Metalinguistic Quantification: Evidence from Japanese Wh-doublets Yasutada Sudo Institut Jean Nicod, CNRS/ENS/EHESS [email protected] May 31, 2013 Abstract Contrary to the widely held presumption that quotations are quantificationally opaque, I argue that quantification into quotational contexts is in fact possible, although only in a very limited way. Specifically, I show that quantification over expressions, or metalinguistic quantification, into quotational contexts is attested in natural language. The main empirical support comes from Japanese wh-doublets, which I claim are indefinites ranging over expressions. This analysis has a consequence on the theories of quotation, as metalinguistic quantification requires quotations to be able to contain variables, which is only possible under certain theories.

1 Introduction It is widely recognized that quotation constitutes an opaque domain for various syntactic and semantic operations (Cappelen & LePore 2012, Fodor 1970:7f, Oshima 2006, Quine 1960, 1966, Schlenker 1999). Specifically, quotations are opaque in that any dependency between one position inside a quotation and another outside of it generally results in unacceptability. For instance, consider the contrast in (1). (1)

Long-distance wh-movement a. *Whati did John ask “Mary should have read ti ”? b. Whati did John ask that Mary should have read ti ?

Here, a wh-phrase what is extracted from the embedded clause. As shown in (1a), the sentence is judged as ungrammatical if the embedded clause is construed as a quotation, while extraction out of an indirect speech like (1b) is perfectly grammatical. The same point can be made with other linguistic phenomena that require a dependency between the matrix and embedded clauses, e.g. long-distance NPI licensing, long-distance semantic binding, de re readings, and dependent plurals.1 (2)

Long-distance NPI licensing a. *John did not say “Bill ever visited Paris.” b. John did not say that Bill ever visited Paris.

(3)

Long-distance semantic binding a. *John told nobody’si mother “Mary likes himi .” b. John told nobody’si mother that Mary likes himi .

(4)

De re reading a. #John told me “The man drinking martini is a beautiful woman.” b. John told me that the man drinking martini is a beautiful woman.

(5)

Dependent plural a. #The first-years’i advisors all said, “Theyi are the smartest student.”

1 (2a)

is strictly speaking not ungrammatical, as ungrammatical expressions can generally be quoted without making the entire sentence ungrammatical. The point here is that the NPI ever cannot be licensed by not in the matrix clause.

1

b.

The first-years’i advisors all said that theyi are the smartest student.

Quotations are also known to be opaque with respect to purely semantic operations. In general, truth-preserving inferences fail to preserve the truth when performed on a quotation, as demonstrated by the examples below. (6)

Existential Generalization Richard wrote “John is the murderer”. œ Richard wrote “Someone is the murderer”.

(7)

Universal Instantiation John said “Everyone is happy.” œ John said “George is happy.”

(8)

Substition of Identicals Charles said “The evening star is Venus.” ø Charles said “The morning star is Venus.”

(9)

Commutativity of Conjuncts “John and Bill met in Somerville”, said Mary. ø “Bill and John met in Somerville”, said Mary.

In addition to these restrictions, Quine (1960, 1966) contends that quotations are quantificationally opaque too. He adduces an example of the following sort. (10)

a. b.

Ralph said “Ortcutt is a spy”. There is someone x such that Ralph said “x is a spy”.

Clearly, (11a) does not entail (11b), since “x is a spy” is not what Ralph said. This has been taken as showing that quantifiers cannot binding into quotational contexts. The quantificational opacity of quotations has been taken for granted in the literature together with the syntactic and semantic opacity alluded to above. Contrary to this long held view, however, I will show that quotations are not fully quantificationally opaque, and one specific kind of quantification is indeed permitted, namely quantification over expressions, or metalinguistic quantification. In other words, I submit that (10a) entails (11) (cf. Kaplan 1968). (11)

There is some expression X such that Ralph said “X is a spy”.

Furthermore I will show that at least some natural languages are equipped with lexical items that are exclusively used to express metalinguistic quantification. My primary empirical evidence comes from Japanese wh-doublets which I claim are interpreted as indefinites ranging over expressions. The organization of the paper is as follows. In Section 2, I will examine the distribution of Japanese wh-doublets in detail and show that it is limited to quotational contexts, and in Section 3 I will closely look at their semantics, and analyze them as indefinites ranging over particular types of expressions. Then in Section 4, I will discuss theories of quotation found in the literature in light of the proposed analysis, especially focusing on whether metalinguistic quantification can be made possible in these theories. Crucially, metalinguistic quantification requires quotations to be able to contain variables, as represented in (11), but as we will see, not all theories are compatible with this. Adopting the Identity Theory of quotation (Washington 1992), I will offer a model-theoretic treatment of the semantics of Japanese wh-doublets in Section 5. Section 6 concludes the paper.

2 Distribution of Wh-doublets The class of Japanese words that I call wh-doublets in this paper take the form of reduplicated wh-words.2 In the Tokyo dialect of Japanese, to which we confine our attention throughout this paper, there are only seven wh-doublets 2 As

has been well discussed elsewhere, wh-words in Japanese have multiple functions depending on the particles they are associated with, and hence semantically indeterminate. For this reason, they are often referred to as indeterminate pronouns (Kuroda 1965, Kratzer & Shimoyama 2002, Shimoyama 2006). As the wh-doublets only have the indefinite use unlike wh-singlets, this aspect is ignored in this paper and wh rather than indeterminate pronoun is used as the mnemonic for them.

2

despite the presence of more wh-words in the language.3 This suggests that the relevant morphological process is not productive. (12)

dare-dare itu-itu dore-dore ikutu-ikutu

‘who-who’ ‘when-when’ ‘which-which’ ‘how.many-how.many’

nani-nani ‘what-what’ doko-doko ‘where-where’ ikura-ikura ‘how.much-how.much’

The following are examples of wh-words that do not have the corresponding wh-doublets. (13)

doitu doo naze

‘who (impolite)’ ‘how’ ‘why’

dochira docchi dono

‘which of the two, to which direction’ ‘which of the two, to which direction (colloquial)’ ‘which (adonominal)’

In this paper, I will leave aside questions such as why there are these gaps in the morphological paradigm, and what the semantic function of doubling is, and will solely be concerned with the semantic properties of the wh-doublets in (12). The rest of the present section investigates the quite peculiar distribution of the items in (12). Above all, whdoublets cannot appear in simple declarative or interrogative sentences as demonstrated by the sharp ungrammaticality/uninterpretability of the sentences in (14). Throughout the present section, I will not attempt to provide accurate English translations of the examples, deferring the discussion of the semantics of wh-doublets entirely to the next section. Instead, I will put the corresponding English wh-doublets in the positions where the Japanese wh-doublets appear. (14)

a. *Taro-ga nani-nani-o katta yo. Taro-NOM what-what-ACC bought PRT (lit.) ‘Taro bought what-what.’ b. *doko-doko-de gakkai-ga arimasu ka? where-where-at conference-NOM exist Q (lit.) ‘A conference is happening where-where?’

As hinted at in Section 1, Japanese wh-doublets are licensed only in quotations. The remainder of this section is devoted to the presentation of empirical evidence supporting this distributional generalization. 2.1 Pure and direct quotations It should first be made clear what I mean by quotation. Descriptively speaking, quotations appear as arguments of two different kinds of predicates (Abbott 2003, Recanati 2000). One class of predicates taking quotations are those that describe some properties of expressions, such as is a noun, is ungrammatical and is not an English word. (15)

a. b. c.

“Dog” is a noun. “Eats John apples?” is ungrammatical. “Natto” is not an English word.

The other kind of predicates taking quotations are attitude predicates that describe actions or states involving a use of language, such as write, say, sing and ask. (16)

a. b.

Richard Montague wrote at the end of his paper, “this is not a reading with which David Kaplan would be content.” Prof. Lettvin is known to have said, “You might regard it as the womb of the Institute.”

These sentences are associated with an inference that the quoted material is what the subject of the attitude predicate actually used. Thus, for instance, if what Montague actually wrote were “this isn’t an interpretation which David Kaplan would be happy with” instead, the sentence in (16a) would be false, although what is conveyed by these sentences are more or less the same. I call this inference a verbatim inference. Generally, quotations used with attitude predicates of the above sort are associated with verbatim inferences. 3 In

many other dialects of Japanese, albeit not in the dialect we focus on, wh-doublets seem to act as plural indeterminate pronouns too, possibly in addition to the metalinguistic indefinite use reported in this paper. I ignore this dialectal variation in this paper.

