Indexical shift and the long-distance reflexive caki in Korean∗
Yangsook Park
University of Massachusetts Amherst
Manuscript
September, 2014
1
Introduction
It has been proposed in the literature that certain elements are obligatorily interpreted de se in
many languages, such as shifted indexicals, logophors, PRO, etc (Chierchia 1989, Schlenker 1999,
2003, Anand and Nevins 2004, Anand 2006, a.o.). In this paper, it will be shown that two types of
elements in Korean, shifted indexicals and the long-distance reflexive caki, must be interpreted de
se within the scope of attitude predicates, and they interact in an interesting way.
(1)
John-i
Mary-ka na-lul cohahanta-ko malhayssta.
John-Nom Mary-Nom I-Acc like-C
said
‘John said that Mary likes {me, John}.’
(2)
John-i
Mary-ka caki-lul cohahanta-ko malhayssta.
John-Nom Mary-Nom caki-Acc like-C
said
‘John said that Mary likes {John, Mary}.’
The embedded first person pronoun na ‘I’ in (1) can be interpreted as the speaker of the utterance
context or the attitude holder John. In (2), caki can also find its antecedent within the embedded
clause, Mary, or across the clause boundary so that it can co-refer with the attitude holder John.
In both examples above, the two elements can refer to the attitude holder only when John utters,
"Mary likes me."
The main purpose of this paper is to examine the uses of indexicals and the long-distance reflexive in Korean as de se elements, and to develop two independent but related analyses for them.
The first part of this paper (Sections 2 and 3) shows that Korean is a language where indexicals can
optionally shift under certain attitude predicates. It will be shown that the shifted-like interpretation of indexicals is the result of context-shifting along the lines of Anand and Nevins (2004) and
Anand (2006). Some cross-linguistically interesting contrasts between person and adverbial indexicals in Korean will also be presented. Section 4 introduces the basic properties of the long-distance
reflexive caki as a logophor, and provides a semantic analysis of binding a logophor with obligatory de se interpretations. Section 5 presents novel data on the interactions between the shifted
indexicals and logophor caki. In particular, I propose a new blocking effect whereby the presence of
∗
I thank Seth Cable, Rajesh Bhatt, and Vincent Homer for their invaluable help, encouragement, and patience
throughout this project. I also thank Angelika Kratzer, Brian Dillon, Lisa Green, Peggy Speas, and Amy Rose Deal
for their helpful comments and insightful discussion at various stages of this work, as well as the members of the Semantics Workshop at UMass, the UMass Syntax-Semantics Reading Group, and audiences at the LSA 2014, GLOW 37
and SALT 24. Of course, all remaining errors are mine.
1
context-shift operators limits the interpretations of the long-distance reflexive. We will also extend
our analysis of caki to the third person pronoun with de se interpretation.
2
Indexical Shift in Korean
The referents of indexicals, such as I, you, here, now, etc., are dependent on the context of utterance
and are fixed for all possible circumstances, while the meanings of indexicals are determined with
respect to certain aspects of the context, e.g. speaker, addressee, time, location, etc. (Kaplan 1989).
For example, the first person pronoun in (3) can only refer to the speaker of the utterance rather
than the matrix subject Mary in English.
(3)
Mary said that I am a hero.
Since indexicals are not affected by modal operators, Kaplan (1989) also argues that these expressions need to be evaluated with respect to a context parameter in addition to other parameters
that are responsible for the semantic evaluation of other expressions. The first person pronoun in
(4b) must refer to the speaker of the context of utterance even under the modal operator. Thus,
unlike (4a), (4b) is false given its truth-conditions: For every world w’ accessible from w, the speaker of
the context of utterance exists in w’.
(4)
a.
b.
(a priori true)
(false)
I exist.
Necessarily, I exist.
Given the fact that indexicals are not affected by any operator in English, Kaplan (1989) makes
a conjecture that there is no operator that can manipulate the context parameter, i.e. no monsters.
However, it has been found that indexicals in the complements to attitude verbs can be interpreted
with respect to the reported context instead of the actual speech context in many languages, such
as Amharic (Schlenker 1999), Zazaki (Anand and Nevins 2004), Uyghur (Sudo 2012), Nez Perce
(Deal 2014), etc., a phenomenon known as ‘indexical shift’. In Zazaki, for example, the first person
pronoun in the embedded clause can refer to the speaker in the actual speech act or the attitude
holder ‘Hensen’, as shown below.
(5)
HEsenij
(m1k -ra) va kE Ezj/k dEwletia
Hessen.obl (I.obl-to) said that I
rich.be-pres
‘Hensen said that {I am, Hesen is} rich.’
Zazaki
(Anand and Nevins 2004)
Although it has not been discussed in the previous literature, both the first person pronoun
na and the locative adverbial yeki ‘here’ in the embedded clause of the attitude predicate can be
ambiguous in Korean, as in (6)-(7).
(6)
Mary-ka nay-ka yengwung-i-lako malhayssta.
Mary-Nom I-Nom hero-be-C
said
‘Mary said that {I am, Mary is} a hero.’
(7)
Utterance in Seoul
Amherst-eyse Mary-ka John-i
yeki-eyse thayenassta-ko malhayssta.
Amherst-at Mary-Nom John-Nom here-at be.born-C
said
‘Mary said in Amherst that John was born in {Seoul, Amherst}.’
2
In (6), the first person pronoun can refer to either the speaker of the utterance or the matrix subject
Mary. Similarly, yeki ‘here’ in (7) can co-refer with the location of the utterance context, Seoul, or
the location of the reported context, Amherst.
In this section, we will examine how the ambiguity shown in (6)-(7) is derived in Korean. I
will first show that Korean is indeed a language in which indexicals can optionally shift in indirect
speech under certain attitude predicates. Then I will present some uncommon contrasts between
person and adverbial indexicals in Korean.
2.1
Strict indexicals in Korean
Following Kaplan (1989) and Schlenker (2003), I distinguish the ‘strict indexicals’, which refer to
the expressions that can only be interpreted indexically, e.g. ‘I’, ‘you’, ‘today’, ‘here’ in English,
etc., from those that can have indexical uses in certain cases, e.g. ‘he’ with a pointing gesture.1 In
what follows, I will first show that the first/second person pronouns and the temporal/locative
adverbials, e.g. ‘here’, ‘today’, ‘now’, etc. are ‘strict indexicals’ in Korean.
The ‘strict indexicals’ contain the information of what the referent would be in the relevant
context (Kaplan 1989). For example, the first person pronoun ‘I’ always refers to the person who
speaks the sentence in any given context c. However, according to Kaplan, since indexicals are
rigid designators, the referent of an indexical that is determined by a context c is always the same
with respect to all possible worlds, while the ordinary description like ‘the speaker’ may co-vary
with a quantifier (Kaplan 1989, Sudo 2012, Deal 2014, a.o.).
(8)
a.
b.
Whenever Obama is speaking, the speaker is president.
#Whenever Obama is speaking, I am president.
(Deal 2014)
The same contrast is observed between the first person pronoun na ‘I’ and the description hwaca
‘the speaker’ or between the second person pronoun ne ‘you’ and chengca ‘the hearer’ in Korean as
well, as shown in (9) and (10).
(9)
(10)
Na ‘I’ vs. Hwaca ‘the speaker’
a. Obama-ka malhal ttyay.mata hwaca-nun taythonglyeng-ita.
Obama-Nom speak whenever speaker-Top president-be
‘Whenever Obama speaks, the speaker is president.’
b. Obama-ka malhal ttyay.mata na-nun taythonglyeng-ita.
Obama-Nom speak whenever I-Top president-be
‘Whenever Obama speaks, I am president.’
(True)
(False)
Ne ‘you’ vs. Cengca ‘the hearer’
a. Obama-ka Biden-kwa malhal ttyaymata chengca-nun pwutaythonglyeng-ita.
Obama-Nom Biden-with speak whenever hearer-Top vice.president-be
‘Whenever Obama speaks with Biden, the hearer is vice president.’
(True)
b. Obama-ka Biden-kwa malhal ttyaymata ne-nun pwutaythonglyeng-ita.
Obama-Nom Biden-with speak whenever you-Top vice.president-be
‘Whenever Obama speaks with Biden, you are vice president.’
(False)
Since the meaning of the first person pronoun does not vary depending on the quantifier, na ‘I’
in (9b) refers to the speaker of this sentence in the actual speech act, i.e. the writer of this paper.
1
Strictly speaking, this distinction is not so accurate as we know that the first and second person pronouns can get
bound variable interpretations (Partee 1989, Kratzer 1998, Rullmann 2004, a.o.). In this paper, however, there will not
be discussion of these cases.
3
Thus, unlike (9a) with the ordinary description ‘the speaker’, (9b) is false given that the writer
of this paper is not president whenever Obama speaks. (10) also shows that the second person
pronoun ne ‘you’ is a strict indexical contrary to the ordinary description ‘the hearer’ in Korean.
In addition to the 1st /2nd person pronouns, the following adverbials are also pure indexicals
in Korean: temporal adverbials (ece ‘yesterday’, onul ‘today’, nayil ‘tomorrow’, cikum ‘now’), and
locative adverbials (yeki ‘here’). The same test as above shows the contrast between the adverbial
indexicals and the ordinary expressions that denote the speech time or location: between the temporal indexical cikum ‘now’ and the ordinary description palhwa sikan ‘the speech time’ (11), and
between the locative indexical yeki ‘here’ and the expression palhwa cangso ‘the speech location’
(12).
(11)
Cikum ‘now’ vs. Palhwa sikan ‘the speech time’
a. Obama-ka malhal ttyay.myun manhun salamtul-i palhwa sikan-ey
Obama-Nom speaks when
many people-Nom speech time-at
pakswuchinta.
clap
‘When Obama speaks, many people clap at the speech time.’
b. #Obama-ka malhal ttyay.myun manhun salamtul-i cikum pakswuchinta.
Obama-Nom speaks when many people-Nom now clap
#‘When Obama speaks, many people clap now.’
(12)
Yeki ‘here’ vs. Palhwa cangso ‘the speech location’
a. Obama-ka malhal ttyay.mata manhun salamtul-i palhwa cangso-ey issta.
Obama-Nom speaks whenever many people-Nom speech location-at be
‘Whenever Obama speaks, many people are at the speech location.’
b. #Obama-ka malhal ttyay.mata manhun salamtul-i yeki-ey issta.
Obama-Nom speaks whenever many people-Nom here-at be
#‘Whenever Obama speaks, many people are here.’
The fact that cikum ‘now’ in (11b) and yeki ‘here’ in (12b) cannot refer to either Obama’s speech time
or location but the actual speech time or location under the scope of the quantifier supports the
claim that these adverbials are in fact pure/strict indexicals in Korean.2
2.2
Direct vs. Indirect Speech
Having shown that the first/second pronouns and several temporal/locative adverbials are pure
indexicals in Korean, let us now turn our attention to the environments where the indexicals can
be interpreted relative to the reported context. As we saw before, in Korean, indexicals in the
embedded clause of certain attitude predicates show an ambiguity similar to those in the languages
that allow indexical shift: they denote the speaker, hearer, time, location of the utterance context
or those of the reported context. In this sub-section, I will show that the shifted-like interpretation
of indexicals in Korean can be derived in non-quoted clauses (or phrases), and thus it is the result
of indexical shift rather than quotations.
One possible approach to the shifted-interpretation of indexicals is to treat the embedded clause
as a direct quotation. In Korean, the special complementizer −lako follows quoted sentences with
any sentence final particles under the attitude predicates, as in (13). In other words, the usage of
2
This test also confirms that other temporal adverbials are also strict indexicals: ece ‘yesterday’ vs. palhwa cen nal ‘the
day before the day of speech’, onul ‘today’ vs. palhwa tangil ‘the day of speech’, nayil ‘tomorrow’ vs. ‘the day after the day
of speech’.
4
the complementizer −lako indicates that the embedded clause is directly quoted.3 Also, when the
first person pronoun is in this type of quoted sentence, it is always interpreted as the speaker of
the quoted sentence, rather than the speaker of the utterance. Thus, in (13), nay ‘I’ can only refer
to Mary, the speaker of the quoted sentence.
(13)
Direct quotation
Mary-ka "Nay-ka ceyil pwuca-eyyo"-lako malhayss-ta.
Mary-Nom "I-Nom most rich.person-Decl"-C said-Decl
‘Mary said, "I’m the richest person."’
On the other hand, when a sentence is embedded under attitude predicates with another complementizer −ko, it could be understood as either a clausal quotation or an indirect reported speech.
Due to this fact, the interpretation of the first person pronoun in the embedded clause can be ambiguous, as shown in (6), repeated below.
(6)
Mary-ka nay-ka yengwung-ila-ko malhayssta.
Mary-Nom I-Nom hero-be-C
said
‘Mary said that {I am, Mary is} a hero.’4
When the embedded clause is a clausal quotation reported by Mary in (6), nay ‘I’ has to refer
to ‘Mary’ as in (13). When the embedded clause is an indirect reported speech, the first person
pronoun can always be interpreted as the speaker of the utterance.
However, it has been argued that indexicals in a non-quoted embedded clause can indeed be
interpreted relative to the reported context instead of the context of utterance in many languages.
