PLACE IN RETURN BOX to remove this checkout from your record.
1‘0 AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

2002 8 1 M,

 

MAR 19 17c“

 

12010'

 

 

 

 

 

 

 

 

 

 

 

 

2/05 czfimeDmJMd-nm

 

Some Topics in Comparative Constructions

By

Ming Xiang

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Linguistics and Germanic, Slavic, Asian and African Languages

2005

ABSTRACT
SOME TOPICS IN COMPARATIVE CONSTRUCTIONS
By

Ming Xiang

Making a comparison, namely, ordering objects along a certain dimension, is a basic and
universal cognitive ability of us, yet the linguistic encoding of comparisons is not simple.
To express different kinds of comparisons, there exist a variety of comparative
constructions. The same kind of comparison can also be expressed differently cross
languages. Through analyzing some specific problems, this thesis aims at addressing
some of the basic semantic and syntactic issues, using both English and Chinese data. For
example, what is the argument structure of comparatives? Is there a single syntactic
structure for all predicate comparatives in English, and cross-linguistically? How do the
semantic and syntactic properties interact to make comparisons? In Chapter 1, I introduce
a few problems that I will discuss later in the thesis. Although the rest of the thesis can be
roughly divided into two parts: a semantic part (chapter 2&3) and a syntactic part
(Chapter 4&5), most of my analyses are rooted in a syntax-semantics interaction of
comparatives. Chapter 2 talks about quantifier licensing in comparatives and its
consequence for the structure of the standard degree argument. I argue that first there
should be a distinction between phrasal and clausal comparatives, and second, for clausal
comparatives, the surface wide scope interpretations of quantifiers are actually the result
of pair-list readings in wh-constructions. Chapter 3 discusses why comparatives are

licensing environments for negative polarity items (NPIs). I argue that the basic

component of a comparison, namely, the scale provided by the gradable adjective, plays a
crucial role to license NPIs. Chapter 4 argues for the existence of a degree argument in
the structure of comparatives, even if this argument is not a constituent on the surface.
Chapter 5 proposes a tentative structure for comparatives that incorporates this

discontinuous degree argument. Chapter 6 is a conclusion.

Copyright by Ming Xiang 2005

All rights reserved

ACKNOWLEDGMENT

I am grateful to finally have this opportunity to express my gratitude to all the people that
have contributed to this thesis and to my professional growth. The first person I want to
thank is my thesis supervisor, Professor Alan Munn. I took one of my first linguistic
classes with him, and from there he led me into the areas of both theoretical linguistics
and psycholinguistics. Alan has helped me at every point of my professional
development, and this thesis has witnessed a lot of his expertise, comments, and
encouragement. I am also indebted to Professor Christina Schmitt. My interest in and
approach to syntax-semantics interface issues are largely influenced by her. She is often
skeptical on ideas and very direct, but it often turns out she is right. I thank Professor
Alan Beretta and Fernanda Ferreira for their guidance in my research in cognitive
science. I did my first research project in cognitive science under Alan’s supervision, and
have benefited from him ever since then. Working in Fernanda’s psycholinguistic lab is
the best thing that happened to me. She is smart, beautifill and humorous. Had she not
been so generous to open her lab, her classes and her office to me, I would almost
certainly be on a different path now that is less desirable. I owe all my phonology to
Professor Grover Hudson and Yen-Hwei Lin. Professor Hudson’s office is wide open to
me whenever I have problems teaching phonology. Professor Lin has given me great help
and encouragement when my spirit is low. I am grateful to Professor Barbara Abbott for
her time and patience whenever I have problems in semantics. I am indebted to Chris
Wilder and Chris Kennedy for their comments on part of the thesis. I also want to thank
all the fellow students in the linguistics department and the psycholinguistics lab. I have

learned a lot from you and I wish everybody a great future. Finally, I want to dedicate

this thesis to my dearest mom, dad, sister and Allan Schenkel, for their love and support

that make me who I am now.

vi

Chapter 1
l .0

1.1.

1.2.
1.3.

1.4.

Chapter 2

2.1.
2.2.

2.3.

2.4.

2.5.
2.6.

Chapter 3

3.0.
3.1.

3.2.

Table of Content

Introduction ...................................................................... 1
Setting the Stage ................................................................ 1
Quantifiers in comparatives ................................................. 3
1.1.1. Clausal and phrasal comparatives .............................. 4
1.1.1.1.The debate in the literature .............................. 5
1.1.1.2.Some evidence to support the phrasal analysis ....... 8
1.1.2. Quantifiers in comparatives ...................................... 11
NPI licensing in comparatives ........................................... 13
The degree argument and the structure of comparatives .............. 17
1.3.1. The traditional analysis ........................................... 17
1.3.2. Kenney (1999) ..................................................... 20
1.3.3. Evaluating the two analyses ..................................... 22
The organization of the thesis ............................................ 23
Quantifiers in Comparative Clauses .................................... 24
Introduction ................................................................ 24
The problems of quantifiers ............................................. 26
2.2.1. The interval based semantics for degrees ..................... 29
2.2.2. A summary ........................................................ 30
The pair-list reading in wh-constructions .............................. 31
2.3.1. Pair-list readings in wh-questions .............................. 31
2.3.2. Pair-list readings in relative clauses ............................ 33
2.3.3. Pair-list readings and comparative clauses ..................... 35
More evidence for the pair-list readings in comparative clauses. . 39
2.4.1. Subject-object asymmetry ........................................ 39
2.4.2. Monotone decreasing quantifiers in subject position ......... 44
2.4.3. Negative islands ................................................... 46
2.4.3.1.The negative islands effect in comparatives .......... 47
2.4.3.2. A diagnosis from cardinality degrees ................. 49
2.4.3.3. Summary ................................................. 55
A few words about phrasal comparatives .............................. 55
Conclusion .................................................................. 56
NPI in comparatives ....................................................... 58
Introduction ................................................................. 58
The problem with NPI ..................................................... 61
3.1.1. Comparatives as non-DE environments ........................ 61
3.1.2. Comparatives are not non-veridical ............................. 64

3.1.3. Existential polarity wh-phrases in Chinese comparatives. . .. 67
3.1.4. The interaction between clan and NPI

in Chinese comparatives ........................................................ 70
3.1.5. Summary and outline .............................................. 72
An analysis of Chinese dou ................................................ 72

vii

3.3.

3.4.

Chapter 4
4.0

4.1.

4.2.

4.3.

4.4.

4.5.

Chapter 5

5.0.
5.1.
5.2.
5.3.

5.4.

3.2.1. An introduction .................................................... 73

3.2.1.1.Dou — the distributor ..................................... 73
3.2.1.2.Dou — even ................................................ 77
3.2.1.3.Dou - already ............................................. 78
3.2.2. The analysis ......................................................... 81
3.2.2.1.Dou — distributivity as the end of a quantity scale. . ..84
3.2.2.2.The analysis of the even use of dou ..................... 95
3.2.2.3.The analysis of the already use of dou ................ 97
3.2.2.3.The leflness condition .................................. 102
3.2.2.4.Summary ................................................... 107
Licensing NPI through a scale ............................................. 107
3.3.1. Cheng (1995) ....................................................... 108
3.3.2. Don and NPI licensing ............................................ 111
3.3.3. NPI licensing in English comparatives ......................... 115
3.3.4. Summary ............................................................ 123
Conclusion ................................................................... 124
The Degree Argument ...................................................... 126
Introduction .................................................................. 126
Do we need a degree argument? .......................................... 126
4.1.1. Quantification over degrees ....................................... 126
4.1.2. The debate about quantification over degrees .................. 128
The definiteness effect ..................................................... 132
4.2.1. The definiteness effect in attributive comparatives ........... 132
4.2.2. Beil (1997) ......................................................... 133
The DE as an indication of the degree argument movement ......... 137
4.3.1. The quantifier intervention effect at LF ........................ 137
4.3.2. A new proposal ..................................................... 140
4.3.3. Predictions .......................................................... 145
The syntactic structure of comparatives ................................. 149
4.4.1. The discrepancy between semantics and syntax ............... 149
4.4.2. Solution one — late merge the than-clause ...................... 152
4.4.3. Solution two — the DegP-shell structure ........................ 155
Summary ..................................................................... 159
The DegP-shell analysis of Chinese Comparatives .................... 160
Introduction ................................................................. 160
The adjective selection constraints ...................................... 164
Chinese phrasal comparatives ........................................... 173
The structure of bare comparatives ....................................... 180
5.3.1. The differential degree argument ................................ 180
5.3.2. The VP—shell structure for double object constructions ...... 184
5.3.3. The DegP-shell structure for bare-comparatives ............... 189
Bi-comparatives ............................................................. 192
5.4.1. Head movement ................................................... 192
5.4.2. Against the adjunct analysis ..................................... 194

viii

5.5.

5.6.

5.7.

Chapter 6

Some consequences of a DegP-shell analysis .......................... 201
5.5.1. Reduplication of the adjectival predicate ....................... 201
5.5.2. The generic reading of the standard degree .................... 205
5.5.3. Summary ............................................................ 213
Lower DegP as the degree argument ..................................... 214
5.6.1. The extraction of the standard degree .......................... 214
5.6.2. Intensifying the degree argument ............................... 219
Conclusion .................................................................. 230
Concluding Remarks ....................................................... 231

ix

Chapter 1

Introduction

1.0. Setting the stage

Comparatives have been of great interest for linguists who study the semantics of
gradable adjectives and the structure of adjectival constructions. lntuitively speaking, a
comparative construction consists of the following elements: two degrees that are under
comparison, a gradable adjective that provides the dimension of the comparison, a degree
morpheme that indicates the ordering relation between the degrees, and a differential
degree that measures the difference between the two compared degrees. Consider the

example in (l):

(I) John is taller than Bill by three inches.

This sentence compares two degrees: John’s degree of height and Bill’s degree of height.
Bill’s degree is often referred to as the standard degree, and John’s degree the reference
degree. The comparison scale, i.e., the scale of height, is provided by the gradable
adjective tall. The comparative morpheme —er indicates an ordering that John’s height
exceeds Bill’s height. The differential degree three inches measures the difference
between John’s height and Bill’s height. In cases where the differential degree is not
Specified as a measure phrase, it still can be some implicit amount. For instance, in (2),

the differential degree is zero, and it is some unspecified amount (bigger than zero) of

degree in (3) and (4). As shown in (3) and (4), the unspecified differential degree can be

intensified by a degree adverb.

(2) John is as tall as Bill.
(3) John is much smarter than Bill.

(4) John is a little less smart than Bill.

There are many different kinds of comparative constructions. The examples we
have seen so far are predicate comparatives, where the gradable adjectives predicate the
subjects. Adjectives can also be used attributively in comparatives, where the adjectives

are part of the NP modifiers. (5) is such an example:

(5) a. John bought a cheaper car than I did.

b. She read more books than I did.

Most comparatives compare two things along the same comparison scale, as we have
seen so far. However, in some languages, English for instance, some comparatives seem
to contain two different scales on the surface. They are called sub-comparatives. The

following are some examples:

(6) a. The table is longer than the door is wide.

b. The Pacific Ocean is deeper than the Rocky Mountains are tall.

This thesis does not attempt to give a comprehensive account for all kinds of
comparatives. I will limit myself mostly to the discussion of predicate comparatives. I
believe the conclusions drawn in this thesis can be extended to other comparative
constructions, although I will leave the exact implementation open. The general research
question of this work centers around the semantics and syntax of comparatives as well as
their interactions. This thesis can be read in two parts. In the first part, I investigate the
following two semantic issues:

0 Quantifier licensing in comparatives (Chapter 2)

0 Licensing Negative Polarity Items (NPIs) in comparatives (Chapter3)
The second part focuses on the syntax of comparatives. I will discuss:

0 The degree argument (Chapter 4)

0 The overall structure of comparatives (Chapter 5)

Hopefully, the investigation of these problems can help us to understand some more
fimdamental issues of comparatives. For instance, what is the argument structure of
comparatives? Is there a single syntactic structure for all predicate comparatives in
English, and crosslinguistically? How do the semantic and syntactic properties interact to
make comparisons? Before we answer these questions, let me first lay out more details of

each problem to be discussed.

1.]. Quantifiers In comparatives
The first syntax-semantics interface problem in comparatives is the quantifier problem.

This is a good case to demonstrate how syntactic properties interfere with a semantic

issue. On the surface, there is a semantic problem of how to license and interpret
quantifiers at the standard degree position. For instance, how is the quantifier everybody

interpreted in (7)? And why is the negative quantifier nobody not licensed in (83), but is

fine in (8b)?
(7) a. John is taller than everybody else is.
b. John is taller than everybody else.
(8) a. *John is taller than no one is.

b. John is taller than no one. Brame (1983)

But I will argue that this is not a purely semantic issue. My analysis to this problem
crucially depends on the structure of the standard degree argument. Specifically I will
assume that there is a difference between standard degrees that are just DPs and those
that are elided CPs. To see clearly what the problem is, let me first introduce some

background on the structure of the standard degree.

1.1.1. Clausal and phrasal comparatives
In English comparatives, the standard degree can be expressed either as a clause or a

phrase, as shown in (9):

(9) a. John is happier than Bill is.(clausal comparative)

b. John is happier than Bill. (phrasal comparative)

Since Chomsky (1977), most people agree that the clausal comparatives are wh-

constructions, with a wh-operator binding a degree variable.

(10) John is taller than [Opj Bill is tj]

The comparative clause generates a set of degrees to which Bill is tall. Assuming the
maximality operator in Rullmann (1995), the maximal degree that Bill is tall to is picked
out as the standard degree, and John’s height is compared to that maximal degree.

Therefore, (10) is interpreted as (1 l):

(l l) 3dl[du> MAX (M2 tall (Bill, dz))][t2|ll (John, (11)]

(l 1) says that there is a degree d], John is tall to dl, and d1 exceeds the maximal degree
d2 that Bill is tall to.

However, there is more controversy about the syntactic structure of phrasal
comparatives. Since the two sentences in (9) have the same semantic interpretations, one
natural question to ask is whether they are syntactically related too. In this section, I will

review some proposals in the literature.

1.1.1.1. The debate In the literature
Following Bresnan (1973), one view assumes that phrasal comparatives are

derived from their clausal counterparts through comparative ellipsis/comparative

deletion. Let’s call this view the clausal analysis for phrasal comparatives. For example,

(9b) is derived from (9a), as shown in (12):

(12) John is happier than [cp Bill is]

Under this analysis, the semantic derivation of phrasal comparatives is the same as their
clausal counterparts. As Larson (1987) suggests, comparative ellipsis is just a special case
for a common ACD (antecedent contained deletion) process. Sentence (13) has the

structure in (14), where the comparative site contains an empty I’.

(13) John grew as tall as / taller than his father .

(14) [[9 John [1’ past[vp ngW[Ap as tall [cp as his father [1’ e]]]]]]

If comparatives are quantifications over degrees, and quantificational elements undergo

QR, the AP in (14) as tall as his father e will move to adjoin to IP at LF, as shown in

(15):

(15) [whim as tall [cp as his father [1~e]]][;p John [1’ past [vp grow [APi e]]]]]

Now the empty I’ in the comparative site can be reconstructed from the l’ in the main

clause, and the desired LP is derived in (I 6):

(16) [[plApi as tall [cp as his father [1’ Past [vp grow [Am e]]]]][1p John [r Past [vp grow

[AH 6111]]

On the other hand, various authors have challenged the clausal hypothesis for
reasons I will come back to shortly. Under this view, the DP Bill in (9b) is not derived
from a clause, instead it is just a simple DP. Let’s call this proposal the phrasal analysis
for phrasal comparatives. Since in (9b) what is being compared is Bill’s degree of
happiness, not him as an individual, one problem for the direct analysis is how to derive
the meaning of a degree from the meaning of an individual without appealing to copying
a predicate, as has been done in (16). Heim (1985) proposed a semantic solution for the
problem. For example, the comparison in (17) is comparing two people along the
dimension of “happiness”. A dimension of comparison is a function from individuals to
degrees, so the dimension “happiness” is translated by Heim as a lambda-iota expression
like ( 18). This function can take the two people in (17) as arguments respectively and

give two degrees. The meaning of the comparative morpheme —er is specified as (19):

(17) John is happier than Bill.

(18) Ax ty [x is y-happy]

(19) “-er<a, b> r: is true iff f(a) > f(b)

Under this analysis, (17) has a semantic interpretation like (20), and (20) is true iff “the

happiness of John” exceeds “the happiness of Bill”.

(20) -er<Iohn, Bill >Ax ty [x is y-happy]

To summarize, both the clausal analysis and the phrasal analysis have a way to
derive the correct semantic interpretation for phrasal comparatives. The former appeals to
the general solution for ACD constructions, and the latter appeals to the semantics of
functions, i.e. the direct analysis. Next I will turn to some syntactic tests that can

distinguish the two structures.

1.1.1.2. Some evidence to support the phrasal analysis
There have been some arguments showing that phrasal comparatives should
receive a different syntactic structure from their clausal counterparts (Brame 1983,
Napoli 1983, Hankamer 1973, among others). The crucial point behind these arguments
is that if the phrasal analysis is correct, phrasal comparatives should reveal a different
distribution from their clausal counterparts. Next I will turn to some examples of this sort.
First, the DP afier than appears with nominative case in clausal comparatives, but

accusative case in phrasal comparatives.

John is taller than she is.

9‘

(21)
b. ‘John is taller than she.
a. ‘John is taller than her is.

b. John is taller than her.

Second, reflexives only appear in phrasal comparatives, which is consistent with the

phrasal analysis.

(22) a. ‘It is impossible for John to be taller than himself is.

b. It is impossible for John to be taller than himself.

Third, extraction out of a clausal comparative is bad, but fine out of a phrasal

comparative.

(23) a. *Who is John taller than is?

b. Who is John taller than?

Fourth, clausal comparatives don’t license negative quantifiers, but phrasal comparatives

do.

(24) a. *John is taller than no one is.

b. John is taller than no one.

Finally, in certain dialects, an explicit wh-word appears in the comparative clause, but we

don’t see the same pattern in phrasal comparatives.

(25) a. She did more than what John did.

b. *She did more than what Bill.

There are more tests we can use than I have listed here. But I hope the point is
already clear, namely, these tests seem to show that phrasal comparatives should have
different structures from clausal ones. However, one point still worth keeping in mind is
that not all the tests above necessarily point to the phrasal analysis. Some of them could
be accounted for in other ways. For example, Lechner (1999, 2001) pointed out the
accusative case test in (21) is still compatible with a clausal analysis if we at the same
time adopt the position that comparatives are similar to coordination structures. He
noticed that the accusative remnants can be licensed in coordinations that involve

gapping, as shown in (26):

(26) a. John is eager to meet them, and me too.

b. *John is eager to meet them, and me is eager to meet them too.

For more details of Lechner’s analysis I refer readers to the original papers. Although the
alternative analyses Lechner provided could account for some of the data above, I don’t
think the clausal analysis can account for all the differences. In particular, it is not clear
how the clausal analysis could accommodate the quantifier data in (24) and the wh-
extraction data in (23). As will become clearer in the next section and in Chapter 2,
acknowledging the structural difference between phrasal and clausal comparatives
provides us a natural explanation of quantifier licensing in comparatives.

Before I demonstrate the quantifier problem in comparatives, I need to point out

that although we see some strong evidence to support the phrasal analysis of phrasal

10

comparatives, all the cases we have looked at are predicative comparatives, which is also
the focus in our discussion of quantifiers. It seems that for some kind of attributive
phrasal comparatives, we do need to give them clausal structures. For instance, Lerner
and Pinkal (1995) discussed two kinds of attributive comparatives: the one with narrow

readings (NRA) and the one with wide readings (WRA).

(27) a. John owns a faster car than this BMW. (NRA)

b. John owns a faster car than Bill. (WRA)

In their analysis, the DP afier than in the NRA kind, e.g. this BMW, is still analyzed as a
simple DP. However, the DP after than in the WRA kind, e.g. Bill, is analyzed as an
elided CP. In addition, Lechner (1999, 2001) also discussed the clausal analysis for some
other attributive comparatives. In the rest of this thesis I will not have opportunity to talk
about phrasal comparatives with clausal structures any more, but it is nevertheless a very
interesting issue to keep in mind.

To summarize, in this section we see that at least for some kind of comparatives
(e,g. predicative comparatives), the phrasal comparatives have different syntactic

structure from their clausal counterparts. This is a crucial point for our discussion next.

1.1.2. Quantifiers in comparatives
Since May (1985), the scope of quantifiers is derived from the syntactic structure at LF.
Quantifier raising (QR) operates at LF and establishes a one to one mapping between the

syntactic structures and the scope interpretations. Consider (28):

ll

(28) Some student read every book.
a. [some student. [every bookk [ ti read tk]]]

b. [every bookk [some student.- [ti read tk ]]]

After applying QR to both subject and object quantifiers, depending on the scope relation
of the two, (28a) and (28b) are two different LF representations of (28). At LF (28a),
some student takes scope over every book, and the sentence means there is a student x,
such that x read every book. On the other hand, every book in (28b) takes wide scope
over some student, so the sentence means that for each book x, there is a student y that
read the book x. In this way, semantic properties, scope ambiguities in particular, are
encoded transparently in the structure at LF.

It is this mapping between syntactic and semantic representations that raises
questions for quantifiers in comparatives. Quantifiers in comparative clauses take wide
scope over the comparative clause. For example, (29) only has the interpretation (293),
where the quantifier every student takes wide scope over the wh-operator. It means that
for each student x, John is taller than x; the (29b) interpretation, where the quantifier
takes narrow scope under the wh-operator, is on the other hand an implausible
interpretation, because it means that John is taller than a maximal degree d such that

every student is d-much tall, which will have to force every student to be of the same

height.

(29) John is taller than every student is.

12

a. Vx (x is a student) [John is taller than [MAX Op: x is di-much tall]]

b. "' John is taller than [MAX Opi Vx (x is a student and x is di-much tall)]

The problem we are going to solve is how to derive the wide scope interpretation
in (29) and why it is an obligatory interpretation. As I will show in more detail in Chapter
two, the wide scope interpretation of (29) is not derivable through QR, as we have done
for (28), since the quantifier in (29) is non-extractable. But this doesn’t mean that the
interpretation in (29a) is not dependent on the syntactic representation any more. Exactly
to the contrary, I will argue that the obligatory wide scope interpretations are heavily
dependent on comparatives being wh-constructions. The solution I will provide is also
extended to explain the facts that negative quantifiers are licensed differently in clausal

and phrasal comparatives, as shown in (30):

(30) a. *John is taller than no one is. (clausal comparative)

b. John is taller than no one. (phrasal comparative)

Therefore, the analysis I will advocate crucially draws a line between phrasal and
clausal comparatives. Quantifiers in clausal comparatives behave the way they do
because they are constrained by wh-movements constraints. The same constraints don’t

exist in phrasal comparatives.

1.2. NPI licensing in comparatives

13

The semantic problem of quantifiers is tied closely to the structure of the standard
degree, which is a necessary component to make a comparative. Another central
component of comparatives is the adjective, which provides a scale of some dimension,
for instance, height, intelligence, beauty, etc. The semantics of scalar adjectives
themselves have been discussed extensively in the literature, but it is less often discussed
what these scales do for comparatives, beyond the basic intuition that comparisons can’t
be made without them. In Chapter 3, I will show that scales can do more for
comparatives. Specifically, they make comparatives a licensing environment of NPls. In
this discussion, we will also find that Chinese data becomes instructive. A cross-linguistic
perspective offers us some insight that is not as easily available through only English
data.

Generally speaking, negative polarity items (NPls) are lexical items that have to

stay within the scope of some decreasing operator, for instance, a negation, as shown in

(31):

(31) a. He didn’t see anybody.
b. *He saw anybody.
c. John won’t ever do that.

(I. *John will ever do that.

When we look at the various analyses that have been proposed to characterize the

distribution of NPls, it is immediately clear that both syntax and semantics have

contributed to the licensing of them. Progovac (1994) analyzes NPI licensing as syntactic

l4

binding, and suggests that NPIs behave like pronominals and have to be bound by a local
A’-binder, which could be a negation or some other functional operator. On the other
hand, there have been various proposals on the semantic properties that can license NPIs.
Ladusaw (1980) argues that downward entailing environments trigger NPI licensing,
Fauconnier (1975), Lee & Horn (1994) and Larihi (1998) all have suggested that NPIs
are licensed through scalar implicature, and Giannakidou (1998, 1999, 2001), Zwarts
(1995) argues for (non)veridicality as the crucial licensing condition.

Interestingly, English comparatives license NPIs, as shown below:

(32) John is taller than anybody.

(33) John is taller than I ever imagined.

There is obviously no negative element in (32) and (33) that will trigger NPI licensing.
As I will show in Chapter three, it is controversial whether comparatives are downward
entailing environments (Schwarzschild & Wilkinson 2002, Zepter 2003) and it is also not
completely clear if one can characterize comparatives as (non)veridical environments.
More importantly, although many languages do license NPIs directly in comparatives
(Haspelmath 1997), some don’t. Chinese belongs to the latter kind. As shown in (34),
NPIs are not licensed in the two kinds of Chinese comparatives. However, adding the
distributive morpheme dou to comparatives will turn comparatives into NPI-friendly
environments again, but only when certain syntactic conditions are met. As shown in

(35), only one kind of comparatives will license an NPI when dou is added.

15

(34) a. *Ta gao shei yidian
he tall who a little
He is taller than anybody.
b. *Ta bi shei gao yidian
he bi who tall a little
He is taller than anybody.
(35) a. *Zhangsan dou gao shei yidian
Zhangsan dou tall who a little
Zhangsan is a little taller than anybody.
b. Zhangsan bi shei dou gao yidian.
Zhangsan than who all tall a little

Zhangsan is a little taller than anybody.

In Chapter 3, I will address the NPI issue in comparatives. Independent investigation of
the distributive morpheme dou suggests that the core function of dou is not distributivity,
instead, it should be a maximality operator that picks out the maximal value on a
contextually determined scale. Consequently, NPIs in Chinese comparatives are licensed
through scalar implicatures. The implication of this analysis for English data is also
discussed in Chapter 3. Importantly, although the exact implementation of this scalar
implicature can be different from language to language (for instance, Chinese needs the
maximality operator dou, but English doesn’t), comparatives nevertheless automatically

provide scales through their gradable adjectives.

l6

1.3.The degree argument and the structure of comparatives

While discussing the two semantic issues raised above, we have already touched upon
some syntactic properties of comparatives, for example, the structure of the standard
degree. In the second part of this thesis, I will discuss the general structure of
comparatives. Again, semantics and syntax interact closely here. The central problem is
whether one should acknowledge a degree argument in the argument structure of
comparatives. The degree argument is motivated on the semantic fact that gradable
adjectives involve quantification over degrees, but it causes a syntactic problem because
this argument is not a constituent on the surface syntax. Let’s look at the problem in more

detail.

1.3.1. The traditional analysis

Gradable adjectives, such as tall, smart, beautiful, are traditionally assumed to
map an individual to a degree of some properties (e.g. Cresswell 1976, E. Klein 1991).
As shown in (36), an adjective introduces two arguments: an individual argument and a
degree argument. (36) is true just in the case that the degree of x on a scale related to b is
at least as great as d. So the question is how to define the value of d. Let’s illustrate it
with two examples, one simple degree construction and the other comparative

construction.

(36) 4) (x, d) (¢ is a gradable adjective such as tall)

17

In simple degree constructions, such as (37), the degree argument d is
determined by context. The adjective tall, smart and beautiful maps the individual Mary

to this contextually determined degree, as shown in (38).

(37) Mary is tall/smart/beautiful.
(38) a. tall (Mary, (1)
b. smart (Mary, d)

c. beautiful (Mary, d)

(37) is true if and only if Mary is at least as tall/smart/beautiful as a degree d, which is
determined by whatever value that is considered as tall, smart or beautiful in the context.
For instance, if 6-feet is commonly assumed to be tall, then for (38a) to be true, Mary
must possess the degree of height d that is at least 6-feet. The advantage of this analysis is
that an implicit degree argument (see (39)) aligns the structure of (39) to those adjectival
predicates with explicit measure modifications, such as (40). Specifically, in both (39)
and (40), the adjective introduces a degree argument in the [Spec, AP] position, as shown

in (41). In (39) this argument is implicit, but in (40) it is expressed by a measure phrase.

(39) Mary is (d-much) tall/smart/beautiful.

(40) Mary is six-feet tall'.

 

' For independent reasons, adjectives such as smart and beautiful can’t take measure phrase modification
(see Schwarzschild 2002)

18

(41) [P

John VP
/\
is AP
/\
DegP A
d/six-feet tall

Along the same line, comparatives are traditionally assumed to be quantificational
expressions that quantify over a degree argument (Scuren 1973, Hellan 1981, von
Stechow 1984, Helm 1985, Lerner & Pinkal 1992, I995, Hazout 1995, Rullmann 1995).
Consider (42) for an example. The adjective tall still introduces two arguments: and
individual and a degree, as shown in (42b). So (42a) is true just in case that John is at
least as tall as a degree argument d. The value of d itself is restricted by the degree
morpheme and the comparative clause, namely, d has to exceed Bill’s degree of height,
as shown by the semantic formula in (42c). Under this analysis, the degree argument still

stays in the [Spec, AP] position, just like other degree constructions, as shown in (42d).

(42) a. John is taller than Bill
b. tall (John, d)

c. 3d[d > dam ][ tall(John,d)]

d. IP
/\
John VP
/\
is AP
/\
DegP A
(-er than Bill) tall

19

Under this analysis, a comparative construction only has two main arguments, an
individual (John in (42)), and a degree argument. The degree argument is a complex
argument. It consists of a comparative morpheme, a standard degree, and (optionally) a
differential degree.

1.3.2. Kennedy (1999)

Different from the traditional analysis, based on how we intuitively make a
comparison, it seems appealing to posit three arguments for comparatives: a standard
degree argument, a diflerential degree argument and a reference degree that the subject is
true of. This is essentially the line taken in Kennedy (1997, 1999)2.

Adopting the extended degree phrase structure in Abney (1987), Corver (1990)

and Grimshaw (1991), Kennedy constructs the semantics of comparatives from the

following structure:
(43) John is taller than Bill is.
IP
/\
John ......
%
Deg’
/\
Deg’ PP
/\ A
Deg AP [than Bill is]
-er tall

Under this analysis, the comparative clause, i.e., the than-phrase, generates the standard

Value of comparison, the adjective tall provides a measure function from individuals to

20

degrees, and the comparative morpheme —er (or less, as...as. . .). The compositional
semantics of comparatives crucially lies on the interpretation of the degree morpheme,

which is defined in (44):

(44) Deg = XGleR(G(X))(d)l

G is the measure function provided by the AP, and d is the standard degree value
provided by the PP. Depending on what the degree morpheme is (-er, less, or as...as..), R
expresses an ordering relation between the standard and the reference values, so R could
be either more, less or equal to relations. In this way, the degree phrase DegP expresses

some property of an individual. For instance, the DegP in (7) expresses the following

prepertyz

(45) DegP: Mlmol‘d tall(X))( dstandardll

When the DegP combines with the subject, we derive the meaning of the sentence as

below:

(46) | I more (tall (John)) (dsmndard) l |

What this means is that the reference value, John’s height, and the standard value needs
to stand in a partial order relation defined by more, defined as in (47). (46) is true if and

only if (48) holds:

21

(47) iim0re(dRerefence)(dStandard)iI = I Iff dReference > dStandard
(48) iimore(dJ0hn)(dStandard)“ = I Iff dJolm > dStandard

1.3.3. Evaluating the two analyses

The difference between these two proposals crucially lies in the degree argument.
Kennedy’s analysis rejects such an argument, and therefore employs different semantic
compositions from the traditional analysis. Therefore, to distinguish the two proposals,
we will have to first answer the question if such an argument is necessary. In this thesis, I
will argue for the existence of this argument (Chapter4). Drawing on data from the
definiteness effect in attributive comparatives, I argue that the LF movement of the
degree argument provides a natural account for the data. The degree argument is a
constituent at LF, but discontinuous in the surface word order.

The existence of such a degree argument raises an interesting problem of syntax-
semantics mismatch, because on the surface the degree morpheme —er and the standard
degree than-phrase don’t form a constituent. How can we maintain such an argument in
the syntax and at the same time derive the correct word order? In chapter 4 and chapter 5
I will provide a tentative analysis to this problem. To preview, the traditional structure in
(42d) will be rejected. Focusing on data from Chinese comparatives, I will argue for a

Degree-shell structure for comparatives, as shown in (49).

22

(49) DegP
/ \

Deg’
/ \
Deg AP
tall-er / \
A DegP

l 63“) ' /\
Deg’
/\

Deg PP
t
____J than Bill

 

Although the Chinese data does suggest that a shell like structure is on the right track, it
has to be acknowledged that some of the analyses in Chapter 5 are speculative, and
certainly raise many questions for future research.

1.4. The organization of the thesis

The goal of this thesis is to address the syntactic and semantic issues laid out above. In
Chapter two I will investigate the quantifier problem in comparatives. Chapter three is a
discussion about NPIs in comparatives. Chapter four discusses the issue of the degree

argument, chapter five discussed the general structure of comparatives, and Chapter six is

the conclusion.

23

Chaper 2

Quantifiers in Comparative Clauses

2.1.Introduction

As introduced in Chapter one, the structure of phrasal and clausal comparatives
has driven some debate. In this chapter, I will show that maintaining a structural
difference between the two provides us an opportunity to explain some data about
quantifiers in comparatives.

Quantifiers in comparative constructions raise two problems. First, quantifiers in
comparatives take obligatory wide scope interpretations. As shown in (1), assuming that
comparatives involve quantification over degrees (Cresswell 1976, Hoeksema 1983, von
Stechow 1984, Heim 2000), the narrow scope interpretation of the quantifier everybody

requires that everybody else is of the same height, and this is not a natural reading of (1).

(1) John is taller than everybody else (is).
a. For everybody x, John’s height exceeds x’s height.

b. *There exists a height h, everybody is h-tall, and John’s height exceeds h.
Second, monotone decreasing quantifiers are not licensed in clausal comparatives, but

they are fine in phrasal comparatives (2a &2b). This is puzzling given that clausal and

Phrasal comparatives are almost equal when they license other quantifiers (20 &2d).

24

(2) a. *John weighs more than nobody/few people/at most five people weighs.
b. John weighs more than nobody/few people/at most five people.
c. John weighs more than everybody else/somebody weighs.

d. John weighs more than everybody else/somebody.

The first problem has attracted lots of attention in the literature (see discussion below).
The second problem, however, has not been discussed in the literature as a related
problem. At the same time, although (23) & (2b) are ofien quoted as an argument for the
distinction between phrasal and clausal comparatives, it is not clear why clausal
comparatives don’t license certain quantifiers.