3

Following the conventions in the literature (Cappelen & LePore 2012, Botterell & Stainton 2005), quotations occurring as arguments of these two types of predicates are henceforth referred to as pure quotations and direct quotations respectively. Thus, the generalization defended here can be stated as (17). (17)

Generalization: Japanese wh-doublets only appear in pure and direct quotations.

I will show below that wh-doublets are licensed in both types of quotations (I will discuss a third type of quotation, mixed quotations in Section 2.4). 2.2 Wh-doublets in pure quotations Generally, wh-doublets are licensed in pure quotations, as demonstrated in (18). (18)

a.

b.

“nani-nani-o taberu” -wa dooshiku-da. “what-what-ACC eat” -TOP verb.phrase-is (lit.) ‘ “what-what eat” is a verb phrase.’ “sono hi dare-dare-ga kita jikan” -wa bunpootekina meishiku-desu. “that day who-who-NOM came time” -TOP grammatical noun.phrase-is (lit.) ‘ “the time at which who-who came on that day’ is a grammatical noun phrase.’

The meanings of those sentences are discussed in Section 3. 2.3 Wh-doublets in direct quotations For direct quotations, there are some language specific confounding factors: certain occurrence of direct quotations are ambiguous between a direct quotation interpretation and an indirect speech interpretation.4 More specifically, direct quotations and non-quotational complement clauses in Japanese both appear with the clause-final marker to, which is often analyzed as a complementizer. Thus Japanese direct quotations are unlike English counterparts which cannot appear with the complementizer that. For example, the following sentence has two interpretations. (19)

John-wa [Bill-no baka-ga ki-ta to] itta. John-TOP [Bill-GEN stupid-NOM came-PAST TO] said a. ‘John said that the stupid Bill came.’ b. ‘John said “the stupid Bill came”.’

In one reading, the attitude expressed by baka ‘stupid’ is attributed to the speaker, while in the other reading, it is attributed to the matrix subject John. This ambiguity is accounted for by analyzing the embedded clause as ambiguous between indirect speech and direct quotation. In fact, in the latter reading, it can be additionally inferred that John uttered the exact word reported by the sentence, which is a verbatim inference, a hallmark of direct quotation. Thus, there is good reason to believe that some direct quotations are string identical to indirect speech reports, but they are semantically distinguishable when words like baka ‘stupid’ are present.5 As expected from our generalization that wh-doublets only appear in quotations, replacing Bill in the above sentence with a wh-doublet dare-dare ‘who-who’ disambiguates the reading. Thus, the following sentence has only the reading under which the attitude holder, John, has a depreciative attitude toward the embedded subject. (20)

John-wa [dare-dare-no baka-ga ki-ta to] itta. John-TOP [who-who-GEN stupid-NOM came-PAST TO] said (lit.) ‘John said “the stupid who-who came”.’

Furthermore, (20) is assocaited with the verbatim inference that John used the expression X-no baka-ga kita for some name expression X, just as in the quotational reading of (19) above. It should be remarked here that (20) does not necessarily imply that John used the wh-doublet dare-dare in his original speech, an interesting semantic property of wh-doublets that will be discussed in greater detail in Section 3. The crucial point here is that this example confirms 4 This

has been the traditional view (Coulmas 1985, Hirose 95, Oshima 2006), but Maier (2008) recently proposes that Japanese does not have the direct-indirect distinction. The data presented in the rest of this section contradicts his claim. I will take up Maier’s claim in some detail at the end of this section, and argue that the data he raises is actually not incompatible with the traditional view. 5 See Oshima (2006) for more tests to distinguish direct quotations from indirect speech in Japanese.

4

the generalization that wh-doublets are licensed only in quotations. There are also cases where direct quotations and indirect speech are morphologically distinguished. For example, quoted questions and embedded questions differ in that quoted questions always appear with to, while embedded questions never do. For instance, the following pair of sentences involving quoted questions do not entail each other. (21)

a.

b.

John-wa [ kinoo Bill-to Mary-ga atta ka to ] kiita. John-TOP [ yesterday Bill-and Mary-NOM met Q TO ] asked ‘John asked, “Did Bill and Mary meet yesterday?”.’ John-wa [ kinoo Mary-to Bill-ga atta ka to ] kiita. John-TOP [ yesterday Bill-and Mary-NOM met Q TO ] asked ‘John asked, “Did Mary and Bill meet yesterday?”.’

Were it not for to, these sentences would entail each other, as the embedded clauses would be construed as embedded questions. Thus, the presence of to necessarily makes an embedded question a quotation, and its absence makes it necessarily a non-quotation. In accordance with the generalization (17), an embedded question containing a whdoublet requires to, as shown in (22). (22)

John-wa [ kinoo dare-dare-ga kita ka *(to) ] kiita. John-TOP [ yesterday who-who-NOM came Q TO ] asked (lit.) ‘John asked, “Did who-who come yesterday?”.’

The above two kinds of data vindicate that wh-doublets are licensed in direct quotations, but not in non-quoted embedded clauses. Yet, I will provide two additional kinds of evidence to corroborate this generalization further. Both kinds of data involve contexts where only quotational interpretation is possible, and therefore they are weaker evidence in that they only show that quotation is a necessary condition for licensing wh-doublets. First, if the embedded sentence involves expressions in a foreign language, it can only be interpreted as a direct quotation. (23)

Galileo-wa [eppur si muove to] itta. Galileo-TOP [eppur si muove C] said ‘Galileo said “eppur si muove”.’

In a situation where (23) is true, it is possible to infer that Galileo used the expression eppur si muove. In other words, (23) is obligatorily associated with a verbatim inference. This suggests that the embedded sentence can only be a direct quotation. As we expect, wh-doublets can appear among embedded foreign words. (24)

Galileo-wa [nani-nani si muove to] itta. Galileo-TOP [what-what si muove C] said (lit.) ‘Galileo said “what-what si muove”.’

Another well known context where an indirect speech interpretation is independently absent is embedded clauses occurring with non-embedding predicates, such as sigh and laugh, which one can confirm by using various tests mentioned at the outset and also by checking the verbatim inference. (25)

John-wa [fuyu-ga moosugu kuru to] tameikiotsuita. John-TOP [winter-NOM soon come C] sighed ‘John sighed, “Winter is approaching”.’

Again, wh-doublets are licensed in such clauses, as expected. (26)

John-wa [dare-dare-ga kuru to] tameikiotsuita. John-TOP [who-who-NOM come C] sighed (lit.) ‘John sighed, “who-who came”.’

To sum up the observations so far, we just saw above empirical evidence that pure and direct quotations license wh-doublets. Also, we already saw some data in (14) demonstrating that wh-doublets are not licensed in matrix clauses. Moreover, some of the data presented in this subsection shows that non-quoted embedded clauses, declarative or interrogative, do not license wh-doublets. All in all, these observations support the generalization in (17) that wh5

doublets only appear in direct and pure quotations. For the sake of completeness, I will examine in the next subsection, a third type of ‘quotation’, called mixed quotation, and show that wh-doublets are not licensed in them. 2.4 Mixed quotations and wh-doublets Many researchers, most notably Abbott (2003) and Recanati (2000), argue that there are two basic types of quotations (Geurts & Maier 2005, Partee 1973, Potts 2007, Stainton 1999). Similar ideas have been around for quite a while, and consequently there are several different terms used to make the same distinction. In this paper, I follow Recanati and call them closed quotation and mixed quotation. Closed quotation is essentially a cover term for pure quotation and direct quotation, and all instances of quotations we have seen so far in this paper are closed quotations. Closed quotations are used to talk about the quoted expression qua expression. Specifically, in the case of pure quotations, the predicate describes the property of the quoted expression, and in the case of direct quotation, the predicate describes how the quoted expression is, was or will be used. It is also noticeable that a closed quotation itself does not inherit the linguistic property of the quoted expression with respect to the non-quoted portion of the sentence. For example, the direct quotation “Eat an apple” in the following sentence does not behave as a predicate, but as a noun phrase, although what is quoted is a predicate. (27)

“Eat an apple” is a predicate.

