You are on page 1of 35

Journal of East Asian Linguistics

https://doi.org/10.1007/s10831-019-09195-3
(0123456789().,-volV)(0123456789().,-volV)

Parallel copying in dislocation copying: evidence


from Cantonese

Jackie Yan-Ki Lai1

Received: 19 March 2018 / Accepted: 6 May 2019


© Springer Nature B.V. 2019

Abstract Cantonese exhibits dislocation copying, a phenomenon whereby some


syntactic material in a canonical sentence is repeated after the sentence particle.
This paper demonstrates that the familiar Move-and-Elide analysis does not readily
accommodate the structural properties of dislocation copying. Instead, the deriva-
tion of dislocation copying involves the formation of parallel Ā-chains. The
apparent copying character of dislocation copying, accordingly, is the result of
neither Move-and-Elide nor spell-out of multiple chain links, both prevalent
mechanisms for yielding repetition. The current proposal sheds a fresh light on an
ill-understood area of Cantonese syntax with regard to dislocation in general, and
lends additional support to utilizing parallel chains in grammar as a syntactic means
of yielding repetition.

Keywords Copying · Dislocation · Ellipsis · Parallel chain · Cantonese

1 Introduction

In recent decades, the Move-and-Elide analysis has become a well-established


option for analyzing ellipsis, as schematized in (1).

(1) [ZP XP \[YP … XP …][]

The Move-and-Elide analysis standardly assumes the following form: an XP first


evacuates out of an ellipsis site (e.g. YP in (1)), and the ellipsis site subsequently
undergoes deletion. This line of analysis has been adopted for a variety of

& Jackie Yan-Ki Lai


jyklai@uchicago.edu
1
Department of Linguistics, University of Chicago, 1115 E. 58th Street, Chicago, IL 60637, USA

123
J. Y.-K. Lai

phenomena across languages, including exceptives (Potsdam 2017 for Malagasy),


fragment answers (Merchant 2004 for English; İnce 2012 for Turkish; Yim 2012 for
Korean), fragment questions (Wei 2018 for Chinese), sluicing (Merchant 2001 for
English; van Craenenbroeck and Lipták 2006 for Hungarian), split questions (Arregi
2010 for Spanish), and even dislocation phenomena such as left dislocation (Ott
2014 for Dutch and German) and right dislocation (Tanaka 2001 for Japanese; Ko
2016 for Korean; Ott and de Vries 2016 for Dutch and German), among many
others.
The goal of this paper is to investigate in detail an understudied elliptical
phenomenon attested in Cantonese, termed dislocation copying (hereafter ‘DC’)
following Cheung’s (2015) terminology.

(2) keoi5 zau2-zo2 laa3 keoi5


3.SG leave-PFV SP 3.SG
‘She left.’
(3) Ming4zai2 wui5 maai5 go2 bun2 syu1 aa3 Ming4zai2
Ming will buy that CL book SP Ming
‘Ming will buy that book.’

A DC sentence consists of a full (i.e. gapless) pre-SP clause, a sentence particle SP,
and some post-SP material which has an overt correlate in the pre-SP clause. In each
of the above examples, the subject appears twice, which yields the impression of
copying.
At first glance, the phenomenon of Cantonese DC does not look at all
remarkable, for it appears to fall perfectly under some sort of Move-and-Elide
analysis, whereby the post-SP phrases in (2)–(3) have evacuated out of an ellipsis
site and are the remnants of deletion. In fact, the first formal study devoted
specifically to Cantonese DC, i.e. Cheung (2015), has proposed exactly such a
Move-and-Elide analysis. This paper offers detailed argumentation that, although
being quite an attractive option, the Move-and-Elide analysis is ultimately ill-suited
for Cantonese DC. Instead, it is proposed that the derivation of Cantonese DC
sentences involves the formation of two parallel chains (à la Chomsky 2008), with
the pre-SP clause and the post-SP clause constituting links of separate chains
sharing the same foot, and the post-SP material arising from a deletion operation.
While deletion is an indispensable ingredient in the proposed analysis and the
Move-and-Elide analysis, the former does not necessitate evacuating movement out
of an ellipsis site.
As will become apparent, the current proposal sheds a fresh light on a poorly
understood area of Cantonese syntax. Although different types of dislocation are
known to exist in colloquial varieties of the language, each type of dislocation has
received drastically different analyses in the existing literature (see Cheung
1997, 2009 for the Dislocation Focus Construction; Cheung 2015, Tang 2015, 2018
for Dislocation Copying; Chan 2013 for Clause-Internal Dislocation; Lee 2017 for a
sub-type of Right Dislocation), and a general theory of Cantonese dislocation is still
lacking. Crucially, I will demonstrate that the current analysis, proposed originally
for DC, successfully predicts the existence of the kind of dislocation structure

