Page 1
1
An overt syntactic marker for genericity in Hebrew1
Yael Greenberg, Bar Ilan University [email protected]
(Pre-Final version, Appeared in 1998 in in S. Rothstein (ed.) Events and Grammar, Kluwer Academic
Publishers)
In this paper I argue for the existence of an overt syntactic marker for genericity in Hebrew
nominal predicative sentences, which will be called Pron (following Doron’s (1983)
terminology), and discuss the implications of this claim for a number of topics dealt with in
current theories of genericity. I begin with a background about the distribution of Pron, and
make the initial claim about the connection between Pron and genericity. In section 3 and 4 I
examine in detail two possible distributions of Pron, and claim that characterizing the generic
/nongeneric distinction they create should be done in terms of both quantificational and kind
referring genericity. In section 5 I reject an analysis of the data in terms of the stage / individual
distinction. Finally, I discuss the compatibility of three Syntax-Semantics interface theories
(Diesing 1992, Chierchia 1992 and Ramchand 1996) with the Hebrew data.
1. The distribution of Pron, and the Predicative/Identity distinction approach.
Present tense nominal (copular) sentences in Hebrew show some distinctive syntactic
properties. First, while past and future tense copular sentences pattern like their English (and
other Germanic) counterparts in having a verbal copula, the copula of the present tense ones in
Hebrew has the form of a nominative third person pronoun. Examples of the verbal and
“pronominal” copula are underlined in (1) and (2), respectively:
(1) a. dani haya xaxam b. dani yihye xaxam
‘Danny was wise.’ ‘Danny will be wise’
Page 2
2
(2) a. dani hu gavoha b. rina hi xaxama c. ha-’etim Seli hem kxulim
Danny 3ms.sg. tall Rina 3fm.sg. smart the pens mine 3ms.pl. blue
‘Danny is tall.’ ‘Rina is smart.’ ‘My pens are blue.’
Doron (1983) shows convincingly that the pronominal copula (“Pron”, in her terminology
and henceforth) lacks any verbal properties, and is a realization of agreement features only,
which is located in INFL, the head of IP. This claim is followed in Rapoport 1987 and Rothstein
1995, and I will assume it in this paper.
Another special property of these nominal sentences in Hebrew is that, unlike the verbal
copula, the (pronominal) copula, i.e. Pron, can sometimes be deleted from the sentence, without
causing any ungrammaticality. Two examples are seen in (3):
(3) a. dani xaxam b. ‘ima Seli mora
Danny wise = ‘Danny is wise.’ mother mine teacher = ‘My mother is a teacher.’
However, as noticed by Doron (1983), the absence of Pron is not possible in all kinds of
sentences. For example, Pron is obligatorily present in sentences like (4):
(4) ha-’iSa ha-zot *(hi) rina
the woman the this 3fm.sg. Rina = ‘This woman is Rina.’
Doron (1983), followed by Rapoport (1987) and Rothstein (1995) claimed that Pron is
obligatory in identity sentences, as in (4), where the post copular element is an argument, and
that this is due to theta assignment reasons (according to Doron), in order to prevent a Case-
Filter violation (according to Rapoport), or in order for a predication relation to hold (according
to Rothstein). In predicative sentences, on the other hand, as in (3), the predicate can fill the
theta assignment, Case assignment or predication relation requirements directly, so Pron is
optional.
Page 3
3
This kind of approach seems to account well for the fact that Pron must appear in identity
sentences. However, with respect to predicative sentences, it has several explicit and implicit
predictions, which are not borne out in reality. First, there exist sentences where Pron must
appear (oblig.+Pron sentences), which are clearly predicative. Two examples are seen in (5)2:
(5 ) a. ‘orvim *(hem) (yecurim) Sxorim b. pariz ??(hi) be-carfat
ravens 3ms.pl. creatures black Paris 3fm.sg. in France
‘Ravens are black (creatures).’ ‘Paris is in France.’
Second, the predicative/identity approach cannot predict the existence of sentences as in (6),
where Pron is obligatorily absent (oblig.-Pron sentences, henceforth):
(6) a. dani *(*hu) ‘ayef ‘axSav b. rina (*hi) yafa ha-’erev
Danny 3ms.sg. tired now Rina 3fm.sg. pretty the night
‘Danny is tired now.’ ‘Rina is pretty tonight.’
Third, the predicative/identity approach crucially relies on the post copular element as
determining the distribution of Pron. But, as seen from (7), there are minimal pairs which show
that in some cases, it is the pre-copular element which makes the difference:
(7) a. zmaxim *(hem) yerukim b. ha-zmaxim ha-elu (hem) yerukim
plants 3ms.pl. green the plants the these 3ms.pl. green
‘Plants are green.’ ‘These plants are green.’
Finally, this approach cannot predict the meaning difference, noted by Bendavid 1971, and
indicated in the glosses, between the +Pron and the -Pron versions of a sentence like (8):
(8) a. ha-Samayim hem kxulim b. ha-Samayim kxulim
the sky 3ms.pl. blue the sky blue
‘The sky is generally blue, blue by their nature. ‘The sky is blue now/today.’
Page 4
4
The predicative/identity approach can account, then, for the obligatory presence of Pron in
identity sentences, but cannot handle the varied distribution of Pron in predicative sentences. Is
there any pattern governing the presence/absence of Pron in such sentences too ?
In this paper I claim that there is. I suggest that there is a semantic distinction which
correlates with Pron in predicative sentences, which we can characterize, in a very intuitive and
informal manner, as a distinction between permanent or general statements or those expressing
some kind of ‘definitions’ as opposed to episodic or temporary ones, which report accidental
facts. Looking at the sentences in (5) - (8) above shows that a semantic distinction along these
lines plays some role in the distribution of Pron. The oblig.+Pron sentences (as in (5)) express
the former type of statements, whereas the oblig.-Pron in (6) express the latter type. This way of
characterizing of the semantic distinction correlating with Pron is, however, though intuitively
on the right lines, clearly too vague and informal. In order to be able to test it, and to draw
predictions from it, we need to characterize it in a much more precise and formal way. There
are two independently motivated semantic distinctions discussed in the literature, which might
be useful for this purpose. The first is the stage/individual level distinction (the s/i distinction,
henceforth), and the second is the generic/nongeneric distinction. In the following three sections
I will argue in detail that the generic/nongeneric distinction is the most appropriate for
accounting for the Hebrew data. In section 5 I will show why, despite its intuitive appeal, the s/i
distinction cannot do so.
