Comparison of complex predicates: ‘and’, ‘or’ and ‘more’

Chapter (PDF Available) · April 2011with 54 Reads
In book: Understanding Vagueness: Logical, Philosophical and Linguistic Perspectives, Publisher: College Publications, Editors: Petr Cintula, Christian G. Fermüller, Lluís Godo, Petr Hájek
Cite this publication
Comparison of complex predicates and, or and more
Galit W. Sassoon, ILLC, University of Amsterdam
Are complex predicates in particular, negated (e.g., not expensive), conjunctive (e.g., expensive
and time consuming) and disjunctive predicates (e.g., tall or bald) associated with a graded
structure, namely a mapping of entities to degrees? On the one hand, most up to date semantic
theories of gradability and comparison in natural language disregard this question. On the other
hand, contemporary fuzzy logical theories provide compositional rules to construct a degree
function for a complex expression based on the degree functions of its constituents. These
composition rules have been found useful for a variety of practical applications. The question is
then whether these rules can correctly represent the interpretation of complex natural language
expressions and its relation to the interpretation of their constituents. The relevance of this
question is enhanced by recent findings from a variety of studies (Ripley, 2010; Serchuk et al.,
2010; Alxatib and Pelletier 2010), according to which high percentages of subjects count
contradictory predicates such as tall and not tall as true of borderline cases (neither short nor tall
entities). While these findings stand in sharp contrast to predictions of vagueness-based theories
of adjectives, they are in accord with the predictions of a fuzzy analysis, as extensively argued by
Kamp and Partee (1995). Given these new findings, then, the fact that fuzzy analyses allow for
non-zero truth values to contradictions can no longer count against them (for a more detailed
discussion see Sauerland, this volume). It is therefore increasingly important to test other
predictions of applications of fuzzy analyses to natural language conjunctions and disjunctions.
To this end, this paper discusses preliminary results based on a questionnaire eliciting judgments
from 35 Hebrew speakers. The results suggest that, counter the predictions of fuzzy analyses,
comparative and equative morphemes cannot apply to conjunctions and disjunctions of gradable
adjectives.
1. Introduction
Part 1 of this paper briefly surveys prominent semantic analyses of natural language comparison
constructions (section 1.1) and coordination constructions (section 1.2). Part 2 presents an
empirical study of the two constructions, focusing on comparisons of conjunctions and
comparisons of disjunctions. Part 3 concludes with the implications this study has concerning the
theoretical debates presented below.
1.1 The comparison morpheme
Linguistic theories of gradability and comparison can be divided into two main approaches,
‗ordinal‘ and ‗numerical‘ (cf. Sassoon 2010a). The ordinal approach attempts to reduce the
interpretation of comparative morphemes such as more and as to ordering relations between
individuals or between their ordinal degrees in predicates P, i.e., >P and P, respectively (cf.
Sapir 1944; Creswell 1977; Moltmann 2006; Bale 2008). In particular, in vagueness based
This work was made possible by the Orgler Scholarship (Tel Aviv University, 2004-2007) as well as the project `On
vagueness -- and how to be precise enough', funded by the Netherlands Organization for Scientific Research (NWO
360-20-201). This paper has benefitted a lot for comments by Uli Sauerland, as well as an additional anonymous
reviewer. Any remaining mistakes are all mine. I warmly thank my special friends and colleagues in Tel Aviv who
helped me carrying out the experimental study for this paper, including Ido Ben Tzvi, Uriel Cohen Priva, Aynat
Rubinstein, Tal Keidar, Meira Marom, Aya Meltzer, Tal David, Lior Laks, Julie Fadlon, Eytan Kidron, Sefi
Potashnik, Iddo Greental, Eugenia Berger, Hillel Taub-Tabib, Lena Fainleib, Lustigman Lyle, Bracha Nir-Sagiv,
Aviad Eilam, Shai Cohen, Dafna Heller, Aldo Sevi, Ziva Wijler, Nirit Kadmon and Adar Weidman.
gradability theories the ordering relations >P and P are derived based on facts pertaining to
membership in P‘s denotation, rather than based on fine grained numerical measurements. On
these theories, an entity is more P than other entities iff it falls under P relative to more
delineations (possible boundary specifications for vague predicates P; cf. Lewis 1970, 1979;
Kamp 1975; Fine 1975; Klein 1980; Landman 1991; van Rooij 2009).
In opposition, the numerical approach provides a unified analysis of comparative morphemes
with and without numerical modifiers. This approach characterizes gradable adjectives as
associated with numerical degree functions, i.e. mapping of entities xDx to a set of degrees
isomorphic to the real numbers r (Russell 1905; Bartsch and Venneman 1972; Klein 1991;
Kamp and Partee 1995; Kennedy 1999; Heim 2000; Schwarzschild and Wilkinson 2002;
Landman 2005; Sassoon 2010). Assuming a -categorial language in the style of Heim and
Kratzer (1998), with basic types x for individuals, t for truth values, and r for numerical degrees,
and basic semantic domains Dx, Dt, and Dr = (sets of individuals, truth values, and numerical
degrees, respectively). Gradable adjectives are interpreted as follows:
(1) Let Tc stand for a set of indices, the worlds (or completions) consistent with a
background context c (cf. Stalnaker, 1978; Kamp 1975).
(2) For any context c, for any tTc and any gradable adjective P:
a. fP,tDx is the degree function of P in t (a function from entities x in the
domain Dx to real numbers r in )
b. P holds true of an object xDx in t iff x‘s value exceeds P‘s cutoff point: fP,t(x)
cutoff(P,t) (Kennedy, 1999).
In particular, theories in this approach tend to assume that gradable predicates map arguments to
degrees for which plus, difference and ratio operations are applicable (von Stechow1984a,b). The
numerical approach is prevalent in the literature, as it provides straightforward semantic accounts
of expressions whose interpretation is mediated by the application of operations on numbers
(identity, multiplication, difference, etc.), such as, e.g., numerical modifies like 2 meters tall,
ratio predicates like twice as happy as Sam, and difference predicates like 2 meters shorter than
Sam. In these theories, the interpretation of more and as involves the application of a difference
operation, as demonstrated in (3), with rm,t being the degree of height of the meter in t (von
Stechow 1984a; Schwarzschild and Wilkinson 2002; Kennedy and McNally 2005; Schwarzschild
2005; Kennedy and Levin 2007:17; Sassoon 2010a).
(3) a. [[Dan is 2 meters taller than Sam]] t = 1 iff ftall,t([[Dan]] t) ftall,t([[Sam]] t) = 2rm,t
b. [[Dan is happier than Sam]] t = 1 iff r>0: fhappy,t([[Dan]] t) fhappy,t([[Sam]] t) = r.
Thus, despite differences in detail between analyses, all in all, a widely employed view is that,
e.g., Dan is taller than Sam (by 2 meters) holds true in an index t iff the difference between Dan
and Sam's degrees in t is a positive real number (twice the degree of a meter unit object in t).
The basic interpretation of phrasal er as a difference operation is, then, roughly,
r2.M<r,t>.r1.M(r1 r2), where the variable M has to be saturated by a degree predicate
like two inches (Schwarzschild and Wilkinson 2002; Landman 2005; Sassoon 2010b) and the
interpretation of as is r2.r1.[[ er]] (r2, r.r0, r1), which reduces to: r2.r1. r1 r2 0.
Given the latter, the use of the comparative, rather than the equative, normally excludes the
possibility that M is r.r0; i.e., in the absence of an overt numerical degree modification, M is
thought to be saturated by the predicate r. r > 0 (‗somewhat‘), implying that r1 r2 > 0 (cf.
Schwarzschild and Wilkinson 2002; Landman 2005; Sassoon 2010b), as illustrated below.
Finally, when er combines with an adjective as in taller, interpretation type shifts to an individual
level: fDx.x2Dx.M.x1D. [[er]] (f(x2))(M)(f(x1)), which reduces to
fDx.x2Dx.M.x1Dx.M(f(x1)- f(x2)).
(4)
On both the vagueness-based and numerical approach, a comparative or equative morpheme
applies to one predicate at a time, e.g. The table is taller than the chair is wide is a comparison of
the table‘s status (degree or denotation membership) with respect to a single predicate (tall), and
the chair‘s status with respect to a single predicate (wide). An open question, then, regards the
interpretation of combinations of comparison morphemes with complex predicates, as in, for
instance, more honest and intelligent and equally expensive and time consuming. How are such
statements interpreted? The next section discusses two different possible answers, a fuzzy and a
Boolean one. Section 2 presents an empirical investigation whose goal is to decide between the
two. Implications to the numerical versus vagueness-based debate are addressed.
1.2 The interpretation of coordination constructions
Sentences with conjunctions and modified-nouns in predicate position usually entail the
sentences resulting from dropping some of the constituents or changing the constituent ordering.
For example, (5a) entails (5b,c) and (6a) entails (6b,c). In addition, (5a) and (6a) are equivalent to
(5d) and (6d) respectively. Such entailment-patterns form the basis for the intersective analysis of
modified-nouns and conjunctions, whereby they denote the intersection of their constituents'
denotations, as stated and illustrated in (7) (Kamp and Partee 1995; Landman 2000; Heim and
Kratzer 1998). The intersection-rule in (7) directly predicts the fact that an item is classified as,
for instance, a four legged animal or an animal which is four legged iff it is classified as an
animal and it is classified as four legged.
(5) a. Tweety is brown and big
b. Tweety is brown
c. Tweety is big
d. Tweety is big and brown
(6) a. Tweety is a four legged animal
b. Tweety is four legged
c. Tweety is an animal
d. Tweety is an animal and is four legged
(7) tT: [[P (and) Q]] t = [[P]] t [[Q]] t
a. [[brown and big]] t = xDx.brown(x) big(x) = [[brown]] t [[big]] t
b. [[brown apple]] t = xDx.brown(x) apple(x) = [[brown]] t [[apple]] t
Dan
x1
r. r > 0
Tall
ftall
Sam
x2
er
f x2.M.x1. M(f(x1)- f(x2))
M.x1. M(ftall(x1)-ftall(x2))
Is
than
Dan is taller than Sam
ftall(x1)-ftall(x2) > 0
x1. ftall(x1)-ftall(x2) > 0
The same basic facts hold in the verbal domain, too. For example, the entailments from (8a)
to (8b-e) are instances of intersective inference patterns in modified verbs. These additional facts
form the basis for the Davidsonian intersective analysis of modified verbs ((9); Landman 2000).
(8) a. Dan ate quickly with a knife
b. Dan ate with a knife
c. Dan ate quickly
d. Dan ate
e. Dan ate with a knife quickly
(9) eE: [[eating]] t(e) Agent(e) = [[Dan]t xDx, Instrument(e) = x [[knife]t(x) …
While using the intersection rule, most up to date semantic theories of gradability and
comparison in natural language do not associate complex predicates (e.g., negated, conjunctive
and disjunctive ones) with graded structures (say, a mapping of entities to numerical degrees).
However, fuzzy logical theories can be used to do precisely that (cf. Hajek, 2009). Fuzzy logic is
a form of multi-valued logic, whereby propositions may have as a truth value any number in the
real interval [0,1]. The disjunction, conjunction and negation operators of Boolean logic exist in
fuzzy logic and are usually defined as the maximum, minimum, and complement, respectively
(Zadeh 1965); when they are defined this way, they are called the Zadeh operators. So for the
fuzzy propositions P(x) and Q(y):
(10) a. [[P(x)]] t = 1 [[P(x)]] t
b. [[P(x) Q(y)]] t = min([[P(x)]] t, [[Q(y)]] t)
c. [[P(x) Q(y)]] t = max([[P(x)]] t, [[Q(y)]] t)
Other definitions exist for conjunctive and disjunctive expressions that are not based merely on a
selection of one of the constituents‘ degrees; rather, these definitions make use of functions tand
and tor (often called t norms and t-conorms for conjunctions and disjunctions, respectively) to
compute a value for the complex expression based on the values of both of its constituents, as
follows:
(11) a. [[P(x)]] t = tnot,t([[P(x)]] t)
b. [[P(x) Q(y)]] t = tand,t([[P(x)]] t, [[Q(y)]] t)
c. [[P(x) Q(y)]] t = tor,t([[P(x)]] t, [[Q(y)]] t)
There are multiple choices for the fuzzy conjunction and disjunction operators. A common choice
is the algebraic product for fuzzy conjunction and algebraic sum for fuzzy disjunction, but there
are an infinite number of other choices (Yen, 1999; Hajek, 2009). Rather than the definitions of
(or axioms constraining) the t-functions, the very possibility that such functions may be relevant
to natural language coordination constructions is the focus of interest of this paper. Recall that
more is analyzed as denoting a difference modifier, e.g., Dan is taller than Sam is true in c iff
ftall,c([[Dan]]c) ftall,c([[Sam]] c) 0 (cf. section 1.1). According to this analysis, more cannot apply
to the interpretations of two predicates two degree functions simultaneously. However, if
natural language semantics is ‗fuzzy‘, i.e. conjunctive and disjunctive predicates are
systematically associated with composed degree functions, more should be capable of accessing
these functions and operating on them. Let fP-and-Q,t be the function xDx. tand(fP,t(x), fQ,t(x)) and
fP-or-Q,t be the function xDx. tor(fP,t(x), fQ,t(x)). A fuzzy natural language semantic theory
predicts the following interpretations for expressions of the form more P and Q and more P or Q.
(12) A fuzzy natural language semantic theory:
a. [[more P and Q]] t = x2Dx.x1Dx. fP-and-Q,t(x1) fP-and-Q,t(x2) 0
b. [[more P or Q]] t = x2Dx.x1Dx. fP-or-Q,t(x1) fP-or-Q,t(x2) 0
On the one hand, psychological findings pertaining to modified nouns seem to support a
fuzzy semantic analysis of complex natural language expressions. Psychological theories
associate concepts (typically, nominal ones) with functions corresponding to the mean of entities
in a variety of dimensions. Hampton (1987; 1988a; 1997a,b) has analyzed ratings of goodness of
example (typicality) of a list of entities in modified-nouns of the form 'Ps which are Qs' (such as,
for instance, pets which are birds) and in their constituents (e.g. pets and birds). The following
patterns emerged.
First, for any item x, it is possible to predict x's typicality rating in a modified-noun, fP-and-Q(x),
from x's ratings in the constituents, fP(x,) and fQ(x), by an equation like (13a). WP and WQ
represent the constituents' weights and WPQ represents the weight of the constituents' interaction
(the product fP(x) fQ(x)). For example, the values for pets which are birds were: Wpets = .30,
Wbirds = .78, and Wpetsbirds = .10.
Second, the typicality ratings in modified-nouns with negated constituents, fP-and-not-Q(x), are
predicted by adding a negative sign to the weight of the negated constituent (WQ). The interaction
term is also negative when significant (13b). For example, for pets which are not birds the
weights were: WP = .32, WQ = .75, and WPQ = .11. Why? Because the better an item is as an
example of Q, the worse it is as an example of not-Q.
Third, given the logical connections between disjunctions, conjunctions and negations (PQ =
(PQ)), and the fact that negation affects the equation by changing the coefficient sign,
Hampton predicted that the typicality ratings in disjunctions like hobbies or games, deg fP-or-Q(x),
would be given by adding a negative sign to the interaction term (WPQ). Why? The value fP-or-
Q(x) ought to be identical to f(PQ)(x), which, in turn, should be given by an equation in which
a negative sign is added to the weight of each negated-constituent (namely by the equation: (
WPfP(x) WQfQ(x) + WPQ(fP(x) fQ(x))). After the elimination of double negative-signs, this
equation reduces to the one in (13c), with the negative interaction-weight. And indeed, using
(13c), Hampton (1988b) could predict the typicality ratings in disjunctions from the ratings in the
disjuncts.
(13) a. fP-and-Q(x) = WPfP(x) + WQfQ(x) + WPQ(fP(x) fQ(x)).
b. fP-and-not-Q(x) = WPfP(x) WQfQ(x) WPQ(fP(x) fQ(x)).
c. fP-or-Q(x) = WPfP(x) + WQfQ(x) WPQ(fP(x) fQ(x)).
On the other hand, the constituent-based equations in (13) seem to be too coarse-grained.