123
Parallel copying in dislocation copying

examined in Lee (2017), which constitutes another virtue in not subsuming


Cantonese DC under the Move-and-Elide analysis.
The rest of the paper is organized as follows. Section 2 documents the core
properties of Cantonese DC, including both existing and new observations.
Section 3 explicates the proposed analysis, under which Cantonese DC involves
parallel chain formation. Section 4 then explores the consequences of the proposal,
demonstrating how it successfully accounts for the noted DC properties; further-
more, some apparent counterexamples are raised and addressed. In Sect. 5, I
compare the two existing analyses of DC (i.e. Tang 2015 and Cheung 2015) with the
current analysis, and discuss several issues which they encounter. Section 6 extends
the current proposal to other types of dislocation in Cantonese. Section 7, the final
section, offers conclusions and further implications of the proposed analysis on the
Cantonese left periphery, the copying mechanism in UG, and the formulation of a
general theory of dislocation in Cantonese.

2 Core properties of Cantonese DC

For ease of reference, I will label the components of a DC sentence as follows.

(4) [XP … α …]host clause SP [YP α]cauda

Hence in (2), for instance, keoi5 zau2-zo2 is the host clause, and the post-SP
element keoi5 is the cauda.
The first property of a Cantonese DC sentence is that the cauda may not contain
any overt SP (see also Cheung 2015:258 for the same observation).

(5) *ngo5dei6 ting1jat6 haau2si3 aa3 ngo5dei6 aa3


1.PL tomorrow exam SP 1.PL SP
‘We are having an examination tomorrow.’

Although the acceptability status of (5) would considerably improve when there
exists an intonational break in between the SP and the cauda, as in (6), such an
apparent improvement is attributable to the irrelevant parse where two independent
sentences are juxtaposed one after another. As (7) demonstrates, Cantonese SPs are
compatible with sub-sentential phrases.

(6) ngo5dei6 ting1jat6 haau2si3 aa3 || ngo5dei6 aa3


1.PL tomorrow exam SP 1.PL SP
(7) ngo5dei6 aa3
1.PL SP
‘(It’s) us.’

This view is further supported by the existence of examples like (8) which remain
ungrammatical regardless of the presence of an intonational break. In this ill-formed
example, the DC parse is unavailable because an overt SP is present in the cauda, and
the (irrelevant) two-sentence parse is also unavailable since (9) is independently bad.

123
J. Y.-K. Lai

(8) *nei5 dou3dai2 heoi3-zo2 bin1dou6 aa3 (||) nei5 dou3dai2 aa3?
2.SG DOUDAI go-PFV where Q 2.SG DOUDAI Q
‘Where the hell did you go?’
(Cheung 2015:258, ex. (104); tones added)
(9) *nei5 dou3dai2 aa3?
2.SG DOUDAI Q

The second property is that the cauda observes a curious restriction which
prohibits it from correlating with an object phrase in the host clause, and hence it is
not the case that any syntactic material in the host clause may serve as the cauda;
see (10) for an illustration.

(10) ngo5 zung1ji3 keoi5 aa3 {ngo5/*keoi5}


1.SG like 3.SG SP 1.SG 3.SG
‘I like him.’