2. Pron and the Generic/ nongeneric distinction
My claim is that the semantic distinction which correlates with the presence/absence of Pron
in (5) - (6), and many other sentences, is best characterized in terms of a generic / nongeneric
Page 5
5
distinction, in the sense of e.g. Krifka et al (1992) and Chierchia (1992). Specifically, I make
the following generalization:
(9) a. oblig.-Pron sentences can be interpreted as nongeneric only.
b. oblig.+Pron sentences are interpreted as generic only.
Before turning to justifying and supporting this claim, there are two points which need to be
clarified. First, the generalization in (9) does not deal at all with cases where Pron is optional as
in (7b) and (8) above. Many such cases can be accounted for by a more general statement which
subsumes (9), namely that the presence of Pron is associated with genericity, and its absence
with nongenericity. However, many other Optional Pron cases need a more developed and
sophisticated analysis. Thus, I concentrate here on the obligatory +/-Pron cases, which seem to
correlate with the generic/nongeneric distinction in a quite natural and clear way. The extension
of the theory here to the more complicated cases of Optional Pron, and the problems involved
with such a move are discussed elsewhere ( Greenberg (1995)).
The second point concerns the meaning of the terms “generic” and “nongeneric”. Following
e.g. Dahl (1975), Krifka (1987) and Chierchia (1992) I will start by defining generic sentences
as those expressing nonaccidental, law-like generalizations. Formally, they are associated with
an unselective generic quantifier (called Gen, henceforth) with modal force, which binds
situation/event variables introduced by predicates, and variables introduced by indefinites. I
follow here Chierchia’s 1992 claim that all predicates, both s-level and i-level predicates have a
situation variable, which can be bound by Gen (contra to Kratzer’s (1989a) claim that i-level
predicates lack such an argument), and his way of representing generics, according to which a
generic sentence as (10a ) is represented as (10b):
(10) a. Ravens are black b. Gen x,s [ravens (x) & C(s,x)] [black (s,x)]
Page 6
6
The C in (10b) is a context variable, restricting the possible situations and worlds we should
consider. (10b) means that any situation involving any dog in the close enough and contextually
relevant worlds, which in our case are the stereotypical worlds with respect to ravens, is a
situation in which a raven is black. This kind of representation accounts for two central
properties of generics: Their “law-like”, nonaccidental nature, which is captured by the
reference to possible worlds (and not only to the actual one), and the fact that they allow for
exceptions, which is captured by restricting the possible worlds by the modal base and ordering
source to only contextually relevant, e.g. normal or stereotypical worlds (in the sense of Kratzer
(1981)). In what follows, then, I will start by using the term “generic” in this sense. We will see
later on that for some oblig.+Pron sentences this definition will not do, and we will have to
modify it. But in the meantime we will take it as a starting point. Consequently “nongeneric”
will mean “a sentence which is not associated with Gen, and does not express a law-like
statement”.
I will now turn to the data supporting the hypothesis in (9).
3. Oblig.-Pron sentences are nongeneric only.
There are not many kinds of oblig.-Pron sentences, and it is relatively easy to show that all of
them are nongeneric only. One kind illustrated in (6) above, consists of sentences modified by
specific temporal adverbials. Similar examples are brought in (11):
(11) a. ha-’agvaniyot ba-Suk (*hen) kvar ‘adumot
The tomatoes in market 3fm.pl. already red = ‘The tomatoes in the market are already red’
b. ha-ma’amar Seli (*hu) ‘adayin lo muxan
the paper mine 3ms.sg. yet not ready = ‘My paper is not ready yet.’
Page 7
7
The sentences in (6) above and in (11) express accidental and episodic statements, which
are true in very specific situations and in the actual world only. Hence they clearly cannot be
associated with a generic operator, which quantifies over all situations in all (close enough,
relevant) worlds. They can be naturally analyzed, then, as “nongeneric only”. Notice, though,
that Pron is also obligatorily absent in antecedents of “when clauses”, as in (12):
(12) kSe-dani (??hu) Samen, hu holex la-rofe
when Danny 3ms.sg. fat he goes to the doctor= ‘When Danny is fat, he goes to the doctor.’
Unlike (6) and (11), the oblig.-Pron sentence in (12) is not episodic. How can we justify the
claim that it is nongeneric only, i.e. cannot be associated with a generic operator ? The answer
lies in the logical representation of “when clauses”. Such sentences were analyzed (e.g. by
Kratzer 1989a and Krifka et al 1992) as tripartite structures headed by a universal or a generic
quantifier, which has scope over the whole sentence, and binds all free variables in both the
antecedent (restriction) and the consequence (scope). A sentence like (12), for example, is
represented as s / Gen s [Fat (Danny,s)] [Goes to the doctor (Danny,s)], and paraphrased as
“Every situation in which Danny is fat is a situation where Danny goes to the doctor”. Thus, if
the antecedent of the “when clause” was allowed to be generic, i.e. contain its own generic
operator, which bind all free variables in it, then the quantifier heading the whole sentence
would have no variable to bind in the antecedent, and this would lead to vacuous quantification.
I.e. (12) would be represented as * s / Gen s [ Gen s [C (s,Danny)] [Fat (s,Danny)]] [Goes to
the doctor (s,Danny)]. A similar situation arises in English when an overt adverb of
quantification (Qadv.) is added to the antecedent of a “when clause”:
(13 ) * When Danny is usually fat, he goes to the doctor.
Thus, the inappropriateness of Pron in (12 ) can be naturally explained by prohibiting the
Page 8
8
presence of an unselective quantifier, like Gen, in the antecedent, i.e. by demanding that the
antecedent will be nongeneric only. It thus also supports clause (a) of generalization (9).
