . λβ⟨1 σ1 ⋯σn , t⟩1 ⋯⟩n .
α(λχ.λγ 1 ⋯λγ n .λg . β(γ 1 )⋯(γ n )(g[(↓ χ)(g)/iτ ]))

(3.6) JthatK = JCdec K = λaa .λp⟨s,t⟩ .λgg . ∀w s.t. w(g) = @(a(g)), p(w)(g)
(7.15) JifK = JCif K = λaa .λpst .λw′s .λgg . @(a(g)) = w′ (g) ∧ ∀w s.t. w(g) = @(a(g)), p(w)(g)
(8.7) JC? K = λaa .λpst .λw′s .λgg ∶ @(a(g)) = w′ (g) . ∀w s.t. w(g) = @(a(g)), p(w)(g)
(6.20) JCrel K = λaa .λPet .λye .λxe .λgg . x(g) = y(g) ∧ P(x)(g)
(alternatives in n. 66, (6.74a), §8.5.3)

210
(3.10) JthinkK = λws .λA⟨a,t⟩ .λxe .λgg . ∀aa s.t. a(g) is compatible with x(g)’s state
of mind in w(g)∶ A(a)(g)
(3.9) JmayK = λws .λr⟨s,at⟩ .λA⟨a,t⟩ .λgg . ∃aa s.t. r(w)(a)(g)∶ A(a)(g)
(7.35) JmayK = λaa .λAat .λA′at .λgg . ∃a′a s.t. a′ (g) ≤ a(g) ∧ A(a′ )(g)∶ A′ (a′ )(g)
(alternative argument structure in (7.36))
(3.8) JTK = λA⟨a,t⟩ .λgg . for aa = λgg .g− , A(a)(g)
(8.8) JQK = λAat .λTt .λgg . for aa = λgg .g− , ∃g′′g ∶ T = [λg′ ∶ g′ ≈ g′′ . A(a)(g′ ) = A(a)(g′′ )]
(7.13) JPDK = λAat .λA′at .λgg . [ιa(g)∗ ∶ A(a)(g)] A′ (a)(g)

(7.60) JDEFK = λP+⟨a,et⟩ .λQ+⟨a,et⟩ .λgg . [ιx(g)∗ ∃a∶ P+ (a)(x)(g)] Q+ (a)(x)(g)

(6.23) JeveryK = λP+⟨a,et⟩ .λQ+⟨a,et⟩ .λgg . [∀xe ∃aa ∶ P+ (a)(x)(g)] Q+ (a)(x)(g)
(universal donkey reading)
(6.54) JeveryK = λP+⟨a,et⟩ .λQ+⟨a,et⟩ .λgg . [∀xe ∃aa ∶ P+ (a)(x)(g)]
∃a′a ∶ P+ (a′ )(x)(g) ∧ Q+ (a′ )(x)(g) (existential donkey reading)
(6.87) JeveryK = λPet .λyae .λQaet .λgg . ∀xe ∃aa ∶ x(g) = y(a)(g)∧P(x)(g), Q(a)(x)(g)
(alternative Q DP argument structure)

(6.90)/(6.99) JXK = J’sK = λxe .λPet .λye .λgg . y(g) = x(g) ∧ P(y)(g)
(alternatives in nn. 65, 66, §6.3.1)

Choice-function pronouns (§§4.4, 6.1, 6.2.1, 6.3, 7.2, 7.7, 8.3)

a. JFi,cf K = λaa .λPτ t .λgg . a(g)(icf)((↓ P)(g))
b. A choice function Fcf is a function F ∶ [X → T] → X s.t. ∀P ∶ X → T ≠ ∅∶ P(F(P)).

B Metasemantics / Metalanguage
(2.1) Models M: (2.2) Domains / Semantic types:
– E: set of entities – De = EDg
– T: set of truth-values, {0, 1} – Dt = {0, 1}Dg
– W: set of worlds – Ds = WDg
– G: set of assignmentsM – Da = GDg
– Dg =domain of assignmentsD
– Dαβ = DDβ α

(4.9) Assignment modification

211
 a. [z/iτ ] ∶= λgg . ιmg ∶ m(iτ ) = z ∧ m(jσ) = g(jσ), for all jσ ≠ iτ
b. gg [⋯]1 . . . [⋯]n ∶= [⋯]1 ○ . . . ○ [⋯]n (g)

Constraints on readings / Interpretation of assignments: §§1, 2.1, 3.1, 3.3, 3.4, 4.1,
4.3.3, 5, 6.3.3, 7.7
Metasemantics: §§2.2, 3.3, 4.1, 6.1.3, 7.2, 8.2, 8.5

(8.11) a. A declarative sentence S is true in c iff JSK(gc ) = 1

b. Tt is an answer to an interrogative sentence S? in c iff (i) T ∈ JS?K , and
(ii) for all g′ ∈ T and assignment-indices ia, g′ (ia) = g−c

C Syntax
Sample trees and examples: e.g., §§3.2, 4, 6.1, 6.2, 6.3.3, 7.2, 7.5, 7.7, 8.2, 8.5

(3.3)
X⟨i,a⟩ CP/C′

[C0 ]⟨j,s⟩ ⋮

C0 tia V tjs
(8.6) S

Q⟨i,a⟩
[C0 ]⟨j,s⟩ IP

C? tia …tjs …
(7.11) CP

PD⟨i,a⟩
C

⟨j,s⟩ IP

if/Cif tia . . . tjs . . .

(6.4) [ DP D0 [ CPrel whrel NP [ C Crel IP ]]]]

212
(4.1) S

T⟨1,a⟩ CP

⟨1,s⟩ IP

Cdec t1a
VP
he1 g1

⟨2,a⟩ CP

thinks t1s
⟨2,s⟩ IP

that t2a
it2 g1 cried t2s
(4.21) ‘Freddoi met Freddok .’
[ S T⟨1,a⟩ [ CP [Cdec t1a ]⟨1,s⟩ [[ DP F1cf g1 [ NP Freddo w1 g1 ]]
[ VP met t1s [ DP F2cf g1 [ NP Freddo w1 g1 ]]]]]]

213
(6.26) S

T⟨1,a⟩ CP

⟨1,s⟩ IP

Cdec t1a

DP⟨2,a⟩ VP

every⟨2,a⟩ CPrel
liked t1s
it3 g2
toy-w1 -g1
C
wh2 g2
baby-w1 -g1 ⟨2,e⟩

Crel t2a t2e

got-w1 -g1
a3 g2
toy-w1 -g1

214
(6.125)‘every boy’s cat’
DP{...⟨2,a⟩... } ∶ ⟨et, t⟩

DP∗∗{⟨2,e⟩,⟨2,a⟩} ∶ ⟨t, t⟩ D⟨3,a⟩ ∶ ⟨et, t⟩

’s⟨3,a⟩ CP
every⟨2,a⟩ CP

C C

F2cf -g2 boy-w1 -g1 …t2a … F3cf -g3 cat-w1 -g1 ⟨1,e⟩

t1e
C t3a R1 g1 t2e

(6.128)‘some child of every parent’

DP {...⟨2,a⟩... } ∶ ⟨et, t⟩

DP∗∗{⟨2,e⟩,⟨2,a⟩} ∶ ⟨t, t⟩ DP∗⟨3,a⟩ ∶ ⟨et, t⟩

some⟨3,a⟩ CP
every⟨2,a⟩ CP

C C

F2cf -g2 parent-w1 -g1 …t2a … F3cf g3 …t3a …

t2e
child-w1 -g1
n. 66 DP: D XP + individual-denoting CPrel (alternative)
a. ‘Every girl baby which Alice likes laughed.’
⟨2,e⟩
b. [ DP every⟨2,a⟩ [ XP girl [[X w1 -g1 ] [ CP [wh2cf∗ -t2a baby] [ C Crel [ IP Alice
likes-w1 -g1 t2e ]]]]]]
c. JXK = λws .λx′se .λP′⟨s,et⟩ .λye .λgg . y(g) = x′ (w)(g) ∧ P′ (w)(y)(g)
JCrel K = λPet .λx′se .λws .λgg . ιy(g) s.t. y(g) = x′ (w)(g) ∧ P(y)(g)
(6.87) QP: Q DP (alternative)
a. ‘every baby’
b. [ QP [ Q0⟨i,a⟩ every P-g] [ DP Fcf -tia baby]]

215
(6.100)D+XP free R prenominal/postnominal genitive LFs (alternative)
a. Alice’s cat / every cat of Alice’s
b. [ DP⟨2,a⟩ def/every⟨2,a⟩ [ XP [R1eet -g1 Alice] [ X ’s [ NP F2cf -t2a cat]]]]
(6.101)D+XP inherent R prenominal/postnominal genitive LFs (alternative)
a. Alice’s brother / every brother of Alice’s
b. [ DP⟨2,a⟩ def/every⟨2,a⟩ [ XP [P1et g1 ] [ X ’s [ NP F2cf t2a [brother Alice]]]]]
(7.16) CP⟨2,a⟩ ∶ ⟨t, t⟩

PD⟨2,a⟩ t

⟨s, t⟩
wh2sf g2 p1st g1
⟨2,s⟩ t

if t2a rained t2s

(7.18) ‘fine if you drive’
NP ∶ ⟨e, t⟩

NP ∶ ⟨e, t⟩ CP⟨2,a⟩ ∶ ⟨t, t⟩

fine w1 g2
PD⟨2,a⟩

⟨2,s⟩
wh2sf -g2 p1st -g1
if t2a you2 -g1 drive-t2s

216
(7.46) Hindi multiple correlative:
‘Which girl played with which boy, she defeated him.’ (Dayal 1996)
S∶t