The fact that the restriction at issue is structural is evident in (11), where the subject
and the object DPs in the host clause bear the same phonological form. Again, the
cauda in this case may only correlate with the subject, and the fact that keoi5 makes
a bad cauda in (10) therefore cannot be attributed merely to identity avoidance (the
Obligatory Contour Principle; Leben 1973, Goldsmith 1976, Tang 2000).

(11) keoi51 zung1ji3 keoi52 aa3 keoi51/*2


3.SG like 3.SG SP 3.SG
‘She likes her.’

Moreover, it is important to note that the current restriction is not simply a quirk of
pronominals.1 In fact, the same restriction on the cauda has been noted in Cheung
(2015:248–249), and his examples illustrate that the object restriction holds for
lexical DPs (the judgments below are as reported in Cheung 2015, which my
consultants share).

(12) Aa3ming4 dou1 faan2-zo2 nguk1kei5 laa1{Aa3ming4/ *nguk1kei5}


Ming also return-PFV home SP Ming home
‘Ming also returned home.’
(Cheung 2015:248, ex. (68); tones added)
(13) keoi5 wui5 hou2 siu2sam1-gam2 tai2 go2-bun2 syu1 gaa3
3.SG will very careful-way read that-CL book SP
{keoi5/ *go2-bun2 syu1}
3.SG that-CL book
‘He will read that book very carefully.’
(ibid.:248, ex. (70); tones added)

1
I thank a reviewer for raising this question.

123
Parallel copying in dislocation copying

The final property is that the SP in a DC sentence must be phonologically overt,


which to my knowledge has not been previously documented. In general,
phonologically overt SPs may alternate with the phonologically null SP, which is
known to exist in Cantonese (Matthews and Yip 2011). The contrast between (14)
and (15) illustrates this.

(14) keoi5 zung1ji3 nei5 {aa3/*∅} keoi5


3.SG like 2.SG SP 3.SG
‘He likes you.’
(15) keoi5 zung1ji3 nei5 {aa3/∅}
3.SG like 2.SG SP
‘He likes you.’

2.1 Section summary

This section has established the core structural properties of Cantonese DC: (i) the
cauda may not contain an overt SP; (ii) the cauda observes a structural object
restriction; and (iii) the SP must be phonologically overt. In the next section, I
develop a novel analysis of Cantonese DC, whereby these observed properties are
shown to fall out.

3 Dislocation copying involves parallel chain formation

In the analysis I will develop here, I propose that DC sentences differ structurally
from canonical run-of-the-mill sentences in having an additional functional
projection sandwiched between TP and CP, which is labelled as ‘GP’ (=GivenP).

I assume that the Chinese SPs are X0-elements (Cheng 1991; Tang 1998; Cheung
2009; Paul 2014 among many others), more specifically C0. Moreover, I assume that
the left-headed C0-element bears the structure-building feature [•TP•], following
Heck and Müller’s (2007) notational system. In their system, edge features ‘do not
have to be neutral with respect to the properties of the items they attract’ (Müller

123
J. Y.-K. Lai

2010:38); an edge feature may attract any XP, as well as specifying the category it
attracts, as in the current case.2
On the other hand, the additional functional head G0 in (16b) bears the general
structure-building feature [•XP•], which is therefore free to attract any XP to its
specifier. The dedicated head G0 will instruct the interface systems to interpret its
specifier as given information, and a convergent derivation is one where the
attracted XP is compatible with such an interpretive property; otherwise, the
derivation crashes.

3.1 An example derivation

Let us examine how DC sentences can be derived making use of these assumptions,
with a concrete example.

(17) ngo5 sik6-zo2 faan6 laa3 ngo5


1.SG eat-PFV rice SP 1.SG
‘I have eaten.’

