Content uploaded by Zhengjie Situ
Author content
All content in this area was uploaded by Zhengjie Situ on Nov 09, 2020
Content may be subject to copyright.
1
An ‘alternative’ theory of generics: the case for the existential reading
Examination Number: B157571
Word Count: 7990
MSc Linguistics
The University of Edinburgh
2019/2020
2
Abstract
Generics are statements expressing generalizations over a kind or, roughly, relevant
members thereof. Despite being prevalent in people’s speech, the exact meaning of bare
plural generics has long been controversial among linguists and philosophers. In this
essay I aim to show that 1. Contextually induced alternatives play a central role in the
interpretation of generics; 2. Generic statements have a basic, existential component that
is insensitive to prevalence conditions; 3. Switching from a quasi-universal-as-default to
an existential-as-default mindset opens up more theoretical possibilities. A brief sketch
of a list of theoretical desiderata is followed by a critical review of Cohen (1996, 2004)’s
probability theory and Leslie (2008)’s cognitively oriented theory. I claim that despite
looking starkly different, they display a surprising amount of commonality, both in terms
of theoretical virtues and weaknesses. After reflecting on the common but unexamined
practice of positing a quasi-universal quantificational structure as default, I argue that a
paradigm shift to a quasi-existential based framework is worth pursuing. At last, making
use of contextually induced alternatives and cognitive heuristics, I develop an alternative
account of generics that combines the strengths of previous accounts while avoiding their
shortcomings.
Key words: Generics, Genericity, Bare plurals, Alternatives, Quantification
1 Introduction
Generics are statements expressing generalizations over a kind or, approximately,
relevant members thereof. Below are examples accompanied by intuitive judgements.
(1) a. Ravens are widespread.
b. Ravens are black.
c. Ravens are black and widespread.
(2) a. Chickens lay eggs.
b. ? Chickens are female.
(3) a. Lions have manes and give birth to live young.
b. Elephants live in Africa and Asia.
(4) a. Ticks carry Lyme disease.
b. Japanese people are good table tennis players.
I use the notation K (F) to characterize what is common among these constructions. First,
they all have a simple subject-predicate surface structure in which some property F is
predicated on the subject nominal K. Second, the subject is in the bare plural (BP) and
contains no overt quantifier (e.g. some, all) indicating how many instances of individual K
to which should (F) apply. Third, they apparently express generalizations not of individual
3
K but of K as a kind. Although the exhibition of genericity in English is not restricted to
sentences with BP subjects, they are the most common and the most studied. I thus leave
other potentially relevant nominals (definite singulars and indefinite singulars) out of my
discussion in the hope that what I propose for BP generics can be extended to them.
Despite being prevalent in people’s speech, the exact meaning of bare plural generics
has long been controversial among linguists and philosophers. In this essay I aim to show
that 1. Contextually induced alternatives play a central role in the interpretation of
generics; 2. Generic statements have a basic, existential component that is insensitive to
prevalence conditions; 3. Switching from a quasi-universal-as-default to an existential-
as-default mindset opens up more theoretical possibilities.
This essay is organized as follows: section 2 combs through linguistic data to identify
some puzzles that any theory of generics should account for; Section 3 presents two
influential existing accounts: Cohen (1996, 2004) and Leslie (2008). I then discuss their
advantages and shortcomings. Section 4 highlights some common practices among
discussions of generics, analyzes the implicit assumptions made and offers evidence to
challenge those assumptions. I then demonstrate how those unexamined practices are
linked to the problematic theoretical commitments made by Cohen and Leslie. Section 5
presents a set of data demonstrating the existence of an existential reading of generics.
Section 6 proposes a theory that combines the strengthens of previous accounts while
overcoming their weaknesses. I usher in an alternative framework to think about generics,
and explain why it enjoys unique advantages over the old framework. Section 7
demonstrates the workings of my theory. Section 8 concludes.
2 A survey of the theoretical landscape
2.1 Direct kind predication vs characterizing sentences
Generics do not form a homogeneous group. A distinction made in the literature can be
highlighted by comparing (1)a and (1)b, repeated as (5)a and (5)b:
(5) a. Ravens are widespread.
b. Ravens are black.
No individual raven can be widespread, only the kind ‘ravens’ can. Predicates such as
‘widespread’, which demand a kind-denoting nominal serving as subject, are called kind-
selecting predicates. Tthe interpretation of (5)a is relatively clear: it roughly means a
predication of the property ‘widespread’ to the kind entity ‘ravens’, which can be
schematized by a simple bipartite structure below:
Ravens (widespread)
Sentences like (5)a are called direct predications (Krifka et al. 1995). The evaluation of its
truth value is straightforward: checking whether the relevant property is instantiated by
4
the relevant kind. If so, the sentence is true, otherwise it is false.
Let’s turn to (5)b. An initial impression one gets is that here the property ‘black’ is
predicated of some characteristic instances of the kind ‘ravens’ rather than the kind itself
(sentences like (5)b are thus called characterizing sentences). This impression is backed
by intuition: it is hard to conceive how a kind can be black when some of its members,
e.g. albino ravens, display non-black colours. This impression is further supported by
linguistic tests. Note that (5)a cannot be modified by an overt adverb of quantification
while (5)b can:
(6) a * Ravens are usually widespread.
b Ravens are usually black.
Since the instantiation of the property ‘widespread’ by the kind ‘ravens’ is not temporally
confined, inserting a frequency adverb to a direct kind predication results in oddness.
Another test employs scope. Characterizing sentences, but not direct kind predications,
display scope ambiguities. To illustrate, the characterizing sentence (7)a can mean either
that each salmon has a (possibly unique) spawning ground or there is one spawning
ground being generally favored by salmon. In contrast, the direct kind predication (7)b
can only mean that salmon as a kind is facing extermination.
(7) a. Salmon have a favorite spawning ground.
b. Salmon are at risk of distinction through overfishing.