Thus, if the same mechanism that allows indexical shift exists in Korean, nay ‘I’ in (6) could still
refer to Mary even when the embedded clause is not a sentential quotation. Therefore, we need
to determine how the shifted-like reading of (6) is derived in Korean: either only under direct
quotation or both under the direct and indirect speech.
There are a number of ways to distinguish direct and indirect speech that have been introduced
in the literature. One well-known property is that a matrix clause and a direct quotation are syntactically opaque to each other, i.e. no A’-extraction, NPI licensing, etc. (Schlenker 1999, Anand
and Nevins 2004, Maier 2008, Sudo 2012, a.o.). In Korean, which is a wh-in-situ language, a whphrase in an embedded clause can take a matrix scope. By contrast, a wh-phrase inside a sentential
quotation cannot take a matrix scope due to the ‘syntactic opacity’ of clausal quotations. Thus, if
a wh-phrase inside an embedded clause can take wide scope, we can confirm that the embedded
clause is not a direct quotation.
(14)
Wh-phrase in embedded clauses
a. Mary-ka "nwu-ka John-ul coahay-yo"-lako malhayss-ni?
Mary-Nom who-Nom John-Acc like-Decl-C
said-Q
3
In Korean, there are various sentence final particles for each sentence type given the degree of formality/colloquialism, and respect, etc. These particles are only allowed to be used in the sentence final position of the
matrix clause. The quoted sentences followed by −lako can take any sentence final particles, even those that can be only
used in the matrix clauses, because a quoted sentence is an independent sentence by itself.
4
Since only a few sentence final particles are allowed in the embedded clause when it is followed by the complementizer −ko, the quoted sentence with the colloquial declarative particle -eyyo in (13) cannot be followed by −ko.
(i)
*Mary-ka nay-ka ceyil pwuca-eyyo-ko
malhayss-ta.
Mary-Nom "I-Nom most rich.person-Decl"-C said-Decl
5
b.
‘Did Mary say, "Someone likes John."?’
Mary-ka nwu-ka John-ul coahanta-ko malhayss-ni?
Mary-Nom who-Nom John-Acc like-C
said-Q
(i) ‘Did Mary say that someone likes John?’
(ii) ‘Who did Mary say likes John?’
In (14a), where a sentential quotation is embedded with the complementizer -lako, the sentence is
only interpreted as a yes-no question by the interrogative marker in the matrix clause with the existential interpretation of the wh-phrase, because the wh-phrase inside the direct quotation cannot
take a matrix scope due to the ‘syntactic opacity’.5 On the other hand, the embedded wh-phrase
under the complementizer -ko in (14b) can take either a wide scope or get an existential interpretation. Thus, if an indexical in the complements to attitude verbs can be interpreted relative to
the reported speech with a wide-scope interpretation of an embedded wh-phrase, we can be assured that the indexical can be shifted in an indirect speech. Let us, then, consider the following
examples.
(15)
a.
b.
(16)
a.
Mary-ka nwuka na-lul coahanta-ko malhayss-ni?
Mary-Nom who I-Acc like-C
said-Q
‘Who did Mary say likes {me, Mary}?’
Mary-ka John-eykye nwuka ne-lul coahanta-ko malhayss-ni?
Mary-Nom John-to
who you-Acc like-C
said-Q
‘Who did Mary say to John likes {you, John}?’
Conversation in Seoul
New York-eyse Mary-ka nwuka
yeki-eyse thayenassta-ko malhayss-ni?
New York-at Mary-Nom who-Nom here-in be.born-C
said-Q
‘Who did Mary say in NY was born in {Seoul, New York}?’
b.
Conversation on January 8th
Cinan cwu-ey Mary-ka nwuka
nayil
ttenanta-ko malhayss-ni?
Last week-in Mary-Nom who-Nom tomorrow leave-C
said-Q
‘Who did Mary say a week ago would leave on {January 9th , January 2nd }?’
As shown in (15) and (16), the shifted interpretations of the person/adverbial indexicals are still
available with the matrix scope interpretation of the wh-phrase in the same embedded clause. This
shows that the shifted interpretations can indeed be derived in indirect speech.
Another piece of evidence in favor of indexical shift in Korean (for adverbial indexicals only)
comes from the usage of the 3rd person pronoun or the long-distance reflexive caki inside an embedded clause with adverbial indexicals, as shown in (17).
5
Because yo in the quoted sentence is a declarative marker, the wh-phrase can only get the existential interpretation.
Even though the sentence final particle in the embedded clause is replaced by the interrogative marker ni, as in (i) below,
the entire sentence is still a yes-no question rather than a wh-question, showing that the embedded wh-phrase cannot
take a wide scope.
(i)
Mary-ka "nwu-ka John-ul coaha-ni"-lako malhayss-ni?
Mary-Nom who-Nom John-Acc like-Q-C
said-Q
‘Did Mary say, "Who does like John"?’
6
(17)
a.
b.
New York-eyse Johni -i
kui /cakii -ka yeki-eyse thayenassta-ko malhayssta.
New York-at John-Nom he/caki-Nom here-in be.born-C
said
‘John said in New York that hei was born {here, in New York}.’
Ece
Johni -i
kui /cakii -ka nayil
ttenanta-ko malhayssta.
Yesterday John-Nom he/caki-Nom tomorrow leave-C
said
‘Yesterday, John said that hei would leave {today, tomorrow}.’
Logically speaking, the embedded clause in (17) cannot be a direct quotation, since the 3rd person
pronoun ku or the long-distance reflexive caki cannot be directly uttered by John to refer to himself.6
Therefore, the shifted-like interpretation of the adverbial indexicals in (17) cannot be due to direct
quotation. Based on the evidence presented above, we can conclude that both the person and
adverbial indexicals in Korean can be interpreted with respect to the reported context, as well as
with respect to the utterance context, in an embedded clause of indirect speech.
Although we have proved that indexicals can be shifted in non-quoted embedded clauses, one
might argue that the shifted interpretation is a result of some special quotation, i.e. mixed/partial
quotation, inside an indirect report (Maier 2007, 2008). Maier (2007) argues that the shifted interpretation is due to a mixed quotation of indexicals, rather than the whole sentential quotation.
However, as noted by Anand (2006), the mixed quotation approach is too weak to account for
the crucial constraint on indexical shift, i.e. Shift Together (Anand and Nevins 2004). If only an
indexical can be quoted in an indirect report, we would expect the first person pronoun in (18) to
be able to be quoted so that it can refer to the matrix subject Mary rather than the speaker of the
utterance context, while the second person pronoun refers to the addressee of the actual context
by not being quoted. Unlike the prediction, however, more than one indexical in the same clause
cannot find its reference from different contexts, as illustrated below.
(18)
Context: John and Sue are having a conversation.
John: Mary-ka Tom-eykey nay-ka ne-lul cohahanta-ko malhayssta.
Mary-Nom Tom-to
I-Nom you-Acc like-C
said
‘Mary said to Tom that I (=John) like you (=Sue, *Tom).’
‘Mary said to Tom that I (=Mary) like you (=*Sue, Tom).’
Finally, as Sudo (2012) points out from Uyghur and Japanese, this analysis can also not explain
why the two different types of complements to the attitude verb (CP vs. NP) differ with respect
to indexical shift. That is, as in Uyghur and Japanese, only CP complement clauses can allow
indexical shift in Korean, while NP complement clauses can not.7
(19)
a.
CP complement clause
Mary-ka nay-ka nwukwu-lul ttaylyessta-ko malhayss-ni?
Mary-Nom I-Nom who-Acc
hit-C
said-Q
‘Who did Mary say {I, Mary} hit?’
b.
NP complement clause
Mary-ka nay-ka nwukwu-lul ttaylyessta-nun sasil-ul malhayss-ni?
Mary-Nom I-Nom who-Acc
hit-NM
fact-Acc said
Unlike the adverbial indexicals, the shifted person indexicals cannot be used with either the 3rd person or the longdistance reflexive caki. I will come back to this issue later.
7
It should be noted that it is unclear even under the indexical shift analysis why NP complement clauses do not allow
indexical shift unlike CP complement clauses.
6
7
Lit. ‘Who did Mary say the fact {I, *Mary} hit?’
As shown in (19), the verb malhata ‘say’ can take a CP argument direlctly or an NP clause with a
complement CP as its argument. If it is the case that nay ‘I’ is quoted so that it can refer to Mary in
the CP complement clause in (19a), nothing would prevent the same indexical nay ‘I’ from being
quoted in the NP complement clause to refer to the attitude holder rather than the speaker of the
utterance. However, the first person pronoun in the NP complement clause must be interpreted
relative to the utterance context, i.e. the speaker of the utterance.
In summary, we have seen that the shifted-like interpretation of indexicals inside an embedded
clause of attitude predicates is indeed possible without any kind of quotation. That is, indexicals
inside an indirect speech can be interpreted with respect to the reported speech by a mechanism
that is different from direct or partial quotation. We will discuss later in detail how indexicals can
get the shifted interpretations.
2.3
2.3.1
Basic properties of shifted indexicals
Shift Together
Anand and Nevins (2004) and Anand (2006) propose a cross-linguistic constraint on indexical shifting, i.e. Shift-Together Constraint. This constraint is illustrated below.
(20)
Shift-Together Constraint
All indexicals within a speech-context domain must pick up reference from the same context.
a. CA [. . . CB . . . [i A . . . i A ]]
B
B
b. *CA [. . . CB . . . [i . . . i ]]
(Anand and Nevins 2004)
A
B
B
A
This constraint also holds for both person and adverbial indexicals in Korean. In other words,
when there is more than one person or adverbial indexicals in the same embedded clause of an
attitude predicate, they must find reference from the same context, either the context of utterance
or the reported context. Consider the following examples with the two person indexicals in (21)
and with the two adverbial indexicals in (22).
(21)
Context: John and Mary are having a conversation.
John: [Bill-i
Sue-eykey [nay-ka ne-lul cohahanta-ko malhayssta.] ]
Bill-Nom Sue-to
I-Nom you-Acc like-C
said
Lit. ‘Bill said to Sue that I like you.’
(No Shift)
a.‘I’ (=John), ‘you’ (=Mary)
(Both Shift)
b.‘I’ (=Bill), ‘you’ (=Sue)
(22)
c.*‘I’ (=John), ‘you’ (=Sue)
(2nd
d.*‘I’ (=Bill), ‘you’ (=Mary)
(1st person Shift)
person Shift)
Context: John and Mary are having a conversation in Boston on July 3rd .
John: [Tom-i
ece
cenyek New York-eyse [Sue-ka ece
yeki-ey
Tom-Nom yesterday night New York-at
Sue-Nom yesterday here-at
wassta-ko malhayssta] ].
came-C said
8
Lit. ‘Tom said in New York last night that Sue came here yesterday.’
a.‘here’ (= Boston), ‘yesterday’ (= July 2nd )
b.‘here’ (= New York), ‘yesterday’ (= July 1st )
c.*‘here’ (= New York), ‘yesterday’ (= July 2nd )
d.*‘here’ (= Boston), ‘yesterday’ (= July 1st )
(No Shift)
(Both Shift)
(Location Shift)
(Time Shift)
Both in (21) and (22), the interpretations where one indexical picks up reference from the context of
utterance and the other from the reported context are excluded among the four logically possible
readings (c-d). It is only possible for the indexicals of same class to shift together or not to shift.
2.3.2
Obligatory de se interpretation
It has been proposed that shifted indexicals receive obligatory de se readings in many languages,
e.g. 1st and 2nd person pronouns in Amharic, both 1st /2nd person pronouns and temporal/locative
indexicals in Zazaki, etc (Schlenker 1999, 2003, Anand 2006).8
In Korean, as in Zazaki, both the person and adverbial shifted indexicals receive obligatory de
se interpretations. As shown in (23)−(26) below, the shifted interpretations of nay ‘I’, ne ‘you’, yeki
‘here’, and onul ‘today’ are infelicitous under the non-de se(/de te/de nunc/de hic) contexts.
(23)
1st person pronoun
S1: John says, "I am the smartest."
S2: John took an exam, and later saw the top 10 scorers with the respective ID numbers.
He forgot his own ID number, so didn’t know who is who. Pointing to the top score, he
remarked "This guy is the smartest!" But it turned out that he was talking about himself.
(context from Sudo (2012))
John-i
nay-ka ceyil ttokttokhata-ko malhayssta.
John-Nom I-Nom most smart-C
said
‘Johni said that hei is the smartest.’ [S1, #S2]
(24)
2nd person pronoun
S1: John says to Tom, "You should leave."
S2: John is hosting a party. He hears that a certain waiter named Tom is being a nuisance.
John tells the nearest waiter, "Tom should go home." Unbeknownst to him, he’s talking to
Tom.
(context from Sudo (2012))
John-i
Tom-eykey ne-ka
ttena-ya hanta-ko malhayssta.
John-Nom Tom-to
you-Nom leave
shoud-C said
‘John said to Tomi that hei should leave.’ [S1, #S2]
(25)
Locative adverbial
S1: John says in Seoul, "Mary was born here."
S2: John visited his friends in Seoul, and he and his friends were looking at old photos of
various cities. John pointed at an old photo of a city, and he said "I recognize this building
in this photo! Mary was born in this city.", without knowing that the city he pointed at was
actually the same city as he was in.