What distinguishes mixed quotations from closed quotations is that the quoted material in a mixed quotation does not lose the linguistic properties of the quoted material. Thus, when a predicate occurs in a mixed quotation, it behaves as a predicate just as much as it does in non-quotational contexts. This is illustrated by the following examples. (28)

a. b.

According to Quine, quotation “has a certain anomalous feature”. Alice said that life is ‘difficult to understand’.

(Cappelen & Lepore 1997:433)

Although they do not lose their morpho-syntactic properties with respect to the non-quotated parts of the sentences, they are quotations in the sense that they are associated with verbatim inferences just like direct quotations, i.e. they necessarily imply a use of the quoted expressions. For example, (28a) implies that Quine used the phrase has a certain anomalous feature, and (28b) implies that Alice actually uttered the phrase difficult to understand.6 Interestingly, wh-doublets are not licensed in mixed quotations, as shown in the following examples. (29)

a. *Quine-niyoruto, inyoo-wa “nani-nani-no seishitsu-o motteiru”. Quine-according.to quotation-TOP “what-what-GEN property-ACC have” b. *Alice-ga iuniwa, jinsee-wa “nani-nani-ni-wa fukuzatsuda”. Alice-NOM according.to, life-TOP “what-what-DAT-TOP difficult”.

An implication of this observation is that closed and mixed quotations must be given different theoretical treatments. In particular, I assume that the crucial property of closed quotations that licenses wh-doublets is that the quoted material is treated as expressions. This point will be elaborated in the subsequent sections. Although some authors such as Cappelen & Lepore (1997), Cappelen & LePore (2012) and Davidson (1979) pursue a uniform analysis where the two types of quotations are treated as essentially sharing the semantic core, this type of approach is not empirically warranted given the considerations here (see also Stainton 1999 for other criticisms against it). In this paper, I will only give an analysis of closed quotation, and leave it undiscussed how mixed quotations should be analyzed (see Clark & Gerrig 1990, Recanati 2000, Saka 2005 and papers in de Brabante 2005 for various ideas). 2.5 Excursus: Maier (2008) on Japanese speech reports Before moving on, I would like mention that the data in (29) argues against Maier’s (2008) analysis of Japanese speech reports (cf. fn.4). He claims that Japanese does not have genuinely direct quotations and puts forward an account where alleged direct quotations are analyzed as mixed quotations. However, the data we have seen above suggests that Japanese does have a distinction between direct quotation, indirect speech and mixed quotations. In this subsection I will review the empirical facts he raises in support of his claim. 6 The

verbatim inference with a mixed quotation is less linguistically specified, compared to that of a direct quotation. For instance, the attitude holder and/or the attitude predicate are not explicitly mentioned in the sentence containing the mixed quotation. See papers in de Brabante (2005) for examples and theories.

6

Looking in particular at the interpretation of indexical and perspective-sensitive expressions in embedded sentences, Maier points out that they are sometimes interpreted relative to the current context of utterance, even if they are in a clause with a quotational feature (cf. Kuno 1988). Let us look at his examples one by one. Suppose that my boss said to me the following. (30)

asatte madeni kono shigoto-o yar-e! day.after.tomorrow until this work-ACC do-IMP ‘Finish this work before the day after tomorrow!’

One day later, I can report this to a third party by uttering (31). (31)

[ashita madeni sono shigoto-o yar-e] to jooshi-ni iw-are-mashi-ta. [tomorrow until that work-ACC do-IMP] TO boss-by say-PASS-POLITE-PAST ‘I was told by the boss that I should finish that work by tomorrow.’

Maier claims that since the embedded clause in (31) is in the imperative mood, it has to be a quotation, but the indexical ashita ‘tomorrow’ is interpreted with respect to the context of the current utterance, rather than that of the original utterance. He then concludes that this is neither a true direct nor true indirect speech report (cf. Kuno 1988). However, the premise of this argumentation that imperatives cannot be embedded seems to be incorrect. Recent work on imperatives suggest that they are in fact embeddable, at least to some extent (Crniˇc & Trinh 2009a,b, Kaufmann 2012, Thomas 2012). A closer look reveals that imperatives are indeed generally embeddable under the verb iw- ‘say’ in Japanese. For instance, consider the following example. (32)

[shukudai-o yar-e to] okaasan-ni iw-are-mashi-ta. [homework-ACC do-IMP C] mother-by say-PASS-POLITE-PAST ‘I was told by my mother to do the homework.’

This example contains an imperative verb yar-e ‘do-IMP’ just as in the Maier’s example above. However, the implication that the speaker’s mother used this verb verbatim is not at all obligatory. Sociolinguistically, the imperative suffix -e indicates impoliteness and rudeness, and typically a mother does not use it when addressing to her children.7 Thus, the felicity of (32) indicates that the verbatim requirement, a hallmark of direct quotations, is not observed with embedded imperatives in Japanese. This suggests that they are not quotations at all, and that Maier’s data in (31) is not showing the points he is making. Furthermore, a wh-phrase taking a long-distance scope may appear in an embedded imperative in Japanese. (33)

[nani-o yar-e to] okaasan-ni iw-are-mashi-ta ka? [what-ACC do-IMP C] mother-by say-PASS-POLITE-PAST Q ‘What were you told by your mother to do?’

This example also shows that Japanese imperatives are embeddable. Moreover, an NPI like nanimo ‘anything’ can be long-distance licensed as shown in (34). (34)

[nanimo yar-e to] iw-are-nakat-ta yo. [anything do-IMP C] say-PASS-NEG-PAST PRT ‘I wasn’t told to do anything.’

Thus, Maier’s first example does not show that Japanese has no genuine direct quotation. Maier’s other examples involve items that are perspective sensitive, such as ik-u ‘go-PRES’ and yar-u ‘give-PRES’ (the examples are originally due to Oshima 2006). (35)

a.

b.

kinoo, Matsushima-kun-wa [kyoo boku-no uti-ni ik-u to] it-ta. yesterday, Matsushima-TOP [today me-GEN home-to go-PRES TO] say-PAST ‘Yesterday, Matsushima said that he would go to my home today.’ kinoo, Matsushima-kun-wa [kyoo boku-ni purezento-o yar-u to] it-ta. yesterday, Matsushima-TOP [today me-DAT present-ACC give-PRES TO] say-PAST ‘Yesterday, Matsushima said to me that he would give me a gift today.’

7 Instead,

the gerundive suffix -te is generally used, which sounds less peremptory. There seems to be other semantic differences between these two types of imperatives, but their details do not concern us here.

7

The indexicals kyoo ‘today’ and boku ‘I’ in these examples are interpreted with respect to the current context of utterance, but the perspective verbs reflect the attitude holder’s point of view, rather than the speaker’s point of view. Generally, in matrix contexts a use of ik- is infelicitous if its subject is moving toward the speaker (or the speaker’s home base). Similarly, the receiver of yar- cannot be the speaker. This is shown in (36). (36)

a. #Matsushima-wa kyoo boku-no uti-ni iku. Matsushima-TOP today me-GEN home-to go ‘Matsushima is going to my place.’ b. #Matsushima-wa kyoo boku-ni purezento-o yaru. Matsushima-TOP today me-to present-ACC give ‘Matsushima will give me a present.’

Therefore, the perspective verbs in (35) suggest that the embedded clauses are quotations. However, the interpretation of the indexicals is incompatible with this possibility. Given this, Maier claims that these clauses involve mixed quotations, rather than direct quotations. However, there is a problem for Maier’s account: The above sentences do not give rise to verbatim inferences, and thus do not imply that the perspective verbs were used by Matsushima in his original speech, contrary to what is expected for mixed quotations. For instance, they can report the following utterances which contain different verbs. (37)

a.

b.

Matsushima: ashita omae-n ti yor-u ne tomorrow you-GEN place stop.by-PRES PRT ‘Matsushima: I’ll stop by your place tomorrow.’ Matsushima: ashita ii mon age-ru yo. tomorrow good stuff give-PRES PRT ‘Matsushima: I’ll give you something good tomorrow.’

I would like to point out here a different way of looking at Maier’s data in (35). It is a possibility that the shifted perspectives reflected on ik- and yar- in these examples are due to some shifting mechanism. That is, in attitude contexts perspective verbs like ik- and yar- can be interpreted from the perspective of the attitude holder, rather than that of the speaker, as suggested by Kuno (1988) and Oshima (2006), among others.8 Although I have to leave the details of perspective shifting for another occasion, given the availability of perspective shifting in the above contexts, the sentences are actually compatible with the view that Japanese has a distinction between direct and mixed quotations, contrary to Maier, but in harmony with our claim that Japanese has distinctions between closed and mixed quotations and wh-doublets only appear in the former.

