Yosuke 09 Coordination-De
Yosuke 09 Coordination-De
Yosuke 09 Coordination-De
1. Introduction
Ellipsis in natural language has received a significant attention in the linguistic literature since
the proper analysis of this phenomenon provides us with an ideal important window into the way
the various components of grammar mutually interact. Accordingly, various attempts have been
made to identify and elucidate various types of ellipsis (e.g. VP ellipsis, DP/NP ellipsis, sluicing,
generative tradition, deletion and copy theories have been major competitors in the area of
ellipsis since they were first proposed by Ross (1967), Sag (1976) and Williams (1977). The
debate between the two theories are still live until now, though theoretical advances since then
have yielded a variety of new analytical alternatives that incorporate insight from both theories.
The purpose of this paper is to address issues raised by ellipsis concerning linguistic
interfaces through a case study with gapping constructions in Japanese, as illustrated in (1).
∗
I am very grateful to Caroline Heycock and two anonymous JL reviewers for extremely valuable comments on
both the content and structure of an earlier version of this paper. All remaining errors are my own responsibility.
(1) Takesi-ga zassi-o ØV, sosite Kaori-ga hon-o katta.
I propose that gapping in Japanese is best analyzed as the product of the Coordinate and
Dependent Ellipses within the anaphoric theory of deletion presented in Williams (1997) and
further articulated in Ackema and Szendröi (2002). According to this theory, a coordinate
structure arises from the projection of a bivalent lexical item of the form [X, X] P → XP and XP.
William’s theory analyzes English gapping as an instance of the projection of the double-headed
verb, one of whose heads is null, as shown in (2a), and anaphoric to the corresponding element in
the full-fledged conjunct. This process, named Coordinate Ellipsis/CE, yields gapping as in (2b).
b. [give the book to Mary] VP and [0V the record to Sue] VP (Williams 1997: 621)
The CE head, in turn, licenses further ellipsis of its grammatical dependent(s) such as its
complement. This process, dubbed Dependent Ellipsis/DE (Ackema and Szendröi 2002), is
illustrated in (3a, b). ((3a, b) is from Williams 1997: 621; (3c) is from Williams 1997: 623.)
-2-
(3) a. John gave Mary a book today and 0V 0NP a record yesterday.
b. * John gave Mary a book today and gave 0NP a record yesterday.
c. John saw pictures of Mary on Tuesday and [0V [0N of Sue] NP on Wednesday].
The contrast between (3a) and (3b) shows that the DE of the indirect object in the gapped
conjunct is parasitic on the CE of the verb in the full-fledged conjunct. The grammaticality of (3c)
shows that DE may target only the head of the dependent of the 0V (i.e. pictures).
I demonstrate in this paper that the CE/DE theory briefly illustrated above provides a
languages) related to the information structure/pitch accent of overt remnants in gapping, the role of
parallel coordination, and the parallelism constraint. The present analysis also correctly predicts more
subtle properties of gapping, some unique to Japanese and others more crosslinguistically robust (e.g.
homonyms, left branch conditions, etc), that would remain mysterious under earlier analyses of this
This paper is organized as follows. The following section present a detailed exposition of
William’s (1997) anaphoric theory of deletion. Section 3 shows that this theory directly derives core
properties of gapping in Japanese. Section 4 compare our analysis with several existing analyses for
gapping and other elliptic phenomena in Japanese and other languages such as Korean and English.
-3-
The past analyses examined in this section include PF movement analyses (Maling 1972, Kuno 1978,
Saito 1987, Koizumi 2000), LF Copy analyses (Abe and Hoshi 1997), PF Deletion analyses (Fukui
and Sakai 2003; Mukai 2003; Hartmann 2000), Movement + PF Deletion analyses (Jayaseelan 1990,
Sohn 1994; Kim 1997, 1998), and Argument Ellipsis analyses (Oku 1998, Kim 1999, Saito 2003,
Takahashi 2006, 2008). It is shown that these analyses are faced with various empirical problems that
can be successfully resolved under our present analysis of gapping. Section 5 explores the
comparative syntax of gapping in English and Japanese. Section 6 discusses one potential challenge
to the proposed analysis originally pointed out by Abe and Hoshi (1997) and shows that it can be
resolved as a special case of the parallelism constraint at work in real-time processing of gapping
2. Coordinate Ellipsis, Dependent Ellipsis, and the Disanaphora Law: Williams (1997)
As stated briefly in the introductory section, Williams (1997) proposes that a coordinate structure
arises as the projection of a bivalent lexical item of form [X, X] P → XP and XP. CE is the
process whereby one of the two heads of the lexical item remains null, as in (2a). The CE head,
in turn, licenses ellipsis of (the head of) its grammatical dependent(s) such as the verb
proposes the Disanaphora Law. This principle, informally defined in (4a, b) (see Williams 1997:
622 for the original definition and background), is essentially a Gricean-economy principle of
-4-
pronunciation, which requires that an unpronounced element in the incomplete conjunct be
referentially dependent on the structurally parallel element in the full-fledged conjunct whereas
its overt counterpart must not in the same context. This law is illustrated in (5a-c). ((5a, b) are
(5)a. John gave Billi a book today and 0V 0NP a record yesterday.
c. John gave Billi a book today and gave himi a record yesterday.
In (5a), 0NP in the gapped conjunct must refer back to Bill in the first conjunct. In (5b), its overt
counterpart him must be disjoint from Bill. Note the Disanaphora Law applies only when deletion
applies, since the pronoun can refer back to Bill in (5c), where neither CE nor DE has applied.
Subsequent research on deletion has accumulated evidence that this anaphoric theory has
considerable descriptive and explanatory payoffs. Thus, Ackema and Szendröi (2002) show that
subject and object determiner-sharing constructions (McCawley 1993), Lin as illustrated in (6a, b),
-5-
(6)a. Too many Irish setters are named Kelly, too many German shepherds are named Fritz,
b. Bob gave too many magazines to Jessica and too many newspapers to Joanne.
(6a) arises when the CE of the T-head causes the DE of the head of the dependent in its specifier.
Similarly, (6b) arises when the CE of the verb triggers the DE of the head of its object. Hernández
(2007) argues that this analysis captures the fact that gapping satisfies characteristics commonly
the c-command requirement between the antecedent and dependent(s); see Koster 1987; Neeleman and
van de Koot 2002). The same analysis, she notes, allows us to explain why languages like English
allow only forward gapping in terms of Relativized Minimality (Rizzi 1990) (see section 5.2).
typologically different languages such as Japanese and Korean.1 This is the task in the next section.
In this section, I demonstrate that our present theory directly explains general syntactic, semantic, and
prosodic properties of gapping in Japanese and other languages. Though our database in this paper is
1
Hernández (2007) leaves this issue open because no agreement has been reached as to whether (1a) is a case of
backward gapping or right-node-raising. However, I show in section 4 that Japanese has true backward gapping/CE.
-6-
primarily from Japanese, I will introduce data from gapping in Korean (Sohn 1994; Kim 1997, 1998;
Lee 2005) and English, where relevant, to confirm the crosslinguistic validity of the present analysis.
What constituent is coordinated in Japanese gapping? As is well known, the left conjunct of
sosite ‘and’ can never contain tense morphemes, as shown by the ungrammaticality of (7).
Given the independent constraint that the conjuncts of a coordinate structure share the same
categorial status, (7) suggests that the gapped counterpart in (1) involves vP/VP-coordination.
One might counter, based on (8), that gapping should rather involve TP-coordination because
temporally conflicting adjuncts, kinoo ‘yesterday’ and kyoo ‘today’, occur in the conjuncts.