Negated constituents, for instance, sometimes have a decreased weight, because some dimensions
are treated as characterizing both the predicate and its negation. For example, animate often
characterizes both birds and entities that are not birds, and bird-hood characterizes both robins
and non-robins. In general, the typicality ratings in modified-nouns are better fitted by a
composite-prototype representation, wherein the weight of each dimension is adjusted by a
special function.
More importantly, these findings may not extend to adjectival conjunctions and disjunctions.
The graded structure of adjectival predicates is rather different in nature from that of nominal
ones. The interpretation of adjectives like tall and expensive, for example, directly relate to
conventional measurements of height and cost, rather than to a weighted mean in a set of
dimensions. For one, these adjectival degree functions are unbound from above. In addition,
while adjectives can combine with more (or er) to create within-predicate comparisons (as in,
e.g., two meters taller), nouns and noun phrases do not combine with more, as illustrated by the
infelicity of, for example, *Tweety is more a bird/birder than Tan, *x is (a) more midget giant
than y and *x is (a) more fat bald man than y. When licensed, the comparative morpheme either
associates with the modifier most adjacent to it alone, as in x is a fatter bald man than y, or has to
be modified by of, as in x is more of a midget giant than y. The latter statement is interpreted as if
the noun phrase is modified with typical, as in x is more typical of a midget giant than y, where
the adjective typical forms the argument of more and the noun phrase only provides typicality
dimensions for typical to bind. In opposition, combinations of more with conjunctions or
disjunctions of adjectives are fine. For example, x is more expensive and elegant than y is
perfectly grammatical, and so is x yoter shamen ve kerea’x’ (‗x is more fat and bald than y‘) in
Hebrew and similar languages, whereby the comparative morpheme always surfaces as an
independent word (yoter; ‗more‘) and never as a dependent morpheme similar to English er.
Moreover, rather than to relate to a unique graded structure of the conjunctive concepts in
question, intuitively, these phrases seem to convey more expensive and more elegant and
fatter and balder, respectively. The same phenomenon pertains also to conjunctive multi-
dimensional adjectives, like typical with respect to flying and singing or healthy with respect to
blood pressure and pulse.
Thus, gradability is different in nouns and adjectives. While nominal functions may be
correctly described using some sort of fuzzy semantics, it is questionable whether adjectival
functions can be so described. Perhaps adjectival conjunctions and disjunctions, such as tall
and/or fat, are not systematically associated with degree functions at all. If so, more should not be
capable of combining with conjunctive or disjunctive predicates directly; rather, in constructions
of the form more P and Q or more P or Q, the Boolean operators and and or, respectively, should
take wide scope with respect to more, so that more would modify each conjunct/ disjunct
separately, operating on one basic degree function at a time (Sassoon 2007; Bale 2007). Such a
non-fuzzy, classically ‗Boolean‘ natural language semantic theory predicts the following
interpretations for expressions of the form more P and Q and more P or Q.
1
(14) A Boolean natural language semantic theory:
a. [[more P and Q]] t = [[more P and more Q]] t
= x2Dx.x1Dx. (fP,t(x1)- fP,t(x2) 0)
(fQ,t(x1)- fQ,t(x2) 0)
b. [[more P or Q]] t = [[more P or more Q]] t
= x2Dx.x1Dx. (fP,t(x1)- fP,t(x2) 0)
(fQ,t(x1)- fQ,t(x2) 0)
While linguists belonging to the numerical degree approach (cf. section 1.1 above) such as
Kennedy (1997) sometimes argue that this approach resembles fuzzy logic, they never, to the best
of my knowledge, actually study the question of whether conjunctions and disjunctions of
morphologically gradable adjectives are also morphologically gradable, i.e. felicitously licensing
more. A basic way to test whether the interpretation of expressions of the form more P and/or Q
is fuzzy (cf. (12)) or non-fuzzy (cf. (14)), is by presenting subjects with pairs of entities differing
along one conjunct/ disjunct (say P) but otherwise identical (equally Q), and asking them
1
The requirement for a unique dimension in the use of a comparative morpheme can only be abandoned in between-
predicate comparisons. Conjunctive and disjunctive concepts seem to be felicitous and to receive interpretations with
the connective in narrow scope in such comparisons (cf. (i)-(iii)), though more systematic future research needs to
carve out the precise set of interpretations that may be assigned to such statements.
i. This is more a kitchen utensil than an electronic device.
ii. This is more a piece of furniture and a game than a kitchen utensil or an electronic device.
iii. Dan is more fat, bald and unhappy than good-looking, energetic and funny.
whether these pairs stand in the relations (i) more P and Q, (ii) less P and Q and/or (iii) equally P
and Q and (iv) more P or Q, (v) less P or Q and/or (vi) equally P or Q. A non-fuzzy analysis
predicts that such pairs stand in none of the three conjunctive relations (i)-(iii) (because they
stand in none of the relations more P and more Q, less P and less Q and equally P and equally Q)
and in the two disjunctive relation (iv) and (vi) (because they stand in the relations more P or
more Q and equally P or equally Q). In sharp contrast, a fuzzy analysis does not allow for these
possibilities; pairs of entities ought to stand in one and only one conjunctive relation and
disjunctive relation, depending on their composed tP-and-Q and tP-or-Q degrees, respectively. Entity
pairs standing in no relation or in more than one relation are at least not straightforwardly
accounted for by a fuzzy analysis.
A small questionnaire was designed to examine what the facts actually are, i.e. whether they
are more easily fitted by a fuzzy analysis such as the one provided in (12) or by a Boolean
analysis such as the one in (14).
2. And, or and more, a general judgments questionnaire
2.1 Method
Subjects
The subjects were 35 native speakers of Hebrew, 21 females and 14 males, in the age range 20-40
with three exceptions of ages 41, 44 and 59 (average age 31) and with academic education of at
least one year (17 graduate students).
Design and Material
The subjects received a written questionnaire. An opening paragraph included general
instructions. This opening paragraph, translated from Hebrew to English for the purpose of
presentation in this paper, is as follows:
The goal of this questionnaire is to understand the way people think and the way they use certain words. Hence,
there are no right and wrong answers. For each question provide the answer which on your opinion is the most
reasonable and accurate. A slot for comments follows each section. We will be happy with any comment pertaining
to the reasons for which you choose to answer the way you do or to uncertainty you might have. Filling in the
comment slot is not obligatory. While some sections look alike, it is very important that you relate to each one of
them separately. While the questions in this questionnaire are given in masculine forms, they are addressed to both
genders. Many thanks in advance for your patience in filling in the questionnaire.
On every section, draw a circle around the answer you select.
The questionnaire included 13 sections, 5 of which are relevant for the present research. Each
section included a short paragraph with a brief description of two characters consisting of their
values or relative status in two gradable properties. The paragraph was followed by either yes-no
questions (followed by Yes/No), or two-valued questions, followed by two names (for instance,
Dan/Sam). The yes/no questions asked whether the two characters stand in a certain comparison
relation with respect to a given predicate; the two valued questions asked which entity ranks
higher in asymmetric comparison relations. The questions can be divided to various different
conditions depending on the type of comparison relation and the type of predicate.
Predicate types included basic gradable adjectives such as tall, conjunctions and disjunctions
of gradable adjectives such as bald and/or tall, and nominal constructions modified by gradable
adjectives, such as bald tall [one]. Notice that adjectives stand alone (with no overt noun phrase to
modify) significantly more easily in Hebrew than in English, as is apparent from the translations of
some of the questions below. Therefore, some material that did not occur in the original
questionnaire has been added in square brackets to enhance clarity of the English translation (e.g.
[one] in the nominal example given above).
Comparison types included simple comparisons (as in, e.g., taller, less tall, and equally tall)
and complex comparisons, mainly of ease of classification (as in easier/ less easy / equally easy
to determine that Moshe is tall than that Danny is tall) and difficulty of classification (harder,
less hard / equally hard to determine…), but also, on few sections, comparisons of typicality (as
in more typical of a tall person), fitness (as in fits more to be a subject in a scientific experiment
studying properties of tall people) and certainty (as in if Danny fits, Moshe definitely fits). Asking
questions with different forms of comparison can reveal whether different ways to relate to the
relative ordering of entities along dimensions like height produce similar or different answers. At
the same time they may serve to test reliability. Notice that comparative adjectives like taller in
Hebrew are construed of two separate words: yoter gavoha (‗more tall‘; ‗taller‘), thus the data
below is presented in the Hebrew way, e.g. using more fat rather than fatter, even where English
speakers would prefer the latter.
For example, on section 1 the subjects read the following description of two characters called
Moshe and Danny: ―Assume Moshe weighs 100kg and Danny weighs 90kg and they are alike in
other things (for instance, height). The questions following this paragraph include simple
comparison, as well as comparison of ease and difficulty of classification, but only in relation to a
basic adjective (fat), for this section was introductory in nature, with the goal to check whether
subjects understand the general logic of the questions in the questionnaire:
(15) Section 1, the questions:
The basic condition, simple comparison
a. Is Moshe more fat than Danny? Yes/No
The basic condition, complex comparison
b. Is it easier to determine that Moshe is fat than that Danny is fat? Yes/No
c. Is it harder to determine that Danny is fat than that Moshe is fat? Yes/No
On section 3, the subjects read the following description: ―Assume Moshe weighs 100kg and he
is 1.95 cm tall, and Danny weighs 70 kg and is he is 1.95 cm tall.‖ (i.e., Moshe is fatter than
Danny, but they are equally tall). The questions included both basic (fat, tall) and conjunctive
adjectives (fat and tall), as well as simple and complex comparisons. If natural language semantics
is generally fuzzy, subjects should regard fat and tall as assigning Moshe and Danny degrees in a
bound interval isomorphic to the real interval [0,1]. Then, based on these degrees, subjects should
try to compute Moshe and Danny‘s degrees in fat and tall, thereby judging either Moshe or
Danny as more fat and tall.
In opposition, if natural language semantics is not generally fuzzy (the ‗Boolean‘ hypothesis),
subjects would not have access to degrees in complex predicates like fat and tall; rather they will
interpret more fat and tall with and scoping over more; i.e. they will try to determine whether it is
Moshe or Danny that is more fat and more tall (i.e. fatter and taller). Since neither one is fatter
and taller, nor are they equally fat and tall in the sense of being equally fat and equally tall,
subjects are expected to say that (i) Moshe is not more fat and bald”; (ii) Danny is not more
fat and bald” and (iii) They are not equally “fat and bald”.
Three measures where taken in order to bias subjects against such ‗Boolean‘ answers with wide
scope for and/or with respect to more, and towards ‗fuzzy answers, whereby degrees and ordering
relations are construed for complex predicates. First, the following introductory comment preceded
the questions of section 3:
(16) Section 3, introductory comment:
An important comment regarding sections in the questionnaire of the form:
(1) Is Moshe more tall and fat than Danny?
(2) Is Moshe more tall or fat than Danny?
The intention is not to ask whether Moshe is more tall and/or whether Moshe is more fat;
rather, the intention in (1) is to ask whether Moshe exemplifies better the complex property
fat and tall. The intention in (2) is to ask whether Moshe exemplifies better the complex
property fat or tall.‖
Second, in each and every section, the adjectival conjunctions and disjunctions were
underlined in all the questions under concern, so that they will be processed as relating to a single
unified property.
Third, section 9 presented the two figures as equally fat and one balder than the other (i.e. the
same pattern as in section 3); however, this section begins with 9a directly asking whether Moshe
is both more fat and more bald than Danny is (an unambiguously wide-scope and/or question)
and immediately continued by asking whether Moshe is more fat and Bald than Danny is (9b).
On this setup, subjects are expected to try to interpret 9b as asking for something different than
9a, thereby interpreting and within the scope of more. Likewise, sections 3 and 4 begin by asking
who is more fat and who is more tall/bald, except in two separate subsections (3a,c and 4a,b)
rather than in a conjoined question.
Finally, section 4 presented characters in inverse relations (Aharon fatter and Danny balder)
and section 7 involved a conjunctive typicality adjective referencing two typicality features; this
adjective translates roughly to typical of a flying and calling creature. A complete presentation of
the relevant sections (1,3,4, 7 and 9) is found in the appendix (see also the tables in the result
section below).
Procedure:
The subjects have received the questionnaire by email. They were asked to fill it in themselves and not to consult
with anyone but me if they have questions. When they were undecided, they were encouraged to nonetheless select
the answer that fits best their opinion. They have received as much time as they needed to fill in answers and were
encouraged to add comments on each section.
2.2 Results
The results of sections 1, 3 and 4 are presented first, divided to results of basic conditions (atomic
predicates), followed by conjunctive, disjunctive and modifier-position conditions.
2
Only then are
the results of sections 9 and 7 presented, which appear to reflect mainly the effect of repetition of
judgments on all conditions.
2.2.1 Judgments for atomic (‘basic’) predicates
Simple comparison
The answers pertaining to basic predicates confirmed expectations on sections 1 and 3 Moshe was
generally judged fatter, with 89% and 94% agreement when his 100kgs were compared to Danny‘s
90k and 70kg, respectively. Apparently, the bigger weight difference in 3a vs. 1a explains the
higher percentage of agreement in 3a.
The judgments of the few subjects that did not agree to say that Moshe is fatter (even in the 30kg
difference condition) are probably explained by the fact that (as often mentioned in the comment
sections throughout the questionnaire) ordering judgments in adjectives like fat (as well as tall) may
be based on both weight and height as well as on general look (since mere weight may reflect
muscles rather than fat).
2
The results for complex comparisons are basically the same as the results for simple ones, except somewhat weaker.
However, as discussed below, they appear to have been affected by a methodological problem. Thus, these results
are presented separately from those for simple comparisons; readers that are only interested in the main question the
paper asks can skip the paragraphs pertaining to complex comparisons and still capture the main findings.

