Krifka et al. (1995)

In the semantic literature, the special lawlikeness associated with generic sentences is modelled by assuming a covert generic operator, Gen, which, among other things, contributes an intensional meaning component.³⁰ The consensus in the literature is that Gen is a relational operator, similar to a covert adverb of quantification, which expresses a relation between its restrictor and its scope.³¹

Krifka et al. (1995) introduce the general structure for the dyadic generic operator Gen in(39).

(39) Gen[x₁, . . . , x_i;y₁, . . . y_j](Restrictor; Scope) (Krifka et al. 1995:26)

The variablesx₁, . . . , x_i, y₁, . . . y_j in the bracket after the generic operatorGen are the list of all free variables that occur in the restrictor and the scope of the generic operator.

By convention, the variables x₁, . . . , x_i in front of the semicolon are interpreted as

30An alternative formalization, which is based on treatments of rules in the literature on artificial intelligence, models generic sentences as expressing defeasible inferences in default logics. Cf.Asher and Pelletier(1996). For a modal account of generic sentences which replicates the properties of these extensional systems seeEckardt(2000).

31Carlson (1977) analyzes all types of generic sentences as predication over kinds. For reasons of space, this alternative proposal will not be discussed. For a detailed discussion and comparison with the modal relational analysis, seeCondoravdi(1994).

bound by Gen. The variables y₁, . . . y_j, on the other hand, occur only in the scope of the operator, and are interpreted as existentially bound.

In the semantic literature on generic sentences, indefinite singular noun phrases and bare plural noun phrases are assumed to be Heimian indefinites, i.e. expressions that contribute free variables which are either bound by appropriate operators or via ex-istential closure. Hence, the definition in (39) has to be read as follows: only in case the variable contributed by an indefinite expression is bound by the generic operator, the descriptive content of the indefinite is interpreted in the restrictor of the operator.

Descriptive material belonging to individual variables that are bound by the existen-tial quantifier is interpreted in the scope ofGen. The same distribution also applies to situation variables and any descriptive content depending on them.

The distribution of variables across the restrictor and the scope captures the behavior of indefinite expressions in generic sentences. Consider(40-a) and its formalization in (40-b).

(40) a. Professors usually have a cup of tea after lunch.

b. Gen[x;y](professor(x); cup-of-tea(y) & has(y)(x))

Since the variable x that is contributed by the bare plural is bound by the generic operator, the bare plural noun phrase professors is interpreted in the restrictor of Gen. Consequently, (40-a) expresses a regularity about professors. The variable y contributed by a cup of tea, on the other hand, is existentially bound in the scope of Gen. As a result, professors in(40-a)is interpreted as generically quantified anda cup of tea as existentially quantified. Which variables are bound byGen and which are not, is determined by sentential stress, topicality, and various other factors. The default case for individual variables in indefinite singular and bare plural generic sentences is that Gen binds only the variable contributed by the indefinite noun phrase in subject position; all other nominal and/or verbal material is interpreted in the scope of Gen (Krifka et al. 1995:26).³²

The partitioning into a restrictor and a scope also captures the intuition that generic sentences express a lawlike relation between two states of affairs.

The structure in(39) as such does not make any claims regarding the interpretation of the generic operator, though. The intuition shared in the literature is that the

32The following sentence constitutes a counterexample to this default split:

(i) Typhoons arise in this part of the pacific.

(Krifka et al. 1995:26)

The most natural reading of the example above is that it expresses a regularity for a certain part of the pacific: regularly some typhoons arise there. In this case, situations involving the part of the pacific under discussion form the restrictor, and the bare plural in subject position,typhoons, is part of the scope ofGen. The other, less accessible, reading is given by the default restrictor-scope arrangement:

‘Typhoons in general arise in this specific spot of the pacific ocean’. (Krifka et al. 1995:26).

generic operator contributes (quasi-)universal quantification over individuals and/or situations of a certain kind that may or may not be part of the world of evaluation.

This is illustrated in the examples in (41) (repeated from above).

(41) a. Lions have four legs.

b. A lion has four legs.

Intuitively, the generic sentences in(41)state that all relevantly normal lions (i.e. those that are not legitimate exceptions) have four legs. How can this intuition be captured?

The dyadic operator account introduced in (39) is in principle compatible with many different formal analyses of the interpretation of Gen. In fact—apart from modal ac-counts, which are discussed below—various different interpretations forGen have been proposed in the literature: e.g. (quasi-)universal quantification over relevant/possible entities, prototypes, or stereotypes. A detailed discussion of these accounts is beyond the scope of this dissertation; for a summary see Krifka et al.(1995).