(8) [vP John-ga kinoo ringo-o] sosite [vP Tom-ga kyoo banana-o tabe]-ru.
‘Lit. John (ate) an apple yesterday, and Tom eats a banana today.’
-7-
As Hirata (2006: 87) points out, this reasoning is based on the assumption that T-like heads must
license temporal modifiers. This assumption is questionable in light of examples as in (9), where
the temporal modifier kinoo ‘yesterday’ can occur despite the lack of any T-like element in it.
(9) kinoo-no-sinbun
yesterday-Gen-newspaper
In fact, there is independent evidence for the vP/VP-coordination analysis.2 Oerhle (1987) (see also
Siegel 1987) observe that this analysis is supported from the relative scope of negation in (10a).
c. Mrs. J can’t live in Boston or Mr. J. can’t live in L.A. (Oehrle 1987: 205)
Suppose that (10a) involves TP-coordination. If (10a) were derived from (b) by gapping, (10a)
would mean [not p & not q], the meaning of (10b). Oehrle observes that this is not what (10a)
actually means; it means [not (p&q)], which is equivalent to [not p or not q], the meaning of (10c).
2
I thank an anonymous reviewer for brining my attention to this line of argument for the vP/VP-coordination
analysis. What follows is modeled on Lee’s (2005) arguments for the same analysis of Korean gapping.
-8-
This wide-scope reading is hard to explain under the TP coordination analysis, which would
necessarily give rise to the narrow-scope reading due to the TP-level coordination. The vP/VP-
level analysis correctly derives this wide scope reading since under this analysis negation c-
A parallel argument is available in Japanese. (11a) allows wide scope readings of negation.
= It is not the case that John bought apples and Mary bought bananas.’ [not (p & q)]
Facts concerning negative polarity licensing, as illustrated in (12a, b), further confirms our
‘Lit. John (didn’t eat) apples or/and Mary didn’t eat anything.’
-9-
b. John-ga nanimo (sosite) Mary-ga ringo-o tabe-nak-atta.
‘Lit. John (didn’t eat) anything or/and Mary didn’t eat apples.’
(12b) is the crucial case here. The TP coordination analysis incorrectly predicts that (12b) is bad, for
the TP dominating the second conjunct should block the negation from c-commanding the negative
polarity item in the first conjunct. This problem does not arise under the vP/VP-level analysis,
according to which negation c-commands into the first and second conjuncts due to the lack of TP.3
Let us now consider how the CE/DE analysis works with verb gapping in Japanese. The
3
Lee (2005) makes an additional argument for the vP/VP-coordination analysis based on his observation that the honorific
morpheme si, analyzed as the Agr head, never appears in the first conjunct. I won’t construct this argument in Japanese
here since there is huge fluctuation of grammaticality concerning the crucial data he uses to support the relevant analysis.
- 10 -
(13) The Syntactic Derivation of the Gapping Sentence in (1)
TP
VP T
VP (sosite) VP -ta
DP V′ DP V′
Takesi-ga DP 0V Kaori-ga DP V
In this derivation, the verb in the first conjunct remains null as 0V. Due to the Disanaphora Law,
this null verb must be obligatorily referential to the overt verb in the second conjunct.
The proposed analysis provides a natural explanation for core properties associated with
gaping in Japanese (and other languages). First, it has been acknowledged (Sag 1976; Kuno
1976) that the application of gapping is dependent on semantics. Kuno (1976: 310) observes
that “constituents deleted by Gapping must be contextually known. On the other hand, the two
constituents left behind by Gapping necessarily represent new information and, therefore, must
be paired with constituents in the first conjunct that represent new information.” According to
Sag (1976: 192ff), this semantic contrast is realized in the form of focal stress. This
- 11 -
(14)a. Takesi-ga zassi-o, (sosite) (*Takesi-ga) hon-o katta.
This observation can be explained as follows. Due to the Disanaphora Law, a lexical element in
the incomplete conjunct must be disanaphoric to its antecedent in the complement conjunct. This
means that the subject in (14a)/the object in (14b) are obligatorily disjoint from Takesi and zassi
‘magazine’, respectively. This disjointness requirement, thus, assigns an overt element in the
Second, when conjuncts are not identical with respect to the number of remnants and
correspondents or linear order, gapping yields ungrammatical results. This observation has been
(15)a. Takesi-ga LGB-o kayoobi-ni 0V, (sosite) Kaori-ga MP-o suiyoobini yonda.
- 12 -
b. * Takesi-ga LGB-o kayooni-ni 0V, (sosite) Kaori-ga MP-o yonda.
(16)a. Takesi-ga Hisako-ni LGB-o 0V, (sosite) Kaori-ga Masa-ni MP-o kasita.
‘Lit. Takesi (rented) Hisako a copy of the LGB, and Kaori rented Masa a copy of MP.’
‘Lit. Takesi (rented) Hisako a copy of the LGB, and Kaori rented a copy of the MP to Masa.’
This constraint is also explained as a natural consequence of the Disanaphora Law. This law
would not be met if two conjuncts were not parallel with respect to the number or linear order of
remnants and their correspondents. Specifically, in (15b), the disjointness of the overt adjunct
kayoobi ‘on Tuesday’ in the CE clause must be checked against its structural correspondent in
the non-CE clause. However, there is no such overt correspondent in the full conjunct in (15b),
unlike suiyoobini ‘on Wednesday’ in (15a). The absence of this correspondent yields the
ungrammaticality of (15b). (16b) is ungrammatical because the structural position for the two
dative arguments does not match up in the two conjuncts in (16b). Thus, there is no way to check
the disjointness requirement of the dative arguments, in violation of the Disanaphora Law.
- 13 -
Finally, the proposed analysis also provides a straightforward account for the observation
first made by Abe and Hoshi (1997) about P-omission in Japanese gapping. As the contrast
between (17a) and (17b) illustrates, Japanese gapping allows omission of postpositions such as
nituite ‘about’ from the gapped conjunct, unlike English gapping. Sohn (1994) and Kim (1997)
note that the same observation holds for Korean gapping, as shown in (17c).
‘Lit. John (talked) (about) Bill, and Mary talked about Susan. ‘
‘Lit. John (talked) (about) Bill, and Mary talked about Susan. ‘
For the purposes of this paper, I adopt Jayaseelan’s (1990) analysis of gapping. Jayaseelan
- 14 -
(Susan in (17a)) followed by VP ellipsis. This analysis explains the impossibility of P-drop in
(17a) since the Heavy DP Shift is independently known to prohibit P-stranding, as shown in (18).
(18) * John depended ti on yesterday [DP the man who he had seen three days ago]i.
The question, then, is why “P-stranding” is possible in Japanese/Korean gapping. Our analysis
predicts precisely this pattern. It is the result of the DE. Consider the derivation of (17b) in (19).