On 3c subjects generally did not agree that any one of the 1.95cm tall characters is taller than the
other (6% agreement). On section 4, subjects‘ answers to 4a,b unequivocally indicate that they
agree that Aharon is fatter (100%) and Danny is balder (97%), as expected given that Aharon is
30kg fatter than Danny, but is not bald, respectively.
‗Yes‘ answers
‗No‘ answers
%‗Yes
Section 1
Moshe -100kg; Danny -90 kg
31
4
89%
a. Is Moshe more fat than Danny?
24
11
69%
b. Is it easier to determine that Moshe is fat
than that Danny is fat?
21
14
60%
c. Is it harder to determine that Danny is fat
than that Moshe is fat?
Table 1: The basic conditions
(i) Given his higher weight, Moshe is generally judged fatter (sections 1,3; same with Aharon in 4); (ii) given Moshe
and Danny‘s equal heights, none is judged taller (sections 1 and 3) and (iii) given their inverse classification as bald and
not bald, Danny is judged to be balder (section 4).
Complex comparison
The same pattern is found with comparison of ease of classification (e.g., easier to determine that x
is fat than that y is), but with smaller percentages of agreement (69% vs. 89% in 1b and 3b,
respectively). Using comparison of difficulty of classification in 1c (e.g., harder to determine that
y is fat than that x is), yields even smaller percentage (60%). One comment regarding 1b indicates
that it is easy to determine that both are fat. Apparently, this yields the use of easier to determine
less appropriate than the use of fatter; it yields the use of harder to determine even less appropriate.
Also, given that subjects‘ negative answer to (3c) implies that the characters are equally tall,
most (25) subjects ignored question 3d concerning whether it is easier to determine that one of them
is tall; however, four of the 35 subjects (12%) did agree that tallness is easier to determine for one
than for the other, justifying their answers by assigning a role to their very different weights.
2.2.2 Judgments for conjunctive predicates
Simple comparison
The answers to the conjunctive questions in sections 3 and 4 are generally supportive of the
Boolean theory. Recall that on section 3, the characters were equally tall but Moshe was 30kg
fatter. Despite this difference on one of the conjuncts, Moshe was generally judged neither more fat
and bald nor less fat and bald than Danny, with 6% and 0% agreement in 3e and 3f, respectively.
Obviously, the two characters are not equally fat and bald (cf. the results for question 4e below).
Thus, this result suggests that the questions tend to be interpreted with and scoping over more, i.e.
‗Yes‘ answers
‗No‘ answers
%‗Yes’
Section 3, a-d
Moshe -100kg; Danny -70 kg;
both -1.95 cm tall
33
2
94%
a. is Moshe more fat than Danny?
31
4
89%
b. Is it easier to determine that Moshe is fat
than that Danny is fat?
2
30
6%
c. is one of them more tall than the other?
4
6
40%
d. Is it easier to determine that he is tall?
‗Danny‘
‗Aharon‘
% ‗Danny
Section 4a,b
Aharon -100kg; Danny -70 kg;
Aharon not bald; Danny bald
0
33
0%
a. Who is more fat
32
1
97%
b. Who is more bald