Modal accounts for the meaning ofGenare motivated by the non-accidental nature of the generalizations, as well as their tolerance for the existence of exceptional individuals in the actual world. In these accounts, Gen is interpreted as a modal operator in analogy to Kratzer’s account for the meaning of modals (cf. Section 3.2). The various proposals differ with respect to the set of accessible worlds associated with Gen, and the degree of influence that is accorded to the properties denoted by the subject noun phrase and the predicate on the make-up of the accessibility relation (cf. Krifka et al.

1995; Papafragou 1996; Drewery 1998;Eckardt 2000; Greenberg 2007).

As stated above, the first modal account that will be discussed in detail is the pro-posal inKrifka et al.(1995), which is based on Heim’s (1988 [1982]) account for “generic indefinites”. Heim proposes that indefinite noun phrases in generic sentences act like if-clauses in conditional sentences. That is, they restrict a covert necessity modal (cf.

Appendix A3). The covert modal involved in generic sentences differs from the covert epistemic necessity modal found with indicative conditional sentences, though, which is assumed to only take realistic modal bases and ordering sources. Since the covert modal in generic sentences expresses lawlike regularities, Heim proposes that it is in-terpreted with respect to non-realistic modal bases and stereotypical ordering sources.

Krifka et al. (1995) flesh out Heim’s proposal as follows.

(42) Gen[x₁, . . . , x_i;y₁, . . . , y_j](Restrictor;Matrix) is true in w relative to a modal base B_w and an ordering≤_w iff:

∀x₁, . . . , x_i∀w⁰ ∈B_w[Restrictor[x₁, . . . , x_i](w⁰)→

∃w⁰⁰∈B_w[w⁰⁰≤_w w⁰ & ∀w⁰⁰⁰[w⁰⁰⁰ ≤_w w⁰⁰ →

∃y_i, . . . , y_j[Matrix[x₁, . . . , x_i, y₁, . . . , y_j](w⁰⁰⁰)]]]]

(adapted fromKrifka et al. 1995:52)

The specific modal flavor of the generic operator is such that the property contributed by the material in the restrictor is evaluated against the worlds in the modal baseB_w, which are “most normal” with respect to the ordering≤_w.

The truth-conditions of the generic sentence in (43-a), for example, come out as in (43-b).

(43) a. Lions have bushy tails.

b. ∀x∀w⁰ ∈ B_w[lion(x)(w⁰) → ∃w⁰⁰ ∈ B_w[w⁰⁰ ≤_w w⁰ & ∀w⁰⁰⁰[w⁰⁰⁰ ≤_w w⁰⁰ →

∃y[bushy-tail(y)(w⁰⁰⁰) & has(y)(x)(w⁰⁰⁰)]]]]

(Krifka et al. 1995:52)

The formula states that “a world which contains a lion without a bushy tail is less normal than a world in which that lion has a bushy tail” (Krifka et al. 1995:52), i.e.

in the most normal worlds from the point of view of the world of evaluation, all lions have bushy tails.

If the limit assumption is made, the account proposed by Krifka et al. can be stated as in (44).

(44) ∀x∀w⁰ ∈O(f ∩ {w⁰⁰: lion(x)(w⁰⁰)},g, w)[has-a-bushy-tail(x)(w⁰)]

Krifka et al. argue that Heim’s analysis of the modal flavor of Gen and of the covert necessity modal in conditionals is too restrictive. They argue that Gen, as well as the covert necessity modal involve many different combinations of modal bases and ordering sources. For generic sentences in particular, different “most normal” worlds are required to capture the different types of generic statements found in the data.

Consider the different possible readings forGen in(45).

(45) a. Two and two equals four. (tautology)

b. A spinster is an old, never-married woman. (definition)

c. This machine crushes oranges. (design)

d. Mary smokes cigarettes. (behavior)

e. Bob jumps 8.90 meters. (ability)

f. A lion has a mane. (stereotype)

(Krifka et al. 1995:53f)

For the examples in (45), the set of accessible worlds vary among most normal with respect to e.g. mathematical laws, the interpretation of English, and the specific design of an artifact. Hence, the parameters that model the modal base and the ordering source seem to allow a similar array of possible values as for modals (cf. Section3.2).

The account in Krifka et al. (1995) successfully captures that generic sentences ex-press non-accidental regularities, and that they tolerate exceptions. Since Krifka et al.

assign a modal interpretation to generic sentences, not only actual facts and situations

are considered when a generic sentence is evaluated. As discussed above, generaliza-tions that only hold for the world of evaluation may be accidental, and therefore do not necessarily depend on an underlying regularity. By evaluating the generic sentence against a specific set of possible worlds, it is ensured that accidental generalizations that are not based on an underlying regularity come out as false.

The observed tolerance for exceptions found with generic sentences is also captured.