In this chapter, I will argue for a single analysis that will connect (1) and (2)
together. Two points are crucial for the analysis. First, there is a real distinction between
phrasal and clausal comparatives, at least for the kind of comparatives in (2) (for a debate
on this issue see Bresnan 1973, Brame 1983, Heim 1985, Lerner and Pinkal 1995,
Lechner 1999, among others); second, the properties of quantifiers in clausal
comparatives fall under the general properties of quantifiers in wh-constructions. In a
nutshell, clausal comparatives license quantifiers through their property of being wh-
constructions, which poses constraint on what kind of quantifiers can be licensed. On the
Other hand, phrasal comparatives, at the least the kind we are concerned with here, don’t
involve wh—constructions, hence are not constrained to reject negative quantifiers. As will
be demonstrated later, phrasal comparatives are relatively more straightforward in this
Case, and I leave them to the end of the discussion. A large amount of the discussion will

be devoted to the more puzzling behavior of clausal comparatives.

25

The organization of this chapter is the following: in section two, I will lay out in
more details what the problem is and point out some problems for a previous analysis. In
section three, I will introduce the general analysis I adopt for the pair-list readings in wh-
constructions, and my analysis of the problem. More evidence for the current analysis is
provided in section four. Section five is devoted to quantifiers in phrasal comparatives,

and section six is a conclusion.

2.2. The problem of quantifiers

It is not a simple question how quantifiers such as everybody in (1) can be
licensed in clausal comparatives. The problem is that the quantifier in comparative
clauses takes obligatory wide scope over the comparative clause (Larson 1988), which is
in conflict with some standard assumptions about scope interpretations. By common
assumptions, the scope relation between two quantificational elements is established
through the c-command relation between the two at the LF structure (May 1985). To put

it in an informal way, let’s take (3) as our assumption:

(3) For or to take scope over B, or and B are quantificational elements, or needs to c-

command [3 at LF.
Let’s illustrate the problem with the example in (4). At LF, the quantifier

everybody can either move to a higher position to scope over the maximality operator

(Rullmann 1995), or it can stay within the scope of the maximality operator. In other

26

words, (4) could be ambiguous because of the scope ambiguity between everybody and
the comparative clause. The two interpretations are shown in (4a) and (4b).
(4) John is taller than [MAX everybody else is].

a. 3d’ (Vx(person(x)) & MAX (M tall (x, d)) 9 (d’ >d & tall (John, d’)))

b. ad’s...“ Ad Vx(person(x) & tall (x, d) -) (d’ >d & tall (John, (1’)

(4a) and (4b) have different truth conditions. (4a) is true if for each person, John is taller
than the maximal height of that person. This is obviously a possible reading of (4). On the
other hand, the truth condition of (4b) is less straightforward. What (4b) says is that there
is a set of degrees of height and we pick the maximal one, and each person in the context
is that tall. In other words, for (4b) to be true, we have to force everybody to be the same
height. It is not obvious if (4) has this interpretation. The problem is that when (4a) is true,
it could possibly entail (4b) in the circumstances that everybody else just happens to be
the same height, although it is not an obligatory context for (4a) to be true. Therefore, it is
hard to decide whether (4b) is a genuine interpretation. We need a case where the
entailments of one interpretation don’t interfere with the other interpretation. Let’s
consider such an example in Wilkinson (1998), where by using non-monotonic

quantifiers the two readings are logically independent of each other.

(5) Hubert is taller than exactly 5 of the others are.

3. Find the largest height h, where exactly 5 individuals other than Hubert are h

tall, and Hubert is taller than h. (narrow scope)

b. There are exactly 5 individuals that are shorter than Hubert. (wide scope)

27

If (5a) were the available interpretation, (5) could be true in a context as in (6), where
Hubert is 5'8", there are 5 people that are 5'7", and some other people are 5'5", 5'3", 5' 1"
etc, because we do find a unique maximal height, namely 5'7", that exactly 5 people are

of that height, and they are shorter than Hubert.

(6)
5'8" Hubert

5'7" T, Q, P, A, C

 

5'5" W, B, V
5'3" Y
5'1" 0

However, native speakers’ intuition is that (5) is false in a situation like (6), exactly
because the quantifier has to take wide scope over the maximality operator. Therefore
there could be only five people that are shorter than Hubert.

Afier establishing the wide scope interpretation fact, the question is how this
interpretation is derived. Unfortunately the wide scope interpretation in (4a) can’t be
derived by QR everybody out of the than-clause, as pointed out in Schwarzschild &
Wilkinson (2002). One problem for QR is that comparative clauses are generally
considered as wh—constructions (Chomsky 1977), as shown in (7), so it should form an
extraction island that blocks QR. We can find the parallel illicit extraction in wh-

questions and QR, as shown in (8).

(7) John is taller than [everybodyi [Opj MAx [xi is dj-rnuch tall]]
I l

 

28

(8) a. *Whoi is John taller than t.- is?
b. Someone said that everyone left. \lElV

*VEI

2.2.1. The Interval based semantics for degree

Schwarzschild & Wilkinson (2002) (henceforth S&W) proposes an analysis
which solves the problem by using an interval based semantics for degrees. Traditionally,
degrees are assumed to be points on a scale (Cresswell 1976, Hoeksema 1983, von
Stechow 1984, Heim 2000, among others.) S & W argue that if we revise this assumption
to another one which represents degrees as an interval on a scale, we would be able to
derive the correct interpretation of (4) without appealing to QR. More specifically, under
this analysis (4) is true as long as we can find an interval on a height scale containing the
height of everyone but John, and above that interval there is an interval of John’s height,

as shown in (9). This is certainly possible and we do not have to force everybody to be

the same height. At the same time, no QR is needed.

(9) John is taller than everybody else is.

 

 

 

 

John interval 1
B, K, L, Y... interval 2
J, W

T

 

 

 

 

29

This analysis elegantly derives the correct semantics for (4) and avoids the problem of
QR. However, it faces an empirical problem, i.e. being a pure semantic analysis, it is
blind to the syntactic difference between clausal comparatives and phrasal comparatives.
Assuming the semantics of degrees is constant for both clausal and phrasal comparatives,
one would predict that clausal and phrasal comparatives would behave the same way with

respect to quantifiers. We already know that is not the case, data repeated below:

(10) a. *John weighs more than nobody/few people/at most five people weigh.

b. John weighs more than nobody/ few people/at most five people.

One could argue that the analysis in S & W (2002) is only designed to address the
wide scope interpretation issue of everybody, and does not intend to address the general
licensing of quantifiers in comparatives. As a matter of fact, negative quantifiers are not
discussed in S&W’s proposal. Therefore, one possibility is to maintain the analysis for
the interpretation of everybody, but pursue another analysis for negative quantifiers. The
point is well taken. However, we still need to ask, is there an analysis that actually unites
the two kinds of quantifiers in comparatives? In this chapter, I will pursue an analysis that

does this job.
2.2.2. A summary

To summarize, we have two problems that need an account. First, for clausal

cOmparatives, quantifiers such as everybody take obligatory wide scope over the than-

30

clause, and negative quantifiers are not licensed; second, for phrasal comparatives,
quantifiers such as everybody behave the same, but negative quantifiers are also licensed.
The analysis I will develop in this chapter crucially takes the advantage of the fact
that clausal comparatives involve wh-constructions, whereas phrasal comparatives don’t.
Drawing parallels between comparative clauses and other wh-constructions, I argue that
the wide scope interpretation in (4) is actually a pair-list reading. Independent constraints
on pair-list readings will rule out monotone decreasing quantifiers in clausal
comparatives. On the other hand, quantifiers in phrasal comparatives can easily take wide
scope through QR, and there is no obvious constraint on monotone decreasing quantifiers

in phrasal comparatives.

2.3. The pair-list reading in wh-constructions
In this section I will introduce the analysis of pair-list readings in wh—question and
relative clauses. My analysis of comparatives will be built upon the assumptions fiom
this section.
2.3.1. Pair-list readings in wh-questions
Wh-questions with quantifiers are usually ambiguous between having a functional
answer, a pair-list answer and an individual answer. For example, the question in (11) can
have three different answers.
(11) Who does everybody love?
a. His Mom. (functional)
b. Mary loves Bill, John loves Alice, Jason loves Jane... (pair-list)

c. Mary. (individual)

31

(11a) provides a function that maps an individual to his mom. This is clearly different
from the individual answer in (11c), which only provides an individual as the answer.
However, it seems reasonable to think that the pair-list reading in (11b), which provides
pairs of lovers, is a special case of the functional reading because the pair-list answer is in
the extension of the firnctional answer (Engdahl 1986, Chierchia 1993). To derive a pair-
list answer, we only need to apply the function, e. g. a function mapping a person to his
lover, to each individual in the relevant domain of the function, e.g. Mary, John, Jason,
etc. There are some differences between a functional reading and a pair-list reading, but
since they are not crucial for this chapter, I will leave them aside for the moment.
Chierchia (1993) proposes the syntactic structures in (12) and (13) to derive the

pair-list reading.

(12) who, [everybodyj [tj loves t§]]
(13) [[whoi everybodyj] [tj loves :91]
(14) APElA [W (everybody, A) A P (Ap [Elf e [A —+people] 3x 6 A[p = "loves(x,

fiX))])l

In (12), the wh-operator leaves a functional trace behind, which consists of a domain part
and a range part. A process called Absorption happens to form a complex operator by
adjoining who and everybody, as shown in (13). The exact syntactic process of absorption
is lefi open. Chierchia temporarily suggests that it could be an adjunction of the DP

everybody to the Spec of CP. What is important is that afier the absorption, we have a

32

complex operator, the who-part binds the range of the functional trace, and the
everybody-part binds the domain. As shown by its logical form in (14), (13) amounts to
saying that there is a firnction f that maps people to their lovers. The domain of f is fixed
by extracting a minimal witness set from the quantifier everybody. For every x in the
domain of f; x loves f(x). One can easily derive the pair-list reading by feeding the
arguments to the function f.

Under this analysis, the individual answer is derived if the wh-operator leaves an

individual type trace behind, as shown below:

(15) who; [everybodyj [tj loves ti]]

(16) AP P (Ap [3x person (x) A p = "Vy [person (y) —) love (x, y)]])

2.3.2. Pair-list readings In relative clauses

Sharvit (1999) argued that pair-list readings exist in relative clauses like (17) and

(18).

(17) The woman every mani loves invited him,

(18) The picture of himself} which every student; hated annoyed his, friends.

Notice that the quantifiers above face the same problems as the quantifiers in
comparatives. For example, every man in (17) needs to scope out the relative clause and
bind the pronoun his, but direct binding through QR is impossible because the relative

clause forms an extraction island. Sharvit argues that the binding between the quantifier

33

and the pronoun is indirect, and brought out by a pair-list reading, as shown in (19). In
(19), the relative operator leaves a firnctional trace behind; the quantifier every man
adjoins to the relative operator and leaves a trace that binds the argument variable of the
functional trace. The relative clause is interpreted as a function f that maps men to women
they love. The domain of f is fixed by extracting a minimal witness set from the

quantifier everybody. For every x in the domain off, f(x) invited x.

(19) IP
/\
DP‘; 1P
/\ A
D NP t; invited him;
the /\
NP CP

woman /\
()plj (:9
every man; Op; [t; loved tj']

Sharvit’s solution to the indirect binding problem in relative clauses leads us to think that
quantifiers in comparative clauses can be treated along the same line. lntuitively there are
two reasons to draw this parallel. First, the problem for relative clauses is similar to the
problem of comparative clauses; second, Sharvit’s solution is crucially based on the fact
that relative clauses are wh-constructions, which is also true for comparative clauses. In

addition, as I will show in the next two sections, this approach also finds some empirical

support.

34

2.3.3. Pair-list readings and comparative clauses

The goal of this section is to lay out an explicit analysis of quantifiers in
comparative clauses. I will show that the previous analyses of pair-list readings in other
wh-constructions can be applied to comparative clauses as well.

Following Larson (1987), I assume that the predicate in the ellipsis site of the
comparative clause is treated essentially the same as the antecedent contained deletion
(ACD) scenarios. Therefore the example in (20) undergoes the following operations. First,
the degree argument moves to a higher position, as shown in (21); second, the predicate
in the main clause is copied into the ellipsis site within the comparative clause, as shown
in (22).

The task at this point is to explain how the quantifier everybody in (22) gets wide
scope interpretation over the than-clause. To simplify the description, I will focus only on
the than-clause in the analysis in (23). Each node on the tree is labeled to show the

compositional derivation of the semantics.

(20) John is taller than everybody else is.
(21) {-er than everybody else is e]; John is [d; -tall]

(22) [-er than everybody else is d-tall]; John is [d ; -tall]

(23) PP

/\
than CP (e)
Oph/c>\ C’ (d)
Eleverybody else] ;(b) Op,- (3) t; is d;'-tall

35

. APM [1D [Dom (f) = P & [Vd (deD—)3x(xeDom(f) & f(x, d)))] &

[Vx(xeDom(f)—->Eld(deD& f(x, d)))]]]
(The operator Op generates a set of degrees. f is a function that maps individuals

to degrees, and P is the domain of f)

. 1R (R=W(AP[Vx(people(x)—)P(x))]))
(The minimal witness set of the quantifier everybody else provides the domain P

for the function f. This gives us a set of people in the context as the domain.)

. Af [1D [Dom (f) = People & [Vd (deD—>Elx(xeDom(f) & f(x, d)))] &

[Vx(xeDom(f)—>Eld(deD& f(x, d)))]]]

(The combination of everybody else and 0p gives us a complex operator)

. Axkd [d=tall(x)]
(This supplies a fimction that maps an individual to his degree of height)

. [1D [Dom (f) = People & [Vd (deD—>3x(xeDom(f) & [d=tall(x)]))] &
[Vx(xeDom(f)—)3d(deD& [d=tall(x)]))]]]
(Feeding the function generated in ((1) into the complex operator, we derive a set

of degrees by mapping every individual in the domain P to their height)

The degree operator in (23) binds a firnctional trace, which consists of a domain

and a range. The basic semantics of this degree operator (23a) is to generate a set of

degrees. To achieve that, we need to set up the value of two variables: P sets up the

values for the domain of the firnction, and f gives us the relation between the members in

the domain and the members in the range. By absorption the quantifier everybody else

adjoins to the degree operator and leaves a trace that binds the domain variable of the

functional trace. The minimal witness set of this quantifier provides a value for P (23b),

36

namely, the domain of the function is the relevant people (23c) in the context. By
adjoining the quantifier and the degree operator, we derive a complex operator whose
domain is set, but we still need to find a relation to satisfy f (23d). The comparative
clause supplies this function (23e). The comparative clause is interpreted as a function f
that maps people to their height. Combining (23d) and (23e), at the top of the
comparative clause (23 f), we derive a set of degrees of height by matching each person in
the context to their height.

The set of degrees derived from the comparative clause is operated upon by the

maximality operator, as shown in (24):

(24) Max [1D [Dom (f) = People & [Vd (deD—>3x(xeDom(f) & [d=tall(x)]))] &

[Vx(xeDom(f)—>Eld(deD& [d=tall(x)]))]]]

The Max operator will pick out the tallest height from the set. This is a welcome result
because we have achieved the correct semantics and avoided the QR of the quantifier at
the same time.

A question that immediately arises is that in Chierchia’s original analysis, the wh-
operator does not have to leave a functional trace; it can leave an individual trace and
correspondingly have an individual answer. This means that one possible alternative
structure for (23) is (25), in which the degree operator only leaves an individual trace

behind.

(25) John is taller than [cp Op,- [everybody else is dj-much tall]]

37

With the pair-list reading unavailable in (25), and the QR process non-applicable, (25)
gives rise to the incorrect narrow scope reading of everybody else, i.e. everybody else is
the same height. Since this reading is not the natural reading of (1), we are forced to the
conclusion that, different from other wh-constructions, in (23) the degree operator has to
leave a firnctional trace. A closer look at the semantics of gradable adjectives supports
this conclusion.

Gradable adjectives are generally argued to denote functions from individual to
degrees along certain dimensions (Seuren 1973, Cresswell 1976, Kennedy 1997, among
others). Take an adjective R, it provides a fimction that maps an individual to a degree on
a scale that R specifies, as in (26). For example, if the adjective is tall, it provides a
function fi'om individuals to degrees on a scale of height, as in (27).

(26) R (x, d)

(27) tall (x, (1)
So a degree cannot be an individual type entity, because its identity is always dependent
on a certain individual and a certain scale. It is also well known that the comparative
scale is a relative scale within certain comparison class (Klein 1982). For example, the
scale of height for human might be different from the scale of the height for giraffes, so a
tall man is probably still shorter than a short girafle. In other words, there is no individual
degree of gradable adjectives in the world, there are only degrees of individuals along
certain dimensions. So, the trace left by a degree operator can not be an individual trace
either. It has to be a functional trace and its interpretation is fixed by applying a measure

function to the individuals in the domain of that function. This explains why the sentence

38

John is taller than everybody else is has an obligatory pair-list reading, hence obligatory

wide scope reading of the quantifier on the surface.

2. 4. More evidence for the pair-list readings in comparative clauses

In the previous section, I demonstrated that applying the analysis of pair-list
readings in Chierchia (1993), we can account for the wide scope interpretations of
quantifiers in comparative clauses. In this section, I will show that this analysis not only
derives the correct interpretation of quantifiers, it also gives a unified analysis of a range
of different phenomena, including the subject-object asymmetry of quantifiers, the
behavior of monotone decreasing quantifiers at subject position, and the negative island
effect in comparative clauses. The latter two phenomena have been discussed in
Rullmann (1995) and given a semantic account based on maximality, I will show that the

present account is more adequate to cover the relevant facts.

2.4.1. Subject-Object Asymmetry

The idea that the unexpected wide scope interpretation of quantifiers in
comparative clauses is induced from its wh-construction properties is not new. Moltrnann
(1992) already tried to draw a parallel between quantifiers in comparatives and
quantifiers in wh-constructions. She noticed that, like wh-questions, quantifiers in
comparatives demonstrate subject-object asymmetries. For example, if we compare (1),
repeated in (28), where the quantifier is in the subject position of the than-clause, with
(29) where the quantifiers occupy the object positions of the than-clause, different scope

interpretations arise.

39

(28) John is taller than everybody else is.
(29) a. More students read a book than read every newspaper.
b. More students read only this book than read many articles.

0. Fewer students read this book than read most articles.

As discussed earlier, everybody in (28) is interpreted as having wide scope interpretation
over the than-clause. (28) is interpreted as for every x, John is taller than x is. On the
other hand, the quantifiers in (29) seems to be interpreted as having narrow scope with
respect to the than-clause. Consider (29a). The wide scope and the narrow scope

interpretations of every newspaper are true under the following scenarios respectively:

Scenario A: (wide scope interpretation)
The relevant newspapers are: New York Times, Wall Street Journal and
Washington Post. For each of them, the number of students that have read a book
exceeds the number of students that have read that newspaper.

Scenario B: (narrow scope interpretation)
The relevant newspapers are: New York Times, Wall Street Journal and
Washington Post. The number of students that have read a book exceeds the

number of students that have read every newspaper.

40

Native speakers’ intuition seems to accept scenario B, but reject scenario A. For (29b)
and (29c), although less straightforward, native speakers still only get the narrow scope
interpretation of the quantifier.

This patterns the same as the subject-object asymmetry observed in wh-questions.
A universal quantifier in subject position can induce a pair-list reading, but quantifiers in

object position can not, as shown in (30).

(30) a. --Who does everybody like?

«John likes Edgar, Bill likes Kary. ..

b. «Who likes everybody?

-- *John likes Edgar, Bill likes Kary. ..

Similarly, as shown by Sharvit (1999), in relative clauses, only quantifiers in subject
position can license indirect binding; quantifiers in object position can not, as shown in
(3 1).

(31) a. The woman who everybody; loves invited him;.

b. ‘The woman who loves everybody; invited him;.

Chierchia (1993) explains the absence of the pair-list reading in (30b) as a Weak
Crossover (WCO) effect. To have a pair-list reading the wh-operator in both (30a) and
(30b) needs to leave a functional trace behind, as shown in (32a) and (32b). For (32a), in

order to bind the domain part of the firnctional trace, the subject quantifier everybody

41

needs to undergo absorption to adjoin to the wh-operator to form a complex operator, as
shown in (32b). (32b) can be interpreted as a function f that maps people to people they
like, and the domain of f is fixed by everybody. Exactly the same process needs to be
applied to (33a) in order to derive the pair-list reading. However, when everybody moves
from the object position to adjoin to the wh-operator, it crosses over its bindee, i.e. the
domain part of the functional trace. This is reminiscent of a typical WCO violation, as
shown in (34). Since (33b) is an illegitimate representation, a pair-list reading is

unavailable because the domain of the function can’t be fixed by everybody.

(32) a. Who; everybody; like t9
b. [[Who; everybody;] [t; like :9 1]
(33) a. Who; :9 likes everybody;

b. [[Who; everybodyfl t9 1i1ceJ t;]]

/\

(34) * Who; does his; mother love t;
l

X l

 

The same reasoning can be applied to quantifiers in comparatives (29), examples
repeated as (3 5).
(35) a. More students read a book than read every newspaper.
b. More students read only this book than read many articles.

c. Fewer students read this book than read most articles.

42

Consider (35a) again. If (3 5a) were to have a pair-list reading, it would mean that for
each newspaper, the number of students who read that newspaper is less than the number
of students who read a book. That is to say, we can find pairs of newspapers and the
corresponding number of readers, and (35a) is true if for each pair we find, there are more
students who read a book. To derive this reading, (35a) should roughly have a structure
like (36), in which there is a firnctional trace lefi by the degree operator, and the domain
of the function is fixed by every newspaper.

(36) More students read a book than [cp Op; d;’-many students read [every newspaper];]

For (36) to make sense, we face exactly the same problem as (33), i.e. when every
newspaper obligatorily undergoes movement in order to bind the domain part of the
functional trace, it inevitably commits a WCO violation.

Therefore, to avoid the WCO violation, the only legitimate representation of (35a)

is for the degree operator to leave an individual trace, as shown in (37):

(37) More students read a book than [cp Op; d;-many students read every newspaper]

The quantifier every newspaper in (3 5a) cannot get wide scope interpretation through the
pair-list reading, therefore it is interpreted as having a narrow scope within the than-
clause. It is worth pointing out that in my earlier discussion in section 3, I claimed that
degrees of gradable adjectives are not of individual type, and that is why pair-list

readings (hence wide scope interpretations) are the only available interpretations for

43

quantifiers in the than-clause. However, as shown in (3 7), the degree of cardinality can
leave individual type trace, and therefore the quantifiers in the than-clause could have
narrow scope interpretations. At this point, the contrast between adjective degrees and
cardinality degrees is only motivated based on empirical data fiom comparatives. This
contrast will become relevant again for the discussion of negative islands in section 2.3,
and there I will demonstrate that the contrast between the two kinds of degrees with

respect to their semantic types can be supported on independent grounds'.

2.4.2. Monotone Decreasing Quantifiers in Subject Position
Further evidence for the present analysis comes from the fact that monotone

decreasing quantifiers in the subject position make comparatives ungrammatical.

(38) a. *John weighs more than nobody weighs.
b. *John weighs more than few people weigh.
c. *John weighs more than less than five people weigh.

d. *John weighs more than at most five people weigh.

Rullmann (1995) explained (38) in terms of maximality. Recall that the
maximality operator operates on the set of degrees generated from the comparative clause,

and picks out the maximal one. Therefore for (38a) to be true, John needs to weigh more

¥

I The proposed distinction between adjectival degrees and cardinal degrees suggests that the sentence
below should be ambiguous, which the cardinality degree takes either the wide or the narrow scope.
However, it is not obvious that the example below has the narrow scope reading that everybody else read
the same number of books. Unfortunately, I have no further suggestion for this problem.

(1) John read more books than everybody else did.

44

than the maximal degree that nobody weighs. Obviously this does not make sense
because we cannot construct a maximal degree out of a set of degrees that nobody weighs,
since that set does not have an upper bound. The other examples in (3 8) are ruled out for
the same reason, i.e. when a monotone decreasing quantifier is in the subject position, we
can’t construct a maximal degree out of the comparative clause.

This is an elegant solution, but it cannot be completely correct, because it cannot
cover cases fiom phrasal comparatives. In contrast with clausal comparatives in (38),
monotone decreasing quantifiers in phrasal comparatives are fine, or at least much more

improved, as shown in (39).

(39) a. John weighs more than nobody.
b. John weighs more than few people.
c. John weighs more than less than five people.

d. John weighs more than at most five people.

If, as Rullmann argues, monotone decreasing quantifiers cannot satisfy the maximality
requirement, it is a puzzle why such quantifiers are fine in (39). To the extent that (3 8)
and (39) should have almost the same semantic interpretations, the contrast between them
is very likely a result of their syntactic difference, i.e. the standard degree in (3 8) is
instantiated by a wh-clause, but in (39) it is instantiated by a phrase.

Under the current analysis, it is easy to explain how negative quantifiers are ruled
out by the wh-clauses they are in. It is known that wh-constructions in general pose

certain constraints on negative quantifiers. For example, in (40) and (41), when the

45

subject is nobody or at most two people, we cannot answer the question with a pair-list

answer.

(40) a. Who did nobody like?
*John didn’t like his teacher, Bill didn’t like his boss...
b.. Who did at most two people love?

*John loves his teacher, Bill loves his boss...

In Chierchia’s (1993) analysis, a domain that is extracted from the subject quantifier is
crucial to derive the pair-list reading, because a pair-list answer (i.e. a set of ordered pairs)
can only be derived by applying a function to each member of the domain. For quantifiers
like nobody and at most two people, the minimal witness set of them only contains an
empty set. If there is no member in the domain, pair-list readings are unavailable. Now if
as I suggested, the clausal comparatives in (3 8) involve pair-list readings, as shown in

(41), Chierchia’s solution will rule out the data in (38) correctly.

(41) *John weighs more than [Op; nobody; weighs d3]
As for phrasal comparatives (39), since there is no clause in the comparative site,
functional traces and pair-list readings become irrelevant and hence pose no constraints

on the quantifiers.

2.4.3.Negative Islands

46

2.4.3.1. The negative islands effect in comparatives

The last piece of evidence to support the pair-list reading in comparatives comes
from the negative island effect in comparatives. Negative islands refer to negations
contained within the comparative site, which usually yield ungrammaticality, as shown in
(42):

(42) a. *John weighs more than Bill doesn ’t weigh.

b. *John weighs more than Bill never weighed.
Rullmann (1995) again appeals to maximality to explain (42). Since the set of degrees
which Bill does not weigh has no upper bound, we cannot find a maximal degree of that
set either, so (42a) has to be ruled out, and (42b) is ruled out for the same reason.

The proposal I sketched is consistent with (42) too. The negative island effect is
reminiscent of the weak island effect known for wh-extractions. Weak islands refer to
islands that block some, but not all wh-extractions. Negation in wh-questions is a typical
weak island. As shown in (43), extracting a who-argument out of negation is fine, but not

a how-adjunct.

(43) a. Who didn’t you invite?

b. *How didn’t you behave?

Szabolcsi and Zwarts (1993, 1997) argues that wh-phrases that range over individuals,
e.g. who, are immune to weak islands, but wh-phrases that range over partially ordered
domains, e.g. properties, amounts and manners, are sensitive to weak islands. In other

words, this implies an extraction of non-individual types is subject to the weak island

47

effect, whereas an extraction of an individual type is not (also see Homstein 1995, Munn
2000). This predicts that a wh-question that contains a weak island cannot have pair-list
answers, since pair-list answers crucially hinge upon the existence of functional traces,
whereas a legitimate extraction out of a weak island has to leave an individual type trace

behind. This prediction turns out to be true, as shown in (44):

(44) Who didn’t everybody like?
a. Mary

b. *John didn’t like his boss, Bill didn’t like his student...

The lack of the pair-list answer in (44) provides an explanation for (42). As I

proposed earlier, the examples in (42) have the structures given in (45):

(45) a. John weighs more than [Op; Bill; doesn’t weigh dlj]

b. John weighs more than [Op; Bill; never weighed as]

The trace bound by the degree operator is not an individual type trace, instead, it is a
functional trace. Also following the traditional assumption that degrees are points on a
scale, and a scale is a partially ordered domain, so (42) can be fully explained as a weak
island effect.

Now both Rullmann’s proposal based on maximality and the current analysis

based on pair-list readings can account for the negative island effects in comparatives. Is

48

there a way to distinguish the two? In the next section, I will turn to a possible distinction

between the two.

2.4.3.2. A diagnosis from cardinality degrees

So far both the present account and the maximality account derive the same result
for the negative island data. Building on the analysis in Szabolcsi and Zwarts (1997), for
the present proposal, what is at stake is the semantic type of the variable left within the
negative island. The examples in (42) are ungrammatical because the variables within the
negative island are not e type variables, which are expected from the property of gradable
adjectives. On the other hand, what is important for the maximality account is whether
we can pragmatically construct a maximal degree from the context. The following
examples offer some empirical support for the former proposal.

Lechner (2002) noticed some systematic exceptions to negative island effect in

comparatives. For example, (46) is fine.

(46) Mary read more books than she didn’t read.

Although it seems like the match between the subject of the comparative clause and the

matrix subject is relevant for the grammaticality of (46), the following examples provided

by Lechner shows that is not correct.

(47) a. *Mary is taller than she isn’t.

b. *Mary read a longer book than she didn’t read.

49

c. *Mary read more poetry than didn’t read.

The descriptive generalization Lechner reached is that the negative island effect is
alleviated only if the comparison relation operates on degrees that keep track of
cardinality (as in d-many books), but not if these degrees cannot be mapped to the
individual count domain (d-tall in (47a) and d-long in (47b) or mass amounts d-much
poetry in (47c)).

Lechner’s observation immediately poses a problem for Rullmann’s proposal, in
which the maximality operator is not sensitive to the type of degrees generated from the
comparative clause. 1 will come back to the challenge that the maximality account has to
face. Now let’s consider first why the cardinality degree in (46) could survive the
negative island.

It has been observed that how-many questions can be ambiguous sometimes (e. g.

Heycock 1995):

(48) How many people did Mary decide to hire this year?
a. What is the cardinality n such that Mary decided to hire n-many people?
b. What is the cardinality n such that there is a set of n-many people that Mary

decided to hire?

The (a) interpretation refers to the situation where Mary has a number n in mind, maybe
as the plan of the year, and she needs to hire n-many people. She could have decided the

number it even before interviewing anybody. The (b) interpretation refers to the situation

50

where afier the interview, Mary decided on a set of qualified applicants, and the speaker
asked about the cardinality of that set.
However, when embedded within a weak island, only (b) interpretation survives.

(see Heycock 1995, Cresti 1995, Munn 2000, among others).

(49) How many pe0ple did Mary wonder whether to hire this year?
a. *What is the cardinality n such that Mary wondered whether to hire n-many
people?
b. What is the cardinality n such that there is a set of n-many people that Mary

wondered whether to hire?

(48) has both interpretations, whereas the weak island in (49) blocks the first
interpretation.

Different from how-many questions, which sometimes survive weak-islands as
shown in (48b), how—much questions and degree questions in general do not survive

weak-islands.

(50) *How much wine do you wonder whether I drank?

(51) *How difficult did you wonder whether she was?

Following Frampton (1990), Dobrovie-Sorin (1992), Szabolcsi & Zwarts (1993,

1997) and Cresti (1995), I will assume that an extraction survives weak-islands only if

the trace variable ranges over individuals. Under these proposals, the how-many

51

quc
aml

5111'

till
182

CO"

lili

pm

[hi
no
it;
til
car

OUI

l 2;

questions could leave either higher type traces or individual type traces, causing the
ambiguity in (48). Only the interpretation with the individual type trace, i.e. (49b),
survives in the environment of weak islands. On the other hand, how-much questions and
degree questions do not leave individual type traces, and consequently do not survive the
weak islands.

From the above discussion, cardinality degrees are different fi'om degrees of mass
amount and gradable adjectives with respect to their ability to escape weak islands. Since
this seems to be true in general cases, it is not surprising that we can extend the same
reasoning to comparatives. Lechner’s puzzle is therefore explained in a way that is
consistent with the present analysis of quantifiers in comparatives.

How would the maximality account deal with Lechner’s puzzle? It is true that for
the grammatical case (46), native speakers share the intuition that the sentence
presupposes a set of salient books in the context. This means that when we partition the
set of books into two groups, one is the set of books that Mary has read, and the other is
the set of books that she has not, there is an upper bound on the number of books she has
not read. In this sense, the maximality account can successfully rule in (46). The problem
with this account is that it predicts whenever there is a presupposed set in the context,
which makes it possible to construct a maximal degree from the comparative clause, we
can avoid the negative island effect in comparatives. This is not entirely consistent with

our data. Consider (47) again, data repeated in (52).

(52) a. *Mary is taller than she isn’t.

b. *Mary read a longer book than she didn’t read.

52

0. *Mary read more poetry than she didn’t read.

It is difficult to construct a scenario for (52a) where a set of heights is presupposed. But it
is not difficult to presuppose the relevant sets for (52b) and (52c). Consider the following
scenario. Every week Mary’s professor assigns a list of five books for the class to read.
Each student can pick any one out of the five books to read and do a book review. Last
week Mary read book A. This week the assignment is the list of books {B, C, D, E, F},
and Mary read book B. It happens to be the case that book A is longer than C, D, E or F.
Under this scenario, since the books that Mary didn’t read this week are {C, D, E, F}, it is
possible to construct a maximal degree of length from this set, namely, the longest book
in the set. Therefore the maximality constraint is satisfied. There is also nothing unnatural
in the pragrnatics that can rule out our hypothesized scenario. However, it is still

impossible to utter (53), which is slightly modified from (52b) to suit our scenario:

(53) *Mary read a longer book last week than she didn’t read this week.

We can create a similar scenario for (52c) as well. Suppose Mary’s interest is 20"1 century
American poetry, since we approximately know how much 20th century American poetry
work there is, it is possible to construct an upper bound for the amount of poetry that
Mary did not read. It is clear that even under this reasonable scenario, (52c) is still not an
acceptable utterance. These examples suggest that although maximality could be relevant
for quantifiers to avoid negative island effect in comparatives, it is not sufficient to

explain all the data.

53

Parallel to the maximality account for comparatives, those wh-NPs that are easily
extractable from weak islands, such as which person, who, what, can be argued to be D-
linking (discourse linked) DPs. However, as pointed out in Szabolcsi and Zwarts (1993,
1997), D-linking is relevant only insofar as it facilitates individuation by providing
contextually salient items, but ranging over individuals, rather than ranging over
contextually salient items, is the critical factor in extraction. Quoting Romanian data from
Dobrovie-Sorin (1992), they illustrate that how-many questions can escape weak islands
even if they range over non-D-linked individuals, which are not morphologically marked
by clitic doubling. They also give another piece of evidence with wh-the-hell phrases.
Given an appropriate context where individualization is available, the clearly non-D-
linked wh-the-hell phrases can escape weak islands too. For example, (54) can be uttered

in a context where someone was seen “madly searching through the dictionary” (p.262):

(54) What the hell do you still not know how to spell?

For (54) to be felicitous, there does not have to be a discourse linked set of entities in the
context. What is important is the understanding that the person is searching for a
particular item among a set of individual entities.

In conclusion, we have seen that the maximality account is not an adequate
account to explain the negative islands effect in comparatives, whereas the present

analysis based on pair-list readings can capture all the patterns.