T⟨1,a⟩ CP ∶ t

⟨1,s⟩ IP ∶ t

Cdec t1a

⟨2,a⟩
CP1 ∶ ⟨t, t⟩ IP1 ∶ t

PD⟨2,a⟩

w1 g1 she2 -g2 girl-w1 -g1

defeated t1s
⟨2,s⟩ him3 -g2 boy-w1 -g1

Cif t2a

wh2 -g2 girl-w1 -g1

played-with t2s
wh3 -g2 boy-w1 -g1

217
(7.57) Individual correlative (DP-adjunction)
S∶t

T⟨1,a⟩ CP ∶ t

⟨1,s⟩ IP ∶ t

Cdec t1a

DP ∶ ⟨et, t⟩ VP ∶ ⟨e, t⟩

tall t1s

⟨2,a⟩ DP1 ∶ ⟨et, t⟩

CP1 ∶ ⟨t, t⟩

PD⟨2,a⟩ she2 -g2 girl-w1 -g1

w1 g1
⟨2,s⟩

Cif t2a
standing t2s
wh2 -g2 girl-w1 -g1

218
 (8.39) Correlative question
S∶t

Q⟨3,a⟩ CP ∶ t

CP⟨2,a⟩ ∶ ⟨t, t⟩ CP ∶ t

PD⟨2,a⟩
w1 g3
w1 g1 ⟨1,s⟩

⟨2,s⟩
C? t3a
Cif t2a tall t1s d1 g3
standing t2s she2 -g2 girl-w1 -g1
wh2 -g2 girl-w1 -g1
(7.67) S∶t

T⟨1,a⟩ CP

CP ∶ t
C∅ t1a

CP⟨2,a⟩ ∶ ⟨t, t⟩ CP ∶ t

PD⟨2,a⟩ XP⟨1,s⟩ ∶ ⟨t, t⟩ C∶t

⟨2,s⟩ then2sf -g2 p1st -g1 e IP

wh2sf -g2 p1st -g1
if t2a it2 -g1 heads-t2s
I1 -g2 in-WL t1s

219
 (8.40) S ∶ ⟨t, t⟩

Q⟨3,a⟩ CP ∶ t

CP⟨2,a⟩ ∶ ⟨t, t⟩ CP ∶ t

PD⟨2,a⟩
w1 g2
⟨1,s⟩

⟨2,s⟩
wh2sf -g2 p1st -g1 C? t3a I1 g2
it2 -g1 heads-t2s rich t1s d1 g3
if t2a
(8.43) S ∶ ⟨t, t⟩

Q⟨3,a⟩ CP ∶ t

CP⟨2,a⟩ ∶ ⟨t, t⟩ CP ∶ t

PD⟨2,a⟩ XP⟨3,s⟩ ∶ ⟨t, t⟩

t3s
⟨2,s⟩ then2sf -g2 p1st -g1 ⟨1,s⟩
wh2sf -g2 p1st -g1
if t2a it2 -g1 heads-t2s C? t3a I1 -g2 rich-t1s d1 -g3

References
Abbott, Barbara. 2002. Donkey demonstratives. Natural Language Semantics 10,
285–298.

Abney, Steven Paul. 1987. The English noun phrase in its sentential aspect. Ph.D.
thesis, MIT.

Abusch, Dorit. 1994. The scope of indefinites. Natural Language Semantics 2,

83–135.

220
Adamou, Evangelia. 2011. Temporal uses of definite articles and demonstratives
in Pomak (Slavic, Greece). Lingua 121, 871–889.

Anand, Pranav & Andrew Nevins. 2004. Shifty operators in changing contexts:
Indexicals in Zazaki and Slave. In Proceedings of SALT 14. Evanston, IL.

Aoun, Joseph & Norbert Hornstein. 1992. Bound and referential pronouns.
In C.-T. James Huang & Robert May (Eds.), Logical structure and linguistic
structure: Cross-linguistic perspectives, pp. 1–23. Dordrecht: Kluwer.

Arsenijević, Boban. 2009. Correlatives as types of conditional. In Lipták (2009a),

pp. 131–156.

Bach, Emmon, Eloise Jelinek, Angelika Kratzer, & Barbara H. Partee (Eds.).
1995. Quantification in natural languages. Dordrecht: Kluwer.

Bagchi, Tista. 1994. Bangla correlative pronouns, relative clause order, and D-
linking. In Miriam Butt, Tracy Holloway King, & Gillian Ramchand
(Eds.), Theoretical perspectives on word order in South Asian languages, pp. 13–
30. Stanford: CSLI.

Baltin, Mark. 2012. The structural signature of pronouns. MS, NYU.

Baltin, Mark & Chris Collins (Eds.). 2001. The handbook of contemporary
semantic theory. Oxford: Blackwell.

Barbosa, Pilar, Danny Fox, Paul Hagstrom, Martha McGinnis, & David
Pesetsky (Eds.). 1998. Is the best good enough? Optimality and competition in
syntax. Cambridge, Mass.: MIT Press.

Barker, Chris. 2002. The dynamics of vagueness. Linguistics and Philosophy 25,
1–36.

Beaver, David. 2001. Presupposition and assertion in dynamic semantics. Stanford:

CSLI Publications.

Bernstein, Judy. 2001. The DP hypothesis: Identifying clausal properties in the

nominal domain. In Baltin & Collins (2001), pp. 536–561.

Bernstein, Judy. 2008. Reformulating the Determiner Phrase analysis. Language

and Linguistics Compass 2, 1246–1270.

221
Beshears, Anne. 2017. The demonstrative nature of the Hindi/Marwari correlative.
Ph.D. thesis, Queen Mary University of London.

Bhat, D.N.S. 2004. Pronouns. New York: Oxford University Press.

Bhatt, Rajesh. 1999. Covert modality in non-finite contexts. Ph.D. thesis, University
of Pennsylvania.

Bhatt, Rajesh. 2002. The raising analysis of relative clauses: Evidence from
adjectival modification. Natural Language Semantics 10, 43–90.

Bhatt, Rajesh. 2003. Locality in correlatives. Natural Language and Linguistic

Theory 21, 485–541.

Bhatt, Rajesh & Anikó Lipták. 2009. Matching effects in the temporal and
locative domains. In Lipták (2009a), pp. 343–372.

Bhatt, Rajesh & Roumyana Pancheva. 2006. Conditionals. In Everaert & van
Riemsdijk (2006), pp. 638–687.

Bianchi, Valentina. 1999. Consequences of antisymmetry: Headed relative clauses.

Berlin: Mouton de Gruyter.

Bianchi, Valentina. 2000. The raising analysis of relative clauses: A reply to

Borsley. Linguistic Inquiry 31, 123–140.

Bittner, Maria. 2001. Topical reference for individuals and possibilities. In

Rachel Hastings, Brendan Jackson, & Zsofia Zvolenszky (Eds.), Proceed-
ings from SALT XI, pp. 36–55. Ithaca, N.Y.: CLC Publications.

Boeckx, Cedric. 2001. Mechanisms of chain formation. Ph.D. thesis, University of

Connecticut.

Boeckx, Cedric & Kleanthes K. Grohmann. 2004. SubMove: Towards a unified

account of scrambling and D-linking. In David Adger, Cécile de Cat, &
George Tsoulas (Eds.), Peripheries, pp. 241–257. Dordrecht: Kluwer.

Boeckx, Cedric, Norbert Hornstein, & Jairo Nunes. 2007. Overt copies in
reflexive and control structures: A movement analysis. In Anastasia Conroy,
Chunyuang Jing, Chizuru Nakao, & Eri Takahashi (Eds.), University of
Maryland working papers in linguistics 15, pp. 1–45. College Park: University of
Maryland.

222
Bošković, Željko & Jairo Nunes. 2007. The copy theory of movement: A view
from PF. In Corver & Nunes (2007), pp. 13–74.

Brasoveanu, Adrian. 2007. Structured nominal and modal reference. Ph.d. thesis,
Rutgers University.

Browning, M.A. 1996. CP recursion and that-t effects. Linguistic Inquiry 27, 237–
255.

Bruening, Benjamin. 2009. Selectional asymmetries between CP and DP suggest

that the DP hypothesis is wrong. In Laurel MacKenzie (Ed.), Proceedings of the
32nd Annual Penn Linguistics Colloquium, pp. 26–35. University of Pennsylvania
Working Papers in Linguistics.

Burge, Tyler. 1973. Reference and proper names. Journal of Philosophy 70, 425–
439.

Büring, Daniel. 2004. Crossover situations. Natural Language Semantics 12, 23–
62.

Büring, Daniel. 2005. Binding Theory. Cambridge: Cambridge University Press.

Cable, Seth. 2009. The syntax of the Tibetan correlative. In Lipták (2009a), pp.
195–222.

Cable, Seth. 2010. The grammar of Q: Q-particles, wh-movement, and pied-piping.

Oxford: Oxford University Press.

Campbell, Richard. 1996. Specificity operators in SpecDP. Studia Linguistica 50,

161–188.

Caponigro, Ivano. 2002. Free relatives as DPs with a silent D and a CP comple-
ment. In Vida Samiian (Ed.), Proceedings of the Western Conference on Linguistics
2000 (WECOL 2000). Fresno: California State University.

Caponigro, Ivano. 2003. Free not to ask: On the semantics of free relatives and
wh-words cross-linguistically. Ph.D. thesis, UCLA.

Caponigro, Ivano. 2012. Acquiring the meaning of free relative clauses and plural
definite descriptions. Journal of Semantics 29, 261–293.

223
Cardinaletti, Anna & Michal Starke. 1999. The typology of structural defi-
ciency: A case study of the three classes of pronouns. In Henk van Riemsdijk
(Ed.), Clitics in the languages of Europe, pp. 145–233. Berlin: Mouton de Gruyter.