TP
VP T
VP (sosite) VP -ta
DP V′ DP V′
John-ga PP 0V Mary-ga PP V
DP 0P DP P hanasu
In this derivation, the first head of the bivalent verb remains unpronounced as 0V. This CE head
licenses the head of its complement PP to be null as 0P. Note that this pattern of DE is
- 15 -
independently observed in English gapping, as shown in 20), where the DE applies only to the
(20) John saw pictures of Mary on Tuesday and [0V [0N of Sue] NP on Wednesday]. (Williams 1997: 623)
Note that the postposition nituite/eytayhay ‘about’ cannot be omitted within the non-gapped
conjunct. Our analysis correctly derives this correlation between the verb gapping and P-
An anonymous reviewer asks whether the optionality of P-stranding can be accounted for
and what triggers this operation given that the presence/absence of the DE operation has
coreference/disjointness requirement of this law holds for DPs since they have ability to refer. In
contrast to DPs, postpositions (or case-particles as we see in section 6) lack reference but merely
serve as functional categories that link DPs to their predicators in a given sentence. I maintain,
therefore, that the Disanaphora Law vacuously applies to postpositions without any appreciable
semantic consequences at LF, giving the appearance of a truly optional process. This does not
mean that this functional category does not convey any new information in given sentences. As
Kuno (1972: 272) observes, we have two concepts to distinguish here: one related to lexical
items (anaphoricity) and the other related to discourse-semantic relations that lexical items bear
- 16 -
in sentences. Thus, examples as in (21) show that made ‘to’ in the gapped conjunct cannot be
omitted since it forms a contrastive pair with kara ‘from’ in the full-fledged conjunct and thereby
‘Lit. Taro (ran) to Tokyo station, and Jiro ran from Tokyo station.’
To summarize, I have shown that the anaphoric theory of deletion proposed provides a natural
explanation for core properties associated with gapping in Japanese. In the next section, I show that
this theory also correctly derives more subtle properties that would remain mysterious under earlier
well as other characteristics suggestive of on-line processing constraints. The investigation below
shows that our analysis is particularly suited to capture this multi-modular nature of gapping.
The aim of this section is to compare the proposed analysis with several existing approaches
gapping in Japanese/Korean. I show that these approaches are each faced with a different set of
- 17 -
problems (P-omission, subjacency, LF identity, overgeneration, left branch conditions, etc.)
and demonstrate that they are successfully resolved by the proposed analysis.
4.1. Movement Analyses: Maling (1972), Kuno (1978), Saito (1987), Koizumi (2000)
Following Kuno (1978) (see also Maling 1972), Saito (1987) argue that gapping as in (1) are
analyzed as involving Right-Node-Raising, in the manner seen in (22) (cf. Saito 1987: 320).
TP
TP V1
TP sosite TP katta
DP VP DP VP
Takesi-ga DP t1 kaori-ga DP t1
Raising
As pointed out in the literature on Japanese/Korean gapping (see Abe and Hoshi 1997; Sohn
1994; Kim 1997, 1998; Mukai 2003), however, there are several empirical drawbacks within the
Right-Node-Raising analysis. I mention only two of them for reason of space: see also note 6.
- 18 -
First, it is not obvious whether the structure given in (22) is permissible because the TP (or S in
Saito’s 1987: 320 original representation) is not headed by T. Suppose it is headed by T. Then,
gapping would involve standard TP-coordination. Thus, we would wrongly predict (7) to be
grammatical with katta ‘bought’ in the first conjunct. This problem does not occur in the present
analysis because it assumes that the construction under discussion involves vP-level coordination.
The second potential problem with Saito’s analysis is that it cannot derive the P-drop
pattern observed earlier in (17b, c), where the postposition nituite/eytayhay ‘about’ in the gapped
conjunct is omitted under identity with its second occurrence in the full-fledged conjunct. The
structure for the P-less gapping in (17b) would be as in (23) under Saito’s account.
TP
TP V1
TP sosite TP nituite-hanasu
DP VP DP VP
John-ga PP t1 Bill-ga PP t1
DP t2 DP t2
Raising
- 19 -
In this representation, the shared elements, nituite ‘about’ and hanasu ‘talk’, are moved across the
board to the TP-adjoined position. However, this movement is impossible under the commonly
held assumption (see Bresnan 1974; Hankamer 1971; Reinhart 1991) that the Right Node Raising
(24) * He tried to persuade, but he couldn’t convince, them that he was right. (Bresnan 1974: 615)
As an anonymous reviewer notes, this assumption has been challenged by Wilder (1997) for
based on examples as in (25); see also Hartmann (2000: 57) for examples from German.
(25) Alan looked a word with ten and Adriana looked a word with twenty letters up
However, it is not clear whether the underlined phrases do not really form a syntactic constituent.
For example, it is not technically impossible to imagine a derivation where the phrases in question
are a constituent in (24-25) at a certain stage of syntactic derivation under Larson’s (1988)/Kayne’s
(1994) framework, according to which the more leftward an element appears in linear order, the
higher position it occupies. Indeed, Phillips’ (2003) top-down derivational approach to constituency
destruction predicts that they should form a unitary constituent. This analysis, of course, begs the
- 20 -
question of why (24) is bad. I adopt Abbott’s (1976) claim in this paper that the unacceptability of
(24) has its source in syntax-external processing difficulties or stylistic considerations, though I
leave the exact nature of such extra-grammatical constraints for future investigation.
One might attempt to save Saito’s analysis by reanalysis. Specifically, hanasu ‘talk’ and
nituite ‘about’ are reanalyzed as a derived transitive verb; this verb then undergoes the across-
the-board movement. This attempt fails. If reanalysis were involved in the derivation of the P-
less gapping, the P-drop pattern should not be available with postpositions that cannot undergo
reanalysis with a verb. As Abe and Hoshi note, this prediction is falsified by examples like (26).4
‘Lit. John (was fired) this reason, and Mary was fired for that reason.’
As we saw in section 3.2, our proposed naturally accounts for the P-drop option. The CE head of
the bivalent lexical item licenses DE of the head of its dependent, as shown in (19), on a par with
4
Abe and Hoshi (1997: 111, 112) does not provide an explicit definition of reanalysis in this context. Provided that
they take it that nituite ‘about’ and hanasu ‘talk’ can be reanalyzed while de ‘for (reason)’ and kubininaru ‘fire’
cannot, I take the liberty of assuming that only a pair of a verb and the prepositional head of its complement can
undergo reanalysis. See Hornstein and Weinberg (1981) for much relevant discussion.
- 21 -
As an anonymous reviewer notes, the core argument Saito makes for the Right-Node-
Raising analysis for gapping in Japanese concerns the distribution of null complementizers. Thus,
it is worthwhile to see how the relevant pattern is explained without invoking this operation.
Saito (p. 312) observes that “in some western dialects of Japan, some verbs allows their S′ [=CP]
b. [CP *(that) the teaching was lying] Ben already knew. (Saito 1987: 313)
- 22 -
Saito adopts Stowell’s (1981) ECP-based account, which states that empty categories including null
Cs must be governed by an overt verb. (27a)/(28a) are fine because the null C is governed by the
verb. (27b)/(28b) are bad because the null C is not governed by the verb. Saito then argues, based on
the different behavior of null Cs in English and Japanese, that gapping in Japanese is the product of
Right-Node-Raising rather than V-gapping. Consider (29a-d) and (30a-d) (Saito 1987: 317, 318).
(29)a. John said that we should go to London, and Bill [V e] that we should go to Paris.
(30)a. John-ga Koobe-ni iku te, soide Mary-ga Tookyo-ni iku te, yuuta.
‘John said that he was going to Kobe, and Mary said that she was going to Tokyo.’
- 23 -
The paradigm in (29) illustrates that gapping blocks C-deletion only in the conjunct that it has
applied to though the C in the first conjunct may or may not be deleted. This pattern is naturally
accounted for under the V-gapping analysis if empty verbs are not proper governors (Torrego
1984). Now, if Japanese backward gapping were simply V-gapping as in English, (30b) would
remain mysterious because the null C in the second conjunct is properly governed by the verb
yuuta ‘said’ on a par with (29a) and (30a). Saito argues that this example is naturally accounted for
if the identical verb in both conjuncts undergoes Right-Node-Raising into the TP-adjoined position.