First, the TP ngo5 sik6-zo2 faan6 is built. As G0 merges with TP, the [•XP•] feature
which it bears will attract an XP to its specifier. Suppose that it attracts TP. Notice
that as the functional head G0 and TP enter into a Spec-Head relation, the TP will
satisfy the ‘Given Criterion’ and be inaccessible to further movement operations,

2
A reviewer wonders whether the current assumption is compatible with the claim that Chinese SPs may
be enclitics (Lin 2010). Notably, Lin (2010) likewise posits an edge feature on the Mandarin SP le, which
triggers vP-movement that would later satisfy the SP’s encliticization requirements. For an overview of
the syntax of Chinese SPs, see Simpson (2014).

123
Parallel copying in dislocation copying

exemplifying the effects of Criterial Freezing (Rizzi 2006, 2010).3 For example, the
complex wh-phrase in (19a), which has satisfied the Q-Criterion, may not undergo
further movement (these facts are originally from Lasnik and Saito 1984, 1992).4
(19) a. John wonders [which book Q [Mary likes \which book[]]
b. *which book does John wonder [\which book[ Q [Mary likes
\which book[]]?

As C0 is merged into the structure, it thus may not attract the TP-copy in Spec-GP,
since that TP-copy, having satisfied a ‘Criterion’, is frozen in place. In order to
discharge its [•TP•] feature, C0 therefore attracts the same syntactic object TPk
which G0 has targeted. As a result, two parallel Ā-chains (TPi, TPk) and (TPj, TPk)
are sequentially formed.
The entire structure in the workspace then undergoes Transfer. At the semantic
component, the TP-copy in Spec-GP is interpreted as given information (an instruction
executed by the dedicated head G0), which in this case results in non-deviant
interpretation: given the presence of TPi within the same structure which may well serve
as its antecedent, TPj therefore (trivially) counts as given. At the phonological
component, a DC-specific deletion operation may apply to the TP-copy in Spec-GP, and
the VP sik6-zo2 faan6 which it contains is targeted for deletion. For reasons of chain
linearization, the tail of each of the two Ā-chains obligatorily deletes, as in the ordinary
case of Internal Merge (a process known as Chain Reduction; see Nunes 2004). The
derivation ultimately converges, yielding the illustrative example (17) as its output.
The current proposal involves three main ingredients: (i) an additional functional
projection in the DC structure; (ii) parallel chain formation; and (iii) a deletion
operation which applies to the TP-copy in Spec-GP. In what follows, I motivate
each of these ingredients in turn.

3.2 The nature of the additional functional projection

The exact identity of the additional functional projection in the DC structure should
ideally reflect whatever grammatical differences there are in pairs like (20).

3
One may wonder how criterial heads can be distinguished from non-criterial heads under the present
system, since the embedded C0 in (i) below, which presumably also bears an edge feature, does not induce
freezing.
(i) which book does John think Mary likes?

This is perhaps related to the more general question of how criterial freezing can be enforced under a free-
Merge system. For a recent proposal, see Gallego (2018).
4
A reviewer raises the question of whether (19b) would be incorrectly generated with parallel chains.
One prerequisite for parallel chain formation is the presence of two probes which target the same
syntactic object. In (19b), one of those two probes (i.e. the wh-probe on the matrix C0) is simply too far
away to target the lowermost copy which book, and hence such a derivation would be ruled out on
standard phase-theoretic assumptions. Moreover, there is evidence that the English interrogative C0 may
not be assigned any edge feature.
(i) *John wonders to Susan which book Mary gave.

123
J. Y.-K. Lai

(20) a. ngo5dei6 ting1jat6 haau2si3 aa3 [ordinary sentence]


1.PL tomorrow exam SP
‘We are having an examination tomorrow.’
b. ngo5dei6 ting1jat6 haau2si3 aa3 ngo5dei6 [DC]
1.PL tomorrow exam SP 1.PL
‘We are having an examination tomorrow.’

Although it is hard to exactly characterize the interpretive differences between (20a)


and (20b), native speakers share the opinion that the DC example expresses
additional emphasis on the subject DP. This accords also with Cheung’s (2015)
characterization of the cauda as expressing ‘minor emphasis’ (ibid.:262).
Nevertheless, the problem with such impressionistic statements is that it remains
unclear whether the emphatic effect accurately reflects the exact semantic
contribution of the additional functional projection, for the emphatic effect may
well arise as a result of repetition of syntactic material. In this connection, it is worth
noting that no restriction exists on the information-structural status of the copied
material. As (21)–(22) demonstrate, both moved and base-generated topics may
serve as the cauda.