Cariani, Fabrizio, Stefan Kaufmann, & Magdalena Kaufmann. 2013. De-

liberative modality under epistemic uncertainty. Linguistics and Philosophy 36,
225–259.

Carlson, Robert. 1994. A grammar of Supyire. Berlin: de Gruyter.

Chang, Henry Y. 2012. Nominal tense in Tsou: Nia and its syntax/semantics.
In Thomas Graf, Denis Paperno, Anna Szabolcsi, & Jos Tellings (Eds.),
Theories of everything: In honor of Ed Keenan, pp. 43–58. UCLA Working Papers
in Linguistics.

Charlow, Nate. 2010. Restricting and embedding imperatives. In Maria Aloni,

Harald Bastiaanse, Tikitu de Jager, & Katrin Schulz (Eds.), Logic, lan-
guage and meaning: Selected papers from the 17th Amsterdam Colloquium, pp.
223–233. Berlin: Springer.

Charlow, Nate. 2011. Practical language: Its meaning and use. Ph.D. thesis,
University of Michigan.

Charlow, Nate. 2013a. Conditional preferences and practical conditionals. Lin-

guistics and Philosophy 36, 463–511.

Charlow, Nate. 2013b. What we know and what to do. Synthese 190, 2291–2323.

Charlow, Simon & Yael Sharvit. 2014. Bound ‘de re’ pronouns and the LFs of
attitude reports. Semantics and Pragmatics 7, 1–43.

Cheng, Lisa L.-S. & C.-T. James Huang. 1996. Two types of donkey sentences.
Natural Language Semantics 4, 121–163.

Chierchia, Gennaro. 1992. Anaphora and dynamic binding. Linguistics and

Philosophy 15, 111–183.

Chierchia, Gennaro. 1993. Questions with quantifiers. Natural Language Seman-

tics 1, 181–234.

224
Chierchia, Gennaro. 1994. Intensionality and context change: Towards a dy-
namic theory of propositions and properties. Journal of Logic, Language and
Information 3, 141–168.

Chierchia, Gennaro. 1995. Individual-level predicates as inherent generics. In

Gregory N. Carlson & Francis Jeffrey Pelletier (Eds.), The generic book,
pp. 176–223. Chicago: Chicago University Press.

Chomsky, Noam. 1977. On wh-movement. In Peter Culicover, Tom Wasow,

& Adrian Akmajian (Eds.), Formal syntax, pp. 71–132. New York: Academic
Press.

Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.

Chomsky, Noam. 1993. A Minimalist Program for Linguistic Theory. In Hale &
Keyser (1993), pp. 1–52.

Chomsky, Noam. 1995. The minimalist program. Cambridge: MIT Press.

Cinque, Guglielmo. 1995. On the evidence for partial N-movement in the Ro-
mance DP. In Italian syntax and Universal Grammar, pp. 287–309. Cambridge:
Cambridge University Press.

Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-lingusitic per-

spective. Oxford: Oxford University Press.

Cinque, Guglielmo. 2013. Typological studies: Word order and relative clauses. New
York: Routledge.

Cinque, Guglielmo. 2015. Three phenomena discriminating between ‘raising’ and

‘matching’ relative clauses. Semantics-Syntax Interface 2, 1–27.

Citko, Barbara. 2009. What don’t wh-questions, free relatives, and correlatives
have in common? In Lipták (2009a), pp. 49–79.

Collins, Chris. 1993. Topics in Ewe syntax. Ph.D. thesis, MIT.

Collins, Chris. 1998. A note on extraction from conditionals. In Niken

Adisasmito-Smith & Toby Doeleman (Eds.), Cornell workings papers in lin-
guistics, vol. 16, pp. 66–75. Ithaca, N.Y.: Cornell University Press.

225
Collins, Chris & Edward Stabler. 2016. A formalization of minimalist syntax.
Syntax 19, 43–78.

Corver, Norbert & Jairo Nunes (Eds.). 2007. The copy theory of movement.
Philadelphia: John Benjamins.

Cresswell, M.J. 1990. Entities and indices. Dordrecht: Kluwer.

Crnič, Luka & Uli Sauerland (Eds.). 2014. The art and craft of semantics: A
festschrift for Irene Heim. Cambridge, Mass.: MITWPL.

Culy, Christopher. 1990. The syntax and semantics of internally headed relative
clauses. Ph.D. thesis, Stanford University.

Cumming, Samuel. 2008. Variabilism. Philosophical Review 117, 525–554.

Davison, Alice. 2009a. Adjunction, features and locality in Sanskrit and

Hindi/Urdu correlatives. In Lipták (2009a), pp. 223–262.

Davison, Alice. 2009b. Correlative clause features in Sanskrit and Hindi/Urdu. In

Paola Crisma & Giuseppe Longobardi (Eds.), Historical syntax and linguistic
theory, pp. 271–291. Oxford: Oxford University Press.

Dayal, Veneeta. 1996. Locality in WH quantification. Dordrecht: Kluwer Aca-

demic Publishers.

Dayal, Veneeta. 2006. Multiple-wh-questions. In Everaert & van Riemsdijk

(2006), pp. 275–326.

Dayal, Veneeta. 2016. Questions. Oxford: Oxford University Press.

Déchaine, Rose-Marie & Martina Wiltschko. 2002. Decomposing pronouns.

Linguistic Inquiry 33, 409–442.

Déchaine, Rose-Marie & Martina Wiltschko. 2015. When and why can 1st
and 2nd person pronouns be bound variables? In Patrick Grosz, Pritty Patel-
Grosz, & Igor Yanovich (Eds.), NELS 40: Semantics workshop on pronouns, pp.
1–50. Amherst: GLSA.

Déchaine, Rose-Marie & Martina Wiltschko. 2017a. Bound variable

anaphora. In Everaert & van Riemsdijk (2017).

226
Déchaine, Rose-Marie & Martina Wiltschko. 2017b. A formal typology of
reflexives. Studia Linguistica 71, 60–106.

Demirdache, Hamida. 1996. ‘The chief of the United States’ sentences in

St’át’imcets (Lillooet Salish): A cross-linguistic asymmetry in the temporal inter-
pretation of noun phrases and its implications. In International Conference on
Salishan and Neighbouring Languages 31, pp. 79–100. Vancouver: University of
British Columbia.

Diessel, Holger. 1999. Demonstratives: Form, function, and grammaticalization.

Amsterdam: John Benjamins.

den Dikken, Marcel. 2009. Comparative correlatives and successive cyclicity. In

Lipták (2009a), pp. 263–306.

Dimitrova-Vulchanova, Mila & Giuliana Giusti. 1996. Quantified noun

phrase structure in Bulgarian. In Jindrich Toman (Ed.), Formal approaches to
Slavic linguistics, pp. 123–144. Ann Arbor: Michigan Slavic Publications.

Donati, Caterina & Carlo Cecchetto. 2015. (Re)labeling. Cambridge, Mass.:

MIT Press.

Dowell, J.L. 2012. Contextualist solutions to three puzzles about practical condi-
tionals. In Russ Shafer-Landau (Ed.), Oxford studies in metaethics, vol. 7, pp.
271–303. Oxford: Oxford University Press.

Drummond, Alex, Dave Kush, & Norbert Hornstein. 2011. Minimalist con-
strual: Two approaches to A and B. In Cedric Boeckx (Ed.), The Oxford
handbook of linguistic minimalism, pp. 396–426. New York: Oxford University
Press.

Dummett, Michael. 1973. Frege: Philosophy of language. London: Duckworth,

2nd edn.

Elbourne, Paul D. 2001. E-type anaphora as NP-deletion. Natural Language

Semantics 9, 241–288.

Elbourne, Paul D. 2005. Situations and individuals. Cambridge: MIT Press.

Elbourne, Paul D. 2013. Definite descriptions. Oxford: Oxford University Press.

227
Embick, David & Rolf Noyer. 2001. Movement operations after syntax. Linguistic
Inquiry 32, 555–595.

Enç, Mürvet. 1991. The semantics of specificity. Linguistic Inquiry 22, 1–25.

Etxeberria, Urtzi. 2005. Quantification and domain restriction in Basque. Ph.D.

thesis, University of the Basque Country.

Etxeberria, Urtzi & Anastasia Giannakidou. 2010. Contextual domain restric-

tion and the definite determiner. In François Recanati, Isidora Stojanovic,
& Neftalí Villanueva (Eds.), Context-dependence, perspective and relativity, pp.
93–126. Berlin: Mouton de Gruyter.

Everaert, Martin & Henk van Riemsdijk (Eds.). 2006. The Blackwell companion
to syntax. Oxford: Blackwell.

Everaert, Martin & Henk van Riemsdijk (Eds.). 2017. The Wiley Blackwell
companion to syntax, Second edition. Blackwell.

Fabb, Nigel. 1990. The difference between English restrictive and nonrestrictive
relative clauses. Journal of Linguistics 26, 57–78.

Falco, Michelangelo & Roberto Zamparelli. 2016. The only real pro-nouns:
Comparing English one and Italian ne as Noun Phrase pro-forms. In Grosz &
Patel-Grosz (2016), pp. 107–134.

Fara, Delia Graff. 2011. You can call me ‘stupid’,… just don’t call me stupid.
Analysis 71, 492–501.

Fara, Delia Graff. 2015. Names are predicates. Philosophical Review 124, 59–117.

Felser, Claudia. 2015. Syntax and language processing. In Tibor Kiss & Artemis
Alexiadou (Eds.), Syntax: Theory and analysis, vol. 3, pp. 1875–1911. Berlin: de
Gruyter.