Sudo (2012) shows that the shifted 2nd person pronoun does not obligatorily get de se (or de te) reading in Uyghur.
However, he argues that the optional de te reading of the 2nd person pronoun is due to the fact that the 2nd person
pronoun is a definite description rather than an indexical. Deal (2014), on the other hand, argues that the locative
adverbial in Nez Perce, which is a pure indexical, does not have to be interpreted de se unlike the person indexicals.
8
9
John-i
Seoul-eyse Mary-ka yeki-eyse thayenassta-ko malhayssta.
John-Nom Seoul-at Mary-Nom here-at be.born-C
said
‘John said in Seouli that Mary was born therei .’ [S1, #S2]
(26)
Temporal adverbial
S1: Last Monday John said, "Mary is leaving today."
S2: John knew that Mary was going to leave on Monday last week. On that day, John
somehow thought it was Sunday rather than Monday, and said "Mary leaves tomorrow, on
Monday."
John-i
cinan.cwu welyoil-ey Mary-Nom onul ttenanta-ko malhayssta.
John-Nom last.week Monday-on Mary-Nom today leave-C
said
‘John said last Mondayi that Mary would leave that dayi .’ [S1, #S2]
When we replace the first or second person pronoun with the third person pronoun ku in (23)
and (24), the sentence is felicitous both under the de se (S1) and non-de se (S2) context. Unlike the
third person pronoun, the shifted first or second person pronouns can only be used in the contexts
where the attitude holder identifies the referents of the pronouns as the counterpart of himself or
his addressee. Similarly, the attitude holder must recognize that the time and location he refers to
are the counterparts of the attitude time or location in order for the shifted adverbial indexicals to
be used.
3
Two Context-shift Operators
3.1
Person vs. Adverbial indexicals
Thus far, we have seen that indexical shifting in Korean shares various properties with those that
have been found in other languages, e.g. shift together and obligatory de se interpretation. In this
subsection, I will present some exceptional properties of indexical shifting in Korean. Specifically,
we will see some key contrasts between the person and adverbial indexicals with respect to indexical shifting.
First, while the first and second person pronouns can be shifted only under the predicates of
communication, e.g. ‘say’, ’tell’, ‘claim’, etc., the temporal/locative adverbials are shiftable under
other attitude predicates, e.g. ‘believe’, ‘think’, etc., as well as the speech verbs.
(27)
a.
b.
(28)
Mary-ka nwuka na-lul coahanta-ko malhayss-ni?
Mary-Nom who I-Acc like-C
said-Q
‘Who did Mary say like {me, her}?’
Mary-ka nwuka na-lul coahanta-ko sayngkakhass-ni?
Mary-Nom who I-Acc like-C
thought-Q
‘Who did Mary think likes {me, *her}?’
Context: John asks the following question in Seoul.
a. John: New York-eyse Mary-ka nwuka yeki-eyse thayenassta-ko malhayss-ni?
New York-at Mary-Nom who here-in be.born-C
said-Q
‘Who did Mary say in NY was born in {Seoul, New York}?’
b. John: New York-eyse Mary-ka nwuka yeki-eyse thayenassta-ko sayngkakayss-ni?
New York-at Mary-Nom who here-in be.born-C
think-Q
‘Who did Mary think in NY was born in {Seoul, New York}?’
10
Both in (27) and (28), the sentences in (a) and (b) differ only in the matrix verb: ‘say’ in (a) and
‘think’ in (b). The data above, then, show that the verb ‘say’ allows indexical shifting of both the
person and adverbial indexicals in its complement clause, while only the adverbial indexicals are
shiftable under the verb ‘think’.9
Secondly, we have seen that the shift together constraint holds for Korean when more than
one indexical of the same class occurs in an embedded clause. Given that fact, one might predict
that a person and an adverbial indexical must shift together when they co-occur. Interestingly,
contrary to such prediction, person indexicals and adverbial indexicals do not have to shift together
in Korean. That is, person indexicals can shift independent of adverbial indexicals, and vice versa.
Thus, there is a four-way ambiguity in (29), where we have the first person pronoun nay and the
locative adverbial yeki ‘here’ in the embedded clause.
(29)
Context: John and Mary are having a conversation in Seoul.
John: Tom-i
New York-eyse nay-ka yeki-eyse thayenassta-ko malhayssta.
Tom-Nom New York-at I-Nom here-at be.born-C
said
Lit. ‘Tom said in New York that I was born here.’
(No Shift)
a.‘I’ = John, ‘here’ = Seoul
(Location Shift)
b.‘I’ = John, ‘here’ = New York
(Person Shift)
c.‘I’ = Tom, ‘here’ = Seoul
(Both Shift)
d.‘I’ = Tom, ‘here’ = New York
In fact, a similar pattern was observed in Nez Perce by Deal (2014), illustrated below.10
(30)
Context: My friend is calling me on his cellphone and describing his location. He is trying
to make it to Lapwai, but he is lost.
prosubj hi-hi-ce-∅
[prosubj kíne ∅-paay-ca-∅]
met’u weet’u prosubj
pro
3subj-say-imperf-pres [pro
here 1subj-arrive-imperf-pres] but not
pro
hi-paay-ca-∅
kíne
3subj-arrive-imperf-pres here
colloquial: He says he is arriving here, but he is not arriving here.
literal: Hei says Ii am arriving herek , but hei is not arriving herek .
(Deal 2014)
In (30), only the first person in the embedded clause is shifted so that it refers to the attitude holder,
while the locative kíne ‘here’ refers to the location of the speaker of utterance, Lapwai, instead of the
location of the attitude holder. This similar behavior of person and adverbial indexicals in Korean
9
Besides yeki ‘here’, other temporal adverbial indexicals, such as ece ‘yesterday’, onul ‘today’, nayil ‘tomorrow’, can be
shifted under various attitude predicates.
Conversation on January 8th
(i)
Cinan cwu-ey Mary-ka nwuka
nayil
ttenanta-ko sayngkakayss-ni?
Last week-in Mary-Nom who-Nom tomorrow leave-C
said-Q
‘Who did Mary think a week ago would leave on {January 9th , January 2nd }?’
10
According to Deal, however, there is a three-way ambiguity rather than four-way when a person indexical and
locative adverbial occur together. That is, among the four logically possible readings, the reading where only the locative
indexical shifts is impossible in Nez Perce.
11
and Nez Perce suggests a cross-linguistic variation in indexical shift whereby the two types of
indexicals can shift independently.
Finally, we can find one more contrast between the person and adverbial indexicals with respect
to their interactions with the third person pronoun ku or the long-distance reflexive/logophor caki
in Korean. That is, when occurring in the same clause as the third person pronoun or the longdistance reflexive caki referring to one of the matrix speech participants, person indexicals do not
receive the shifted interpretation, but adverbial indexicals can.11
(31)
Context: John and Mary are having a conversation.
John: Tom-i
Sue-eykey ku/caki-ka ne-lul cohahanta-ko malhayssta.
Tom-Nom Sue-to
he/caki-Nom you-Acc like-C
said.
a.‘Tomi said to Sue that hei likes you (=Mary).’
b.*‘Tomi said to Sue that hei likes you (=Sue).’
(32)
(No Shift)
(Shift)
Context: John and Mary are having a conversation in Seoul.
John: Tom-i
New York-eyse ku/caki-ka yeki-eyse thayenassta-ko malhayssta.
Tom-Nom New York-at he/caki-Nom here-at be.born-C
said
a.‘Tomi said in NY that hei was born here (=Seoul).’
b.‘Tomi said in NY that hei was born here (=New York).’
(No Shift)
(Shift)
Having presented the contrasts between the person and adverbial indexicals, I will discuss how
these properties of indexical shift in Korean should be analyzed in the following subsection.
3.2
Analysis
3.2.1
Background assumptions
Let me first introduce my basic assumptions, which are based upon various key works in the semantics of context and indexicality. Following Kaplan (1989), I will relativize the interpretation
function to a context c, an index (point of evaluation) i, and an assignment function g, [[ . ]]c,i,g . I
assume that both contexts and indices are represented as tuples of coordinates, as <author, hearer,
time, location, world> (Kaplan 1989, Schlenker 1999, a.o.). Based on this, I will also assume that the
extensions of indexicals are insensitive to the evaluation points (indices) but are dependent on the
contexts.
(33)
Lexical entries for indexicals
a. J I Kc,i,g = auth(c)
b. J you Kc,i,g = addr(c)
c. J here Kc,i,g = loc(c)
Moreover, based on the assumption that an index (point of evaluation) is a quintuple <a, h, t,
l, w>, I will also assume that propositions, the arguments to attitude predicates, are sets of such
quintuples. For the semantic composition of attitude predicates and propositions, I will use the
following definition of Intensional Function Application (adopted from Heim and Kratzer (1998)):
11
This contrast will be discussed in detail in Section 4 & 5.
12
(34)
Intensional Function Application
If α is a branching node and {β, γ} the set of its daughters, then, for any context c, index i,
and assignment g: if JβKc,i,g is a function whose domain contains λi’. JγKc,i’,g , then JαKc,i,g =
JβKc,i,g (λi’. JγKc,i’,g ).
In addition to this, I also assume that the parameter i and c are equal at the root level, i = c =
<author*, hearer*, l*, t*, w*> (Stalnaker 1978, Anand 2006).
Having an additional context parameter c on the interpretation function does not affect the
meanings of other expressions other than indexicals, as shown below.
(35)
Sample Lexical Entries
a. JJohnKc,i.g = John
b. JheroKc,i,g = λx. x is a hero in wi at ti at li
c. JsayKc,i,g = λp<κt> . [λx. ∀i0 ∈ Say(x,i): p(i0 ) =T] (to be revised)12
Given the denotations of indexicals and other expressions shown in (33)-(35), we can see that the
first person pronoun under the attitude predicate still refers to the speaker of the context of utterance, as we expect.
(36)
3.2.2
a.
b.
John says that I am a hero.
The Truth-Conditions for (36a):
JJohn says that I am a heroKc,i,g = T iff ∀i’ ∈ Say(John,i): auth(c) is a hero in wi0 at ti0 at li0
Two context-shift operators in Korean
Although Kaplan (1989) made a conjecture that there is no operator that can manipulate the context
parameter, which he calls monsters, indexical shifting data found in other languages show that such
"monsters" exist. Following Anand and Nevins (2004) and Anand (2006), I assume that indexical
shift is the result of a context-shift operator that overwrites the context parameter on the interpretation function with the index. Anand (2006) introduces the following semantics of context-shift
operator, adapted from Stalnaker (1978)’s diagonal operator:
(37)
(38)
JOP∀ Kc,i,g = λχκκt . χ(i)(i)
JOP∀ α Kc,i,g = J α Ki,i,g
The context-shift operator in (37), a function from Kaplanian characters (or propositional concepts
in Stalnaker (1978)) to truth-values, overwrites all the coordinates of the context-parameter with
the index.13 Thus, we would expect that every type of indexical will be shifted all together under
such operator.14 Indeed, the empirical pattern observed in Zazaki, where all indexicals must shift
12
κ is the semantic type of the set of quintuples <a, h, t, l, w>.
The character of an expression α is a function from utterance contexts into intensions (Kaplan 1989). Since I assume
that both the context and index are the set of quintuples <a, h, l, t, w>, they have the same semantic type κ.
14
Notice that the operator in (37) takes a character as its argument rather than a proposition. In order to avoid a type
mismatch, then, Anand adds an additional compositional rule called ‘Monstrous Function Application’, as illustrated
below.
13
(i)
Monstrous Function Application
If α is a branching node and {β, γ} the set of its daughters, then, for any context c, index i, and assignment
0 0
0 0
g: if [[β]]c,i,g is a function whose domain contains λc’.λi’.[[α]]c ,i ,g , then [[α]]c,i,g = [[β]]c,i,g (λc’.λi’.[[α]]c ,i ,g ).
(adapted with minor changes from Anand (2006))
13
together when they shift in the complements to the verb ‘say’, can be accounted for under this
operator approach.
It has also been found that not all but some particular type of indexicals shift in other languages:
only 1st /2nd person indexicals shift under the verb ‘say’ in Amharic, Aghem, Navajo or only the 1st
person pronoun under the verb ‘say’ in Slave, etc (Speas 2000, Schlenker 1999, Anand and Nevins
2004, Anand 2006). To account for the cross-linguistic variation, Anand and Nevins (2004) and
Anand (2006) also provide the following partial context-shift operators that only replace certain
coordiates of the context parameter with the index:
(39)
a.
b.
JOPauth α Kc,i = J α Kj,i , where j=<author(i), addr(c), time(c), loc(c), world(c)>
JOPper α Kc,i = J α Kj,i , where j=<author(i), addr(i), time(c), loc(c), world(c)>
In the languages where the operator shown in (39a) exists, only the 1st person indexical is able to
shift, as in Slave. Similarly, the pattern where the person indexicals shift while adverbial indexicals
never shift, as in Navajo and Amharic, can be explained under (39b).
Recall, however, that person and adverbial indexicals can but do not have to shift together
in Korean. If we assume that there is only one type of context-shift operator that is responsible
for overwriting either all or some of the coordinates of the context parameter, we would not expect person and adverbial indexicals to shift independently in Korean or Nez Perce. Based on the
context-shift operator approach, Deal (2014) proposes two operators for person and locative indexicals for Nez Perce. Along this, I also argue that there are two separate operators, OPP ER and
OPADV , respectively for person and adverbial indexicals in Korean.