These observations suggest that direct kind predications and characterizing sentences
deserve different treatments. Some theorists thus posited that characterizing generics’
meaning is represented by a tripartite structure consisting of a GEN(eric) Operator, a
Restrictor, and a Matrix. So a characterizing generic in the form of K (P) has the following
logical form, where x stands for a free variable:
GEN (x) [K (x) ] [P (x)]
This is the received view about the logical form of characterizing generics, despite much
disagreements among theorists about the exact workings of the GEN operator. However,
the possibility of co-predicating different types within a single sentence calls for a uniform
account (1(c), below repeated as (8)):
(8) Ravens are black and widespread.
2.2 Chicken failure: quantification with domain restrict?
Last subsection seems to suggest that characterizing sentences pick out ‘the majority’ or
‘the normal/typical members’, since black ravens constitute a majority, are normal and
are typical. An initial conjecture would be that generic sentences involve
5
universal/majority quantification with some normality/typicality operator. A quick look at
other examples (2(a-b), repeated as (9)a-b below), however, shows that things are more
complicated. What (9)a expresses cannot be that the majority of chickens lay eggs: (9)a
would still be perceived as true even if it happened to be the case that fertile female
chickens constitute less than 50% of the chicken population. Nor does it express some
normality or typicality claim since roosters are not abnormal or atypical chickens. The
closest paraphrase is ‘Fertile female chickens lay eggs’. Facing with such data, a theorist
is motivated to revise her account accordingly: being informed by the selective
constraints placed by its predicate, characterizing sentences apparently can restrict its
domain of application to relevant members only.
(9) a. Chickens lay eggs.
b. ? Chickens are female.
Such an account is too liberal, however, for it overgenerates. Whatever mechanism
proposed that allows for a domain restriction to only ‘female chickens’ in (9)a, seems
equally capable of allowing for that domain restriction in (9)b. That results in a reading of
(9)b as ‘female chickens are female’ which should be trivially true. Yet people judge (9)b
to be utterly unacceptable.
This problem, which so far has plagued all accounts of generics that make use of ideas
similar to a domain restriction, is called ‘chicken failure.’ (Nickel 2016)
2.3 The Chimera Paradox
Even if ‘chicken failure’ is handled, the quantificational account faces other problems.
Copredications predicating incompatible properties to the same nominal are allowed in
characterizing sentences, as illustrated by (3)a-b, repeated below as (10)a-b:
(10) a. Lions have manes and give birth to live young.
b. Elephants live in Africa and Asia. (Nickel 2016)
For (10)a, while it is possible restrict the domain to two sub-groups such that all/most
members of each either have manes or give birth to live young, it is impossible to locate
a sub-group such that all/most members do both, since only male lions have manes and
only female lions give birth. The same applies to (10)b, since not a single elephant both
generally lives in Africa and generally lives in Asia. This set of data poses problems not
only for the quantificational component of previous accounts, but equally for the
normality/typicality components. Suppose a ‘Chimera kind member’, such as a lion that
both has a mane and gives birth to live young or an elephant that does live in both Africa
and Asia, is posited to exist out of theoretical necessity. That move backfires since such
Chimera kind member is intuitively neither normal nor typical. I thus term this puzzle the
Chimera Paradox of generics.
2.4 Striking property generalization and the Port Royal Puzzle (PRP)
6
At last, consider example (4)a-b, repeated below as (11)a-b.
(11) a. Ticks carry Lyme disease.
b. Japanese people are good table tennis players.
Examples like (11)a, called ‘striking property generics’, were first highlighted by Leslie
(2008). Carriers of Lyme disease constitute only a tiny minority of the tick population (less
than 1%), so a simple universal/majority quantificational account is implausible. Disease-
carrying capacity is not a characteristic feature of a species either (to show a contrast,
note it is normal for a child to expect every newly encountered bird to lay eggs but not
so to expect every newly encountered insect to transmit diseases). Claiming that only
disease-carrying ticks are normal/prototypical members of the tick kind is counter-
intuitive too. It is puzzling, then, what license such statements’ easy acceptability. One
plausible suggestion by Leslie (2008) is the nature of predicates: note that the potential
for carrying disease is perceived to be especially dangerous, harmful or striking. Learning
about members of a kind’s potential to do harm is of high value even if the actual
prevalence is low, since the consequences can be devastating in case that harmful
potential gets realized.
The case of (11)b, adapted from the Port Royal Puzzle sentence ‘Dutchmen are good
sailors’, are similar to striking property generics in many ways. Good Japanese table tennis
players constitute a tiny fraction of the whole Japanese population. Being a good table
tennis player cannot be considered a characteristic/normal/prototypical feature of a
Japanese person either. These observations suggests that (11)b should be handled in the
same way as (11)a. However, the crucial difference between them blocks this possibility
(at least under Leslie’s account), since the property of being a good table tennis player is
not at all dangerous, harmful or striking.
3 Existing accounts
Having outlined a list of desiderata, thenI review two approaches which have been
influential.
3.1 Cohen’s account
Cohen (1996, 2004) offered a comprehensive probability-based account which is one of
the most promising within the last two decades. One main innovation of this account is
the use of contextually induced alternatives. To illustrate, for (9)a, alternatives to the
predicate ‘lay eggs’ can be ‘give live birth’, ‘undergo mitosis’ and such. Left as a pre-
theoretic notion, alternatives roughly correspond to constructions of the same type which
can replace parts of the original characterizing sentence. The presentation here might
suggest that alternatives are only explicitly recalled upon reflection, but Cohen’s claim
about alternatives is more substantial: they are posited to implicitly feature in the meaning
computation of characterizing sentences. For a characterizing statement in the form of K
(F), where ALT(F) represents the set containing F as well as all contextually induced
7
alternatives to F, the truth condition for K (F) is proposed to be:
Absolute generic reading: K (F) =1 iff P(F/A∩ V ALT(F))>0.5
Informally, under this reading, K (F) is true if and only if an arbitrary member of K is more
likely to satisfy the predicate F than not, given that it satisfies some predicate in ALT(F).