3 Semantics of Wh-doublets In the previous section, I raised empirical evidence that the distribution of wh-doublets is strictly limited to closed quotations. Now we are ready to turn to their semantics. Above all, there is a remarkable semantic difference between wh-doublets and ordinary expressions occurring in (closed) quotations. That is, wh-doublets semantically behave as if they are not quoted. For example, the following sentence does not entail John uttered the expression dare-dare, but rather, it means that John uttered some expression of the form X-ga kita, where X is a word denoting a person.9 (38)

John-wa “dare-dare-ga kita” to itta. John-TOP who-who-NOM came C said

For instance, (38) can be used to report John’s utterance of Bill-ga kita. Interestingly, not all utterances of the form X-ga kita can be reported using (38). Specifically, I observe that only utterances containing a referring expression in place of dare-dare can, and those containing a quantificational expression cannot. More concretely, (38) above cannot be used to report utterances of (39). 8 Perspective

shifting is, in this sense, very similar to indexical shifting, where indexicals are interpreted relative to the reported context, rather than the current context of utterance (Schlenker 1999, 2003, Anand & Nevins 2004, Anand 2006, Sudo 2012). However, these two mechanisms cannot be the same, as indexicals are not shifted in (36) (see Sudo 2012 for a claim that Japanese indexicals can optionally shift). 9 Strictly speaking, there is an alternative, less salient interpretation that John uttered the ungrammatical sentence dare-dare-ga kita. Thus, whdoublets can actually be quoted, but what is interesting is that they do not have to be quoted. We will come back to this ambiguity in fn.18.

8

(39)

a.

b.

John: ‘minna-ga kita.’ everybody-NOM came John: ‘Everybody came.’ John: ‘nan-nin-ka-no onnanoko-ga kita.’ what-CL-KA-GEN girl came John: ‘Some girls came.’

This pattern extends to specific indefinites. Thus, (40) cannot be reported by (38) either. (40)

John ‘aru otoko-ga kita.’ certain man-NOM came John: ‘A certain man came.’

By contrast, the following sentences involving demonstrative subjects can be reported by (38), just as much as a sentence involving a proper name, e.g. Bill-ga kita, can. (41)

a.

b.

John: ‘sono onna-ga kita.’ that lady-NOM came John: ‘That lady came.’ John: ‘soitu-ra-ga kita.’ that.guy-PL-NOM came John: ‘Those guys came.’

Thus, dare-dare appears in place of an expression referring to a person or persons in the original report, but cannot replace a quantificational expression, including specific indefinites.10 This is basically the same for the other whdoublets except for the restriction on the referent of the replaced expression. For instance, doko-doko ‘where-where’ replaces an expression referring to a place or places, itsu-itsu replaces an expression referring to a time or times. An exception to this pattern is nani-nani ‘what-what’, which acts as a ‘wild-card’: it does not impose any restriction on the replaced expression, and thus the following sentence can report any of the sentences in (40) and (41).11 (42)

John-wa “nani-nani-ga kita” to itta. John-TOP [what-what-NOM came C] said

In order to account for this replacement effect, I propose that the Japanese wh-doublets are interpreted as indefinites ranging over referring expressions. More concretely, I analyze them as involving existential metalinguistic quantification in the following manner. I henceforth use typewriter font to denote expressions in the metalanguage. (43)

(38) is true if and only if there is an expression X such that a. X refers to a person or persons; and b. John uttered the expression X-ga kuru

The other wh-doublets can be analyzed in the same manner but with different restrictions on the variable X in (43a). For instance, for doko-doko ‘where-where’, it will be ‘X refers to a place or places’. There is one aspect of (38) that is not captured by (43). That is, typically it implies that the speaker does not know the exact word that John used in place of dare-dare ‘who-who’, or that the speaker considers it unnecessary or overinformative to specify the exact word. This ignorance/indifference inference is not encoded in the analysis in (43). However, the ignorance/indifference inference should not be hardwired in the semantics of wh-doublets themselves, as it does not arise in certain sentences. For instance, consider (44) which involves another quantifier in the matrix sentence, in place of John. (44)

hotondono hito-ga “dare-dare-ga kuru” to itta. most person-NOM “who-who-NOM come” C said

10 This generalization

conflicts with the view that specific indefinites are referring expressions (Fodor & Sag 1982, Constant 2012), but is in line with other views that treat specific indefinites as quantificational such as Schwarzschild (2002). I will not discuss potential theoretical implications on the analysis of specific indefinites here. 11 One caveat here is that (42) is likely to be read with an extra inference that the speaker is not sure if what John used was a referring expression or not. This can be understood as a sort of conversational implicature: Because the cooperative speaker did not use a more specific form, the hearer infers that there must be a reason for this, namely because the speaker does not have a belief compatible with the specific form.

9

We are interested here in the surface scope reading where the matrix subject takes the widest scope. Under this construal, the sentence is judged true in a context where more than half people uttered expressions of the form X-ga kuru where X is an expression referring to a person or persons. These truth conditions of the sentence are captured by the proposed analysis. (45)

For most of the people y, there is an expression X such that a. X refers to a person, and b. y uttered X-ga kuru.

Importantly, (44) is not associated with an ignorance/indifference inference.12 Thus, (44) does not suggest that the speaker doesn’t know which person uttered which expression, or that the speaker is not interested in specifying who uttered what. Rather, it is associated with a different inference that not everyone said the same thing. This inference, which I call a heterogeneity inference, does not directly follow from the analysis in (45) per se. Although these chameleonic inferences call for an explanation, I leave the semantics¢aof wh-doublets simply existential as represented above (see Appendix for a pragmatic explanation for the inferences). Particularly crucial is the claim that wh-doublets are indefinites ranging over expressions. This is vindicated by (44), where wh-doublets interact with other scope bearing elements in the matrix clause. Finally, it should also be remarked that there are cases where the quantificational force seems to be generic rather than mere existential. The generic reading is especially prominent with pure quotations, which typically are definitional or characterizing statements. To see this concretely, one of the earlier examples is repeated here with a rough translation. (46)

“nani-nani-o taberu” -wa dooshiku-da. “what-what-ACC eat” -TOP verb.phrase-is ‘It is generally the case for an expression X that X-o taberu is a verb phrase.’

The quantificational force here is arguably not universal, since the expression derived by replacing X can well be ungrammatical with a value of X that is not a noun phrase or that is ungrammatical itself, which would make the sentence trivially false. Rather, the quantificational force in this sentence should be understood as generic. Such generic interpretations of wh-doublets are not problematic for the present analysis, since the wh-doublets are indefinites, and indefinites are known to be able to have generic interpretations, especially in characterizing sentences such as (47) (cf. Cohen 2001, Krifka, Pelletier, Carlson, ter Meulen, Link & Chierchia 1995). (47)

A lion has a mane.

(Krifka, Pelletier, Carlson, ter Meulen, Link & Chierchia 1995:15)

I will not be committed to a particular analysis of generic indefinites here, but suffice it to say that whatever accounts for the generic reading of normal indefinites is applicable to the generic use of of wh-doublets. To repeat the main claim of the present section, wh-doublets are indefinites ranging over expressions. Given this semantics, their peculiar distributions restrictions naturally follow. They can only appear in contexts where expressions appear as expressions, i.e. closed quotations. We will discuss this in more precise terms in Section 5. Before that, we will examine implications of our analysis of wh-doublets on theories of quotation. In particular, being quantificational, the semantics of wh-doublets requires a view where quotations can contain variables.

4 Theories of Quotation As remarked in Section 1, it has long been believed that quantification into quotation is entirely impossible. As a result, most of the theories of quotation have been built on this premise and designed to prohibit any form of quantification. Since the semantics of wh-doublets instantiates metalinguistic quantification into quotation, the strongest form of quantificational opacity of quotations needs to be rejected, and consequently those theories that entirely ban quantification into quotation need to be reconsidered. In what follows, I will examine five representative theories of quotation with a particular emphasis on whether they allow metalinguistic quantification.13 I will claim that only one of these five theories allows metalinguistic quantification. 12 When dare-dare ‘who-who’

is read as taking wide scope over the subject quantifier, the sentence will be associated with an ignorance/indifference inference. 13 The discussion to follow is far from comprehensive with respect to considerations motivating each theory. The interested reader is referred to the survey article by Cappelen & LePore (2012) and the references below.