The paradigm in (30a-d), however, receives a principled explanation without evoking Right
Node Raising. Drawing on Pesetsky’s (1992) analysis of the null C as an affix, Bošković and
Lasnik (2003) develop a comprehensive account of the distribution of null Cs in English, according
to which null Cs must undergo Morphological/PF Merger (Halle and Marantz 1993; Bobaljik
1995) to structurally higher [+V] elements to circumvent the Stranded Affix Filter (Lasnik 1981).
(30c, d) are bad because there is no [+V] host for the null C in the gapped conjunct. How about
(30b)? As first observed by Kuno (1973: 10), gapping in Japanese is acceptable when there is an
intonational boundary between the verb in the second conjunct and the dependent that immediately
precedes it. Thus, (1) is acceptable when it is read as in (31a), but not as in (31b).
- 24 -
Bošković (2001) provides independent evidence that intonational boundaries block affixation.
Thus, (30b) is ruled out due to the Stranded Affix Filter because the pause between the null C
and the verb blocks the Morphological/PF Merger of the former to the latter. To the extent that
this analysis is correct, (30a-b) do not necessarily support the Right Node Raising analysis.
Japanese that draws on the string-vacuous overt verb raising. He analyzes (32) as shown in (33)
(33) TP
vP T
vP sosite vP v T
VP tv VP tv ager-
DP V′ DP V′
John-ni DP tV Bob-ni DP tV
ringo-o banana-o
- 25 -
In this derivation, the identical subject Mary undergoes the across-the-board movement into
[Spec, TP]. The identical verb moves overtly through v to T in the across-the-board manner.
Elegant though it may be, Koizumi’s analysis immediately faces the problem of P-stranding in
gapping, as illustrated in (17b); see also Fukui and Sakai (2003) for additional problems with
Koizumi’s analysis. Under his analysis, the P-less gapped conjunct would arguably involve overt
scrambling of the complement of the preposition nituite ‘about’ (Alex and Sue), but this option is
unavailable in Japanese because scrambling does not tolerate scrambling, as shown in (34).
‘Susan1, Mary talked about t1.’ (Abe and Hoshi 1997: 111)
The reanalysis option would also not do since, as we saw above, the P-drop option is available
even with prepositions that are unlikely to be reanalyzed with a verb, as we saw in (26).
Drawing on the data and analysis presented by Jayaseelan (1990), Abe and Hoshi (1997) propose
an alternative LF Copy account of gapping in Japanese that builds on the LF leftward movement
- 26 -
of the remnants in a gapped conjunct and their correspondents in a full-fledged conjunct. Under
Copy at LF
In (35), Sue undergoes leftward LF movement, leaving nituite ‘about’ behind, and adjoins to the I′.
The lower I′ in the second conjunct is then copied at LF onto that of the first conjunct designated here
by e. Their analysis adopts the crucial assumption (Huang 1982; Aoun 1985) that P-stranding is
allowed universally at LF for leftward movement, but not for rightward movement. The P-less
DP Shift) cannot tolerate P-stranding, as shown in (18). Abe and Hoshi’s analysis, thus, correctly
predicts the availability of P-stranding in Japanese gapping and the lack thereof in English gapping.
Abe and Hoshi’s analysis, however, is difficult to sustain for the following reason. 5
Assuming that not only overt movement but also covert/LF movement obeys subjacency
conditions (Nishigauchi 1986, 1990, Choe 1987, Pesetsky 1987, Reinhart 1991), Abe and Hoshi
(p.115) observes that gapping in Japanese exhibits island effects, as shown in (36a, b).
5
See section 6 for another problem with Abe and Hoshi (1997) posed by multiple P/case-omission in Japanese gapping.
- 27 -
(36)a. ?? Harry-ga imiron, sosite Alfonse-ga toogoron-o (relative clause island)
‘John had a meal before he studied math, and Mary English.’ (Abe and Hoshi 1997:115)
This pattern is exactly what is predicted under the LF Copy Analysis. The contrasted elements in
the full conjunct undergo LF movement to create an I′-structure to be copied onto the gapped
conjunct. Consider the LF representations for (36a, b), shown in (37a, b), respectively.
[IP Alfonse-ga [I′ toogoron1 [I′ [NP [IP t2t1-o kenkyuusiteiru] gengogakusya2]-ni atta]]]
[IP Mary-ga [I′ eego1 [I′ [AdvP pro t1-o benkyoosuru mae-ni] syokuzisita]]]
- 28 -
The LF movement of the NP toogoron ‘syntax’ in (37a) and that of the NP eego ‘English’ in
(37b) violates the relative clause island and the adjunct clause island, respectively. However, the
grammaticality judgments reported by Abe and Hoshi are controversial. Mukai (2003) designs a
carefully constructed grammaticality judgment task concerning (38), structurally akin to (36a), and
makes two important discoveries suggesting that subjacency violations are not involved in (36a).
‘Mike saved the man who was being attached by a lion, and Tom a bear.’
First, her result of the survey reveals that 37 out of 43 subjects accepted (38). Second and more
importantly, there was no subject who rejected (38) but accepted (39) (Mukai 2003: 216).
‘Lit. Tom (thought that Bulls would defeat Lakers), and Mike thought Bulls would defeat Nets.’
- 29 -
If the subjacency were at stake in (38) for those subjects who found it unacceptable, then the LF
copy theory would predict that those speakers would judge (39) (without the complex NP
constraint) as grammatical. The total absence of subjects who report this pattern of judgment
suggests that subjacency is irrelevant to (36a)/(38), contrary to Abe and Hoshi’s claim.
Furthermore, Kato (2006) observes that the reduced acceptability of (36a, b) is not due to
movement-induced locality but rather to the omission of case particles. Thus, she (p.50) points
out that “the grammaticality of the sentences improves when the accusative Case-marker of the
object NP is not omitted.” The relevant examples are given in (40a, b) (Kato 2006: 51, 52).
- 30 -
Therefore, Mukai and Kato’s findings indicate that the LF movement analysis is difficult to
sustain.6 Our analysis, however, provides correctly predicts the lack of subjacency violations in
(36a, b)/(38a, b) because focused overt elements in the gapping clause do not need to move at all
4.3. PF Deletion Analyses: Fukui and Sakai (2003), Mukai (2003), Hartmann (2000)
Mukai (2003) proposes that verb gapping is derived by the PF operation she dubs String Deletion
that applies to a phonetic string, regardless of its syntactic constituency. See also Fukui and
Sakai (2003) for a PF-based approach to gapping. Mukai assumes (p. 211) that “the only
structural condition on String Deletion is that the target is continuous and contains a verb.”
Under this analysis, the example in (38) is derived as in (41) (adopted from Mukai 2003: 211).
6
As Mukai (2003) shows, the lack of subjacency effects also argues against Saito’s (1987). Consider (i).
(i) [TP raion-ni Mike-ga [VP [DP ti osowareta otoko-o] tasuketa]]] (sosite)
[TP kuma-ni Tom-ga [VP [DP ti osowareta otoko-o tasuketa]]]
This example is derived by scrambling raion-ni ‘lion-Dat’/kuma-ni ‘bear-Dat’ to the conjunct-initial position, as
shown in (i), followed by right-node-raising the shared VP to the adjoined position of TP. Saito’s analysis would
predict (i) as ungrammatical since the scrambling violates the Complex NP Constraint.