Fenton, Donna. 2010. Multiple functions, multiple techniques: the role of method-
ology in a study of Zapotec determiners. In Andrea L. Berez, Jean Mulder,
& Daisy Rosenblum (Eds.), Fieldwork and linguistic analysis in indigenous lan-
guages of the Americas, pp. 125–145. Honolulu: University of Hawai’i Press.

Fiengo, Robert & Robert May. 1994. Indices and identity. Cambridge, Mass.:
MIT Press.

228
Fillmore, Charles J. 1971/1997. Lectures on deixis. Stanford: CSLI Publications.

von Fintel, Kai. 1994. Restrictions on quantifier domains. Ph.D. thesis, University
of Massachusetts, Amherst.

von Fintel, Kai. 1998. The semantics and pragmatics of quantifier domain restric-
tion. Vilem Mathesius Lectures, Prague.

von Fintel, Kai & Irene Heim. 2011. Intensional semantics. MS, MIT.

von Fintel, Kai & Sabine Iatridou. 2005. What to do if you want to go to Harlem:
Anankastic conditionals and related matters. MS, MIT.

von Fintel, Kai & Sabine Iatridou. 2009. Lecture notes for LSA 220: Morphol-
ogy, syntax, and semantics of modals. MIT.

von Fintel, Kai & Lisa Matthewson. 2008. Universals in semantics. The
Linguistic Review 25, 139–201.

Fodor, Janet Dean. 1970. The linguistic description of opaque contexts. Ph.D. thesis,
MIT.

Fodor, Janet Dean & Ivan A. Sag. 1982. Referential and quantificational indefi-
nites. Linguistics and Philosophy 5, 355–398.

Fox, Danny. 2000. Economy and semantic interpretation. Cambridge, Mass.: MIT
Press.

Fox, Danny. 2002. Antecedent-contained deletion and the copy theory of move-
ment. Linguistic Inquiry 33, 63–96.

Fox, Danny. 2003. On logical form. In Randall Hendrick (Ed.), Minimalist

syntax, pp. 82–123. Oxford: Blackwell.

Fox, Danny & Jon Nissenbaum. 2004. Condition A and scope reconstruction.
Linguistic Inquiry 35, 475–485.

Franco, Ludovico. 2013. A typological rarum in Sogdian: Overt complmentizers

in indicative root clauses. Lingua Posnaniensis 55, 55–67.

Frank, Anette. 1996. Context dependence in modal constructions. Ph.D. thesis,

University of Stuttgart.

229
Fukui, Naoki & Margaret Speas. 1986. Specifiers and projection. In Naoki
Fukui, Tova R. Rapoport, & Elizabeth Sagey (Eds.), MIT working papers
in linguistics 8: Papers in theoretical linguistics, pp. 128–172. Cambridge, Mass.:
MITWPL.

Gair, James W. 1998. Studies in South Asian linguistics: Sinhala and other South
Asian languages. Oxford: Oxford University Press.

Gair, James W., Barbara Lust, K.V. Subbarao, & Kashi Wali (Eds.). 1999.
Lexical anaphors and pronouns in selected South Asian languages: A principled
typology. Berlin: de Gruyter.

Gajewski, Jon. 2008. On the semantics of Hindi-Urdu multiple correlatives. Lin-

guistic Inquiry 39, 327–334.

Geis, Michael. 1970. Adverbial subordinate clauses in English. Ph.D. thesis, MIT.

George, Benjamin. 2011. Question embedding and the semantics of answers. Ph.D.
thesis, UCLA.

Geurts, Bart. 1997. Good news about the description theory of names. Journal of
Semantics 14, 319–348.

Geurts, Bart. 2004. On an ambiguity in quantified conditionals. MS, University

of Nijmegen.

Gil, Kook-Hee, Stephen Harlow, & George Tsoulas (Eds.). 2013. Strategies of
quantification. Oxford: Oxford University Press.

Gillies, Anthony S. 2010. Iffiness. Semantics and Pragmatics 3, 1–42.

Gillon, Carrie. 2013. The semantics of determiners: Domain restriction in

Skwxwú7mesh. Newcastle upon Tyne: Cambridge Scholars Publishing.

Giusti, Giuliana. 1991. The categorial status of quantified nominals. Linguistiche

Berichte 136, 438–452.

Giusti, Giuliana. 1992. Heads and modifiers among determiners: Evidence from
Romanian and German. In University of Venice Working Papers in Linguistics, vol.
1.3, pp. 1–19. University of Venice.

230
Giusti, Giuliana. 1997. The categorial status of determiners. In Liliane Haege-
man (Ed.), The new comparative syntax, pp. 95–123. London: Longman.

Giusti, Giuliana. 2002. The functional structure of noun phrases: A bare phrase
structure approach. In Guglielmo Cinque (Ed.), Functional structure in DP and
IP: The cartography of syntactic structures, Vol. 1, pp. 54–90. New York: Oxford
University Press.

Giusti, Giuliana & Nedžad Leko. 1995. On the syntax of quantity expressions
in Bosnian. In Laura Brugè (Ed.), University of Venice Working Papers in
Linguistics, vol. 5.2, pp. 23–47. University of Venice.

Grimshaw, Jane. 1997. Projection, heads, and optimality. Linguistic Inquiry 28,
373–422.

Grimshaw, Jane. 2005. Words and structure. Stanford: CSLI.

Groat, Erich & John O’neil. 1996. Spell-Out at the LF Interface. In Werner
Abraham, Samuel David Epstein, Höskuldur Thráinsson, & Jan-Wouter
Zwart (Eds.), Minimal ideas: Syntactic studies in the minimalist framework, pp.
113–139. Philadelphia: John Benjamins.

Groenendijk, Jeroen & Floris Roelofsen. 2009. Inquisitive semantics and

pragmatics. MS, ILLC.

Groenendijk, Jeroen & Floris Roelofsen. 2010. Radical inquisitive semantics.

MS, ILLC.

Groenendijk, Jeroen & Martin Stockhof. 1982. Semantic analysis of wh-

complements. Linguistics and Philosophy 5, 175–233.

Groenendijk, Jeroen & Martin Stockhof. 1984. Studies on the semantics of

questions and the pragmatics of answers. Ph.D. thesis, University of Amsterdam.

Groenendijk, Jeroen & Martin Stockhof. 1991. Dynamic predicate logic.

Linguistics and Philosophy 14, 39–100.

Groenendijk, Jeroen & Martin Stokhof. 1990. Dynamic Montague Grammar.

In L. Kalman & L. Polos (Eds.), Proceedings of the Second Symposium on Logic
and Language, pp. 3–48. Budapest: Adakemiai Kiado.

231
Groos, Aneeke & Henk van Riemsdijk. 1981. Matching effects in free relatives: a
parameter of matching effects in free relatives: A parameter of core grammar. In
Adriana Belletti, Luciana Brandi, & Luigi Rizzi (Eds.), Theory of marked-
ness in generative grammar, pp. 171–216. Pisa: Scuola Normale Superiore di Pisa.

Grosu, Alexander. 1996. On the proper analysis of ‘missing P’ free relative

constructions. Linguistic Inquiry 27, 41–58.

Grosu, Alexander & Fred Landman. 1998. Strange relatives of the third kind.
Natural Language Semantics 6, 125–170.

Grosz, Patrick & Pritty Patel-Grosz (Eds.). 2016. The impact of pronominal
form on interpretation. Berlin: de Gruyter.

Guéron, Jacqueline & Teun Hoekstra. 1995. The temporal interpretation of

predication. In Anna Cardinaletti & Maria-Teresa Guasti (Eds.), Small
clauses, pp. 77–107. San Diego: Academic Press.

Hacquard, Valentine. 2006. Aspects of modality. Ph.D. thesis, MIT.

Hacquard, Valentine. 2010. On the event relativity of modal auxiliaries. Natural

Language Semantics 18, 79–114.

Haegeman, Liliane (Ed.). 1997. Elements of grammar: Handbook of generative

syntax. Dordrecht: Kluwer.

Haegeman, Liliane. 2010. The movement derivation of conditional clauses. Lin-

guistic Inquiry 41, 595–621.

Haïk, Isabelle. 1984. Indirect binding. Linguistic Inquiry 15, 185–223.

Haiman, John. 1978. Conditionals are topics. Language 54, 565–589.

Hale, Kenneth & Samuel Jay Keyser (Eds.). 1993. The view from Building 20:
Essays in linguistics in honor of Sylvain Bromberger. Cambridge, Mass.: MIT Press.

Hamblin, C.L. 1973. Questions in Montague English. Foundations of language 10,

41–53.

Hanitramalala, Rita & Paul Ileana. 2012. Malagasy quantifiers. In Keenan &
Paperno (2012), pp. 613–645.

232
Hankamer, Jorge & Line Mikkelsen. 2012. CP complements to D. MS, UC Santa
Cruz & UC Berkeley.

Haspelmath, Martin. 1995. Diachronic sources of ‘all’ and ‘every’. In Bach et al.
(1995), pp. 363–382.

Heim, Irene. 1982. The semantics of definite and indefinite noun phrases. Ph.D.
thesis, University of Massachusetts Amherst.

Heim, Irene. 1990. E-type pronouns and donkey anaphora. Linguistics and Philos-
ophy 13, 137–177.

Heim, Irene. 1991. Articles and definiteness. In Arnim von Stechow & Dieter
Wunderlich (Eds.), Semantics: An international handbook of contemporary re-
search, pp. 487–535. Berlin: de Gruyter.

Heim, Irene. 1994. Interrogative semantics and Karttunen’s semantics for know.
In Rhonna Buchalla & Anita Mitwoch (Eds.), IATL 1, pp. 128–144. Hebrew
University of Jerusalem.

Heim, Irene. 2001. Features of pronouns in semantics and morphology. MS.