54

2.4.3.3.Summary

To summarize, by treating the wide scope interpretation of quantifiers in
comparatives as a special case of the general properties of quantifiers in wh-constructions,
we can explain the wide scope interpretations easily. Since for gradable adjectives, the
degree variables are obligatorily functional, which means the pair-list reading is
obligatory, we also account for the obligatory wide scope interpretation for free. In
addition, this analysis connects together a few seemingly different phenomena, including
the subject-object asymmetry problem, the negative quantifier and the weak island
problems. In the next section, I will show that the contrast between the clausal and

phrasal comparatives with respect to quantifiers also naturally follows from the present

proposal.

2.5. A few words about phrasal comparatives

I have shown that by analyzing the wide scope reading of everybody in clausal
comparatives as pair-list readings, we can naturally account for a few previously
separated problems, among which is the constraint on monotone decreasing quantifiers.
Now let us turn to the other half of the problem we laid out in the beginning of the
chapter—phrasal comparatives.

Recall that everybody in phrasal comparatives gets wide scope interpretation, just
as in clausal comparatives; however, phrasal comparatives license monotone decreasing

quantifiers that are disallowed in clausal comparatives. The relevant data are repeated

below.

55

(55) a. John is taller than everybody.

b. John is taller than nobody/few people/at most five people.

Since under the current analysis pair-list readings depend on wh-constructions, and
phrasal comparatives do not involve wh-constructions, it is obvious that the wide scope
interpretation of everybody in phrasal comparatives has to be derived in a different way,
i.e. it is not derived from pair-list readings. If that is the case, it is not surprising that
monotone decreasing quantifiers are licensed in (55), because in this analysis, pair-list
readings and the constraint on monotone decreasing quantifiers are two effects of the
same mechanism.

The straightforward solution for (55) is QR. Wh-exaction from a phrasal
comparative construction is fine, as shown in (56). This suggests that QR movement of
the quantifier in principle should be allowed to derive the wide scope interpretation, as

shown in (57?.

(56) Who is John taller than?

(57) [everybody else; [John is taller than t;]]

2.6. Conclusion
In this chapter, I argue that the unexpected wide scope interpretations of
quantifiers in comparative clauses are actually pair-list readings. This supports the idea in

Moltmann (1992) that quantifiers in comparative clauses behave the way they do because

56

comparative clauses are wh-constructions. I also show that my solution has advantages
over the interval degree approach in Schwarzschild & Wilkinson (2002) because the
current approach can tie a few different phenomena together including some contrast
between phrasal and clausal comparatives. It is worth noting, however, although the
current chapter argues for an alternative approach to the interval degree semantics, it is
only to address the problem of quantifiers in comparatives. Some recent research in other
domains has shown that an analysis of degrees as intervals could be motivated on
independent grounds (for example, Kennedy 1999, 2001). If the current analysis is on the
right track, we have more evidence to believe that the structural distinction between

phrasal and clausal comparatives should be maintained.

¥

2 .One potential problem for the QR approach is that one has to force the obligatory QR in (57), since
Similar to clausal comparatives in (l), the quantifier in (57) takes obligatory wide scope interpretation over
the than-phrase.

57

Chapter 3

NPI in Comparatives

3.0. Introduction

As discussed in the Introduction, intuitively a comparative construction consists
of four elements: a reference degree, a standard degree, a comparative relation provided
by the comparative morpheme, and a scale of comparison provided by the gradable
adjective. In this chapter I will focus on the function of the scale that is introduced by
gradable adjectives. All comparatives need at least one gradable adjective. This makes
the use of a scale easily available to comparatives, since gradable adjectives introduce
scales for free. However, the semantic consequence of an inherently built scale has not
been firlly explored in the literature. In this chapter I will make use of the scalar property
of comparatives to explain the licensing condition of negative polarity items (N PIs) in
comparatives. In a nutshell, I will argue that NPIs are licensed in comparatives through
scalar implicature.

NPIs are lexical items whose distributions are constrained by certain licensing
conditions. Any is the typical example of an NPI in English. Since negation is the
classical environment that licenses these lexical items, they are usually called negative
polarity items. For example, as shown in (la) and (1b), affirmative statements don’t
license any, but negative statements do. Negation is not the only environment that
licenses NPIs. For instance, questions and conditionals, among many others
environments, also license NPIs, as shown in (lo) and (1d) (see Haspelmath 1997 for a

cross-linguistic survey).

58

(1) a. *He saw anybody.
b. He didn’t see anybody
c. Did you see anybody?

(I. If you saw anybody, let me know.

On the surface NPIs are ambiguous between existential indefinite and universal
interpretations. For instance, in the above examples, anybody is interpreted existentially.

In (2), it is interpreted as universal quantifier, and in (3), it is ambiguous between the two.

(2) a. Anybody knows that.
b. Anything will work.

c. Anybody can solve that problem.

(3) Would you do anything to help me?

Comparatives constitute another environment that license NPIs, as shown in (4a).

Moreover, the NPIs in comparatives are unambiguously interpreted as universal. For

instance, (4a) is best characterized as (4b), not (4c).

(4) a. John is taller than anybody else.

b. John is taller everybody else.

59

 

that
mm
mm
C611
Olll

uni

res
ass
qu;
dis
Tilt
0pc

(19‘,

c. John is taller than somebody.

In this chapter, we will look at how NPIs are licensed in comparatives. The fact
that NPIs are licensed in comparatives is a problem because as we will see next,
comparatives don’t have the general properties that are commonly assumed to be
necessary to license NPIs. Therefore to explore how NPIs are licensed in comparatives
certainly will add to our understanding of NPIs in general. Besides the licensing issue,
our story also needs to explain how and why the existential indefinites are interpreted as
universals in comparatives.

One clarification before we proceed. It’s known that items like anybody can be
interpreted existentially, as in (1), or universally in some contexts, as shown in (2). Under
the universal interpretations, these items are often called free choice items (F CI). Free
choice items on the surface look like universal quantifiers, hence it seems reasonable for
them to be treated as fundamentally different from NPI indefinites. So could it be the case
that anybody in (5) is actually a PC], not a NPI, and we therefore are asking a wrong
research question here? Following Giannakidou (2001) and Lee & Horn (1994), I will
assume that the categorical distinction between so called NPIs and FCIs in terms of their
quantificational force (NPIs are existential and FCIs are universal) is not the correct
distinction in the first place. Rather, both share the same source as existential indefinites.
The surface distributional and interpretational difference arises from different kinds of
operators in the context that bind the indefinites. Moreover, quoting data from Rullmann

(1995), Zepter (2003) argues that since comparatives also license other NPIs that are not

60

conl

C011

15)

lic

CO

3.1

3.1.

confounded as possible FCIs, as shown in (5), we still have the same problem that why

comparatives should license any NPI at all.

(5) a. He told me more jokes than I cared to write down.

b. He said the sky would sooner fall than he would budge an Inch.

Therefore, through out my discussion, I will treat anybody in (4) as a real NPI. The
problem we face then is to explain how they are licensed in comparatives and why they
have universal interpretations.

The organization of this chapter is as follows. In section 1 I will lay out what
problems comparatives pose for previous analyses of NPI licensing, and introduce some
basics about Chinese NPIs. Since the analysis I will give is closely related to dou, a
Chinese functional morpheme, I will give an analysis of dou in section 2. On completely
independent grounds, dou is shown to be an operator that picks out the maximal point on
a context dependent scale. Section 3 is our analysis of NPI licensing in comparatives,
both in Chinese and in English. I will argue that at least for comparatives, NPIs are
licensed through scalar implicature. Chinese dou explicitly spells out this licensing

condition for NPIs. Section 4 is the conclusion.

3.1. The problems with NPI

3.1.1. Comparatives as non-DE environments

61

In Ladusaw (1979) (see also Hoeksema (1983), Van der Wouden (1994), among
others,) NPIs are characterized as being licensed in downward entailing (DE)

environments.

(6) Downward entailing function:
A function f is downward entailing iff for every arbitrary element X and Y, it

holds that X ;Y —-> f(Y) g f(X)

According to this definition, the downward entailment function reverses entailment
relations. Consider negation as an example. In an affirmative statement (7), since car is a
superset of red car, the entailment in (7a) is valid, but not vice versa. When we add
negation to it, the entailment is reversed, as in (8).
(7) a. John bought a red car. —) John bought a car.

b. John bought a car. —l-) John bought a red car.
(8) a. John didn’t buy a red car. L) John didn’t buy a car.

b. John didn’t buy a car. —-> John didn’t buy a red car.

Rullmann (1995) and Hoeksema (1983) have argued that comparatives are indeed
DE environments, and that is why they license NPIs. However, as Schwarzschild &
Wilkinson (2002) point out, comparatives are not qualified as DE environments
. according to the definition in (6). If comparative clauses were downward entailing, we
should see a general pattern of reversing entailment relations, since that is the basic

characterization of downward entailing environment. That means that given the

62

entailment relation in (9), the reversed entailments in (10) should be true, contrary to fact.
On the other hand, in a real downward entailing environment like negation, the reversed

entailments are fine, as shown in (11).

(9) a. Exactly 7 of my relatives are rich. —> At least 4 of my relatives are rich.
b. (Given that there are elephants in this room) Almost every elephant in this
room is heavy. —> Some elephant in this room is heavy.
b. Most of the high tech stocks were overvalued. —> At least 2% of the high tech

stocks were overvalued.

(10) a. John is richer than at least 4 of my relatives were. A John is richer than
exactly 7 of my relatives were.
b. My car is heavier than some elephant in this room is. \9 my car is heavier than
almost every elephant in this room is.
c. Nissan is currently more overvalued than at least 2% of the high tech stocks
were. /—) Nissan is currently more overvalued than most of the high tech stocks

were.

( l l) a. It isn’t true that at least 4 of your relatives are rich. —) it isn’t true that exactly 7
of your relatives are rich.
b. It is impossible that some elephant in this room is drunk. ——> It is impossible that

almost every elephant in this room is drunk.

63

c. He never admitted that at least 2% of the tech stocks were overvalued. —> He

never admitted that most of the tech stocks were overvalued.

Cross-linguistic data also suggests that the DE fails to capture the correct
licensing condition for NPIs in comparatives. Since the property of being a DB
environment is a semantic property, if comparatives are qualified as DE environments,
we expect them to be DE environments as well in languages other than English. This
prediction isn’t borne out. For example, Chinese comparatives don’t license NPIs, as

shown in (12):

(12) a. *Ta gao shei yidian
he tall who a little
He is taller than anybody.
b. *Ta bi shei gao yidian
he bi who tall a little

He is taller than anybody.

3.1.2. Comparatives are not non-veridical
Based on Greek data, Giannakidou (1998, 1999, 2001) proposed (non)veridicality
sensitivity as the licensing environment for polarity items (also see Zwart 1995, and Lin

1996 for a similar idea based on Chinese data). As a subset of polarity items, the

64

distribution of NPIs are argued to be sensitive to (non)veridicality too. The approximate

definition of (non)veridicality is given as below (Giannakidou 1999 p3 84):

(13) The definition of (non)veridicality:
Let Op be a monadic propositional operator. The following statements hold:
(1) Op is veridical just in case Op p —-> p is logically valid. Otherwise, Op is
nonveridical.
(ii) A nonveridical operator is antiveridical just in case Op p —> —.p is logically

valid.

According to this definition, normal affirmative statements are obviously veridical. If
John went to school is true, then we know John went to school. Therefore, assuming there
is an implicit veridical operator, the following logical representation is valid. This is also

why affirmative statements never license NPIs cross-linguistically.

(14) Opp->19

John went to school. —9 John went to school.

On the other hand, negation is the typical antiveridical environment, as shown below:

(15) notP-) —1P

John didn’t go to school. -) —1 John went to school.

65

Negative environments in general license NPIs, as shown in (16):
(16) John didn’t see anybody.

English NPI any can also be licensed in a looser environment, the non-veridical

environment. For instance, conditionals.

(17) ifP -\ P

if John went to school 9 John went to school.

By saying if John went to school, one can’t guarantee the statement John went to school
is true or false, hence conditionals are non-veridical, which license English any in

conditionals:
(18) If John goes to any place, he will call me before he leaves.

English any can be licensed in other non-veridical environments too, such as restrictions
of universal quantifiers, imperatives, polar questions, etc. Therefore we can say that the
licensing condition for English any is non-veridicality.

Just like the DE story, although (non)- veridicality can capture a good amount of
the NPI data, it doesn’t seem to work for comparatives. As shown earlier, English

comparatives license NPIs, but it is not obvious that comparatives are non-veridical.

66

 

As

Ct":

 

 

 

 

 

W0

Ch

3.}

pi

Assuming an implicit veridical operator, the following logical statement is true for
comparatives, and it actually suggests comparatives are veridical.
(19) Op 9 -> 9

Op John is taller than Bill/anybody. —) John is taller than Bill/anybody.

To summarize, since Chinese comparatives don’t license NPIs, we have reason to
believe that there is something else more than the semantic property of comparatives that

works here. Before I pursue this line firrther, let me first introduce some basics on

Chinese NPIs.

3.1.3. Existential Polarity Wh-phrases (EPW) In Chinese Comparatives

The closest Chinese counterparts to English any are existential polarity wh-
phrases (EPWs, borrowing the term from Lin (1996)). As the name suggests, these
polarity items are wh-words in Chinese. Their distribution overlaps with the English NPI
any in some environments, but also overlaps with the English PPI some in some other
environments. So strictly speaking, they are polarity items, instead of negative polarity
items. Also as shown by the name EPW, when used as polarity items, these wh-words are
interpreted as existential indefinites. As far as I know, Lin (1996) gvies the best survey of
EPWs up to now. In this section I will just briefly report his discussion.

EPWs are polarity items because they are sensitive to the environment in which
they can occur. They can’t occur in simple affirmative sentences, as shown below:
(20) a. *shei/shenme ren xihuan Mali.

who/what person like Mary

Somebody/anybody likes Mary.

67

 

 

'Zl

b. *wo xihuan shei/shenme ren.
I like who/what person

I like somebody/anybody.

EPWs are licensed in typical environments that license the NPI any (e.g., negations,

conditionals, yes-no questions)

Negation
(21) W0 mei zuo shenme.
I not do what

I didn’t do anything.

Similar to English any, the EPW isn’t licensed if it is out of the scope of negation.

(22) *Shei mei qu xuexiao.
Who not go school

Somebody/anybody didn’t go to school.

If-conditionals
(23) Ruguo ni you shenme wenti, wo hui bang ni.
If you have what problem I will help you

If you have any problems, I will help you.

68

Yes-no cgrestions
(24) Shei qifir ni le ma?
Who bully you perf. Q

Did anybody bully you?

(25) Ni xie-guo shenme ma?
You write-perf. what Q

Have you written anything?

On the other hand, EPWs can also be licensed in environments that don’t allow any, but

allow some.

Imperatives
(26) a. Nimen shei qu bang ta yixia.
You who go help him once
Somebody/*anybody lend a hand to him.
b. Ni fangjia de shihou qu na-er wanwan ba.
You vacation de time go where play PAR.

Go somewhere/*anywhere (to relax) when you have vacation time.

Uncertainty modality (possible, perhaps, must have)

 

(27) a. Keneng shi nage lingdao lai-le.

Possibly is which official come-perf.

69

 

 

fat
Ch
mls

the

3.1.

Possibly some/*any official came. (otherwise the traffic wouldn’t be that bad)
b. Yexu ni yinggao qu xue xian shenme shouyi.

Perhaps you should go learn little what skill

Perhaps you should go to learn something/*anything.
c. Ta yiding shi zai shenme difang chu-le cuo.

He must is at what place appear-perf. mistake

He must have made a mistake somewhere/*anywhere.

Other condition_aI_s (unless, as long as)
(28) a. Chufei shei lai jiu ni, buran ni si ding le.
Unless who come save you, otherwise you die sure perf.
Unless somebody/*anybody comes to save you, otherwise you are finished.
b. Zhiyao shei ling-ge-tou, women jiu dou qu.
As-long-as who lead we then all go

As long as somebody/*anybody goes first, then we will all go.

I have only given a much simplified picture of the facts discussed in Lin (1996). But the
facts above are enough to show that EPWs are indefinite polarity items in Mandarin
Chinese. Although their distributions don’t overlap with English NPIs completely, the
mismatch doesn’t concern us too much with respect to comparatives. For simplicity of

the discussion, in the rest of this chapter, I will loosely call Chinese EPWs NPIs.

3.1.4. The Interaction between dou and NPIs In Chinese comparatives

70

 

Bl

CC

As shown earlier, neither bare-comparatives nor bi-comparatives license EPWs,

data repeated below:

(29) a. *Ta gao shei yidian
he tall who a little
He is taller than anybody.
b. *Ta bi shei gao yidian
he bi who tall a little

He is taller than anybody.

Both examples above could have an interpretation as the wh-question who is he taller
than ?, but neither is grammatical under the polarity item interpretation. This suggests that
comparatives are not intrinsically DE environments.

Without changing the truth conditions, we can license the NPIs in comparatives
by adding a firnctional morpheme dou. However, as shown in (30), this only rescues NPls
in bi-comparatives, but not in bare-comparatives.

(30) a. Zhangsan bi shei dou gao yidian.
Zhangsan than who all tall a little
Zhangsan is a little taller than anybody.

b. *Zhangsan dou gao shei yidian

Zhangsan dou tall who a little

Zhangsan is a little taller than anybody.

71

(30a) is equivalent to its English counterpart, as shown by the English translation. The
existential indefinite wh-phrase who is interpreted as if it was a universal quantifier. The
bare comparative (30b), on the other hand, is unacceptable].

NPIs in Chinese comparatives are interesting for two reasons. First, they clearly
show that comparatives themselves don’t share the DE property with other NPI licensing
environments, such as negation and conditionals. An immediate question is how NPIs are
licensed in English comparatives then. Chinese comparatives could shed some light on
this issue because there is an explicit morpheme we can work on, the morpheme dou.
Here is our working strategy. I will assume that whatever mechanism dou employs to
license NPIs in Chinese comparatives is the same mechanism that works in English
comparatives. The only difference is that in Chinese the licensing condition is spelled out

through dou, but in English, it is implicit.

3.1.5. Summary and outline

To summarize, the problem we need to solve in this chapter is this: how to license
NPIs in comparatives such that the existential indefinite gets a universal interpretation.
The approach I am going to take is to start from the Chinese data, and then extend the
analysis to English. Since an understanding of dou is crucial to solve the problem, I have
to make a digression and spend quite some time in the next section to give a detailed
analysis of dou. After that, I will apply the analysis of dou to NPI licensing in

comparatives.

3.2. An analysis of Chinese Dou

 

I (30b) is acceptable if it is interpreted as a wh-question.

72

3.2.1. An introduction

Don is a very productive functional morpheme in Chinese. In general it has three
usages: as a distributor, as even, and as already. The distributor interpretation has been
widely discussed, whereas the already use seems to not attract much attention. In this

section I will introduce the basic data of each usage.

3.2.1.1. Dou -— the distributor
As shown in (31), dou obligatorily distributes the predicate over the subject
argument. This is in contrast with (32), which is ambiguous between a distributive and a

non-distributive (collective) reading.

(31) Tamen dou chi-1e yi- ge- pingguo pai
They Dis. eat-perf. one-CL-apple pie
They each ate an apple pic.

(32) Tamen chi-1e yi- ge- pingguo pai

They eat-perf. one-CL-apple pie
a. They each ate an apple pie.

b. They ate an apple pie (together)

Since dou is used as a distributor, the argument it distributes over has to be distributable.

This explains the ungrammaticality of (3 3).

(33) *Ta dou chi-1e yi- ge- pingguo pai

73

he Dou eat-perf. one-CL-apple pie

he each ate an apple pie.

Following Lin (1998), I will temporarily give dou the following semantic representation,

which is similar to the English all.

(34) dou :>APAXVy[yeX—>P(y)] where X is a variable over plural individuals and y

is a variable over singular atomic individuals.

According to (34), dou distributes the property P to each member of the plurality object
X. For example, in (31), dou operates on the property of eating an apple pie. This
property is distributed over each member of the relevant group, and we get an
interpretation and each of them ate an apple pie.

Although dou is similar to English all in meaning, it doesn’t behave exactly like
all. One significant difference between them is that dou can distribute over anything that
is distributable, including DPs that are explicitly marked by universal quantifications; but
as a universal quantifier, all can’t operateon DPs that are already universal. This contrast

is shown by the following examples.

(35) a. Suoyoude ren dou lai-1e.
all people dou come-perf.
All the people came.

b. *All the people all came.

74

(36) a. Tamen quan dou lai-1e.
They all/completely dou come-perf.
They all came.

b. "' They all all came.

Dou also doesn’t behave exactly like the English distributor each. As shown in (37) and
(3 8), each can’t distribute collective predicates, presumably because each distributes
down to the atomic individual members of the plural DP, but the collective predicate on
the other hand can’t be true for a single individual. Dou, however, is compatible with

collective predicates.

(37) a. Tamen dou shi pengyou.
they dou be friend
They are friends.

b. *they each are fiiends.

(38) a. Tamen dou jian-guo—mian
they dou meet-perfi-face
They all met before.

b. *They each met before.

Two more properties of dou we need to keep in mind is that first, dou has to stay

to the left of the predicate, as shown in (39):

75

(39) a. Tamen dou lai-1e.
They dou Come-perf.
They all came.
b. *Tamen lai-1e dou.
They come-perf. dou

They all came.

Second, dou can only distribute an argument to its left. Therefore if it is the object that is
being distributed over, the object has to be scrambled. This is usually called the Leftness

Condition.

(40) a. *Ta dou yao-le mei-ge-pingguo yi-kou

he Dou bite-perf. every-CL-apple one-bite
He took a bite on each apple.

b. Ta mei-ge-pingguo dou yao-le yi-kou
he every-CL-apple dou bite-perf. one-bite
He took a bite on each apple.

c. Mei-ge-pingguo, ta dou yao-le yi-kou
every-CL-apple, he dou bite-perf. one-bite

He took a bite on each apple.

76

3.2.1.2. Dou — even
The even use of dou is most clearly shown in the (lian). . .dou construction. The

morpheme lian is optional, but it has been suggested that lian, or lian...dou together,

marks focus (Shyu 2004, Paris 1998).

(41) a. (Lian) shagua dou zhidao zhege.

(Focus) idiot dou know this
Even idiots know this.

b. (Lian) Zhangsan dou qu-le.
(Focus) Zhangsan dou go-perf.
Even Zhangsan went.

c. Ta (lian) zhoumo dou zai gongzuo.
he (focus) weekend focus prog. Work

He even works on weekends.

As shown by their English translations, these examples all contain a certain scalar
implicature induced by the focus even. (41a) implies that the issue under discussion is an
easy one. Informally this implicature is derived in the following way. There is a
contextually determined scale that orders different people along certain dimensions.
There are probably different dimensions we can choose to elaborate upon, and I will
choose a scale that orders our expectations of different people to solve the problem. If
John is the smartest person in the group, it is expected that he knows how to solve the

problem. If Bill is less smart, it is less expected that he would know how to solve the

77

problem. In this way, we can order people from least expected to the most expected. By
common sense, idiots would be the least expected to solve the problem and they stay at
the bottom of the scale. Notice that we can also say idiots are ordered on the top of the
scale if the ordering is from the least unexpected to the most unexpected. The choice of
the exact scale might be arbitrary, but in either way, idiots is on the end of the scale. The
other two examples have the same end of the scale implicature. (41b) implies that it is
very unexpected that Zhangsan went, and (41c) implies that it is very unexpected that

somebody would work on weekends.

3.2.1.3. Don — already
This use of dou as already is the least discussed in the literature, although it is

very productive. The following are some examples.

(42) Don ji dian ne? (ni zenme hai mei shui.)

dou what time Q you how still not sleep

What time is it already? (How come you haven’t gone to bed yet!)
(43) (Yizhuanyan,) haizi dou da le.

in a blink, children dou grown perf.

(time flies!) In a blink of time, the children have already grown up.
(44) Liuyue dou guo wan le. (Zenme hai zheme leng.)

June dou pass finish perf. How still this cold

It is the end of June already. (How come it is still this cold!)

78

Although it seems to be translated as already, dou in this usage shouldn’t just mean
already for two reasons. First, in all the examples above, one can add an explicit already
to it. One example is given in (45). If dou already means already, it is not clear why an

extra already is allowed.

(45) Don yijing ji dian ne? ni zengme hai mei shui.
dou already what time Q you how still not sleep

What time is it already? How come you haven’t been to bed yet!

Second, all the examples above need to occur in particular contexts, which don’t
constrain already as much. For example, based on the translation what time is it already,
the speaker of (42) seems to ask a simple question, and expect an answer of a certain
time. But this is not what (42) means. It is true that the hearer could answer 2 o ’clock in
the morning, however, asking a question is not the point for the speaker. It might well be
the case that the speaker knows exactly what time it is, or he doesn’t care about the exact
answer. The message that the speaker tries to convey is simply that it is very late now,
you should have gone to bed. It is very unexpected at this time to see you awake. Because
of this implication of unexpectedness, the first part of (42), namely, the question part, has
to be uttered in a context that the speaker has a certain expectation of the timeline of
some event, and in (42), it is the timeline of going to bed. Without the context, an
utterance out of the blue like this would be very awkward. For (43), if we just say that the
children have already grown up, it could be a simple statement of the fact. By adding dou

to the sentence, the speaker expresses certain emotion about time and life. Here is a

79

scenario that (43) would be uttered. Thirty years ago, a group of young people in their
early twenties graduated from university, and they had a reunion thirty years later. It is
amazing and emotional to everybody how thirty years have gone by without notice. What
are the things that can reflect the fast time change? You could have changed a couple of
jobs, or married a couple of times, or traveled all over the world. If there is a scale that
can order all your different experiences together, for most people a grown up next
generation is probably the best witness for the time change. Our last example (44) also
expresses the “end of the scale” kind of implication. Normally we expect after March, the
weather starts getting warm, and at the end of May, it starts getting hot. So on a scale of
expectation of cold weather, the end of June is relatively at the end. Since the relevant
scale is context dependent, one can utter (44) in other environments too. For example, if
John is writing his dissertation, and the plan for June is to finish one chapter, but at the

end of June, he hasn’t written a word yet, then he could say (46):

(46) Liuyue dou guo-wan le. (wo hai mei kaishi xie)
June dou pass-finish perf. I still not start write

It is the end of June now. (But I haven’t start writing)

From this discussion, it is clear that dou in its third use is closer to even than to
already, because sentences with dou are not just simple statements, they convey end of
scale implications. Note that under some analyses, already also expresses certain scalar

implicatures (Michaelis 1996, van der Auwera 1993), and this might be why sometime

80

dou can be translated as already. A detailed discussion of the semantics of already is

beyond the scope of this thesis, and I will leave it for some other occasion.

3.2.2. The analysis
After surveying the three main uses of dou in the last section, in this section I will lay out

a unified semantics of dou that covers all three uses.

(47) The following are the main ingredients of the analysis:
0 Don operates on events
0 Don
(i) Presupposes: a set of alternative events which are ordered on a
contextually determined scale

(ii) Asserts: the event on the top of the scale is true

By positing that dou operates on events, we naturally explain the syntactic
constraint on the position of dou, i.e. dou has to stay to the immediate left of the
predicate. In the literature, dou has been analyzed as the VP level adverb (Cheng 1995),
or a functional head that takes VP as its complement (Wu 1999). Since the difference
between the two is not crucial for our discussion here, I will adopt the analysis in Cheng
(1995) without further justification.

The scalar part of the analysis makes dou similar in meaning to English even. It is
worth reviewing the basic assumptions with regards to the meaning of even. The main

contribution of the focus marker even is to introduce a presupposition. It presupposes that

81

there is a set of alternatives, which are introduced by the focused element (Rooth 1985,
1992), that are compared with the even-associated argument along a likelihood scale, and
the even-associated argument is the least likely on the scale. Let’s illustrate this idea with

a concrete example.

(48) a. Even [John]; was invited.

Presuppose: among all the guests (Mary, Bill, Chris. . .), John was the least
likely to be invited.
Assert: John was invited.

b. John even invited [Bill];.
Presuppose: among all the guests (Mary, Bill, Chris...) that John invited, it is
the least likely that he would invite Bill.
Assert: John invited Bill.

0. John even [brought a present]; to Bill.
Presuppose: among all the things John can do for Bill, bringing a present for
Bill is the least likely to happen.

Assert: John brought a present to Bill.

The focused element in (48) is marked by a subscripted f. In (48a), the theme argument
John is focused. An alternative argument is the argument we can use to substitute John
and still make the sentence true. Suppose Mary, Bill and Chris are the alternatives
provided by the context. The focus marker even presupposes a likelihood scale that orders

John and other alternatives, as shown in (49), and John is at the bottom of the scale,

82

namely, John is the least likely to be invited. In (48b) it is still the theme argument, but in
the object position, and in (48c) it is the predicate that is focused, but the presuppositions

are derived in similar ways.

(49) Mary

Bill decreasing likelihood to be invited

 

5..
O
3'

/

\

Although similar to even in the sense that both dou and even involve scales in
their meaning, dou is different from even in at least two respects. First of all, dou operates
on events, and its syntactic position is relatively restricted. English even, on the other
hand, assumes a flexible syntactic position when it associates to different focuses of a
sentence, as shown in (48). Secondly, even is often associated with the likelihood scale,
but as I will show later, the scale related to dou is unspecified and is largely determined
by the particular context. In addition to these distinctions, there seems to be some
difference between the two in terms of their semantic contributions, but I will not go into
more detailed discussion here. As mentioned earlier, even only contributes to the
presupposition part, but not the truth condition, of the sentence. This is clearly shown in
(48). For all the examples there, the presence of absence of even does change the
presuppositions, but doesn’t change the truth conditions. In contrast to this, the presence
or absence of dou sometimes changes the truth conditions. This is mostly clearly shown
by the distributor use of dou, as we will see next, where an originally ambiguous sentence

can be disambiguated towards the distributive interpretation by dou.

83

3.2.2.1. Dou — distributivity as the end of a quantity scale

My analysis of dou will draw heavily on the previous research of pluralities in
both the nominal and the event domains (see Link 1983, Bach 1986, Lasersohn 1995,
Schwarzschild 1996 among others). Let me briefly sketch the main system that will be
crucial for the analysis.

It is often observed that distributivity and cumulativity are connected. The
utterance in (50) is ambiguous between a distributive reading and a cumulative reading.
Under the distributive reading, each person bought a house. Following the analysis of
plural nouns in Link (1983, 1998), this event can be taken as a plural event (Bach 1986)
because it contains plurality of sub-events in it (Lasersohn 1995), namely, the sub-events
of Mary’s buying of a house, John’s buying of a house and Bill’s buying of a house. Each
of the sub-events can’t be divided any further, because there couldn’t be an event of half
of Mary bought a house or Mary bought half of a house. Therefore, each sub-event is an
atomic event and under the distributive reading, (50) expresses three atomic events. The
cumulative reading, on the other hand, only contains one atomic event, namely, the

buying of a house that involves three agents at the same time.
(50) Mary, John and Bill bought a house.
a. Mary, John and Bill each bought a house. (distributive)

b. Mary, John and Bill bought a house together. (cumulative)

Interestingly there exist intermediate readings that are not completely distributive

or completely cumulative. (51) is a classical example from Gillon (1987).

84

(51) The men wrote musicals.

Suppose the DP the men denotes the set containing Rodgers, Hammerstein, and Hart.
There is at least one reading that (51) is true if Rodgers and Hammerstein wrote musicals
together, and Rodgers and Hart collaborated together. That is to say, the plural event
expressed in (51) contains two atomic events el and e2. e1 is an event of musical writing
that involves Rodgers and Hammerstein, and e2 is an event of musical writing that
involves Rodgers and Hart. Under this interpretation, there is no strict distributivity that
each of the three people wrote musicals independently, and there is also no strict
cumulativity that three people collaborated to write musicals.

Gillon (1987), Higginbotham (1981) and Schwarzschild (1996) discuss the idea of
a cover to account for all three readings above. A cover is a partition of the plurality P if

and only if the following criteria are met (Schwarzschild 1996, p64):

(52) C is a cover of P if and only if:
(i) C is a set of subsets of P
(ii) Every member of P belongs to some set in C.

(iii) Q is not in C

For example, if P is a set that contains four members: John, Matt, Bill and Chris, we
could have different ways to partition this set into smaller subsets. In other words, we

could have different covers. The following are some possibilities:

85

(53) C1: { {1.M.B.C}}
C2: {11. M}, {B, C}}
C3: {{1}, 1M. B.C}}

Cn: {shim}. {B}. {Cl}

Moreover, Schwarzschild proposed the following interpretive principle (revised from

Gillon 1987, Higginbotham 1981):

(54) [s NPpm; VP] is true iff there is a cover C of the plurality P denoted by NP such

that VP is true for every element in C. (p64)

According to (54), how we interpret the sentence the men wrote musicals depends on
how we choose the cover C. If we choose C1, that is we are taking the group of men as
one plural object, since there is only one element in C1, by applying (54), the predicate
wrote musicals is true for this plural object. The sentence would mean that John, Matt,
Bill and Chris collaborated to write musicals. This is the collective reading. On the other
hand, if we choose Cn as our cover, there are four elements in Cn, and the predicate
wrote musicals is true for each of the four elements, we end up having a distributive
reading that each person wrote musicals independently. The intermediate readings arise if
we choose intermediate covers. As shown in (53), there are many possibilities of
intermediate readings because there are different ways to partition the set of four men. If
we choose C2 as our cover, the sentence is true if John and Matt are collaborating, and
Bill and Chris are collaborating. If we choose C3, the sentence is true if John is working

by himself, and Matt, Bill and Chris are collaborating, and so on and so forth. Notice that

86

under this analysis, all these interpretations involve universal quantification, which is
restricted to the contextually determined cover over the domain instead of the domain
itself. To give a more general analysis to all these interpretations, we can adopt the
generalized D-operator proposed by Schwarzschild. Suppose that or is a variable over
predicates, D is a distributive operator, x and y are variables over the relevant domain in

the discourse, the claim in (54) suggests the following analysis:

(55) X 6 ll D(C0V)(0t )ll iff

||Cov|| is a cover ofx AVy [yellCovll —> yellotll]

The cover ||Cov|| is a set of sets, and its value is assigned by contextual (including non-
linguistic) factors. The generalized distributor D doesn’t operate directly on the atomic
individuals in the domain, instead it operates on the members of the cover set.

Now let’s turn to our analysis of dou. Lin (1998) analyzed dou as a generalized
distributor because in the context of collective predicates, dou doesn’t distribute down to

the atomic individuals in the domain. One example is given below:
(56) Naxie ren dou shi fuqi
those people dou be husband-and-wife

Those people are all couples.

Suppose now the set those people has a denotation of Mary, Bill, Ann and Chris. Because

the predicate being a couple can only be true for a pair of people, it can’t be distributed

87

over each single member of the set. The set has to be partitioned first, i.e., we need a
cover to work on. Suppose our cover contains pairs {M, B} and {A, C}, dou as a
generalized distributor can distribute the predicate being a couple into each member of
the cover. In this respect, dou is different fiom the real distributor each in English. Each
is incompatible with collective predicates, because its inherent distributive property
makes it operate upon atomic members of the set, not the covers. Although this analysis
gives the correct interpretation of (56), I suspect it at most lends additional empirical
support for the idea of a context-dependent cover, because in this case obviously the
lexical property of the predicate excludes any cover that doesn’t partition the set into
member-paired sets, but it doesn’t suggest that dou is the explicitly marked generalized
distributor D. For dou to be the generalized-D, it has to be able to operate on different
covers, as long as no other factors can rule out those covers (e.g. contextual factors). Lin

uses the following example to prove this to be the case.