4. Oblig.+Pron sentences are generic only.
The class of oblig.+Pron sentences is much more heterogeneous than the class of oblig.-Pron
ones. In attempting to show that all such sentences are indeed generic only, and in order to give
more content to the term “generic only”, I will compare each kind of oblig.+Pron sentences with
minimally contrasting subject-predicate strings in which Pron is optional. Unlike the
oblig.+Pron sentences, such strings are often ambiguous between a generic and a nongeneric
reading (as in example (8) above). Again I emphasize that I do not attempt make a general claim
about the ambiguity of all Optional Pron sentences, because the situation there is too complex to
deal with here. Rather, the purpose of using them here is only to make my claim about the
oblig.+Pron cases clearer.
For a start, compare (14a) and (15a) to the Optional Pron sentences (14b ), (15b) and (16):
(14 ) a.‘orvim *(hem) Sxorim b. ha-’orvim Seli (hem) Sxorim
ravens 3ms.pl. black the ravens mine 3ms.pl. black
‘Ravens are black.’ ‘My ravens are black.’
(15) a. ‘orev *(hu) Saxor b. ha-’orev ha-ze (hu) Saxor
raven 3ms.sg. black the raven the this 3ms.sg. black
‘A raven is black.’ ‘This raven is black.’
(16 ) a. ‘orvim (hem) be-derex klal/ af pa’am lo Sxorim
raven 3ms.pl. usually never black = ‘Ravens are usually / never black.’
The bare plural and the indefinite singular subjects in (14a ) and (15a ), respectively, can be
understood as generic only, and Krifka et al 1992 and Chierchia 1992 represent such sentences
Page 9
9
as involving a generic operator with modal force, in the manner of the (10b) above. (We will
come back to another possible representation of (14a) later on). In contrast, the subject-
predicate strings in (14b ) and (15b), with definite subjects and optional Pron, allow both for a
generic and a nongeneric, or accidental readings. For example in its generic reading the English
translation of (14b ) can mean that my ravens are black since I grow only black ravens, whereas
in the nongeneric reading it means that this is an accidental fact. The fact that Pron becomes
optional if an explicit Qadv. is inserted in (14a), as in (16), further supports the claim that the
generic operator involved in the oblig.+Pron sentences like (14-15a) must be a modalized
operator. According to Chierchia 1992), among others, explicit Qadvs. differ from the generic
operator in that they may quantify only over relevant situations/events, and need not involve
quantifying over possible worlds, i.e. they are not (necessarily) modalized.
Oblig.+Pron sentences are not limited, though, to those with indefinite subjects only.
Another kind of such sentences have quantified subject. Compare (17) with (18):
(17) kol yeled ??(hu) ‘amic
every child 3ms.sg. brave = ‘Every child is brave.’
(18) kol yeled ba-kita Seli (hu) ‘amic
every child in the class mine 3ms.sg. brave = ‘Every child in my class is brave.’
As shown by Carlson 1989 and in a more formally developed theory by Kratzer 1989b,
universally quantified sentences are often ambiguous between a generic and an accidental
reading. Carlson cites the following example, (adapted from Dahl (1975)):
(19) Every friend of John’s votes for Socialists
and claims that it may express either a generic or ‘intensional’ reading, where “the sentence
expresses a requirement for anyone who might be a friend of John, and it applies to all of John’s
potential friends, as well as his present ones” (p.185). Alternatively, the sentence is understood
Page 10
10
as nongeneric or ‘extensional’ and “describes a possibly accidental common characteristic of
John’s present friends” (p.186). Unlike (19), however, the Hebrew sentence in (17) is not
ambiguous, and has the generic reading only. What supports this observation is the fact that (17)
entails the truth expressed by (20):
(20) If I were a child, I would be brave.
It is not entirely clear yet how to represent NýP quantifiers with generic meaning. One
possibility is to give the semantics of the generic every as ranging over all potential members
of the set denoted by the noun, in all possible worlds (i.e. over a property), or to use the
situation semantic framework developed in Kratzer (1989b) and make the universally quantified
sentence true in all or none of the situations of a world. Alternatively we can keep the
traditional, extensional every, which will bind the subject variable, and add to the representation
a generic operator which will bind the situation/event variable. This move will capture the fact
that, unlike traditional generic sentences, sentences like (17) and (19) do not seem to allow for
exceptions. In any case, (17) is indeed generic only, as predicted by generalization (9a). In
contrast, the Optional Pron (18) is more similar to Carlson’s original example, in being
ambiguous between generic reading, where perhaps only brave children are accepted to this
class, and an accidental reading, where the noun denotes a located and existing set of children,
and where it simply happens to be the case that all children in this class are brave. Indeed, (18)
does not entail the truth expressed by (21):
(21 ) If I were a child in this class, I would be brave.
Another kind of Oblig.+Pron sentences is (22a). (22b)3 and (23) are the Optional Pron version
Page 11
11
(22 ) a. pariz ??(hi) be-carfat b. ha-nacig Seli (hu) be-carfat
Paris 3fm.sg. in France the representative mine 3ms.sg. in France
‘Paris is in France.’ ‘My representative is in France.’
(23) Karega, ramat -ha-golan be-yisrael (ve-laxen ha-memshala xayevet la-azor la-toshavim)
Now height the Golan in Israel and thus the government must to-help to-the-residents
‘At the moment, the Golan Heights are in Israel, so the government must help the residents.’
Unlike the oblig.+Pron sentences in (14a), (15a) and (17), the genericity of (22a) cannot be
attributed to a generic interpretation of the subject. Nor can it be due to the predicate itself,
which is in this case a locative PP, traditionally classified as stage level, which expresses
transitory properties. Rather, (22a) is generic-only due to the interaction between the stationary
subject and the locative PP. Locations, expressed by locative PPs, are understood as inherent,
nonaccidental properties of stationary, nonmobile objects. Following Chierchia, we can
represent (22a) as Gen s [C(s, Paris)] [in (s, Paris, France)], which means that any situation
involving Paris in all the close enough relevant worlds (where, cities, for example, do not tend
to walk), is a situation where Paris is in France. In contrast, the Optional Pron (22b), with
‘mobile’ subject, is ambiguous between a generic reading (where my representative is located in
France) and a nongeneric one (where he is there episodically)4. Pron is optional also in contexts
were international treaties affect the location of borders, thus treating locations of geographical
areas as temporary properties, as in sentence (23).