Heim, Irene. 2008. Features on bound pronouns. In Daniel Harbour, David

Adger, & Susana Béjar (Eds.), Phi theory: Phi-features across modules and
interfaces, pp. 35–56. Oxford: Oxford University Press.

Heim, Irene. 2012. Lecture notes for 24.973: Advanced semantics. MIT.

Heim, Irene & Angelika Kratzer. 1998. Semantics in Generative Grammar.

Oxford: Wiley-Blackwell.

Heine, Bernd. 1997. Possession: Cognitive sources, forces, and grammaticalization.

Cambridge: Cambridge University Press.

Hendriks, Herman L.W. 1993. Studied flexibility: Categories and types in syntax
and semantics. Amsterdam: ILLC.

Hestvik, Arild. 1992. LF movement of pronouns and antisubject orientation.

Linguistic Inquiry 23, 557–594.

Heycock, Caroline. 1995. Asymmetries in reconstruction. Linguistic Inquiry 26,

547–570.

233
Higginbotham, James. 1983. Logical form, binding, and nominals. Linguistic
Inquiry 14, 395–420.

Higginbotham, James. 1985. On semantics. Linguistic Inquiry 16, 547–593.

Higginbotham, James & Robert May. 1981. Questions, quantifiers and crossing.
The Linguistic Review 1, 41–80.

Hiraiwa, Ken. 2005. Dimensions of symmetry in syntax: Agreement and clausal

architecture. Ph.D. thesis, MIT.

Hiraiwa, Ken. 2017. Internally headed relative clauses. In Everaert & van
Riemsdijk (2017).

Hiraiwa, Ken, George Akanlig-Pare, Samuel Atintono, Adams Bodomo,

Komlan Essizewa, & Fusheini Hudu. 2017. A comparative syntax of internally-
headed relative clauses in Gur. Glossa 2, 1–30.

Holmberg, Anders. 1993. On the structure of predicate NP. Studia Linguistica 47,
126–138.

Hornstein, Norbert. 1995. Logical form: From GB to Minimalism. Oxford:

Blackwell.

Hornstein, Norbert. 2001. Move! A minimalist theory of construal. Oxford:

Blackwell.

Hornstein, Norbert. 2007. Pronouns in a minimalist setting. In Corver & Nunes

(2007), pp. 351–385.

Iatridou, Sabine. 1991. Topics in conditionals. Ph.D. thesis, MIT, Cambridge.

Iatridou, Sabine. 1994. On the contribution of conditional then. Natural Language

Semantics 2, 171–199.

Iatridou, Sabine. 2000. The grammatical ingredients of counterfactuality. Lin-

guistic Inquiry 31, 231–270.

Iatridou, Sabine. 2013. Looking for free relatives in Turkish (and the unexpected
places this leads to). In Umut Özge (Ed.), Proceedings of the 8th Workshop on
Altaic Formal Linguistics (WAFL8), pp. 129–152. Cambridge, Mass.: MITWPL.

234
Iatridou, Sabine, Elena Anagnostopoulou, & Roumyana Izvorski. 2001. Ob-
servations about the form and meaning of the perfect. In Kenstowicz (2001),
pp. 189–238.

Iatridou, Sabine & Anthony Kroch. 1993. The licensing of CP-recursion and
its relevance to the Germanic verb-second phenomenon. Working papers in
Scandinavian syntax 50, 1–24.

Ihsane, Tabea & Genoveva Puskás. 2001. Specific is not definite. In Ur Shlonsky
& Tabea Ihsane (Eds.), Generative grammar in Geneva 2, pp. 39–54. University
of Geneva.

Ilkhanipour, Negin. 2016. Tense and modality in the nominal domain. Linguistica
56, 143–160.

Isaacs, James & Kyle Rawlins. 2008. Conditional questions. Journal of Semantics
25, 269–319.

Izvorski, Roumyana. 1996. The syntax and semantics of correlative proforms. In

Kiyomi Kusumoto (Ed.), Proceedings of NELS 26, pp. 133–147. Amherst: GLSA
Publications.

Jacobson, Pauline. 1995. On the quantificational force of English free relatives. In

Bach et al. (1995), pp. 451–486.

Jacobson, Pauline. 1999. Towards a variable-free semantics. Linguistics and

Philosophy 22, 117–184.

Jacobson, Pauline. 2000. Paycheck pronouns, Bach-Peters sentences, and variable-

free semantics. Natural Language Semantics 8, 77–155.

Janssen, Theo M.V. 1997. Compositionality. In Johan van Bentham & Alice
ter Meulen (Eds.), Handbook of logic and linguistics, pp. 417–473. Amsterdam:
Elsevier.

Jenks, Peter. 2018. Articulated definiteness without articles. Linguistic Inquiry 49,
501–536.

Jensen, Per Anker & Carl Vikner. 1994. Lexical knowledge and the semantic
analysis of Danish genitive constructions. In Steffen L. Hansen & Helle We-
gener (Eds.), Topics in knowledge-based NLP systems, pp. 37–55. Copenhagen:
Samfundslitteratur.

235
Jespersen, Otto. 1924. The philosophy of grammar. London: George Allen &
Unwin.

Johnson, Kyle. 2012. Toward deriving differences in how wh movement and QR

are pronounced. Lingua 122, 529–553.

Kamp, Hans. 1971. Formal properties of ‘now’. Theoria 37, 227–274.

Kamp, Hans. 1981. A theory of truth and semantic representation. In Jeroen

Groenendijk, T. Janssen, & M. Stokhof (Eds.), Formal methods in the study of
language, pp. 277–322. Amsterdam: Mathematical Centre.

Kanazawa, Makoto. 1994. Weak vs. strong readings of donkey sentences in a

dynamic setting. Linguistics and Philosophy 17, 109–158.

Kandybowicz, Jason. 2007. On fusion and multiple copy spell-out: The case of
verbal repetition. In Corver & Nunes (2007), pp. 119–150.

Kandybowicz, Jason. 2008. The grammar of repetition: Nupe grammar at the

syntax-phonology interface. Philadelphia: John Benjamins.

Kaplan, David. 1989. Demonstratives. In Joseph Almog, John Perry, & Howard
Wettstein (Eds.), Themes from Kaplan, pp. 481–563. Oxford: Oxford University
Press.

Karttunen, Lauri. 1977. Syntax and semantics of questions. Linguistics and

Philosophy 1, 3–44.

Kayne, Richard. 1991. Romance clitics, verb movement, and PRO. Linguistic
Inquiry 22, 647–686.

Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge: MIT Press.

Kayne, Richard S. 2002. Pronouns and their antecedents. In Samuel David

Epstein & T. Daniel Seely (Eds.), Derivation and explanation in the Minimalist
Program, pp. 133–166. Oxford: Blackwell.

Keenan, Edward L. 2008. Quantification in Malagasy. In Matthewson (2008),

pp. 319–352.

Keenan, Edward L. & Denis Paperno (Eds.). 2012. Handbook of quantifiers in

natural language. Dordrecht: Springer.

236
Kennedy, Christopher. 2007. Vagueness and grammar: The semantics of relative
and absolute gradable adjectives. Linguistics and Philosophy 30, 1–45.

Kennedy, Christopher. 2014. Predicates and formulas: Evidence from ellipsis. In

Crnič & Sauerland (2014), pp. 253–277.

Kenstowicz, Michael (Ed.). 2001. Ken Hale: A life in language. Cambridge, Mass.:
MIT Press.

King, Jeffrey C. 1994. Anaphora and operators. Philosophical Perspectives 8, 221–

250.

King, Jeffrey C. 2004. Context dependent quantifiers and donkey anaphora.

Canadian Journal of Philosophy 34 (supplement 1), 97–127.

Kobele, Greg. 2010. Inverse linking via function composition. Natural Language
Semantics 18, 183–196.

Kolodny, Niko & John MacFarlane. 2010. Ifs and oughts. Journal of Philosophy
pp. 115–143.

van Koppen, Marjo. 2007. Agreement with (the internal structure of) copies of
movement. In Corver & Nunes (2007), pp. 327–350.

Kotek, Hadas. 2014. Composing questions. Ph.D. thesis, MIT.

Kratzer, Angelika. 1977. What ‘must’ and ‘can’ must and can mean. Linguistics
and Philosophy 1, 337–355.

Kratzer, Angelika. 1981. The notional category of modality. In Hans-Jürgen

Eikmeyer & Hannes Rieser (Eds.), Words, worlds, and contexts: New approaches
in word semantics, pp. 38–74. Berlin: de Gruyter.

Kratzer, Angelika. 1991. Modality/Conditionals. In Arnim von Stechow &

Dieter Wunderlich (Eds.), Semantics: An international handbook of contem-
porary research, pp. 639–656. New York: de Gruyter.

Kratzer, Angelika. 1995. Stage-level and individual-level predicates. In Gre-

gory N. Carlson & Francis Jeffrey Pelletier (Eds.), The generic book, pp.
125–175. Chicago: University of Chicago Press.

237
Kratzer, Angelika. 1998a. More structural analogies between pronouns and
tenses. In Devon Strolovitch & Aaron Lawson (Eds.), Proceedings of Seman-
tics and Linguistic Theory (SALT) 8, pp. 92–110. Ithaca, N.Y.: CLC Publications.

Kratzer, Angelika. 1998b. Scope or pseudoscope? Are there wide-scope indef-

inites. In S. Rothstein (Ed.), Events and grammar, pp. 163–196. Dordrecht:
Kluwer Academic Publishers.

Kratzer, Angelika. 2009. Making a pronoun: Fake indexicals as windows into the
properties of pronouns. Linguistic Inquiry 40, 187–237.