(57) Xiaoming, Xiaohua han Dabao dou shi tongxue
Xiaoming Xiaohua and Dabao Dou be classmate

Xiaoming, Xiaohua and Dabao are all classmates.

The predicate being classmates again can’t be true for an atomic individual argument, but
it is true if the three people is taken to be a plural object. That is to say we can have a
cover that contains one member in it, as shown in (58a). Under this interpretation, the
three people are classmates at the same time. Lin noticed that there is another possible

plurality cover, as in (5 8b), which contains three pairs. If the predicate being classmates

88

is true for each pair in (58b), we should have a reading that X.m and X.h are classmates,
X.h and D.b are classmates, and D.b and X.m are classmates. Crucially, there is no
obvious reason to force the three classes to be the same. The interesting question is if
there is an interpretation that the three pairs are in different classes. This interpretation is
arguably present. Therefore, Lin (1998) took the two interpretations in (58) as showing
that dou can distribute over members of different plurality covers, therefore it is a

generalized distributor.

(58) a. C1: {{Xiaoming, Xiaohua, Dabao}}

b. C2: {{Xiaoming, Xiaohua}, (Xiaohua, Dabao}, {Dabao, Xiaoming}}

There are two problems with this argument. First, I don’t think the ambiguity in
(58) is real. It is possible that the interpretation under C2 would entail the interpretation
under C1. With a plurality cover C2, I agree with Lin that it is unspecified if the three
classes should be at the same time or not. But if they are, the reading under which is
pragrnatically equivalent to that under Cl, because if a and b are classmates of class A,
and b and c are also classmates of class A, then it is true that a, b and c are all classmates
to each other. Second, there exist other possible covers for this case, but the
interpretations under those covers don’t seem to be available. For example, we can

partition the set of three people into two pairs, as shown in (59):

(59) C3: {{Xiaoming, Xiaohua}, {Xiaoming, Daobao}}

89

If dou can operate on this cover as a generalized distributor, we should have a reading
that Xiaoming and Xiaohua are classmates, Xiaoming and Daobao are classmates. If the
two classes are different, it is possible that Xiaohua and Daobao are never classmates.
However the original sentence (57) would definitely be false under this context.

To analyze dou as a generalized distributor also fails to explain the commonly
observed distinction between sentences with dou and those without it. The following is
such an example. With dou present, (60) only has a distributive reading, which says each
person bought a car. Without dou, (61) is ambiguous between a distributive and a
collective reading. According to (55), it is the cover, which in turn is determined by
contextual factors, not the generalized distributor itself that determines the available
interpretations. Since (60) and (61) are not different with respect to the context in which
they occur, we are led to the conclusion that dou is not the explicitly marked generalized
distributor in Chinese, because dou has to do something to contribute to the semantic

difference between the two.

(60) Zhangsan he Lisi dou mai-le yi—liang-che.
Z. and L. dou buy-perf. one-CL-car
Zhangsan and Lisi each bought a car.

(61) Zhangsan he Lisi mai-le yi-liang-che.

Z. and L. buy-perf. one-CL-car
a. Zhangsan and Lisi each bought a car.

b. Zhangsan and Lisi bought a car together.

90

In agreement with Lin (1998), I reject the analysis that dou is a simple distributor, as
English each. I instead suggest that dou is a scalar operator that operates on events
ordered on a quantity scale. To entertain this analysis, we need to take a closer look at the
idea of cover again. As mentioned earlier, a cover is a set of sets derived from partitions.
Using our earlier example, a set with four members John, Matt, Bill and Chris, can be
partitioned in different ways. In other words, we can have different covers for the plural

argument John, Matt, Bill and Chris.

(62) C1: { {J, M, B,C}}
C2: {0. M}. {3. Cl}
C3: {0}. {M, B. C1}

CH! {{1}, {M}. {3}, {Cl}

Some of these covers will be excluded on the basis of contextual concerns. For
descriptive simplicity, I will just assume that all these covers are possible under certain

contexts. According to (54), we can make the following assumption.

(63) [GP John, Matt, Bill and Chris VP] is true iff there is a cover C of the plurality

argument denoted by the subject such that VP is true for every element in C

Let’s assume VP is not a collective predicate for the moment, for example, buy a car.
Since buying a car could be true for one individual argument, an atomic event of buying
a car should contain just one individual argument. If we choose C1 in (62) as our cover,
we derive one event of buying a car. Let’s call this event E1. E1 is a singular event

because the only sub-event it contains is El itself. Crucially for our purpose, E1 doesn’t

91

contain any atomic sub-events, since the only sub-event it contains involves four
individuals as its argument. If we choose C2 as our cover, since the VP needs to apply to
each member of C2, we will end up having a plural event E2, which contains two sub-
events el and e2. The first sub-event e1 has the set {J, M} as its agent, and the second
sub-event e2 has the set {B, C} as its agent. Again, there isn’t any atomic sub-event.
Under the cover C3, however, there is an atomic sub-event, the sub-event that has the
individual {J} as its argument. Repeating this process, it is not hard to see that with
different covers, we could derive different events, with respect to how many atomic sub-
events they have. When we move on to the cover Cn, we are going to have a plural event
with four atomic sub-events, which is the largest number of atomic sub-events possible.
In (62), the cover C1 gives us the collective reading, and C11 gives the distributive
reading, because under the former cover, the predicate is true for everybody together as a
whole, and under the latter cover, the predicate is distributed down to each individual
member. There has always been an intuition that collective and distributive readings are
the two ends of a scale from the least distributive to the most distributive readings, and
intermediate readings are in the middle part of the scale (van der Does & Verkuyl 1996).
Along this distributivity scale, the presence of dou forces the top of the scale reading, i.e.
the most distributive reading. Different from Lin (1998), dou is not the implicit D in (4),
instead, dou operates on the representation in (4), and picks out the cover that will yield
the most distributive event. I will informally call this kind of cover a distributive cover,

e. g. the cover C11 in (62). Adding dou to (55), we can have the following representation:

(64) x ell Dou D(Cov)(or )u iff

92

||Cov|| is a distributive cover of x AVy [ye ||Cov|| ——> ye ||or||]

Let’s now turn to the situations where the predicate in (63) is collective, for

example, being classmates, or meet each other, data repeated in (65).

(65) a. Tamen dou shi tongxue
they dou be classmate

They are classmates

b. Tamen zuotian jian-le-mian.
They yesterday meet-perfl-face

They met yesterday.

Since a cover is always constrained by contextual factors, it is not surprising that the
lexical properties of collective predicates impose certain requirements on the cover. For
example, being classmates, or meet each other, requires that each atomic event involves
two individuals, and an event with only one individual as its argument can’t be true. As a
result, if the relevant set has three members John, Mary and Bill, among the following
covers, covers like C2, C3, and C4 are ruled out since they contain sub-events that have
only one individual as the argument. Among the rest of the covers, Cn contains the
largest number of atomic events possible. dou picks the cover Cn, and (65a) means that
John and Mary are classmates, Mary and Bill are classmates, and John and Bill are

classmates, and (65b) means that John met Mary yesterday, Mary met Bill yesterday and

93

John met Bill yesterday. Importantly, since the cover Cn does not impose a time interval
on each event, so (65a) could be true in a situation where the three being classmates
events occurs at a different time, i.e., John and Mary’s class is different from Mary and
Bill’s class or John and Bill’s class. Similarly, the three meeting each other event in (65b)

could occur at different time too.

(66) C1: { {J, M, B, }}
C21 {{J, M}. {BJ}
C31 {11}, {M, B.}}
C41 {{J}. {M}. {B}}

(311.1: {{J, M}: {J, B}}
CHI {{J, M}. {M, B}. {J, 3}}
My analysis of dou bears some resemblance to Huang (1996), who analyzed dou

as a sum operator on events, as shown in (67):

(67) DOU (e, Pred) = U { CPREDI, 69115132, ePRED3-"i

e is an event of minimum size consistent with the semantics of the PRED.

The events of minimal size are also called minimal events in Huang (1996), which refer
to events that involve minimal number of arguments requirement by the predicate. For
example, dance only requires minimally one member for its argument, but a collective
predicate such as meet will require minimally two members for its argument. It is clear
that the intuition Huang had is close to the current analysis. In her analysis, distributivity
is achieved by letting dou Operate on the predicate and yield a plural event that contains

all the minimal events. Although both analyses account for the distributive reading of

94

dou, by employing the idea of a scale, the current analysis can be extended to the even
use and the already use of dou as well, which I will turn to next.

It is worth noting that although we have some evidence to show that dou and all
are not exactly the same (section 3.2.1.1.), the current analysis of dou does share some
similarities with a particular analysis of all, namely the analysis in Brisson (1998)2. In a
nutshell, Brisson argues that all is not a universal quantifier, contrary to the most
common assumptions, and she claims all is a modifier, an adverb specifically, which
modifies the generalized D operator (in the sense of Schwarzschild 1996) and
strengthens the distributivity. Assuming this analysis, both all and dou are non-
quantifiers, and both are adverbs that strengthen distributivity on the predicate. The
strengthening effect of all in Brisson (1998) is a little different fiom our analysis of dou.
All strengthens distributivity by excluding non-maximal interpretations, whereas dou
achieves the strongest distributivity through the end of scale effect. A detailed
comparison between the two analyses is beyond the scope or our discussion here and I

will leave it open for future work.

3.2.2.2. The analysis of the even use of Don

For the even use of dou, I will largely adopt the analysis in Portner (2002) with
some modifications. For the (lian)...dou construction, Portner’s basic idea is that dou
quantifies over a presupposed set of alternatives to the focused element. This proposal is

formalized as below:

¥

2 This is brought to my attention by Alan Munn.

95

(68) D [lian X [pred. . .dou. . .]], D is an implicit topical set of alternatives to X and X at
the extreme, end of a contextually given scale on D:
(i) asserts PRED (x)

(ii) implicates VxeD [Pred (x)]

Portner’s original analysis assumes dou is associated with the focused argument X. Since
we assume that dou operates on events, we need a slight modification on (68) to make it

suit the new assumption, as shown in (69):

(69) [lian X [pred. . .dou. . .]], D is an implicit topical set of alternatives to X and X at
the extreme end of a contextually given scale on D:

(i) asserts PRED (X)

(ii) implicates Vy(yeD) —)3e [Pred (e, y)]

(iii) De is the set of events derived in (ii) and the event PRED (X) is at the

extreme end of a likelihood scale on D,

Let’s illustrate this analysis with an example.

(70) (Lian) shagua dou hui zuo zhe-dao-ti.

lian idiot dou can do this-CL-problem

Even idiots know how to solve this problem.

96

In (70), the argument marked by lian is in the focus position. Following standard
assumptions, focused elements introduce a set of alternatives (Rooth 1985, 1992). The
word idiot in (70) introduces a set of alternative people in the context, for example, John,
Bill, Mary, etc. This set of people can be ordered along a scale of intelligence, and the
idiot by common sense is the least intelligent one, so it would stand at one end of the
intelligence scale. Applying the predicate knowing this problem to the set of people, we
derive a set of events. They are all events of knowing the problem, with different agents.
If we order these events on a likelihood scale, since the idiot is the least intelligent one,
by common sense his knowing of the problem is the most unlikely, or we can say, the
least likely event. In either way, the event that (70) expresses is on one end of the

likelihood scale.

3.2.2.3. The analysis of the already use of Don

As discussed in section 3.2.1.3, although dou seems to be translated as already
sometimes, it is not equivalent to already. I have provided two pieces of arguments for
that. First, dou and already can co-occur, and second, to be translated as already, dou has
to occur in a specific kind of context. The following data further supports the distinction
between dou and already. Let’s consider (71) in a scenario that we need 10 students to

open a graduate class, but there are only six people enrolled so far}.

(71) #a. You liu-ge xuesheng dou xuan-le zhe-men-ke,
there six-CL student dou enroll-pert“. this-CL-class
dan hai yuanyuan bu go
but still far-far not enough
Six students have already enrolled for this class, but it is far from enough.

 

3 Data modified fi'om Chen (2004).

97

b. You liu-ge xuesheng yijing xuan-le zhe-men-ke
there six-CL student already enroll-perf. this-CL-class
dan hai yuanyuan bu go
but still far-far not enough
Six students have already enrolled for this class, but it is far from enough.
In this scenario, with certain expectation being unsatisfied, dou is not acceptable in (7 la),

but already is acceptable in (71b). Interestingly, dou becomes acceptable if the number of

enrolled students exceeds 10.

(72) You shiwu-ge xuesheng dou xuan—le zhe-men-ke,

there fifteen-CL student dou enroll-perf. this-CL-class

ba jiaoshi zuo man-1e

BA classroom sit firll-perf.

Fifteen students have enrolled and the classroom is full.
This suggests that the use of dou is related to certain expectation standard. Along this
line, I will propose that Don is used to express the speaker’s subjective emotion, opinion
or expectations. Crucially, in the context when the utterance is made, the speaker could
have chosen from an alternative set of events to describe, but the event he/she finally
utters, is the one that at the point could best describe the speaker’s feeling or expectation.

This idea may sound vague so far. Let’s illustrate it with the high school reunion

example we discussed earlier. The data is repeated below:

(73) Shijian guode zheng kuai ya! Yizhuanyan, haizi dou da le.

time pass really fast a blink, children dou grown perf.

Time flies! In a blink of time, the children already grown up.

98

Recall that this sentence is uttered in a reunion party thirty years after the graduation. It
could be a very emotional occasion seeing that time has changed the teenage boys and
girls into middle aged men and women. To express one’s impression towards the
changing of time, i.e. e1 under our terminology, the e2 that the speaker has chosen to
evaluate e1, is that the next generation has grown up. During that last thirty years, many
events have happened, but the growing up of a new generation is probably the most
powerful one to serve the speaker’s purpose here, namely, it speaks most convincingly
about the change of time and life. In this sense, the e2 stands on the end of an evaluation
scale that orders the set of alternative events that the speaker could have chosen. There

could be different ways to state this proposal. The following is one way to do it.

(74) For the discourse [d;scom e1... D dou e2], D is an implicit set of alternative events
to e2, there is a scale on D that orders the events based on their effect to evaluate
e1, the purpose of the evaluation is determined by the context. E2 is on the end of

the scale, namely, it achieves the best evaluation effect.

We can also use the following ordered scale to express the same intuition.
(75) [Some changes have happened in our lives], time really flies.

Beyond expectation —> e1: The next generation has grown up
e2: The speaker has married twice expectaion
e3: The speaker changed different jobs scale

 

Let’s take a look at another example (76). The two events under discussion are e1:

I haven ’t started writing; and e2: June has passed. Recall that the background is that the

99

plan of the month of June is to finish one chapter. Suppose that it was only May, e1
would be judged as an expected event; if it was early June, e1 can be judged as
acceptable too. But the closer to the end of June, the more likely e1 will be judged as
unacceptable. To pick e2 as the passing of the end of June, the speaker has chosen the
strongest case that reflects the unexpectedness of el and possibly the disappointment of
the speaker. In this sense, the event that June has gone stands on the end of a scale that
orders the time passing events according to their effect to evaluate the expectedness of

the other event.

(76) Liuyue dou guo-wan 1e. (wo hai mei kaishi xie)
June dou pass-finish perf. I still not start write

It is the end of June now. (But I haven’t start writing)

Again, the following expectation scale can capture native speakers’ intuition:
(77) [It is time t at the speaking time], and I haven’t started writing my thesis.

Beyond expectation —> e1: It Is June
e2: It is March expectation scale
e3: It is Jan

 

 

The property of dou to mark the end of the scale event is also illustrated by the
degree constructions as below. The following sentences express meanings similar to

English so...that... constructions.

100

(78) a. Mali qi de dou ku le
Mary upset de dou cry perf.
Mary was so upset that she broke into tears.
b. Zhangsan deyi de dou wang 1e ziji jiao shenme
Zhangsan happy de dou forget perf. Self call what
Zhangsan is so happy that he forgot his own name.
0. W0 deng de taiyang dou xiashan le. ta ye mei chuxian.
I wait de sun dou go down perf. he still not show up
I have waited for so long that the sun went down, but he still didn’t show up.
All these examples describe two events. The second event expresses a high degree along
a contextually determined dimension that applies to the first event. The morpheme de
marks the degree construction. For example, in (78a), the crying event measures a very
high degree of Mary’s anger; in (78b), the forgetting his own name event measures the
extreme happiness of Zhangsan; and in (780), the event that the sun has gone down
measures extreme inappropriate long time he has let me wait.
Finally let’s reconsider the graduate course scenario. We can also explain the data

using the following scale:

(79) [There are n students enrolled], and we need 10 to open a graduate class4

n=20
(\lDou)Beyond expectation ‘{ n=15
(?Dou) Expected n=10
n=9

(*Dou) Under the expectation { n=8

 

 

4 See Chen (2004) for a more detailed analysis on the issue of presupposition of dou and distributivity.

101

When the expectation standard is explicitly set, as in this case, anything event that
expectation satisfies the requirement of dou.

To summarize, I have unified the three uses of dou through a scalar analysis. The
basic meaning of dou is immediately related to the end of a contextually determined
scale, which could either be a scale of quantity (i.e. the number of atomic events), or a
scale of expectation of some kind. This analysis not only captures our intuition about the
semantics of dou, it also shed some light on the syntactic condition dou poses on its
argument, which I will turn to in the next section.
3.2.2.4. The Leftness Condition

The leitness condition refers to the syntactic requirement that one argument in the
dou-construction needs to stay to the left of dou, although not necessarily adjacent to dou.
For example, in its distributor use, the DP that is being distributed over stays to the left of

dou. If the relevant DP is originally an object, it has to move to the left of dou.

(80) a. *Ta dou yao-le mei-ge—pingguo yi-kou
he dou bite-perf. every-CL-apple one-bite
He took a bite on each apple.
b. Ta mei-ge-pingguo dou yao-le yi-kou
he every-CL-apple dou bite-perf. one-bite
He took a bite on each apple.
0. Mei-ge-pingguo, ta dou yao-le yi-kou
every-CL-apple, he dou bite-perf. one-bite

He took a bite on each apple.

102

In the even use of dou, the focused DP is the argument that is constrained by the leftness
condition. Again, if the focused argument is the object, it needs to move to the left of

dou.

(81) a. *Ta dou bu langfei yifenqian
he dou not waste a penny
He doesn’t waste even a penny.
b. Ta yifenqian dou bu langfei.
he a penny dou not waste
He doesn’t waste even a penny
c. Yifenqian ta dou bu langfei.
a penny he dou not waste

He doesn’t waste even a penny

In the already use of dou, the leftrress condition doesn’t seem to apply. Imagine John and
Bill are in a game to compete drinking beer, and John drinks much faster than Bill. Both
examples below are grammatical:
(82) a. John dou he-wan yI tong 1e, Bill cai he dao dier bei.

John dou drink-finish one bucket perf. Bill only drink to second glass

John already finished a whole bucket, but Bill is only at his second glass.

b. John yi tong dou he-wan 1e, Bill cai he dao dier bei.
John one bucket dou drink-finish perf. Bill only drink to second glass

John already finished a whole bucket, but Bill is only at his second glass.

103

As shown above, although the object argument could be fronted, probably for focus
reasons, the presence of dou doesn’t force the fronting. For the moment, I will put aside
the case of this dou, and only focus on the first two uses of dou. As will be clear later,
once the nature of the leftness constraint is clear, it naturally follows why it doesn’t apply
to the already use of dou.

Under our scalar analysis of dou, the distributive and the even use of dou are
different outputs of the same core, which is that dou marks the top of a contextually
determined scale. Note that under the present analysis, dou itself is not quantificational5
(contra Cheng 1995). But the output of dou could be. For example, the distributive
reading could have a quantificational analysis, as reviewed earlier from Schwarzschild

(1996), and the basic analysis is repeated in (83).

(83) x ell dou D(Cov)(ot )|| iff

||Cov|| is a distributive cover of x AVy [ye ||Cov|| —) ye ||or||]

Recall that or is a variable over predicates, D is the generalized distributive operator, x
and y are variables over the relevant domain in the discourse, and the function of dou is
to pick out the cover that will give the distributive reading, because that is the reading

that contains the largest number of sub-events along a quantity scale. I will propose that

 

5 Careful readers might find this point seems to be inconsistent with the discussion in chapter two, where
we assume that dou is a quantificational element that induces the kind of quantifier barrier effect in Beck
(1996). This is not a problem because although dou itself is not a quantificational element, the distributive
structure is. There might be more than one ways to implement a revision for our earlier discussion in
Chapter 2. One possibility is that dou can mark the position of the implicit generalized distributor D.
Because D is quantificational, our arguments in Chapter can be maintained.

104

in this case the leftness condition holds because the universal quantifier has to take wide
scope over the existentially quantified event argument. To see this more clearly, we can

modify the representation in (83) to (84), which has the same interpretation:

(84) Suppose X is the distributee argument and e is the event argument
[cp X. .. dou D (Cov) (e)] is true iff:

(i). ||Cov|| is a distributive cover of X

00- W [0'6 NOW") 9 36 (600)]

As shown in (84), achieving the distributive meaning requires that the universal
quantifier, which quantifies over the atomic individuals of X, has to scope over the
existentially quantified event. It is known that in Chinese the scope relation is explicitly
marked by the surface word order (Aoun & Li 1993). If X in (84) happens to the object
argument, it has to scramble to the front to scope over the event.

We can extend the same scope analysis to the leftness condition in the even use of
dou. The basic analysis for the even use of dou is repeated in (85). What is crucial for our
purpose here is that through scalar implicature the universal quantification over the
members of the alternative set has to take scope over the existentially quantified event
argument. Again, assuming that the scope relation in Chinese has to be explicitly marked
by the surface word order, it is not surprising that the focused argument X, which

introduces the alternative set of arguments, has to scramble to the left of the predicate.

105

(85) [(lian) X [pred. . .dou. . .]], D is an implicit topical set of alternatives to X and X at
the extreme end of a contextually given scale on D:

(i) asserts PRED (X)

(ii) implicates Vy(yeD) —>Ele [Pred(e,y)]

(iii) De is the set of events derived in (ii) and the event PRED (X) is at the

extreme end of a likelihood scale on De

To summarize, the leftness condition takes place to satisfy the scope relation
between an argument and the event that this argument is part of. If this characterization is
correct, we have a tool to explain why the leftness condition doesn’t constrain the already
use of dou. Under the already use of dou, we are looking at two events in the discourse,
the event that dou operates on is the one that can best evaluate the other event along a
contextually specified dimension. We don’t need to front any argument because dou is

not operating on any individual argument. Consider (82) again, data repeated in (86):

(86) John dou he-wan yi tong 1e, Bill cai he dao dier bei.
John dou drink-finish one bucket perf. Bill only drink to second glass

John already finished a whole bucket, but Bill is only at his second glass.

(87) [Bill finished x-amount], and Bill is drinking his second glass.

Beyond expectation —) e1: John finished 2 bucket
e2: John finished 1 bucket expectation scale
e3: John finished 3 glasses

 

 

106

The alternative events that dou could have operated on, namely, John finished two
buckets, John finished one bucket, or John finished three glasses, etc, are evaluated along
an expectation scale with respect to another event, namely, the event that Bill is drinking
his second glass. Crucially different from the other two kind of cases I have just
discussed, these are two separate events, instead of an alternative set of arguments and an
event that these argument could be part of. Therefore, there is no scope relation at stake
here.
3.2.2.5 Summary

In this section we argued for two main points. First, the semantics of dou is to
specify the end of a scale. Second, the intended semantics of dou interacts with structural
constraints. Specifically, if the semantic output of dou is quantificational, it complies
with the general word order constraint on scope relations in Chinese, which leads to overt

movement when it is necessary.

3.3. Licensing NPI through a scale

In this section we are going to use the semantics of dou as a tool to poke at the
question of how NPIs are licensed in comparatives. Note that in Chinese dou can also
license NPIs in constructions other than comparatives. For instance, the wh-word in (88)

is interpreted as a polarity item, not an interrogative word.

(88) Ta shenme dou bu chi
he what dou not eat

He won’t eat anything.

107

Ideally we would like to see that when NPIs are licensed by dou, whether it is in
comparatives or not, dou works the same way. Cheng (1995) proposed an analysis to
explain how dou licenses NPIs in general. Although she didn’t discuss comparatives
specifically, probably she would carry over the same analysis to comparatives too. Let’s
take a look at her analysis first. After that, I will make a proposal to explain how NPIs are
licensed in Chinese comparatives, and extend this analysis to account for English
comparatives too. Finally, I will show that the current analysis fits nicely with the general
rationale of NPI licensing.

3.3.1. Cheng (1995)

The analysis in Cheng (1995) mainly consists of two steps: first, the polarity items
constructed fi‘om wh-words are variables that need a binder (Cheng 1991, Heim 1982,
Reinhart 1997); second, dou as a quantificational adverb binds wh-words and give them
universal quantification force. The binding takes place when at LF dou is at a position
that m-commands the wh-word.

My main opposition to this analysis is the assumption that dou carries universal
quantification force and that is the source of the universal quantification on NPIs. In
section 3.2.1.1. I have shown some evidence that supports a different position. The main
argument there is that dou is compatible with DPs that are already explicitly marked as
universal. If dou carries universal force itself, we would have a doubly quantified DP.

The relevant data are repeated below:

108

(89) a. Suoyoude ren Dou lai-1e.
all people Dou come-perf.
All the people came.
b. *All the people all came.
(90) a. Tamen quan Dou lai-1e.
They all/completely Dou come-perf.
They all came.

b. * They all all came.

Portner (2002) also provides some interesting and compelling evidence to argue
against the quantificational status of dou. Pomer noticed that if dou can bind indefinite
variables and give them universal interpretation (in the fashion of Heim 1982), we can’t
explain the contrast between the following two examples (examples are different from

Portner’s to form a minimal pair).

(91) a. Ta yi-fen-qian dou jinjinjijiao.
he one-penny-money dou calculate
He calculates even a penny (i.e. he calculates on everything)
b. *Ta yI-fen-qlan dou hua
he one-penney-money dou spend

He spends everything.

In both sentences we have the same indefinite a penny. If dou is the source for universal

109

quantification, it is puzzling why only in (91a) a penny is interpreted universally as
everything, but the same interpretation in (91b) is ungrammatical. Note that the only
difference between the two lies in the predicates. It suggests that universal force actually
is related to the meaning of the predicate. This intuition is fiirther confirmed by the
observation that a penny in (92) can be interpreted universally if we add negation to the

predicate, as shown below:

(92) Ta yi—fen-qian dou bu hua
he one-penney-money dou not spend

He doesn’t spend anything.

The puzzle is explained under our analysis of dou. Under the present analysis, dou
itself is not quantificational, it only picks out the end point of a contextually determined
scale. The output of the dou operation could be quantificational. In the case of (91a), by
common sense, calculating a penny could be the end point of a likelihood scale because it
is not expected for most people, when we look at how generous they are with money. If
this event on the extreme end of the likelihood scale is true, by inference we can conclude
that he calculates everything, i.e., we can draw an inference with universal
quantificational force. In contrast to this, spending a penny in (91b) is not the least likely
event we expect. (91b) is ungrammatical because the end-of-scale requirement of dou
isn’t satisfied. On the other hand, not spending a penny in (92) is a very unusual event.
Again, if this end—of-scale event is true, we can draw an inference that he doesn ’t spend

anything.

110

To summarize, if dou doesn’t carry universal quantification force, Cheng’s
analysis on how dou license wh-NPIs can’t be maintained. In the next section, I will

make a proposal that dou licenses NPIs through a scale.

3.3.2. Don and NPI licensing
Recall that dou only licenses NPIs in bi-comparatives, but not in bare

comparatives (data repeated in (93)):

(93) a. Zhangsan bi shei dou gao yidian.
Zhangsan than who all tall a little
Zhangsan is a little taller than anybody.

b. *Zhangsan dou gao shei yidian
Zhangsan dou tall who a little

Zhangsan is a little taller than anybody.

This is not unexpected because we have seen that as part of the leftness constraint, in
order to be licensed, the NPI has to stay to the left of dou, which is violated in (93b).
Therefore, because of the inherent syntactic property of the bare comparatives, they can’t
host dou to license NPIs. In the rest of this chapter, I will leave aside bare comparatives,
and concentrate only on bi-comparatives.

The basic idea I am going to pursue is that the indefinite NPI gets interpreted as a
universal quantifier through the end of the scale implicature. Specifically, since dou

contributes to the end of the scale implicature, the comparatives that contain a dou-

111

licensed NPI will be interpreted as superlatives. To see how this works, let’s start with an

example without an NPI.

(94) Ta bi Yaoming dou gao.
He bi Yaoming dou tall

He is even taller than Yaoming.

According to the semantics of dou, the sentence in (94) expresses three things.
First there is an alternative set of people compared to Yaoming, suppose they are {a, b,
c. . . }. If we apply the comparative predicate to these alternative arguments, we have a set

of alternative events, as shown in (95):

(95) {he is taller than a, he is taller than b, he is taller than c. . .}

Second, all these events, including the one that he is taller than Yaoming, can be ordered
on a likelihood scale. He is taller than Yaoming is on the end of the scale, i.e., it is the
least likely, or the most unlikely event. This is so because by common sense, Yaoming as
a NBA player, is much taller than the majority of the Chinese. Finally, dou picks out the
event on the top of the scale. That is to say, it picks out the event asserted in (95).

Since the presence of dou forces the assertion of the least likely event, both
examples below are ungrammatical, or at least very odd because the scalar requirement

isn’t satisfied.

112

(96) a. *John bi xiao haizi dou gao.
John bi little kid dou tall
John is even taller than little kids.
b. *John bi Yaming dou ai.
John bi Yaoming dou short

John is even shorter than Yaoming

(96a) is odd because it is not the least likely event that John is taller than little kids. The
fact is probably to the contrary that he is very likely to be taller than little kids. (96b) is
odd for the same reason. If by common sense Yaoming is much taller than average
height, instead of being the least likely event, the chance is high that John is shorter than
Yaoming. Therefore, in both case, the intended semantics of the sentences are in
contradictions with the semantics of dou.

Turning now to the NPIs in comparatives, data repeated in (97). We find that dou

works very much the same way for NPIs as for non-NPIs in (94).

(97) Zhangsan bi shei dou gao yidian.
Zhangsan than who all tall a little

Zhangsan is a little taller than anybody.

First of all, the licensed NPI has to move to the left of dou, as expected from the leftness

constraint. This is satisfied automatically because the surface structure of bi-

comparatives. The focus on the indefinite NPI induces a set of alternatives. The

113

semantics of dou makes sure that the comparative event associated with the NPI is the
least likely event. In the context of comparatives, the least likely event is a superlative.
For instance, if we compare the height of John with other people in a statement John is
taller than..., the least likely to happen is that John is taller than the tallest person in the
comparison set. If even this event is asserted to be true, by scalar implicature, any other
alternative event, which involves an alternative argument to the NPI, is true too. This is
semantically equivalent as saying for each member X in the alternative set, there is an
event such that John is taller than X. In this way the universal quantification over the set
of alternatives scopes over the existentially quantified event argument. The overt
movement of the NPI ensures the surface word order complies with the scope relation.
Our discussion up till now builds upon a simple logic: since the NPI licensing in
Chinese comparatives is closely related to the presence of dou, if we understand how dou
works, we gain some insights on how NPIs work in comparatives. To compile what we
have done so far, to license NPIs in comparatives and achieve universal interpretation

through the indefinite NPI, three criteria need to be met:

(i) There needs to be an enlarged set that contains more than just the NPI. In the
case of Chinese, and probably many other languages, we derive such a set by putting a
focus on the NPI, since focus induces an alternative set.

(ii) There needs to be a scale that orders all the events, namely, the event
associated with the NPI and the events associated with the members in the alternative set.

(iii) There needs to be a pointer that makes sure the event associated with the NPI,

i.e., the currently asserted event is the least likely event. In Chinese, the (ii) and the (iii)

114

points are achieved through the operation of dou.

If all of these are met, by scalar implicature, the indefinite NPI will give us a universal

interpretation.

3.3.3. NPI licensing in English comparatives

On the surface, the current analysis on Chinese comparatives is similar to Zepter
(2003) analysis of English comparatives. Although the basic idea is very similar, I will
argue that Zepter’s analysis can’t be maintained on its original form.

Zepter’s analysis is partly inspired by the analysis in Kadmon and Landman
(1993) (thereafter K&L), where they explored the pragmatic function of using NPIs.
K&L proposed that the semantic contribution of any is to reduce tolerance of exceptions,
i.e. to make a stronger statement than the non-any version. The following are the main

ingredients of their analysis.

(98) a. any NP = the corresponding indefinite a NP plus the additional
semantic/pragmatic characteristics (widening, strengthening) contributed by any.
h. WIDENING
In an DP of the form any NP, any widens the interpretation of the common noun

phrase (NP) along a contextual dimension.

115

c. STRENGTHENING

Any is licensed only if the widening that it induces creates a stronger statement,
i.e., only if the statement on the wide interpretation entails the statement on the
narrow interpretation.

Let’s illustrate their point with the example below.
(99) I don’t have any potatoes.

The DP any potatoes in this examples is treated like a normal indefinite DP. However,
any can widen the domain of the interpretation of potatoes in different ways, depending
on the context. For instance, consider a situation where you ask me if I can cook potatoes

for dinner, and I could utter either (100a) or (100b) as my reply:

(100) a. I don’t have potatoes.

b. I don’t have any potatoes.

Note that in this scenario, the default understanding is that only the set of potatoes that I
can cook for you should be relevant to your question. Therefore, even if I do have a few
rotten potatoes, or plastic potatoes, (100a) can still be considered to be true. On the other
hand, for (100b) to be true, I am implying that I don’t even have rotten ones, or plastic
ones. Therefore, any in (100b) widens the domain to include normal potatoes or
rotten/plastic potatoes. We can also consider a situation where you want me to cook
potatoes for 50 people. Now the quantity of potatoes becomes relevant. If I have one or

two potatoes in my kitchen, I can still utter (100a) as a reply to your request. However, if

116

I utter (100b), I am implying that I don’t have even one potato. In this case, any widens
the domain of potatoes to include the smallest quantities. Therefore the widening
associated with any is often related to some salient features of the context. In addition to
contributing a widened domain associated with the common noun, any needs to be
constrained by the strengthening requirement. Let’s still consider the example of potatoes

in (99). Since any potatoes in (99) is just an indefinite DP, (99) could be represented as

(101) —13x [potato (x) A I have x]

Alter any widens the domain, the strong/wide interpretation we have is I don ’t have
potatoes, neither cooking ones nor others. This entails a narrow interpretation that I don ’t
have rotten potatoes. The strengthening condition is satisfied, and any is licensed in (99).

On the other hand, if we use any in (95a), the strengthening condition will not be

satisfied.

(102) a. *I have any potatoes.

b. 3x [potato (x) A I have x]

(102a) can be represented in (102b). After any widens the domain of potatoes, the wide
interpretation of (102) is that I have some kind of potatoes, cooking or rotten ones, and
this doesn’t entail the narrow interpretation that I have cooking potatoes or I have rotten

potatoes. Therefore, any is not licensed.