(40)
The two context-shift operators
a. J OPP ER K<Ac , Hc , Tc , Lc >,i,g = λχκκt . χ(c(Ac /Ai ,Hc /Hi ) )(i)
b. J OPADV K<Ac , Hc , Tc , Lc >,i,g = λχκκt . χ(c(Tc /Ti ,Lc /Li ) )(i)
(41)
a.
b.
J OPP ER [α] K<Ac , Hc , Tc , Lc >,i,g = JαK<Ai , Hi , Tc , Lc >,i,g
J OPADV [α] K<Ac , Hc , Tc , Lc >,i,g = JαK<Ac , Hc , Ti , Li >,i,g
OPP ER in (41a) only overwrites the author and hearer coordinates of the context parameter with
those of the index parameter, while OPADV in (41b) overwrites the location and time coordinates.
All of the coordinates are overwritten when the two operators co-occur. Therefore, whenever there
is more than one person or adverbial indexical under OPP ER or OPADV , respectively, the indexicals
with the same type must shift together. For example, under the existence of OPP ER in (42), both
the speaker and addressee must be interpreted relative to the reported context.
(42)
Illustrations of Shift Together
a. LF structure:
CP
say
OPP ER
I like you
Following Anand, I will also assume that the ‘Monstrous Function Application’ is required in the derivation to combine
the context-shift operator with a character.
14
b.
Truth-conditions:
J John said [ OPP ER I like you]Kc,i,g = 1 iff
∀i’ ∈ Say(John,i): author(i0 ) likes addr(i0 ) in world(i0 )
On the other hand, the different types of indexicals do not have to shift together if there is
only one type of context-shift operator. The four-way ambiguity shown in (29), repeated below, is
illustrated in (43)− (46).
(29)
Context: John and Mary are having a conversation in Seoul.
John: Tom-i
New York-eyse nay-ka yeki-eyse thayenassta-ko malhayssta.
Tom-Nom New York-at I-Nom here-at be.born-C
said
Lit. ‘Tom said that in New York that I was born here.’
(No Shift)
a.‘I’ = John, ‘here’ = Seoul
(Location Shift)
b.‘I’ = John, ‘here’ = New York
(Person Shift)
c.‘I’ = Tom, ‘here’ = Seoul
(Both Shift)
d.‘I’ = Tom, ‘here’ = New York
(43)
Illustrations of the four-way ambiguity (1)15 − Both shift
a. LF structure:
say
CP
OPP ER
OPADV
I was born here
b.
Truth-conditions:
J John says [ OPP ER OPADV I was born here] Kc,i,g = 1 iff
∀i’ ∈ Say(John,i): author(i0 ) was born in loc(i0 ) in world(i0 )
As we can see in (43), both the 1st person pronoun and the locative adverbial end up shifting
together under the two context-shift operators, OPP ER and OPADV , in the embedded clause. The
T-conditions shown in (43b) confirms this: For all the points of evaluation i0 compatible with what John
says at ti at li in wi are such that author(i0 ) was born in loc(i0 ) in world(i0 ).
(44)
15
Illustrations of the four-way ambiguity (2) - Person shift
a. LF structure:
See the Appendix for the full derivations.
15
say
CP
OPP ER
I was born here
b.
(45)
Truth-conditions:
J John says [ OPP ER I was born here]Kc,i,g = 1 iff
∀i’ ∈ Say(John,i): author(i0 ) was born in loc(c ) in world(i0 )
Illustrations of the four-way ambiguity (3) - Adverbial shift
a. LF structure:
say
CP
OPADV
I was born here
b.
Truth-conditions:
J John says [ OPADV I was born here] Kc,i,g = 1 iff
∀i’ ∈ Say(John,i): author(c ) was born in loc(i0 ) in world(i0 )
With only one type of operator, then, we get the expected T-conditions where only one type of
indexical shifts while the other does not, as in (44) and (45).
(46)
Illustrations of the four-way ambiguity (4) - No shift
a. LF structure:
CP
say
I was born here
b.
Truth-conditions:
J John says I was born hereKc,i,g = 1 iff
∀i’ ∈ Say(John,i): author(c ) was born in loc(c ) in world(i0 )
We see that the optional existences of the two separate context-shift operators in Korean predict the four different interpretations of the sentence with both types of indexicals in the same
embedded clause. Moreover, it also captures the pattern of shift-together observed between the
same type of indexicals. Finally, the contrast between the person and adverbial indexicals in terms
of the type of predicates that allow indexical shifting can be attributed to the lexical properties of
attitude predicates in Korean That is, only the verbs of communication allow either operator in its
complement clause, while other attitude predicates like ‘think’ and ‘believe’ only allow OPADV .
The novel Korean data we observed provide an additional pattern to the cross-linguistic vari16
ation in indexical shifting that has been collected in the previous studies (Anand 2006, Sudo 2012,
Deal 2014).
(47)
Language
Aghem
Amharic
Navajo
Nez Perce
Slave
Uyghur
Zazaki
Korean
3.3
What Shifts
1st & 2nd person
1st & 2nd person
1st & 2nd person
1st & 2nd person and locative
1st person
1st & 2nd person
1st person (& 2nd person)
all indexicals
1st & 2nd person
locative & temporal adverbial
When
say
say
say (want, think)
say, think
say, want, think
tell
all attitude predicates
say
say
all attitude predicates
how
optional
optional
optional
optional
obligatory under say
optional
obligatory
optional
optional
optional
Context-Shift Operators under multiple embeddings
Let us now consider further consequences of the two context-shift operators analysis in Korean.
Given that OPP ER and OPADV can independently occur under an attitude predicate, we would
predict that various indexical shifting would be possible under multiple embeddings, where there
is more than one position in which the context-shift operators can occur. I will show that this
prediction is in fact borne out for Korean.
First, consider the following example with a person indexical both in the intermediate and the
most embedded clauses.
(48)
Tom: [CP 1 John-i
[CP 2 Bill-i
na1 -eykey [CP 3 Mary-ka na2 -lul cohahanta]-ko
John-Nom
Bill-Nom I-to
Mary-Nom I-Acc like-C
malhayssta]-ko malhayssta.]
said-C
said-C
Lit. ‘John said that Bill said to me that Mary likes me.’
a.
b.
c.
d.
na1 = Tom (no shift)
na1 = Tom (no shift)
na1 = John (shift to CP1)
na1 = John (shift to CP1)
na2 = Tom (no shift)
na2 = Bill (shift to CP2)
na2 = John (shift to CP1)
na2 = Bill (shift to CP2)
These four different interpretations in (48) are expected under the following structures, given the
presence/absence and the position of OPP ER .
(49)
a.
b.
c.
d.
[CP 1 John said [CP 2 Bill said to me [CP 3 Mary likes me] ] ]
[CP 1 John said [CP 2 Bill said to me [CP 3 OPP ER Mary likes me] ] ]
[CP 1 John said [CP 2 OPP ER Bill said to me [CP 3 Mary likes me] ] ]
[CP 1 John said [CP 2 OPP ER Bill said to me [CP 3 OPP ER Mary likes me] ] ]
Note that we cannot get a reading where na1 refers to the utterer, while na2 shifts to the matrix
reported context so it is co-referential with John. This is because the presence of OPP ER in CP2 ,
which triggers the shifting of na2 to CP1 , would also force the first person pronoun in CP2 to shift
17
to CP1 .16
Due to the possibility of independent occurrence of OPP ER and OPADV , we get more complicated interpretations of the person and adverbial indexicals under multiple embeddings. Consider
the interpretations of the following example in (50).
(50)
Utterance by Tom in Amherst
[CP 1 John-i
Seoul-eyse [CP 2 Bill-i
yeki-eyse [CP 3 Mary-ka na-lul
John-Nom Seoul-at
Bill-Nom here-at
Mary-Nom I-Acc
cohahanta]-ko malhayssta]-ko malhayssta.]
like-C
said-C
said
Lit. ’John said in Seoul that Bill said here that Mary likes me.’
a.
b.
c.
d.
e.
f.
yeki = Amherst (no shift)
yeki = Amherst (no shift)
yeki =?Amherst (no shift)
yeki = Seoul (shift to CP1)
yeki = Seoul (shift to CP1)
yeki = Seoul (shift to CP1)
na = Tom (no shift)
na = John (shift to CP1)
na = Bill (shift to CP2)
na = ?Tom (no shift)
na = John (shift to CP1)
na = Bill (shift to CP2)
We can get various interpretations of person and adverbial indexicals, as shown in (50).17 These
can also be explained under our account in terms of two separate context-shift operators. Each
interpretation in (50) can be illustrated as in (51).18
(51)
a.
b.
c.
d.
e.
f.
[CP 1 John said in Seoul [CP 2 Bill said here [CP 3 Mary likes me] ] ]
[CP 1 John said in Seoul [CP 2 OPP ER Bill said here [CP 3 Mary likes me] ] ]
[CP 1 John said in Seoul [CP 2 Bill said here [CP 3 OPP ER Mary likes me] ] ]
[CP 1 John said in Seoul [CP 2 OPADV Bill said here [CP 3 Mary likes me] ] ]
[CP 1 John said in Seoul [CP 2 OPP ER OPADV Bill said here [CP 3 Mary likes me] ] ]
[CP 1 John said in Seoul [CP 2 OPADV Bill said here [CP 3 OPP ER Mary likes me] ] ]
As we saw, person and adverbial indexicals can shift independently even under multiple embeddings. Thus, these data provide further evidence in favor of our two context-shift operators analysis
of indexical shift in Korean.
The Long-distance Reflexive caki
4
We have seen that the 1st person pronoun can refer to the attitude holder under attitude predicates
with the presence of the context-shift operator OPP ER . In this section, I will introduce another
element in Korean that can co-refer with the attitude holder, the so-called ‘long-distance reflexive’
caki. In addition to the coreference with the attitude holder, the shifted 1st person pronoun and the
long-distance reflexive caki share the property that they must be interpreted de se.
16
We can see the same pattern with adverbial indexicals under OPADV .
It should be noted, however, that some readings are not as preferred as the others. For now, I have no explanation
why there is such difference in preference.
18
As predicted by our analysis, we also get various interpretations with clausemate person and adverbial indexicals
in the most embedded clause.
17
18
(52)
Mary-ka [John-i
na/caki-lul cohahanta]-ko malhayssta.
Mary-Nom [John-Nom I/caki-Acc like]-C
said
‘Maryi said that John likes heri .’
As illustrated in (52), both na ‘I’ and caki embedded under ‘say’ denote the same reference, the attitude holder Mary. Furthermore, (52) with either na ‘I’ or caki can only be felicitous in the situation
where Mary says, "John likes me."
One might wonder, then, whether the long-distance reflexive (henceforth "LD reflexive") caki
behaves just as the shifted indexicals do, as shown in (52). We will see, however, that caki is another
type of de se element that is to be distinguished from shifted indexicals.
4.1
Background on the LD reflexive caki
Among various facts about the LD reflexive caki in Korean that have been introduced in the previous literature, this subsection lays out a few basic properties that are relevant for our discussion
(Yang 1983, Yoon 1989, Kang 1998, Kim and Yoon 2009, a.o.). Specifically, our discussion will center on caki in a specific environment, namely, under the scope of attitude predicates. We will see
that caki is categorized as a logophor when it is bound by an attitude holder.
Before we turn our attention to caki under attitude predicates, let us first address some basic
properties of caki. The reflexive caki allows both local and long-distance binding. Thus, (53) is
ambiguous: both the matrix subject John and embedded subject Tom can be an antecedent of caki.
(53)
John-i
Tom-i
caki-lul silhehanta-ko sayngkakhanta.
John-Nom Tom-Nom self-Acc dislike-C
think
‘Johni thinks that Tomj dislikes himi /himselfj .’
Regardless of whether it is local-binding or long-distance binding, two general restrictions hold
for antecedents of caki: a) the 1st or 2nd person pronoun cannot be an antecedent of caki, and b)
antecedents of caki must be animate NPs. Given these properties, (54) is unambiguous since only
the 3rd person Tom can be an antecedent of caki. In (55), the embedded subject ‘the local newspaper’
is not qualified as an antecedent of caki due to its inanimacy.
(54)
a.
b.
(55)
Na-nun Tom-i
caki-lul silhehanta-ko sayngkakhanta.
I-Top Tom-Nom self-Acc dislike-C
think
‘I think that Tom dislikes himself/*me.’
Tom-un ne-ka
caki-lul silhehanta-ko sayngkakhanta.
Tom-Top you-Nom self-Acc dislike-Comp think
‘Tomi thinks that you dislike himi /*yourself.’19
John-i
ciyek sinmwun-i
caki-lul piphanhayssta-ko malhayssta.
John-Nom local newspaper-Nom caki-Acc criticize-C
said
‘John said that the local newspaper criticized him.’
