Island Sensitivity in Japanese Sluicing and Some Implications[*]
Teruhiko FUKAYA
Gunma Prefectural Women's University
This paper investigates island effects in Japanese sluicing and demonstrates that case-marked sluicing, a subtype of sluicing where the wh-phrase is case-marked, is sensitive to such syntactic islands as the relative clause and the adjunct islands, although it appears otherwise at first glance. This is demonstrated by close examinination of readings available and unavailable in case-marked sluicing cases where the element in the first conjunct corresponding to the wh-expression in the second is a non-indefinite expression. It is then shown that the behaviors of Japanese case-marked sluicing can be accounted for by the theory of ellipsis resolution proposed in Fukaya & Hoji 1999 in combination with a version of Merchant's (2001: ch.5) proposal for local movement in sluicing resolution. Another subtype of Japanese sluicing called non-case-marked sluicing, where the wh-phrase is not cased marked, is also examined and shown to exhibit no island effects, in contrast to case-marked sluicing. It is claimed that its properties can be accounted for by assuming a structure which is radically distinct from that of case-marked sluicing. English sluicing is then re-examined from the perspective gained from the investigation of Japanese sluicing, and it is demonstrated that some cases of English contrast sluicing exhibit island effects while others do not. It is then suggested that two subtypes of sluicing be distinguished in English as well which correspond to case-marked and non-case-marked sluicing in Japanese.
1. Introduction
Ross (1969) was the first to bring to light the construction that he referred to as sluicing as in the second conjunct in (1b).
(1) a. Somebody just left--guess who just left.
b. Somebody just left--guess who.
(Ross 1969: 252)
In the forthcoming discussion, I will refer to the wh-phrase in sluicing as the remnant and to the element in the first conjunct that corresponds to the remnant as the correlate, for ease of reference. In (1b), for example, somebody is the correlate, and who is the remnant.
Ross (1969: 276-277) observed that the island effects are weaker in sluicing than in its non-elliptical counterpart, as in (2)-(4).
(2) [Complex NP Constraint (relative clause)]
a. *She kissed a man who bit one of my friends, but Tom doesn't realize which one of his friends she kissed a man who bit.
b. ?She kissed a man who bit one of my friends, but Tom doesn't realize which one of his friends.
(Ross 1969: (72a-b))
(3) [Sentential Subject Constraint]
a. *That he'll hire someone is possible, but I won't divulge who that he'll hire is possible.
b. ??That he'll hire someone is possible, but I won't divulge who.
(Ross 1969: (73b))
(4) [Coordinate Structure Constraint]
a. *Irv and someone were dancing together, but I don't know who Irv and were dancing together.
b. ??Irv and someone were dancing together, but I don't know who.
(Ross 1969: (71))
Observing these facts, Ross gave the informal statement regarding island effects in sluicing in (5).
(5) If a node is moved out of its island, an ungrammatical sentence will result. If the island forming node does not appear in surface structure, violations of lesser severity will (in general) ensue.
(Ross 1969: (75))
Although Ross maintained that English sluicing still exhibits some island effects, it has been the general consensus in the literature that island effects are not observed in English sluicing (see Chomsky 1972, Levin 1982, Chung et al. 1995, Merchant 2001, 2008, Lasnik 2001, and Fox & Lasnik 2003 (F&L, henceforth), among others). (6) and (7) are taken from Merchant's works.
(6) [Complex NP Constraint (relative clause)]
a. *They want to hire someone who speaks a Balkan language, but I don't remember which Balkan language they want to hire [someone who speaks _ ].
b. They want to hire someone who speaks a Balkan language, but I don't remember which.
(Merchant 2001: ch.3 (5))
(7) [Adjunct Clause Constraint]
a. *Ben will be mad if Abby talks to one of the teachers, but she couldn't remember which (of the teachers) Ben will be mad [if she talks to _ ].
b. Ben will be mad if Abby talks to one of the teachers, but she couldn't remember which.
(Merchant 2008: (13a-b))
Schematically, if the structure in (8) obtains in the first conjunct and sluicing is acceptable, it has been considered as evidence that there are no island effects.
(8) ... [ISLAND ... correlate ... ] ...
Many researchers have followed Ross 1969 and maintain that the remnant has undergone regular wh-movement.
The issue of why island effects are nullified in sluicing has been one of the prominent topics in the literature on sluicing. Chung, et al. (1995) propose an analysis where there is no movement involved in the derivation of sluicing; Merchant (2001) pursues an account in which only non-island-violating local movement is involved in the derivation of sluicing in island contexts; F&L and Merchant (2008) propose that syntactic islands are PF phenomena and that deletion of the violating structure ameliorates the island violation. Kimura (2010) argues that the remnant wh-phrase stays in-situ in some cases of sluicing.