subjects try to determine whether Moshe is more fat and more tall in 3e and whether he is less fat
and less tall in 3f. The latter is clearly not the case (testified by across-the board disagreement);
the former is not the case because the characters are equally tall. Counter the prediction of a fuzzy
semantic theory, then, subjects did not judge Moshe more fat and tall, presumably because they
did not compose a degree function for the conjunctive concept fat and tall.
Concerning section 4, recall that subjects‘ answers to 4a,b unequivocally indicate that they agree
that Aharon is fatter (100%) and Danny is balder (97%). Probably precisely because of that,
counter the prediction of a fuzzy theory, subjects did not appear to compute degrees for the
conjunctive concept fat and bald; rather, in line with the Boolean hypothesis, the answers to 4a-f
unequivocally indicate that subjects agree to say neither that any of the two characters is more fat
and bald (0% agreement to 4c), nor that they are equally fat and bald (9% agreement to 4e). This
suggests that these questions are interpreted with and scoping over more; subjects were trying to
determine whether Aharon is more fat and more bald or whether he is less fat and less bald in 4a
(both are clearly not the case) and whether Aharon and Danny are equally fat and equally bald in
4c (which again is clearly not the case).
In sum, counter the prediction of a fuzzy theory, subjects did not judge any character to be
more fat and bald, and at the same time judged them not to be equally fat and bald. This seems to
indicate that they did not compose a degree function for the conjunctive concept fat and bald.
3
Table 2: The first conjunction condition:
Despite a 30kg weight difference (all other things being equal), Moshe is generally not judged more fat and tall.
Table 3: The second conjunction condition:
When Aharon is fatter and Danny balder, generally, neither is judged to be more fat and bald.
Complex comparison
The pattern of results with comparisons of ease of classification (e.g., easier to determine that x is
fat and tall) is similar to the pattern found with direct comparison (e.g. more fat and bald), except
for larger percentage (almost 40%) of agreement to say, on section 3g, that it is easier to
3
To the best of my understanding, product based t-norms and sum based t-conorms cannot account for these data,
except perhaps by virtue of a residua given an assumption of interaction between, e.g., fat and bald. But such an
assumption is not justified, for fat and bald are (intuitively) independent.
‗Yes‘
‗No‘
%‗Yes’
Section 3, e-f
Moshe -100kg; Danny -70 kg;
both -1.95 cm tall
2
33
6%
e. Is Moshe more fat and tall than Danny?
0
32
0%
f. Is Moshe less fat and tall than Danny?
13
21
38%
g. Is it easier to determine that Moshe is fat
and tall than that Danny is?
4
30
12%
h. Is it harder to determine that Moshe is fat
and tall than that Danny is?
‗Yes‘
‗No‘
%‗Yes’
Section 4c-f
Aharon -100kg; Danny -70 kg;
Aharon not bald; Danny bald
0
34
0%
c. Is any of them more fat and bald than the
other?
1
34
3%
d. Is it easier to determine that one of them is
fat and bald than that the other is?
3
31
9%
e. Are they equally fat and bald?
2
31
6%
f. Is it equally easy to determine that they are
fat and bald?