Similarly, the reflexive ziji in Chinese can also be bound by a local subject or a long-distance subject. Given that, Huang and Liu (2001) propose that there are two different uses of ziji: a syntactic
anaphor that is subject to the principle of binding, and a logophor that is related to some discourse
roles. For instance, it is an anaphoric ziji when it is bound by the embedded subject Zhangsan in
(56), while it is a logophor when it is bound by the matrix subject Lisi. They argue that only the
This shows that there is no blocking effect by the 1st /2nd person pronoun in Korean, unlike in Chinese, which was
originally observed in Huang (1984).
19
19
logophor ziji does not need to be bound by a local subject as the anaphor ziji does, but rather by a
‘logophoric center’.
(56)
Lisi shuo Zhangsan chang piping ziji.
Lisa say Zhangsan often criticize self
‘Lisii says that Zhangsanj often criticizes himi /himselfj .’
(Huang and Liu 2001)
Huang and Liu (2001) present some special properties of ziji as a logophor, following the previous works on logophoricity. First, the antecedent of the logophor ziji must be a ‘logophoric center’.
Sells (1987) defines the notion of logophoricity with the following three primitive notions.
(57)
source: one who is the intentional agent of the communication
self: one whose mental state or attitude the content of the proposition describes
pivot: one with respect to whose (space-time) location the content of the proposition is
evaluated
(Sells 1987, p.457)
Following Sells (1987), Huang and Liu (2001) show that an antecedent of the logophor ziji can be
characterized as one of these three notions. Thus, the matrix subject in (56), which is a source, is
a proper antecedent of the logophor ziji. Given this, notice that an attitude holder of any attitude
predicates qualifies as an antecedent of the logophor ziji, because it is a ‘logophoric center’: either
a source or self.
Next, they argue that ‘consciousness’ is one of the properties of the logophor ziji. That is, the
logophor ziji can only be used when an antecedent has the conscious knowledge of the event that
is reported. Indeed, the logophor ziji must be interpreted de se under attitude predicates (Pan 1997,
2001, Huang and Liu 2001, Anand 2006).
(58)
S1: Zhangsan says, "That thief stole my purse!"
S2: Zhangsan says, "That thief stole that purse!" (not aware that it was his purse)
Zhangsan shuo pashou
tou-le
ziji-de pibao.
Zhangsan shuo pickpocket steal-Perf ziji
purse
!S1, #S2]
‘Zhangsani said that the pickpocket stole hisi purse.’ [
Anand 2006)
(Huang and Liu 2001,
Under the context S2 in (58), ziji cannot be used to refer to the attitude holder Zangsan, because
the antecedent does not have conscious knowledge of the event that the purse that the thief stole is
actually his own purse. Therefore, the fact that ziji, which co-refers with the attitude holder, must
be interpreted de se shows that ziji is used a logophor in that environment.
Huang and Liu (2001) also present a restriction on the clausemate zijis under multiple embeddings, which was originally presented by Pan (1997): more than one logophor ziji in the same
embedded clause must find the same antecedent.
(59)
[ZS renwei [LS zhidao [WW ba ziji1 de shu song-gei-le ziji2 de pengyou.]]]
ZS think LS know WW BA ziji Gen book gave-to-Perf ziji Gen friend
Lit. ‘ZS thinks that LS knows that WW sent self’s book to self’s friend.’
a. ziji1 = ziji2 = WW or LS or ZS
b. ziji1 = WW, ziji2 = LS or ZS
c. ziji1 = ZS or LS, ziji2 = WW
d. *ziji1 = ZS, ziji2 =LS
e. *ziji1 = LS, ziji2 =ZS
20
Notice in (59) that one of the zijis can co-refer with either attitude holder, while the other ziji is
bound by the local subject WW. However, when both zijis are long-distance bound, they must refer
to the same attitude holder, either ZS or LS. Thus, the interpretations in (59d)-(59e) are impossible.
Based upon these properties of the logophor ziji proposed by Huang and Liu (2001), I will
show that the long-distance reflexive caki within the scope of attitude predicates possesses the
same properties as the logophor ziji. First, caki bound by an attitude holder is a logophor, since
its antecedent is a ‘logophoric center’. Unlike the logophors/LD reflexives are subject-oriented in
many languages, a logophoric center seems to be able to be an antecedent of caki, even though it is
not a subject.
(60)
a.
b.
Johni -i
Maryj -eykey cakii/∗j -ka am-i-lako malhay-ss-ta.
John-Nom Mary-to
caki-Nom cancer-be-C said
‘John said to Mary that he has cancer.’
Johni -i
Maryj -lopwute cakii/j -ka am-i-lako tul-ess-ta.
John-Nom Mary-from
caki-Nom cancer-be-C heard
‘John heard from Mary that he/she has cancer.’
(Yoon 1989)
According to Sells’ (1987) notions, only Mary in (b) but not in (a) is the logophoric center, i.e. source.
The attitude holder John can always be an antecedent of caki, since John is a logohporic center in
either case: source in (a), and self in (b).
Secondly, it is also true in Korean that the LD caki under the attitude predicates is always interpreted de se. In other words, caki that is co-referential with the attitude holder satisfies the ‘consciousness’ requirement as the logophor ziji does. As ziji in (58), caki in (61) cannot be used in the
embedded clause when John does not know that the purse that is stolen by the thief is actually his
own purse.
(61)
S1: John says, "That thief stole my purse!"
S2: John says, "That thief stole that purse!" (not aware that it was his purse)
John-i
somaychiki-ka caki-uy cikap-ul hwumchy-ess-tako malhay-ss-ta.
John-Nom pickpocket-Nom caki-Gen purse-Acc steal-Past-C
say-Past-Decl
!
‘John said that the pickpocket stole his purse.’ [ S1, #S2]
However, as observed in Pollard and Xue (2001), the LD reflexive ziji in Chinese is not obligatorily
interpreted as de se when it occurs in a relative clause.
(62)
S1: Zhangsan can identify Fred, the man who saved his life by "That man saved my life!"
S2: Zhangsan is trapped in a burning building and faints. When he wakes up, he is safely
outside. He thinks he was lucky, but in fact was saved by a passerby.
Zhangsani zai mei you jian-guo jiu-le
ziji ming de na-ge ren
Zhangsan again not have see-PERF save-PERF self life DE that-CL person
‘Zhangsani didn’t see again the personj who saved hisi/j life.’ (
Xue 2001)
!S1, !S2)
(Pollard and
Unlike ziji that is long-distance bound by an attitude holder, ziji inside an relative clause is not
obligatorily de se even when it is long-distance bound by the matrix subject. Also, the long-distance
antecedent Zhangsan is not a logophoric center.
21
As in Chinese, the long-distance caki is felicitous both in de se and non-de se situations when it
occurs in relative clauses. This might tell us that the distance between the antecedent an caki is not
one of the conditions for the logophoric use of caki. I will leave this issue as a future research topic.
(63)
John-i
caki-lul silheha-nun salam-ul mannassta.
John-Nom caki-Acc hate-RC
person-Acc met
‘Johni met the personj who hates himi /himselfj .
Finally, multiple cakis in the same embedded clause under attitude predicates must be coreferential when they refer to an attitude holder, i.e logophoric center. Consider the interpretations
of the following sentence with multiple cakis in the most embedded clause.
(64)
John-i
[Bill-i
[caki-uy emma-ka
caki-lul silhehanta]-ko sayngkakhanta]-ko
John-Nom Bill-Nom caki-Gen mother-Nom caki-Acc hate-C
think-C
malhayssta.
said
a. ‘Johni said that Billj thought that hisi mother hates himi .’
b. ‘Johni said that Billj thought that hisj mother hates himj .’
c. *‘Johni said that Billj thought that hisi mother hates himj .’
d. *‘Johni said that Billj thought that hisj mother hates himi .’
e. ‘Johni said that Billj thought that hisi motherk hates herk .’
f. ‘Johni said that Billj thought that hisj motherk hates herk .’
The readings in (64a)-(64d) show that multiple cakis must find the same antecedent in order to refer
to an attitude predicate, as observed in Chinese (Pan 1997, Huang and Liu 2001). However, we can
observe from (64e)-(64f) that the two cakis can have different antecedents when the second caki is
locally bound. That is, one caki can be locally bound while the other can take a logophoric center
as its antecedent.
To summarize, we observed the following properties of caki within the scope of attitude predicates: (i) it finds a ‘logophoric center’ as its antecedent, (ii) it must be interpreted de se, and (iii)
multiple clausemate logophor cakis must find the same antecedent. Given these properties of caki,
I will assume that caki can behave like a logophor in a certain environment, namely, under attitude
predicates.
4.2
The semantics of the logophor caki
Let us recall that the logophor caki shares some properties with the shifted person indexical, especially the 1st person pronoun. As we saw before, the shifted 1st person pronoun co-refers with
an attitude holder, which can be classified as logophoric centers, and it is always interpreted de se
as well. Furthermore, we also saw that multiple shifted indexicals in the same embedded clause
must find a reference from the same context, i.e. Shift Together Constraint.
In light of these similarities between the logophor caki and shifted first person pronoun, both
Schlenker (2003) and Anand and Nevins (2004) provide a uniform account for logophors and
shifted indexicals with slight differences in their semantics. For instance, Anand and Nevins (2004)
propose the following lexical entires for logophors.
(65)
a.
b.
Jlog-authKc,i,g = auth(i)
Jlog-addrKc,i,g = addr(i)
Anand and Nevins (2004) assume that the semantics of indexicals are dependent on the coordi22
nates of the context parameter, e.g. auth(c), addr(c), etc. Similarly, the semantics of logophors also
denote the author or addressee coordinates of the index parameter, as shown in (65). Given the semantics in (65a), then, the logophor indicates the speaker of the points of evaluation. However, the
semantics in (65) predict that logophors can even occur outside the scope of attitude predicates,
contrary to the property of logophors in many languages that they have to appear under the scope
of attitude predicates. Therefore, they propose the following additional restriction.
(66)
Context blocking: Do not use a logophor when an indexical could be used.
Nevins 2004)
(Anand and
In the matrix clause, what a logophor denotes, auth(i), would be identical with what the 1st person
indexical indicates, auth(c). Given the restriction in (66), then, the indexical should be used instead
of the logophor in an unembedded sentence.
However, there are problems with this unified account. The first problem has to do with the
interpretation of caki under multiple embeddings. Under more than one attitude predicate, the
logophor caki can find its antecedent from either the matrix clause or the intermediate clause, as
illustrated below.
(67)
John-i
Mary-ka caki-ka ttokttokhata-ko sayngkakhanta-ko malhayssta.
John-Nom Mary-Nom caki-Nom smart-C
think-C
said
20
‘Johni said that Maryj thinks selfi/j is smart.’
Given the semantics of the logophor shown in (65), however, it is hard to explain how the meaning of the logophor caki can depend on either point of evaluation under the two intensional operators. It is only expected to be determined by the closest intensional operator. The presence of the
context-shift operator would also be ineffective, because it only overwrites the context parameter
but not the index parameter.
Another problem arises in a sentence where a shifted indexical and a logophor co-occur. In the
previous sections, we have seen that indexical shifting is subject to the so-called ‘Shift Together’
constraint so that more than one indexical in the same clause must find their reference from the
same context. Similarly, we also saw that multiple logophor cakis in the same clause must have
the same antecedent. Assuming shift indexicals and logophors are the same type of elements, the
unified approach predicts that a shifted indexical and a logohpor in the same embedded clause
should not be able to find references from different contexts.
This is, however, a false prediction. First, caki and the shifted first person pronoun can never
be co-referential, unlike clausemate shifted indexicals or LD cakis.21 Rather, they can be disjoint,
as illustrated below.
(68)
[John-i
[Bill-i
[caki-uy emma-ka
na-lul silhehanta]-ko malhayssta]-ko
John-Nom Bill-Nom caki-Gen mother-Nom I-Acc hate-C
said-C
malhayssta.]
said
Lit. ‘Johni said that Billj said that selfj (=Bill)’s mother hates mei (=John)’22
There are two logically possible references of caki and na ‘I’ in (68), namely, the matrix subject John
20
No matter which attitude holder the LD caki takes as its antecedent, caki must be interpreted de se in either case.
In the next section, we will see that this is due to the person feature of caki.
22
Interestingly, the other reading where caki refers to the matrix subject and na ‘I’ the intermediate subject is impossible. Section 5 provides a thorough discussion of this issue.
21
23
and the intermediate subject Bill. In (68), each element can independently find its reference so
that caki refers to the intermediate subject Bill while na ‘I’ refers to the matrix subject John. Such
interpretation is unpredictable if we assume caki is an instance of indexical-type elements. The
uniform approach, thus, is faced with an empirical challenge from the Korean data above.
In contrast to the unified account of the shifted indexicals and logophors, Anand (2006) divides
various de se elements into two types: (i) the semantic type, which is affected by the context-shift
operator, and (ii) the syntactic type, which is bound by a syntactic operator.
(69)
a.
b.
Semantic (context-overwriting): shifted indexicals, Mandarin1 ziji, Malayalam taan
Syntactic (binding by operator): Yoruba oun, English dream-selves, Icelandic sig, Japanese
zibun, Mandarin2 ziji
(Anand 2006, p.11)
Given the empirical data from Korean for shifted indexicals and caki, I will also assume that two
different mechanisms are required for the de se ascriptions by shifted indexicals and the logophor
caki, respectively. For the mechanism that derives indexical shift, I will maintain the contextshifting operator approach, following Anand and Nevins (2004) and Anand (2006). In addition
to this, I will also assume that the logophor caki must be bound by a relevant operator unlike indexicals, which are not bound by any operator but are interpreted relative to the index parameter
under the context-shifting operators.