In the next kind of oblig.+Pron sentences, it is the predicate which forces genericity:
(24) ha-cipor ha-zot *(hi) xasida
the bird the this 3fm.sg. stork = ‘This bird is a stork.’
(25) a. ha-’is ha-ze (hu) more b. ha-cipor ha-zot yefefiya
the man the this 3ms.sg. teacher the bird the this 3fm.sg. beautiful
‘This man is a teacher.’ ‘This bird is beautiful.’
Page 12
12
Why would (24) be generic only, while (25a,b) not ? (25a) shows that it is not the categorial
status of the predicate (i.e. being nominal or adjectival) which makes the difference. In addition,
the predicates in all three sentences are traditionally classified as individual level, i.e. as
expressing permanent and “tendentially stable” properties. In Chierchia’s 1992 theory, for
example, all three sentences will involve a generic quantifier, since for him, individual level
predicates are inherently associated with a generic quantifier, and sentences with such
predicates are automatically generic-only. A representation with Gen is justified in Chierchia’s
theory in cases like (25a,b), because it is supposed to capture the tendential stability of such
predicates, by letting Gen quantify over all situations in all the “inertia worlds” (where nothing
unexpected happens). The data above, however, suggest that this cannot be the right story, at
least not for Hebrew. Although all three predicates above are i-level and express tendentially
stable properties, only the one in (24) must appear with Pron. This suggest that ‘generic-only’,
at least in the case of Hebrew nominal sentences, is stronger than merely ‘tendentially stable’.
Indeed, the oblig.+Pron cases discussed so far, expressed not merely permanent properties, but
inherent and nonaccidental ones. This seems also the property which distinguishes (24) from
(25a,b). Be a stork expresses a more inherent and nonaccidental property of its subject than be a
teacher or be beautiful. The first property cannot be changed, whereas, although in (25a) and
(25b) the predicates express permanent properties, we can much more easily imagine situations
were this bird was once beautiful, but it is now ugly, or where this man was a teacher for five
years, but now he found another job. Moreover, even if we imagine worlds far away from ours,
where, e.g. a witch got angry at this bird and turned it from a stork to an apple pie, then the NP
this bird cannot be used anymore to refer to the same individual as before. In contrast, if this
bird became ugly, or this man stopped being a teacher, we can still use the same NPs and refer
to the same individuals as before. It seems, then, that in (24) the property expressed by the
predicate must hold of the subject element in all situations in the close enough relevant worlds.
Page 13
13
The predicate thus expresses a nonaccidental, inherent property of the subject and (24) can only
be generic, like the other oblig.+Pron cases already considered. We thus represent (24) in the
same manner, as Gen s [C(this bird,s)] [stork (this bird,s)]. Of course be beautiful or be a
teacher can also be asserted to be inherent, nonaccidental properties of their subjects, and in
such cases they will also be represented as involving a generic operator. The crucial point is,
though, that unlike (24), this is optional. Notice also that this representation can be possible
only if we assume together with Chierchia (1992) and Condoravdi (1992) and contra Kratzer
(1989a), that both s- and i-level predicates have an event argument. Otherwise Gen would have
no variable to bind in (24), which has a definite subject. The data here also implies that, contra
Chierchia’s claim, “tendentially stable” is not necessarily generic, and that there are predicates
(maybe even most i-level predicates) which express permanent but accidental properties.
The generic quantifier in all the examples of oblig.+Pron sentences discussed so far had a an
epistemic modal base and a stereotypical or ‘normal’ ordering source. The relevant worlds
which we consider in evaluating the truth value of these sentences are the worlds where nothing
unexpected happens -with respect to what we know about the laws of nature, the behavior of
stationary objects, or the worlds where our stereotypes (e.g. about children) hold. But, as Krifka
et al (1992) show for English, I-generic sentences can vary much more with respect to these
modal dimensions, and express also mathematical, deontic or linguistic necessities. In this
sense the oblig.+Pron sentences in Hebrew are not different, and can express the same kinds of
statements, seen in (26a - c), respectively:
(26) a. Stayim ve-’od Stayim *(hem) ‘arba b. yeled tov *(hu) yeled menumas
two and two 3ms.pl four boy good 3ms.sg boy polite.
‘Two and two are four.’ ‘A good boy is a polite boy.’
c. genikolog *(hu) rofe naSim
Gynecologist 3ms.sg. doctor women = ‘A gynecologist is a women-doctor.’
Page 14
14
Until now, then, the hypothesis in (9) above seems to be true; all the cases of oblig.-Pron are
necessarily nongeneric and cannot be represented with Gen, whereas all the cases of
oblig.+Pron are necessarily generic, and must be represented with Gen. This kind of correlation
between the presence / absence of Pron in the syntax, and the presence / absence of Gen in the
semantic representation, leads to a very tempting hypothesis, namely that Pron is a syntactic
realization of Gen (a “Realization Hypothesis”). Unfortunately, examining the data more closely
shows that this hypothesis cannot be right in this form.
First, although all the oblig.+Pron sentences we have discussed above had only generic
readings, it is not clear that all of them should be represented as associated with a generic
operator. Generically interpreted bare plural subjects are analyzed by Krifka et al 1992 and
Wilkinson 1991 as ambiguous between a quantificational reading, of the sort we gave above,
and another, kind referring, reading. The latter interpretation can be seen clearly when bare
plurals appear with kind level predicates like being exterminated, as in (27 ):
(27) Dinosaurs were exterminated three millions years ago
The generic interpretation of the subject in (27) cannot be attributed to binding by Gen. The
property of being exterminated is not, and cannot be, attributed to all individual dinosaurs in all
the relevant close enough possible worlds, but rather, to the kind itself. The bare plural in (27),
is then, kind referring and not bound by Gen, and, as analyzed by Krifka et al (1992 ), it
functions semantically like a proper name (in a manner suggested originally by Carlson (1977)).