117

The question for comparatives is that what is special about comparatives that will
make sure an NPI in it is interpreted as having a widened domain and a strengthened
meaning. Fauconnier (1975) noticed that the existential quantification over the highest
point of a scale will give rise to universal quantification. Elaborating on this idea,

Zepter (2003) argued that NPIs are interpreted as universals in comparatives because they
are interpreted as superlatives. Comparatives naturally provide a partially ordered scale.
An NPI can zoom in to the highest point on the scale. If John is taller than the highest
point on the scale, he is taller than everybody in the domain. For a concrete

demonstration of this analysis, let’s look at the following examples.

(103) a. John is taller than someone in the class.

b. John is taller than anybody else in the class.

In (96a) the normal existential indefinite someone is used, in (103b) the existential
indefinite NPI anybody is used, and only the latter gives a universal interpretation. Zepter
argues that similar to normal existential indefinites, an NPI carries an open variable that
gets existentially closed in the nuclear scope (Heim 1982). However, different from
normal existential indefinites, an NPI also carries a firnction 6 that restricts the domain of
the existential quantification over a set. The formal definition of an NPI by Zepter is

given below:

(104) The NPI any, ever are functions that restrict the domain of the existential

quantification over a set: AP [6(P)](x). (p217)

118

In the context of a comparative as (103b), the set of people are ordered on a scale by their
heights. The NPI function 8 not only widens (in the sense of Kadmon and Landman

1993, see below) the set from one person to all the people ordered by the scale, but also

zooms in to the highest end on the scale. As a result we have a semantic representation as

below:

(105) a. 5(people) = the highest person in the relevant context

b. John is taller than 3x [[5(people)](x)]

The representation in (105b) means John is taller than a person, who is the tallest in the
relevant context, hence by inference John is taller than everybody in the context.

This analysis explains NPI in comparatives without appealing to DB environment. It also
catches our intuition that comparatives involve scale of some sort and that gives a
universal interpretation from an existential indefinite. Although I agree with this intuition
and the current analysis also embraces the insight about a scale, I don’t think this is the
complete analysis. The problem is that the function 8 defined in ( 105a) seems somewhat
arbitrary. This function 8 not only widens the domain, it also picks up the highest end on
the scale. Although we can give a semantic definition to make 8 do both the widening

and zooming in jobs, we don’t know exactly where it comes from. It seems to be only
invented to operate upon NPIs such as any or ever in comparatives. However, we know

that neither the lexical properties of NPIs nor the semantics of comparative constructions

119

necessarily carry such a widening function. Otherwise Chinese comparatives should have
licensed NPIs automatically, without appealing to the presence of dou.

What is missing in Zepter’s analysis, therefore, is a more elaborated analysis of
the mysterious 8 function. 8 does two things: widen the domain to include other relevant
members, besides the NPI indefinite; and pick out the end of the scale point. Compare
this to Chinese, the first thing is achieved by focus, and the second thing is done by dou.
This suggests that a language probably has its own way to express the same functions that
8 is set up to do. Instead of proposing a very vague 8, we should be able to compose the
same semantics from specific lexical and syntactic expressions.

For this reason, I will adopt the analysis in Lee and Horn (1994) for English NPIs,
although it isn’t addressing comparatives specifically. Lee and Horn (1994) analyze NPI
and free choice any as semantically equivalent to the indefinite determiner a. Different
from the normal indefinite determiner a, the FC any contains an incorporated focus even,
which presupposes a scale in the discourse context. It is this incorporated even, which is
implicit on the surface, that is responsible for the limited distribution and the widened
domain (Kadmon and Landman 1993) of NPI and FC any. Notice that under this analysis
both NPI and FC any are analyzed as indefinite, which means even the universal reading
of the FC any is not analyzed as coming from universal quantification, instead, it is
achieved through the semantics/pragmatics of even.

Even associates with a focus and presupposes a contextually related scale, and the

DP associated with even usually refers to the lowest point on the scale. Consider (106):

(106) Even Bill ate the spinach.

120

(106) presupposes two things:

(i). Bill is the least likely person to eat spinach (e.g. he hates vegetables)

(ii). Everybody else in the context ate the spinach.

It is clear from the inferences above that even in (106) imposes a scale of the likelihood
of eating spinach, and Bill is at the bottom of the scale, as in (107):

(107) The likelihood of eating spinach

John -)
Mary 9
iiirr —)

 

 

Now consider the following cases in Horn and Lee (1994).

(108) a. There isn’t any person here.
b. There isn’t even a single person here.

(109) a. There isn’t any food left.

b. There isn’t even the least amount of food left.

(110) a. I like any apple.
b. I like even the least delicious apple.
(111) a. Any puppy is cute.

b. Even the ugliest puppy is cute.

121

All the (a) sentences can be paraphrased as the corresponding (b) sentences by using
even. This suggests that just like even, any imposes a contextually relevant scale in these
examples too, and the DP associated with any is at the bottom of the scale. For (108) and
(109), the scale is along the dimension of quantity, therefore the bottom of the scale is the
least amount of quantity, which could be zero. For (1 10) and (111), it is a scale of kind,
therefore the bottom of the scale is the kind of entities that is least likely to satisfy the
predicate. As we can see fi'om these examples, if the predicate holds even for the bottom
of the scale, it will automatically hold for the whole scale. The existence of an implicit
even in English NPIs finds support in other languages, where an overt morpheme even is
part of the NPI form, Hindi, for instance (Lahiri 1998).

Applying Lee and Hom’s analysis to comparatives, English comparatives are
analyzed in a similar way as Chinese comparatives. The implicit focus marker even in
(112a) imposes a set of alternatives that will make the statement John is taller than X
true. Among all the alternatives, the statement in (112a) describes the least likely event.
In a comparative context as (112a), the least likely event would be paraphrased as (112b),
i.e. John is taller than tallest person. If this event is true, by scalar inference, we get the
universal interpretation that John is taller than everybody. This analysis confirms Zepter’s
intuition that comparatives containing NPIs are interpreted as if they were superlatives,

but different from Zepter’s analysis, now we can explain why they are interpreted this

way.

(112) a. John is taller than anybody.

b. John is taller than even the tallest person.

122

 

 

 

It is worth noting that different fiom common assumptions that the DP associated with
even expresses the lowest point on a relevant scale (Haspelmath1997, Lee and Horn
1994), under the present analysis, it is an event that is associated with the lowest point of
the scale. Specifically we are concerned with the least likely event. This point is most
clearly demonstrated by the paraphrase of (1 12a), as shown in (112b), where the DP
associated with even is actually the highest point on a scale of height, but the even that

John is taller than the tallest person is still at the lowest point of a likelihood scale6.

3.3.4. Summary

Mainly based on Chinese data, in this section we give a scalar analysis of NPIs in
comparatives, and I also show that this analysis can be extended to English as well. To
license an NPI in comparatives, we need to have first, a widened domain, probably

induced by focus; second, a scale that orders all the alternative events along a likelihood

 

6 This could explain an argument against the even analysis of NPIs like any or ever (Heim 1984). Heim
argued that it is the class of minimizers, not normal NPIs as any or ever, that contain an implicit even.
Minimizers are a class of NPIs that on the surface expresses minimal quantity or degrees, for instance, lift a
finger, utter a ingle word, lend a red cent, etc. One critical property that Heim claims to distinguish
minimizers on one hand and any or ever on the other, is that minimizers add rhetorical flavor to questions,
but any or ever doesn’t (also see Ladusaw 1979, Wilkinson 1996, Han 1998). Guerzoni (2004) attributes
the negative bias of minimizers to the scope interaction between the implicit even and the interrogative

operators. In the following examples, the (a) questions, which contain minimizers, are biased towards a no
answer, but the (b) questions are not biased to either yes or no answer.

(1) a. Will Mary lift finger to help? Negatively biased
b. Will Mary do anything to help? Neutral

(2) a. Do you have the faintest idea? Negatively biased
b. Do you have any idea? Neutral

(3) it. Did he say a single word? Negatively biased
b. Did he say anything? Neutral

This objection wouldn’t apply to NPI in comparatives because as shown in (I 12) the DP associated with
the NPI is tied to the highest point on a scale, and the event is presupposed as the least likely event. This is
different from minimizers, whose association with the negative rhetoric force, as pointed out by Guerzoni,

is closely tied to the fact that they express the lowest point on a scale themselves, and the event related to
them are presupposed to be the most likely event.

123

scale; and thirdly a way to pick out the event on the top of the scale, i.e., the least likely
event. Crucially, different languages have their own ways to implement these ingredients.
For example, Chinese overtly moves the NPI for focus reasons and uses an adverb on the
predicate to achieve the end of the scale interpretation, on the other hand, For English
NPIs, the same functions can be encoded in the lexicon through an implicated
incorporated even. Since none of these ingredients is an inherent requirement on either
NPIs or comparatives, it is also expected if some languages don’t license NPIs in
comparatives at all, which means those languages lack specific implementations for one,
or even all the above ingredients for comparatives. However, I will leave a full

exploration to another occasion.

3.4.Conclusion

NPI in comparatives poses some challenges to some previous analyses of NPIs in
general. We have shown in this chapter that a scalar analysis explains the data better.
How about the previous analyses then, for instance, the downward entailing analysis and
the (non)veridicality analysis? Independent investigations outside of comparatives have
shown that they each can account for some part of the NPI data. Moreover, the scalar
analysis has its own limitations too. For instance, Rullmann (1996) has criticized Lee and
Hom’s analysis based on Dutch data. Therefore, the purpose of this chapter is not to
replace other proposals with the scalar analysis. In fact, it is notoriously difficult, if not
completely impossible, to give one single analysis to all the variations of NPIs
crosslinguistically. Comparatives provide a good opportunity for us to see one way,

probably among many other ways, that a NPI can not only be licensed, but also be

124

 

interpreted as universal. The next step of the exploration would be looking for the link
behind all these different proposals that can place them together under the same big

picture.

125

Chapter 4

The Degree Argument

4.0. Introduction

While discussing the semantic issues in previous chapters, we have seen that the
semantics and syntax in comparative constructions are closely related. Starting from this
chapter, I will address in more details some issues about the syntax of comparatives. As
mentioned in the Introduction, to address the structure of comparatives, one has to
address a critical question first, that is if we need a degree argument in the structure.
There have been some debates in the literature over this issue. In this chapter I will

provide evidence to support the existence of such an argument. And in the next chapter, I

will propose a structure that integrates the degree argument.

4.1. Do we need a degree argument?

4.1.1. Quantification over degrees

Gradable adjectives are relational expressions that relate individuals to degrees on
a scale (Cresswell 1976, Hoeksema 1983, von Stechow 1984, Kennedy 1997, Heim 1985,
2002). For instance, when we say x is smart, we map x to a point of degree d'on the scale

of intelligence. The semantic representation of the adjectival predicate smart is as stated

in (1):

(1) smart (x, d)

 

‘ I am assuming degrees are points on the scale. For discussions about degrees as extents/intervals on the
scale, see Kennedy (1997, 1999), Schwarzschild & Wilkinson (2002).

126

If the statement x is smart maps x to a degree on the scale of intelligence, the comparative
statement x is smarter than y is should also map x to its degree of intelligence, let’s call it
dx. At the same time, dx is compared to another degree of intelligence (1,. Traditionally, it
is assumed that comparative constructions involve existential quantification over degrees.
That is to say, the adjectival predicate takes a degree argument. The logical

representation of the statement x is smarter than y is is stated as (2):

(2) X is smarter than Y is.

3dx[dx>(hdy smart (y, d,))] [smart(x, dx)]

In traditional terms, dy is called the standard degree. Also following von Stechow (1984)
and Rullmann (1995), I will assume that the comparative clause, i.e., the than-clause, is
under the operation of a maximality operator. That is to say, for (2) to be true, X has to be
smarter than the maximal degree of Y’s intelligence. The maximality operator is defined

as (3). Adding this maximality operator, the logical representation of (2) is stated as (4):

(3) MAX (D) = Id 6 1) [Vd’ e D: dZd’]

(4) X is smarter than Y is.

adedx> MAX (My smart (y, d,))][smart(x, dx)]

127

Although the semantic representation in (4) captures the correct meaning, there
are two concerns as to whether we should consider comparative constructions as
involving quantification over degrees. The first question is that, as Kennedy (1997)
pointed out, some scope facts seem to show that degree arguments don’t exist. Secondly,
even if the degree argument does exist, it is unclear how to fit it into the syntactic

structure of comparatives. I will turn to these two issues one by one.

4.1.2. The debate about quantification over degrees
Semantic ambiguity arising from scope ambiguity between two quantificational
elements is very common in languages. For example, depending on the scope interaction

between the two quantificational DPs in (5), example (5) has two different meanings.

(5) Everybody likes a teacher.
a.Vx3y [teacher (y) & x likes y]

b. 3ny[teacher (y) & x likes y]

Under the interpretation of (5a), each person likes a different teacher; but under (5b) there
is a specific teacher that everybody likes. If the inherent semantics of adjectives involves
quantification over a degree argument, we expect to see its interaction with other
quantificational elements, e. g. quantifiers, negation, and intensional operators. Kennedy
(1997) discussed these issues and arrived at a negative answer. Let’s take a look at one

example in which the subject of the comparative construction is a strong quantifier.

128

(6) Every squirrel on campus is bigger than my cat.
a. VXlsquirrel (X)l[3d1[d1>MAx(M2 big (my cat. d2))ll[big(X. dl)]]

b. 3d.[d.>M,.x(r.d2 big (my cat, d2))][Vx[squirrel (x)] [big(x. do]

If the degree argument of the adjective big (i.e. d;) participates in scope ambiguities in
sentence (6), we should have two interpretations as represented in (6a) and (6b). In (6a),
the universal quantifier every squirrel takes scope over the degree argument d;. (6a) says
every squirrel possesses a degree of bigness d;, and d; exceeds d2, which is the degree of
bigness of my cat. This amounts to say that each squirrel is bigger than my cat, and that is
the natural reading we get for (6). When the scope between the two inverts, as shown in
(6b), we get an interpretation that there exists a degree d; which exceeds d2, the bigness
of my cat, and every squirrel is d;-big. (6b) forces a reading that every squirrel is the
same size, and that is not a natural reading we get from (6). Kennedy argued that similar
facts also hold for negation and intensional operators, i.e. the degree argument fails to
scope over them and give rise to semantic ambiguity. These facts cast serious doubts on
the quantificational nature of the degree argument and Kennedy takes them further to
argue that there doesn’t exist such a degree argument.

Heim (2000) argued that although Kennedy’s observations are largely correct, the
degree argument does show a scope ambiguity in a limited number of cases. In particular,
although the degree argument can’t scope over a higher quantificational DP, as in (6), nor

can it scope over a higher negation, it can scope over some intensional predicates, e. g. a

possibility sentence as in (7).

129

(7) (This draft is 10 pages) The paper is allowed to be less long than that.
a. allowed [less than that]; [the paper is (I; -long]
i.e. the paper is not allowed to be longer than 10 pages.
b. [less than that]; [required] [the paper is d; -long]

i.e. the paper is not allowed to be as long as 10 pages

Heim acknowledged that the conclusion drawn from intensional predicates (also see
Stateva 1999) is tentative, because besides the movement of the degree argument, there
might be other ways to explain the ambiguity in (7). She suggests that a stronger
argument for the degree argument comes fi‘om its behavior in comparatives that contain
Antecedent Contained Deletion (ACD), where the covert movement of such an argument
is detectable. A standard diagnosis for covert movement of a quantificational DP (QR) is
its ability to license ACD. For example, in (8), to resolve the ellipsis site after did one
needs to copy its antecedent, i.e. the higher predicate VPl, into VP2. However, since VP2
is contained within VPl, such an operation will suffer from the infinite regress problem.
QR can be used to rescue the situation. After the DP is moved out of WI , there is a

simple VPl read that can be copied into VP2, as shown in (9).

(8) I [vm read [or every book that you [VP2 did A]]].

(9) [09 every book that you [vpz did A]]; [I [van read t;]]

130

To the extent that (8) is a perfectly fine sentence despite the ACD problem on the surface,

covert movement in (9) of the quantificational DP must have happened at LF. Following
Wold (1995), Heim argued that the same argument can be carried over to the degree

argument in comparatives, as shown in (10).
(10) John was climbing higher trees than Bill was.

To understand (10), we have to first understand a similar sentence in (11).

(1 1) *John was climbing trees that Bill was.

Carlson (1975) noticed that ACD is degraded in a relative clause as (11), in which the
head of the relative clause is a weak indefinite, especially an existentially read weak
plural. This was explained in Diesing (1992) as the conflict between two contrary
conditions. On the one hand, to resolve ACD, the DP trees that Bill was has to undergo
QR; on the other hand, on independent ground Diesing claimed that existential bare
plurals do not QR, instead they stay inside VP to get existentially bound via existential
closure. In contrast to (11), the ACD in (10) can be resolved by QR of the degree

argument while leaving the bare plural inside the VP, as shown in (12).

(12) [-er than Bill was elimbing—el..—high—trees]; [John was climbing (1; -high trees]

131

 

To summarize, the ACD facts in comparatives seem to strongly support the
quantificational status of the degree argument. In next section, I will discuss another issue
that also involves ACD in comparatives. That is, the definiteness effect in attributive

comparatives. I will show that using the quantificational degree argument we can provide

a simple account for the definiteness effect.

4.2. The Definiteness Effect

In previous studies, the scope property of the degree argument was used to
examine whether there is a quantificational degree argument or not. Recall that the main
argument in Kennedy (1997) is that since we don’t seem to find evidence that the degree
argument enters into scope ambiguities with other quantificational elements, there
shouldn’t be a degree argument in the representation. Part of the counter arguments in
Heim (2000) also focused on looking for scope ambiguity between the degree argument
and other quantificational elements. To the extent that scope ambiguity does exist in
some limited cases, we still have reason to believe the quantification over degree
arguments. In this section, I will present some new evidence to support the
quantificational view. The test I am using is the intervention effect at LF. As proposed in
Beck (1996), LF movements are blocked by intervening quantifiers. I will show that the

LF intervention effect accounts for the definiteness effect in comparatives, which in turn

supports LF movement for degree arguments.

4.2.1. The definiteness effect In attributive comparatives

Lerner and Pinkal (1995) noticed that attributive comparatives show a

definiteness effect (DE). As shown below, the comparative DP has to be an indefinite

DP.

132

(13) a. George owns a/some/a few faster car(s) than Bill (does).

b. *George owns every/the faster car than Bill (does).

They also noticed that the indefiniteness effect is dependent on the explicit occurrence of

a complement phrase (i.e., the than-phrase), since all of the examples in (11), which have

no complement, are acceptable.

(14) a. George owns a faster car.
b. George owns every faster car.

c. George owns the faster car.

Beil (1997) attributes the DE above to a requirement on strong DPs that their domain has
to be presupposed in previous context. Although his proposal might be independently
needed for comparatives anyway, I will argue in the next section that it is not sufficient to

account for the whole range of data of the definiteness effect in comparatives.

4.2.2. Beil (1997)
Beil (1997) noticed that for strong determiners, the observation about the lack of a

complement phrase is correct only when there is a presupposed comparison set in the

previous context. Consider the examples below:

(15) a. Of those cars, Sue bought one. George bought every faster car.

133

b. Sue bought a car. *George bought every faster car.

c. Sue bought a car. George bought a faster car.

lntuitively (15a) is preferred over (15b) because in (15a) there is a contextually given
comparative instance (namely, Sue’s car) that is taken from a contextually presupposed
set (namely, of those cars). The absence of such a presupposed set does not matter for the
weak DP in (15c). Beil attributes the difference to a general rule that strong quantifiers
always presuppose their domain (Moltmann 1996), but weak ones do not. The
interpretation of every faster car in (15a) depends on the presupposition of a set of cars,
and that set is provided in context. Notice that the single instance Sue ’s car has to be a
subset of some set of cars whose definition can have various instantiations, but
mentioning Sue’s car in the context of (15b) does not eliminate the ungrammaticality.
Based on the contrast between (153) and (15b), Beil argues that it is crucial that the
domain of the strong quantifier is not just non-empty, but is also defined in the previous
context.

The presupposition difference between strong and weak DPs is independently
motivated by some other linguists too. Milsark (1974) argues that strong detenniners are
unambiguously presuppositional, whereas weak detenniners are ambiguous between a
presuppositional interpretation and a nonpresuppositional one, for example, the cardinal

reading. For instance, the strong determiners in (16) presuppose the existence of ghosts.

If there are no ghosts, the truth value of ( 16) is undefined.

(16) a. Every ghost danced in my house.

134

b. Most ghosts danced in my house.

On the other hand, the weak determiners in (17) are ambiguous. (17a) only asserts the

existence of ghosts, it does not presuppose it. If it turns out there is no ghost in the house,
the sentence is false, but if there are ghosts in the house the sentence is true. The sentence
in (17b), usually with a stressed reading of the determiner, carries a presupposition of the

existence of the ghost. It is read more like a partitive determiner, such as some of the

ghosts, three of the ghosts.

a. There is/are a/some/a few ghost(s) dancing in my house.

(17)
b. A/SOME/A FEW ghost(s) is/are dancing in my house, (the others are

dancing in the street).

However, the presupposition property alone is not sufficient to explain the
definiteness effect in attributive comparatives. For example, it is not clear how Beil’s
analysis can be extended to explain the cases where the comparative complements are

explicit, although Beil claims it does. If we add a context to the original examples in

Lerner and Pinkal (1995), the judgments are still similar:

Of those cars, Bill owns some. George owns a/some/a few faster car(s) than

( 18)
Bill (does).

Of those cars, Bill owns some. *George owns every/the faster car than Bill

( 1 9)
(does).

135

As Beil notes himself, in the following examples, even if there is a presupposed set
(because contestant lexically presupposes the set of participants in a contest), there is a

difference between the weak comparative DP and strong comparative DP.

(20) Sue defeated a stronger contestant than A1.
a. Sue defeated a contestant that is stronger than Al is.
b. Sue defeated a stronger contestant than Al did. (i.e. the contestant

defeated by Sue is stronger than the contestant defeated by Al.)

(21) ?Sue defeated every stronger contestant than A1.2
a. Sue defeated every contestant stronger than Al.

b. *Sue defeated every stronger contestant than Al did (i.e. every contestant

defeated by Sue is stronger than every contestant defeated by Al).

(20) has two interpretations. In the first reading, we are comparing the contestant
defeated by Sue with another person Al; in the second reading we are comparing the
person Sue defeated and the person Al defeated. When we change the indefinite DP in
(20) to a definite DP in (21), we lose the second interpretation. If the domain
presupposition requirement is the only difference between a weak and a strong DP, the

contrast above is left unexplained. The same problem exists in the following examples:

 

2 For some reasons, (21) sounds degraded to native speakers. Most people would prefer a word order like
“ . . .every contestant stronger than AL”. But the contrast between (18a) and (18b) is nevertheless robust.

136

(22) 21. ?Of those cars, Sue bought a BMW. George bought every faster car than

that BMW.
b. *Of those cars, George bought every faster car than Sue did.

The context “of those cars, ...” provides a presupposed set of cars in both examples.

However, the DE is avoided only in the (a) example.

To summarize, the presupposition account alone is not the desired solution. The

alternative I will propose reduces DB in comparatives to an LP intervention effect laid

out in Beck (1996). The basic idea is that at LF, in order to yield a final interpretation, the

degree argument of the gradable adjective has to move across the DP head, namely, the

determiner. Being quantificational, strong detenniners will block such movements,- but

weak detenniners will not. In the next section, I will review the discussion on the

quantifier intervention effect, and propose my analysis of the DE.

4.3. The DE as an Indication of the degree argument movement

4.3.1. The quantifier Intervention effect at LF
The main finding in Beck (1996) is informally schematized as below, which

indicates that quantifiers block LF movements.

(23) *[....X;....[Q...[...t;...]]]

The main arguments come from German split-quantifiers, as shown in (24) and (25).

137

(24) a. Wen alles hat Luise gesehen?
Whom all has Luise seen
Who-all did Luise see?

b. Wen hat Luise alles gesehen?
Whom has Luise all seen

Who-all did Luise see?

(25) a. Wen von den Musikern hat Luise getroffen
Whom of the musicians has Luise met
Which of the musicians did Luise meet?
b. Wen hat Luise von den Musikern getroffen
Whom has Luise of the musicians met

Which of the musicians did Luise meet?

In (24a), wen alles is originally one argument, but it can split in (24b); 3 similar case
holds for wen von den musikern in (25). Since in the (b) sentences what is left behind is
the restriction of the wh-phrase, the natural conclusion is that there are LF movements for
the (b) sentences in order to have a complete quantifier argument to derive the correct
interpretation. So alles in (24b) and von den Musikern in (25b) will move at LP to join

wen. Interestingly, quantifier splitting is not allowed if there is another intervening

quantifier.

138

(26) a. ?? Wen hat niemand alles gesehen?
Whom has nobody all seen
Who-all did nobody see?
b. ?? Wen hat keine studentin von den Musikern getroffen
Whom has no student of the musicians met

Which of the musicians did no student meet?

In (26), the quantifier niemand ‘nobody ’ is intervening between the split quantifiers.
Since that degrades the acceptability of the sentence, Beck concludes that the LF
movement of alles and von den Musikern is blocked.
The above discussion is formalized into the following two definitions:
(27) Quantifier-Induced Barrier (QUIB):

The first node that dominates a quantifier, its restriction, and its nuclear scope

is a Quantifier-Induced Barrier.

Minimal Quantified Structure Constraint (MQSC)

If an LF trace B is dominated by a QUIB or, then the binder of B must also be

dominated by or.

Beck also notices that indefinite DPs do not generate the same kind of LF
blocking effect as other quantificational structures. She suggests that only inherently
quantificational elements can induce the LF blocking effect. Following Heim (1982),
indefinites are variables with no inherent quantification force, so they do not induce the

LF blocking effect.

139

Interestingly, the distinction between indefinite DPS and other DPs in terms of
their quantificational status seems to find support in the discussion in Diesing (1992).
Diesing argues that strong and weak DPs differ with their ability to QR. Strong DPs are
quantificational; hence they can be raised by QR. On the other hand, only
presuppositional weak DPs are quantificational and undergo QR. Cardinal (non-
presuppositional) weak DPs are not quantificational and stay in-situ at LF. If this analysis
is on the right track, it suggests that strong determiners or weak but presuppositional
determiners will induce the LF intervention effect, but weak determiners will not. This
assumption will become crucial for the discussion in the next section.

To summarize, strong detenniners block LF movements, because they are
genuinely quantificational, but weak non-presuppositional detenniners are just variables
with no inherent quantificational force, so they do not block LF movements. In my
analysis in the next section, I will show that the DE arises as the result of an interaction

between the LF movement of the degree argument and the strong/weak detenniners.

4.3.2. A new proposal
Consider (28). Following Lerner and Pinkal (1995), (28a) is elliptical just as (28b)
is. I also follow Chomsky (1977) in treating the clausal complement as a wh-

construction.

(28) a. George ate a bigger cake than [Op; Bill e;].

b. George ate a bigger cake than [Op Bill did].

140

Following the traditional QR solution for ACD cases, (25a) will receive a derivation at

LF as in (29):

(29) a. [a bigger cake than [Op Bill [e]]]2 [Georger [t; ate t2]]
U

b. [a bigger cake than [Op; Bill [ate t;]]]; [George; [t; ate t2]]

As noted by Lerner and Pinkal, there is a problem for the above syntactic analysis. In the
resulting structure in (29b), the wh-operator binds an individual variable. We know it is

of an individual type because it is copied from another individual type variable, t2.
However, to derive a sensible interpretation, the wh-operator has to bind a degree

variable. How can we connect the individual cake introduced by t; to its degree of bigness?
Lerner and Pinkal’s solution is to introduce an implicit anaphoric element that relates
individuals to degrees. Let’s call this anaphoric component Po and give it a semantic

interpretation as (30), and the semantics of the comparative site of (29b) is given in (31):

(30) XQ3ylPo(y. d)AQ(y)l

(31) 3y[Po(y, d) A ate (Bill, y)]

Now we have a degree variable d that is derived from the individual variable. The wh-
operator needs to be re—interpreted as binding the degree variable.
Although this analysis can give us the desired semantics, it is not an ideal

solution. We not only need to introduce a hidden anaphoric element, but also have to

141

reinterpret the wh-operator with no obvious syntactic support. I will suggest that we can
apply QR again to solve the problem. Since the problem only arises for the DP in (29b),
in my discussion I will separate it from the rest of the structure. The problem now is how

to derive a degree variable that can replace the individual variable t; in (32):
(32) [Dp a bigger cake than [Op; Bill [ate t;]]]

If the gradable adjective big takes a degree argument, (32) actually has the structure in
(33), namely [-er than Bill ate] is the degree argument of big. Following the tradition that

modifiers are generated at [Spec, NP] position, (33) has a tree structure as (34):

(33)[Dp a [-er than [Op; Bill [ate t;]]] big cake]

DP

./ >“x

a
D P NP
A68 cake

[-er than Bill ate tj] big

(34)

Since the degree argument is quantificational, it is able to QR again, as shown in (35):

(35) [-er than [Bill ate [e]] ]; [Dp a [gg d;-much big CM

l-er than [Op; Bill ate tj]]i DP

D P
a
DegP NP
A cake
d ; -much big

142

Now a degree variable is available inside the NP. What we need is only copying that NP

into the object position of ate, and that gives us (36):

(36) [-er than [Op; Bill ate [d;-much big cake]] ]; [Dp a [Np d;-much big cake]]

In (36), we let the wh-operator bind the degree variable that is copied from the NP.

Combining (36) with (29b), the final LF structure we derive is (37):

(37) [Dp[-er than [Op; Bill ate [d;-much big cake]] ]; [up a [Np d;-much big cake]]];

[Georget [ti ate tzll

Now we can read the meaning of (37) with no problems. (37) says George ate a cake that
is d;-much big such that d; exceeds the degree of bigness d; which is the size of the cake
eaten by Bill.

One point is worth mentioning. In (36) what is copied is only the NP, not the DP,
because there are reasons to believe that the determiner is not copied back to the ellipsis
site (Lerner & Pinkal 1995, Lechner 1999). As shown below, the sentence George ate a
bigger cake than Bill only has the interpretation in (38a), not (38b). The same point is

shown by examples with other determiners, as in (39).

(38) George ate a bigger cake than Bill

a. George ate a bigger cake than [Op; Bill ate any d;-much big cake] (i.e.

George ate a bigger cake than any of the cakes Bill ate)

143

b. *George ate a bigger cake than [Op; Bill ate a d;-much big cake] (i.e.

George ate a bigger cake than a cake Bill ate)

(3 9) George ate at least two/three/at most three bigger cakes than Bill
a. George ate at least two/three/at most three bigger cake than [Op; Bill ate
any d;-much big cake]
b. *George ate a bigger cake than [Op; Bill ate at least two/three/at most

three d;-much big cake]

To summarize, we derive the interpretation of the sentence George ate a bigger
cake than Bill by applying QR twice. First, the object DP undergoes QR in order to
resolve ACD and recover the materials in the ellipsis site; second, the degree argument
undergoes QR in order to establish a degree variable in the than-phrase. It is at the second
step that the DE arises. As discussed in the last section, strong and weak determiners
have different ability to block LF movements. When we QR the degree argument, as in
(3 5), if there is a strong determiner in the way, that movement will be blocked. The
derivation crashes because the final interpretation can not converge, as shown in (40).
But weak determiners would not intervene, so the derivation will go through fine:

(40) A DP

D/\NP

/ every
>\ DegP NP
/\ cake

her than Bill fig] big
1

 

QR

144

4.3.3. Predictions

Since the current analysis crucially makes use of the movement of the degree
argument, it makes two predictions. First, in cases where movement is not needed to
derive the final interpretation, the DE will not emerge; second, if movement has to cross
over an island, the derivation will fail. I will argue in this section that both predictions are
borne out.

So far in my discussion the movement of the object DP is necessary in order to
resolve ACD, and consequently one needs to move the degree argument. However, not
all comparatives face the ACD issue. Lerner and Pinkal (1995) distinguish two kinds of

attributive comparatives: the narrow reading (NRA) as in (413), and the wide reading

(WRA) as in (41b):

(41) a. George owns a faster car than this BMW.

b. George owns a faster car than Bill.

Crucially, Lerner and Pinkal argue that we should treat the two readings differently. The
NRA readings are genuine phrasal comparatives, and the WRA constructions are
genuinely elliptical ACD constructions. To derive the WRA readings, the missing
material in the comparative site (i.e. the materials contained in the than-phrase) has to be
recovered first. The structural difference between two readings is demonstrated below:
(42) NRA: a. George owns a faster car than [DP this BMW]
b. Sue defeated a stronger contestant than [DP Al].

c. Sue defeated every stronger contestant than [or A1]

145

(43) WRA: a. George owns a faster car than [CP Bill does]
b. Sue defeated a stronger contestant than [or Al did].

c. Sue defeated every stronger contestant than [CP Al did]

For more details of this analysis, I refer readers to Lerner and Pinkal (1995). At this
point, what is crucial for my purpose is that the NRA readings do not involve
movements, but the WRA readings do. Interestingly, this explains some old data that are

problematic for Beil (1997). Consider his original data again.

(44) Sue defeated a stronger contestant than A1.
3. Sue defeated a contestant that is stronger than Al is.
b. Sue defeated a stronger contestant than Al did. (i.e. the contestant defeated

by Sue is stronger than the contestant defeated by Al.)

(45) ?Sue defeated every stronger contestant than A1.
a. Sue defeated every contestant stronger than Al.
b. *Sue defeated every stronger contestant than Al did (i.e. every contestant

defeated by Sue is stronger than every contestant defeated by Al).
Recall the problem is that the presence of a weak determiner in (44) allows two

interpretations of the sentence, but the strong determiner in (45) blocks one interpretation.

In other words, the DE only arises for the (b) interpretations, but not the (a)

146

interpretations. Interestingly, the (a) interpretations are exactly the same kind of
interpretation as the NRA readings of (42), and the (b) interpretations are the same as the
WRA readings of (43). That is to say, no movement is needed to derive the (a)
interpretations, but (b) interpretations require movements. The contrast between (44) and
(45) suggests that the DE is sensitive to syntactic movements, and this is exactly what the
current analysis can capture. Specifically, weak determiners never induce the DE, as
shown in (44). Strong detenniners induce DE only when there is movement, as shown in

(45). I summarize the results in the following table:

 

 

 

 

Involves Ellipsis? Definiteness Effect?
Phrasal comparatives NO NO
(the (a) interpretations)
Clausal comparatives YES YES
(the (b) interpretation)

 

 

 

The same approach extends to another contrasting pair observed earlier, repeated

below for convenience:

(46) a. ?Of those cars, Sue bought a BMW. George bought every faster car than
that BMW.

b. *Of those cars, George bought every faster car than Sue did.

Again, (46a) is a case of NRA, and (46b) a WRA. The strong determiner every only

shows a DE effect in (46b).

147

 

Let’s use the following data to test the second prediction. As shown in (47), there

exists a contrast between a pre-nominal modifier and a post-nominal modifier.3

(47) a. John met a taller girl than Bill did.

b. *John met a girl who is taller than Bill did.