Let us now discuss the details of binding of the logophor caki by a syntactic operator. First, we
assume that this type of de se element, e.g. the logophor caki, is a variable that bears some special
syntactic feature, e.g. [+log], rather than having lexical entries as indexicals (Maling 1984, Sells
1987, Kratzer 1998, Stechow 2003, a.o.). Note also that caki cannot have the first or second person
pronoun as its antecedent unlike ziji in Chinese. We explain this in terms of inherent third person
feature of caki. That is, caki carries the presupposition as the 3rd person pronoun : it cannot refer
to the author or the hearer of the context c (Schlenker 2003, Heim 2008).
(70)
c,i,g
[[cakin ]]
(
g(n),
if g(n) is not auth(c) or addr(c)
=
undefined, otherwise
Recall also that we treat indices as quintuples <speaker, addressee, time, location, world>, and
the complement clause of an attitude predicate is a function from indices, i.e. quintuples, to truthvalues. We assume, then, that a complement clause contains each abstractor for two individuals
(speaker and addressee), time, location, and world variable in the syntax.23 Following Chierchia
(1989), we assume that the de se element caki, which is a variable with the feature [+log], is bound
by an individual abstractor that is introduced by an attitude predicate (see also Stephenson (2007)
and Pearson (2013)).
Lastly, following the locality conditions for binding a variable proposed in the previous work
(Fox 2000, Anand 2006), we also assume that the variable with [+log] must be bound by a closest
23
(i)
More specifically, we assume the following abstractors:
a.
b.
c.
d.
e.
Jλxn . XPKc,i,g = λx: JXPKc,i
Jλy. XPKc,i,g = λy: JXPKc,i
(auth/x) ,g (n/x)
(addr/y) ,g
c,i(time/t) ,g
Jλt. XPKc,i,g = λt: JXPK
Jλl. XPKc,i,g = λl: JXPKc,i
(loc/l) ,g
Jλw. XPKc,i,g = λw: JXPKc,i
(world/w) ,g
24
binder that also carries the feature [+log]. I will also elaborate the lexical entry for the attitude
predicates for de se interpretation as in (71) (Lewis 1979, Chierchia 1989).
(71)
JsayKc,i,g = λp<κt> . [λx. ∀i0 ∈ Say(x,i): p(i0 ) =T],
where Say(x,i) = {<y,z,l’,t’,w’>}: it is compatible with what x says in w at l at t, and x is y in
w’ at t’ at l’}
(72)
Illustrations of binding of the logophor caki
a. JJohn says [CP1 Mary likes caki1 ] Kc,i,g .
b.
c.
(auth/x,addr/y,time/t,loc/l,world/w) ,g (1/x)
JCP1 Kc,i,g = λx1 λy λt λl λw JMary likes caki1 Kc,i
= λx1 λy λt λl λw Mary likes x at t at l in w.
JsayKc,i,g (JCP1 Kc,i,g )(John) =
∀<x’,y’,t’,l’,w’>∈SayJohn,<x,y,t,l,w> : Mary likes x’ at t’ at l’ in w’
Under our current system, every attitude predicate introduces an individual abstractor that
can bind the logophor caki, whien both the abstractor and caki bear [+log]. Let us examine, then,
whether our account can be extended to the interpretation of the logophor caki under multiple
embeddings. To satisfy the locality restriction, we would predict that caki in the most embedded
clause can only be bound by the closer abstractor so that it is co-referential with the intermediate
subject.
(73)
Predictions
a. [John thinks [. . . λx7 [+log] . . . Tom says [. . . λy8 [+log] . . . Mary likes caki8 [+log] ] ] ]24
b. *[John thinks [. . . λx7 [+log] . . . Tom says [. . . λy8 [+log] . . . Mary likes caki7 [+log] ] ] ]
The LF as in (73a) is predicted, since caki is bound by the closest binder λy8 [+log] , rather than
λx7 [+log] . (73b), on the other hand, is not expected to be legitimate because of the violation of
the locality restriction. In other words, the interpretation where caki is coreferential with the matrix subject cannot be derived under our current account. We have already seen, however, that caki
can take either attitude holder as its antecedent, e.g. John or Tom in (73).
In order for our analysis to correctly predict the data regarding caki under multiple embeddings,
I will alter our assumption that every individual abstractor takes [+log]. Instead, I now assume that
the abstractor takes either [+log] or [−log], although it is still the abstractor with [+log] that can
bind the LD caki.
(74)
New Predictions
a. [John thinks [λx7 [+log] . . . Tom says [λy8 [+log] . . . Mary likes caki8 [+log] ] ] ]
b. [John thinks [λx7 [+log] . . . Tom says [λy8 [−log] . . . Mary likes caki7 [+log] ] ] ]
The logophor caki in (74b) can now be bound by the higher abstractor so that it can refer to the
matrix subject, because the lower abstractor is not qualified as a legitimate binder of caki due to the
feature mismatch.25
It should also be noted that our proposal related to the [±log] feature on the individual abstractor that is responsible for binding caki correctly captures the interactions between multiple
24
Note that other abstractors for world, time, etc. are omitted in the schema for the sake of simplicity.
Rajesh Bhatt (p.c.) suggests an alternate account of the feature match between the abstractor and caki. Instead of
assuming that the logophor caki bears [+log] feature, we could suppose that the feature [log] is unspecified on caki, e.g.
[uLog], and caki can bear [+log] only via agreement with the abstractor with [+log] feature. Given that caki also has a
non-logophoric use as well, this account might be able to be extended to caki in general. In this paper, however, I will
maintain the view that some de se elements, e.g. logophor caki, always bear the feature [+log].
25
25
clausemate cakis. Since the logophor caki must be bound by the closest operator that carries the
[+log] feature, more than one caki in the same clause must be bound by the same abstractor.
(75)
Multiple cakis
a. John said [λj +log
b. *John said [λj +log
c. *John said [λj +log
d. John said [λj +log
Bill said [λk +log
Bill said [λk +log
Bill said [λk +log
Bill said [λk −log
cakik +log ’s mother hates cakik +log ] ]
cakik +log ’s mother hates cakij +log ] ]
cakij +log ’s mother hates cakik +log ] ]
cakij +log ’s mother hates cakij +log ] ]
Neither (75b) nor (75c) is derivable, because the locality condition of binding logophor caki is violated in these cases, namely that one of the cakis, cakij +log , is not bound by the closest binder with
[+log] . As we saw for single caki under multiple embedding, both cakis in the most embedded
clause can take the matrix subject as its antecedent only when the closer binder does not carry
[+log] so that the one in the higher clause can bind them, as shown in (75d).
In summary, we have seen that the logophor caki is a different type of de se element from shifted
indexicals, and it must be bound by a closest individual abstractor with [+log] that is introduced
by an attitude predicate. Under our account, attitude predicates play a crucial role in both binding
the logophor caki and deriving the obligatory de se interpretation of caki within the scope of them.
4.3
Person indexicals and caki
Having provided the separate operators for shifted indexiclas and the LD caki, let us turn our
attention to the unavailability of co-reference between caki and shifted person indexicals. Recall
that a shifted person indexical cannot find its reference from the context where the logophor caki
finds its antecedent. On the other hand, there is no such restriction between a shifted adverbial
indexical and caki. Consider the contrast between (76) and (77) below.
(76)
Context: John and Mary are having a conversation in Seoul.
John: Tom-i
Amherst-eyse caki-ka yeki-eyse thayenassta-ko malhayssta.
Tom-Nom Amherst-at caki-Nom here-at be.born-C
said
‘Tomi said in Amherst that hei was born here (=Seoul, Amherst).’
(77)
Context: John and Mary are having a conversation.
John: Tom-i
Sue-eykey caki-ka ne-lul cohahanta-ko malhayssta.
Tom-Nom Sue-to
caki-Nom you-Acc like-C
said.
‘Tomi said to Sue that hei likes you (=Mary, *Sue).’
Both in (76) and (77), caki in the embedded subject position can only be interpreted as the attitude
holder Tom. While caki co-refers with the attitude holder in the matrix clause, the adverbial indexical yeki ‘here’ in (76) can get a shifted interpretation, which ensures that both operators, OPADV
and the abstractor, exist in the embedded clause. Notice, however, that the person indexical ne
‘you’ in (77) can only refer to the addressee of the utterance context, Mary, but not the addressee
of the reported context, Sue.
Owing to the properties of indexicals, there is a question about why there is such a difference
between person and adverbial indexicals. One might stipulate that only OPADV can co-occur with
the abstractor, while OPP ER is incompatible with it. Under our system, however, the individual
abstractor that can bind the logophor caki is always introduced by an attitude holder. Therefore, the
26
contrast we saw above cannot due to the (in)compatibility between the context-shifting operator
and the abstractor.
Instead, I propose that this somewhat striking contrast between the two types of shifted indexicals and caki has nothing to do with special relationship between the context-shift operator and the
binder of caki. Rather, it is merely due to the person feature of caki. Given our semantics of caki, caki
cannot refer to the speaker or the hearer of the context as the third person pronoun. Suppose, then,
that the context-shift operator OPP ER occurs in the embedded clause under the attitude predicate.
As illustrated below, caki in the embedded clause will end up referring to the speaker of the context
that is overwritten due to the presence of OPP ER . Thus, the whole sentence is undefined due to
the pressuposition of caki.
(78)
a.
b.
c.
John says [CP1 λxn λy λt λl λw OPP ER my mother likes cakin ].
JCP1 Kc,i,g =
(A/A(i’),H/H(i’)) ,i0 ,g (n/x)
λxn λy λt λl λw Jmy mother likes cakin Kc
,
where i0 =i(A/x,H/y,T/t,L/l,W/w)
= λxn λy λt λl λw x0 s mother likes x at t at l in w.
JsayKc,i,g (JCP1 Kc,i,g )(John) =
∀<x’,y’,t’,l’,w’>∈SayJohn,<x,y,t,l,w> : the mother of x0 likes x’ at t’ at l’ in w’
On the other hand, caki can refer to the attitude holder even with the presence of OPADV . Since
the speaker and addressee coordinates of the context parameter are not overwritten under OPADV ,
caki does not refer to the speaker of context, and thus it satisfies the presupposition. This is illustrated in (79).
(79)
a.
b.
c.
John says [CP1 λxn λy λt λl λw OPADV cakin was born here].
JCP1 Kc,i,g =
(T/T(i’),L/L(i’)) ,i0 ,g (n/x)
λxn λy λt λl λw Jcakin was born hereKc
,
0
(A/x,H/y,T/t,L/l,W/w)
where i =i
= λxn λy λt λl λw x was born in l at t in w.
JsayKc,i,g (JCP1 Kc,i,g )(John) =
∀<x’,y’,t’,l’,w’>∈SayJohn,<x,y,t,l,w> : x was born in l’ at t’ in w’
These data show that our analysis of the two context-shift operators and the binder for caki correctly
captures how shifted indexicals and caki behave in the same embedded clause.
5
Interactions between the shifted indexicals and LD reflexive caki
In the previous sections we saw that there are, at least, two de se elements in Korean that are derived
by the two different machineries: shifted indexicals by context-shift operators and the logophor by
an individual abstractor, respectively. According to Anand (2006), where he argues for the syntactic
and semantic operators for different types of de se elements, the two mechanisms are independent
from each other. Thus, in a language in which the two different types of de se elements exist, such
as Korean, we would predict that there is no restriction on the interpretations of the two de se
elements derived by each operator. In other words, given the independence of these operators, an
embedded clause could contain only one operator or both under an attitude predicate.
In this section, we will see that that there is indeed a restriction on the interactions of the two
operators. First, I will present a new blocking effect of the logophor caki caused by shifted indexicals, i.e. ‘IS-Blocking effect’. That is, no context-shift operator can intervene between caki and
its antecedent. I will propose that the ‘IS-blocking effect’ is due to the relationship between the
27
two operators: the [+log] feature on the context-shift operators forces the individual abstractor to
carry the same feature when they co-occur in an embedded clause. Moreover, we will also see
that our account can be extended to the interaction between the third person pronoun with de se
interpretation and shifted indexicals.
5.1
A New Blocking Effect: IS (Indexical Shift)-Blocking Effect
In this subsection, I will introduce an interesting interaction between shifted indexicals and the
logophor caki in Korean. Assuming that there is an abstractor with [+log] that is responsible for
the binding as well as the de se interpretation of the logophor caki, we would predict that the abstractor and the context-shift operators do not restrict each other’s occurrence in the embedded
clauses. Then, various interpretations of caki and shifted indexicals are expected under multiple
embeddings. For example, caki and a shifted indexical in the same embedded clause might get
their reference from different contexts, given the positions of the context-shift operator and the
abstractor that bears [+log].
(80)
Some expected schemata
a.
[+log]
[CP 1 N P1 ...[CP 2 λj
. . . N P2 ...[CP 3 OPPER ...cakij ...ind2 ...]]]
b.
[+log]
[CP 1 N P1 ...[CP 2 OPPER . . . N P2 ...[CP 3 λk
...cakik ...ind1 ...]]]