The crucial point for us, however, is that if bare plurals are semantically ambiguous between the
kind referring and the quantificational reading, there is no way to guarantee that they are not
ambiguous in the same way, even when combining with predicates which are not kind level5.
Thus, besides the reading we gave to (14a) above, there is no way to rule out a reading where
Page 15
15
the property of being black is predicated of the kind ravens, with no generic operator involved at
all. In other words, (14a) can be represented also as (28):
(28) Black (RAVENS) (where capitals stand for a kind)
Since this is the situation, we cannot claim anymore that in all the oblig.+Pron sentences in
Hebrew Pron must be a syntactic realization of the generic operator. Looking at a wider set of
oblig.+Pron sentences seems to support this conclusion. Consider the sentences in (29)-(31):
(29) xol *(hu) cahov
sand 3ms.sg. yellow = ‘Sand is yellow.’
(30) daniel *(ze) (?tamid) Sem populari
“Daniel” 3ms.sg. (imp.) always name popular = ‘”Daniel” is (?always) a popular name.’
(31) ‘iSun *(hu) (? bederex klal) hergel mesukan
smoking 3ms.sg. usually habit dangerous = ‘Smoking is (?usually) a dangerous habit.’
(29) poses a similar problem to the “realization hypothesis” as (14a) above. Krifka et al
1992 shows that, like bare plurals, mass nouns can combine with kind level predicates. So, in
their generic use they are also analyzed as ambiguous between a quantificational and kind
referring interpretation. One may still claim, however, that although the subjects in (14a) and
(29) may be kind referring, and thus, need not be bound by Gen, Gen must be nonetheless
present in them in order to bind the situation argument of the predicate. I.e. that e.g. (14) should
be represented as Gen s [C(s,RAVENS] [black(s,RAVENS] (meaning “Every situation in the
close enough possible worlds involving the kind ‘ravens’, is a situation where this kind is
black”) , and therefore that Pron must be a syntactic realization of Gen after all. As suggested by
Krifka et al 1992, this kind of reading for sentences like (14a) and (29) is possible, but I don’t
Page 16
16
think there is a way to show that it is obligatory. Furthermore, examining the oblig.+Pron (30)
and (31) shows that even this kind of reading is impossible, and Gen cannot be present at all. In
(30) and (31), even without the overt Qadvs., Gen cannot bind the subjects. Being a popular
name is not a property of many individual callings or utterances of the name “Daniel”, and
being a dangerous habit is not a property of many individual smoking events or states. Rather,
the predicates necessarily express properties of the name “Daniel” as an abstract kind, and of
smoking as a kind of event (or state). Indeed, Krifka et al (1992) and Carlson (1977) analyze
abstract nouns and gerunds as in (30) and (31), respectively, as kind referring only. Moreover,
in these sentences the obligatoriness of Pron cannot be attributed to the obligatory presence of
Gen which binds the situation argument, since, unlike what happens in (16) above, Pron does
not become optional even when we insert in (30 ) and (31) explicit Qadvs. which bind the
situation variable6. The reason seems to be that the presence of a kind referring subject is
enough to necessitate the presence of Pron, and unlike (14a ) above, the subjects of (30) and
(31) must be kind-referring. The consequence is that whereas Pron is even more clearly a
marker of genericity, we cannot claim, based on these oblig.+Pron sentences, that Pron
syntactically realizes a generic operator.
To conclude this section, all the cases of oblig.+Pron sentences above seem to be analyzed as
generic only, and thus support clause (a) of hypothesis (9). The only change we have to make is
to give a wider interpretation to the term “Generic”, so it includes both quantificational and
kind referring generics. There are two implications to this move. First, we cannot claim that
Pron is a syntactic realization of Gen. Second, the fact that both kinds of genericity are
expressed in Hebrew in the same syntactic construction (namely oblig.+Pron sentences) implies
that despite the semantic differences between the two, there may share a common semantic
property, which is responsible for their common syntactic distribution.
Page 17
17
5. Generic/nongeneric or stage/individual level ?
I want to turn now to briefly examine the second possibility of characterizing the semantic
distinction which correlates with Pron, namely the stage/individual level distinction (s/i
distinction, henceforth), originally discussed by Carlson 1977, and then by Kratzer 1989a,
Diesing 1992, and Chierchia 1992, to name just a few. I will not discuss specific theories of the
s/i distinction here, but talk more generally about assumptions shared by most current theories
(Ramchand’s 1996 theory is different in kind and I deal with it in section 6).
Crucially, the s/i distinction is a distinction between predicates. I-level predicates are
traditionally those expressing permanent properties of their subjects (e.g. know French, be tall,
be a mammal), while s-level ones typically express transitory properties (e.g. be tired , be wet,
speak French, eat ice-cream). The s/i distinction has been shown to be involved in various
semantic and syntactic phenomena, which are taken in the literature to be, in fact, diagnostics
for classifying predicates as s-level or i-level. For example, i-level, but not s-level predicates,
yield a generic-only interpretation of bare plural subjects, and are not compatible with there-
insertion sentences, “when clauses” (with specific subjects), and temporal or locative modifiers.
Turning back to the Hebrew data, we might try to claim that the presence/absence of Pron is
another such diagnostic, and that it is a marker for the s/i distinction, and not the
generic/nongeneric distinction. Specifically, we may hypothesize (32):
(32) a. Predicates in oblig.+Pron sentences can be interpreted as i-level only
b. Predicates in oblig.-Pron sentences can be interpreted as s-level only
Page 18
18
Such an hypothesis, together with the already observed diagnostics to the s/i distinction, can
account for some of the data discussed above, such as the occurrence of only oblig.-Pron
sentences with temporal modifiers, as in (6), and in “when clauses”, as in (12), and why the
generic-only interpretation of the bare plural subjects in oblig.+Pron sentences (as in (14a)).