In case the accessibility relation determined for a given generic sentence is a non-realistic relation (e.g. ideal-based, see Section 3.2), the world of evaluation is not a member of the set of most normal worlds.³³ Consequently, the facts in the world of evaluation are not considered when the validity of the universal statement about individuals in the accessible worlds is evaluated; individuals in the world of evaluation that do not conform to the regularity that is expressed do not have an effect on the truth or falsity of the generic sentence.

Krifka et al.’s (1995) account is not without shortcomings, though. One issue is already pointed out in Krifka et al. (1995): defining the accessible worlds as the “most normal” worlds seems counter-intuitive for some generic sentences. Krifka et al. observe that the “normalcy” that is required for the truth of these sentences seems highly abnormal from the point of view of the world of evaluation. Consider (46).

(46) Turtles die old.

Example(46)is a true statement about turtles given what is known about their biology.

For this generic sentence to come out as true, though, the most normal worlds with respect to considerations of biology need to be those where all turtles die old. However, these worlds are highly abnormal from the point of view of the actual world. In fact, most turtles die young because they are killed by predators.³⁴

Example (47) illustrates another problematic aspect for the accessibility relation assumed in Krifka et al.

(47) Cats bear live young.

Since (47) expresses a true generalization about cats, the account predicts that in the biologically most normal worlds all cats bear live young. This means that the biologically most normal worlds are those where only female cats of the right age exist.

However, worlds like these are highly abnormal, if not impossible: how can the existing female cats bear any young, if no male cats exist?³⁵ One possibility to account for

33An accessibility relationRis realistic iff∀w∈W[wRw], i.e. ifRis reflexive. Accessibility relations that are not realistic are said to be non-realistic.

34Similar observations are made e.g. inPapafragou(1996),Eckardt(2000) andGreenberg (2007).

35Krifka et al.(1995) propose that this problem can be solved by assuming that the generic operator quantifies over “most normal” situations instead of most normal worlds. For reasons of space, the situation theoretic account cannot be introduced at this point. For details, cf. Krifka et al.(1995).

(47) in Krifka et al.’s account is to argue that the noun phrase in subject position is contextually restricted to female cats. The following two reasons speak against this argument. First, in contrast to nominal quantifiers, generic sentences do not allow for contextual restrictions (cf. Dahl 1975; Condoravdi 1994; Krifka et al. 1995, pace Drewery 1998;Greenberg 2007). Compare (48) and (49).

(48) a. Uttered at UCLA: All professors wear a tie.

b. At UCLA, all professors wear a tie.

(Krifka et al. 1995:45)

(49) a. Uttered at UCLA: A professor wears a tie.

b. At UCLA, a professor wears a tie.

(Krifka et al. 1995:45)

Nominal quantifiers, like all in (48-a), may be understood as contextually restricted, i.e.(48-a)may be understood as(48-b). For generic sentences, contextual restriction is impossible. Even if(49-a)is uttered at the UCLA campus, Krifka et al. argue, it cannot be understood as expressing(49-b), which contains an explicit spatial expression.

Even if one were to assume that contextual restriction is an option—as argued for in Drewery (1998) and Greenberg (2007)—no adequate restriction for (47) could be found. Assume for example, that cats is contextually restricted to female cats of the right age. This restriction is not strong enough. It does not exclude e.g. female cats of the right age that have a birth defect and are infertile, or female cats of the right age that are kept apart from male cats and never bear any young. In fact, many different reasons can be found for why female cats of the right age might not bear live young. All of the individuals that have at least one of the properties that make them exceptions will have to be excluded by the contextual restriction for Gen in (47). However, the disjunction of all of these properties seems to be a highly unnatural “salient property” to be assigned contextually. In other words, while a contextual restriction ofall professors toall professors at UCLAseems plausible enough, the restriction ofcats tofemale cats of the right age that neither have a birth defect, nor are kept apart from male cats, nor . . . does not.

The second argument against appealing to contextual restrictions is that, for some sentences, incompatible restrictive properties would have to be assumed for the subject noun phrase. Consider(50).

(50) Peacocks lay eggs and have beautiful feathers.

Since only female peacocks lay eggs and only male peacocks have the characteristic, beautiful feathers, the subject noun phrase peacocks would have to be restricted to individuals that are both male and female. Hence, the assumption thatGen is

contex-tually restricted cannot solve the problems that are raised by the accessibility relation assumed in Krifka et al.(1995).

Two further issues for Krifka et al.’s (1995) modal account arise because the set of accessible worlds B_w does not depend on the material in the restrictor of Gen.