Kratzer, Angelika. 2013. Modality for the 21st century. In Stephen R. Ander-
son, Jacques Moeschler, & Fabienne Reboul (Eds.), The language-cognition
interface: Proceedings of the 19th International Congress of Linguists, pp. 179–199.
Geneva: Librairie Droz.

Kratzer, Angelika & Junko Shimoyama. 2002. Indeterminate pronouns: The

view from Japanese. In Y. Otsu (Ed.), Proceedings of the Third Tokyo Conference
on Psycholinguistics. Tokyo: Hituzi Syobo.

Kripke, Saul. 1979. A puzzle about belief. In Avishai Margalit (Ed.), Meaning
and use, pp. 239–283. Dordrecht: Reidel.

Kripke, Saul. 1980. Naming and necessity. Cambridge, Mass.: Harvard University
Press.

Kusumoto, Kiyomi. 2005. On the quantification over times in natural language.

Natural Language Semantics 13, 317–357.

Laenzlinger, Christopher. 2005. French adjective ordering: Perspectives on DP

internal movement types. Lingua 115, 645–689.

Larson, Brooke. 2016. The representation of syntactic action at a distance: Mul-

tidominance versus the copy theory. Glossa 1, 1–18.

Larson, Richard. 1985. On the syntax of disjunction scope. Natural Language and
Linguistic Theory 3, 217–264.

Lasersohn, Peter. 1996. Adnominal conditionals. In Teresa Galloway &

Justin Spence (Eds.), Proceedings of SALT 6, pp. 154–166. Ithaca, N.Y.: CLC
Publications.

238
Lasnik, Howard. 1989. Essays on anaphora. Boston: Kluwer.

Lebeaux, David. 1983. A distributional difference between reciprocals and reflex-

ives. Linguistic Inquiry 14, 723–730.

Lecarme, Jacqueline. 1996. Tense in the nominal system: The Somali DP. In
Jacqueline Lecarme, Jean Lowenstamm, & Ur Shlonsky (Eds.), Studies in
Afroasiatic grammar, pp. 159–178. The Hague: Holland Academic Graphics.

Lecarme, Jacqueline. 1999a. Focus in Somali. In Georges Rebuschi & Laurie

Tuller (Eds.), The grammar of focus, pp. 275–309. Amsterdam: John Benjamins.

Lecarme, Jacqueline. 1999b. Nominal tense and tense theory. In Francis Cor-
blin, Carmen Dobrovie-Sorin, & Jean-Marie Marandin (Eds.), Empirical
issues in formal syntax and semantics 2, pp. 333–354. The Hague: Thesus.

Lecarme, Jacqueline. 2004. Tense in nominals. In Jacqueline Guéron &

Jacqueline Lecarme (Eds.), The syntax of time, pp. 441–475. Cambridge, Mass.:
MIT Press.

Lecarme, Jacqueline. 2008. Tense and modality in nominals. In Jacque-

line Guéron & Jacqueline Lecarme (Eds.), Time and modality, pp. 195–226.
Springer.

Lecarme, Jacqueline. 2012. Nominal tense. In Robert I. Binnick (Ed.), The

Oxford handbook of tense and aspect, pp. 696–718. Oxford: Oxford University
Press.

Lechner, Winfried. 2006. An interpretive effect of head movement. In Mara

Frascarelli (Ed.), Phases of interpretation, pp. 45–71. Berlin: Mouton de
Gruyter.

Lee, Felicia. 2003. Anaphoric R-expressions as bound variables. Syntax 6, 84–114.

Leung, Tommi Tsz-Cheung. 2009. On the matching requirement in correlatives.

In Lipták (2009a), pp. 309–341.

Lewis, David. 1975. Adverbs of quantification. In Edward L. Keenan (Ed.), Formal

semantics of natural language, pp. 3–15. Cambridge: Cambridge University Press.

Lewis, David. 1979. Attitudes de dicto and de se. Philosophical Review 88, 513–543.

239
Lewis, David. 1980. Index, context, and content. In Stig Kanger & Helle Ohman
(Eds.), Philosophy and grammar, pp. 79–100. Holland: D. Reidel.

Lewis, David. 1986. On the Plurality of Worlds. Oxford: Blackwell.

Li, Yen-Hui Audrey. 1992. Indefinite wh in Chinese. Journal of East Asian

Linguistics 1, 125–155.

Lidz, Jeffrey & William J. Idsardi. 1998. Chains and phono-logical form. Uni-
versity of Pennsylvania Working Papers in Linguistics 5, 109–125.

Link, Godehard. 1983. The logical analysis of plurals and mass terms: A lattice-
theoretical approach. In Paul Portner & Barbara H. Partee (Eds.), Formal
semantics: The essential readings, pp. 127–146. Malden: Blackwell.

Lipták, Anikó (Ed.). 2009a. Correlatives cross-linguistically. Amsterdam: John

Benjamins.

Lipták, Anikó. 2009b. The landscape of correlatives: An empirical and analytical

survey. In Lipták (2009a), pp. 1–46.

Lipták, Anikó. 2012. Correlative topicalization. Acta Linguistica Hungarica 59,

245–302.

Lipták, Anikó & Georges Rebuschi. 2009. Types and distribution of Northern
Basque correlatives. In Nguyen Chi Duy Khuong Richa & Samar Sinha
(Eds.), The 5th Asian GLOW: Conference proceedings, pp. 194–218. New Delhi:
FOSSSIL.

Longobardi, Giuseppe. 1994. Reference and proper names: A theory of N-

movement in syntax and logical form. Linguistic Inquiry 25, 609–665.

Longobardi, Giuseppe. 2001. The structure of DPs: Some principles, parameters,

and problems. In Baltin & Collins (2001), pp. 562–604.

Luján, Marta. 2004. Determiners as pronouns. In Ana Castro, Marcelo Fer-

reira, Valentine Hacquard, & Andres Salanova (Eds.), Romance, Op. 47:
Collected papers on Romance syntax, pp. 129–148. Cambridge, Mass.: MITWPL.

Lycan, William. 2001. Real conditionals. New York: Oxford University Press.

Lyons, Christopher. 1999. Definiteness. Cambridge: Cambridge University Press.

240
MacFarlane, John. 2009. Nonindexical contextualism. Synthese 166, 231–250.

MacFarlane, John. 2014. Assessment sensitivity: Relative truth and its applications.
Oxford: Clarendon Press.

Martí, Luisa. 2002. Contextual variables. Ph.D. thesis, University of Connecticut.

Matthewson, Lisa. 1998. Determiner systems and quantificational strategies: Evi-

dence from Salish. The Hague: Holland Academic Graphics.

Matthewson, Lisa. 1999. On the interpretation of wide-scope indefinites. Natural

Language Semantics 7, 79–134.

Matthewson, Lisa. 2001. Quantification and the nature of crosslinguistic varia-

tion. Natural Language Semantics 9, 145–189.

Matthewson, Lisa (Ed.). 2008. Quantification: A cross-linguistic perspective. Bin-

gley: Emerald.

Matthewson, Lisa. 2013. Strategies of quantification in St’át’imcets and the rest of

the world. In Gil et al. (2013), pp. 15–38.

May, Robert. 1977. The grammar of quantification. Ph.D. thesis, MIT.

May, Robert. 1985. Logical form. Cambridge: MIT Press.

McNeill, David. 1970. The acquisition of language. New York: Harper & Row.

Montague, Richard. 1970. English as a formal language. In Richmond H.

Thomason (Ed.), Formal philosophy, pp. 188–221. New Haven: Yale University
Press.

Mortensen, David. 2003. Two kinds of variable elements in Hmong anaphora.

MS, University of California–Berkley.

Moulton, Keir. 2015. CPs: Copies and compositionality. Linguistic Inquiry 46,
305–342.

Nasu, Norio. 2010. Conditions on copy realization. English Linguistics 27, 126–158.

Ninan, Dilip. 2012. Counterfactual attitudes and multi-centered worlds. Semantics

and Pragmatics 5, 1–57.

241
Noguchi, Tohru. 1997. Two types of pronouns and variable binding. Language 73,
770–797.

Nordlinger, Rachel & Louisa Sadler. 2004. Nominal tense in crosslinguistic

perspective. Language 80, 776–806.

Nunes, Jairo. 2004. Linearization of chains and sideward movement. Cambridge,

Mass.: MIT Press.

Nyvad, Anne Mette, Ken Ramshøj Christensen, & Sten Vikner. 2017. CP-
recursion in Danish: A cP/CP-analysis. The Linguistic Review 34, 449–477.

Özyıldız, Deniz & Rodica Ivan. 2017. The Somali microscope: Personal pro-
nouns, determiners and possession. In Vera Hohaus & Wanda Rothe (Eds.),
Fieldwork perspectives on the semantics of African, Asian and Austronesian lan-
guages 3, pp. 56–69. Universität Tübingen.

Palmer, F.R. 2001. Mood and modality. Cambridge: Cambridge University Press,
2nd edn.

Partee, Barbara. 1975. Montague grammar and transformational grammar.

Linguistic Inquiry 6, 203–300.

Partee, Barbara H. 1973. Some structural analogies between tenses and pronouns
in English. In Partee (2004), pp. 50–58.

Partee, Barbara H. 1978. Bound variables and other anaphors. In Partee (2004),
pp. 110–121.

Partee, Barbara H. 1979. Semantics — Mathematics or psychology? In Rainer

Bäuerle, Urs Egli, & Arnim von Stechow (Eds.), Semantics from different
points of view, pp. 1–14. Berlin: Springer.

Partee, Barbara H. 1983. Genitives: A case study. In Partee (2004), pp. 182–189.

Partee, Barbara H. 1984. Nominal and temporal anaphora. Linguistics and

Philosophy 7, 243–286.

Partee, Barbara H. 1986. Noun phrase interpretation and type-shifting principles.