- 31 -
(41) 1st conjunct IP 2nd conjunct IP
NP I′ NP I′
Mike-ga VP I Tom-ga VP I
NP V NP V
In this structure, the underlined portion of the first conjunct, which is continuous and contains a
verb, is identical to the underlined portion of the second conjunct. Thus, String Deletion applies
to the underlined portion of the first conjunct. This analysis correctly predicts the absence of
island effects in gapping since the (LF) movement of the correlate (raion ‘lion’) is not necessary
in this string-based deletion approach). It also accounts for the P-drop pattern since ni-
osowareta otoko-o in the first conjunct in (41) has the identical PF string in the second conjunct.
Mukai provides independent evidence from honorification that gapping is sensitive to the
string identity at PF. In Japanese, mesiagaru is the honorific suppletive form for the plain verb
taberu ‘eat’ that a speaker uses to show respect to the socially superior person denoted by the
subject, as shown in (42a). (42b) is unacceptable with mesiagaru since it is awkward to show
- 32 -
(42)a. Kootyoosensei-ga tempura-o mesiagatta/*tabeta.
‘I ate tempura.’
Mukai shows that String Deletion is sensitive to the surface verb form because mesiagaru and
taberu are distinct for the application of this operation. This point is illustrated by (43).
‘I ate tempura, and the principal ate sushi.’ (Mukai 2003: 212)
The ungrammaticality of this example shows that String Deletion is sensitive to the PF identity
condition. Specifically, this example has the derivation in (44). Thus, the first conjunct includes
the unacceptable honorific form, just as the first conjunct in (42a) (with mesiagaru) does.
- 33 -
There are two problems with the String Deletion analysis. Firstly, just because the
application of String Deletion is sensitive to surface verb forms does not mean that this is the
sole condition on String Deletion. If the PF identity were a sufficient identity condition for
gapping, this analysis would make a wrong prediction concerning gapping examples involving
homonyms.7 Specifically, examples like (45) involving the homonymic kumo ‘spider, cloud’
would allow four possible interpretations given here, contrary to facts. The ungrammaticality of
(46a, b), where the gapped verb is homophonous, makes the same point.
‘John showed Mary a cloud/a spider, and Bill showed Susan a cloud/spider.’
→ * John showed Mary a spider, and Bill showed Susan a cloud. (adopted from Mukai 2003: 213)
‘Intended: John (made) this plan, and Bill set up that goal.’
7
This issue was raised by Hajime Hoji, according to Mukai (2003: 213).
- 34 -
b. * John went to Chicago, and Mary went crazy. (Kim 1997: 160)
This semantic constraint would remain totally unexplained if the only structural condition on String
Deletion were that the target is continuous and contains a verb.” (p. 211). Mukai (p. 213) does
conclude in light of (45) that “LF identity, as well as PF identity, is relevant in the case of verbless
conjunction,” but she does not clarify the nature of the relevant LF identity requirement or why such
a requirement is imposed on gapping in the first place. Our proposed analysis can accommodate this
constraint if we take the notion of “anaphoricity” in the Disanaphora Law seriously. Let us suppose
that the elided element in the first conjunct (i.e. kumo) be strictly identical in its sense to the overt
element in the second conjunct. Thus, if kumo is interpreted as ‘cloud’ in the second conjunct, then
its first occurrence in the gapped clause must be also interpreted as ‘cloud’.
The necessity of some LF identity requirement on ellipsis is not the exclusive property of
gapping but also of ellipsis in general. To take one example, Chung et al (1995: 248) observe that
the sprouting operation for implicit complements within their LF Copy analysis of sluicing is
sensitive to whether the verb in the sluiced TP is identical to its antecedent verb in the preceding
TP in terms of argument structure. Consider the ditransitive verb serve. This verb has two distinct
argument structures shown in (47a, b), which generate the examples in (48a, b), respectively. The
sluicing examples relevant for our purposes are given in (49a, b).
- 35 -
(47)a. server <meal (diner)>
b. * She served the soup, but I don’t know who. (Chung et al. 1998: 248)
The contrast between (49a) and (49b) shows that licensing of an implicit argument in the sluiced
TP is sensitive to the particular argument structure of the verb serve. (49a) is fine because the
sprouting of the implicit Goal argument to whom is licensed by the argument structure in (47a).
(49b) is bad because this argument structure frame does not include the DP realization of the Goal
argument. Chung et al’s findings here, therefore, provide independent evidence that the elided
element in an incomplete clause must be identical to its correspondent not only in terms of PF
Another problem with Mukai’s analysis is that String Deletion overgenerates. Specifically,
since the only condition is that the target is continuous and contains a verb, it would incorrectly
8
As a reviewer points out, Takahashi and Fox (2005) argue that the parallelism constraint is also relevant to another
instance of ellipsis, VP-ellipsis in re-binding configurations where the variable within the elided clause is bound by an
element outside of that clause. Takahashi (2008) argues that the interpretation of quantificational null arguments in
Japanese is sensitive to the scope parallelism. One implication of these arguments is that LF parallelism is a necessary
condition for deletion in general.
- 36 -
predict that this operation should apply not only in coordinate structures but also subordinate
structures. However, as is widely known since Jackendoff (1970: 22), “gapping occurs only
when the clauses are connected by the coordinating conjunctions and, or, and nor. “ This
‘Intended Interpretation: After John bought a magazine, and Mary bought a book.’
This overgeneration is sufficiently constrained under the present theory, for gapping results
only when one of the bivalent heads is V0 and this head occurs only in coordinate structures.
the product of the PF string-based deletion process not subject to syntactic constituency.
However, she argues that the relevant operation is constrained so as to conform with the Major
Constituency Condition (Hartmann 2000: 144), originally due to Hankamer (1973), that
9
This problem arises with Abe and Hoshi’s (1997) analysis as well since their postulated LF movement should not
be prohibited from creating the I′ constituent to be copied into the gapped conjunct in non-coordinate structures.
- 37 -
remnants of this process must be major constituents, “a constituent ether immediately
(Hankamer 1973: fn.2). Although her analysis works for (17a), where Susan is not a major
ungrammatical on a par with (17a). This analysis also predicts that (20) should be
ungrammatical. The PF deletion should not be able to target pictures alone since this nominal
is not the major constituent in Hankamer’s sense. As we have seen earlier, these examples are
naturally explained under our proposed analysis due to the DE process. Therefore, I conclude
that our analysis is superior to purely string-based deletion accounts such Mukai and Hartmann.
Kim (1997, 1998) and Sohn (1994) propose that gapping results from the overt syntactic movement
of a focused remnant followed by PF deletion. Since the analyses proposed by Kim and Sohn
essentially make the same point, I restrict my discussion here on Kim’s (1997, 1998) analysis. Kim
proposes that gapping is the result of overt focus movement of remnants and correspondents
followed by TP deletion at PF. Consider the derivation of (17b) in (51) (cf. Kim 1997: 175).10
10
Kim assumes that the [+EPP] feature is weak in Korean and Japanese. Thus, subjects do not undergo overt
syntactic movement out of VP in (51).