In this paper, I will first examine case-marked sluicing (cm sluicing) in Japanese, in which the remnant is marked with a case-marker or a postposition, and demonstrate that despite appearing otherwise, it is in fact sensitive to the relative clause and the adjunct islands. I will then show that the copy theory of ellipsis resolution proposed in Fukaya & Hoji 1999 (henceforth, F&H) combined with a version of Merchant's (2001: ch.5) proposal for local movement in sluicing in propositional island contexts can account for what appears to be the peculiar behavior of Japanese sluicing.[1] I will also investigate non-case-marked sluicing (non-cm sluicing) in Japanese, where the remnant is not marked with a case-marker or a postposition, and argue that its apparent island insensitivity is due to the availability of a copula structure with an empty pro as the subject. I will then revisit English sluicing and demonstrate that it also exhibits relative clause island effects and suggest that the relative clause island is not a PF representational island in English, contra the widely-held view. To account for the new facts of English sluicing discussed in this paper, it will be suggested that two types of sluicing be distinguished in English, which correspond to cm and non-cm sluicing in Japanese.
This paper is organized as follows. Section 2 observes island sensitivity in cm sluicing in Japanese. Section 3 introduces F&H's theory of ellipsis resolution adopted in this paper. Section 4 examines how the facts observed in section 2 are accounted for under the theory of ellipsis resolution in section 3 combined with a version of Merchant's (2001) proposal for local movement. Section 5 discusses island sensitivity in non-cm sluicing in Japanese and shows how the facts are accounted for under the copula analysis. Section 6 re-examines English sluicing from the perspective gained from the examination of Japanese sluicing. Section 7 concludes the paper.
2. Island sensitivity in Japanese sluicing
In this section I will show that Japanese sluicing respects syntactic islands, although in some cases it appears to be insensitive to them as in the case of English sluicing. I will examine two types of islands for that purpose: the relative clause island and the adjunct island. Before getting into discussion on island effects, let us first give an overview of sluicing in Japanese with respect to what types of elements are allowed as the correlate.
2.1. Japanese sluicing
Inoue (1976, 1978) first observed that Japanese has a construction similar to English sluicing as in (9). (9a) is a simplex sentence example, and (9b) is an embedded sentence example.[2]
(9) a. John-wa dareka-o suisensita ga, boku-wa [dare-o ka] siranai.
John-top someone-acc recommended but I-top who-acc Q know:not
'John recommended someone, but I don't know who <John recommended>.'
b. John-wa Mary-ga dareka-o suisensita to itteita ga,
John-top Mary-nom someone-acc recommended that said but
boku-wa dare-o ka oboeteinai
I-top who-acc Q remember:not
'John said that Mary recommended someone, but I don't remember who <John said that Mary recommended>.'
(Inoue 1978: 56)
These are cases where the correlate is an indefinite. Not only an indefinite but also a definite can be the correlate in Japanese sluicing, as in the case of what Merchant (2001: 36, 2008) refers to as contrast sluicing in English. (10a) is an example where a simplex sentence appears to be missing, and (10b) is an example where a complex sentence appears to be missing.
(10) a. Boku-wa [USC-ga Susan-o suisensita no]-o sitteiru ga,
I-top USC-nom Susan-acc recommended that -acc know but
Bill-wa dare-o ka siranai.
Bill-top who-acc Q know:not
'I know that USC recommended Susan, but Bill doesn't know who <USC recommended _>.'
b. Boku-wa [John-ga [USC-ga Susan-o suisen sita to] itteita no]-o sitteiru ga,
I-top John-nom USC-nom Susan-acc recommended that said that-acc know but
Bill-wa [dare-o ka] siranai.
Bill-top who-acc Q know:not
'I know that John said that USC recommended Susan, but Bill doesn't know who <John said that USC recommended>.'
In addition to indefinite and definite NPs, in-situ wh-phrases can be the correlate in Japanese sluicing as in (11). (11a) is a case where a simplex sentence appears to be missing, and (11b) a case where a complex sentence appears to be missing.
(11) a. Boku-wa [USC-ga dare-o suisen sita ka] sitteiru ga,
I-top USC-nom who-acc recommended Q know but
Bill-wa dare-o ka siranai.
Bill-top who-acc Q know:not
'(lit.)I know who USC recommended, but Bill doesn't know who <USC recommended>.'
b. Boku-wa [John-ga [USC-ga dare-o suisensita to] itteita ka] sitteiru ga,
I-top John-nom USC-nom who-acc recommended that said Q know but
Bill-wa dare-o ka siranai.
Bill-top who-acc Q know:not
'(lit.)I know who John said that USC recommended, but Bill doesn't know who <John said that USC recommended>.'