I hope it is clear now how (47a) can be derived. It should be derived in the same way as
(28a). The exact same process applies to (47b), but problems arise as shown below. First,
the whole object DP undergoes QR, and the predicate VP is copied into the ellipsis site,
as shown in (48a) and (45b); in (48b) what is needed is a degree variable in the ellipsis
site, but we have an individual variable, so the degree argument, as shown in (48c), needs
to undergo QR again, as in (48d). It is at this step problems arise. The degree argument
can not undergo QR because it is within a relative clause island. Since (48d) and (48e)
can not proceed, the semantic interpretation would fail because there is no degree

generated in the ellipsis site.

(48)

I”

[a girl who is taller than Bill [e]]; [John met t;]

b. [a girl who is taller than Bill [met t]]; [John met t;]

c. [a girl who is [-er than Bill met t] tall]; [ John met t ;]

(1. *[[-er than Bill met t].. [ a girl who is dr -much tall]] j I John met til

e. *[[-er than Bill met [girl who is d-much tall]]k [ a girl who is d k -much

tall]]; [ John met t ;]

 

 

3 I thank Chris Wilder for pointing out this to me.

148

To summarize, I reviewed the debate about the existence of a quantificational degree

argument. New evidence from the definiteness effect lends support to the traditional view
that comparatives involve quantification over a degree argument. The immediate problem
is how to fit this argument into the syntactic structure. I will address this issue in the next

section.

4.4. The syntactic structure of comparatives
4.4.1. The discrepancy between semantics and syntax

If comparatives involve quantifications over degree arguments, assuming that the
argument is generated at the specifier position of its head, we can derive a syntactic
structure of comparatives as in (49a). The DegP [-er than Bill is] denotes the degree
argument of the adjective head tall and they are in a [Spec, head] relation. One advantage
of this structure is that it makes a comparative construction as in (49) analogous to a
simple adjectival predicate as in (50). In both sentences, the predicate tall takes a degree
argument DegP, so both mean John is d—much tall. The difference between the two only
lies in the content of the degree argument. As (49b) shows, (49) means John is tall to a
degree d1, and d1 is defined as a degree exceeding the maximal degree d2 to which Bill

is tall. (50) also means John is tall to a degree d1, but d1 is defined as six-feet.

(49) John is taller than Bill is.

149

a. John is [AP [933p —er than Bill is] tall]

AP

DegP A’

Deg/ \P lA

-er than Bill is tall

b. 3d1[d1> MAX (M2.tall(Bill, d2))] [tall (John, d1)]

(50) John is six-feet tall.

AP
D4 \ A’
six-feet I

A
tall

b. 3d1[six-feet(dl)] [tall (John, dl)]

One problem with the structure in (49) is that on the surface the degree argument
DegP is not a constituent. To derive the correct word order, Bresnan (197 3) proposed to
extrapose the than-clause to the right. However, extraposition can’t solve a deeper
problem of (49). As Corver (1997) noticed, it is impossible to extract the DegP out of the
adjective phrase, instead, one has to extract the whole AP to ask a question, as shown in
(51) and (52).
(51) John is [Ap[Degp-er than Bill is] tall]

a. *How; is John [ti tall]?

b. How tall is John?

150

(52) John is [Ap[DegP six-feet] tall]
a. *How; is John [ti tall]?

b. How tall is John?

This is unexpected under a structure as in (49), because there is no reason that we
shouldn’t be able to extract the specifier of a maximal projection.

From the discussion above, we know that the structure in (49) can’t be right. It
over-generalizes and wrongly rule in cases like (51) and (52). The problem can be
avoided if we adopt an extended DegP structure for comparatives (see Abney 1981,

Corver 1997, Kennedy 1997), as shown in (53).

(53) John is [W [DegP —er L... tall] 1 [PP than Bill is]]

DegP
l
Deg
/\
Deg’ PP
/\

9

Deg AP than Bill is
-er I

A
tall

syntactically, (53) can solve a couple problems. First, it preserves the surface word order;
second, the degree morpheme —er and the PP than Bill is don’t form a constituent, so we

can’t extract them without extracting the whole DegP, as in (51) and (52).

151

However, although the structure in (53) is independently motivated on syntactic
grounds, it does speak against the kind of semantics that assumes quantification over
degree arguments. This is so exactly because now the head of the projection is the degree
morpheme —er, and it takes two arguments, the gradable adjective tall and the standard
degree than Bill is. The adjective head tall isn’t taking any degree argument, and in fact it
can’t because the degree argument [-er than Bill is] isn’t a constituent at any point of the
derivation. To maintain this structure, one has to give up the degree argument. This is the
approach taken in Kennedy (1997). To maintain the degree argument, which seems
necessary from discussion in the last section, one has to give up the extended DegP
structure, namely, one needs to go back to a structure like (49). Could we have another
alternative that can rescue us from this dilemma? What we need is a structure that
maintains the degree argument, but at the same time maintains the surface word order and

possibly the extended DegP structure. Next, I will review two proposals in the literature.

4.4.2. Solution one—late merge the than-clause

To maintain that the degree morpheme —er and the than-clause together form a
degree argument on the one hand, and on the other hand they are discontinuous on the
surface, Bhatt and Pancheva (2004) proposed that the than-clause is countercyclicly
merged to the -er after —er undergoes QR. The lower copy of -er is spelled out, whereas
the than-clause is merged to the higher -er at the QR position. This explains why on the
surface —er and the than-clause are not pronounced as a constituent, but they
semantically behave as one degree argument. Let’s illustrate this proposal in more detail.

As shown in (54a), the degree argument is merged with the adjective head at the [Spec,

152

AP] position. The degree argument at this point only contains the degree morpheme —er.
Being a quantificational argument, the DegP is able to undergo QR, as shown in (54b).
The lower copy of —er (-er;) is the one that is spelled out in pronunciation, whereas the

higher —er is interpreted at LF.

(54) a. AP
/\
DegPi A
I tall
Deg
|
er
b XP
/\
. . .XP DegPi
l l
. . Deg
AP i
/\
DegPi A -er2
l tall
Deg
-CI'|

At the high DegP position, the than-clause is merged, as shown in (55).

153

(55)

XP
/\
XP DegPi
. . . .. Deg’
DegPi A -er2 A
I tall than-clause

Deg
-CI'|

As shown in (55), the degree argument [-er2 than-clause] gets interpreted at LF as one
constituent, but it is —er1, not —er2 that is pronounced.

One advantage of this approach is that since the than-clause is base generated
high, instead of being moved from a lower position, we can explain nicely the absence of

the Condition C effects in (56).

(56) a. * I will tell him; a sillier rumor about John;.

b. I will tell him; a sillier rumor (about Ann) tomorrow than Mary told John;.

(56a) demonstrate a typical Condition C violation. The absence of the Condition C effect
in (56b) suggests that the than-clause is situated out of the VP to escape being c-
commanded by the indirect object him. The late merge approach explains why the

than-clause could escape the c-command domain of him.

154

The late merge approach wasn’t originally implemented on an extended DegP
structure. As shown in (54), it is assuming a traditional AP structure, which we argued
against in the last section. However, I believe this proposal can be implemented with the

extended DegP structure as well.

4.4.3. Solution two—the DegP-shell structure

Based on the semantic and syntactic parallel between DP structure and DegP
structure, Larson (1991) motivated the DegP-shell structure for comparatives on
independent grounds.

Under generalized quantifier theory, detenniners are interpreted as relational
items, namely, they express relations among sets. For example, the sentence all students
dance is true if the set of students and the set of dancers stand in such a relation that the
former is a subset of the latter; some students dance is true if the set of students and the
set of dancers stand in such a relation that their intersection is not empty. The following

are some examples of how detenniners are interpreted:

(57) a. all (A, B) =1 iff A g B
b. some (A, B) =1 iff AnB ¢Q
e. no (A, B) =1 iffAnB =3

d.Clat least five (A, B)Cl=l iff |AnB| 2 5

Under this analysis, a determiner behaves like a binary transitive predicate. It has two

arguments, one restriction and one scope. The common noun part is the restriction, the

155

main predicate is the scope, and the determiner contributes to the relation that links the
two, in particular, how many things restricted by A also satisfy the scope B.

All the determiners in (57) are dyadic determiners, namely they express relations
between two sets. The same analysis can be extended to triadic determiners, which
express relations among three sets. For example, the complex determiner more than
expresses relation between two arguments and a predicate. For More women than men
came to be true, we need to consider three sets: X the set of women, Y the set of men and
Z the set of people who came. The intersection of X and Z has to exceed the intersection

of Y and Z.

(58) a. more than (Z, Y, X) Cl=1 iff |XnZ| Z |YnZ|
b. more women than men came iff
|{x: woman (x)} n {2: people who came (z)}| 2

|{y: man (y)}r\ {2: people who came (z)}|

Ideally we would like a single structure that can cover both kinds of determiners in (57)
and (58). The commonly assumed DP structure for dyadic determiners is as shown in
(59). The determiner some takes one argument students in its complement position, and

the other argument will be provided by the predicate of the sentence that this DP appears

in‘.

¥

4 In Larson’s analysis, the second argument (and for the triadic determiners, the third argument too), is
generated in pro at the [Spec, DP] position. The value of this pro given by the main predicate at LF. In this
way, all the arguments of a determiner can be generated within the maximal projection of that determiner.
The exact details of this analysis are beyond the scope of our discussion.

156

(59) DP
/ \
D,

/\
D NP

some students

To add the third argument for the triadic determiner, Larson proposed a DP-shell
structure shown in (60). Under this structure, the determiner starts at the lower D position
to link two of its arguments. Then it is raised to the higher D position to derive the correct
word order. The derivation conveniently accounts for the discontinuous dependency
between more and the than-phrase.

(60) DP
/ \
D’
/ \
D DP
more / \
NP D’
women /\
D PP
t A

] than men

 

 

Now let’s turn to comparative constructions. Assuming that adjectival expressions
such as comparatives are quantificational in nature, Larson argued that quantificational
degrees have a parallel semantics to quantificational determiners, namely they are both
relational items. The degree morpheme such as —er/more expresses a three-way relation
between an adjective and two DPs. To link these three things together, syntactically
-er/more first combines with the than-phrase, then with the adjective, as shown in (61),

and finally in order to introduce the third argument, -er moves to the higher DegP

157

projection. Putting aside technical details, the subject argument at this point is only a pro,

its value will be identified later in the derivation as the subject argument.

 

 

(61) DegP
/ \
pro Deg’
/ \
Deg DegP
-er / \
AP Deg’
tall /\
Deg PP
t A
1 than Bill

A slightly different version of the DegP-shell structure was proposed in Izvorski (1995)
and I will not go into more details of those discussions. What is important for our purpose
here is that at some stage of the derivation, the degree morpheme and the than-phrase
forms one constituent, i.e. the Deg’ structure in (61). Interestingly, the DegP-shell
structure is reminiscent of the VP-shell structure in double object constructions (Larson
1988). For instance, the example in (62) has a structure as in (63). I will have more
discussion about double object constructions and comparatives in the next chapter.

(62) John sent Mary a letter.

(63) VP]

V DP a letter
rj ti

 

 

 

 

 

158

4.5. Summary

In this section, we see two proposals that can incorporate the degree argument
into the syntactic structure and at the same time satisfy the surface word order
requirement. In the late merge analysis, the degree morpheme and the than-phrase forms
a constituent at the late merge position; whereas in the DegP-shell structure, the same
constituent can be formed at the lower DegP structure.

The two proposals are originally motivated on different grounds. The late merge
approach was mainly motivated by scope and condition C interactions, whereas the
DegP-shell analysis was motivated by the similarity between DPs and DegPs. As far as I
know, the two proposals cover more or less the same range of facts for English
comparativess. However, when we move to other languages, it seems that the DegP-shell
analysis makes more sense. As will be discussed in Chapter five, the head movement in
the shell structure is actually lexically implemented in Chinese comparatives and gives us
two different comparative constructions that are underlyingly related. If we believe that
crossliguistically there should be one unified structure for comparatives, Chinese data
might lend support to the DegP-shell structure for English data as well. However, I also
want to point out that before we find any direct empirical evidence in English to defend
the DegP-shell structure against the other alternatives, it is just a convenient assumption.
Moreover, instead of completely adopting Larson’s structure in (61) for Chinese
comparatives, substantial revisions will also be made for both conceptual and empirical

reasons.

 

5 It is not clear at this point if the DegP shell analysis can account for the Condition C facts discussed in
Bhatt and Pancheva 2004.

159

Chapter 5

The DegP-shell analysis of Chinese comparatives

In Chapter 4, I argued that comparatives indeed involve quantification over degree
arguments, as has been assumed traditionally. The question left is how to build this
degree argument into the overall structure of comparatives. The literature has provided at
least two approaches: the late merge approach and the DegP-shell approach. In this
chapter I will look at the structure of Chinese comparatives, which lends support to the
DegP—shell like structure, although some revisions need to be made to Larson’s original
structure. Some of the arguments here might still be speculative at this point, and many
questions are left open, but nevertheless I hope that the general properties of Chinese
comparatives will add to our understanding of comparatives from a cross-linguistic

perspective.

5.0. Introduction

Li & Thompson (1981) summarized different types of Chinese comparatives.
According to him, the basic pattern of Chinese comparatives is (l):

(1) X comparison word Y (adverb) dimension
In (1), X is the subject or topic under comparison, Y is the standard degree that X is
compared to, and dimension refers to the predicate which specifies the dimension along

which the comparison is made. In Li’s survey, there are three types of comparison words:

(i) Superiority: bi (‘than’)
(2) Ta bi ni gao

He than you tall

160

He is taller than you.

(ii) Inferiority: mei you/bu ru. . .(name) (literally ‘not as..as. . . ’)
(3) Ta meiyou/buru ni (name) gao
He not you (that) tall

He isn’t as tall as you.

(iii) Equality: gen...yiyang (literally ‘with. . .same. ..’)
(4) Ta gen ni yiyang gao
He with you same tall

He is as tall as you.

Throughout this thesis, I will only investigate the first type, superiority comparatives. I
believe the conclusions derived from this discussion will also apply to the second and the
third types, but I will leave this as an empirical question for firture research.

Chinese actually has two types of superiority comparatives. One type is like the
sentence in (2), and (5a) below, in which the standard degree is introduced by a
morpheme bi (than), I will call this kind bi-comparatives. The standard degree can also
be introduced without bi, as shown in (5b), I will call this a bare comparative. The word
order of the two types is different: in bare comparatives, the gradable adjective predicate
precedes the standard degree argument, whereas in the bi—comparatives, this order is

reversed.

161

(5) Superiority comparatives
a. Zhangsan bi wo gao yicun (bi-comparative)
Zhangsan than me tall one-inch

“Zhangsan is one inch taller than me”.

b. Zhangsan gao wo yicun (bare comparative)
Zhangsan tall me one-inch

“Zhangsan is one inch taller than me”.

Since the two examples in (5) express the same meaning, it is natural to ask if they share
similar syntactic structures too. Ideally, we would like to see that there is a unified
syntactic structure that can account for both word orders, and this is exactly what I am
going to argue for in this chapter.

Before we get into the details of Chinese comparatives, I will briefly introduce
some background on Chinese adjectives. There has been some debate as to whether
Chinese has a genuine class of gradable adjectives. It is so because Chinese gradable
adjectives show behaviors like verbs in many respects. I will keep agnostic to the final
conclusion of the debate, but I will show some facts that suggest a close relation between
gradable adjectives and verbs in Chinese.

First, Chinese gradable adjectives, just like verbs, can bear aspectual markers.
(6) Huaer bong-1e

Flower red-perf.

The flowers have become red.

162

(7) Ta ye cengjin meili guo.
She also ever beautiful perf.

She was beautiful before.

Second, when an adjective is used as a predicate, there isn’t a copular preceding the

adjective. This is different from a nominal predicate.

(8) a. Ta (*shi) hen piaoliang.
she is very pretty. 1
She is very pretty.
b. Ta *(shi) ge xuesheng
She is CL student

She is a student.

Although based on these facts we can’t definitely conclude that Chinese gradable
adjectives are verbs, it is worthwhile to keep these facts in mind when we discuss
Chinese comparatives.

This chapter is organized as follows. Since bare comparatives and bi-
comparatives don’t select the same set of adjectives, in section 5.1 I will introduce more
data about this issue. I discuss the structure of the standard degree in Chinese
comparatives in section 5.2, and argue that at least for the kind of comparatives we are
discussing here, the standard degree is expressed as a DP, not an elided CP. Section 5.3.

and 5.4. discuss the structure of bare comparatives and bi-comparatives. I argue that the

163

DegP-shell analysis can unite the two kinds of comparatives through head movements.

Some consequences of the DegP-shell analysis are explored in section 5.5. Since in the

DegP-shell analysis, the lower DegP structure is claimed to be the degree argument, I will

investigate the LF movement of this argument in section 5.6. The last section is a

conclusion.

5.1. The adjective selection constraints

Bi-comparatives and bare comparatives are not completely interchangeable. Although we

can always use any adjective in bi—comparatives, bare-comparatives seem to be severely

constrained in terms of what adjectives can be used. The following are some examples:

(9)

a. Ta bi wo gao/ai/zhong/qing /piaoliang/xixing /gaoxing /youqu yidian
he bi me tall/short/heavy/light/pretty/carefirl/happy/interesting a little
He is a little taller/shorter/heavier/lighter/prettier/more carefid/happier/more
interesting than me.

b. Zhe-gen-shenzi bi na-gen-shenzi Chang/duan/cu/xi . yidian
this-CL-rope bi that-CL-rope long/short/thick/thin a little
This rope is a little longer/shorter/thicker/thinner than that one.

c. Zhe-jian-fangjian bi na-jian ganjing/shufir /mingliang haoduo
this-CL-room bi that-CL clean /comfortable/bright much

This room is much cleaner/more comfortable/brighter than that one.

164

(10) a. Ta gao/ai /zhong /qing wo yidian
he tall/short/heavy /light me a little
He is a little taller than me.
b. *Ta piaoliang/xixing /gaoxing /youqu wo yidian
she pretty /careful /happy /interesting me a little
She is a little more prettier/careful/happier/interesting than me.
e. Zhe-gen-shenzi chang/duan/cu/xi na-gen-shenzi yidian
this-CL-rope long/short/thick/thin that-CL-rope a little
This rope is a little longer/shorter/thicker/thinner than that one.
(I. *Zhe-jian-fangjian ganjing/shufu /mingliang na-jian haoduo
this-CL-room clean /comfortable/bright that-CL much

This room is much cleaner/more comfortable/brighter than that one.

This shows that some adjectives, such as pretty, careful, happy, interesting, clean,
comfortable and bright can only be used in bi-comparatives, not bare-comparatives;
whereas other adjectives, such as tall, short, fast, slow, long, thick, can be used in both. In
the following discussion, I will call the former the constrained adjectives, and the latter
the free adjectives.

What is the difference between the two types of adjectives? I am going to argue
that dimensional adjectives (DA) are free to occur in both structures, but evaluative
adjectives (EA) are constrained in their occurrences. I have adopted the two terms DA
and BA from Bierwisch (1989). Bierwisch found that adjectives like long, short, old,

young (DA) are very different fiom adjectives such as lazy, industrious, pretty, ugly

165

(EA), although the distinction is not completely clear-cut all the time. One common test

for a DA is that positive DAs can be modified by a measure phrase, as shown in (11):

(1 1) a. 2-feet tall
b. 10-inches long

c. 10-years old

Presumably this is because dimensional adjectives describe explicit scales, such as
height, length, size, etc. In contrast, evaluative adjectives are not related to explicit
measurement. Although we can still modify the degree using degree adverbs such as

very, a little, we can not use measure phrases to modify EAs.

(12) a. a little pretty
b. very happy
0. *two-?? careful

(I. *10-?? comfortable
Interestingly, all the free adjectives in (10a) and (10c) satisfy the measure phrase test
(recall we only need to test the positive one), but not the constrained adjectives in (10b)

and (10d), as shown below:

(13) a. liang-mi gao

two- meter tall

166

b. liang-mi chang
two-meter long

c. liang-bang zhong
two-pound heavy

d. san-cun cu

three-inch thick

(14) a. *liang-‘l? piaoliang
two-?? pretty
b. *san-?? xixing

three-?? careful

Interestingly, there are some adjectives that don’t pass the measure phrase test, but they
can still show up in bare-comparatives. For instance, pang (fat), kuai (fast) in the

following examples.

(15) a. *10—bang pang
10-pound fat
b. * 3-fenzhong kuai
3-minutes fast
(16) a. Zhangsan pang wo shi-bang.
Zhangsan fat me 10-pound.

Zhangsan is fatter than me by 10 pounds.

167

b. Zhangsan kuai wo san-fenzhong.
Zhangsan fast me 3-minutes.

Zhangsan is faster than me by 3-minutes.

I think these examples don’t really challenge the generalization that the adjectives in
bare-comparatives are DAs. Although fat and fast can’t be directly modified by measure
phrases, they are still related to dimensions that are measurable, for instance weight, and
speed. In this sense, they should be categorized as DAs. As we will see in a moment that
fat and fast, as well as other DAs, can be categorized under one class by a negation test in
Chinese.

Based on Bierwisch (1989), the main difference between DAs and EAs is that
antonymous DA pairs such as long/short, large/small, refer to the same scale of a given
dimension and differ in the ordering on the scale, whereas antonymous EA pairs such as
pretty/ugly, lazy/industrious, clever/stupid, refer to different scales. For example, both
long and short are mapping individuals to a scale of length. Under a simplified view each
individual would correspond to a degree point on the scale, and we can order these
degrees along the scale. But depending on whether we choose a positive DA (e.g. long)
or a negative DA (e.g. short) there will be different ordering of the degrees. So if A is

longer than B and B is longer than C, we have the following ordering:

(17) A > long B >long C

But the above ordering is reversed if we order the three degrees along the dimension of

shortness, although the fact under description is still the same, as shown below:

168

(18) C >short B >short A

This explains why the following two statements are equal in truth values:

(19) Beethoven’s fifth symphony is longer than a Beatle’s song.

®A Beatle’s song is shorter than Beethoven’s fifth symphony.

On the other hand, the positive EA and the positive EA don’t necessarily refer to the
same scale. So although a short song nevertheless has some length, a lazy person is not

industrious to some extent. The following two statements are not equal either.

(20) John is more industrious than Mary. as Mary is lazier than John.

The first statement leaves open whether two people are industrious or not, but the second
statement implies both people are lazy to some extent. So the two statements in (20) don’t
have the same relation as the two statements in (19).

With this distinction in mind, I will now turn to a negation test that can
distinguish free adjectives and constrained adjectives. The test proceeds as follows: take
an adjective A, add a negation in front of it, then use the degree intensifier adverb hen

(very) to modify the negated adverb. Interestingly, the adjectives that can surface in bare-

169

comparatives are the adjectives whose negated versions refuse to be modified by hen.

Some examples are given below‘:

(21) Free adjectives
negated form: bu-gao/chang/cu/zhong

not-tall/long/thick/heavy

hen-modification: * hen bu-gao/chang/cu/zhong

very not-tall/long/thick/heavy

(22) Constrained adjectives:
negated form: bu-piaoliang/ganjing/shufir/qingkuai

not-pretty/clean/comfortable/industrious

hen-modification: hen bu-piaoliang/ganjing/shufir/qingkuai

very not-pretty/clean/comfortable/industrious

Since hen (very) in Chinese is a degree intensifier adverb that can modify any gradable
adjective, the contrast in (21) and (22) suggests that the negated adjectives in (21) don’t
behave as lexical adjectives, although they can surface as predicate negation, as shown in

(23). On the other hand, the negated adjectives in (22) are themselves lexical items that

 

' I intentionally only tested positive adjectives. For some reason there is no difference with respect to this
test among the negative adjectives, no matter they are from the DA group, such as ai (short), xi(thin), or
from the EA group, such as zang (dirty), chou (ugly). Similar facts hold for English, as discussed in Horn
(1989) p274-275.

170

can be modified again by hen (very). Note that the negated forms in (22) can also serve

as predicate negation, as shown in (24), which is not relevant for our purpose here.

(23) X bu gao/chang/cu/zhong
X not tall/long/thick/heavy
X is not tall/long/thick/heavy
(24) X bu piaoliang/ganjing/shufir/qingkuai
X not pretty/clean/comfortable/industrious

X is not pretty/clean/comfortable/industrious

I take the contrast in (21) and (22) to demonstrate the rule of morphological blocking.
Let’s take gao (tall) and piaoliang (pretty) as our examples. If the negated adjective not-
A refers to the same scale of the base adjective, but with an opposite direction of ordering
of the degrees on the scale, since gao(tall) and ai(short) are two lexical items co-existing
in the lexicon and they already map to the opposite orderings of the same scale, any
other lexical items that refer to the opposite ordering of gao(tall) on the same scale
should be blocked. On the other hand, piaoliang (pretty) and chou (ugly) are mapping
individuals to two different scales, therefore the existence of chou (ugly) will not block
not-pretty. The blocking rule works the same way in English. Although negative prefixes
are productive for English EAs, as shown in (253), they can never attach to the positive

English DAs, as shown in (25b).

171

(25) a. unhappy, unhealthy, imposSible, inconsistent, immoral, nonsymmetric

b. *un/im/in-tall, *un/im-long, ‘un/im- thick

Interestingly, the two semi-DAs earlier fat and fast pass our negation-test.

(26) a. *hen bu-pang
very not-fat

b. *hen bu-kuai

very not-fast

Now I can safely conclude that free adjectives are DAs, and constrained adjectives are
EAs. To put it in another way, bare-comparatives only license DAs, whereas bi-

comparatives license all kinds of adjectivesz. One remaining puzzle is that we don’t know

 

2 The following data seems to show that if the standard degree arguments are implicit, EAs are fine in
bare—comparati ves too:

(i) Ta piaoliang/xixing /gaoxing /youqu (*wo) yidian
She pretty /careful /happy linteresting me a little
She is a little more prettier/careful/happier/interesting (than me).

(ii) Zhe-jian-fangjian ganjing/shufu lmingliang (*na-jian) haoduo
This-CL-room clean /comfortable/bright that-CL much
This room is much cleaner/more comfortable/brighter (than that one).

Examples (i) and (ii) have to be uttered in a context that the standard degree arguments are provided by
context. For instance, (i) and (ii) can be uttered as answers to the question “compare Mary and me, who is
pretty?”. So the generalization is not that EAs can not occur in bare-comparatives per se, it is rather that
EAs can’t be followed by explicit standard degree arguments. Since the standard degree arguments stay in
different syntactic positions for bare-comparatives and bi—comparatives, we see the artifact that the choice
of adjectives is conditioned by the choice of a particular structure. The problem with this approach is that
the judgments are very vague and fuzzy as to whether (i) and (ii) have the comparative meaning “She is
prettier” without claiming any property of prettiness or the absolute meaning “She is pretty”. Since the
data is very unclear, I will not discuss it further.

172

why the two comparative forms are constrained by the lexical property of adjectives. I
will leave this problem open.
5.2. Chinese phrasal comparatives

As shown by all the examples we have looked at until now, the standard degree in
Chinese comparatives is in general expressed by a DP. Recall that we had some
discussion in Chapter one about the relation between phrasal and clausal comparatives. In
this section I will show that Chinese phrasal comparatives are better analyzed with a
direct analysis than a comparative deletion analysis. Since all the facts I will discuss
below apply equally to bi-comparatives and bare comparatives, and as discussed in the
last section, bi-comparatives are less constrained by adjective selections, I will primarily
use bi-comparatives as my examples. The particular kind of data we are looking at here is

as below:

(27) Zhang san bi Lisi gaoxing
Zhangsan than Lisi happy

Zhangsan is happier than Lisi.

Following the clausal analysis, the DP Lisi above has been analyzed as an elided
clause in Liu (1996). Xiang (2003) argued that for three reasons it is better to analyze Lisi
as a simple DP, not an elided clause. First, Chinese does not allow long distance wh-
movement. English clausal comparatives allow long distance wh-movement, as shown

below:

173

(28) John is taller than [cp Op; [cp Max thought [Cp Bill was t;]]]

Since in general Chinese allows long distance wh-movement, e.g. long distance wh-
movement in relative clauses (Huang 1982), one would expect that Chinese comparatives
would behave the same as (28), if they are really clausal, but this is not the case, as

shown in (29).

(29) *Zhangsan bi Wangwu renwei Lisi zuotian gaoxing.
Zhangsan than Wangsu think Lisi yesterday happy.

Zhangsan is happier than Wangwu thought Lisi was yesterday.

Second, Chinese does not allow subdeletion. Comparatives like (30) below are called
subdeletion in English. The biggest difference between subdeletion and other
comparatives is that it is a full clause that is contained in the comparative site. The
standard analysis of it is to posit a degree variable in the comparative site, which is bound

by a wh-operator, as shown in (31):

(30) The table is longer than the door is wide.

(31) The table is longer [than [or Wh; the door is d;-much wide]].

Surprisingly, Chinese doesn’t allow subdeletions like (30). Example (32) below is bad in

Chinese.

174

(32) *Zhuozi bi men kuan chang
table than door wide long
The table is longer than the door is wide.
Notice that if the clausal analysis of Chinese comparatives is right, a comparative

sentence like (33) has an LF representation like (34):

(33) Zhuozi bi men chang
Table than door long

The table is longer than the door.

(34) Zhuozi [bi [CP Wh; men d;-much changk]] changk
table [than [CP wh; door d;-much longk]] long,

The table is longerk [than [cp wh; the door is d;-much longk] ]

Comparing (34) and (31), we see that they share the same mechanism of deriving the
standard degree: there is a degree variable bound by a wh-operator. The only difference
between (34) and (31) is that (34) contains an elided clause at the comparative site, so
reconstruction has to be used to recover the predicate, whereas in (31), no ellipsis is
involved, hence no reconstruction. As we have shown in (32), the counterpart of (3 1) in
Chinese is bad. Now we have a puzzle: if (34) is a valid representation, why is (32) out in
Chinese? One stipulation is that Chinese comparatives can host an elided CP, but not a

firll CP. However, it is surprising that a marked construction like ellipsis would be

175

preferred over an unmarked one like a normal clause, and there is no natural explanation
for this.

A third problem for the clausal analysis concerns the interaction between the
Chinese distributor dou and comparatives. I already had a discussion about dou in
Chapter 3. Recall that one usage of it is to behave as a distributor that distributes over a
plurality argument to its left (see Lin 1998, Wu 1999 among others). If the subject of the
clause is singular and not distributable, the presence of dou is not licensed, as shown in

(35a).

(35) a. *Zhangsan dou zai jia.
Zhangsan dou at home.
Zhangsan is all at home.
b. Mei-ge-ren dou zai jia.
Every-CL-person dou at home

Everybody is (all) at home.

In (35b), the subject is a strong quantifier, and dou is allowed to distribute over the
subject. Moreover, the presence of dou is obligatory in this situation, as shown by the

contrast between (a) and (b) in (36).

(36) a. Mei-ge-ren dou xihuan youyong.

Every-CL-person dou like swim

Everyone likes swimming.

176

b. *Mei-ge-ren xihuan youyong.
Every-CL-person like swim

Everyone likes swimming.

With this small background on dou in mind, let’s now turn to comparatives.
(37) Mei-ge-nanhaizi dou bi mei-ge-nuhaizi gao
Every-CL-boy dou than every-CL-girl tall

Every boy is taller than every girl.

In (37) there are two strong quantifiers, one is the subject of the matrix clause, the other
is inside the bi-phrase. Dou can be licensed by the first quantifier, i.e. the subject of the
matrix clause. The grammaticality of (37) is unpredicted by the clausal analysis. If every-
girl in (37) is the subject of an elided CP, as in (38), the sentence is wrongly predicted to

be bad just like (36b), because there is no dou in this CP.

(38) Mei-ge-nanhaizi dou [PP bi [cp mei-ge—nuhaizi e ]] gao
Every-CL-boy dou than every-CL-girl tall

Every boy is taller than every girl.

The clausal analysis would also predict that if we add another dou in the elided CP, it

could be licensed by the subject every-girl, and the sentence should be fine. This is an

incorrect prediction too, as shown in (3 9).

177

(39) * Mei-ge-nanhaizi dou [pp bi [Cp mei-ge-nuhaizi dou e ]] gao
Every-CL-boy dou than every-CL-girl dou tall

Every boy is taller than every girl.

Examples (37) to (39) strongly suggest a phrasal status of every girl in the comparative
site. However, there seem to be counter-examples. For instance, the presence of dou is

fine in (40a).

(40) a. Zhangsan bi mei-ge-ren dou gao.
Zhangsan than every-CL-person dou tall
Zhangsan is taller than everybody.

b. Zhangsan bi mei-ge-ren gao.
Zhangsan than every-CL-person tall

Zhangsan is taller than everybody.

On the surface, (40a) seems to support a clausal analysis of everybody, since if we

assume everybody is the subject of an elided CP, the detailed structure of (40a) is like

(41), and dou would be required in this CP.

(41) Zhangsan [pp bi [cp mei-ge-ren dou e ]]gao

Zhangsan than everybody dou tall

178

What is strange for this analysis is that (40b), which also has an LP like (41), isn’t bad,
although there is no dou following everybody. This is unexpected since the structure in
(41) predicts a contrast between the pair in (40), just like the contrast between the pair in
(36).

Now we face a dilemma. If we analyze (40a) as (41), we can’t explain the
grammaticality of (40b), on the other hand, if we don’t analyze everybody in (40a) as the
subject of a clause, why is dou licensed in (40a)? As has been noticed in the literature,
the licenser of dou needs to be at the left of dou (the leftness constraint, Lin 1998), but it
can be in various syntactic positions, for example, a topicalized or scrambed object, as in

(42), or an indirect object, as (43).

(42) a. Mei-ben-shu, Zhangsan dou du-le.
Every-CL-book, Zhangsan dou read-perf.
Zhangsan has read every book.

b. Zhangsan mei-ben-shu dou du-le.
Zhangsan every-CL-book dou read-perf.

Zhangsan has read every book.

(43) Zhangsan gei mei-ge-ren dou mai-le liwu.

Zhangsan for every-CL-person dou buy-perf. present

Zhangsan has bought present for everybody.

179

These facts suggest that (40) can’t provide evidence to show that the standard degree DP
is the subject of an elided CP, because dou not only licenses subjects to its left, it also
licenses objects to its left. This is an interesting fact for our discussion later, because as it
will turn out the standard degree in Chinese comparatives does behave like an object of
the predicate.

Based on these arguments, I will adopt the direct analysis position for Chinese
phrasal comparatives, namely the element denoting the standard degree is a DP and not a

clause.

5.3.The Structure of bare-comparatives
5.3.1. The differential degree argument

The differential degree argument measures the difference between the two
degrees under comparison. For instance, if John is taller than Mary, there must be a
difference between John’s height and Mary’s height. This point seems to be trivial,
because the difference between the two degrees comes for free as long as we have two
degrees to compare. Therefore one would think that we don’t have to include the
differential degree argument in our formal representation of comparatives. Moreover, in
many languages, e.g. English, the differential degree argument is also not obligatorily
expressed in narrow syntax. Therefore, the syntactic status of this argument has been
ignored most of the time. However, in this section I will argue that the differential degree
argument might not be as simple as has been assumed. First, it has crucial contribution to
semantic interpretations for some kind of comparatives; second, even for simple

comparatives, it is obligatorily expressed in some languages, for example, Chinese.