However, there are three main problems with this hypothesis. First, it cannot distinguish
between cases like (24), and (25a,b), since, as explained above, in all of these cases the
predicates are typical i-level with respect to all diagnostics. Secondly, the fact that the s/i
distinction is a distinction between predicates, leads to a similar problem we encountered with
the predicative/identity approach. There are cases where what makes Pron obligatorily present is
the subject, and not the predicate or the interpretation of the predicate. This can be seen again
when comparing minimal pairs like (14a)-(14b) and (17)-(18). In these cases the predicates are
the same, and there is no apparent meaning difference between them (in all cases they seem to
have the same “level of permanence”). It would be problematic to justify a claim that what
seems to be the same predicate must be i-level in (17), but can be also s-level in (18) (as
required by hypothesis (33)), without independent motivation, since the difference seems to lie
only in the subject. In this sense, the generic/nongeneric distinction, which is phrased in terms
of whole sentences, is preferable as accounting for the distribution of Pron.
The third problem with hypothesis (32 ), which highlights a serious independent problem in
the classification of predicates into s- and i-level predicates, is that the presence and absence of
Pron seems to be sensitive not to the classification of predicates as originally s-level or i-level,
but to their level of permanence in the given sentence. E.g. a typical s-level predicate as in
France appears in an oblig.+Pron sentence (in (22a)), since it is understood there as permanent,
and a typical i-level predicate like pretty appears in an oblig.-Pron sentence, as in (6b) above,
since it is temporally modified, and thus understood episodically. On the surface it seems that
we can easily overcome this problem if we simply assume that when a predicate like pretty is
Page 19
19
temporally modified, as in (6b), it is ‘coerced’ from i-level to s-level. This kind of solution is
independently and explicitly argued for in the theories of e.g. Kratzer 1989 and Chierchia 1992.
However, examining it more closely, it turns out that with respect to temporally modified i-level
predicates this solution is in general problematic. The problem is that once we take a such a
predicate to be a real s-level one, we should expect it to behave like an s-level one with respect
to other diagnostics for the s/i distinction. e.g. bare plural subjects, and there-insertion
sentences. But this expectation is not borne out in reality, as shown in (33) and (34):
(33) a. Tomatoes are green now.(Generic only) b. Israeli girls are blond today(Generic only)
(34) a. *There are tomatoes green in the market now b. *There are Israeli girls blond today.
Having a bare plural subject with a generic only interpretation, and being ungrammatical in
there-insertion sentences, are two of the main traditional properties of i-level predicates. Being
compatible with temporal modification, on the other hand, is taken to be a property of s-level
predicates. What (33) and (34) show is that the originally i-level predicates blond and green
continue to behave like i-level predicates even when temporally modified. Thus, the claim that
in such cases they turn into s-level is very problematic. This conflict between the diagnostics
has two general implications, and one implication for the Hebrew data. First, it questions
seriously the traditional classification of s- and i-level predicates into two clear ‘natural classes’
of predicates. Second it poses an even more serious problem for theories which attempt to
account for the different behavior of the predicates in these diagnostical constructions, using a
single underlying factor, such as the presence of an event argument (in Kratzer’s 1989 theory),
an obligatory association with Gen (in Chierchia’s 1992 theory), or two different kinds of INFL
(in Diesing’s 1992 theory). It seems that these theories should either give up their central claims
about the underlying difference between the two kinds of predicates, or alternatively, give up
one of the diagnostics as central evidence for positing these underlying differences.7 The
Page 20
20
implication for Hebrew is that since the classification of s/i level predicates and the notion of
coercion are problematic in general, the claim that in Hebrew sentences like (6b), the predicate
has really turned into s-level, and consequently the hypothesis in (32), are problematic as well.
This is a further reason to reject an s/i level solution for accounting for the distribution of Pron.
6. The syntax-semantics interface. Or - Why Pron ?
Until now I examined the semantic part of the syntactic-semantic correlation in Hebrew
nominal sentences. The question which is naturally raised now is why would a generic /
nongeneric distinction be syntactically manifested in Hebrew in the presence/absence of this
copular element ? A beginning of an answer can be found in an independent claim made by
both Rapoport 1987 and Rothstein 1995 about a structural difference between +Pron and -Pron
sentences, which I will assume here. Both claim that +Pron sentences are full clauses, in which
Pron realizes the INFL node, whereas in -Pron ones, there is no INFL node at all. To use
Rapoport’s terminology, they are “matrix small clauses”. The implication of this claim is that
the subjects in the two kinds of sentences are located in a different positions. Subjects of +Pron
sentences are located in the external specIP position, whereas subjects of -Pron sentences are
located in the internal position, or the specifier of the ‘matrix small clause’, (or specXP).
Turning back to generalization (9), we can now phrase it like this:
(35) a. A sentences whose subject must be in specIP, external position is generic only.
b. A sentence whose subject must be in specXP, internal position, is nongeneric only.
Page 21
21
An interesting point to note is that there are reports in the literature about similar correlations
between the external/internal position of the subject, and the generic/nongeneric reading of the
sentence, respectively, in other languages where both syntactic positions are available for
subjects at S-structure. Diesing 1992 cites examples from German and Dutch (the latter are
taken from Reuland (1988)) which show that generically interpreted subjects appear in the
external position at S-structure, whereas existentially interpreted subjects appear in the internal
position. In a recent paper, Ramchand 1996 cites data in Scottish Gaelic, which is even more
similar to the Hebrew data. As in Hebrew, when the subject in Scottish Gaelic is external, the
sentence has a permanent /inherent reading only, and the predicate cannot be temporally
modified. In addition, generic indefinites only appear in the syntactic external position. What
these reports from German, Dutch and Scottish Gaelic suggest, is that the syntactic-semantic
correlation found here for Hebrew is not a language particular phenomena, but part of a wider
cross linguistic pattern.
This is already very nice. But we may want to make a step further and try to find the reason
for this correlation in Hebrew. Diesing, Chierchia, and Ramchand all offer explanations for the
correlation between the subject position and the generic/permanent reading of the sentence. We
may try to use one of these theories to explain the correlation in Hebrew too.