determine that Moshe is more fat and tall. At any rate, most subjects still disagree to rank Moshe
higher.
Two points reveal that the results concerning ease of classification should be taken with a
grain of salt.
First, subjects commented that Danny is obviously not fat and tall (because he is by no means
fat), rendering this comparison inappropriate; again, we see evidence supporting the hypothesis
that comparisons of the form easier to determine that x is P than that y is P are only appropriate
if it is possible to determine that both x and y are P, but it isn‘t too easy to do so or to reject doing
so (P-hood is somewhat uncertain or dubious, but is definitely a live option). But if classification
as, P, not P or undetermined is the main issue at stake in ease-of-classification comparisons, their
acceptance is compatible with a Boolean theory; hence, these comparisons are not ideal means to
distinguish between fuzzy and boolean interpretations.
Second, probably subjects would have been more willing to say that it is easier to determine
whether Moshe is fat and tall than whether Danny is. This issue is left for future research to
resolve. Notice, however, that reinterpretation of that as whether may explain the unexpected
12% (rather than 0%) agreement to 3h (it is harder to determine that [ whether] Moshe is fat and
tall), since it is very easy to determine that Danny falls outside this conjunction (because he is by
no means fat), while Moshe‘s classification is uncertain. Indeed, some subjects commented that
they could only have answered the questions positively if easier to determine that would have
been substituted for easier to determine whether and others said that they have answered as if the
question included a whether-, rather than a that-complementizer.
At any rate, the results pertaining to section 4 are pretty clear. Subjects agree to say neither
that for any of the two characters it is easier to determine that he is fat and bald (3% agreement to
4c), nor that for both this is equally easy to determine (6% agreement to 4f). This suggests that
these questions are interpreted with and scoping over more; subjects were trying to determine
whether, e.g. for Aharon, it is easier to determine that he is fat and easier to determine that he is
bald or whether it is less easy to determine that he is fat and less easy to determine that he is bald
in 4b (both are clearly not the case) and whether for Aharon and Danny it is equally easy to
determine that they are fat and equally easy to determine that they are bald in 4d (which again is
clearly not the case).
2.2.3 Judgments for disjunctive predicates
Simple comparison
The answers to the disjunctive questions confirmed the Boolean theory‘s predictions, although to a
lesser extent.
Notice that if subjects were to construct a degree function for fat or bald, their answers to 4e-
g should have been that none is more fat or bald, for both characters have a high degree in one
disjunct and a low degree in the other disjunct, rendering their degrees in fat or bald more or less
equal. However, first, we have already seen that they are not judged equally fat and bald (cf.
Table 3, 4e,f). Second, the answers to 4g indicate that ¾ of the subjects agree that one of the two
characters is more fat or bald (74% agreement in 4g). Third, importantly, less than half of the
subjects (46%) were willing to provide a single name indicating who is more fat or bald, Danny
or Aharon (only 16/35 answers for 4h).
Thus, first and foremost, the 19 subjects (54%) not providing an answer to this question
indicate that they interpreted more fat or bald as more fat or more bald, classifying both
characters as such. In other words, these subjects interpreted or as scoping over more, trying to
determine whether Aharon is more fat or more bald and whether Danny is more fat or more bald;
both are clearly the case as Aharon is fatter and Danny balder, rendering 4g true and at the same
time making it impossible to choose one answer to 4h.

Second, of the 16 answers, the majority (10, which make 63%) selected Danny as their
candidate, explaining that Danny is clearly bald and hence fat or bald, while Aharon is not clearly
fat, thus not clearly fat or bald. Thus, these subjects were using a Boolean union rule for
classification under disjunctions. The remaining 6 subjects did not indicate why they selected
Aharon.
Third, one could argue that the 16 subjects answering 4h were using fuzzy disjunctive
degrees; however, had this been the case, these 16 subjects would have agreed to consider one of
the characters as more fat and bald in the second conjunctive condition, but they did not (cf. 0%
and 3% agreement to 4c,d, Table 3). Thus, these results do not indicate fuzzy reasoning.
4
Table 4: The first disjunction condition:
When Aharon is fatter and Danny balder, generally, both are judged to be more fat or tall.
Complex comparison
Similar patterns are found in the ease of classification questions. Most subjects agree to say that
for one of the two characters it is easier to determine that he is fat or bald (68% agreement in 4i),
and 77% of the 22 subjects that provided an answer to 4j have selected Danny. Danny is clearly
bald and hence fat or bald, while Aharon is not clearly fat, thus not clearly fat or bald. For this
reason, it is easier to determine that Danny is fat or bald, assuming a union rule for classification
under disjunctions.
Moreover, 13 subjects did not give an answer to this question, indicating that they probably
classified both as ones for whom is easier to determine fat or bald, as predicted by the Boolean
theory. They have probably interpreted these questions with or scoping over more, trying to
determine whether for Aharon it is easier to determine that he is fat or easier to determine that he
is bald or whether for Danny it is easier to determine that he is fat or easier to determine that he
is bald; both are clearly the case (as Aharon is fatter and Danny balder), making it impossible to
choose one answer.
Also, the results of 4s-w with fitness comparisons and certainty comparisons support
reliability. On 4s, 92% of the subjects agree that Danny fits more to an experiment for which fat
or bald subjects are needed. Likewise, on 4t, 82% refused to rank Aharon higher (to say that if
Danny fits, Aharon definitely fits), while on 4w, 61% agreed to rank Danny higher (if Aharon
fits, Danny definitely does).
4
The use of product and sum functions for and and or cannot account for these data, as it predicts, for any pair of
entities standing in the relation ―equally P and Q‖, that in order to also stand in the relation ―more P or Q‖, their P
values should both be greater than both their Q values (or vice versa.) This condition is not satisfied in the present
scenario, since Danny‘s degree in bald exceeds Aharon‘s degree in fat (Danny is definitely bald, but 100kg is not
definitely fat), but Aharon‘s degree in bald doesn‘t (Aharon is ‗maybe not bald ‗). Similarly, Aharon‘s degree in fat
exceeds his degree in bald, but not Danny‘s.
‗Danny‘ /
‗Yes‘
Aharon/
‗No‘
%‗Danny/
%‗Yes’
Section 4g-j
Aharon -100kg; Danny -70 kg;
Aharon not bald; Danny bald
25
9
74%
g. Is one of them more fat or bald than the
other?
10
6
63%
h. Who is more fat or bald?
23
11
68%
i. Is it easier to determine that one of them is
fat or bald than that the other is?
17
5
77%
j. For whom is it easier to determine that?

Table 5: The second disjunction condition:
When Aharon is fatter (100kg vs. 70kg) and Danny is balder, generally, Danny ‘fits more‘ and is ‗more definitely
classified in‘ the category fat or tall.
2.2.4 Judgments for predicates in modifier position
On this condition the adjectives occur with no overt coordination marker; rather, they are in
a modifier position of a null noun (perhaps the rightmost adjective functions as a noun in this
construction). On 4o-4r, it is bald that is more adjacent to the empty noun head (or perhaps itself
functions as a noun), whereas in 4k-4n fat is in the more ‗nominal‘ position. At any rate, subjects
could not take more in the modified noun condition to refer only to the adjective not adjacent to
(or not itself functioning as) the head noun (trying to determine whether any of the characters is a
fatter bald man in 4k-n or a balder fat man in 4o-r), because for this interpretation to occur in
Hebrew the comparative morpheme and the adjective it modifies should follow the noun they
modify rather than precede it.
The results on this condition are similar to the results on the conjunction condition, though a
bit weaker. The majority of the subjects refused to agree that any of the characters is more fat
bald [person] (13% agreement in 4k) and bald fat [person] (23% agreement in 4o), with less than
a third of the subjects 9 and 11 answering the questions which one is in 4l and 4p, of which
89% and 91% respectively selected Danny. This suggests that most subjects interpreted these
questions with a Boolean conjunctive operator scoping over more, as explained above for more
bald and fat.
The results for the ease of classification comparisons are similar but slightly weaker, as in
previous sections, suggesting that the subjects answered reliably. Most subjects refused to say for
any of the two characters that it is easier to determine that he is fat bald (27% agreement in 4m) or
that he is bald fat (25% agreement in 4q), with about a third of the subjects 11 and 14
answering the questions ‗which one is‘ in 4n and 4r, of which 82% and 100% respectively
selected Danny to be better in the modified construction.
The selection of Danny rather than Aharon by subjects in 4l,n,p,r may be merely due to
Danny‘s description as bald and Aharon‘s description as maybe not bald, as opposed to the
absence of direct descriptions of their status in fat. Still, it seems puzzling that 70kg counts as fat
here. Thus, the order of conditions may have affected the results. The present condition directly
followed the disjunction condition, whereby Danny ranked higher for obvious reasons (cf. 2.2.3).
‗Danny‘ /
‗Yes‘
Aharon/
‗No‘
%‗Danny/
%‗Yes’
Section 4s-w
Aharon -100kg; Danny -70 kg;
Aharon not bald; Danny bald
22
2
92%
s.Fat or bald subjects are needed to fill in a
questionnaire for a scientific experiment. Who
fits more?
6
27
18%
t. It is true that if Danny fits Aharon definitely
fits?
20
13
61%
w. It is true that if Aharon fits Danny
definitely fits?