- 38 -
(51) The Syntactic Derivation of the Example in (17b)
FocP
t2 V′ t2 V′
AgrP2 V1 AgrP2 V1
t1 V′ t1 V′
V2 V2
tV tV
Adopting the idea of Checking-through-Adjunction (Saito 1994; Sohn 1994), Kim assumes that a
contrastive phrase can adjoin iteratively to another contrastive phrase and the highest one,
containing all contrastive phrases, can move to the specifier of FocP. In (51), the lower focused
phrase (Bill-nituite in the first conjunct and Susan-nituite in the second conjunct) adjoins to the
higher focused phrase (John-ga in the first conjunct and Mary-ga in the last conjunct) to check
- 39 -
its strong [+focus] feature against the strong [+focus] feature of the latter. (17b) results when TP
deletion applies to the first conjunct under PF identity, after the remnants have all undergone
There are two problems with this hybrid movement + deletion account. The first problem
lies in Kim’s assumption that gapping in Japanese/Korean involves TP deletion. Recall that gapping
involves vP/VP-coordination. Kim argues (p.176) that gapping is TP deletion precisely for this
reason; “Since Tense never appears in the gapped conjunct, …. Gapping is PF deletion of TP.”
However, this reasoning not only begs the question of why non-gapped examples as in (7) would still
prohibit the past tense morpheme from appearing in the first conjunct but leaves unexplained the
wide scope reading of negation and negative polarity licensing discussed in section 3.1. Note that
Kim’s analysis cannot be saved by assuming that the focus-driven movement of contrastive phrases
into the specifier of the vP-internal Foc head, followed by vP deletion since it is well known since
Kuno (1978) that Japanese lacks this process entirely, as shown by the ill-formedness of (52).
‘Intended Interpretation: Since Taroo buys a book, Hanako also buys a book.’
- 40 -
The second problem concerns the Left Branch Condition effect, an issue Kim (1997: 215)
himself has noted as a residual problem for future research. I reproduce his discussion in
Japanese; see Kim (1997: 215) for Korean data. Japanese observes the Left Branch Condition, as
shown in (53a-c). Kim’s analysis, however, predicts that (54) should be bad since it involves
(54) John-ga [NP Mary-(no) imooto]-o aisiteiru sosite Bill-ga [NP Sue-(no) imooto]-o aisiteiru.
‘Lit. John (loves) Mary’s (sister) and Bill loves Sue’s sister.’
- 41 -
Under Kim’s analysis, (54) is derived by first extracting the possessor Mary/Sue out of the larger
NP into [Spec, FocP], followed by deletion of the TP in the first conjunct. This derivation,
however, is impermissible in Japanese due to the Left Branch Condition. Note that one could not
resort to the so-called “repair-by-deletion” (Merchant 2001). Since Japanese gapping involves
vP/VP-level coordination but lacks vP/VP deletion, there is no constituent whose deletion
simultaneously eliminates the offending structure from which extraction has taken place (i.e. NP)
and aisiteiru ‘love’ in (54). On the other hand, the example in question is exactly what the
present analysis predicts. The verb in the first conjunct undergoes CE. This CE, in turn, licenses
DE of the head of its DP complement, as in the English non-constituent gapping seen in (20).11
4.5. Argument Ellipsis Analyses: Oku (1998), Kim (1999), Saito (2003), Takahashi (2006, 2008)
Various researchers on Korean and Japanese, including Oku (1998), Kim (1999), Saito (2004), and
Takahashi (2006, 2008), have recently proposed that certain cases of null arguments in these
languages are analyzed as NP/DP ellipsis rather than empty pronouns (Kuroda 1965). For reasons
of space, I repeat only one argument in favor of this analysis made in Takahashi (2008) based on
what he calls quantificational null objects; see the above-mentioned works for various arguments
based on the sloppy reading (Oku 1998), inalienable possession constructions (Kim 1999), and
“parasitic gap-like” constructions (Takahashi 2006). Consider examples in (55a, b) and (56a, b).
11
Obviously, Kim’s (1997) analysis cannot explain why gapping is restricted to coordinate structures since the two
components of his analysis for gapping, scrambling and PF deletion, are known to apply elsewhere.
- 42 -
(55)a. Hanako-ga taitei-no sensei-o sonkeisiteiru.
b. Taroo-mo e sonkeisiteiru.
Taroo-also respect
Both the null argument e in (55b) and the lexical pronoun karera in (56b) are intended to take the
object quantifier taiteino sensei ‘most teachers’ in (55a)/(56a). Takahashi (p. 310) observes that the
lexical pronoun in (56b) functions as an E-type pronoun in the sense of Evans (1980); (56b) ‘means
only that Taroo respects those teachers that Hanako respects.’ Importantly, however, (55b) with the
null object not only allows this E-type pronoun reading but also ‘the interpretation that Taroo
respects most teachers, where the null object serves as a full-fledged quantifier meaning most
- 43 -
teachers (thus, the set of teachers that Taroo respects can be different from the set of teachers that
Hanako respects). If the empty null object in (55b) were analyzed as a case of null pronominal (as
argued on various grounds in Hoji 1985, Kuroda 1965, and Saito 1985), this reading would not be
accounted for. On the other hand, the argument ellipsis account provides a natural analysis for this
reading, as shown in (57a, b), where the null object is a genuine quantifier, not an empty pronoun,
that undergoes ellipsis in the PF component under identity with the object.
Given the wide range of cases of null arguments in Japanese/Korean that supports the Argument
Ellipsis account, it would be important to see whether this analysis could also extend to gapping
there seem to be significant differences between gapping and argument ellipsis. Among others,
argument ellipsis can take place across sentential boundaries, as shown in (58a, b). This does not
hold for gapping, which can take place only in coordinate structures, as illustrated in (59a, b).12
12
Admittedly, there are cases of gapping as in (ia, b) that seem at first sight to span sentence boundaries.
- 44 -
(58)a. Speaker A: John-ga zibun-no gakusei-o hometa.
Bill-also praised
According to the Argument Ellipsis analysis, the object in (58b) is elided at PF under identity with the
object in (58a). The ungrammaticality of (59a) shows that this analysis cannot be extended to gapping.
However, following Neijt (1979) and Hernàndez (2007), these cases may be analyzed as collaboration of two people
in what amounts to a single sentence. Note that (ib), without the coordinator and, is ungrammatical, a signature
property of gapping in English. I will not discuss this type of gapping further for reasons of space in this paper.
- 45 -
5. The Comparative Syntax of Gapping: Japanese vs. English
In this section, I examine differences between Japanese and English with respect to the number of
remnants in the gapped clause and the directionality of gapping. I show that these differences are
5.1. The Number of Remnants in Gapping: Rightward Movement + Deletion vs. CE/DE Ellipsis
Jackendoff (1971), Kuno (1976) and subsequent work note that there is a restriction on the number of
remnants in English gapping. Most typically, a gapping construction has two remnants in the gapped
clause. Thus, gapping with three remnants is poor, as in (60a, b). However, gapping with three
remnants is perfect in Japanese, as in (61a, b). See Kim (1997: 161-162) for relevant data from Korean.
(60)a. ?* Willy put the flowers in a vase, and Charlie the book on the table.
b. ?* Charlie entered the bedroom at 5:30, and Vera the kitchen at 6:00. (Jackendoff 1971: 24)
‘Lit. Willy (put) the flowers in a vase, and Charlie put the book on the table.’
‘Lit. Charlie (entered) the bedroom at 5:30, and Vera entered the kitchen at 6:00.’
- 46 -
There do exist cases as in (62) (from Schwartz 1999: 316, as cited in Winkler 2005: 193 with
(62) SOME talked with YOU about POLITICS and OTHERS talked with ME about MUSIC.