We have illustrated that (i) an indefinite, (ii) a definite, and (iii) an in-situ wh-phrase can be a correlate in Japanese sluicing. Next, we turn to island sensitivity in Japanese sluicing. In the following discussion, correlates of the types (ii) and (iii) will play a crucial role for the reason that only with these correlate types can the existence of island effects be demonstrated.
2.2. The relative clause island
Let us begin with the relative clause island. Before getting into discussion on sluicing, however, I will first demonstrate that relative clauses are indeed syntactic islands in Japanese. Consider (12).
(12) a. *[soko3-no kogaisya]-ni [[kanari-no kazu]-no nikkei kigyoo]3-ga
that:place-gen subsidiary-to a:large:number-gen Japanese company-nom
[ _ kekkan buhin-o noonyuu siteita meekaa]-o uttaeta.
defective parts-acc was:supplying maker -acc sued
'(lit.) To its3 subsidiary, [a large number of Japanese companies]3 sued a manufacturer that had been supplying defective parts _.'
b. *[[[[kanari-no kazu]-no nikkei kigyoo]3-ga [ _ kekkan buhin-o
a:large:number-gen Japanese company-nom defective parts-acc
noonyuu siteita meekaa]-o uttaeta] no]-wa [soko-no kogaisya]3-ni da.
was:supplying maker -acc sued that-top that:place-gen subsidiary-to cop
'(lit.) It is to its3 subsidiary that [a large number of Japanese companies]3 sued a manufacturer which had been supplying defective parts _.'
(12a) and (12b) are examples of scrambling and the cleft construction, respectively. I have used cases involving bound variable anaphora to make sure that the dislocated element is related by means of movement to the position marked by an underscore that is within the relative clause.[3] The intended reading here is roughly that for each of the large number of Japanese companies x, x sued a manufacturer who had been supplying defective parts to x's subsidiary. In neither of these examples would this covariant reading be available. This contrasts with the availability of such a reading in (13), which involves no dislocation of the phrase containing the dependent term from its canonical position.
(13) [[kanari-no kazu]-no nikkei kigyoo]3-ga [[soko3-no kogaisya]-ni
a:large:number-gen Japanese company-nom that:place-gen subsidiary -to
[kekkan buhin]-o noonyuu siteita meekaa]-o uttaeta.
defective parts-acc was:supplying maker -acc sued
'(lit.) [A large number of Japanese companies]3 sued a manufacturer that had been supplying defective parts to its3 subsidiary.'
The unacceptability of (12) shows that the dislocated element cannot be related to the underscored position, and this indicates that a relative clause is indeed a syntactic island in Japanese.
Now let us return to sluicing. If we follow the criterion for island insensitivity in sluicing mentioned in section 1, i.e., whether sluicing is possible when the first conjunct is of the structure in (8), sluicing in Japanese appears not to exhibit island effects. Consider (14).
(14) keisatu-wa [a [pro2 Los Angeles-de [aru yuumee zin]-ni mayaku-o utta]
police-top Los Angeles-in a celebrity -to drug-acc sold
otoko2]-o taihosita rasii ga, boku-wa [dare-ni ka] siranai.
man -acc arrested seem but I-top who-to Q know:not
'(I heard) the police arrested a man who had sold drugs to a celebrity in LA, but I don't know to whom.'
In (14), the correlate is located within a relative clause island; nevertheless, sluicing is possible. This indicates that Japanese sluicing with an indefinite correlate does not exhibit island effects.[4] In sluicing with a definite correlate, island effects do not seem to be observed, either, as can be seen in (15) and (16).
(15) [keisatu-wa [ISLAND [pro2 [Tanaka giin]-ni wairo-o okutta]
police-top Rep. Tanaka-to bribe-acc gave
otoko2]-o taihosita ga, Bill-wa [dono giin-ni ka] siranakatta rasii.
man -acc arrested but Bill-top which Rep. -to Q knew:not seem
'The police arrested the man who gave a bribe to Representative Tanaka, but it seems that Bill didn't know to which Representative.'
(16) boku-wa keisatu-ga [ISLAND [ pro2 [Tanaka giin]-ni wairo-o okutta] otoko2]-o
I-top police-nom Rep. Tanaka-to bribe-acc gave man -acc
taihosita no-wa sitteiru ga, [[hoka-no dono giin]-ni ka]-wa siranai.
arrested that-top know but other-gen which Rep.-to Q -top know:not
'I know that the police arrested the man who gave a bribe to Representative Tanaka, but I don't know to which other Representative.'
Sluicing with a wh-phrase correlate does not seem to exhibit island effects, either, as is observed in (17) and (18).
(17) boku-wa [[ISLAND [sensyuu pro2 [dono giin]-o
I-top last:week which congressman-acc
hihansita] sinbunsya2]-ga boikotto sareteiru ka] sitteiru ga,