Unfortunately, none of these theories can account for the full range of Hebrew data, either
because of independent problems, or because of specific incompatibilities8. Diesing (1992), for
example, claims that “Material in specIP is mapped onto the restrictive clause, and material in
specXP is mapped onto the nuclear scope” (the “Mapping Hypothesis”). This may explain why
(14a)-(15a) are oblig.+Pron, since their subjects are bound by Gen, and thus must appear in the
restriction. However, as shown above, in Hebrew there are also cases of universally quantified
sentences, or those bound by Qadvs., which must also be analyzed as appearing in the
restriction, but which may appear without Pron (see again (16)-(17)). In our terms, these are
Page 22
22
subjects which may be syntactically internal although semantically they must appear in the
restriction. This seems to be a serious problem for the Mapping Hypothesis. An additional
problem with Diesing’s theory lies in her claim that s-level and i-level predicates differ in the
kind of inflection they combine with, which in turn influence their subjects positions. Diesing
herself, however, shows in a footnote (p. 140) that the s/i distinction shows also in INFLless
constructions in English. The Hebrew data reinforces this problem, since oblig.-Pron, which
many times appear with s-level predicates, have no INFL node at all.
The fact in that (16), with a Qadv., the subject may appear without Pron (i.e. in specXP) is
also a problem for Chierchia’s (1992) splitting algorithm. According to Chierchia the generic
operator and explicit Q-advs. dictate the same splitting process of the sentence, where “the
restriction is drawn from the material which is external to the scope” (p.23). Like Diesing,
Chierchia believes that in English this splitting is done at LF. But in languages where subjects
may appear either externally at S-structure, like German and presumably Hebrew, indefinites
bound by Gen, as well as Qadvs. should appear externally at S-structure. Chierchia’s theory
cannot account, then, for the differences between (14a) and (16). What distinguishes between
them seems to be the modal character of (14a), but this is not supposed to play a role in the
syntax. In addition, it seems that neither Chierchia’s, nor Diesing’s theories can account for why
kind referring subjects must appear, like subjects of quantificational generics, In specIP.
Finally, I want to turn to Ramchand’s 1996 theory, which treats the s/i distinction as a
distinction between two kinds of syntactic constructions. Ramchand’s follows the intuition that
‘i-level predicates are about individuals, whereas s-level ones are about situations’. She
formalizes this intuition by claiming that, although both s- and i- level predicates have a
classical Davidsonian event argument (e), they differ in the kind of syntactic external argument
that they project. Specifically, the external argument of i-level sentences is an individual
variable (x variable), whereas that of s-level sentences is a situation variable (s variable). In
Page 23
23
languages like Scottish Gaelic, where the mapping between the syntax and semantic is very
direct, the syntactic external argument is also the semantic subject of predication (what the
sentence is ‘about’), which is anchored to tense, and which restricts default quantifiers (like Q-
advs., Gen or sentential existential closure), and crucially, there is only one such external
argument per sentence. Consequently, in i-level sentences subjects can only appear in specIP
position, and Qadvs. or Gen may bind only the external x variable of an indefinite subject. As to
i-level sentences with definite and referential subjects, like John is careful, Ramchand
represents such a sentence as x: x=j [y [e [‘careful’ (y,e)]]](x), (meaning “There is an
individual, who is John, who has the property of participating in a (nonepisodic) event of being
careful”), and claims that the property, which is directly predicated of the external individual
subject, must be understood as ‘inherent and enduring’. In contrast, in s-level sentences it is the
s variable which appears in the external position Thus, indefinite subjects can only appear
internally, in specXP, and cannot be generic, since Qadvs. or Gen may bind only the s, not the x
variable. The s-level version of sentences with referential subjects like John is careful can have
two different interpretations. The external s variable may be bound by a temporal modifier or
existential closure, leading to an episodic reading of the sentence. In this case, Ramchand
represents it as s [e[‘careful’ (j,e)]](s) , (meaning - “there is an episodic situation which has
the property of being a careful event/property of John”). Alternatively, the s variable may be
bound by Gen, yielding a reading which is truth conditionally very similar to the i-level version
of the sentence, where John is ‘habitually’ careful.
Turning back to the Hebrew data, it seems that several aspects in Ramchand’s theory are
especially promising. Specifically, we may account for the data concerning oblig.+Pron
sentences, whose subjects are in specIP, by claiming that these are i-level constructions only,
with an external x variable. This may immediately account for all the cases where the IP
subjects are bound by Gen ((14a-15a) and maybe also (16)). As for the oblig.+Pron sentences
Page 24
24
with referential subjects (as in (22) and (24)) and those with kind referring subjects (as in (29-
31), we may claim, following Ramchand, that the ‘inherent’ meaning arises because of an
obligatory direct predication of the external individual variable. In contrast, we may claim, the
optionality of ( 25a-b) arises because a ‘habitual’ reading where Gen binds a situation variable
is also available. The advantage of such an analysis lies in the possibility to give a unified
account to all the oblig.+Pron sentences, both the D- and the I-generic ones, by analyzing both
as those in which the predication is of individuals (and not over situations). In addition,
Ramchand’s treatment of the s/i distinction as a distinction between whole sentences, and not
between predicates, seems also desirable in terms of the Hebrew data, and overcomes some of
the problems with the s/i distinction noted in section 5.
A closer examination, however, shows that, as it stands now, the theory faces some serious
problems, which the Hebrew data highlights. First, since in Ramchand’s theory only one
variable per sentence is allowed to be external and restrict quantifiers, Gen cannot be restricted
by both the x and s variables. But in Hebrew this reading is possible for cases such as (14a-15a),
and there is no independent way to rule it out (this is even more clearly true for the English A
bird flies - with a generic subject and habitual predicate). Since the predicate in these sentences,
as well as in (22a) and (24) can get ‘habitual’ interpretation, i.e. be bound by Gen, as in the
Chierchian-style representation we gave above, it is not clear why such cases cannot appear in
an s-level construction too (i.e. without Pron). Secondly, like Diesing’s and Chierchia’s
theories, which do not distinguish Qadvs. from the modalized Gen, Ramchand’s theory cannot
account for the optionality of Pron in cases like (16), which has an explicit Qadv. binding the x
variable. Another independent problem for Ramchand’s theory, relevant here, is that it does not
really explain why direct predication of individuals, as in i-level predication, makes the
predicates express inherent /enduring properties of the subjects, especially since Ramchand’s
Page 25
25
allows in principle the x variable to be anchored to tense. It seems that a more developed notion
of “individual” is in need here (maybe along the lines of Carlson (1977)).