180

Kennedy (1997) discussed one kind of comparative that depends crucially on the
differential degree argument to derive the correct semantic interpretation, namely, the

comparison of deviation. Consider the contrast in (44):

(44) a. San Francisco Bay is more shallow than Monterey Bay is.

b. San Francisco Bay is more shallow than Monterey Bay is deep.

(44a) is the kind of comparative we normally see. It only compares the depth of two bays.
There is no implication about the actual depth of them. They could be both very deep or
shallow, or one deep and the other shallow. As long as San Francisco Bay is more
shallow, the sentence is true. However, (44b) implies that San Francisco Bay is shallow
and Monterey Bay is deep, and what is being compared are the two differential degrees,
namely, the degree to which San Francisco Bay is shallow and the degree to which the
Monterey Bay is deep. The first degree is the differential degree between the actual depth
of the bay and the commonly assumed standard depth that can be called shallow, and the
second degree is the differential degree between the actual depth of Monterey Bay and
the standard depth that can be called deep. One can’t derive the correct interpretation of
(44b) without employing the differential degree argument.

One would ask except for the complex cases such as (44b), do we need the
differential degree argument in our formal representations? Beck (1997) briefly suggested
that even for simple comparatives, we should encode a differential degree argument in
the semantic representations. For instance, although there is no explicit differential

degree argument for (453) the two sentences in (45) could have the same semantic

181

representation as in (46), in which the degree head —er takes a differential degree

argument d.

(45) a. Luise is taller than Otto is.

b. Luise is 3 cm taller than Otto is.

(46) [[-er]] (D|)(d)(D2) =1 iffthe max d2D2(d2) = (I + the max d1D1(d|)

Schwarszchild and Wilkinson (2002) also highlighted the function of a differential degree
argument in deriving semantic interpretations. But as far as I know the syntactic status of
this argument hasn’t attracted much attention. Fortunately, Chinese provides us an
opportunity to see that the differential degree argument is indeed a non-trivial syntactic
argument, even if sometimes it doesn’t appear in the surface syntax explicitly.

The measure phrase, which expresses the differential degree, is always obligatory
in Chinese bare comparatives, as shown in (47). This is very different from their English
counterparts, where measure phrases are always optional, as shown in (48). This suggests

that Chinese requires an explicit spell-out of the differential degree argument.

(47) Zhangsan gao wo *(yicun/yidian)

Zhangsan tall me one-inch/a little

“Zhangsan is one inch/ a little taller than me”.

(48) John is (two-inches) taller than me.

182

The obligatory differential degree argument also plays a role in sentences that have
gradable adjective as predicates, but don’t involve explicit comparison. As shown in (49),
in Chinese, a bare adjective usually can’t be used as a predicate unless it is modified by

some degree phrase.

(49) a. ??Ta gao
He tall

He is tall

b. Ta hen/you dian er/ feichang/bu zenme/tai gao
He very/a little /very/ not how / extremely tall

He is very/ a little / not very much/extremely tall

This has been considered merely as an idiosyncratic property of Chinese adjectives. But
by taking the differential degree argument into account we can give a more principled
explanation for it. Following Kennedy (1997), we take the adjectival predicate in (49) as
a comparative construction too, but with an implicit standard degree, e.g. a degree of
height by most people’s standard to be called tall. When we say x is tall, we mean if the
standard height d that we consider to be tall is six feet, x is at least six-feet tall, or
probably taller. The degree modification in (49b) actually marks the difference between

x’s height and d. If the difference is big, he is very/extremely tall is true; if the difference

183

is small, he is a little tall is true. Therefore, (47) and (49) actually pattern the same in the
sense that they both require an obligatory differential degree argument.

The discussion so far suggests that the differential degree argument is a necessary
syntactic argument for comparative constructions. A comparative predicate is actually a
three-way predicate: it relates an external argument, the subject, to its two internal
arguments, the standard degree argument and the differential degree argument. Next I
will turn to the question how to represent this three-way predicate in the syntactic

structure.

5.3.2. The VP-shell structure for double object constructions

Based on the discussion in the previous section, the argument structure of comparatives,
therefore, is similar to a double object construction, in the sense that they both have three
syntactic arguments, one external argument and two internal arguments. In this section I
will show that comparatives indeed share some structural similarities to double object
constructions.

The structure of double object has generated lots of discussion in the literature.
Without doing justice to all the alternatives, I will assume the VP-shell structure in
Larson (1988) and variants of the same kind (e.g. Johnson 1991) for double object
constructions. Under a VP-shell analysis, the sentence in (50) contains a VP structure as
in (51):

(50) John sent Mary a letter.

184

(51) VPt

/\
v9
/\
V VP2
send; /\
DP; V’
Mary /\
V’ DP
/\
V DP 3 letter

1; ti

 

 

 

 

 

In (51), the verb head takes the indirect object as its complement at the early stage of the
derivation, namely, at the lower VP shell. The verb then undergoes head movement to the
higher VP shell, and the indirect object also moves up through XP movement. One
important insight this analysis captures is that we end up having a small clause like
structure for the two objects (also see Johnson 1991, Kayne 1984). That is to say, at some
point in the derivation the indirect object becomes the subject of a small clause-like

structure, and asymmetrically c-commands the direct object, as shown in (52):

(52) ...VP2
/ \

DP;0 V’
Mary / \
V DPDO
send a letter
Larson argued that the asymmetric c-command relation in (52) accounts for a number of
important asymmetries between two objects, as discussed in Barss and Lasnik (1986).

First, since reflexives and reciprocals must be c-commanded by their antecedents, (52)

explains the asymmetric pattern with respect to anaphor licensing in (53):

185

(53) a. I showed Mary herself.

b. *I showed herself Mary.

Similarly, a bound pronoun must be c-commanded by its binder, so the asymmetry in

(54) is explained too.

(54) a. I gave every worker; his; paycheck.

b. *1 gave its; owner every paycheck;.

Thirdly, it is known that a wh-phrase can’t move over a c-commanding phrase if the c-
commanding phrase contains a pronoun that is co-referential with the wh-phrase. This is
called a weak crossover effect. In (55), the ungrammaticality of (55b) indicates that the
wh-phrase must have moved fi'om a position c-commanded by the indirect object his
mother. Whereas (55a) is grammatical meaning that the wh-phrase is moved from a

higher position than the direct object his paycheck.

(55) a. Which man; did you send t; his; paycheck?
l J

 

b. ‘Whose; pay did you send his; mother t; ?
l 1

Finally, negative polarity items are asymmetrically licensed too in double object

constructions, as in (56).

186

(56) a. I showed no one anything.

b. *I showed anyone nothing.

So the asymmetric relation between the two objects is a crucial property that any
analysis should capture. Interestingly, Chinese comparatives show some similar variable
binding facts that indicate the standard degree argument should asymmetrically c-

command the differential degree argument.

(57) Zhe-gen shengzi chang na-gen shengzi yiban
This-CL rope long that-CL rope half

This rope; is longer than that rope; by half (of that rope*;/;)

In (57), the differential degree argument half can be called a relational noun, because it
has to be interpreted with respect to something else, i.e. one can’t talk about halfby itself,
it has to be half of something. Relational nouns are usually analyzed as containing an
implicit argument that is bound by an antecedent (Barker 1991). In the case of (57), that
means the standard degree that rope is binding the implicit argument variable contained

in half, as shown in (58):

(58) that rope;. . ..[e; half]

The same fact also holds for bi—comparatives, as shown in (59). Since the variable

binding relation requires asymmetric c-command, (57) and (59) suggest comparatives, at

187

least Chinese comparatives, have a structure similar to double object construction as in

(52).

(59) Zhe-gen shengzi bi na-gen shengzi chang yiban
This-CL rope bi that-CL rope long half

This rope; is longer than that rope; by half (of that rope*;/;)

Notice that since the variable binding relation requires asymmetric c-command, if the c-
command relation in (58) doesn’t hold, we lose the binding interpretation too, as shown
in (60). In (60), that rope doesn’t c-command half. It is the head of the relative clause the
box that c-command half. Therefore, we only have an interpretation that it is the half of
the length of the box, not that rope. (60) is a bi-comparative, and bare comparatives work

the same way.

(60) Zhe-gen shengzi bi zhuang na-gen shengzi de hezi chang yiban
This-CL rope bi contain that-CL rope part. box long half
This rope is longer than [the box that contains that rope;] ; by half (of that

rope*;/;)

To summarize, in the previous section I showed that a comparative predicate actually has
two internal syntactic arguments, the standard degree argument, and the differential
degree argument. This raises the possibility that they are similar to a double object

construction, which has two internal arguments too. In this section, we see further

188

evidence that the standard degree argument and the differential degree argument have to

stand in an asymmetric c-commanding relation3.

5.3.3.The DegP-shell structure for bare-comparatives

As discussed briefly at the end of Chapter 4, Larson’s analysis of comparatives in terms
of a DegP-shell structure intuitively looks like the VP-shell analysis of double object
constructions. The discussion in the previous two sections in this chapter suggests that
comparatives and double object construction do share some structural properties.
Therefore, Larson’s DegP-shell structure looks very promising to capture the structure of
comparatives. However, instead of completely adopting Larson’s analysis, I will argue
that some revisions have to be made.

Larson’s DegP-shell structure for comparatives is repeated in (128):

 

(128) IP
/\
DegP
Deg’

/\

Deg DegP

“er /\

AP Deg’
tall /\

Deg PP
t A
1 than Bill

 

g

3 However, one problem left unsolved is that the c-command relation, for some unknown reason, needs to
be very local. As shown in (22) and (24), although the subject this rope also c-commands half, there is no
binding interpretation between them.

189

As discussed in Chapter 4, one advantage of this structure is that it can be mapped onto
the standard semantics of gradable adjectives, namely, gradable adjectives involve
quantification over degree arguments. For instance, since the adjective tall takes two
argument, an individual argument and a degree argument (because tall maps an
individual to a degree of height), the sentence John is taller than Bill actually means John
is [-er than Bill] tall, with [-er than Bill] as the degree argument of the adjective. Since
the degree head —er and the than-phrase don’t form a constituent on the surface, we need
a syntactic structure that views them as a constituent at some level of the representation.
In this sense, the DegP-shell structure in (128) is appealing, since at the lower Deg’
structure, the degree head —er and the than-phrase form a constituent. However, also
based on the semantics of the adjective tall, the degree argument [-er than Bill] should be
an argument of the adjective tall, and this is not reflected in the structure in (128). In fact,
what (128) shows is that the adjective is an argument of the degree head —er. To keep the
essence of the DegP-shell structure and at the same time make the degree argument an
argument of the adjective, I will suggest to revise the structure in (128) in a way that the
AP structure is sandwiched in between the two DegP structures. I use the following

example of Chinese bare comparatives to demonstrate the structure:

(61) W0 gao Lisi liang-cun

I tall Lisi two-inch

I am two inches taller than Lisi

190

IP

 

 

 

/\
I DegP
l \
Deg AP
(exceedk)-tall; /\
L181; A ’
/\
A DegP
(tallh /\
Lisi; Deg’
1
Deg DiffP
(exceed);( 2 inches

 

There is no explicit degree morpheme in Chinese comparatives. On the surface it seems
that the adjective tall itself expresses both the adjective and the comparative meanings.
Following the lexical decomposition approach (see Hale and Keyser 1993, 2002, Huang
1996, Lin 2001) that there exists in syntax phonetically empty heads that are decomposed
from single lexical items, I assume there is a phonetically null degree morpheme which I
call exceed. The degree morpheme merges with its two internal arguments first, the
standard degree argument and the differential degree argument, as shown by the lower
DegP structure in the tree. The degree morpheme internally merges with the adjective
through head movement, and the standard degree DP Lisi moves to the [Spec, AP]
position for EPP feature checking, thus we derive a small-clause like structure at the AP
level. Finally, in order to introduce the external argument, the complex head exceed-tall
moves to the higher Deg-head through head movement.

One empirical advantage of the revision made in (61) is that it also leads to a

natural explanation of the bi-comparatives, which I will turn to in the next section. In a

191

nutshell, by generating the AP structure in between the two DegP projections, bare
comparatives and bi-comparatives are united under one single structure through the head

movement.

SARI-comparatives

5.4.1. Head movement

Given that bare-comparatives and bi—comparatives express similar meanings, a natural
thought is that they are structurally related. There is an easy way to implement the idea.
In the structure of bare comparatives in (61), the Deg-head moves to the higher DegP
shell position to introduce its external argument. If at this point of the derivation, instead
of doing the head movement, the morpheme bi can be externally merged with the AP and

project the higher DegP-shell. In this way a bi-comparative is derived, as shown in (62).

(62) IP
/\
Zhangsan l’
/\
DegP
/\
Deg AP
bi /\
L181; A ’
/\
A DegP
exceed-tall /\
(Lisi); Deg’

 

 

Deg DiffP
(exceed). 2 inches

 

 

 

Notice that different from bare-comparatives, for bi-comparatives, the differential degree

argument is optional, as shown in (63).

192

(63) Ta bi wo gao (yidian)
He bi me tall (a little)

He is (a little) taller than me.

This is not surprising because of the idiosyncratic nature of the argument structure. It is
known that not all the arguments have to be explicitly expressed.

I will call the analysis in (62) a unified analysis, since it is derivationally related
to the bare-comparatives. This is the analysis I am going to pursue. Conceptually a
unified analysis has obvious advantage because it reduces different patterns of
comparatives to one single syntactic structure. However, empirically it faces a few
challenges. The most important challenge to the unified analysis is that there seem to be
independent reasons for us to believe that the bi-phrase should be analyzed as an adjunct
(e.g. Liu 1996, Xiang 2003). As shown below, if the adjunct analysis is correct, bi-

comparatives are very different from bare-comparatives in structure.

(64) Adjunct analysis

IP
/ \
Zhangsan I’

/\
I AP
/\
PP AP
A gaoxing (happy)
bi (than) Lisi

193

This section is organized as following. First I will review the existing and potential
arguments for the adjunct analysis of bi-comparatives and show that none of these
arguments are really strong enough to motivate the adjunct analysis. I will also provide
evidence to show that the adjunct analysis is empirically inadequate. Second, I will
discuss some welcome consequences if we adopt the unified analysis.
5.4.2. Against the adjunct analysis

On the surface, bi—comparatives are similar to English than-comparatives, except
that bi-phrase stays before the adjective, whereas than-phrase stays after the adjective.
Liu (1996) analyzed bi—phrase as a left adjunct of the predicate, and Xiang (2003)
adopted this position. There are two basic arguments for the adjunct status of bi-phrase,
as shown below. However, I will show that these tests are not decisive.

First, the coordination test seems to show that bi—phrase is a constituent, and it is

taken by Liu (1996) as an argument to show that the bi-phrase is a prepositional adjunct.

(65) Zhangsan bi Lisi huozhe bi Wangwu gao
Zhangsan bi Lisi or bi Wangwu tall

Zhangsan is taller than Lisi or Wangwu.

However, there are other ways to explain (65) without appealing to (64). To see why the
alternative might be more reasonable, let’s first look at the category of bi. Based on the
similarity between bi and English than, it is tempting to draw the conclusion that bi is a
preposition in Chinese. However, a closer look at the data shows that this conclusion is

artificial. Instead of being a preposition, bi behaves more like a verb. For instance, bi can

194

be used alone as a verb to mean compare, as shown in (66). Since (66) is describing a
perfect event, bi is followed by an existential perfect marker guo, which ofien attaches to

verbs.

(66) a. Ni bl-guo na-liang-ben shu de jiaqian ma?
You compare-perf. that-two-CL book DE price Q

Have you compared the price of those two books?

b. Ta lao na wo he Mary bi.
He always take me and Mary compare

He always compares me with Mary.

Also, when used as a bound morpheme, bi can be combined with other morphemes to

form verbs. As shown in (67), these verbs often have the meaning of comparison of some

kind.

(67) a. bi-jiao
compare-compare
compare

b. bi-shi
compare-try

compete

195

c. pan-bi
climb-compare
compete
According to these arguments, I will assume that bi is a verb, not a preposition. However,
when used in comparative constructions, bi seems to lose the status of a main verb. For
instance, different from (66a), aspect markers need to follow adjectives, instead of bi, in

comparatives, as shown in (68).

(68) Ji nian bu-jian, ta bi(*-le) wo gao-Ie hau-duo
several year not-see, he bi me tall-perf. much

After not seeing him for a few years, he became much taller than me.

The current analysis can handle (68) easily. In the shell structure, bi is not a complete
verb anymore, instead, it forms a complex predicate together with the adjectival head and
the lower degree head at LF though head movements.

Assuming the verbal status of bi and the shell structure of comparatives, let’s look
at the coordination fact in (65) again. There are at least two ways to derive (65) under the

current analysis. It could be done though coordination plus deletion, as shown in (69).

(69) Zhangsan [bi Lisi goo] huozhe [bi Wangwu gao]

Zhangsan [bi Lisi tall] or [bi Wangwu tall]

Zhangsan is taller than Lisi or Wangwu.

196

In this way, the two constituents that or relates are DegPs, not PPs. Alternatively, it could
be derived through a rightward across the board movement (ATB movement), as shown

in (70).

(70) Zhangsan [[bi Lisi geek] huozhe [bi Wangwu geek] gaok]
Zhangsan [[bi Lisi tel-11‘] or [bi Wangwu talk] tallk]

Zhangsan is taller than Lisi or Wangwu.

I will leave it open which implementation is more correct, but it at least shows that the

coordination test does not necessarily mean that the bi-phrase is a prepositional phrase.
The second argument for the adjunct analysis is that the relative position between

the bi—phrase and other VP adverbs is flexible. Since Chinese has relatively free order

between VP adverbs, this fact is taken to mean bi-phrase is an adjunct too.

(71) a. Zhangsan (zhongshi) bi Lisi (zhongshi) gao yidian
Zhangsan (always) bi Lisi (always) tall a little

Zhangsan is always a little taller than Lisi.

b. Zhangsan (bu) bi Lisi (bu) gao duoshao.

Zhangsan (not) bi Lisi (not) tall much

Zhangsan isn’t much taller than Lisi

197

The position of adverbials is notoriously flexible, depending on the scope of the adverbial
modification, as well as the type of the adverbials. It is true that if the bi-phrase is an
adjunct, the above data could be explained as the free ordering among adjuncts. But note
that this argument is only valid if the position of an adjunct is only free relative to another
adjunct, which is not true. For instance, an adjunct can often appear either before or after

the verb phrase in English.

(72) a. She quietly left.

b. She left quietly.

It is true that in Chinese adjuncts often precedes the verb phrase, as shown below:

(73) Ta qiaoqiaode likai-le (*qiaoqiaode)
she quietly leave-perf. (*quietly)
She quietly left.

However, it is also common that adjuncts can appear between a verb phrase and a higher
functional projection, and the relative position between an adjunct and the higher
functional projection is flexible. Consider the well-known BA-structure in Chinese. The

functional morpheme BA triggers the SOV word order, as shown below:

(74) a. Ta chi-le na-ge pingguo. (Without BA, 3V0)

she eat-perf. that-CL apple

198

She ate that apple.
b. Ta BA na-ge pingguo chi-le (with BA, SOV)
She BA that-CL apple eat—perf.

She ate that apple.

Some analyses have treated BA as a functional morpheme that heads its own projection
(Sybesma 1999, Zou 1993), some have treated it as a verb (Hashimoto 1971; Ross, 1991;
Bender 2000). Analyses also vary in many other respects, for instance, whether or not the
object DP is moved from a post-verbal position. I will leave these questions open, but
crucially for our purpose here, it is less controversial that BA is often analyzed as a head
that takes a VP as its complement, as shown below:

(75) XP

/\
x° VP

that apple VP
A
ate

Interestingly, adjuncts can appear either before BA, or between BA and part of the verb

phrase, as shown below.
(76) Ta (dakou—dakou-de) BA na-ge pingguo (dakou-dakou-de) chi-1e

She (big-bite) BA that-CL apple (big bite) eat-perf.

She ate that apple in big bites.

199

This suggests that adverbials in Chinese, have possible flexible positions not only relative
to other adverbials, but also to other fimctional projections. If this generalization is
correct, the data in (71) pose no problem for the current analysis anymore. Similar to the
BA-structures, in comparatives, adverbials could stay either before bi, or sandwiched
between bi and the lower part of the DegP-shell structure.

The adjunct analysis also makes the wrong prediction about binding. Recall that
in section 3.3.2., we saw the standard degree argument can bind the variable within the
differential degree argument (data repeated below). Since the binding relation is built
upon the c-command relation between the two arguments, if the standard degree is
generated within an adjunct, as shown in (78b), we can not derive the necessary c-
command condition to satisfy the binding requirement, contrary to the data shown in

(78a).

(78) a. Zhe-gen shengzi bi na-gen shengzi chang yiban/san-fen-zhi-yi
This-CL rope bi that-CL rope long half/a third

This ropei is longer than that rope,- by halfla third (of that rope*i/j)

b. IP
/ \
this rope I’
/ \
I AP
EP AP
long tj half/a third
bi (than) that rope,-

binding

200

5.5.Some consequences of a DegP-shell analysis

5.5.1. Reduplication of the adjectival predicate

In the DegP-shell analysis, the bare comparatives are related to bi-comparatives through
head movement of the adjectival predicate. Let’s take a look at the bare comparative
again. After the complex head [(exceed)-tall] undergoes head movement to the higher
DegP-shell position, there is a copy left in the AP position, as shown in (79a), the

structure is repeated in (79b):

(79) a. Yuehan gao Mali yidiandian
John tall Mary a little

John is a little taller than Mary.

b. John [Degpl [Dego(exceed)j-tallk]i [Ap (exceed),~ta-llq. [Degpz exceed,- a little]]]

Since in (79b), the head exceed is a null morpheme anyway, I will only refer to the copies
in the chain as higher tall and lower tall from now on. Normally in a chain of movement,
only the highest copy of the chain is pronounced (spelled out), and lower copies in the
chain are not. Therefore in (79a) only the higher tall is pronounced. Under the current
analysis, the bi—comparatives are derived by externally merging bi to the higher DegP-

shell position, as shown in (80):
(80) a. Yuehan bi Mali gao yidiandian

John bi Mary tall alittle

John is a little taller than Mary.

201

b. John [DegP] [0,30 bi] [Ap (exceed)j-tallk [Degm exceedj a little]]]

Comparing (79) and (80), if the lower exceed—tall in (79) could also be spelled out, we
would have a clear minimal pair in which only the highest position of the chain is
different. Before we look for such cases, let’s detour a little bit to the issue of spelling out
a chain formed by movement.

Under the copy theory of movement, the moved element leaves copies on its way
to the target, therefore we have a chain of copies. Usually only one copy on a chain
would be pronounced. In overt movement, it is usually the highest copy in the chain that

is spelled out, and other copies are deleted at PF. (81) is such an example of passives.

(81) a. John was invited John. (spell out the higher copy)

b. arm was invited John. (spell out the lower copy)

Nunes (1996, 1999) analyzes the reduction/deletion of copies on a movement chain as the
result of PF linearization considerations. Following the Linear Correspondence Axiom
(LCA) (Kayne 1994), if in a syntactic structure X asymmetrically c-commands Y, then at
PF X precedes Y. In the movement of (81), the higher copy John is asymmetrically c-
commanding the lower copy. To spell out both copies in (81) would give us a
contradictory result that John is preceding John itself, because the two are non-distinct
copies. However, some languages allow multiple copies on the same chain to be spelled

out, as shown below:

202

(82) a. German
Mit wem glaubst du mit wem Hans spricht
With whom think you with whom Hans talks
With whom do you think Hans is talking?
b. Frisian
Wer tinke jo wer’t Jan wennet
where think you where-that Jan lives

Where do you think that Jan lives?

Capitalizing on the proposal in Chomsky (1995) that LCA applies afler morphology and
hence doesn’t apply word-intemally, Nunes argues that the lower copy on a chain can be
spelled out because it has been turned into a phonological word by some morphological
process, and hence isn’t subject to LCA anymore. Specifically for the above examples,
Nunes suggests that the intermediate wh-copy is adjoined to an intermediate
complementizer and the morphology of these languages can convert the adjunction
structure [co choC°]] into a phonological word, as shown schematically in (83). Such
operation is most clearly shown by the Frisian data, where the intermediate
complementizer is cliticized to the wh-trace.

(83) CP

203

In (83), if the intermediate wh-word and the complementizer have been converted to one
word, and if LCA only linearizes strings at the word level, it won’t matter for the word
intemall wh-element. Therefore there is no reason the lower wh-element can’t be
pronounced.

Now let’s go back to comparatives. The following examples seem to suggest that
more than one copy involved in the head movement of comparatives (see (79)) can be

spelled out.

(84) a. Zhangsan gao Lisi gao bu liao yi diandian
Zhangsan tall Lisi tall not particle a little
Zhangsan isn’t much taller than Lisi.

b. Zhangsan bi Lisi gao bu liao yi diandian
Zhangsan bi Lisi tall not particle a little
Zhangsan isn’t much taller than Lisi.

c. Zhangsan gao Lisi jiu gao yi diandian
Zhangsan tall Lisi only tall a little
Zhangsan is only a little taller than Lisi.

d. Zhangsan bi Lisi jiu gao yi diandian
Zhangsan bi Lisi only tall a little

Zhangsan is only a little taller than Lisi.

In the bare comparatives (84a) and (84c), the adjectival predicates are reduplicated. This

is possible because in (84a) the lower copy of toll is housed within a reformed word tall-

204

not-particle, and in (840) the lower copy tall can be housed within the complex word
formed by the incorporation of focus only and the predicate tall. If this is correct, (84a)
and (840) reflect the head movement in bare comparatives, and (84b) and (84d) reflect
the alternative process to the head movement, i.e. bi-insertion. All together they support

the structural relation between bare comparatives and bi—comparatives.

5.5.2. The generic reading of the standard degree
Another difference between the bare-comparative and the bi-comparative is that
the standard degree in the bi—comparatives can have a generic reading, but the standard

degree in the bare-comparatives can not.

(85) a. Yuehan shenzhi bi changjinglu gao (yidian)
John even bi giraffe tall (a little)
John is even (a little) taller than giraffes/a giraffe
b. Yuehan shenzhi bi na-zhi changjinglu gao (yidian)
John even bi that-CL giraffe tall (a little)
John is even (a little) taller than that giraffe
(86) a. *Yuehan shenzhi gao changjinglu yidian
John even tall giraffe a little
John is even a little taller than giraffes/a giraffe
b. Yuehan shenzhi gao na-zhi changjinglu yidian
John even tall that-CL giraffe a little

John is even a little taller than that giraffe

205

In (85a), the bare singular DP girafle expresses a generic reading. This sentence isn’t
comparing John with a particular giraffe. Instead, John’s height is compared with the
standard height any giraffe would have, and John’s height exceeds that height. In
contrast, (85b) compares John’s height with the height of a particular giraffe, and the
sentence is fine too. On the other hand, if we use bare-comparatives to express the same
two sentences, as shown in (86), the standard degree has to be of a particular giraffe. The
generic DP in (86a) leads to ungrammaticality. The following examples demonstrate the

same point.

(87) a. Zhe zhi gang-chusheng de xiao luotuo bi ma da (yidian).
This CL just-bom DE little camel than horse big (a little).
This new-bom camel is a little bigger than a horse.
b. Zhe zhi gang-chusheng de xiao luotuo bi na-pu-ma da (yidian).
This CL just-bom DE little camel than that-CL-horse big (a little).

This new-bom camel is a little bigger than that horse.

(88) a. *Zhe zhi gang-chusheng de xiao luotuo da ma yidian.
This CL just-bom DE little camel big horse alittle.
This new-bom camel is a little bigger than a horse.
b. Zhe zhi gang-chusheng de xiao luotuo da na-pu-ma yidian.
This CL just-bom DE little camel big that-CL-horse a little.

This new-born camel is a little bigger than that horse.

206

So the descriptive generalization is that the bi—comparatives allow generic readings of the
DP that denotes the standard degree argument, but the bare comparatives don’t. In the
rest of this section I will offer an explanation for this contrast based on the syntactic
difference between the two comparatives.

DPs with generic interpretations make references to kinds instead of any

individuals or objects, as shown by the italic DPs in following examples:

(89) a. Monkeys live in trees.
b. The dragon comes from China.

c. A whole consumes a lot.

Generic DPs have a semi-universal quantificational force. The sentences in (89) roughly

mean (90):

(90) a. Vx (x is a monkey) (x lives in trees)
b. Vx (x is a dragon) (x comes from China)

0. Vx (x is a whale) (x eats a lot)
The reason that they are only semi-universal is because they allow exceptions. If among

all the monkeys, one monkey actually lives in the water, (89a) can still be judged to be

true. In the literature, a generic operator Gen is proposed for generic sentences (see

207

Kriflca. et al. 1995 for a review and discussion). The sentences in (89) actually have

semantic interpretations as in (91):

(91) a. Gen x (x is a monkey) (x lives in trees)
b. Gen x (x is a dragon) (x comes from China)

c. Gen x (x is a whale) (x eats a lot)

How to handle the possible exception cases in (89) and how to draw a fine grained line
between the universal operator and the generic operator are not issues of our concern
here. To simplify the discussion below, I will roughly assume that generic DPs do
involve universal quantification, as demonstrated in (90).

The second assumption of my analysis is that all predicates have an explicit event
argument. Since Davidson (1967), the event argument has been widely used to study
event structure. In the original proposal, only active verbs carry event arguments. I will
adopt a version of the Neo-Davidsonian analysis (Parsons 1990) and assume that all
predicates contain event arguments. So a sentence like (92) describes an existentially
quantified event of eating where John is the eater and an apple is the thing that got eaten
and slowly is predicated of this event.

(92) John ate an apple slowly.

3e [eat (e) & agent (e, John) & theme (e, an apple) & slowly (e)]

208

When the predicate is stative, such as (93), there is also an existentially quantified event
argument under the Neo-Davidsonian VICW4. Therefore (93) says that there is an event, it

is an event of being bald and the theme of the event is John.

(93) John is bald.

3e [bald (e) & theme (e, John)]

The event quantification in (93) is flat. The existential quantifier is simply
attached to other event components that are conjoined together. A refined structured
representation of event quantification has been proposed by a few authors. In general, in
the style of the tripartite structure of quantifiers proposed in Heim (1982), an event
quantification can also be represented as consisting of three parts: a quantifier, the
restriction of the quantifier, and the scope of the quantifier. For example, Herburger
(2000) maps focus structure to a tripartite event quantification, with the background
information in the restriction, and the focused assertion in the scope. Larson (2004)
suggests that sentence final adverbs are mapped into the scope of the event structure, and
they are predicates of events. The most well-known example is probably the mapping
hypothesis proposed in Diesing (1992). The problem she tries to answer is that there is a
contrast with respect to the interpretation of bare plurals at a subject position, depending

on whether the predicate is stage-level or individual-level. Consider (94) and (95):

(94) Stage-level predicate

 

" Note however, the View that state predicate has event argument is not uncontroversial (e.g. Katz 2003)

209

Firemen are available.

a. 3,, [x is a fireman][ x is available]

b. Gen,“ [x is a fireman and t is a time] [x is available at t]
(95) Linguists are intelligent.

a. *3, [x is a linguist][ x is intelligent]

b. Gen,“ [x is a linguist and t is a time] [x is intelligent at t]

The subject DP in (94) is ambiguous. Either it is existentially quantified, and the sentence
means that there are some firemen available at some time, or it is bound by a generic
operator and the sentence means that in general it is the property of firemen that they are
available. On the other hand, the subject DP in (95) only has the generic reading. (95)
only means that in general it is the property of linguists that they are intelligent.

To account for this contrast of stage-level and individual level predicates in terms
of the interpretation of bare plural subjects, Diesing suggests that there is a simple
mapping between the clausal structure and the logical representation, namely, the VP
structure consists of the nuclear scope, and the residue structure is the restriction, as
shown below:

(96)

\

 

 

Y. .
Restriction Scope

210

Diesing further assumes that subjects of stage-level predicates can be mapped into either
the restriction or the scope, as shown by the two subject positions above, but subjects of
individual level predicates must stay in the restriction. Crucially, only DPs that remain in
the scope will be bound by obligatory existential closure (Heim 1982), therefore receive
existential readings. The subjects of individual level predicate, being in the restriction,
will be bound by the generic operator in the text and that is why they are only interpreted
generically.

With this much background information, now we are ready to explain the
observation in (87) and (88). Recall that the descriptive generalization is that the bi-
comparatives allow generic readings of the DP that denotes the standard degree
argument, but the bare comparatives don’t. There are two main ingredients in the current
solution. First, I will suggest that the event quantification maps to the clausal structure of
the two kinds of comparatives differently. More specifically, I assume that the position of
the adjective decides the content of the scope. Since adjectives in bare comparatives and
bi—comparatives occupy different syntactic positions, the structure is consequently
mapped onto logical representations differently. For bare comparatives, the whole DegP-
shell structure, plus some higher functional projections, is included in the scope; but for
bi-comparatives only the lower part of the shell is included in the scope. Second, the
standard degree argument could be interpreted at the base position, or it could be
interpreted at the case checking position, and the case is checked at some functional

projection higher than the DegP-shell. This is shown below:

211

(97) bare comparatives

 

 

 

 

 

DPSD X’
Lisi /\
X DegP 1
(exceed)—tall /\
Deg AP
/\
t DegP 2
IjSi A
k . a littl/e
scope

(98) bi-comparatives

DPSD X’
Lisi /\

 

 

 

 

t A’
Lisi /\
J A DegP;
(exceed)-tall A
k a littlj
Y
scope

212

As shown above, at LF, the standard degree DPSD moves to [Spec, XP] to check case. In
bare comparatives, the adjective head (plus the phonologically null degree head) can
move to the higher functional head XO position. Assuming that the position of the
adjective determines the boundary of the scope of the event quantification, the case
checking position of the standard degree argument DPSD, as well as its base position, will
be included within the scope. That is to say, the standard degree argument always stays
within the scope. Assuming the default existential closure on DPSD within the scope area,
DPSD in bare comparatives will always be interpreted existentially. On the other hand, in
bi-comparatives, the functional head bi prevents the head movement of the adjective tall
to a higher functional position. Therefore, although the base position of DPSD is within
the scope, the case checking position is in the restriction. Consequently, DPSD could
either be interpreted existentially within the scope, or it could pick up some generic

operator in the restriction.

5.5.3. Summary

In this section I argued that comparatives have a DegP-shell structure. The two kinds of
comparative structures in Chinese are related through head movement. Recall that in
Chapter four, I suggested that a DegP-shell structure can help us to implement the
semantically motivated degree argument into syntax, because the degree argument can be
represented as one constituent in the lower DegP-shell structure. If this approach is on the
right track, we expect to see that the lower DegP-shell structure can move as a normal

argument. In the next section, I will provide evidence to show that by assuming that the

213

lower DegP-shell is the degree argument and it can move at LF, we can nicely explain

some problems.