This set of truth conditions correctly predicts the truth of statements such as (5)b and (9)a:
an arbitrary raven that is of some color is more likely to be in black than in any non-black
color, and an arbitrary chicken that engages in some form of reproduction is more likely
to lay eggs than to give live birth or undergo mitosis.
One major improvement of Cohen’s theory over prior accounts is its ability to
accommodate ‘striking property’ generics. Consider 11(a) ‘Ticks carry the Lyme disease’.
Here ALT(F) would be {carry the Lyme disease, carry the West Nile virus, carry cholera……}.
Prompted by this alternative set, people only consider disease-carrying ticks in the
meaning computation process. Since any disease-carrying tick is more likely to carry
Lyme disease than not, the statement is correctly predicted to be true.
Yet this absolute reading cannot explain PRP. Consider 11(b). Here ALT(F) would be {are
good football players, are good baseball players……}. In order for (11)b to be true under
this reading, an arbitrary Japanese person must be more likely to be a good table tennis
player than not, given that he/she is good at some sports. That is simply not the case:
good Japanese players in all other sports together far outnumber good Japanese table
tennis players. So (10)b is incorrectly predicted to be false.
To tackle this issue, Cohen claims that on top of the absolute reading, generics can receive
another ‘relative’ reading. In addition to ALT(F), truth conditions laid out by this relative
reading requires the computation of a set consisting of the subject K and alternatives to
K, designated as ALT(K). Crucially, ALT(K) is not purely contextually supplied. All members
of ALT(K) must satisfy at least some property listed in ALT(F).
Relative generic reading: K (F) =1 iff P(F/K∩ V ALT(F)) > P(F/ ALT(K)∩ V ALT(F))
To illustrate, the truth condition of (11)b is roughly that an arbitrary good Japanese
sportsperson is more likely than an arbitrary good ‘International’ sportsperson to be a
good table tennis player. Although far from being perfect, this holds some promise
towards aligning with speaker intuitions.
Informally, under this reading, K (F) is true if and only if the probability that an arbitrary
member X of K satisfies the predicate F given that X satisfies some predicate in ALT(F), is
higher than the probability that an arbitrary member Y of ALT(K) satisfies the predicate F
given that Y satisfies some predicate in ALT(F).
8
However, even with this second reading Cohen’s account faces difficulty dealing with the
‘chicken failure’. Consider (9)b ‘Chickens are female’. Here ALT(F) would be {female, male}.
Since female chickens are being industrially farmed at a much larger scale than their male
counterparts, an arbitrary chicken (of some sex) is more likely to be female than male. It
is thus predicted that (9)b is true under the absolute reading, contrary to speaker intuition.
Resorting to the relative reading does not help either, since it is also the case that an
arbitrary chicken (of some sex) is more likely than an arbitrary animal (of some sex) to be
female, assuming all other animals have a roughly balanced sex ratio.
To salvage the situation Cohen offered another stipulation in the form of the
‘homogeneity constraint’:
Homogeneity: if the kind K can be partitioned into different ‘cells’ in some psychologically
salient way, the truth conditions proposed must hold for all cells.
Gender obviously qualifies as a salient partitioning, so chickens are divided into the cell
of ‘male chickens’ and ‘female chickens’. Since the probability of satisfying the property
‘are female’ is always 0 with respect to the ‘male chickens’ cell, (9)b is ruled as false under
any reading, according to the Homogeneity constraint.
This constraint proves to be too blunt a device. Although it correctly rules out (9)b, it also
incorrectly rules out (9)a by the same criteria: the probability of satisfying the property
‘lay eggs’ is also always 0 with respect to the ‘male chickens’ cell. Since Cohen’s account
fail to account for all the relevant judgement data, ‘chicken failure’ remains an
outstanding challenge to it.
Other than the technical issue of making incorrect predictions, Cohen’s account suffers
from a number of more fundamental conceptual issues. First, the account entails
systematic structural ambiguity: it holds that every characterizing sentence has two
possible readings without giving any detail of how the interpreter disambiguates between
them and settle for the correct reading eventually. The widespread structural ambiguity
posited is already against people’s intuitions. To make matters worse, Cohen’s account
makes the even more implausible stipulation that one component of the meaning
computation for one of the readings is optional. Recall that to get the absolute reading,
only alternatives to the predicate need to be induced. To get the relative reading,
however, alternatives to the subject, on top of those of the predicate, need to be induced.
Since the account itself offers no guidance as to which reading should be prioritized, the
only clue for an interpreter to decide the matter is the truth value of each reading. But
the truth value for the relative reading cannot be computed unless one has already
induced alternative to the subject! This dilemma forces interpreters to induce alternatives
to subject all the time while those alternatives so induced will only be useful sometimes.
This would be a highly inefficient and thus implausible mechanism for language
comprehension. Moreover, this entails significant disambiguation cost sensitive to
behavioural testing, which has so far not been observed.
9
To summarize, while Cohen’s theory accounts for quasi-majority quantificational data,
striking-property data and PRP data, it fails to account for data regarding ‘chicken failure’.
Its very framework entails systematic structural ambiguity, an undesirable prediction that
is counter-intuitive and not corroborated by empirical testing.
3.2 Leslie’s account
Despite widespread internal disagreement, theorists within the formal semantics tradition
have generally embraced the received view about the logical form of generics, i.e. the
existence of GEN as some kind of quantificational operator. More recently, however, an
alternative approach paying more attention to cognitive processes is gaining traction
within psychology circles. On this approach, genericity is not so much a semantic
phenomenon but a psychological one: the ‘quirky’ truth conditions of generic sentences,
ever defying any systematic semantic theorizing, comes not as a surprise since out basic
mechanisms of categorization and generalization are open to various kinds of biases. The
most popular view within this approach is the Generics-as-Default (GaD) view which
holds that generic sentences are manifestations of the mind’s innate, default and
cognitively primary mode of generalization (Leslie 2007, 2008). One virtue of this view is
its alignment with acquisition data which is not often considered by formal semanticists.