10

4.1 Name Theory According to the Name Theory, a quotation is an unstructured name that refers to the expression enclosed by the quotation marks (Carnap 1947, Quine 1951, Reichenbach 1940). For example, the quotation “black dog” is analyzed as the name that refers to the expression black dog. This view is primarily attributed to Tarski (1956) and Quine (1951).14 Evidently this theory cannot accommodate metalinguistic quantification, simply because a name cannot contain a variable as its subpart. As Cappelen & LePore (2012) remark, furthermore, many other arguments against this theory have been offered by a number of authors, and the theory has long been considered inapproriate (Cappelen & Lepore 1997, Cappelen & LePore 2012, Davidson 1979, Geach 1957, Saka 1998, Washington 1992). 4.2 Description Theory Under the Description Theory, a direct quotation is an amalgam of one or more referring expressions. There are two minor variants of this view. According to Geach (1957) and Quine (1960), each letter/phonological segment that appears in a quotation refers to itself. For instance, a occurring in a quotation refers to the letter a. On the other hand, in the version due to Tarski (1956), each word in a quotation refers to itself, e.g. dog occurring in a quotation refers to the word dog. In either view, a quotation is decomposed into pieces, unlike in the Name Theory. Under the Description Theory, a quotation has subparts, and in principle quantification over such parts should be possible if variables rangeing over letters/segments are available. However, this way of formulating metalinguistic quantification is not adequate for describing the semantics of the Japanese wh-dobulets. Recall that the wh-doublets (except for nani-nani ‘what-what’) have restrictions on their range. For example, dare-dare ‘who-who’ ranges over referring expressions denoting individuals. This is a problem for the Geach-Quine version of the theory, as such a restriction cannot be appropriately modeled with quantification over segments/letters. Also in the Tarski version, metalinguistic quantification will necessarily involve quantification over words. However, wh-doublets are more flexible than this. For instance, nani-nani may range over any kind of expression including expressions smaller than a word. For example: (48)

“nani-nani-ru”-wa genzaikee-no dooshi-da “what-what-PRES”-TOP present.tense-GEN verb-is ‘ “X-ru” is a present tense verb (for any X)’

In Japanese, a verbal stem is a bound morpheme that is smaller than a word, but nani-nani can perfectly range over such morphemes at a sub-word level, as demonstrated by (48). Moreover, the domain of a wh-doublet may contain expressions bigger than a word. We have already seen such a case in (41a), where the domain of dare-dare must contain a complex DP sono onna ‘that woman’, which is made up of two words. Thus, although it allows metalinguistic quantification, the Description Theory is not suited for formalizing the semantics of wh-doublets. See Cappelen & LePore (2012), Davidson (1979) and Washington (1992) for more criticisms of this theory independent of metalinguistic quantification. However, it should be remarked that the above discussion also suggests a solution. Notice that the sole problem of the Description Theory in accommodating the semantics of wh-doublets is its inflexibility with respect to the size of the variable. That is, this theory assumes the variables to be of a particular size, but wh-doublets range over expressions of any size. In fact, the view that I will adopt in the end is a version of this view with variables of many different sizes. 4.3 Demonstrative Theory The Demonstrative Theory was originally proposed by Davidson (1979) and has been widely entertained (Bennett 1988, Cappelen & Lepore 1997, Christensen 1967, García-Carpintero 91, Goldstein 1984, 1985, Partee 1973, Predelli 2008, Prior 1971). The main tenet of this theory is that the quotation marks are a sort of demonstrative expression that refers to (the shape of) the quoted material. For example, the sentence in (49a) can be paraphrased as (49b) using a demonstrative this (this example is originally due to Davidson 1979:90). (49)

a.

‘Alice swooned’ is a sentence.

14 But

Bennett (1988:401) writes: “Although distinguished philosophers have seemed to commit themselves to this theory [i.e. the Name Theory], I don’t believe that any of them have meant to do so. In particular, it is not credible that Frege, Tarski and Quine were unaware that what is displayed in a quotation is systematically related to what it names.” See also Saka (1998:114f) and the references in Cappelen & LePore (2012).

11

b.

Alice swooned. The expression of which this is a token is a sentence.

According to this theory, a sentence containing a quotation is semantically made up of two separate components, i.e. the presentation of the quoted portion (Alice swooned in (49b)), and a sentence containing a demonstrative (The expression of which this is a token is a sentence in (49b)). This feature make the theory fundamentally incompatible with metalinguistic quantification, as the variable occurring in the first component will not be bound by the quantifier in the second component. More concretely, consider the sentence in (50), for instance. (50)

‘dare-dare swooned’-wa eigo-no bun-da ‘who-who swooned’-TOP English-GEN sentence-is

According to the Demonstrative Theory, (50) is analyzed as follows (with a generic meaning). (51)

X swooned. It is generally the case that if X refers to a person, the expression of which this is a token is a sentence in English.

The second sentence here is predicted to be always false, since a sentence containing a variable like X is not a sentence in English. This example involves a generic quantifier, but nothing hinges on this. The exact same argument can be made with a sentence like (52). (52)

John-wa “dare-dare-ga kita” to itta. John-TOP “who-who-NOM came” C said

This is analyzed as (53). (53)

X-ga kita. There is an expression X that refers to a person such that John uttered the expression of which this is a token.

Again, the demonstrative in the second sentence will refer to a sentence that contains a variable. This does not capture the reading of (52). See Botterell & Stainton (2005), Cappelen & LePore (2012), Recanati (2000), Saka (1998) and Washington (1992) for other independent criticisms against the Demonstrative Theory. 4.4 Demonstration Theory Clark & Gerrig (1990), Recanati (2000) put forward the Demonstratin Theory where the quotation marks are analyzed as simply indicating that the quoted words are being demonstrated. By assumption, the intended referent of the demonstration is determined pragmatically. An appeal of this theory is that various different referents of quotations can be given a uniform theoretical treatment. However, this theory is also fundamentally inappropriate for metalinguistic quantification. Specifically, in order to allow quantification over expressions, ‘demonstration’ needs to be understood as something that can contain variables over expressions. Then, however, demonstration cannot conceived of as an actual event or action, as Clark & Gerrig (1990) and Recanati (2000) envisage, since such an event or action cannot contain variables, which are in their essence part of some language and do not exist spacio-temporally. 4.5 Identity Theory Frege (1892), Reichenbach (1940), Saka (1998, 2005), Searle (1969), Reimer (1996), Botterell & Stainton (2005) and Washington (1992) among others are proponents of the Identity Theory of quotation, according to which quoted expressions just refer to themselves. For example, the quotation “Bill likes Sue” denotes the expression Bill likes Sue. Under this theory, it is possible to implement metalinguistic quantification in such a way that the semantics of wh-doublets can be appropriately formalized. Consider again the sentence in (50), which can be analyzed as follows. (54)

It is generally the case that if X denotes a person, X swooned is a sentence in English.

This appropriately captures the meaning of the sentence. Similarly, (52) can be analyzed as (55). (55)

There is an expression X such that X denotes a person and John uttered the expression X-ga kita.

12

To be more precise, I take it that expressions denoted by quotations may refer to the types, tokens, shapes, etc. of those expressions depending on the use. This is akin to the indeterminacy of referents for expressions like book names. For example, Word and Object can refer to a specific token of a book, or its type, as in the following examples modeled after Reimer’s (1996). (56)

a. b.

That book is Word and Object Work and Object was first published in 1960

I assume that the referents of quotations show a similar kind of variability, but do not attempt here to explain how their referents are actually determined, as our focus is on a particular aspect of the semantics of wh-doublets that does not necessarily require a specific theory of referents of quotations (see García-Carpintero 91, Saka 1998, 2005 for discussion). In the following section, we present a model theoretic treatment of the semantics of Japanese wh-doublets that is built upon the Identity Theory of quotation.