Table 6: The modifier-position condition:
When Aharon is fatter (100kg vs. 70kg) and Danny is balder, generally, neither is more [of a] fat bald [person] and
neither is more [of a] bald fat [person]; but if one is selected, Danny is.
We see that, as in previous sections, the results support a Boolean theory rather than a fuzzy
theory for complex concepts, over and above the fact that complex concepts where underlined
and an explicit comment asked subjects to consider the whole concept rather than each
constituent separately (cf. section 2.1). We can, therefore, conclude that the results for sections
1,3 and 4 support the existence of a Boolean bias, whereby Boolean operators (conjunction or
disjunction) are interpreted in wide scope with respect to comparative morphemes (or phrases).
2.2.5 Judgments for section 9 - the ‘repetition’ conditions
To overcome this Boolean bias, section 9 begins by directly asking whether Moshe is both
more fat and more bald (9a) versus whether Moshe is more fat and Bald than Danny (9b). The
idea is that on this setup, subjects will try to interpret 9b as asking for something different than
9a, thereby interpreting and within the scope of more. Recall that on section 9, the characters were
equally fat (100kg) but Moshe was balder. This setup is, then, very similar to the one in section 3.
The results, however, are different.
Subjects generally disagreed to answer 9a positively (9% agreement); however, while most
subjects still disagreed to answer 9b positively, much fewer did (43% agreement). These results
indicate that an alternative fuzzy interpretation may be accessed, although it is not the preferred
or natural interpretation for contexts such as those given in the questionnaire.
Even more striking is the 91% and 79% agreement to it being easier to determine that Moshe
is fat and bald (9c) and bald and fat (9d) than that Danny is, respectively. As in previous sections,
replacing direct comparison with comparisons of ease of classification facilitates agreement to
rank Moshe higher, but this time facilitation is extreme.
5
Nonetheless, two problems in the experimental design make it impossible to conclude that
under the conditions specified above (when sentences such as 9b are adjacent to sentences such
as 9a) speakers generally tend to interpret and within the scope of more, i.e. to compute a graded
structure degrees and/or ordering for conjunctions, in line with a fuzzy semantic theory.
The first problem is that the order of presentation of sections and questions was one and the
same for all subjects. Thus, the high percentages in 9b-d may also result from an effect of
repetition of the same sort of questions, together with a desire on the part of the subjects to be
cooperative, i.e., to answer positively, a desire that was repressed by virtue of the Boolean bias.
5
Recall that on section 3 Moshe was generally judged neither more fat and tall nor less fat and tall than Danny, with
6% and 0% agreement respectively, and that less than 40% agreed to say that it is easier to determine that Moshe is
more fat and tall.
‗Danny‘ /
‗Yes‘
Aharon/
‗No‘
%‗Danny/
%‗Yes’
Section 4k-r
Aharon -100kg; Danny -70 kg;
Aharon not bald; Danny bald
4
27
13%
k. Is any of them more fat bald than the other?
8
1
89%
l. Who is more fat bald?
9
24
27%
m. Is it easier to determine that one of them is
fat bald than that the other is?
9
2
82%
n. For whom is it easier to determine that?
7
23
23%
o. Is any of them more bald fat than the other?
10
1
91%
p. Who is more bald fat?
8
24
25%
q. Is it easier to determine that one of them is
bald fat than that the other is?
14
0
100%
r. For whom is it easier to determine that?

But having answered more than 20 conjunctive/disjunctive questions before getting to section 9
(4 questions on section 3 plus 19 questions on section 4), subjects tendency to disagree that
and/or can scope under more may have slowly diminished. Perhaps, then, repetition in itself
facilitates the composition of a unified degree function or ordering for the complex category.
The second problem regards an additional difference between section 3 and section 9. If both
characters of section 9 are considered possibly fat (100kgs) and either definitely or possibly bald,
this may have facilitated positive answers. Recall that on section 3 subjects commented that
Danny is obviously not fat and tall, and that, for this reason, comparison of ease/difficulty of
classification under these concepts is inappropriate; it may well be the case, then, that on question
9 more subjects were willing to accept the use of easier to determine that because no character
was obviously not fat and obviously not bald. Furthermore, it may well be that more subjects
answered as if the relevant questions included whether instead of that on section 9 than on section
3, due to the fact that 9 occurred after 3. But if it is only the status of entities as bald or not bald
(or undetermined) that matters, the interpretation, in both cases, could be Boolean (supplemented
by a representation of epistemic ignorance and/or vagueness, cf. part 3), with wide-scope for and
with respect to the comparison phrase, as discussed earlier.
The same considerations apply to the modified noun conditions in 9e-h and even more so to
the disjunction conditions, as seen in 9i-o.
Table 7a: The ‗repetition‘ condition - conjunction
Table 7b: The ‗repetition‘ condition modifier-position
Table 7c: The ‗repetition‘ condition - disjunction
‗Danny‘ /
‗Yes‘
‗Moshe‘ /
‗No‘
%‗Danny/
%‗Yes’
Section 9a-d
Moshe is bald;
Danny is less bald (maybe not bald);
Both-100kg
3
32
9%
a. Is Moshe both more fat and more bald?
15
20
43%
b. Is Moshe more fat and Bald than Danny?
32
3
91%
c. Is it easier to determine that Moshe is fat
and bald than that Dan is fat and bald?
27
7
79%
d. Is it easier to determine that Moshe is bald
and fat than that Dan is bald and fat?
18
15
55%
e. Is Moshe more fat bald than Danny is?
25
10
71%
f. Is it easier to determine that Moshe is fat
bald than that Danny is fat bald?
19
16
54%
g. Is Moshe more bald fat than Danny is?
23
10
70%
h. Is it easier to determine that Moshe is bald
fat than that Danny is bald fat?
35
0
100%
i. Is Moshe more fat or Bald than Danny?
31
3
91%
j. Is it easier to determine that Moshe is fat or
bald than that Danny is?
35
0
100%
k. Is Moshe more Bald or fat than Danny?
32
1
97%
l. Is it easier to determine that Moshe is bald
or fat than that Dan is?
2
28
7%
m. Fat or bald subjects are needed to fill a
questionnaire for a scientific experiment.
Who fits more?
27
8
77%
n. It is true that if Danny fits Moshe definitely
fits?
5
29
15%
o. It is true that if Moshe fits Danny definitely
fits

The answers to section 7 were basically Boolean in nature. Subjects were presented with two
creatures not satisfying the first conjunct (flying) and differing along the second (calling), the
latter being more typical of a calling creature than the former (non-calling creature). In this case,
both creatures obviously do not satisfy the conjunction and the former obviously does not satisfy
the disjunction. Most subjects disagreed about the latter being more typical of the modified noun
flying calling [creature] (29% agreement for 7a) and conjunction flying and calling [creature] (37%
agreement for 7b). Conversely, most subjects agreed about it being more typical of the disjunction
flying or calling [creature] (80% agreement for 7c).
The results on section 7 are somewhere in between those of section 3 and those of section 9.
This supports the hypothesis that the results on section 9 are explained by the two above
mentioned problems, i.e. an effect of repetition, together with characters not obviously violating
any of the constituents of the complex predicates in section 9.
Table 7d: The ‗repetition‘ condition - general
3. General discussion
We have seen that Hebrew speakers tend to dislike ordering entities under conjunctive or
disjunctive concepts when the two entities do not stand in one and the same ordering relation (e.g.
more) in all the constituents. The subjects have taken care to make this point clear in their
comments, stating that questions that ask them to do so are impossible to answer, are inappropriate,
that the complex (conjunctive or other) comparison will not be used by a person sensitive to the
language in such circumstances and so on. One subject has written that these questions are of the
same type as questions such as is a cucumber more long or more green. We cannot tell. In fact
answering the latter is impossible for precisely the same reasons; two properties that are not easily
comparable are involved. They are not easily comparable precisely because they do not consist of
mappings of entities to degrees on a bound interval isomorphic to the real interval between 0 and 1.
In accordance with these comments, the results presented in part 2 show that a truly
compositional interpretation, i.e. a degree function (or ordering) of a conjunction/disjunction of
properties, composed based on the degree functions (or ordering relations) of the constituents, is
hardly ever occurring naturally in the absence of a particularly encouraging context. Evidence for
this is, for example, the fact that in section 3, the two characters are described as equally tall
(1.95m), but Moshe is 30kg fatter, and still most subjects refuse to say that either Moshe or Danny
is more fat and tall (3e-h). These results suggest that speakers are, more often than not, unwilling to
interpret and inside the scope of more, i.e. more P and Q is interpreted as more P and more Q, not
as More (P and Q). The same holds for or. These results are in line with a Boolean analysis more
than with a fuzzy one. However, these preliminary findings are restricted to contexts of the nature
the questionnaire examined. Other types of contexts should be investigated in the future, as well as
the effect of repetition on interpretation. In addition, other languages should be studied.
6
6
Novel findings (currently in process) suggest that English is similar to Hebrew in that conjunctive and disjunctive
predicates such as expensive and time consuming appear incompatible with gradable morphology and interpretation.
‗Yes‘
‗No‘
%‗Yes’
Section 7
Two non-flying creatures; The first one
doesn‘t call; The second one is more typical of
a calling creature
10
25
29%
a. Is the second more typical of a flying calling
[creature]?
13
22
37%
b. Is the second more typical of a flying and
calling [creature]?
28
7
80%
c. Is the second more typical of a flying or
calling [creature]?