As convincingly argued by Leslie:
“A puzzling question now arises: how does a language learner ever come to master generics?
Not only is the interpretation of Gen rather complicated, the operator is not even phonologically
realized... To make matters all the more puzzling, it happens that generics are acquired quite
early on. Children start using generics by two years of age, which is significantly earlier than
explicit quantifiers (Gelman, 2003; Roeper, Strauss, & Zurer Pearson, 2006). That children ever
master generics is perplexing; that children master them more readily than explicit quantifiers
borders on the paradoxical. This is a phenomenon that demands explanation. (2008, p. 19)”
This ‘Paradox of Acquisition’ is one of the main motivations for Leslie to hold that generic
generalizations are both simpler and qualitatively different from quantificational ones.
Treating generic sentences as reflecting basic mechanisms of cognition confers an
additional advantage: cognitive biases of the human mind becomes one’s theoretical
ammunition. It was no coincidence that Leslie was the first to highlight the set of ‘striking
property’ generics. If the trait of risk avoidance, due to its high survival value, is deeply
encoded in our ancestors’ minds and selected for through evolutionary forces, it is no
surprise that generic sentences predicating some ‘dangerous, harmful or striking
properties’ are especially prone to acceptance, assuming that generic sentences reflects
our most primitive way of conveying generalizations.
To account for the acceptability of generic statements such as ‘chickens lay eggs’, where
relevant members of the kind constitute a minority, Leslie again make a connection
between the nature of the predicate and our cognitive processes of categorization and
generalization. It is posited that children view every natural kind as having a set of
10
characteristic dimensions. For example, each kind of animal has a characteristic way of
locomotion, way of reproduction, a noise, a biological ornament etc. Any encounter with
an individual member of a kind serves as an invitation for a child to fill in values for those
dimensions. This tendency to label a kind by some characteristic features explains why
certain generics are exception-tolerant: exceptional members are regarded as
uncharacteristic of the kind. Note this proposal solves part of the ‘chicken failure’
mentioned earlier.
Leslie does not attempt to explain all linguistic data in terms of cognitive heuristics or
biases, however. For more run-of-the-mill generic sentences she simply leans back to
the formal semanticist tradition of a majority quantification. That introduces a
complication: the negation of a striking-property generic, such as ‘Ticks do not carry the
Lyme disease’ would come out true under majority-quantification. To solve this issue,
Leslie introduces an important distinction between positive and negative alternatives:
“I propose a powerful factor here is whether the counter-instances are positive rather than
negative. The distinction I have in mind is as follows: A positive counter-instance to Ks are F
occurs when an instance of the kind K has a concrete alternative property, that is, when it has a
positive alternative to the property F, while negative counter-instances occur when an instance
simply fails to be F. Whether a counter-instance counts as positive or negative is highly
dependent on the property being predicated. (2007, p.66)”
The idea is that negative counterinstances are more easily ‘overlooked’ by people’s
cognition. This stipulation kills two birds in one stone. One the one hand, it tackles the
negation of striking-property generics, since the counterinstances to such claims all have
a positive (striking) property. On the other hand, it solves what remains of the ‘chicken
failure’: the statement ‘Chickens are female’ are false because male chickens display the
positive property of being male rather than simply failing to be female.
Putting everything together, for a generic sentence in the form of ‘
Ks are F’
, Leslie
(2008:43) proposes the following metaphysical truth conditions:
“The counterinstances, if any, are negative; and
If F lies along a characteristic dimension for the Ks, then some Ks are F, unless K is an artifact
or social kind, in which case F is the function or purpose of the kind K;
If F is striking, then some Ks are F and the others are disposed to be F;
Otherwise, almost all Ks are F.”
Although being well-integrated with insights from both semantics and psychology,
Leslie’s account does not escape objections. I raise two here.
First, similar to Cohen’s, this account suffers from over-positing structural ambiguity. The
three-way ambiguity is arguably worse. The situation is even more complicated when
there are in-between categories. Some belong to more than one type of statements.
11
Consider ‘Firearms are lethal’, which apparently is a characteristic property predication
and striking property predication simultaneously. Some have subjects and predicates
which cannot nicely fit into any category posited. Consider ‘Countries which do no respect
womens’ rights do not respect general human rights’. Is the subject a social kind? Is its
human rights condition a characteristic feature of a country? Is the predicate ‘not respect
general human rights’ striking enough? Intuitions are not clear at all here. Leslie’s theory
needs to be further refined to escape from this line of attack.
Second, while the concept of ‘negative counter-instances’ serves the purpose of reining
in overgeneration, the concept of ‘positive counter-instances’ might allow too liberal a
construal. Since Leslie admitted that alternatives are contextually induced and
psychological in nature, there could be drastically different positive counter-instances to
one and the same predicate. If the exceptions to a generic sentence are negative counter-
instances in one construal but positive counter-instances in another, this spells trouble
for Leslie’s theory. Consider the ‘chicken failure’, which is supposedly solved by applying
the ‘characteristic-filling heuristic’ and the positive/negative counterinstance distinction.
Facing the statement ‘Chickens lay eggs’, it is of course natural to assume that male
chickens count as negative counterinstances by simply failing to lay eggs. But it is equally
natural to assume that male chickens count as positive counterinstances since they
fertilize eggs. Since there is nothing in Leslie’s theory to block such a construal, the
consequence is that ‘Chickens lay eggs’ can be either true or false depending on one’s
construal, a highly undesirable result.
4 Compare and Contrast
This section compares and contrasts the two accounts reviewed in last section. I argue
that despite looking starkly different, they display a surprising amount of commonality,
both in terms of theoretical virtues and weaknesses.