In (80a), caki finds its antecedent in the matrix clause owing to the presence of the abstractor with
[+log] under CP2 , while the person indexical is shifted to the intermediate subject NP2 by the
OPP ER under CP3 . The interpretation shown in (80b) is also expected under the different positions
of the operators, i.e. OPP ER under CP2 and the abstractor with [+log] under CP3 . We would make
the same prediction for the adverbial indexicals with OPADV as well.
Contrary to the prediction, however, we will see that only (80b), but not (80a), is possible in
Korean. First, let’s consider the interpretations of the sentence where a person indexical and caki
co-occur in the most embedded clause under multiple embeddings.
(81)
a.
[CP 1 John-i
[CP 2 OPP ER Bill-i
[CP 3 λk [+log] cakik -uy emma-ka na-lul
John-Nom
Bill-Nom
caki-Gen mom-Nom I-Acc
silhehanta]-ko malhayssta]-ko malhayssta.]
hate-C
said-C
said
Lit. ‘Johni said that Billj said that self (=Bill)’s mother hates me (=John)’
b.
*[CP 1 John-i
[CP 2 λj [+log] Bill-i
[CP 3 OPP ER cakij -uy emma-ka na-lul
John-Nom
Bill-Nom
caki-Gen mom-Nom I-Acc
silhehanta]-ko malhayssta]-ko malhayssta.]
hate-C
said-C
said
Lit. *‘Johni said that Billj said that self (=John)’s mother hates me (=Bill)’
In (81a), when the 1st person pronoun is interpreted as John, entailing the presence of OPP ER at
CP2 , caki can have Bill as its antecedent. However, in (81b), if the 1st person pronoun is interpreted
as Bill, entailing the presence of OPP ER at CP3 , then caki cannot have John as its antecedent.
The same contrast is shown with the shifted adverbial indexical and caki, too, as in (82a)-(82b).
28
(82)
a.
[CP 1 Seoul-eyse John-i
[CP 2 OPADV Bill-i
Amherst-eyse [CP 3 λk [+log]
Seoul-in John-Nom
Bill-Nom Amherst-in
cakik -ka yeki-se thayenassta]-ko malhayssta]-ko malhayssta.]
caki-Nom here-at be.born-C
said
said
Lit. ‘John said in Seoul that Bill said in Amherst that self (=Bill) was born here (=Seoul).’
b.
Amherst-eyse [CP 3 OPADV
*[CP 1 Seoul-eyse John-i
[CP 2 λj [+log] Bill-i
Bill-Nom Amherst-in
Seoul-in John-Nom
cakij -ka yeki-se thayenassta]-ko malhayssta]-ko malhayssta.]
caki-Nom here-at be.born-C
said
said
Lit. ‘John said in Seoul that Bill said in Amherst that self (=*John) was born here
(=Amherst).’
Again, in (82a), where the context-shift operator occurs in the higher clause, caki and yeki ‘here’ can
find their reference from different contexts.26 The presence of OPADV in CP3 , which intervenes
between the abstractor with [+log] in CP2 and caki, blocks the long-distance binding of caki.
What we can observe from these data is that the binding of caki is impossible once indexical
shifting occurs in the lower clause. That is, the context-shift operators cannot intervene between
caki and its antecedent, if they are separated by more than one clause boundary. On the basis of
this observation, the interaction between the shifted indexicals and caki can be described and be
schematized, as in (83).
(83)
IS-Blocking Effect
If caki and its antecedent are separated by more than one clause, a context-shift operator
cannot intervene between them.
X
[+log]
∗[CP 1 A1 ...[CP 2 λj
B2 ...[CP 3 OPPER/ADV ...cakij ...ind2 ...]]]
If we assume that the abstractor with [+log] and context-shift operators are independent from
each other, this one-way blocking effect between the shifted indexicals and caki is not expected to
occur. How can we, then, account for this interesting interaction between the two elements?
5.2
Analysis
Let me first present, again, the basic assumptions that I adopt from various studies. As I have
shown in the previous sections, I will assume that there are separate routes to de se of shifted indexicals and that of the LD reflexive (Anand 2006). That is, the shifted indexicals are de se elements
that are derived by the existence of context-shift operators (Anand and Nevins 2004, Anand 2006),
while the long-distance reflexive is a de se element that is bound by an individual abstract with
[+log] within the scope of the attitude predicate. I will also adopt that the de se elements like caki
always bear the syntactic feature, e.g. [+log]. The de se elements that bear [+log] must be bound by
the closest operator that also takes the [+log] feature. I also presented a new assumption that an
abstractor introduced by attitude predicates can take either [+log] or [−log] feature.
Based on these assumptions, I propose new assumptions in order to account for the interaction
between the shifted indexicals and caki. First, the context-shift operators and indexicals always
bear [+log], as the obligatory de se element caki does. Secondly, whenever the abstractor and the
26
As we saw, caki can also refer to John since the OPADV does not manipulate the speaker or hearer coordinates of the
context parameter (Section 4.3).
29
context-shift operator occur in the same embedded clause, they must agree in the feature [log].
In other words, the individual abstractor introduced by an attitude predicate must carry [+log]
feature whenever a context-shift operator occurs in the same clause.
Given the assumptions made above, we can now derive the ‘IS-Blocking Effect’. In (84), the abstractor in CP3 obligatorily takes [+log] because it co-occurs with the context-shift operator. Consequently, the abstractor in CP3 must bind caki as it fulfills the requirements of binding caki: the
locality restriction and the feature match. Therefore, the blocking effect occurs in (84), since caki
is not bound by the closest binder with [+log], i.e. λk +log , but by the one in the higher clause, i.e.
λj +log .
(84)
Deriving the IS-Blocking effect
0
∗[CP 1 John said [CP 2 λj [+log] Billi said[CP 3 λk [+log] OP PER [+log] cakij [+log] s mother hates mei ]]]
X
On the other hand, shifted indexicals and caki can find references from different contexts with
the absence of the context-shift operator in the most embedded clause. In (85), where the contextshift operator is posited in CP2 , caki can refer to the intermediate subject Bill by being bound by
the closest binder λk +log , while the 1st person pronoun is interpreted as the matrix subject John.
Notice that the abstractor with relevant feature in CP2 , λj +log , cannot compete with λk +log because
of the locality condition.
(85)
No Blocking effect
0
[CP 1 Johnl said [CP 2 λj [+log] OP PER [+log] Billi said[CP 3 λk [+log] cakik [+log] s mother hates mel ]]]
X
Because both OPP ER and OPADV obligatorily take [+log] feature, and we naturally find the
same blocking effect between adverbial indexicals and caki, as below.
(86)
a.
b.
IS-Blocking Effect with OPADV
*[CP 1 John said in Seoul [CP 2 λj +log Bill said in Amhersti [CP 3 λk +log OPADV +log cakij +log
was born herei ] ] ]
No Blocking Effect with OPADV
[CP 1 John said in Seoull [CP 2 λj +log OPADV +log Bill said in Amherst [CP 3 λk +log cakik +log ’s
mother hates mel ] ] ]
In summary, the two different de se elements exhibit the IS-blocking effect in their interactions.
I proposed that this is due to the feature match between the context-shift operators, which always
takes [+log], and the individual abstractor. The core insight behind this account is that the two
mechanisms interact with each other rather than be independent, given the same property that
both of them derive de se.
5.3
Third person pronoun as a de se element
Lastly, I will point out one further interaction between shifted indexicals and the 3rd person pronoun. As in English, the 3rd person pronoun in Korean can get a de se or (non-de se) de re interpretation under an attitude predicate. Thus, the example in (87) can be used either when John has a de
se belief or not about the individual expressed by the third person pronoun.
30
(87)
John-un ku-ka chencayla-ko sayngkakhanta.
John-Top he-Nom genius.be-C think
‘Johni thinks that hei is a genius.’
It has been controversially discussed in the literature whether the de se interpretation of the third
person pronoun is derived by dedicated machinery or as a special case of de re (Chierchia 1989,
Percus and Sauerland 2003, Maier 2011, a.o.). In this subsection, I will provide new evidence in
favor of the dedicated mechanism for the de se reading of the third person pronoun (Percus and
Sauerland 2003).
Under our account, the logophor caki is a variable that bears the [+log ] feature, and it has
to be bound by the individual abstractor, which is introduced by an attitude predicate, with the
same [+log] feature. Moreover, caki carries the presupposition that it cannot refer to a speaker or
hearer of the context. Given this, we can extend our account to other elements that bear the [+log]
feature. In other words, if there was a variable with [+log] feature other than caki, we would be
able to observe similar properties and restrictions in terms of the de se interpretation.
Let us suppose, then, the third person pronoun optionally takes [+log] feature, and the third
person pronoun with [+log] must be bound by the closest binder that also carries the [+log] feature.
Given the same mechanism, we would expect that the third person pronoun would also be subject
to the ‘IS blocking effect’ under multiple embeddings with clausemate shifted indexicals. This
is indeed borne out in Korean. Consider the example in (88), where caki is replaced by the third
person pronoun from the example of ‘IS blocking effect’.
(88)
[John-i
[Bill-i
[ku-uy emma-ka na-lul silhehanta]-ko malhayssta]-ko malhayssta.
John-Nom Bill-Nom he-Gen mom-Nom I-Acc hate-C
said-C
said
‘Johni said that Billj said that hisi mother hates me (=Bill).’
(de re reading only)
As we saw in the previous section, caki is unable to refer to the matrix subject John when the clausemate na ‘I’ shifts to the intermediate subject Bill. Note that the 3rd person pronoun is seemingly
different from caki in that it can refer to the matrix subject, as shown in (88). Interestingly, however,
the 3rd person pronoun must be interpreted as (non-de se) de re in (88), even though the 3rd person pronoun can usually be interpreted either de se or de re. Thus, (88) is infelicitous in a situation
where John says: "Bill said that my mother hates him." Rather, it is only felicitous when John says:
"Bill said that that guy’s mother hates him.", and unbeknownst to John, he is that guy.
I argue that this is another case in which the ‘IS-blocking effect’ occurs, so it can be explained by
the same analysis we have for caki: the de se 3rd person pronoun also carries [+log] as the logophor
caki, and it must be bound by the closest binder with [+log] under attitude predicates. Owing to
the same reason, namely, the feature match between the context-shift operator and the binder, the
third person pronoun with [+log] cannot be bound by another binder in the higher position, as
shown in (89).
(89)
The IS-blocking effect of the de se 3rd person pronoun
*John said [λj +log Billi said [λk +log OPP ER +log hej +log ’s mother hates mei ] ]
Unlike the logophor caki that must be interpreted de se by this particular mechanism we introduced,
the third person pronoun can be interpreted (non-de se) de re under different mechanism than that
is responsible for de se interpretation. In other words, in cases where the third person pronoun
does not bear [+log] feature, it is subject to another machinery for its de re interpretation. These
Korean data argue for, then, the analysis of the dedicated LF for de se pronouns, which is distinct
31
from the one for de re reading (Percus and Sauerland 2003).27
Having shown that the de se third person pronoun is the same type of de se element as the
logophor, let us address another similarity between caki and the third person pronoun due to their
person feature.
(90)
[[hen ]]c,i,g
(
g(n),
=
undefined,
if g(n) is a single male and is not auth(c) or addr(c)
otherwise
Given that caki and the third person pronoun carry the same presupposition, we would also predict
that person and adverbial indexicals interact with the third person pronoun in a different way, as
with caki, under a single embedded clause. We indeed saw such examples earlier.
Recall that the 3rd person pronoun cannot refer to the author or hearer of the reported context
when any person indexical in the same embedded clause is shifted. Consider, again, the example
(31), which is repeated below.
(31)
Context: John and Mary are having a conversation.
John: Tom-i
Sue-eykey ku-ka ne-lul cohahanta-ko malhayssta.
Tom-Nom Sue-to
he-Nom you-Acc like-C
said.
a. ‘Tomi said to Sue that hei likes you (=Mary).’
b. *‘Tomi said to Sue that hei likes you (=Sue).’
(No Shift)
(Shift)
In (31), the 3rd person pronoun ku can co-refer with the attitude holder Tom when the 2nd person
pronoun ne in the embedded clause refers to the addressee of the context of the utterance, Mary.
However, when ne ‘you’ gets the shifted interpretation, Sue, then ku ‘he’ must refer to someone else
but not Tom, who is the speaker of the reported context.
On the other hand, there is no restriction for the 3rd person pronoun when it co-occurs with a
shifted adverbial indexical. That is, the 3rd person pronoun can refer to the author or hearer of the
reported context even when an adverbial indexical is shifted. Thus, in (32), repeated below, both
the adverbial indexical yeki ‘here’ and the 3rd person pronoun ku ‘he’ can find their referent from
the same context, i.e. the reported context.
(32)
Context: John and Mary are having a conversation in Seoul.
John: Tom-i
New York-eyse ku-ka yeki-eyse thayenassta-ko malhayssta.
Tom-Nom New York-at he-Nom here-at be.born-C
said
a.‘Tomi said in NY that hei was born here (=Seoul).’
b.‘Tomi said in NY that hei was born here (=New York).’