7. Summary and Conclusions - What can Hebrew tell us about Genericity ?
In this paper I claimed that the obligatory presence and absence of the pronominal copula -
Pron - in Hebrew nominal sentences correlates with a semantic distinction, best characterized in
terms of a generic/nongeneric distinction, which can be either quantificational or kind referring
genericity. In this sense, Pron can be seen as a syntactic marker for genericity. Comparing
oblig.+Pron sentences with oblig.-Pron and optional Pron ones, raised several implications for
more general theories of genericity. First, the Hebrew data seems to support the view that
quantificational genericity involves modal force. Second, it suggests that, unlike Chierchia’s
1992 claim, not all i-level predicates are necessarily associated with a generic operator, and in
general, the fact that a property is ‘tendentially stable‘ is not enough to make it generic. In
addition, it shows that unlike Kratzer’s 1989a theory, both s- and i-level predicates have an
event argument, which can be bound by Gen. This claim is made independently also in
Chierchia (1989a), and Condoravdi (1992), among others. The Hebrew data also highlights a
serious independent problem for those theories which bring the diagnostics for the s/i
distinction as evidence for the existence of a single underlying difference between the two kinds
of predicates. Finally, although the correlation in Hebrew between the generic / nongeneric
distinction and the external / internal position of the subject seems to be part of a cross
linguistic pattern, it cannot be accounted for yet by either of the current theories dealing with
similar syntactic-semantic interface processes. A fruitful direction to concentrate on in further
research would be the syntactic manifestations of the semantic properties shared by both
Page 26
26
quantificational and kind referring genericity. In any case, although this and many other
theoretical questions are still left open, the Hebrew data seems useful in showing the directions
in which the answers will be eventually found.
References:
BenDavid, A. (1971) Leshon ha-mikra ve-lshon xaxamim. Dvir, Tel-Aviv, Israel.
Carlson, G. (1977) “Reference to Kinds in English”.Ph.D. dissertation, Umass, Amherst.
Carlson, G. (1989) “On the Semantic Composition of English Generic Sentences” in G.
Chierchia, B.H. Partee and R. Turner (eds.) “Properties, Types and Meaning. II, 167-192.
Dordecht, Kluwer.
Chierchia G. (1992) “ Individual Level Predicates as Inherent Generics”, ms. Cornell
University.
Condoravdi, C. (1992) “Individual Level Predicates in Conditional Clauses”, ms.
Dahl, O. (1975) “on Generics”, in E. Keenen (ed. ) Formal Semantics of Natural Language,
Cambridge University Press, Cambridge, 99-111.
Dowty, D. (1979) Word Meaning and Montegue Grammar , D. Reidel, Dordecht.
Diesing, M. (1992) Indefinites, MIT Press, Cambridge, Mass.
Doron (1983) Verbless Predicates in Hebrew, Ph.D. Dissertation, University of Texas at
Austin.
Greenberg, Y. (1995) “Hebrew nominal sentences and the generic / nongeneric distinction”,
ms. ýBar-Ilan University.
Kratzer, A. (1981) “The Notional Category of Modality”, in H.-J. Eikmeyer &H. Reiser (eds.)
Worlds and Contexts. New Approaches to World Semantics, Berlin, de Gruyter.
Kratzer, A. (1989a) “Stage Level and Individual Level Predicates” in Papers on
Quantification, NSF Grant Report, Umass, Amherst.
Page 27
27
Kratzer, A. (1989b) “An Investigation Into the Lumps of Thoughts” Linguistics and
Philosophy 12, 607-653.
Krifka, M., J. Pelletier, G. Carlson, Ter-Meulen, G., G. Link and G. Chierchia (1992),
“Genericity: An Introduction”, ms. University of Texas at Austin.
Ramchand, G. (1996) “Two Subject Positions in Scottish Gaelic: The Syntax-Semantic
Interface” in Natural Language Semantics, volume 4, no.2, 165-191.
Rapoport T.R.(1987) “Copular, Nominal and Small Clauses: A study of Israeli Hebrew, Ph.D.
Dissertation, MIT, Mass.
Reuland, E. (1988). “Indefinite Subject” In Proceedings of NLS 18, GLSA, Umass, Amherst.
Rothstein, S. (1995) “Small Clauses and Copular Constructions” in A. Cardinalrtti, and M.T.
Gausti (eds.) Syntax and Semantics 28.
Wilkinson, K. (1991). “Studies in the Semantics of Generic NPs” Ph.D. Dissertation, Umass,
Amherst.
1 I want to thank Susan Rothstein for her continuous professional guidance and moral support
in the various stages of working on this paper. I also want to thank Fred Landman, Gennaro
Chierchia, Talila Cohen, Edit Doron, Anita Mittwoch, Daphna Heller, Adam Wyner Carina
Wilkinson, Malka Rappaport Hovav, and especially Angelika Kratzer for helpful comments for
this and earlier versions of the paper.
2 The oblig.+Pron sentences here, and those discussed below can appear without Pron only
when uttered with contrastive stress on the subject or a strong intonation break after it.
3 Thanks to Edit Doron for pointing out this example to me
4 Dowty (1979) (p.173-4) reports that a similar contrast causes a progressive/simple aspect
alternation in English.
Page 28
28
5 Thanks to Angelika Kratzer for suggesting this point to me.
6 The Q-advs. are a bit odd with (31-32) but can be understood as quantifying over e.g. relevant
situations of historical periods in (31 ) or those of average health in (32)).
7 A possible solution for this problem lies in chracterizing the difference between s- and i-level
predicates, which is responsible for their different behavior in the diagnostical constructions,
not by using a single underlying factor, but rather as a cluster of properties which many times,
but not necessarily, go together. And by analyzing the diagnostics as testing differnt porpeties of
this cluster.
8 Because of space limitations I will be able to relate here only to very specific details of these
three much richer theories.
Yael Greenberg
English Department, Bar-Ilan University, Ramat-Gan 52900, ISRAEL.
E-mail: [email protected]