4.1 Common insights: contextually induced alternatives and psychological saliency
Both accounts make use of some notion of contextually induced alternatives. Cohen’s is
more explicit on that, having incorporated the process of alternative set generation into
the meaning computation of any generic sentence. Leslie’s use is more implicit in positing
the distinction between negative and positive counter-instances. Positive
counterinstances roughly correspond to predicate alternatives in Cohen’s theory.
Both accounts acknowledge the significance of psychological processes. Leslie’s goes
further in claiming that genericity is a psychological phenomenon through and through,
while Cohen’s is more conservative, only borrowing the concept of ‘psychological saliency’
to carve up the subject kind term and motivate the homogeneity constraint.
4.2 Common failings: the dogma of quasi-universal quantification
Common to many theories of generics is an implicit assumption that generic sentences
involve some kind of universal/majority quantification. Most articles (especially those
12
within the formal semantics tradition) on generics start by displaying example generic
sentences the interpretations of which look most similar to universal/majority
quantificational constructions. Almost every single article on generics makes some
comments that cannot be made sense of unless one embraces the assumption: it is said
that generics display this ‘surprising feature of exception-tolerance’, that a generic can
be true ‘despite the fact’ that members satisfying the predicate constitute a minority of
the population, that a theorist cannot help but wonder ‘How many exceptions can a
generic endure before being rendered false?’. To illustrate, consider
(12) Birds fly.
(13) All/Most birds fly. (adapted from Lazaridou-Chatzigoga, 2019)
Statement (12) is said to 'allow for exceptions' despite being true. This characteristic is
contrasted with the universally/majority quantified (13), which either does not allow
exceptions or allows a specific number of exceptions.
I take issue with this assumption. That is, I think the attempt to model generic sentences
as universal/majority quantifications in disguise is mistaken. I shall first substantiate my
claim by showing how obsession with this assumption gets both Cohen and Leslie into
various kinds of trouble. I’ll then offer some additional motivation.
Evidence that Cohen makes this assumption in coming up with his proposal is obvious.
His truth condition requiring a chance of greater than 0.5 is just majority quantification
cast in probabilistic terms. Another place where the effect of this assumption can be seen
is his formulation of the homogeneity constraint: it is stipulated his set of truth conditions
should hold with respect to every cell, despite the common intuition that ‘Chickens lay
eggs’ is true so long as one single cell (fertile, female chickens) fulfills the predicate (‘lay
eggs’). Leslie’s embrace of this assumption is equally obvious in her formulation of
‘majority generics’.
To show why such a theoretical commitment gets them into trouble, I present two
arguments: the ‘Runner-up challenge’ and the ‘multiple characteristic-existential
problem’.
4.2.1 The ‘Runner-up challenge’
In section 3.1 I demonstrated how Cohen’s relative reading apparently account for PRP
generics such as 11(b) ‘Japanese people are good table tennis players.’ Recall that the
truth conditions of (11)b are roughly that an arbitrary good Japanese sportsperson is
more likely than an arbitrary good ‘International’ sportsperson to be a good table tennis
player. Suppose an equal proportion of good sportspeople are good table tennis
players in both Japan and China, and China’s population of good table tennis players
keeps increasing until 99% of good ‘International’ sportspeople are Chinese table tennis
players. In such a situation, an arbitrary good Japanese sportsperson is much less likely
13
than an arbitrary good ‘International’ sportsperson to be a good table tennis player. So
according to Cohen, the generic would become false, contrary to people’s intuition.
4.2.2 The ‘multiple characteristic-existential problem’
Consider following sentences
(14) Roses are red.
It can be seen that such examples pose the same problem of the ‘runner-up challenge’
to Cohen’s account. If we choose not to invoke homogeneity, so long as non-red roses
outnumber red ones Cohen’s theory would predict (14) to be false under the absolute
reading. If we choose to invoke homogeneity, since colour is obviously a salient way of
partitioning, roses would be divided into red roses, black roses, white roses etc. Because
members of all cells of non-red roses consistently have 0 chance of being red,
homogeneity is not satisfied and (14) is still predicted to be false. Resorting to a relative
reading does not help either, since it is not necessarily the case that an arbitrary rose is
more likely to be red than an arbitrary flower. Even if that happens the case, we can
always think up the possible scenario where some other red flowers, say red tulips,
predominates just to nullify the relative reading.
Data like (14) also poses problem for Leslie’s account. First, it is not clear into which
category of generic should it fall. The colour of a flower seems to be a characteristic
feature, but since a rose can be of many mutually incompatible colours, does it mean a
natural kind can have multiple mutually incompatible characteristic features? If that is
the case, then why do children usually settle into one stable characteristic generic
statement for most natural kind items rather than positing multiple, equally legitimate
characteristic generic statements? Second, even if we grant that (14) is not a
characteristic generic statement, problems remain. Obviously (14) is not a striking-
property generic statement. That means it can only be considered a majority generic
statement, but it is not necessarily the case that most roses are red. Third, note that
exceptions to (14) are positive counterinstances by any account: any non-red rose must
have some colour, no rose can ‘simply fail to be red’. It seems that a commonly
acceptable generic is ruled by Leslie’s theory to be false through and through.
Similar comments can be said about (3)b. Note that both Cohen and Leslie cannot
account for the truth of (3)a unless they claim that the intra-sentential conjunction in
(3)a is actually sentential conjunction in disguise. Making that move, however, compels
them to apply the same treatment to (3b), in turn committing them to the truth of both
statements ‘Elephants live in Africa’ and ‘Elephants live in Asia’. Objections raised in the
last two paragraphs can then be applied against them.
4.3 A way out: existential quantification?
Section 4.2 demonstrates two points. First, the truth of certain generics can be totally
insensitive to both absolute prevalence level within a kind and relative proportion
14
across alternative kinds. This strongly argues against any universal/majority
quantificational approach. Second, certain generics behave as if they have a quasi-
universal quantificational flavour. In next section I offer more examples of this type to
motivate an alternative framework that starts from existential quantification.