In Partee (2004), pp. 203–230.

242
Partee, Barbara H. 1989. Binding implicit variables in quantified contexts. In
Partee (2004), pp. 259–271.

Partee, Barbara H. 1995a. Lexical semantics and compositionality. In Lila R.

Gleitman & Mark Liberman (Eds.), Invitation to cognitive science, vol. 1, pp.
311–360. Cambridge: MIT Press.

Partee, Barbara H. 1995b. Quantificational structures and compositionality. In

Bach et al. (1995), pp. 541–602.

Partee, Barbara H. 2004. Compositionality in formal semantics: Selected papers by

Barbara H. Partee. Malden: Blackwell.

Partee, Barbara H. & Vladimir Borschev. 2003. Genitives, relational nouns,

and argument-modifier ambiguity. In Ewald Lang, Claudia Maienborn, &
Cathrine Fabricius-Hansen (Eds.), Modifying adjuncts, pp. 67–112. Berlin:
Mouton de Gruyter.

Partee, Barbara H. & Mats Rooth. 1983. Generalized conjunction and type
ambiguity. In Rainer Bäuerle, Christoph Schwarze, & Arnim von Stechow
(Eds.), Meaning, use, and interpretation of language, pp. 361–383. Berlin: de
Gruyter.

Patel-Grosz, Pritty & Patrick Grosz. 2010. On the typology of donkeys:

Two types of anaphora resolution. In Martin Prinzhorn, Viola Schmitt, &
Sarah Zobel (Eds.), Proceedings of Sinn und Bedeutung 14, pp. 339–355. Vienna:
University of Vienna.

Patel-Grosz, Pritty & Patrick G. Grosz. 2017. Revisiting pronominal typology.

Linguistic Inquiry 48, 259–297.

Percus, Orin. 2000. Constraints on some other variables in syntax. Natural

Language Semantics 8, 173–229.

Percus, Orin & Uli Sauerland. 2003. On the LFs of attitude reports. In Matthias
Weisgerber (Ed.), Proceedings of Sinn und Bedeutung 7, pp. 228–242. Konstanz:
Universität Konstanz.

Pesetsky, David. 1978. Category switching and so-called so-called pronouns. In

Donka Farkas, Wesley M. Jacobsen, & Karol W. Todrys (Eds.), Proceedings
of the Chicago Linguistics Society 14, pp. 350–361. Chicago: CLS.

243
Pesetsky, David. 1987. Wh-in-situ: Movement and unselective binding. In Reu-
land & ter Meulen (1987), pp. 98–129.

Pesetsky, David. 1998. Some optimality principles of sentence pronunciation. In

Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, & David
Pesetsky (Eds.), Optimality and competition in syntax, pp. 337–383. Cambridge,
Mass.: MIT Press.

Pesetsky, David. 2000. Phrasal movement and its kin. Cambridge: MIT Press.

Pesetsky, David & Esther Torrego. 2001. T-to-C movement: Causes and conse-
quences. In Kenstowicz (2001), pp. 355–426.

Peterson, Thomas H. 1974. On definite restrictive relatives in Mooré. Journal of

West African Linguistics 9, 71–78.

Postal, Paul. 1969. On the so-called ‘pronouns’ in English. In David A. Reibel

& Sanford A. Schane (Eds.), Modern studies in English: Readings in transforma-
tional grammar, pp. 201–224. Englewood Cliffs, N.J.: Prentice-Hall.

Potts, Christopher, Ash Asudeh, Seth Cable, Yurie Hara, Eric Mc-
Cready, Luis Alonso-Ovalle, Rajesh Bhatt, Christopher Davis, Ange-
lika Kratzer, Tom Roeper, & Martin Walkow. 2009. Expressives and identity
conditions. Linguistic Inquiry 40, 356–366.

Quine, Willard Van Orman. 1960. Word and object. Cambridge: MIT Press.

Rabern, Brian. 2012. Monsters in Kaplan’s logic of demonstratives. Philosophical

Studies pp. 1–12.

Radford, Andrew. 1997. Syntactic theory and the structure of English: A minimalist
approach. Cambridge: Cambridge University Press.

Rappaport, Gilbert C. 1986. On anaphor binding in Russian. Natural Language

and Linguistic Theory 4, 97–120.

Rawlins, Kyle. 2008. (Un)conditionals: An investigation in the syntax and semantics

of conditional structures. Ph.D. thesis, University of California, Santa Cruz.

Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics.

Philosophica 27, 53–94.

244
Reinhart, Tanya. 1983a. Anaphora and semantic interpretation. Chicago: Univer-
sity of Chicago Press.

Reinhart, Tanya. 1983b. Coreference and bound anaphora: A restatement of

anaphora questions. Linguistics and Philosophy 6, 47–88.

Reinhart, Tanya. 1987. Specifier and operator binding. In Reuland & ter
Meulen (1987), pp. 130–167.

Reinhart, Tanya. 1997. Quantifier scope: How labor is divided between QR and
choice functions. Linguistics and Philosophy 20, 335–397.

Reinhart, Tanya. 2006. Interface strategies: Optimal and costly computations.

Cambridge, Mass.: MIT Press.

Reinhart, Tanya & Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24, 657–
720.

Reuland, Eric. 2001. Primitives of binding. Linguistic Inquiry 32, 439–492.

Reuland, Eric J. & Alice G.B. ter Meulen (Eds.). 1987. The representation of
(in)definiteness. Cambridge, Mass.: MIT Press.

van Riemsdijk, Henk. 2006. Free relatives: A syntactic case study. In Everaert &
van Riemsdijk (2006), pp. 338–382.

Rijkhoff, Jan. 2008. Synchronic and diachronic evidence for parallels between
noun phrases and sentences. In Folke Josephson & Ingmar Söhrman (Eds.),
Interdependence of diachronic and synchronic analyses, pp. 13–42. Amsterdam:
John Benjamins.

Rizzi, Luigi. 1997. The fine structure of the left periphery. In Haegeman (1997),
pp. 281–337.

Romero, Maribel. 2014. ‘fake tense’ in counterfactuals: A temporal remoteness

approach. In Crnič & Sauerland (2014), pp. 47–63.

Rouveret, Alain (Ed.). 2011. Resumptive pronouns at the interfaces. Philadelphia:

John Benjamins.

245
Rullmann, Hotze. 2003. Bound-variable pronouns and the semantics of number.
In Brian Agbayani, Paivi Koskinen, & Vida Samiian (Eds.), Proceedings of the
Western Conference on Linguistics, WECOL 2002, pp. 243–254. Department of
Linguistics, California State University, Fresno.

Rullmann, Hotze. 2004. First and second person pronouns as bound variables.
Linguistic Inquiry 35, 159–168.

Rullmann, Hotze & Sigrid Beck. 1998. Presupposition projection and the
interpretation of which-questions. In Devon Strolovitch & Aaron Lawson
(Eds.), Semantics and Linguistic Theory (SALT) 8, pp. 215–232. Ithaca, N.Y.: CLC
Publications.

Safir, Ken. 1984. Multiple variable binding. Linguistic Inquiry 15, 603–638.

Safir, Ken. 1999. Vehicle change and reconstruction in a-bar chains. Linguistic
Inquiry 30, 587–620.

Safir, Ken. 2004. The syntax of (in)dependence. Cambridge, Mass.: MIT Press.

Safir, Ken. 2017. Weak crossover. In Everaert & van Riemsdijk (2017).

Salmon, Nathan. 1986. Frege’s puzzle. Cambridge: MIT Press.

Santorio, Paolo. 2010. Modals are monsters: On indexical shift in English. In

Nan Li & David Lutz (Eds.), Proceedings of SALT 20, pp. 289–308. Ithaca, N.Y.:
CLC Publications.

Santorio, Paolo. 2012. Reference and monstrosity. Philosophical Review 121, 359–
406.

Santorio, Paolo. 2013. Descriptions as variables. Philosophical Studies 164, 41–59.

Sauerland, Uli. 1998. The meaning of chains. Ph.D. thesis, MIT.

Sauerland, Uli. 2003. Unpronounced heads in relative clauses. In Kerstin

Schwabe & Susanne Winkler (Eds.), The interfaces: Deriving and interpreting
omitted structures, pp. 205–226. Amsterdam: John Benjamins.

Sauerland, Uli. 2004. The interpretation of traces. Natural Language Semantics

12, 63–127.

246
Saul, Jennifer M. 1998. The pragmatics of attitude ascription. Philosophical Studies
92, 363–389.

Schein, Barry. 2003. Adverbial, descriptive reciprocals. Philosophical Perspectives

17, 333–367.

Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy 26,
29–120.

Schlenker, Philippe. 2004. Conditionals as definite descriptions (A referential

analysis). Research on Language and Computation 2, 417–462.

Schlenker, Philippe. 2005. Minimize Restrictors! (Notes on definite descriptions,

Condition C and epithets). In Emar Maier, Corien Bary, & Janneke Huitink
(Eds.), Proceedings of Sinn und Bedeutung 9, pp. 385–416. Nijmegen: Nijmegen
Center for Semantics.

Schlenker, Philippe. 2006. Ontological symmetry in language: A brief manifesto.

Mind and Language 21, 504–539.

Schlenker, Philippe. 2009. Local contexts. Semantics and Pragmatics 2, 1–78.

Schlenker, Philippe. 2010. Local contexts and local meanings. Philosophical

Studies 151, 115–142.

Schwabe, Kerstin. 2016. Sentential proforms and argument conditionals. In

Werner Frey, André Meinunger, & Kerstin Schwabe (Eds.), Inner-sentenial
propositional proforms: Syntactic properties and interpretive effects, pp. 211–240.
Amsterdam: John Benjamins.