5 A Model-Theoretic Fragment This section presents a model theoretic fragment with quotations and metalinguistic quantification. It is presented as an extension of the extensional semantics of Heim & Kratzer’s (1998), but nothing crucial hinges on this. However, some familiarity with it is presupposed. Firstly, a new atomic type u is added to the inventory of basic types (cf. Maier 2007, Potts 2005, 2007). This is meant to be the type of expressions. (57)

The set of types T is the smallest set such that a. e, t, u P T b. If σ, τ P T , then pσ, τ q P T

The domain Dτ for a type τ is defined in the usual way, except that Du is added. (58)

a. b. c. d.

De is the set of individuals Dt is the set of truth-values, t0, 1u Du is the set of expressions Dpσ,τ q “ DτDσ

The ontology of Du is simple-mindedly assumed to contain any ‘expression’, simplex or complex, interpretable or 15 uninterpretable, linearly ordered or hierarchically structured, verbal Ťor gestural, etc. In this setting, the interpretation function v w can be conceived of as a partial function from Du to τ PT Dτ . It is a partial function, since Du contains uninterpretable expressions by assumption. Now, let us introduce a new compositional rule to allow metalinguistic predication. (59)

Metalinguistic Functional Application (MFA) If α is a branching node with β and γ as its daughters, and v β wQ,g is of type pu, τ q for some type τ , then v α wQ,g “ v β wQ,g pQpγqq.

Ignoring Q for the moment, this rule states that when there is a predicate β that takes an expression, the sister node γ of β will not be interpreted by v w. MFA comes into play when the sentence involves a predicate of expressions, such as is a noun phrase or say, whose lexical entries are given below. (60)

a. b.

v is a noun phrase wQ,g = λXu . X is a noun phrase v say wQ,g = λXu .λye . y utters X

Note that separate lexical entries are assumed for the quotational say and the familiar propositional say (cf. Potts 2007).16 For examples involving non-embedding verbs like sigh, e.g. (25), we can assume that the quotation is part of 15 I

will not go into the discussion of possible ramifications that this assumption brings into the theory. See Cappelen & LePore (2012:§3.4, fn.10) and Postal (2004) for relevant discussion. 16 Usually, when one utters some expression, they stand in a certain relation (saying, demanding etc.) with the meaning of that expression. For example, that John said “Bill loves Mary” seems to entail that John said that Bill loves Mary. However, this does not always hold, as von Fintel (2004) argues, and hence it is not specified in the lexical entry in (60b), and left entirely for pragmatics.I thank Michela Ippolito (p.c.) for discussion on this point.

13

an adverbial clause involving a silent version of the quotational say. Thus, (25) is analyzed as meaning ‘John sighed, saying “Winter is approaching”’. (25)

John-wa [fuyu-ga moosugu kuru to] tameikiotsuita. John-TOP [winter-NOM soon come C] sighed ‘John sighed, “Winter is approaching”.’

In order to allow metalinguistic quantification, several additional assumptions are necessary. Firstly, we deploy syntactically complex indices that are represented as ordered pairs of an integer and a type (cf. Heim & Kratzer 1998, Sudo 2012). Furthermore, indices with u trigger a special rule that allows metalinguistic quantification: (61)

Metalinguistic Predicate Abstraction (MPA) If α is a branching node whose daughters are an index xi, uy for some integer i and β, then v α wQ,g “ X{xi,uy ,g λXu .v β wQ .

This rule modifies the function Q in a similar way that Predicate Abstraction of Heim & Kratzer (1998) modifies the assignment function g. More precisely, QX{xi,uy denotes that function different from Q at most in that it assigns X to xi, uy. The function Q : Du Ñ Du is a substitution function that replaces indexed traces with expressions in the following manner. If Q “ ri Ñ x, j Ñ y, k Ñ z, ...s where i, j, k, ... are indices with type u, then Qpαq “ αrx{ti , y{tj , z{tk , ...s

(62)

Here, αrx{ti , y{tj , z{tk , ...s is meant to be the expression obtained from α by simultaneously replacing every occurrence of ti , tj , tk , ... in α by x, y, z, ... respectively. In this system, the denotations of the wh-doublets can be analyzed as existential quantifiers over type u elements.17 Schematically, it looks as follows. „  1 1 1 1 1 1 @Q1 , g 1 rX P dompv wQ ,g q ñ pv X wQ ,g P De ^ Rpv X wQ ,g q “ 1qs (63) v wh-wh wQ,g “ λPpu,tq .DXu ^P pXq “ 1 1

1

‘v X wQ ,g P De ’ in this representation ensures that X is a referring term. R here is an inherent restriction of the whdoublet (e.g. person for dare-dare ‘who-who’, and place for doko-doko ‘where-where’). It is assumed that the first argument P of type pu, tq is derived via covert Quantifier Raising (QR) of the wh-doublet which leaves a trace with an index xi, uy for some integer i.18 Finally, in order to prevent wh-doublets from appearing in non-quotational contexts, an additional assumption needs to be made.19 To this end, I assume that v txi,uy wQ,g is undefined for any i, Q, g. For an illustration, let us analyze the sentence in (64). The LF representation with the wh-doublet QR’ed looks like (65). (64)

John-wa “Bill-ga dare-dare-o aishitieru” to itta. John-TOP “Bill-NOM who-who-ACC love” C said ‘For some expression X such that X denotes a person, John said “Bill loves X”.’

17 This

way of treating indefinites is probably too simple given the issues such as exceptional wide scope and quantificational variability, but I believe that implementing it in other approaches (e.g., generalized Skolem functions/choice functions, D-closure, etc.) is not particularly difficult using the ingredients given here. 18 As noted in fn.9, a quotation containing a wh-doublet is inherently ambiguous between two interpretations: one where the wh-doublet is used as a metalinguistic indefinite, and one where it is truly quoted (which more often than results in a quotation of an ungrammatical expression). The latter quoted interpretation can be captured by assuming that the relevant QR is optional. In order for this to work, furthermore, it is crucial to assume that QR is not triggered by semantic considerations (e.g. by type-mismatch), but is an optional syntactic operation. 19 I am indebted to Irene Heim for pointing out an inadequacy of my earlier formalization regarding this point.

14

(65)

✟❍❍ ❍ ✟✟ ❍❍ ✟✟ ✟❍ ✟ ❍ dare-dare ✟ ❍❍ ✟ (who-who) ✟ ❍❍ ✟ ✟❍ x6, uy ✟✟ ❍❍ ✟ ❍❍ ✟ ❍❍ ✟✟ ✟❍ John-wa ❍❍ ✟✟ ✟ ❍ “Bill-ga tx6,uy -o aishiteru” to itta (said)

Assuming that itta has the same semantics as say given in (60b), it triggers, and the quoted expression “Bill-ga tx6,uy -o aishiteru” is fed into QX{x6,uy without being interpreted by v w. As a result, the truth conditions of (64) will be the following. „  1 1 1 1 1 1 @Q1 , g 1 rX P dompv wQ ,g q ñ pv X wQ ,g P De ^ personpv X wQ ,g q “ 1qs (66) v (65) wQ,g ô DXu ^john uttered Bill-ga X-o aishiteiru This captures the meaning of the sentence in question.

6 Conclusion To repeat the main proposal of the paper, natural language allows metalinguistic quantification into quotational contexts, and hence quotations are not fully quantificationally opaque, contrary to Quine’s surmise. I raised novel empirical data of Japanese wh-doublets, which I claim are interpreted as indefinites ranging over expressions. The proposed semantics has theoretical implications on theories of quotation in that it demands quotations to be able to contain variables over expressions. Among the theories we considered in Section 4, only the Identity Theory meets this requirement in an appropriate way. I would like to close the paper with some open remarks about other items than Japanese wh-doublets that have similar metalinguistic functions. For instance, English so and so and such and such and French un tel can appear in quotational contexts without being quoted and behave as indefinites, just like Japanese wh-doublets. This is shown in (67).20 (67)

Ce terme vient de Sam Weller [...] qui [...] ponctue régulièrement ses propos de locutions du this term came from Sam Weller [...] who [...] punctuate regularly his words with phrases of.the type « comme dit untel... ». type “as said untel...”. ‘This term came from Sam Weller who punctuates his words with phrases of the type “as so-and-so said”.’

However, unlike Japanese wh-doublets, they seem to be able to appear outside of quotations as well.21 (68)

On appelle relation un ensemble d’attributs qui définissent un fait – par exemple [...] que son nom est one calls relation a set of-attributes that define a fact – for example [...] that his name is untel, [...] untel, [...] ‘What one calls relation is a set of properties that define a fact – for example, that his name is so-and-so, ...’