However, this pattern requires quite a bit of topic-focus articulation not required in Japanese/Korean
gapping; see Winkler (2005) for relevant discussion. Thus, let us follow the consensus in the
literature that English allows up to two remnants in the gapped conjunct. Our proposed analysis
predicts that by default, there should be no upper limit on the number of remnants permitted by the
CE as long as they meet syntactic and semantic constraints that that we have seen thus far to
characterize gapping. This is what we see in Japanese/Korean gapping. Thus, the question is why
English has a particular limit. Let us assume, following Jayaseelan (1991), that English gapping is the
product of the Heavy DP Shift to the right periphery of the vP, followed by the deletion of the vP at
PF. Then, the “two remnant” restriction directly follows as the combination of two independently
motivated facts: a) the subject in English moves to [Spec, TP] to vacate the vP and b) the Heavy DP
Shift can apply only once. The latter observation is illustrated by (63a, b) (Abe and Hoshi 1997: 103).
(63)a. * John built t1 t2 yesterday [with a hammer]2 [the house that he will live in]1.
b. * John built t1 t2 yesterday [the house that he will live in]1 [with a hammer]2.
- 47 -
5.2. The Directionality of Gapping: Forward vs. Backward Deletion 13
Ross (1970) proposes that there is a strong correlation between the directionality of gapping and the
underlying word order of a language. Specifically, gapping operates forward in right-branching languages
(like English) and backward in left-branching languages (like Japanese). The robustness of this
correlation has been subject to critique by subsequent research: see Junaido (1991/1992) for Hausa,
Rosenbaum (1977) for Zapotec, and Hernàndez (2007) for Persian. Most recently, Hernàndez (2007)
attempts to derive the directionality of gapping from the interaction of the category-specific nature of
coordinators with the Relativized Minimality (Rizzi 1990). Under the asymmetric view of coordination
(Munn 1993; Kayne 1994), she proposes that, in languages like English, which use the single coordinator
and to combine all types of constituents alike, the dependency/c-command relation between the verb in
the first conjunct and the CE head in the second conjunct for the purposes of Full Interpretation is
established. In languages like Japanese, which uses different coordinators for different types of phrases,
the relation in question is blocked by coordinators with the [+V, -N] feature that intervenes between the
two verbs. Hence, forward gapping is impossible in Japanese. However, this analysis obviously cannot
explain why backward gapping is possible in Japanese in the first place because the verb in the second
conjunct should not be able to c-command into the CE verb in the first conjunct, hence the gapped verb
should violate the Principle of Full Interpretation. For this reason, I pursue a different analysis for the
13
Thanks to an anonymous reviewer for encouraging me to think about the question discussed in this subsection.
- 48 -
directionality of gapping, though I adopt certain basic assumptions from her work. Following Johannesen
(1998), suppose that coordinate structures in English and Japanese are as in (64a) and (64b), respectively.
ConjP ConjP
V2 V1
Assuming with Hernàndez that the c-commanding relation between the two Vs can be mediated by their
dominating VPs, VP1 asymmetrically c-commands VP2. English allows only forward gapping because
the feature transfer would not be successful if gapping/CE went backwards. Conversely, Japanese only
allows backward gapping/CE because in (64b), V2 asymmetrically c-commands V1. This analysis thus
ties the directionality of gapping to the difference from independently motivated parametric differences
concerning coordination. Johannessen’s (1998) study provides independent confirmation for this
languages, Johannesen (p. 55, 56) shows that there is a quite robust correlation between the direction of
verb and object and the position of what she calls unbalanced coordination, where one of the conjuncts
shows deviant grammatical properties with respect to case, number, etc. Specifically, SVO languages
- 49 -
(Italian, Norwegian, Old Hebrew, Palestinian Arabic, etc.) have the second conjunct as the deviant
conjunct whereas SOV languages (Amharic, Burushaski, Hopi, etc.) have the first conjunct as the deviant
conjunct. This correlation is naturally explained by (64a, b) since what is selected by the Conj head is the
second conjunct in (64a) but the first conjunct in (64b). Note that gapping is one special case of
unbalanced coordination in that the verb in only one of the conjuncts is affected. Thus, the directionality
of the gapping can be predicted straightforwardly from the parametric difference concerning coordination.
In this section, I discuss one potential challenge to our present analysis and suggest a solution to
Abe and Hoshi (p.133) note, as one issue for future research, that P-/case-drop from a non-
final remnant within the gapped conjunct results in ungrammaticality. This point is illustrated in
(65a), together with its LF representation in (65b) (Abe and Hoshi 1997:133).
‘Mary built the garage with a saw and John the house a hammer.’
[IP John-ga [I′ hammaa1 [I′ ie-o2 [I′ [PP t1-de] t2 tukutta ]]]]
- 50 -
Abe and Hoshi (p. 133) note that their analysis would incorrectly rule (65a) in because “nothing
prevents P-stranding of de because it is created by left movement” and leave this problem for
future research. This problem, in fact, has remained as an unresolved issue in subsequent research
on gapping in Japanese and Korean. Noting that a similar P-stranding prohibition characterizes
Korean gapping, Sohn (1994, 1999) and Kim (1997, 1998) make the observation in (66), stating
that P-stranding in Korean gapping does not follow from anything deeper other than that. (67a, b)
confirms that this observation also characterizes environments for P-omission in Japanese gapping.
(66) The postposition or Case-marker in Korean Gapping may drop only if the host remnant is
‘Lit. Mary (built) the garage with a saw and John the house a hammer.’
‘Mary built the garage with a saw and John the house a hammer.’
Similarly, Sohn (1999: 383) speculates that P/case particle-omission is a PF phenomenon, stating
“conjunction takes the whole contrasted part as one unit, optionally deleting the postposition on
- 51 -
the last NP.” Kato (2006: 124, 125) speculates that P-stranding in Japanese might have to do
with a different intonational contour on the element that immediately precedes the gap (such as
lengthening of the syllable preceding the gap), but leaves the issue for future research.
Following the suggestion of an anonymous reviewer, I pursue a different analysis of the case at
hand in the area of on-line processing whereby the ungrammaticality of (67a, b) is not due to Kim’s PF
constraint itself but ultimately to the violation of the parallelism constraint (see section 2) induced by
that constraint as one processes gapping on line. According to the PF constraint, once we hear the DP
without the postposition or case-particle, we interpret it as marking the end of the first conjunct, in
conformity with the PF constraint. Thus, the first conjunct of (67a) with nokogiri ‘saw’ is interpreted as
in (68). When one hears gareezi-o ‘garage-Acc’, however, we perceive that the parallelism constraint is
immediately violated, for this DP has no correspondent in the first clause. This problem does not occur
in processing (67a) with nokogiri-de since this variant meets the parallelism constraint (69).
- 52 -
Our analysis crucially relies on the idea that the parallelism constraint must be satisfied as one
It is important to note in this present context that Kuno (1976) proposes the following
The two constituents left over by Gapping are most readily interpretable as entering into a
This constraint is different from our constraint in that it was motivated on certain parallelism required
by question-answer pairs. However, the two constraints share the same underlying idea that gapping
7. Conclusions
Gapping in Japanese is best analyzed by the coordinate and dependent ellipsis theory proposed by
Williams (1997). This analysis provides a natural explanation for several syntactic, semantic, and
phonological properties associated with this construction that would remain unexplained under
- 53 -
previous analyses resorting to Right-Node-Raising, LF Copy, PF-String Deletion, Syntactic
Movement + PF Deletion, and Argument Ellipsis. Our proposed analysis has right theoretical
the degree that previous uni-modular analyses cannot. The payoff of our present analysis, of course,
is that it achieves this result from independently motivated principles of coordination, dependency
and their manifestations at the interfaces among syntax, phonology, semantics, and processing.