5 The existence of existential generics
In this section I lay out a set of data called ‘existential generics’. It was first discovered by
von Fintel (1997) and Cohen (2003), further developed in Sterken (2014) and
acknowledged by Assarsson (2012), Hoeltje (2017) and Mcconnell-Ginet (2012). I hope
to show, by displaying a wide variety of evidence, that the existence of a quasi-existential
generic reading is robust and should be accounted for by any semantic theory of generics.
5.1 Emphatic Affirmation
When used as a refutation of a universal negative claim, a generic statement is
existentially interpreted (the notation [X]F indicates contrastive focus on linguistic
constituent X).
(15) A: Nobody in India eats beef.
B: That’s not true! Indians [do]F eat beef!
Here B’s utterance is intuitively true so long as some Indians eat beef. Moreover, those
beef-eating Indians do not have to be significant in number or any other way—it suffices
that they are Indians and they eat beef, even if there exists no relation between the
identity and the dietary practice. The generic statement has a weak, purely existential
force, which is not typical among examples that have featured in most discussions of
generics.
5.2 Emphatic Negation
When used as a refutation of a generic positive claim, a generic statement with negation
appears to be the negation of an existential assertion and tolerates no exception.
(16) Mammals [don’t]F lay eggs.
The existence of one species of mammal, say the platypus, is enough to render (2)
intuitively false. It need not be the case that mammals in general lay eggs.
Carlson (2008) had similar observations and concluded: ‘The truth-conditions for generics
involve an existential assertion; negation of that is the negation of an existential assertion’.
The inquiry seemed to be independently conducted, since he did not cite the two authors
before him.
5.3 Focus-sensitive Particles
15
First, just to demonstrate the general feasibility of embedding generics within focus-
sensitive particles, I offer some natural linguistic data:
Even birds have uses for tools. (In 'Revise GCSE: Biology Study Guide' by I.
Honeysett, p142, published in 2007 )
I believe this sentence above argues strongly for a 'generic existential' reading for two
reasons: first. it is common knowledge that birds do not in general use tools, only
certain sub-species such as ravens actually do use tools with some regularity (much like
the example of 'mammals lay eggs' and platypuses); second, it features in a GCSE
guidebook which should express generic generalizations by nature.
Now, observe that when embedded as the prejacent of some focus-sensitive particle, a
generic statement seems to be existentially interpreted.
(17) Even [mammals]F lay eggs
(18) Only [mammals]F bear live young
Again, the existence of platypuses is enough to render (17) acceptable. Since the
particle ‘even’ presupposes the truth of its adjacent, we reach the conclusion that
‘mammals lay eggs’ is true so long as some mammals do lay eggs—an existential
reading.
The case for (18) is slightly more complicated. First, some preliminaries about the
semantics of ‘only’: it is tempting to assume a simple analysis of (18) that takes "Only
birds lay eggs" to convey 'the existence of birds that lay eggs and the non-existence of
non-birds that lay eggs'. However, such a truth condition is too strong and does not
allow for contextual accommodation. Consider the sentence:
(19) Only North Korean school students sing their national anthem daily.
In a seminar about political system and education in which policies in various countries
is being compared with one another, an utterance of (19) is intuitively true. The hearer
considers school students of US, UK, China, compare their behaviors with that of North
Korean ones, and concludes that (19) is true. However, the simple analysis dictates the
non-existence of any non-North-Korean school student who sings their national
anthem daily. That is too strong. To my knowledge, Singaporean school students also
sing their national anthem daily.
One way to reconcile that piece of information with the intuition that (19) is true is to
say that (19) in some context does not 'touch on' Singaporean school students. To do
that the negation part of its meaning has to be more fine-tuned than simply 'non-
North-Korean school student'. This is where the concept of 'alternative set' comes in.
16
Recent literature on focus sensitive particles arrives at the consensus that the defining
feature of focus is to evoke alternative expressions that can replace the element in
focus. The focused element and the alternatives together form an 'alternative set'. The
construction of this 'alternative set' is online and contextually guided. In the case of (19),
the alternative set could be {US school students; UK school students; Chinese school
students......}. Each alternative can replace the focused element to form a new,
alternative statement. What the negation part of the semantics of 'only' does, then, is to
deny all those alternative statements. By saying that some less-known cases do not
feature in the alternative set, we achieve contextual accommodation.
Now we can spell out the exact semantics of the particle ‘only’: on top of presupposing
the truth of its prejacent, 'only' also demands the falsity of all the contextually relevant
alternatives of its prejacent. To illustrate, suppose 'John' activates the set of his friends
{Mary, Tom......}
(20) a. Only John owns guns.
b. ✓John owns guns.
c. X Mary owns guns.
d. X Tom owns guns.
......
For (20)a to be true, it must be the case that John owns guns and it must not be the
case that Mary owns guns or Tom owns guns or......So (20)a presupposes the truth of
(20)b and demands, for its truth, the falsity of all of (20)c, (20)d,......
I now apply this framework to the interpretation of generic statements. Strictly speaking,
what is 'embedded' is just the bare nominal without the predicate. But that does not
leave the predicate out of the picture, since every time the focused element (the bare
plural) gets replaced by an alternative BP, the predicate still has to combine with that
alternative BP to form an alternative statement to be fed to the ultimate meaning
computation for the original statement. Given the supposition that the prejacent part of
the original statement is interpreted generically (which has been independently
motivated), it is fair to assume that all the alternative statements are also interpreted
generically. To illustrate, consider (21)a:
(21) a. ?Only Mammals give birth to live young.
b. ✓Mammals give birth to live young.
c. X Birds give birth to live young.
d. X Reptiles give birth to live young.
......
Similarly, (21)a presupposes the truth of (21)b and its truth demands the falsity of all of
(21)c, (21)d,......Note that (21)b is true regardless of whether we take an existential
17
approach or a quasi-universal approach to generics since mammals indeed generally
give birth to live young. The presupposition of 'only' is always satisfied.