The semantics of these items are not well-understood, but it is often remarked that they function as place holders for proper names. It appears to me that there is an interesting connection with the idea that proper names themselves involve quotations (Bach 1981, 2002, Burge 1973, Geurts 1997, 1999, Katz 1977, Matushansky 2008), but an analysis of these items is left for future research. Irene Heim (p.c.) and Angelika Krater (p.c.) point out to me that German has expressions like der und der and das und das that seem to have very similar distributions and semantics to Japanese wh-doublets, but Cieschinger & Ebert 20 Taken 21 Taken

from http://fr.wikipedia.org/wiki/Wellerisme (Retrieved (5/31/2013). from http://fr.wikipedia.org/wiki/Mod%C3%A8le_relationnel (Retrieved 5/31/2013).

15

(2010) claim that they cannot be analyzed as indefinites over expressions. I leave detailed examination of these items for another occasion.

A

Pragmatics of Wh-doublets

As briefly mentioned in Section 3, wh-doublets used in simple episodic sentences like (69) are typically associated with ignorance/indifference inferences. (69)

John-wa “dare-dare-ga kuru” to itta. John-TOP “who-who-NOM come” C said ‘John said “So-and-so will come” (but I cannot or don’t want to say who that was).’

This sentence suggests that the speaker does not know or does not care which referring expression John actually used. We also noted that the inference is not always present, for instance, when there is another scope bearing element in the matrix clause. (44)

hotondono hito-ga “dare-dare-ga kuru” to itta. most person-NOM “who-who-NOM come” C said

Instead of an ignorance/indifference inference, (44) is associated with a heterogeneity inference that different people used different referring expressions in place of the wh-doublet dare-dare ‘who-who’. The quality of these inferences is quite similar to the inferences associated with so-called epistemic indefinites such as German irgendein-indefinites and Spanish algún-indefinites (Kratzer & Shimoyama 2002, Alonso-Ovalle & Menéndez-Benito 2010, 2011, 2013, Aloni to appear, Aloni & Port to appear, Alonso-Ovalle & Shimoyama to appear). These indefinites are characterized by ignorance inferences in simple episodic sentences. For instance, the Spanish sentence (70), originally due to Alonso-Ovalle & Menéndez-Benito (2010:2), implies that the speaker does not know how the student in question is. (70)

María se casó con algún estudiante del departamento de lingüııstica. María SE married with ALGÚN student of.the department of linguistics ‘María is dating a linguistics student.’

Furthermore, these indefinites are also known to trigger heterogeneity inferences in the scope of quantifiers like every student (Alonso-Ovalle & Shimoyama to appear). We adopt a pragmatic analysis along the lines of analyses of epistemic indefinites proposed by Kratzer & Shimoyama (2002) and Alonso-Ovalle & Menéndez-Benito (2010, 2011, 2013) where the ignorance/indifference and homogeneity inferences of epistemic indefinites are derived as conversational implicatures.22 Specifically, I assume that the lexical semantics of wh-doublets themselves are merely existential, as proposed in Section 3, but they trigger a pragmatic inference in relation to (referring) expressions in their domains of quantification. For illustration, let us analyze (69). The use of this sentence indicates that there are alternative sentences where the wh-doublet is replaced with a (relevant) referring expression, for instance (71). (71)

a. b. c.

John-wa “Aaron-ga kuru” to itta. John-wa “Ben-ga kuru” to itta. John-wa “Charles-ga kuru” to itta.

Given that these sentences are more informative than (69), when the cooperative speaker obeying the Gricean Maxim of Quantity utters (69), the hearer infers that the speaker would have violated the Maxim of Quantity if any of (71) were used instead. That is, the inference is that the speaker does not hold a belief that (71a) is true, and does not hold a belief that (71b) is true, and does not hold a belief that (71c) is true. This amounts to an ignorance inference that the speaker does not know which expression John actually used in place of the wh-doublet. In order to explain the hetergeneity inference for (44), we need one additional assumption. Following Sauerland (2004), among others, I assume that the conversational implicatures that the speaker doesn’t believe that p, henceforth Bs ppq, that are derived in the above manner can be strengthened to Bs p pq on the assumption that the speaker is opinionated about the truth of p, i.e. either Bs ppq _ Bs p pq. This strengthening operation is dubbed Epistemic 22 This

is not to deny other proposals such as Aloni (to appear), Aloni & Port (to appear).

16

Step. It is assumed that an Epistemic Step from a set of conversational implicatures t Bs ppq, Bs pqq, Bs prq, . . .u to tBs p pq, Bs p qq, Bs p rq, . . .u can be taken, if the latter does not contradict the literal meaning of the sentence. In the case of (69), the implicature cannot be strengthened, because it would entail that the speaker believes that none of the sentences in (71) are true, i.e. the speaker believes that John didn’t utter a sentence of the relevant form. On the other hand, an Epistemic Step can take place in the case of (44). The literal meaning is in (72). (72)

For most of the students, there is an expression X referring to a person and they uttered X-ga kuru.

The relevant alternatives are sentences like the following: (73)

a. b. c.

hotondono hito-ga “Aaron-ga kuru” to itta. hotondono hito-ga “Ben-ga kuru” to itta. hotondono hito-ga “Charles-ga kuru” to itta.

From these alternatives, we obtain conversational implicatures that the speaker doesn’t believe (73a), (73b) or (73c). Then by an Epistemic Step, we strengthen these inferences to that the speaker believes that the negation of these sentences. These inferences are compatible with the assertion and amount to the inference that there are more than one expression that the relevant students used in place of the wh-doublet, which is a heterogeneity inference. Acknowledgments The present work is based on my unpublished MA thesis submitted to the University of Tokyo in March 2007. I would like to thank the members of my thesis committee, Tsuneko Nakazawa, Christopher Tancredi, and Shuichi Yatabe. Some parts of the paper were presented at the International Conference on East Asian Linguistics (ICEAL) in 2006, Sinn und Bedeutung (SuB) 12 in 2007, and at the International Conference on Quotation and Meaning (ICQM) in 2007, and I would like to thank the audiences of these conferences. I am also indebted to many friends, colleagues and mentors with whom I have discussed this work, including Chris Barker, Christopher Davis, Kai von Fintel, Danny Fox, Daniel Gutzmann, Irene Heim, Michela Ippolito, Satoshi Ito, Makoto Kanazawa, Angelika Kratzer, Eric McCready, Mitsue Motomura, Hiroki Narita, Toshiyki Ogihara, David Y. Oshima, Maribel Romero, Jacopo Romoli, Mamoru Saito, Philippe Schlenker, Yael Sharvit, Natsumi Shibata, Satoshi Tomioka, Masahiro Yamada, and Tomoyuki Yoshida. The present work was supported by a Euryi grant from the European Science Foundation (Schlenker:‘Presupposition: A Formal Pragmatic Approach’); the ESF is not responsible for the claims made here. All errors are solely mine.

References Abbott, Barbara. 2003. Some notes on quotation. Belgian Journal of Linguistics 17. 13–26. Aloni, Maria. to appear. On epistemic indefinites: a note on emphatic free choice uses. In Proceedings of Sinn und Bedeutung 16, . Aloni, Maria & Angelika Port. to appear. Epistemic indefinites crosslinguistically. In Proceedings of NELS 41, . Alonso-Ovalle, Luis & Paula Menéndez-Benito. 2010. Modal indefinites. Natural Language Semantics 18(1). 1–31. Alonso-Ovalle, Luis & Paula Menéndez-Benito. 2011. Domain restrictions, modal implciatures and plurality: Spanish Algunos. Journal of Semantics 28(2). 211–240. Alonso-Ovalle, Luis & Paula Menéndez-Benito. 2013. Two views on epistemic indefinites. Language and Linguistics Compass 7(2). 105–122. Alonso-Ovalle, Luis & Junko Shimoyama. to appear. Expressing ignorance i the nominal domain: Japanese Wh-ka. In Proceedings of WCCFL 31, . Anand, Pranav. 2006. De De Se: Massachusetts Institute of Technology Ph.D. dissertation. Anand, Pranav & Andrew Nevins. 2004. Shifty operators in changing context. In Proceedings of SALT 14, . Bach, Kent. 1981. Referential/Attributive. Synthese 49(2). 219–244. Bach, Kent. 2002. Giorgione was so-called because of his name. Philosophical Perspectives 16. 73–103. Bennett, Jonathan. 1988. Quotation. Noûs 22. 399–418. Botterell, Andrew & Robert Stainton. 2005. Quotation: Compositionality and innocence without demonstration. Crítica 37(110). 3–33. Burge, Tyler. 1973. Reference and proper names. The Journal of Philosophy 70(14). 425–439. Cappelen, Herman & Ernest LePore. 2012. Quotation. In Edward N. Zalta (ed.), The Stan-