Schwarz, Anne. 2016. All-in-one and one-for-all: Thetic structures in Buli gram-
mar and discourse. In Doris L. Payne, Sara Pacchiarotti, & Mokaya Bosire
(Eds.), Diversity in African languages, pp. 81–100. Berlin: Language Science Press.

Schwarz, Florian. 2009. Two types of definites in natural language. Ph.D. thesis,
University of Massachusetts, Amherst.

Schwarzschild, Roger. 1996. Pluralities. Dordrecht: Kluwer.

Sells, Peter. 1987. Binding resumptive pronouns. Linguistics and Philosophy 10,
261–298.

247
Shimoyama, Junko. 2001. Wh-Constructions in Japanese. Ph.D. thesis, University
of Massachusetts Amherst.

Showalter, Catherine. 1986. Pronouns in Lyele. In Ursula Wiesemann (Ed.),

Pronominal systems, pp. 205–216. Tübingen: Günter Narr.

Silk, Alex. 2013. Deontic conditionals: Weak and strong. In Proceedings of the 19th
International Congress of Linguists. Département de Linguistique de l’Université
de Genève. URL http://goo.gl/xSGbcJ.

Silk, Alex. 2014a. Accommodation and negotiation with context-sensitive expres-

sions. Thought 3, 115–123.

Silk, Alex. 2014b. Evidence sensitivity in weak necessity deontic modals. Journal
of Philosophical Logic 43, 691–723.

Silk, Alex. 2016. Discourse Contextualism: A framework for contextualist semantics

and pragmatics. Oxford: Oxford University Press.

Silk, Alex. 2017. How to embed an epistemic modal: Attitude problems and other
defects of character. Philosophical Studies 174, 1773–1799.

Silk, Alex. 2018. Commitment and states of mind with mood and modality.
Natural Language Semantics pp. 1–42.

Siloni, Tal. 1997. Noun phrases and nominalizations: The syntax of DPs. Dordrecht:
Kluwer.

Soames, Scott. 2002. The unfinished semantic agenda of Naming and Necessity.
Oxford: Oxford University Press.

Speas, Margaret & Carol Tenny. 2003. Configurational properties of point of

view roles. In Anna Maria Di Scuillo (Ed.), Asymmetry in grammar, pp. 315–
344. Amsterdam: John Benjamins.

Sportiche, Dominique. 2006. Reconstruction, binding, and scope. In Everaert

& van Riemsdijk (2006), pp. 35–93.

Srivastav, Veneeta. 1991. The syntax and semantics of correlatives. Natural

Language and Linguistic Theory 9, 637–686.

Stalnaker, Robert. 1970. Pragmatics. In Stalnaker (1999), pp. 31–46.

248
Stalnaker, Robert. 1978. Assertion. In Stalnaker (1999), pp. 78–95.
Stalnaker, Robert. 1988. Belief attribution and context. In Stalnaker (1999),
pp. 150–166.
Stalnaker, Robert. 1999. Context and content: Essays on intentionality in speech
and thought. Oxford: Oxford University Press.
Stalnaker, Robert. 2014. Context. New York: Oxford University Press.
Stanley, Jason. 1997. Names and rigid designation. In Bob Hale & Crispin
Wright (Eds.), A companion to the philosophy of language, pp. 555–585. Oxford:
Blackwell.
Stanley, Jason. 2002. Nominal restriction. In Gerhard Preyer & Georg Peter
(Eds.), Logical form and language, pp. 365–388. Oxford: Oxford University Press.
Stanley, Jason. 2005. Knowledge and practical interests. Oxford: Oxford University
Press.
Stanley, Jason & Zoltán Gendler Szabó. 2000. On quantifier domain restric-
tion. Mind and Language 15, 219–261.
Starr, William. 2010. Conditionals, meaning, and mood. Ph.d. thesis, Rutgers
University.
von Stechow, Arnim. 2003. Feature deletion under semantic binding: Tense, per-
son, and mood under verbal quantifiers. Proceedings of the North East Linguistics
Society 33, 397–403.
von Stechow, Arnim. 2008. Tense, modals, and attitudes and verbal quantifiers.
Paper presented at ESSLLI 2008.
Stephenson, Tamina. 2007. Judge dependence, epistemic modals, and predicates
of personal taste. Linguistics and Philosophy 30, 487–525.
Sternefeld, Wolfgang. 1998. The semantics of reconstruction and connectivity.
In Arbeitspapier des SFB 340 97, pp. 1–58. Universität Tübingen.
Sternefeld, Wolfgang. 2001. Partial movement constructions, pied piping, and
higher order choice functions. In Caroline Féry & Wolfgang Sternefeld
(Eds.), Audiatur vox sapientiae: A festschrift for Arnim von Stechow, pp. 473–486.
Berlin: Akademie Verlag.

249
Stone, Matthew. 1997. The anaphoric parallel between modality and tense. IRCS
TR 97-06, University of Pennsylvania.

Sudo, Yasutada. 2012. On the semantics of phi features on pronouns. Ph.D. thesis,
MIT.

Swanson, Eric. 2010. On scope relations between quantifiers and epistemic modals.
Journal of Semantics 27, 529–540.

Swanson, Eric. 2011. Propositional attitudes. In Claudia Maienborn, Klaus

von Heusinger, & Paul Portner (Eds.), Semantics: An International Handbook
of Natural Language Meaning, vol. 2, pp. 1538–1560. Berlin: de Gruyter.

Szabolcsi, Anna. 1987. Functional categories in the noun phrase. In István

Kenesei (Ed.), Approaches to Hungarian 2: Theories and analyses, pp. 167–190.
Szeged: JATE.

Szabolcsi, Anna. 1989. Noun phrases and clauses: Is DP analogous to CP? In John
Payne (Ed.), The structure of noun phrases, pp. 151–180. Berlin: de Gruyter.

Szabolcsi, Anna. 1994. The noun phrase. In Ferenc Kiefer & Katalin É. Kiss
(Eds.), Syntax and semantics 27: The syntactic structure of Hungarian, pp. 179–274.
New York: Academic Press.

Szabolcsi, Anna. 2011. Certain verbs are syntactically explicit quantifiers. In

Barbara H. Partee, Michael Glanzberg, & Jurǵis Šķilters (Eds.), The Baltic
international yearbook of cognition, logic and communication, vol. 6: Formal se-
mantics and pragmatics: Discourse, context and models, pp. 1–26. Manhattan, KS:
New Prairie Press.

Takahashi, Shoichi. 2010. Traces or copies, or both — Parts I and II. Language
and Linguistics Compass 4, 1091–1115.

Thurgood, Graham & Randy J. LaPolla (Eds.). 2017. The Sino-Tibetan lan-
guages. London: Routledge, 2nd edn.

Tonhauser, Judith, David Beaver, Craige Roberts, & Mandy Simons. 2013.
Towards a taxonomy of projective content. Language 89, 66–109.

Trutkowsky, Ewa & Helmut Weiß. 2016. When personal pronouns compete
with relative pronouns. In Grosz & Patel-Grosz (2016), pp. 135–166.

250
Vergnaud, Jean-Roger & Maria Luisa Zubizarreta. 1992. The definite deter-
miner and inalienable construction in French and English. Linguistic Inquiry 23,
595–652.

Vikner, Carl & Per Anker Jensen. 2002. A semantic analysis of the English
genitive: Interaction of lexical and formal semantics. Studia Linguistica 56, 191–
226.

Vikner, Sten. 1995. Verb movement and expletive subjects in the Germanic lan-
guages. New York: Oxford University Press.

de Vries, Mark. 2002. The syntax of relativization. Ph.D. thesis, University of

Amsterdam.

Ward, Gregory, Richard Sproat, & Gail McKoon. 1991. A pragmatic analysis
of so-called anaphoric islands. Language 67, 439–473.

Williams, Edwin. 1997. Blocking and anaphora. Linguistic Inquiry 28, 577–628.

Williamson, Janis. 1987. An indefiniteness restriction for relative clauses in

Lakhota. In Reuland & ter Meulen (1987), pp. 168–190.

Wiltschko, Martina. 1998. On the syntax and semantics of (relative) pronouns

and determiners. Journal of Comparative Germanic Linguistics 2, 143–181.

Wiltschko, Martina. 2014. The universal structure of categories: Towards a formal

typology. Cambridge: Cambridge University Press.

Wiltschko, Martina. 2016. Fake form and what it tells us about the relation
between form and interpretation. In Grosz & Patel-Grosz (2016), pp. 13–52.

Yalcin, Seth. 2007. Epistemic modals. Mind 116, 983–1026.

Yalcin, Seth. 2012. A counterexample to modus tollens. Journal of Philosophical

Logic 41, 1001–1024.

Yalcin, Seth. 2014. Semantics and metasemantics in the context of generative

grammar. In Alexis Burgess & Brett Sherman (Eds.), Metasemantics: New
essays on the foundations of meaning, pp. 17–54. Oxford: Oxford University Press.

Yashima, Jun. 2015. On the apparent unbindability of overt third-person pronouns

in Japanese. Natural Language and Linguistic Theory 33, 1421–1438.

251
Yatsushiro, Kazuko. 2001. The distribution of mo and ka and its implications.
In Maria Christina Cuervo, Daniel Harbour, Ken Hiraiwa, & Shinichiro
Ishihara (Eds.), Formal approaches to Japanese linguistics 3, pp. 181–198. Cam-
bridge, Mass.: MITWPL.

Zribi-Hertz, Anne. 2008. From intensive to reflexive: The prosodic factor. In

Ekkehard König & Volker Gast (Eds.), Reciprocals and reflexives: Theoretical
and typological explorations, pp. 591–632. Berlin: Mouton de Gruyter.

252