What about its second presupposition? What are the actual truth values of (21)c and
(21)d? If we were to assume the received view that generic statements have a quasi-
universal flavour by default, (21)c, (21)d...... are all false since no animals other than
mammals generally bear live young. Since both presuppositions of 'only' hold, it is thus
predicted that (21)a should then be acceptable. But this is the wrong prediction! People
judge (21)a to be intuitively unacceptable:
(22) A: Only Mammals give birth to live young.
B: That's not true! Some reptiles do that too.
So the prediction is not born out: the existence of merely some reptiles which bear live
young is enough to render (21)a false. This observation suggests something is amiss in
our analysis. The fault cannot lie in the presupposition of 'only', since it holds regardless.
The only option is to say some of the truth conditions demanded by 'only' fail to hold.
How exactly? A natural move would be to claim that statements like (21)c and (21)d
receive an existential interpretation instead, for that is the only reading the truth of
which require only some birds/reptiles to bear live young. The truth conditions should
be better reflected as in (21')a-d:
(21')a. ?Only Mammals give birth to live young.
b. ✓Mammals give birth to live young.
c. X Birds give birth to live young.
d. ✓ Reptiles give birth to live young.
......
Since the truth of (21')a demands the falsity of (21')d while (21')d is true, a failure to
fulfill the truth conditions explains the unacceptability of (21')a.
2.4 Additives
Note that with an additive a generic statement has an existential reading when following
a relevant positive counterpart. For example, the acceptability of (6)b or (6)c when uttered
right after (6)a only requires the existence of some mammals that lay eggs.
(23)a. Birds lay eggs.
b. Mammals also lay eggs.
c. Mammals lay eggs too.
The existential reading of generics requires an explanation. I have already argued that a
uniform account is superior than an ambiguity account. In pursuit of uniformity one has
to treat one reading as basic and the other as derivative. Cohen (2003) took the second
approach, arguing that the derivation of a quasi-existential from a quasi-universal
reading is mediated by the introduction (or not) of alternatives by contrastive focus or
18
focus-sensitive particles. However, Sterken (2014) objected that quasi-existential
readings can be grasped even in the absence of focus-sensitive particles or any other
formal features of information structure. I agree with her verdict but due to space reasons
take it for granted without offering more examples. I hope my arguments in section 4
and demonstrations in section 7 can convince my readers of the primacy of the existential
interpretation.
6 An ‘alternative’ theory of generics
In this section I aim to synthesize the two theories reviewed in section 4 to offer a new
account that combines their strengths while avoiding their shortcomings. I follow them
in making use of the following concepts: contextually induced alternative sets; subject
alternative sets constrained by predicate alternative sets; the distinction between positive
and negative alternatives; psychologically salient partitioning of a kind. I combine the
‘characteristic property’ and ‘striking property’ in Leslie’s classification into a single
‘distinguishing feature’ cashed out in terms of the contrast between two subject
alternatives in their capability to fulfil a predicate.
I explicitly abandon the following theoretical commitments: probability of greater than
0.5; universal/majority quantification; structural ambiguity, and replace them with a
uniform existential quantification with a twist (the distinguishing feature).
For a bare plural generic K (P), where K is the subject kind term nominal and (P) the
predicate:
(P) activates some contextually alternative predicates: (P1), (P2)…(Pn), which together with (P)
form a set Alt(P);
K activates some contextually alternative subjects: K1, K2…Kn such that for x=1,2…n it is
always the case that at least one member of Kx fulfils at least one member of Alt(P);
There is a salient way of partitioning K into an ad hoc sub-kind K* such that every member of
K* fulfils at least one member of Alt(P); AND
Among the members of K* which (P), their fulfilling (P) can be explained by their being K.
(There exists a Km, where m is a number between 1 and n and a change of identity from K* to
Km, ceteris paribus, guarantees that the members of K* which (P) no longer fulfil (P))
This set of truth conditions is accompanied by a heuristic guiding alternative formation:
Prohibition against negative alternatives: a contextually induced alternative to (P) cannot be
(¬P).
The reason for this is simple. If (¬P) is an alternative to (P), since both of them are
members of the set Alt(P), this means the formation of the subject alternative set would
be unselective: anything in the world can be included because it either satisfies (P) or (¬P).
This results in trivial computations that do not limit semantic space in any meaningful way.
19
Out of efficiency considerations, the prohibition is motivated.
7 The ‘Alternative’ theory put to work
This section demonstrates the workings of my theory with some examples. To start, I show
that it accounts for (24)-(25) and thus can handle the ‘chicken failure’.
(24) Chickens lay eggs
Analysis: ‘Lay eggs’ activates some contextually alternative predicates: ‘bear live young’,
‘undergo mitosis’…, which together with ‘lay eggs’ form a set ‘Alt(lay eggs)’
[Here the contextually relevant consideration is ways of reproduction];
‘Chickens’ activates some contextually alternative subjects: ‘ducks’, ‘cows’…
[Each of which has members which have some way of reproduction];
There is a salient way of partitioning ‘chickens’ into an ad hoc sub-kind ‘chickens*’ such
that every member of ‘chickens*’ fulfils at least one member of Alt(lay eggs)
[✓ e.g. living, fertile, female chickens without genetic defects]; AND
Among the members of ‘chickens*’ which do lay eggs, their egg-laying can be explained
by their being chickens (e.g. instead of cows)
[✓ This explanation is intuitively reasonable since not a single cow reproduces by egg-
laying. Had they been cows instead, they would not lay eggs]
Therefore, ’chickens lay eggs’ is acceptable as a generic.
(25)? Chickens are female
Analysis: ‘Are female’ activates a contextually alternative predicate: ‘are male’, which
together with ‘are female’ form a set ‘Alt(are female)’
[Here the contextually relevant consideration is biological sex]
‘Chickens’ activates some contextually alternative subjects: ‘ducks’, ‘cows’…,
[Each of which has members which have some biological sex]
There is a salient way of partitioning ‘chickens’ into an ad hoc sub-kind ‘chickens*’ such
that every member of ‘chickens*’ fulfils at least one member of Alt(are female)
[?✓ Any arbitrary way of partitioning will do since every chicken has a sex]; HOWEVER
Among the members of ‘chickens*’ which are female, their being female cannot be
explained by their being chickens (e.g. instead of ducks/cows etc.)