(No Shift)
(Shift)
This contrast can be explained under the same account we introduced for caki. The third person
pronoun can only be defined under OPADV , but not under OPP ER , in (31)-(32), because only ‘he’
in (31) ends up referring to the speaker of the overwritten context.
Before we leave this section, let us consider one potential problem for our account. Note that
we argued there is a dedicated LF for the de se third person pronoun, that is distinct from de re.
27
The analysis of the mechanism for de re interpretation of the third person pronoun is beyond the scope of this paper.
It should be noted, however, that we need some mechanism that blocks a de se interpretation from a de re interpretation,
which might be problematic under the view that a de se reading is always derivable from a de re reading. Otherwise, we
should still be able to derive a de se reading in(89) as well as the de re reading.
32
Given that, then, one might predict that we would still get a de re interpretation, instead of de se,
when ‘he’ and a shifted person indexical co-occur.
(91)
John says [. . . λn[+log] . . . OPP ER . . . my mother likes hej ].
In the schema as in (91), ‘he’ is not a de se element with [+log], so it is not bound by the individual
abstractor with [+log] that is introduced by the attitude predicate. Under a distinct mechanism for
the de re third person pronoun, ‘he’ should be able to co-refer with the attitude holder with a de re
interpretation. However, unlike the prediction, even the de re reading of ku ‘he’ in (31) is difficult.
We will have to leave this issue to future study.
To summarize, we have seen that the de se third person pronoun behaves like the logophor caki
under multiple embedding because of the same mechanism that they are subject to for the de se
interpretation. This led us to conclude that the de se third person pronoun is the same type of de se
element as the logophor caki.
6
Conclusions
Having established that Korean is a language in which indexicals can optionally shift under certain
attitude predicates, I have argued that there are two different context-shift operators for person and
adverbial indexicals in Korean to account for the different properties of the two types of indexicals.
I also proposed that another type of de se element in Korean, the logophor caki, must be bound by
the individual abstractor that is introduced by an attitude predicate.
I have also introduced a new blocking effect of the logophor caki as well as the de se third person pronoun that is caused by the interaction between context-shift operators and the individual
abstractor. The interactions between shifted indexicals and other de se elements shown in Korean
shed some light on the general understanding of de se ascription: there are indeed different routes
to de se of shifted indexicals and other de se elements, but the two mechanisms interact with each
other rather than behave independently, given the same property that both of them derive de se.
Furthermore, the Korean data provide evidence in favor of the analysis of the dedicated LF for de
se pronouns, which is distinct from the one for de re (Percus and Sauerland 2003).
References
Anand, Pranav. 2006. De de se. Doctoral Dissertation, Massachusetts Institute of Technology.
Anand, Pranav, and Andrew Nevins. 2004. Shifty operators in changing contexts. In Proceedings of
SALT XIV, ed. Robert B. Young, 20–37. Cornell University, Ithaca, NY: CLC Publications.
Chierchia, Gennaro. 1989. Anaphora and attitude de se. In Semantics and contextual expression, ed.
Renatte Bartsch, Johan van Benthem, and Boas van Embde, 1–32. Dordrecht: Foris.
Deal, Amy Rose. 2014. Nez perce embedded indexicals. In Proceedings of SULA 7, ed. H. Greene.
Amherst: GLSA.
Fox, Danny. 2000. Economy and semantic interpretation. Cambridge, Massachusetts: MIT Press.
Heim, Irene. 2008. Features on bound pronouns. In Phi theory, ed. Daniel Harbour, David Adger,
and Susana Béjar, 35–56. Oxford: Oxford University Press.
Heim, Irene, and Angelika Kratzer. 1998. Semantics in generative grammar. Malden, MA: Blackwell.
33
Huang, C.-T. James, and C.-S. Luther Liu. 2001. Logophoricity, Attitudes and ziji at the Interface. In
Syntax and semantics: Long distance reflexives, ed. Peter Cole, Gabriella Hermon, and C.-T. James
Huang, 150–195. Academic Press.
Huang, Yun-Hua. 1984. Chinese reflexives. Studies in English Literature and Linguistics 10:163–188.
Kang, Beom-Mo. 1998. Grammar and the use of language: Korean reflexives ‘caki’, ‘casin’, and
‘caki-casin’. Kwukehak 31:165–204.
Kaplan, David. 1989. Demonstratives: An essay on the semantics, logic, metaphysics, and epistemology of demonstratives and other indexicals. In Themes from kaplan, ed. Joseph Almog, John
Perry, and Howard Wettstein, 481–564. Oxford: Oxford University Press.
Kim, Ji-Hye, and James H. Yoon. 2009. Long-distance bound local anaphors in Korean - An empirical study of the Korean anaphor caki-casin. Lingua 119:733–755.
Kratzer, Angelika. 1998. More structural analogies between pronouns and tense. In Proceedings
of SALT VIII, ed. Devon Strolovitch and Aaron Lawson, 92–110. Cornell University, Ithaca, NY:
CLC Publications.
Lewis, David. 1979. Attitudes de dicto and de se. Philosophical Review 88:513–543.
Maier, Emar. 2007. Quotation marks as monsters, or the other way around? In Proceedings of the
16th Amsterdam Colloquium, ed. Maria Aloni and Floris Roelofsen, 145–150.
Maier, Emar. 2008. Japanese reported speech: against a direct-indirect distinction. In Proceedings
of LENLS.
Maier, Emar. 2011. On the roads to de se. In Proceedings of SALT 21, ed. Anca Chereches Neil Ashton
and David Lutz, 392–412. Ithaca: CLC Publications.
Maling, Joan. 1984. Non-Clause-Bounded Reflexives in Modern Icelandic. Linguistics and Philosophy
7:211–241.
Pan, Haihua. 1997. Constraints on reflexivization in mandarin chinese. New York: Garland Publishing,
Inc.
Pan, Haihua. 2001. Why the blocking effect? In Syntax and semantics vol. 33: Long distance reflexives,
ed. Peter Cole, C.-T. James Huang, and G. Hermon, 279–316. New York: Academic Press.
Partee, Barbara. 1989. Binding implicit variables in quantified contexts. In Papers from the 25th
Regional Meeting of the Chicago Linguistics Society, Part One, ed. Caroline R. Wiltshire, Randolph
Graczyk, and Bradley Music, 342–365. Chicago Linguistics Society, Chicago Illinois: University
of Chicago.
Pearson, Hazel. 2013. The sense of self. Doctoral Dissertation, Harvard University.
Percus, Orin, and Uli Sauerland. 2003. On the LFs of Attitude Reports. In Proceedings of Sinn und
Bedeutung 7, ed. M. Weisgerber. Konstanz: Universität Konstanz.
Pollard, Carl, and Ping Xue. 2001. Syntactic and nonsyntactic constraints on long-distance reflexives. In Syntax and semantics vol. 33: Long distance reflexives, ed. Peter Cole, C.-T. James Huang,
and Gabriella Hermon, 279–316. New York: Academic Press.
34
Rullmann, Hotze. 2004. First and second person pronouns as bound variables. Linguistic Inquiry
35:159–168.
Schlenker, Philippe. 1999. Propositional attitudes and indexicality. Doctoral Dissertation, Massachusetts Institute of Technology.
Schlenker, Philippe. 2003. A Plea for Monsters. Linguistics and Philosophy 26:29–120.
Sells, Peter. 1987. Aspects of logophoricity. Linguistic Inquiry 18:445–480.
Speas, Margaret. 2000. Person and point of view in navajo. In WCCFL Papers in Honor of Ken Hale,
ed. Eloise Jelinek. Cambridge, Massachusetts: MIT Press.
Stalnaker, Robert. 1978. Assertion. Syntax and Semantics 9:315–332; reprinted in Stalnaker (1999:78–
95).
Stechow, Arnim von. 2003. Feature deletion under semantic binding. In Proceedings of the North
East Linguistics Society, ed. Makoto Kadowaki and Shigeto Kawahara, volume 33, 377–403. Graduate Linguistic Student Association, Amherst, Massachusetts: University of Massachusetts at
Amherst.
Stephenson, Tamina. 2007. Judge dependence, epistemic modals, and predicates of personal taste.
Linguistics and Philosophy 30:487–525.
Sudo, Yasutada. 2012. On the semantics of phi features on pronouns. Doctoral Dissertation, Massachusetts Institute of Technology.
Yang, Dong-Whee. 1983. The extended binding theory of anaphors. Language Research 19:169–192.
Yoon, Jeong-Me. 1989. Long-distance anaphors in korean and their crosslinguistic implications.
Chicago Linguistic Society 25:479–495.
Appendix 1
(92)
Illustrations of the four-way ambiguity (1) - Both shift
a. John said [ OPP ER OPADV I was born here].
b.
Derivations of the Truth-conditions of (92a):
J John says [ OPP ER OPADV I was born here] Kc,i,g = 1 iff
(by FA & Lex.)
c,i,g
i. Jsays [ OPP ER OPADV I was born here]K (John) = 1 iff
(by IFA)
0 ,g
c,i,g
0
c,i
ii. JsayK (λi . J OPP ER OPADV I was born hereK
)(John) = 1 iff
(by MFA)
0 ,g
00 ,i00 ,g
c,i,g
0
c,i
00
00
c
iii. JsayK (λi . JOPP ER K
(λc . λi . JOPADV I was born hereK
))(John) = 1 iff
(by def. of OPP ER )
iv. JsayKc,i,g (λi0 . JOPADV I was born hereK<Ai0 , Hi0 , Tc , Lc >,i’,g )(John) = 1 iff
(by MFA)
c,i,g
0
<A
00
00
<A
0 , Hi0 , Tc , Lc >,i’,g
0 , Hi0 , Tc00 , Lc00 >,i”,g
i
i
v. JsayK (λi . JOPADV K
(λc . λi . JI was born hereK
))(John)
= 1 iff
(by def. of OPADV )
vi. JsayKc,i,g (λi0 . JI was born hereK<Ai0 , Hi0 , Ti0 , Li0 >,i’,g )(John) = 1 iff
(by FA and Lex.)
c,i,g
0
0
0
0
vii. JsayK (λi . auth(i ) was born in loc(i ) in world(i ) )(John) = 1 iff (by Lex.)
viii. ∀i” ∈ Say(John,i): auth(i00 ) was born in loc(i00 ) in world(i00 )
35
(93)
Illustrations of the four-way ambiguity (2) - Person shift
a. John said [ OPP ER I was born here].
b.
(94)
Illustrations of the four-way ambiguity (3) - Adverbial shift
a. John said [ OPADV I was born here].
b.
(95)
Derivations of the Truth-conditions of (93a):
J John says [ OPP ER I was born here] Kc,i,g = 1 iff
(by FA & Lex.)
c,i,g
i. Jsays [ OPP ER I was born here]K (John) = 1 iff
(by IFA)
0 ,g
c,i,g
0
c,i
ii. JsayK (λi . J OPP ER I was born hereK
)(John) = 1 iff
(by MFA)
0 ,g
00 ,i00 ,g
c,i,g
0
c,i
00
00
c
iii. JsayK (λi . JOPP ER K
(λc . λi . JI was born hereK
))(John) = 1 iff
(by def. of OPP ER )
iv. JsayKc,i,g (λi0 . JI was born hereK<Ai0 , Hi0 , Tc , Lc >,i’,g )(John) = 1 iff
(by FA and Lex.)
c,i,g
0
0
0
v. JsayK (λi . auth(i ) was born in loc(c) in world(i ) )(John) = 1 iff
(by Lex.)
00
00
vi. ∀i” ∈ Say(John,i): auth(i ) was born in loc(c) in world(i )
Derivations of the Truth-conditions of (94a):
J John says [ OPADV I was born here] Kc,i,g = 1 iff
(by FA & Lex.)
c,i,g
i. Jsays [ OPADV I was born here]K (John) = 1 iff
(by IFA)
0 ,g
c,i,g
0
c,i
ii. JsayK (λi . J OPADV I was born hereK
)(John) = 1 iff
(by MFA)
0 ,g
00 ,i00 ,g
c,i,g
0
c,i
00
00
c
iii. JsayK (λi . JOPADV K
(λc . λi . JI was born hereK
))(John) = 1 iff
(by def. of OPP ER )
c,i,g
0
<A
,
H
,
T
0 , Li0 >,i’,g
c
c
i
iv. JsayK (λi . JI was born hereK
)(John) = 1 iff
(by FA and Lex.)
c,i,g
0
0
0
v. JsayK (λi . auth(c) was born in loc(i ) in world(i ) )(John) = 1 iff
(by Lex.)
00
00
vi. ∀i” ∈ Say(John,i): auth(c) was born in loc(i ) in world(i )
Illustrations of the four-way ambiguity (4) - No shift
a. John said [ I was born here].
b.
Derivations of the Truth-conditions of (95a):
J John says I was born hereKc,i,g = 1 iff
(by FA & Lex.)
c,i,g
i. Jsays I was born hereK (John) = 1 iff
(by IFA)
0 ,g
c,i,g
0
c,i
ii. JsayK (λi . JI was born hereK
)(John) = 1 iff
(by FA and Lex.)
c,i,g
0
0
iii. JsayK (λi . auth(c) was born in loc(c) in world(i ) )(John) = 1 iff
(by Lex.)
00
iv. ∀i” ∈ Say(John,i): auth(c) was born in loc(c) in world(i )
36