The results are also consistent with the hypothesis that a use of a truly compositional degree
function (or ordering) may be facilitated by repetition of discussion and processing of the
interpretation of complex concepts such as fat and bald (cf., the results of section 9). Also,
mentioning a conjunctive comparison, e.g. more fat and bald, following a conjunction of
comparisons, e.g. more fat and more bald, may trigger such an interpretation. Therefore, it may be
useful to test such contexts in the future, separating the two variables (repetition and conjunction of
comparisons) from one another.
7
Another variable affecting the results is whether the ranked entities obviously violate one of
the atomic concepts in question or not. That membership or non-membership in the categories in
question plays an important role in judgments of ordering is in line with vagueness based
accounts of gradability in natural language. According to theories in this approach, an entity pair
<x,y> is classified as more P iff, roughly, either x is P but y not necessarily P, or alternatively y is
not P but x is not necessarily not P (Kamp 1975; Fine 1975; Klein 1980; van Rooij 2009). The
present study shows that when subjects are forced to give a comparative judgment about the
degrees of instances with respect to conjunctions or disjunctions, judgments seem to be
determined by the characters‘ likelihood of classification in the positive or negative denotations
of the conjunction or disjunction. For example, when Moshe weighed 100 kg and Danny only 70
kg (Moshe was fatter), but Danny was bald and Moshe was not bald (Danny was balder),
according to the few subjects that selected one character to be more fat and/or bald than the
other, Danny scored much better than Moshe relative to fat or bald (questions 4), because it is
easier to determine that Danny is fat or bald (Danny is bald, while Moshe is not bald and is not
necessarily fat); also, Danny scored much better than Moshe relative to fat bald / bald fat,
because it is easier to determine that Moshe is not: fat and bald (because Moshe is not bald),
while Danny might still be fat and bald. A similar pattern occurred in other scenarios as well.
When the two characters were equally tall (1.95m), but Moshe was fatter (100 kg as opposed to
Danny's 70 kg), among those who answered the question, Moshe was regarded as more fat and
tall than Danny (section 3). When the characters where fat, and equally fat (100 kg), but Moshe
was balder, 90%-100% of the subject agreed that Moshe is more fat or bald than Danny (section
9). Future research may, then, profit from asking what happens when knowledge about denotation
membership is more uncertain.
A different issue raising many questions yet to be examined regards different sorts of ways to
refer to ordering relations with adjectives (e.g., more versus less comparatives, as well as simple
versus complex comparison). Different ways to refer seem to differ in a variety of respects, one
of which is the extent to which they raise the expectation that the ordered entities are predicate
members; this expectation may also be stronger for the subject or for the object (Sassoon 2007).
These issues call for further empirical investigation.
The phenomenon this paper investigates pertains also to conjunctive multi-dimensional
adjectives, like typical with respect to flying and singing or healthy with respect to blood pressure
and pulse, when these are combined with more. Our results suggest that, by and large, and takes
wide scope, and more combines with a single function at a time (per a conjunct); e.g., we
understand healthier with respect to blood pressure and pulse to mean healthier with respect to
blood pressure and healthier with respect to pulse.
An important open question is to what extent ordering in conjunctive predicates can be based
on the entities' weighted mean in the constituents, in context in which information about weights
is made available. If, for example, I compare a patient with cancer to a patient with the flu, I may
weigh the cancer as more important, and judge the former patient as having a higher degree of
sickness with respect to the conjunction of these dimensions. When context does not tell us how
7
Novel findings (currently in process) strongly support the assumption that it is repetition, rather than conjunction of
comparisons, that facilitates fuzzy interpretations.

the conjuncts are to be weighed (for instance, if I compare a patient with cancer to a patient with
serious heart problems), this strategy might fail. When it fails, we are left with a wide-scope
interpretation like healthier with respect to cancer and healthier with respect to the heart, and we
are likely to be reluctant to say about any of the patients that he is healthier (in the given
conjunction of respects). Thus, another open question for the future is to examine the effect of
constituents with variable weights.
An alternative strategy for construing an ordering relation for conjunctive and disjunctive
adjectives is by using the mean in the typicality dimensions of the modified nouns corresponding
to them. Any adjective can modify a trivial noun such as object, individual, one. Like other noun
phrases, the noun phrase healthy entity is linked with a set of typicality dimensions such that
entities whose mean degree in these dimensions is high are classified as members in the
denotation (cf. Murphy 2002‘s review of data concerning basic nouns and Hampton‘s 1997a,b
review of data concerning nouns modified by a nominal relative clause, such as birds which are
pet); examples of dimensions typical of healthy people include calm, does not smoke, does not
drink, does not eat fat, eats fruit and vegetables, is regularly involved in sport activities, etc. Such
dimensions may be directly linked to modified nouns such as healthy person, but only indirectly
related to the adjective healthy itself, for otherwise we would expect ordering judgments in
conjunctions and disjunctions of adjectives to be as easy as the typicality judgments are in nouns
modified by a nominal relative clause or in noun-noun compounds. The study presented in this
paper suggests that this is not the case.
In sum, compositionality, in the case of the ordering in conjunctions and disjunctions, more
often than not, fails. In many cases (for many pairs of entities) their ordering in a conjunction (or
in a disjunction) cannot be predicted from their ordering in the constituents. Thus, while a fuzzy
analysis could have been considered a natural extension of the numerical approach (cf. Kennedy
1999), this no longer seems a viable option. Rather, ordering relations for conjunctions and
disjunctions need to be learnt directly based on whether entities are classified in their intersective
and union-based denotations or not, as well as, perhaps, on context-dependent information
regarding the constituents‘ weights or their typicality features. However, knowledge concerning
these issues is often not available. Thus, conjunctions and disjunctions tend not to license
gradable morphology.
Last but not least, what are the implications for the numerical versus vagueness-based debate?
On the one hand, the fact that conjunctive and disjunctive predicates do not appear gradable is
surprising given a vagueness-based approach, for if we use denotations (rather than numerical
measurements) to systematically build ordering relations, why aren‘t we able to systematically
use the denotations of conjunctive and disjunctive concepts to do so? After all, these predicates
are at least as vague as their constituents are. The numerical approach fairs better over here; since
and and or are merely Boolean, conjunctive and disjunctive predicates denote entity sets, not
degree functions, which explains why they are non-gradable. On the other hand, we have seen
that when we do make gradable judgments in relation to complex predicates, they do seem to go
along the line suggested by vagueness based theories. This is probably the case because no other
option (no unified numerical degree function) is available.
Hence, the general moral to draw from all the above must be in favor of a combined
approach. Both measurement-based degree functions and vagueness-based ordering relations play
a role in the semantics of natural language expressions (for a discussion and a detailed model see
Sassoon 2007). In addition, different predicate types may be associated with different types of
degree functions (e.g., numerical versus ordinal; cf. Sassoon 2010a).

References
Alxatib, Sam. and Pelletier, Jeff. (2010). The psychology of vagueness: Borderline cases and
contradictions. Forthcoming in Mind and Language.
Bale, Alan. C. (2007). Boolean AND and the Semantic Correlates of Gradable Adjectives.
International Conference on Adjectives. Lille (France), 13 15, September, 2007.
Bale, Alan. C. (2008). A universal scale of comparison. Linguistics and Philosophy, 31(1), 155.
Bartsch, Renate., & Venneman, Theo. (1972). Semantic structures: 4786. Frankfurt: Anthenäum
Verlag.
Cresswell, Maxwell John. (1977). The semantics of degree. In Barbara Partee (Ed.), Montague
grammar : 261292. New York: Academic Press.
Fine, Kit. (1975). Truth, vagueness and logics. Synthese, 30, 265300.
Hampton, James. (1987), Inheritance of Attributes in Natural concept conjunction. Memory and
Cognition 15(1): 55-71.
Hampton, James. (1988a) Overextension of conjunctive concepts: Evidence for a unitary model of
concept typicality and class diagnosticity. Journal of Experimental Psychology: Learning,
Memory & Cognition, 14, 12-32.
Hampton, James. (1988b), Disjunction of natural concepts, Memory & Cognition, 16(6): 579-
591.
Hampton, James. (1997a), Conceptual combination: Conjunction and negation of natural
concepts, Memory & Cognition, 25 (6): 888-909.
Hampton, James. (1997b), Conceptual Combination. In Koen Lamberts and David Shanks (eds.),
Knowledge, Concepts and Concepts. The MIT Press. Cambridge, MA: 135-162.
Hajek, Petr. (2009). Fuzzy logic. In Edward N. Zalta (eds.), The Stanford Encyclopedia of
Philosophy. Spring 2009 edition.
Heim, Irene. (2000). Degree operators and scope. Proceedings of SALT X. Ithaca, NY: CLC
Publications.
Kamp, Hans. (1975). Two theories about adjectives. In Edward Keenan (ed.), Formal semantics
for natural language (pp. 123155). Cambridge: Cambridge University Press.
Kamp, Hans, & Partee, Barbara. (1995). Prototype theory and compositionality. Cognition, 57,
129191.
Kennedy, Christopher. (1999). Projecting the adjective: The syntax and semantics of gradability
and comparison. New York: Garland (1997 UCSC Doctoral dissertation).
Kennedy, Christopher. (2001). Polar opposition and the ontology of degrees. Linguistics and
Philosophy, 24(1), 3370.
Kennedy, Christopher, & Levin, Beth. (2007). Measure of change: The adjectival core of degree
achievements. In Louise McNally & Christopher Kennedy (eds.), Adjectives and adverbs:
Syntax, semantics and discourse. Oxford: Oxford University Press.
Kennedy, Christopher, & McNally, Louise. (2005). Scale structure and the semantic typology of
gradable predicates. Language, 81, 345381.
Klein, Ewan. (1980). A semantics for positive and comparative adjectives. Linguistics and
Philosophy, 4, 145.
Klein, Ewan. (1991). Comparatives. In Arnim von Stechow & Dieter Wunderlich (eds.),
Semantik/semantics, an international handbook of contemporary research (pp. 673691).
Berlin, NY: Walter de Gruyter.
Landman, F. (1991). Structures for semantics. Dordrecht: Kluwer.
Landman, Fred. (2000). Events and Plurality, Kluwer, Dordrecht
Landman, F. (2005). An almost (but not quite) naïve theory of measures, manuscript, Tel Aviv
University.
Lewis, David. (1970). General semantics. Synthese, 22, 1867. Reprinted In David K. Lewis
(1983). Philosophical Papers, 1, 189229. New York: Oxford University Press.