17

ford Encyclopedia of Philosophy, Metaphysics Research Lab, CSLI, Stanford University. http://plato.stanford.edu/archives/win2008/entries/quotation/. Cappelen, Herman & Ernie Lepore. 1997. Varieties of Quotation. Mind 106(423). 429–450. Carnap, Rudolf. 1947. Meaning and Necessity. Chicago: University of Chicago Press. Christensen, Neils E. 1967. The Alleged Distinction Between Use and Mention. The Philosophical Review 76(3). 358–367. Cieschinger, Maria & Cornelia Ebert. 2010. Mario hat gesagt, der und der hätte dans und das gemacht: On a weird determiner in German. In Proceedings of Sinn und Bedeutung 11, . Clark, Herbert & Richard Gerrig. 1990. Quotations as demonstrations. Language 66. 764–805. Cohen, Ariel. 2001. Relative readings of many, often, and generics. Natural Language Semantics 9. 41–67. Constant, Noah. 2012. Witnessable quantifiers license type-e meaning: evidence from CT, equatives and supplements. Talk given at SALT 20. Coulmas, Florian. 1985. Direct and indirect speech: general problems and problems of Japanese. Journal of Pragmatics 9(1). 41–63. Crniˇc, Luka & Tue Trinh. 2009a. Embedding imperatives. In Proceedings of NELS 39, . Crniˇc, Luka & Tue Trinh. 2009b. Embedding imperatives in English. In Proceedings of Sinn und Bedeutung 13, . Davidson, Donald. 1979. Quotation. Theory and Decision 11. 27–40. de Brabante, Philippe. 2005. Hybrid Quotations. John Benjamins. von Fintel, Kai. 2004. How Multi-Dimensional is Quotation?: Comments on “The Dimensions of Quotation” by Chris Potts. Harvard-MIT-UConn Workshop on Indexicals, Speech Acts, and Logophors. Fodor, Janet Dean. 1970. The Linguistic Description of Opaque Contexts: Massachusetts Institute of Technology Ph.D. dissertation. Fodor, Janet Dean & Ivan Sag. 1982. Referential and quantificational indefinites. Linguistics and Philosophy 5. 355–398. Frege, Gottlob. 1892. Über Sinn und Bedeutung. Zeitschrift für Philosophie und Philosophische Kritik 100. 25–50. García-Carpintero, Manuel. 91. Ostensive signs: Against the identity theory of quotation. Journal of Philosophy 253–264. Geach, Paul. 1957. Mental Acts. New York: Routledge. Geurts, Bart. 1997. Good news about the description theory of names. Journal of Semantics 14. 319–348. Geurts, Bart. 1999. Presuppositions and Pronouns. Amsterdam: Elsevier. Geurts, Bart & Emar Maier. 2005. Quotation in context. Belgian Journal of Linguistics 17(1). 109–128. Goldstein, Laurence. 1984. Quotation of types and types of quotation. Analysis 44. 1–6. Goldstein, Laurence. 1985. The title of this paper is ‘quotation’. Analysis 45. 137–140. Heim, Irene & Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Hirose, Yukio. 95. Direct and indirect speech as quotaions of public and private expression. Lingua 223–238. Kaplan, David. 1968. Quantifying in. Synthese 19(1-2). 178–214. Katz, Jerrold J. 1977. A proper theory of names. Philosophical Studies 31(1). 1–80. Kaufmann, Magdalena. 2012. Interpreting Imperatives. Dordrecht: Springer. Kratzer, Angelika & Junko Shimoyama. 2002. Indeterminate pronouns: The view from Japanese. In Yukio Otsu (ed.), Proceedings of the 3rd Tokyo Conference on Psycholinguistics, 1–25. Tokyo: Hituzi Syobo. Krifka, Manfred, Francis Jeffry Pelletier, Gregory Carlson, Alice ter Meulen, Godehard Link & Gennaro Chierchia. 1995. Genericity: an introduction. In Greg Carlson & Francis Jeffry Pelletier (eds.), The Generic Book, 1–124. Chicago: University of Chicago Press. Kuno, Susumu. 1988. Blended quasi-direct discourse in Japanese. In Papers from the Second International Workshop on Japanese Syntax, 75–102. Kuroda, Shige-Yuki. 1965. Generative Grammatical Studies of the Japanese Language: Massachusetts Institute of Technology Ph.D. dissertation. Maier, Emar. 2007. Mixed quotation: between use and mention. In Proceedings of LENLS 2007, Miyazaki, Japan. Maier, Emar. 2008. Japanese reported speech: against a direct-indirect distinction. In Proceedings of LENLS, . Matushansky, Ora. 2008. On the linguistic complexity of proper names. Linguistics and Philosophy 31(5). 573–627. Oshima, David Y. 2006. Perspectives in Reported Discourse: Stanford University Ph.D. dissertation. Partee, Barbara H. 1973. The syntax and semantics of quotation. In Stephen R. Anderson & Paul Kiparsky (eds.), A Festschrift for Morris Halle, 410–418. New York: Holt, Rinehart, and Winston, Inc. Postal, Paul Martin. 2004. The openness of natural languages. In Skeptical linguistic essays, 173–201. Oxford

18

University Press. Potts, Christopher. 2005. The Logic of Conventional Implicatures. Oxford: Oxford University Press. Potts, Christopher. 2007. The dimensions of quotation. In Christopher Barker & Pauline Jacobson (eds.), Direct Compositionality, 405–431. Oxford: Oxford University Press. Predelli, Stefano. 2008. The demonstrative theory of quotation. Linguistics and Philosophy 31. 555–572. Prior, Arthur. 1971. Objects of Thought. Oxford: Oxford University Press. Quine, Willard Van Orman. 1951. Mathematical Logic. Cambridge, MA: Harvard University Press revised edn. Quine, Willard Van Orman. 1960. Word and Object. Cambridge, MA: MIT Press. Quine, Willard Van Orman. 1966. The Ways of Paradox and other essays. Cambridge, MA: Harvard University Press revised and enlarged edn. Recanati, François. 2000. Oratio Obliqua, Oratio Recta. Cambridge, MA: MIT Press. Reichenbach, Hans. 1940. Elements of Symbolic Logic. Cambridge, MA: MIT Press. Reimer, Marga. 1996. Quotation marks: Demonstratives or demonstrations? Analysis 56. 131–141. Saka, Paul. 1998. Quotation and the Use-Mention Distinction. Mind 107. 113–135. Saka, Paul. 2005. Quotation constructions. In Philippe de Brabante (ed.), Hybrid Quotations, John Benjamins. Sauerland, Uli. 2004. Scalar implicatures in complex sentences. Linguistics and Philosophy 27(3). 367–391. Schlenker, Philippe. 1999. Propositional Attitudes and Indexicality: MIT Ph.D. dissertation. Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy 26(1). 29–120. Schwarzschild, Roger. 2002. Singleton indefinites. Journal of Semantics 19(3). 289–314. Searle, John R. 1969. Speech Acts. Cambridge: Cambridge University Press. Shimoyama, Junko. 2006. Indeterminate phrase quantification in Japanese. Natural Language Semantics 14(2). 139– 173. Stainton, Robert J. 1999. Remarks on the Syntax and Semantics of Mixed Quotation. In Keiko Murasugi & Robert J. Stainton (eds.), Philosophy of Linguistics, 259–278. Boulder, CO: Westview Press. Sudo, Yasutada. 2012. On the Semantics of Phi Features on Pronouns: Massachusetts Institute of Technology Ph.D. dissertation. Tarski, Alfred. 1956. The concept of truth in formalized languages 152–278. Clarendon Press. Thomas, Guillaume. 2012. Embedded imperatives in Mbyá. Talk given at NELS 43. Washington, Corey. 1992. The Identity Theory of Quotation. Journal of Philosophy 89(11). 582–605.

19