References
Abe, Jun and Hiroto Hoshi. 1997. Gapping and P-stranding. Journal of East Asian Linguistics 6: 101-136.
Abbott, Barbara. 1976. Right node raising as a test for constituenthood. Linguistic Inquiry 7: 639-642.
Ackema, Peter and Krista Szendröi. 2002. Determiner sharing as an instance of dependent
Bobaljik, Jonathan. 1995. Morphosyntax: The syntax of verbal inflection. Doctoral dissertation. MIT
Bošković, Željiko. 2001. On the nature of the syntax-phonology interface: Cliticization and
Bošković, Željiko and Howard Lasnik. 2003. On the distribution of null complementizers.
- 54 -
Bresnan, Joan. 1974. The position of certain clause particles in phrase structure. Linguistic
Inquiry 5: 614-619.
Choe, Jae-Woong. 1987. LF movement and pied-piping. Linguistic Inquiry 18: 348-353.
Chung, Sandra, William Ladusaw, and James McCloskey. 1995. Sluicing and logical form.
Fukui, Naoki and Hiromu Sakai. 2003. The visibility guideline for functional categories: Verb
Fox, Danny. 2000. Economy and semantic interpretation. Cambridge, MA: MIT Press.
Halle, Morris and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In: Kenneth
Hale and Samuel Jay Keyser (eds.) The view from building 20. 111-176. Cambridge, MA: MIT Press.
Hankamer, Jorge. 1971. Constraints on deletion in syntax. Doctoral dissertation. Yale University.
Hartmann, Katrina. 2000. Right node raising and gapping: Interface conditions on prosodic
Hernandez, Anna Carrera. 2007. Gapping as a syntactic dependency. Lingua 117: 2106-2133.
Hirata, Ichiro. 2006. Predicate coordination and clause structure in Japanese. The Linguistic
- 55 -
Hoji, Hajime. 1985. Logical form constraints and configurational structures in Japanese.
Hornstein, Norbert and Amy Weinberg. 1981. Case theory and preposition stranding. Linguistic
Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. Doctoral
dissertation. MIT.
Jayaseelan, K.A. 1990. Incomplete VP deletion and gapping. Linguistic Analysis 20: 64-81.
Junaido, Ismail. 1991/1992. The typology of gapping in Hausa. Harsunan Nijeriya XVI: 1-9.
Kato, Kumiko. 2006. Japanese gapping in minimalist syntax. Doctoral dissertation. University of
Washington.
Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press.
Kim, Jeong-Seok. 1997. Syntactic focus movement and ellipsis: A minimalist approach. Doctoral
Kim, Soowon. 1999. Sloppy/strict identity, empty objects, and NP ellipsis. Journal of East Asian
Linguistics 8: 255-284,
Koizumi, Masatoshi. 2000. String vacuous overt verb raising. Journal of East Asian Linguistics 9: 227-285.
- 56 -
Koster, Jan.1987. Domains and dynasties. Dordrecht: Kluwer.
Kuno, Susumu. 1972. Functional sentence perspective - A case study from Japanese and English.
Kuno, Susumu. 1973. Nihon bumpo kenkyu (Studies on Japanese grammar). Tokyo: Taishukan.
Kuno, Susumu. 1978. Japanese: A characteristic OV language. In: Winfred Lehmann (ed.) Syntactic
typology: Studies on the phenomenonology of language. 57-138. Austin: University of Texas Press.
Kuroda, Sigeyuki. 1965. Generative grammatical studies in the Japanese language. Doctoral
dissertation. MIT.
Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19: 335-391.
Lasnik, Howard. 1981.Restricting the theory of transformations: A case study. In: Norbert
Hornstein and David Lightfoot (eds.) Explanations in linguistics. 152-173. London: Longman.
Lasnik, Howard and Mamoru Saito. 1984. On the nature of proper government. Linguistic
Lasnik, Howard and Mamoru Saito. 1992. Move α: Conditions on its application and outputs.
Maling, Joan. 1972. On ‘gapping and the order of constituents’. Linguistic Inquiry 3: 101-108.
- 57 -
McCawley, James. 1993. Gapping with shared operators. In: Guenter, Joshua, Barbara Kaiser, and
Cheryl Zoll (eds.) Proceedings of BLS 19: 245-253. Berkeley: Berkeley Linguistics Society.
Mukai, Emi. 2003. On verbless conjunction in Japanese: In: Makoto Kadowaki and Shigeto Kawahara
Munn, Alan. 1993. Topics in the syntax and semantics of coordinate structures. Doctoral
Neeleman, Ad and Hans van de Koot. 2002. The configurational matrix. Linguistic Inquiry 33: 529-574.
Massachusetts, Amherst.
Oehrle, Richard. 1987. Boolean properties in the analysis of gapping. In: Geoffrey J. Huck and
Almerindo E. Ojeda (eds.) Syntax and Semantics 20: Discontinuous Constituency. 203-240.
Oku, Satoshi. 1998. A theory of selection and reconstruction in the minimalist perspective.
Pesetsky, David. 1987. Wh-in-situ: Movement and unselective binding. In: Eric Reuland and Alice
ter Meulen (eds.) The representation of (in)definiteness. 98-129. Cambridge, MA: MIT Press.
- 58 -
Phillips, Colin (2003). Linear order and constituency. Linguistic Inquiry 34: 37-90
Reinhart, Tanya. 1991. Elliptic constructions – non-quantificational LF. In : Asa Kasher (ed.)
Sag, Ivan. 1976. Deletion and logical form. Doctoral dissertation. MIT.
Saito, Mamoru. 1985. Some asymmetries in Japanese and their theoretical implications.
Saito, Mamoru. 1987. Three notes on syntactic movement in Japanese. In: Takashi Imai and
Saito, Mamoru. 1994. Additional WH-effects and the adjunction site theory. Journal of East
Schwarz, Bernhard. 1999. On the syntax of either…or. Natural Language and Linguistic Theory
17: 339-370.
Siegel, Muffy. 1987. Gapping and interpretation. Linguistic Inquiry 15: 523-530.
Sohn, Keun-Won. 1994. On gapping and right node raising. In: Young-Sun Kim et al. (eds.) Explorations
in generative grammar: A festschrift to Dong-Whee Yang. 589-611. Hankuk Publishing Co.: Seoul.
- 59 -
Sohn, Keun-Won. 1999. Deletion or copying? Right node raising in Korean and Japanese.
Takahashi, Daiko. 2006. Apparent parasitic gaps and null arguments in Japanese. Journal of East
Takahashi, Daiko. 2008. Quantificational null objects and argument ellipsis. Linguistic Inquiry 39: 307-326.
Takahashi, Shoichi and Danny Fox. 2005. MaxElide and the re-binding problem. In: Effi
Georgala and Jonathan Howell (eds.) Proceedings of Semantics and Linguistic Theory 15:
Torrego, Esther. 1984. On inversion in Spanish and some of its effects. Linguistic Inquiry 15: 103-129.
Watanabe, Akira. 1992. Subjacency and S-structure movement of wh-in-situ. Journal of East
Wilder, Chris. 1997. Some properties of ellipsis in coordination. In: Artemis Alexiadou and T. Alan
Hall (eds.) Studies on universal grammar and typological variation. Amsterdam: John Benjamins.
Williams, Edwin. 1977. Discourse and logical form. Linguistic Inquiry 8: 101-139
Williams, Edwin. 1997. Blocking and anaphora. Linguistic Inquiry 28: 577-628.
Winkler, Suzanne. 2005. Ellipsis and focus in generative grammar. Berlin: Mouton de Gruyter.
- 60 -