Lewis, David. (1979). Scorekeeping in a language game. Journal of Philosophical Logic, 8, 339
359. (Reprinted In Philosophical Papers, 1, pp. 233249, by D. K. Lewis (ed.), 1983, New
York: Oxford University Press.
Moltmann, Friederike. (2006). Comparatives without degrees. A new approach. A manuscript for
the workshop on scalar meaning, University of Chicago.
Murphy, Gregory L. (2002). The big book of concepts. Cambridge, MA: The MIT Press.
Ripley, David. (2010). Contradictions at the borders. Forthcoming in: Rick Nouwen, Uli
Sauerland, Hans-Christian Schmitz, and Robert van Rooij, editors. Vagueness in
Communication. Springer, Heidelberg, Germany.
Russell, Bertrand. (1905). On denoting. Mind, 14, 479493 (Reprinted from Essays in Analysis,
pp. 103119, by Russell, B., 1973, London: Allen and Unwin).
Sapir, Edward. (1944). Grading: A study in semantics. Philosophy of Science, 11, 93116.
Sassoon, W. Galit. (2007). Vagueness, gradability and typicality, a comprehensive semantic
analysis. Doctoral Dissertation, Tel Aviv University.
Sassoon, W. Galit. (2010a). Measurement theory in linguistics, Synthese 174(1):151180
Sassoon, W. Galit. (2010b). The degree functions of negative adjectives. Natural language
semantics 18(2), 141-181.
Sauerland, Uli, (2011), Vagueness in Language: The Case Against Fuzzy Logic Revisited. In:
Petr Cintula, Chris Fermüller, Lluis Godo, Petr Hajek (eds.), Reasoning under Vagueness -
Logical, Philosophical, and Linguistic Perspectives, Studies in Logic series of College
Publications.
Schwarzschild, Roger. (2008). The semantics of comparatives and other degree constructions.
Language and Linguistics Compass, 2.2, 308331.
Schwarzschild, Roger, & Wilkinson, Karina. (2002). Quantifiers in comparatives: A semantics of
degree based on intervals. Natural Language Semantics, 10, 141.
Serchuk, Phil, Hargreaves, Ian, and Zach, Richard. (2010). Vagueness, logic and use: Four
experimental studies on vagueness. Forthcoming in Mind and Language.
Seuren, Peter. (1978). The structure and selection of positive and negative gradable adjectives. In
Papers from the Para-session on the Lexicon. 14th Regional Meeting of the Chicago
Linguistic Society (pp. 336346). CLS, Chicago, IL.
Stalnaker, Robert. (1978). Assertion. In Peter Cole, Syntax and semantics 9: Pragmatics (pp.
315332). New York: Academic Press.
van Fraassen, Bas C. (1969). Presuppositions, supervaluations and free logic. In K. Lambert, The
logical way of doing things (pp. 6791). New Haven: Yale University Press.
van Rooij, Robert. (2009). Vagueness and linguistics. In Giuseppina Ronzitti (ed.), The
vagueness handbook. Dordrecht: Springer.
Veltman, Frank. (1984). Data semantics. In Jeroen Groenendijk, Theo Janssen & Martin Stokhof
(eds.), Truth, interpretation and information proceedings of the 3rd Amsterdam colloquium.
von Stechow, Arnim. (1984a). Comparing semantic theories of comparison. Journal of
Semantics, 3, 177.
von Stechow, Arnim. (1984b). My reaction to Cresswell‘s, Hellan‘s, Hoeksema‘s and Seuren‘s
comments. Journal of Semantics, 3, 183199.
Yen, John (1999). Fuzzy LogicA Modern Perspective, Transactions on Knowledge and Data
Engineering 11(1): 153-165.
Zadeh, Lotfi A. (1965). Fuzzy sets. Information and Control 8(3): 338353.

Appendix
On section 1 the subjects read the following description of two characters called Moshe and
Danny: ―Assume Moshe weighs 100kg and Danny weighs 90kg and they are alike in other things
(for instance, height). Here are the questions that followed the paragraph.
(17) Section 1, the questions:
The basic condition
a. Is Moshe more fat than Danny? Yes/No
b. Is it easier to determine that Moshe is fat than that Danny is fat? Yes/No
c. Is it harder to determine that Danny is fat than that Moshe is fat? Yes/No
On section 3, the subjects read the following description: ―Assume Moshe weighs 100kg and he
is 1.95 cm tall, and Danny weighs 70 kg and is he is 1.95 cm tall.‖ (i.e., Moshe is fatter than
Danny, but they are equally tall).
(18) Section 3, the questions:
The basic condition
a. is Moshe more fat than Danny? Yes/No
b. Is it easier to determine that Moshe is fat than that Danny is fat? Yes/No
c. is one of them more tall than the other? Yes/No
d. Is it easier to determine that he is tall? Yes/No
The conjunction condition
e. Is Moshe more fat and tall than Danny? Yes/No
f. Is Moshe less fat and tall than Danny? Yes/No
g. Is it easier to determine that Moshe is fat and tall than that Danny is? Yes/No
h. Is it harder to determine that Moshe is fat and tall than that Danny is? Yes/No
i. Comments: [three empty lines]
On section 4, the subjects read the following description of characters called Aharon and Danny:
―Assume Aharon weighs 100kg and he is not bald, and Danny weighs 70 kg and is bald. They are
alike in other respects.‖ Thus, in this scenario Aharon is fatter and Danny is balder.
(19) Section 4, the questions:
The basic condition
a. Who is more fat? Aharon/ Danny
b. Who is more bald? Aharon/ Danny
The conjunction condition
c. Is any of them more fat and bald than the other? Yes/No
d. Is it easier to determine that one of them is fat and bald than that the other is?
Yes/No
e. Are they equally fat and bald? Yes/No
f. Is it equally easy to determine that they are fat and bald? Yes/No
The disjunction condition
g. Is one of them more fat or bald than the other? Yes/No
h. Who is more fat or bald? Aharon/ Danny
i. Is it easier to determine that one of them is fat or bald than that the other is?
Yes/No
j. For whom is it easier to determine that? Aharon/Danny
The modifier condition

k. Is any of them more fat bald than the other? Yes/No
l. Who is more fat bald? Aharon/ Danny
m. Is it easier to determine that one of them is fat bald than that the other is?
Yes/No
n. For whom is it easier to determine that? Aharon/Danny
o. Is any of them more bald fat than the other? Yes/No
p. Who is more bald fat? Aharon/ Danny
q. Is it easier to determine that one of them is bald fat than that the other is?
Yes/No
r. For whom is it easier to determine that? Aharon/Danny
The disjunction condition - continued
s. Fat or bald subjects are needed to fill a questionnaire for a scientific experiment.
Who fits more? Aharon/Danny
t. It is true that if Danny fits Aharon definitely fits? Yes/No
w. It is true that if Aharon fits Danny definitely fits Yes/No
u. Comments: [3 lines]
Section 9 began with the following description: ―Assume Moshe and Danny both weigh 100kg
and also that Moshe is bald and Danny is less bald (maybe even isn‘t bald at all).‖ Thus, here
Moshe and Danny are equally fat, but Moshe is balder (the same pattern as in section 3). They are
alike in other respects.‖
(20) Section 9, the questions:
The conjunction condition
a. Is Moshe both more fat and more bald? Yes/No
b. Is Moshe more fat and Bald than Danny? Yes/No
c. Is it easier to determine that Moshe is fat and bald than that Dan is fat and bald?
Yes/No
d. Is it easier to determine that Moshe is bald and fat than that Dan is bald and fat?
Yes/No
The modifier condition
e. Is Moshe more fat bald than Danny is? Yes/No
f. Is it easier to determine that Moshe is fat bald than that Danny is fat bald?
Yes/No
g. Is Moshe more bald fat than Danny is? Yes/No
h. Is it easier to determine that Moshe is bald fat than that Danny is bald fat?
Yes/No
The disjunction condition
i. Is Moshe more fat or Bald than Danny? Yes/No
j. Is it easier to determine that Moshe is fat or bald than that Danny is?
Yes/No
k. Is Moshe more Bald or fat than Danny? Yes/No
l. Is it easier to determine that Moshe is bald or fat than that Dan is?
Yes/No
m. Fat or bald subjects are needed to fill a questionnaire for a scientific experiment.
Who fits more? Moshe/Danny
n. It is true that if Danny fits Moshe definitely fits? Yes/No
o. It is true that if Moshe fits Danny definitely fits Yes/No
p. Comments: [3 lines]

On section 7 the subjects read a description of two creatures differing only on whether they call
or not: ―Imagine two creatures that do not fly; also, the first one does not call, while the second is
more typical of a calling creature than the first‖.
(21) Section 7, the questions:
The multidimensional, conjunctive adjective condition
a. Is the second more typical of a flying calling [creature]? Yes/No
b. Is the second more typical of a flying and calling [creature]? Yes/No
c. Is the second more typical of a flying or calling [creature]? Yes/No
d. comments [3 lines]