5.6.Lower DegP as the degree arguments

In Chapter 4 I argued for the existence of a quantificational degree argument, but
we left the question open where the degree argument is in the structure. The data fi'om
Chinese comparatives supports the DegP-shell structure, similar to those proposed in
Larson (1991). If this is correct, the lower DegP structure would express the degree

argument. In this section I will provide some evidence for this.

5.6.]. The extraction of the standard degree
The standard degree in both bare comparative and bi-comparatives is resistant to

extraction, as shown in (99) and (100).

(99) a. Ta bi Zhangsan gao yidian
he bi Zhangsan tall a little
He is a little taller than Zhangsan.
b. *ta bi tj gao yidian DE renj
he bi tj tall a little DE person,-
the person he is a little taller than
(100) a. Ta gao Zhangsan yidian.
He tall Zhangsan a little

He is a little taller than Zhangsan

214

b. *ta gao tj yidian DE renj
he tall tj a little DE person,-

the person he is a little taller than

(99) and (100) show that relativization of the standard degree is impossibles. This might
be expected for bi-comparatives. It is possible that for independent reasons bi can’t be
stranded, and that could block the extraction of the standard degree. For this reason, I am
going to only discuss the extraction in bare-comparatives such as (99).

At first it looks like (99b) can be analyzed on a par with the extraction of an
indirect object from a double object construction. As shown in (101), wh-movements
such as relativization and wh-movement can not extract an indirect object from a double
object construction. Since the standard degree argument in the current analysis holds the
same structural position as an indirect object, the fact in (100) can be explained by the

same mechanism that bans extraction from a double object structure.

(101) a. *The person that you gave a book came to see you.

b. ‘Who did you give a book?

 

5 Note that wh-questions on the standard degree is perfectly fine. To include this data into our discussion,
we essentially need to talk about the debate with regards to whether wh-in-situ involves movement or not
(Huang 1982, Reinhart 1997). To avoid complicating the problem, I will put aside wh-questions for the
present discussion.
(1) Ta gao shei yidian.

He tall who a little

Who is he a bit taller than?

215

There are a couple problems indicating this can not be the whole story. First, as shown in
(102), in Chinese the extraction of an indirect object, although sounds odd, isn’t as bad as
the extraction of the standard degree in (100b).
(102) ??ta gei-le yi-ben-shu DE na-ge-ren

he give-perf. one-CL-book DE that-CL-person

the person that he gave a book
Further more, the extraction in Chinese double object constructions seems to be less
constrained than comparatives. For example, object shift in fine on the indirect object, as

in (103), but object shift of the standard degree is impossible, as in (104):

(103) Ta meigerenj dou gei-1e t j yi-ben-shu
He everybody j all give-perf. t j one-CL-book

He gave everybody a book.

(104) *Ta meigerenj dou gao t ,- yidian.
He everybodyj all tall t J- alittle

He is taller than everybody.

These facts suggest that the extraction of the standard degree argument in comparatives
can’t be reduced completely to its similarity to double object constructions. I will suggest
that the standard degree can’t be extracted because the extraction will lead to a Left
Branch Condition violation at LF. As shown below, at LF if we reconstruct moved

elements back to their base position, or alternatively interpret the lower copy of the

216

moved elements based on the copy theory, we derive a LF that contains a degree

argument, namely the lower DegP-shell structure.

(105) IP
/\
I TP
/\
DegP
/\
Deg AP
Emerita—tall; /\
Lisi,- A’
/\
A DegP
LISIJ' Deg’
/\
DiffP
exceed k 2 inches

The degree argument DegP is the complement of the adjective head. Following the
tradition that degrees are interpreted at the specifier of its adjective head (see Chapter 1),

I will assume that the degree argument DegP moves at LF to the [Spec, AP] position, as

shown below:
(106) AP
/\
Deng A’
A /\
LISI . . . . A tk
tall,

According to (106) above, the extraction of the standard degree Lisi will be ruled out as a

violation of the Left Branch Extraction.

217

One puzzle I don’t have much to say about is the standard degree is extractable in

English, as shown in (107):

(107) a. Who is John taller than?

b. The person that John is taller than left.

Since both Chinese and English comparatives involve quantification over a degree
argument, it is not obvious why the above analysis doesn’t block the extraction of the
standard degree in English as well. However, I want to point out that even in English, the
extraction of the standard degree is not completely fi'ee. This is clearly seen in ACD cases

like (108).

(108) *John is taller than everybody that I am.

[everybody that I am]j [John is taller than tj]

If the standard degree everybody that I am can move at LF, as shown in (108), ACD
should be resolved and (108) should have had a sensible interpretation that for every x I
am taller than, John is taller than x. However, (108) is ungrammatical. Notice that the
ungrammaticality in (108) doesn’t come from some independent properties of
comparative constructions that just block QR in ACD situations. In another slightly
different example (Chris Wilder, p.c.), we see that the standard degree can undergo QR

again.

218

(109) John is taller than more people than I am.
[more people than I am] ,- [John is taller than t j]

[more people than I am taller than] [John is taller than]

(109) is only slightly different from (108). But all of a sudden ACD can be resolved by
extracting the standard degree. (109) means that the number of people that John is taller
than is more than the number of people that I am taller than. It is mysterious why in
Chinese the standard degree in general is unextractable, whereas in English it is so only

in a very limited number of cases.

5.6.2. Intensifying the degree argument

Recall that in Chapter 4, I used the quantifier intervention effect at LF (Beck
1996) to examine the existence of the degree argument. In this section, we are going to
look at the same effect induced by the interaction of the degree adverb hen and the degree
argument.

We have come across the degree adverb hen several times earlier. The degree
adverb hen in Chinese can be used to intensify the degree, hence it is usually called
intensifier". For example, by adding hen to (110a), as shown in (1 10b), it means that the

difference between the two compared heights is not just a bit, it is quite a bit.

 

6 The use of hen in comparatives seems to be conditioned by regional variations. Some consultants accept
xiangdang (very), another morpheme similar to hen, but they don’t accept hen; some other consultants
don’t accept this kind of degree intensification at all. Since in the author’s speech, the hen-modification is ’
fine, the author informally consulted l6 informants in her home community at Yichang City, Hubei
Province, China. The informants were asked to judge the grammaticality of the sentences on a l to 5 scale,
with 5 as the best score. The scores of the core data used in this section are listed below, the last number in
each row is their corresponding position in the text:

219

(110) a. Ta gao wo yidian

he tall me a little

He is a bit taller than me.

b. Ta hen gao wo yidian
he very tall me a little

He is quite a bit taller than me.

The interpretation of (1 10b) suggests that the adverb is modifying the differential degree

a little. But how is the semantic interpretation achieved? As shown in (111), the adverb

hen can never directly modify the differential degree.

 

 

ta
he

gao wo yi-dian
tall me a little

4.4 (110a)

 

ta
he

hen gao wo yi-dian
very tall me a little

4.3 (11Gb)

 

ta
he

bi wo hen gao yi-dian
bi me very tall a little

4.6 (120b)

 

ta
he

hen bi wo gao yi-dian
very bi me tall a little

4.4 (120a)

 

ta
he

bi mei-ge-ren dou hen gao yidian
bi everybody all very tall a little

3.2 (1229)

 

ta
he

bi na-san-ge-ren dou hen gao yi-dian
bi those-three—people all very tall a little

4.0 (122D)

 

ta
he

ta
he

hen bi men-ge—ren dou gao yi-dian
very bi everybody all tall a little

hen bi na-san—ge-ren dou gao yi-dian
very bi those-three-people all tall a little

1.6 (1233)

2.0]123b)

 

ta
he

hen bi mei-ge-ren gao yi-dian
very bi everybody tall a little

1.7 (1258)

 

ta
he

hen bi na-san-ge-ren gao yi-dian
very bi those-three-people tall a little

2.9 (125g

 

220

(111) *Ta gao wo hen yidian
he tall me verya little

He is quite a little taller than me.

At least from the word order, it seems that hen is always an adverb on the predicate. I will
assume that hen is working as an intensifier to modify the whole degree argument. In
other words, by adding the degree adverb in (11%), the sentence not only describes the
degree to which the person is tall, but also intensifies that degree. I will suggest that hen
is a head of an adverb projection, situated on top of AP, and at LF the degree argument
moves to the specifier of that head. Through the [Spec, head] agreement relation between

hen and the degree argument, the degree argument is intensified. This derivation is shown

 

 

in (112).
DegP Ava
hen DegP
/\
Deg AP
(exceedQ-talli /\
mej A’

/\\

A DegP

t1 /\

tj Deg’

/\
Deg DiffP
tk a little

 

 

 

 

If the LF movement of the lower DegP is real, we expect to see the quantifier

intervention effect at LF (Beck 1996): a quantificational element in between hen and the

221

lower DegP should block this movement. The negation data below seems to support this

prediction.

(113) a. Ta hen gao wo yidian
he very tall me a little
He is quite a bit tallerithan me.
b. Ta bu gao wo duoshao
He not tall me a bit
He isn’t taller than me by much.
c. *Ta hen bu gao wo duoshao
He very not tall me abit

He really isn’t taller than me by much.

The example in (113a) is the same as (112). Negation by itself is fine with comparatives,
as shown in (lOSb). However, when we combine (113a) and (1 13b), the sentence
becomes ungrammatical, as shown in (1130). Compare the ungrammatical (113c) with
the grammatical example (113a) and (113b), the only difference is that the negation7 is
intervening between hen and the DegP. Crucially (1130) is not semantically
uninterpretable. It can mean that the differential degree between the two people’s heights
is not just small, it is very small. Under the current analysis, the ungrammaticality of
(113c) is expected because LF movement of the degree argument is illicit, as shown in

(114):

 

7 Yidian in (1133) and duoshao in (1 13b) means the same thing, but they are conditioned by
positive/negative environment, so we can think duoshao as the NPI item.

222

(114) [rhea .. Neg... Degree argument (illegitimate movement)
I

 

The current analysis also predicts that if we combine (113a) and (113b) with a different
word order, namely, if the negation is preceding hen, the output should be fine.
Unfortunately this prediction is not borne out. As shown in (115), when the negation is
preceding hen, there is nothing intervening between the adverb hen and the degree
argument. Therefore nothing can block the LF movement of the degree argument, as

shown in (116), and the sentence should be fine, contrary to fact.
(115) *Ta bu hen gao wo yidian/duoshao
he not very tall me a bit

He isn’t taller than me by much.

(116) ...Neg...Ahen...Degref argument (legitimate movement)

 

The ungrammaticality of (1 15) is puzzling under the present analysis. Does (115) suggest
that the degree adverb hen and the negation are just not compatible for some reason,
therefore (113) is not an argument for the movement of the degree argument after all? Or
we can keep the current analysis for (l 13), and rule out (115) on other independent
grounds? I will argue for the second possibility.

Notice that independent of explicit comparative constructions, hen and negation

are compatible and the relative order between the two is flexible, as shown in (117). This

223

rules out the possibility that (1130) and (115) are both ungrammatical because of the

incompatibility between hen and negation.

(117) a. Ta hen bu gaoxing.
he very not happy
He is very unhappy.
b. Ta bu hen gaoxing.
he not very happy

He is not very happy

Interestingly, as shown by the English translations above, the different word order
between the two seems to give rise to different interpretations. In (117a), the negation is
on the adjective, whereas in (117b) the negation is on the adverb very. Following Huang
(1988), Ernst (1995), Chinese negation cliticizes into the element it precedes. Therefore
in (117a) not-happy actually forms one complex morpheme, and so does not-very in

(1 17b). Following the same line, the negation in (113c) and (115) cliticizes to the
following elements too, which gives us a revised version of (l 14) and (116), as shown

below:

(118) .. .hen. .. Neg-X. . . Degree argument (Neg cliticized into the following element X)

(119) ...Neg—hen. . .Degree argument (Neg cliticized into the following hen)

224

Now let’s see what happens when at LP the degree argument moves to the specifier of the
Ava headed by hen. The movement in (118) is ruled out in exactly the same way as
before: the movement is blocked by the quantificational element, i.e. the negation, on the
way. What is interesting is that in (119) since the negation and hen are merged together
now as a complex head, in order for the degree argument to move over hen, it has to
move over the negation too. This is impossible because as I mentioned in Chapter 4, for
some unknown reason the degree argument can not take scope over negation (Kennedy
1997)

We can use the intensifier hen in bi-comparatives too. As shown below, the
intensifier takes flexible positions: it can be situated higher than bi, which means it is
higher than the high DegP-shell, or it can be situated lower than bi, but higher than the
low DegP-shell. In either case, under the current analysis, the degree argument (e.g. the
lower DegP-structure) moves to the specifier of the hen head, as shown by the examples
below.

(120) a. Zhangsan hen bi wo gao yidian
Zhangsan very bi me tall a little

Zhangsan is much taller than me.

225

Ava

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

/\
DegP Adv’
/\
hen DegP
/\
Deg AP
bi /\
mej A,
/\-
A DegP
[lexceedQ-tall /\
tj Deg’
/\
DiffP
tk a little
b. Zhangsan bi wo hen gao yidian
Zhangsan bi me very tall a little
Zhangsan is much taller than me.
DegP
/\
Deg XP
bi /\
mej Ava
/\
DegP Adv’
/\
hen AP
/\
A DegP
[lexceedQ-tall /\
tj Deg’
/\
DiffP
tk a little

 

 

 

 

 

 

What is interesting for our purpose is that in bi-comparatives we have an opportunity to

add a quantificational structure dou-structure in between the two DegP-shells, hence it is

226

interesting to see if that makes a difference to the high and low position of hen. As
discussed in Wu (1999), dou is a functional head that projects its own structure. It is also
considered to be quantificational, so it can distribute over the DP to its left (Cheng
1995). When the standard degree is distributable, e. g. everybody, the three people, dou

can be used, as shown below:

(121) 3. Ta bi mei-ge-ren dou gao yidian.
He bi every-CL-person all tall a little
He is a little taller than everybody.
b. Ta bi na-liang-ge-ren dou gao yidian.
He bi that-two-CL-person all tall a little

He is a little taller than those two people.

When hen is low, there is no problem to add hen to sentences in (121), as shown in (122):

(122) a. Ta bi mei-ge-ren dou hen gao yidian.
He bi every-CL-person all very tall a little
He is quite a bit taller than everybody.
b. Ta bi na-liang-ge-ren dou hen gao yidian.
He bi that-two-CL-person all very tall a little

He is quite a little taller than those two people.

227

However, when hen is high, it becomes odd to add hen to sentences in (121), as shown in

(123):
(123) a. ??Ta hen bi mei-ge-ren dou gao yidian.
He very bi every-CL-person all tall a little
He is quite a bit taller than everybody.
b. ??Ta hen bi na-liang-ge-ren dou gao yidian.

He hen bi that-two-CL-person all tall a little

He is quite a bit taller than those two people.

This contrast is easily explained under the present analysis that the degree argument in
the low DegP-shell position is a quantificational argument and it moves to the specifier of
hen to intensify the degree. (122) is fine because the movement of the lower DegP
doesn’t cross the quantificational dou, but in (123) dou is intervening between hen and
the low DegP, so the LF movement will be blocked. Let’s demonstrate the two

derivations by the diagrams below:

(124) a. douAhen degree] argument (the movement is fine)

a. ..Ahen drnrf .degref argument (the movement is blocked by dou)
/ \

Interestingly if dou is dropped fiom (123), mei-ge-ren (everybody) still leads to
ungrammaticality, but na-liang-ge-ren (those two people) is much better, as shown by the

contrast below.

228

(125) a. ??Ta hen bi mei-ge-ren (*dou) gao yidian.
He very bi every-CL-person all tall a little
He is quite a bit taller than everybody.
b. Ta hen bi na-liang-ge-ren (*dou) gao yidian.
He hen bi that-two-CL-person all tall a little

He is quite a bit taller than those two people.

The result in (1253) is not surprising. Although dou is absent, everybody is still
quantificational and can block the degree argument movement just like dou does, as

shown in (126).

(126) ”A hen. . .everybody. . .cllegree argument (the movement is blocked by everybody)
What is interesting in (125b). DPs such as those two people are clearly referential, and
the referentiality might not induce blocking effect as much as pure quantificational DPs.
Beck (1996) also observes similar distinction between pure quantificational and
referential DPs. I will leave the exact nature of this distinction open for firrther
investigations.

To summarize, in this section I have argued that the degree intensifier hen triggers
LF movement of the degree argument. This movement is blocked if there is another

quantificational structure standing in between hen and the degree argument.

229

5.7. Conclusion

In this chapter I have shown that independent evidence suggests that Chinese
comparatives are best characterized as having a DegP-shell structure. As a consequence
of this analysis, the degree argument now can be implemented into syntax as the lower
DegP-shell structure. The existence of the degree argument as the lower DegP-shell is
supported by extraction facts in Chinese, and more importantly, the LF intervention effect

in both Chinese and English.

230

Chapter 6

Concluding Remarks

In this thesis I investigated some issues on the semantics and syntax of
comparatives. Although the domain of my investigation is far from complete, I hope I
have drawn a basic picture that can capture our general intuition of how comparatives
work. First, comparatives express relations between individuals and degrees. Parallel to
other simple adjectival predicates, a gradable adjective in comparatives also relates an
individual to its degree of some dimension. Syntactically speaking, a DegP-shell structure
can capture the relation among different arguments and the degree morpheme. Second,
the standard degree argument, which is part of the degree argument, can be encoded
either as an elided clause, or as a simple DP. This syntactic property interacts with the
semantic property of quantifiers and gives rise to unexpected wide scope interpretation of
quantifiers. Thirdly, the inherent scale introduced by gradable adjectives, interacting with
the semantics and pragmatics of indefinite negative polarity items, makes comparatives a

licensing environment of NPIs.

231

Bibliography

Abney, Steven. 1987. The English Noun Phrase in its Sentential Aspect. Doctoral
dissertation, MIT, Cambridge, MA.

Aoun, Joseph; Li, Yen-hui Audrey, 1993. Syntax of Scope, MIT press
Bach, Emmon. 1986. The Algegra of Events. Linguistics and Philosophy. 9. 1. 5-16.
Barker, Chris. 1991. Possessive Descriptions. Ph.D. thesis. UC-Santa Cruz.

Barss, Andrew; Lasnik, Howard. 1986. A note on anaphora and double objects.
Linguistic Inquiry. l7. 2. spring. 347-354

Beck, S. 1996. Quantified structures as barriers for LF movements. Natural Language
Semantics 4, 1, 1-56.

Beck, S. 1997. On the semantics of comparative conditionals. Linguistic and Philosophy.
20, 3, 229-271

Beil, F. 1997. The Definiteness Effect in Attributive Comparatives. In: Proceedings of
SALT VII, 37-54

Bender, Emily. 2000. The syntax of Mandarin Ba: Reconsidering the Verbal Analysis.
Journal of East Asian Linguistics 9(2). pp.105-145.

Bhatt, R and Pancheva R. 2004. Late merger of degree clauses. Linguistic Inquiry 35: 1-
46

Bierwisch, M. 1989. The Semantics of Gradation, in M. Bierwisch and E. Lang (eds)
Dimensional Adjectives, Springer-Verlag, Berlin.

Brame, M. K. 1983, Ungrammatical notes 4: Smarter than me, Linguistic Analysis, 12, 2,
323-338

Bresnan, J. W. 1973, Syntax of the comparative clause construction in English, Linguistic
Inquiry, 4(3), 275-343

Brisson, Christine. 1998. Distributivity, Maximality, and Floating Quantifiers Ph.D.
thesis, Rutgers University.

Carlson, G. N. and Martin, L. W. 1975. This Antecedent isn’t the Right One. 9. 1. 13-24.

Chen, Liping, 2004a, Dou (Dis)harmony in Chinese, talk presented at the 34th NELS

232

Chen, Liping, 2004b. Dou Revisited: Distributivity and Presupposition. ms. Rutgers
University

Cheng, Lisa Lai Shen, 1991. On the Typology of Wh-Questions. Ph.D. Thesis, MIT.

Cheng, Lisa Lai-Shen, 1995, On Dou-quantification, Journal of East Asian Linguistics, 4,
3, 197-234

Chierchia, G. 1993, Questions with Quantifiers, Natural Language Semantics, 1, 2, 181-
234

Chomsky, Noam. 1977. On wh—movement. In Formal Syntax. eds. P. Culicover, T.
Wasow, and A. Akmajian. New York: Academic Press.

Chomsky, Noam. 1995 The Minimalist Program. Cambridge, Mass: The MIT Press.

Corver, N. 1990. The Syntax of Left Branch Extractions. Ph.D. dissertation, Tilburg
University.

Corver, Norbert. 1997. The internal syntax of the Dutch extended adjectival projection.
Natural language and linguistic theory 15, 2: 289-368

Cresti, D. 1995, Extraction and Reconstruction, Natural Language Semantics, 3, l, 79-
122

Cresswell, M. J. 1976. The semantics of degree. In Montague grammar, ed. Barbara
Partee, 261-292. New York: Academic Press.

Davidson, D. 1967. The Logical Form of Action Sentences," in N. Rescher (ed.) The
Logic of Decision and Action :81-120. Pittsburgh: University of Pittsburgh Press.

Diesing, M. 1992. Indefinites. MIT press.
Engdahl, E. 1986, Constituent Questions, Kluwer, Dordrecht

Ernst, Thomas, 1995. Negation in Mandarin Chinese, Natural Language & Linguistic
Theory, 1995, 13, 4, Nov, 665-707

F auconnier, G. (1975) Polarity and the Scale Principle. Chicago Linguistics Society 1 1,
188-199.

Frampton, J. 1990. The Fine Structure of Wh-Movement and the Proper Formulation of
the ECP. Ms., Northeastern University.

Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non)veridical Dependency. John
Benjamins, Amsterdam-Philadelphia. 281 pp.

233

Giannakidou, Anastasia. 1999. Affective Dependencies. Linguistics and Philosophy
22: 367-42 1 .

Giannakidou, Anastasia. 2001. The meaning of free choice. Linguistics and Philosophy
24: 659-735 .

Grimshaw, Jane, 1990. Argument Structure. x+202pp, Cambridge, MA: MIT Press, 1990

Garrido, Joaquin, 1992, Expectations in Spanish and German Adverbs of Change,
Linguistica, 26, 3-4, 357-402

Gillon, Brendan: 1987, ‘The Readings of Plural Noun Phrases in English’, Linguistics
and Philosophy 13, 477—485.

Guerzoni, Elena. 2004. Why Even Ask? On the Pragmatics of Questions and the
Semantics of Answers. Ph.D. Thesis. MIT.

Hale, Kenneth; Keyser, Samuel Jay, 1993. On Argument Structure and the Lexical
Expression of Syntactic Relations. In Hale, Kenneth, & Keyser, Samuel Jay [Eds],
The view from building 20: essays in linguistics in honor of Sylvain Bromberger,
53-109. Cambridge, MA: The MIT Press.

Hale, Ken; Keyser, Samuel Jay, 2002. Prolegomenon to a theory of argument structure.
Cambridge, MA: MIT Press

Hashimoto, Anne Y. 1971. Descriptive Adverbials and the Passive Construction. Unicorn
7, 84-93.

Haspelmath, Marthin. 1997. Indefinite pronouns. Oxford University Press.

Hazout, Ilan, 1995. Comparative Ellipsis and Logical Form. Natural Language &
Linguistic Theory, 13, 1, Feb, 1-37

Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Doctoral
dissertation, University of Massachusetts, Amherst, Massachusetts.

Heim, Irene. 1984. "A Note on Polarity Sensitivity and Downward Entailingness." NELS
14: 98-107.

Heim, I. 1985, Notes on comparatives and related matters, ms. University of Texas,
Austin.

Heim, I. 2000, Degree operators and scope, in Bredan Jackson and Tanya Matthews
(eds.) Proceeding of SALT X 40-64

Hellan, L. 1981. Towards an Integrated Analysis of ComparativesTiibingen: Narr.

234

Heycock, C. 1995. Asymmetries in Reconstruction. Linguistic Inquiry, 26, 4, 547-570

Higginbotham, James: 1981, ‘Reciprocal Interpretation’, Journal of Linguistic Research
1, 97—117.

Hoeksema, J. 1983, Negative Polarity and the Comparative, Natural Language &
Linguistic Theory. l(3):403-434

Horn, Laurence R. 1989. A natural history of negation. Chicago: University of Chicago
Press.

Homstein, N. 1995, Logical Form: From GB to Minimalism, Blackwell, Oxford

Huang, Shi-Zhe, 1996, Quantification and Predication in Mandarin Chinese: A Case
Study of dou, PhD. Thesis, University of Pennsylvania

Huang, C.-T. James. 1982. Logical relations in Chinese and the Theory of Grammar.
Ph.D. thesis. MIT.

Huang, C.-T. James. 1988. Wo Pao De Kuai and Chinese Phrase Structure. Language 64:
274-31 1

Huang, C.-T. James. 1996. On Lexical Structure and Syntactic Projection," Chinese
Languages and Linguistics 3, 000-000, 1996

Izvorski, R. 1995. A DP-shell for comparatives. In: The Proceedings of ConSole 3. eds.

A. Bisetti, L. Bruge, J. Costa, R. Goedemans, N. Munaro, and R. van de Vijver.
99- 12 1 .

Johnson, Kyle, 1991. Object Positions Natural Language & Linguistic Theory, 9, 4, Nov,
577-636

Kadmon, N. and Landman, F. 1993. Any. Linguistics and Philosophy 16:353-422.

Karttunen, L. and S. Peters: 1979, ‘Conventional Irnplicature’, in C. Oh and D. Dineen
(eds.), Syntax and Semantics ll: Presuppositions, Academic Press, New York,
pp. 1—56.

Katz, Graham, 2003. Event Arguments, Adverb Selection, and the Stative Adverb
Gap, in Lang, Ewald, Maienbom, Claudia, & F abricius-Hansen, Cathrine
[Eds]: Modifying Adjuncts. Berlin: Mouton de Gruyter, pp 45 5-474

Kayne, Richard. 1984. Connectedness and binary branching. Dordrecht: F oris.

Kayne, Richard S, 1994. The Antisymmetry of Syntax. xvii+l95pp, Cambridge, MA:
MIT.

235

Kennedy, C. 1999. Projecting the adjective: The syntax and semantics of gradability and
comparison. New York: Garland. (1997 UCSC Ph.D thesis).

Kennedy C. 2001. Polar opposition and the ontology of ‘degrees', Linguistics and
Philosophy 24.33-70.

Klein, E. 1982, The Interpretation of Adjectival Comparatives, Journal of Linguistics 18,
113-136.

Klein, E. (1991) “Comparatives,” in Amim von Stechow and Dieter Wunderlich (eds.)
Semantik/Semantics An International Handbook of Contemporary Research,
Walter de Gruyter, page 673-691.

Krifka, Manfred; Pelletier, Francis Jeffry; Carlson, Gregory N; Ter Meulen, Alice; Link,
Godehard; Chierchia, Gennaro, 1995. Genericity: An Introduction. In Carlson,
Gregory N., & Pelletier, Francis Jeffry [Eds], THE GENERIC BOOK, pp 1-124.
Chicago, IL: The U of Chicago Press

Ladusaw, W. A. 1979. Polarity Sensitivity as Inherent Scope Relations. Ph.D. thesis. UT
Austin.

Lahiri, U. (1998) Focus and Negative Polarity in Hindi. Natural Language Semantics
6:57-123.

Lasersohn, Peter: 1995, Plurality, Conjunction, and Events, Kluwer, Dordrecht.

Larson, R.K. 1987. 'Missing prepositions' and the analysis of English free relative
clauses. Linguistic Inquiry 16: 239-266.

Larson, R. K. 1988a, On the double object construction. Linguistic Inquiry 1/9: 335-391.
Larson, R. K. 1988b, Scope and Comparatives, Linguistics and Philosophy, 11, 1, 1-26
Larson, R. 1991. The Projection of DP and DegP. ms. Stony Brook U.

Larson, R. 2004. Sentence-final adverbs and ‘scope’. In M. Wolf and K. Moulton (eds.)
Proceedings of NELS 34. (pp.23-43) UMASS: GLSA.

Lechner, W. 1999, Comparatives and DP-Structure, Ph.D. thesis, University of
Massachusetts, Amherst

Lechner, Winfiied, 2001. Reduced and Phrasal Comparatives. Natural Language &
Linguistic Theory, 2001, 19, 4, Nov, 683-735

Lechner, W. 2002, Negative Islands in Comparatives. Snippets 5.

236

 

Lee, Y.-S. and Horn, L. 1994. Any as Indefinite + Even. Unpublished manuscript. Yale
University.

Lerner, J-Y and Pinkal, M. 1995, Comparative ellipsis and variable binding. The
proceedings of SALT 5

Li, Charles N. and Sandra A. Thompson. 1981. Mandarin Chinese: A Functional
Reference Grammar. University of California Press, Berkeley

Lin, Jo-Wang, 1996. Polarity Licensing and Wh-Phrase Quantification in Chinese. Ph.D.
thesis.Umass.

Lin, Jo-wang, 1998, Distributivity in Chinese and its implications, Natural Language
Semantics, 6, 201-243

Lin, Tzong-Hong, 2001. Light Verb Syntax and the Theory of Phrase Structure . Ph.D.
thesis. UC Irvin.

Link, G. (1983) “The Logical Analysis of Plural and Mass Terms: A Lattice-Theoretical
Approach,” in R. Bauerle, et al., (eds.), Meaning, Use and Interpretation of
Language, DeGruyter.

Link, Godehard, 1998. Algebraic Semantics in Language and Philosophy. Stanford, CA:
Center Study Language & Information.

Liu, Luther Chen-Sheng, 1996. A Note on Chinese Comparatives. Studies in the
Linguistic Sciences, 26, 1-2, spring-fall, 217-235

May, R. 1985, Logical Form, MIT Press, Cambridge, Massachusetts

Michaelis, Laura A. 1996, On the Use and Meaning of Already, Linguistics and
Philosophy, 19, 5, 477-502

Milsark, G. 1974. Existential sentences in English. Ph.D. thesis. MIT.
Moltmann, F. 1992, Coordination and comparatives. MIT Ph.D.

Moltmann, F. 1996. Domain-related dynamic semantics and the weak-strong distinction
among quantifiers. Ms. CUNY.

Munn, A. 1999, On the Identity Requirement of ATB Extraction, Natural Language
Semantics, 7, 4, 421-425

Munn, Alan, 2000. Explaining Parasitic Gap Restrictions. In Culicover, Peter W., &
Postal, Paul M. [Eds]. Parasitic Gaps. pp 369-392. Cambridge: MIT

237

 

Nunes, Jairo. 1996. The Copy Theory of Movement and Linearization of Chains in the
Minimalist Program. Ph.D. thesis. University of Maryland.

Nunes, Jairo. 1999. Linearization of Chains and Phonetic Realization of Chain Links. in
Epstein, Samuel David, & Homstein, Norbert [Eds], WORKING MIN IMALISM,
pp217-249. Cambridge: MIT.

Paris, Marie-Claude, 1998. Focus Operators and Types of Predication in Mandarin
Cahiers de Linguistique Asie Orientale, 27, 2, 139-159

Parsons, Terence, 1990. Events in the Semantics of English: A Study in Subatomic
Semantics. Cambridge, MA: MIT Press.

Portner, Paul, 2002, Topicality and (Non)Specificity in Mandarin, Journal of Semantics,
19, 275-287

Progovac, Liljiane. 1994. Positive and Negative Polarity: A binding Approach.
Cambridge University Press, Cambridge, MA.

Reinhart, T. (1997). "Quantifier scope: How labor is divided between QR and choice
firnctions." Linguistics and Philosophy 20: 335-397.

Rooth, M. (1985). Association with Focus. Ph.D. Dissertation, University of
Massachusetts at Amherst. Distributed by the GLSA, Department of Linguistics,
University of Massachusetts at Amherst.

Rooth, M. (19923). "A Theory of Focus Interpretation". Natural Language Semantics 1:
75-116.

Ross, Claudia. I991. Coverbs and Category Distinctions in Mandarin Chinese. Journal of
Chinese Linguistics, 19: 79-115

Rullmann, H. 1995. Maximality in the semantics of wh-constructions, Doctoral
Dissertation, University of Massachusetts, Amherst.

Rullmann, Hotze. 1996. Two types of negative polarity itemsIn Kiyomi Kusumoto (ed.)
Proceedings of the North East Linguistic Society, NELS 26, p. 335-350. GLSA
Publication. Department of Linguistics, University of Massachusetts, Amherst.

Schwarzschild, Roger: 1996, Pluralities, Kluwer, Dordrecht.

Schwarzschild, R, Wilkinson, K. 2002, Quantifiers in Comparatives: A Semantics of
Degree Based on Intervals, 10, 1, 1-41

Schwarzschild, R. 2002. The Grammar of Measurement. In Brendan Jackson, ed.,
Proceedings of Semantics and Linguistic Theory XII. Ithaca: CLC Publications.

238

Seuren, P., 1973. The comparative. In Generative grammar in europe, ed. F. Kiefer and
N. Ruwet. Dordrecht: Riedel.

Sharvit, Y. 1999, Functional Relative Clauses, Linguistics and Philosophy, 22, 5, 447-
478

Shyu, Shu-ing, 2004. (A)symmetries between Mandarin Chinese lian...dou and shenzhi.
Journal of Chinese Linguistics, 32, 1, Jan, 81-128

Stateva, P. 1999. In Defense of the Movement Theory of Superlatives. In: Proceedings of
Eastern States Conference on Linguistics (ESCOL), 15, 215-226

Szabolcsi, A. and Zwarts, F. 1993, Weak islands and an algebraic semantics for scope
taking, Natural Language Semantics 1: 235-285. (Reprinted in Szabolcsi, A.
[eds.] Ways of Scope Taking 1997, Kluwer)

Sybesma, Rint. 1999. The Mandarian VP, Kluwer, Dordrecht.

Van Der Auwera, Johan, 1993, ‘Already’ and ‘Still’: Beyond Duality, Linguistics and
Philosophy, 16, 6, 613-653

Van der Does, Jaap; Verkuyl, Henk, 1996, Quantification and Predication, in Semantic

Ambiguity and Underspecification, Van Deemter, Kees& Peters, Stanley [Eds],
Stanford, CA, 27-54

van der Wouden, Ton. 1994. Negative Contexts. Ph.D. thesis. University of Groningen.
Von Stechow, A. 1984. Comparing semantic theories of comparison, Journal of
Semantics 3. 1-77.

Wilkinson, Karina, 1996, The Scope of Even, Natural Language Semantics, 4, 3, 193-215

Wu, Jianxin, 1999, Syntax and Semantics of Quantification in Chinese, PhD thesis,
University of Maryland, Colloege Park

Wilkinson, K. 1998, Comparatives and Quantifier Scope, handout, University of
California, Irvine Colloquium.

Xiang, M., 2003, A Phrasal Analysis of Chinese Comparatives, in Proceedings of CLS
39, Chicago Linguistic Society.

Zepter, Alex, 2003. How to Be Universal When You Are Existential: Negative Polarity
Items in the Comparative: Entailment along a Scale. Journal of Semantics, 2003,
20, 2, May, 193-237

Zou, Ke. 1993. The Syntax of the Chinese BA Construction. Linguistics 31, 715—736.

239

Zwarts, Frans. 1995. Nonveridical Contexts. Linguistic Analysis, 25, 3-4 : 286-312

240

 

Ill‘llllllllll