[X This explanation is intuitively unreasonable since they could still be female had they
been cows instead (or any other animal for that matter), i.e. there exists no alternative
kind a mere identity change to which guarantees they would not be female]
Therefore, ’chickens are female’ is unacceptable as a generic.
I proceed to show that this theory can equally handle the PRP examples such as (26).
(26)Japanese people are good table tennis players
20
‘Good’ is a vague predicate, so to simplify the analysis I make a number of assumptions:
1. ‘Are good table tennis players’ means ‘have table tennis skills above international
average’.
2. International average is the arithmetic average of the national averages.
3. National average is the arithmetic average of the skill levels of all nationals of a country
with non-zero skill level.
4. ‘Are good table tennis players’ is a property of a group rather than individuals. This is
supported by the unacceptability of the indefinite counterpart of (25) ‘A Japanese person
is a good table tennis player.’
Analysis: ‘are good table tennis players’ activates some contextually alternative predicates:
‘are good football players, are good baseball players…’ which together with ‘are good
table tennis players’ form a set Alt(are good table tennis players)
[Here the contextually relevant consideration is being good at some sport];
‘Japanese people’ activates some contextually alternative subjects: ‘Chinese people’,
‘American people’……
[Each of which has members which are good at some sports]
There is a salient way of partitioning ‘Japanese people’ into an ad hoc sub-kind ‘Japanese
people*’ such that every member of ‘Japanese people*’ fulfils at least one member of
Alt(are good table tennis players)
[✓ e.g. Japanese people who are good at some sports]; AND
Among the members of ‘Japanese people*’ which are good table tennis players, their
being good table tennis players can be explained by their being Japanese [instead of e.g.
North Korean]
[✓ This explanation is intuitively reasonable since by definition at least one country’s
national average will be lower or equal to the international average. Had those Japanese
people been nationals of that country instead, they are guaranteed not to be good table
tennis players]
Since the existential flavor is built into my truth conditions, I believe that my account’s
ability to explain (3)a-b cases can be easily appreciated. I omit further illustrating
examples.
8 Conclusion and implications
In this essay I have introduced some main theoretical puzzles surrounding the study of
generics, reviewed Cohen’s probability theory within formal semantic camp alongside
Leslie’s cognitive-default theory within psychology, and made an attempt to synthesize
the two. I suggested that a paradigm shift from quasi-universal quantification to quasi-
existential quantification as default has theoretical advantages to be gained and backed
my claim by revisiting and reorganizing neglected linguistic data. I finally offered an
alternative, eclectic theory of generics. The limitations of my account are many: I leave
out generics’ modal/normative/functional readings because I believe they should be
treated as conversational implicatures. I also cannot explore how my account can be
extended to cover other forms of generic sentences, or to bridge the gap between
21
characterizing sentences and direct kind predications. Nevertheless, I conclude that the
novel approach of combining the existential-quantification-as-default view with a
uniform alternative semantics offers new insight into the study of generics.
(7990 words)
References
Assarsson, E. (2012). A Two-Tiered Theory of Generics: A Synthesis of Leslie’s and
Liebesman’s Theories. Ph. D. Thesis. Lund University.
Cohen, A. (1996). Think generic! The meaning and use of generic sentences. Ph. D.
Dissertation. Carnegie Mellon University.
Cohen, A. (2003). Existential Generics. Linguistics and Philosophy, 27(2), 137-168. doi:
10.1023/b:ling.0000016441.89129.3d
Gelman, S. A. (2003). The essential child: Origins of essentialism in everyday thought.
Oxford University Press, USA.
Cohen, A. (2004). Generics and mental representations. Linguistics and Philosophy, 27(5),
529-556. doi: 10.1023/B:LING.0000033851.25870.3e
von Fintel, K. (1997). Bare plurals, bare conditionals, and only. Journal of Semantics, 14(1),
1-56. doi: 10.1093/jos/14.1.1
Hoeltje, M. (2017). Generics and ways of being normal. Linguistics and Philosophy, 40(2),
101-118. doi: 10.1007/s10988-016-9203-6
Honeysett, I. (2007). Revise GCSE: Biology Study Guide (p. 142). London: Letts and
Lonsdale.
Krifka, Manfred, et al. (1995). Genericity: An Introduction. In: G.N. Carlson and F.J. Pelletier
(eds.), The Generic Book, 1-124. Chicago and London: University of Chicago Press.
Leslie, S. J. (2007). Generics, cognition, and comprehension. Doctoral dissertation. Princeton
University.
Lazaridou-Chatzigoga, D. (2019). Genericity. In C. Cummins & N. Katsos. (Eds.), The
Oxford handbook of experimental semantics and pragmatics (pp. 156–177). Oxford:
Oxford University Press.
Leslie, S. J. (2008). Generics: Cognition and acquisition. Philosophical Review, 117(1), 1-47.
doi: doi.org/10.1215/00318108-2007-023
Mcconnell-Ginet, S. (2012). Generic predicates and interest-relativity. Canadian Journal Of
Linguistics/Revue Canadienne De Linguistique, 57(2), 261-287. doi:
10.1017/s0008413100004771
22
Nickel, B. (2016). Between logic and the world: An integrated theory of generics. Oxford
University Press.
Roeper, T., Strauss, U., & Pearson, B. Z. (2006). The acquisition path of the determiner
quantifier every: Two kinds of spreading. Papers in Language Acquisition, University of
Massachusetts Occasional Papers UMOP, 34.
Sterken, R. (2014). Generics in context : generalisation, context and communication. Ph. D.
Thesis. The University of Oslo and the University of St Andrews.