As observed for generic sentences with impersonally interpretedich anddu in Chap-ter 2, all individuals in the accessible worlds have to behave according to what is most normal. This is the case because the set of worlds B_w is determined entirely contextu-ally, and depends neither on the material in the restrictor, nor on the material in the scope of Gen. Consider the example in (51).

(51) Dogs chase cats.

Since the sentence above expresses a regularity regarding the behavior of animals, it is plausible to assume that only such worlds are considered that are most normal with respect to biological laws (or maybe even specifically with respect to the behavior of animals). This predicts that both the dogs and the cats show “most normal behavior” in these worlds. Intuitively, however, only the individuals that are picked by the material in the restrictor have to be relevantly normal or ideal. For (51), this means that only the most normal behavior of dogs is relevant. Nothing seems to be required regarding the behavior of the cats in these worlds.

The lack of dependence between Gen and the specific form of the subject property leads to another problem (cf. Drewery 1998): the addition of an intersective modifier to the subject property results in simple intersective modification of the material in the restrictor. Compare the examples in (52).

(52) a. Birds fly.

b. Baby birds fly.

If modifiers have the simple effect of restricting the domain of individuals—given the same subject matter—it is predicted that the same set of worlds is picked for both sentences in (52). The truth-conditions of the two sentences are given in(52).

(53) a. ∀x∀w⁰ ∈O(f ∩ {w⁰⁰ : birds(x)(w⁰⁰)}, g, w)[fly(x)(w⁰)]

b. ∀x∀w⁰ ∈O(f ∩ {w⁰⁰ : baby-birds(x)(w⁰⁰)}, g, w)[fly(x)(w⁰)]

Since universal quantifiers are left-downward-monotone, and the set of baby birds is a subset of the set of birds, the truth of (53-b) follows from the truth of (53-a). This is undesirable since, of course, baby birds do not fly, yet.

The final issue that is discussed for Krifka et al. depends on the assumption of strict universal quantification over individuals in the accessible worlds. As the discussion of example(47)above suggests, this assumption is too strong. The quantification needs to be qualified. For the truth of(47)according to Krifka et al.’s account, only female cats

of a certain age may be considered, but worlds in which only these kinds of cats exist may not contain any cats that actually bear live young.³⁶ To ensure that sentences like (47)can be interpreted, quantification over individuals in the accessible worlds cannot be strictly universal. Since contextual restriction of the domain of quantification has been shown to be implausible, the restriction has to be part of the meaning of Gen:

the modal interpretation of generic sentences needs to involve a special type of in-built restriction which picks out the relevantly non-exceptional individuals in the accessible worlds.

The following consideration also supports this assumption. Tolerance for exceptions is captured in Krifka et al. (1995) with the assumption of a non-realistic accessibility relation. Since the actual world is consequently not necessarily a member of the set of accessible worlds, the properties and behavior of actual individuals may turn out to be irrelevant. However, it seems counter-intuitive to assume that the accessibility relation involved with generic sentences is necessarily non-realistic given that generic sentences may express regularities that are observed in the world of evaluation. A similar point is made by Dahl (1975), who argues that physical and biological rules require a realistic accessibility relation. If generic sentences were to involve realistic accessibility relations, though, Krifka et al.’s account would require that the generalizations that are expressed be exceptionless. Even though Dahl suggests that physical and biological laws fall into this class, this is obviously not always the case, as exemplified by the scenario featuring the three-legged lion Paul in Section 3.3.1.

In sum, the modal account in Krifka et al.(1995) successfully captures that generic sentences express non-accidental generalizations, and provides an account for their tol-erance of exceptions. The account is not without problems, though. The problematic aspects identified above point toward two desiderata for modal proposals for the mean-ing of Gen:

(54) a. The set of accessible worlds needs to depend on (at least) the subject property.

b. Universal quantification over individuals needs to be restricted by a generi-city-specific, inbuilt restriction.

To meet the second desideratum, Papafragou (1996) and Eckardt (2000) argue for the introduction of a normalcy predicate that restricts the domain of quantification of Gen to those individuals in the accessible worlds that are relevantly normal or ideal.³⁷

36The only worlds that fulfill the criteria are those in which cats are nearly extinct, only pregnant cats exist, and the last male cat died recently. These worlds, however, do not count as normal in any sense of the word.

37Eckardt (2000) introduces two different operators which filter out the exceptional individuals in the accessible worlds depending on the type of generalization that is expressed. That is, generalizations about normal or ideal course of events are distinguished. Eckardt introduces an operatorN(F) which returns the set of “normal” individuals for the subject propertyF, and an operatorI(F), which returns

With respect to the accessibility relation associated with Gen, they keep the basic

With respect to the accessibility relation associated with Gen, they keep the basic

Im Dokument Impersonally Interpreted Personal Pronouns (Seite 182-190)