(21) pin1 man2 ngo5 tai2-zo2 laa3, pin1 man2


CL article 1.SG read-PFV SP CL article
‘The article, I have read.’
(22) di1 saam1, go3go3 dou1 gaan2 go2 gin6 lo1, di1 saam1
CL garment everyone all choose that CL SP CL garment
‘The garments, everyone chooses that one.’

Moreover, the cauda is also compatible with inherently focussed elements and
exhaustive focus, as (23)–(24) illustrate.5

(23) bin1go3 seng4jat6 gong2 daai6waa6 aa3 bin1go3?


who always tell lie SP who
‘Who always lies?’
(24) zing6hai6 Ming4zai2 lai4-zo2 zaa3 zing6hai6 Ming4zai2
only Ming come-PFV SP only Ming
‘Only Ming came.’

If cauda formation involved direct extraction of the material serving as the cauda
itself, it would be impossible to maintain the claim that the additional functional
projection, which I will here refer to simply as ‘FP’, is a TopP or a FocP, since the
cauda may in fact host both topic and focus elements. Although it is possible to
claim that this FP is sometimes a TopP (21–22) and sometimes a FocP (23–24), we
would then be missing the generalization that the cauda is insensitive to the
information-structural function which its overt correlate bears.

5
I am grateful to a reviewer for pointing out these examples.

123
Parallel copying in dislocation copying

On the other hand, facts like (21)–(24) do not exactly bear on the question of the
nature of FP under the current account, since it is the TP containing the cauda, rather
than the cauda itself, which enters into a Spec-Head relation with the head of FP.
Nevertheless, the assumption that FP is GP is especially natural in this case, for the
TP-copy in Spec-GP would always count as given by virtue of the existence of a
distinct TP-copy in Spec-CP which may serve as its antecedent. The notion of
givenness is well grounded in the semantic and information-structural literature; for
related discussion, see Rochemont (1986, 2013a, 2013b, 2016); Selkirk (1995);
Williams (1997); Schwarzschild (1999); Sauerland (2005); Wagner (2006); Beaver
and Clark (2008); Kučerová (2012) among others.6
Crucially, the current additional functional projection is not tailor-made to
capture simply the DC facts. As I will demonstrate in Sect. 6, the exact same
functional projection is involved in the derivation of another type of dislocation in
Cantonese, and hence GP is not specific to the DC structure.

3.3 Parallel chains

Chomsky (2008) proposes that the derivation of English wh-questions like (25a)
involves the formation of parallel chains. Contrary to traditional accounts, the wh-
phrase who in (25a) does not undergo movement from Spec-TP to Spec-CP.

(25) a. who supported Mary?


b. whoi [C [whoj [T [whok v* [support Mary]]]]]

The parallel chain analysis, according to Chomsky, enables the elimination of mixed
chains of the Ā-A-A type. In the current example, two independent chains (whoi,
whok) and (whoj, whok) anchored to the same foot are formed throughout the
derivation. Thus the wh-copies whoi and whoj in (25b), being links of two separate
chains, do not bear any relation to one another. Subsequent works have utilized
parallel chain formation in handling various repetition phenomena in different
languages (Kandybowicz 2008; Aboh and Dyakonova 2009; Lohndal 2010; Duncan
2016 among others).
The notion of parallel chain formation can be characterized as follows, taken
from Kandybowicz (2008).

(26) Two chains α and β are related by parallel chain formation if and only if:
i. Tail (α)=Tail (β), and
ii. Head (α)≠Head (β)
(Kandybowicz 2008:115)

It is worth emphasizing that the current proposal for Cantonese DC requires parallel
chain formation, but not parallel movement. The two notions are distinct, as
Kandybowicz (2008) emphasizes, and a parallel chain need not be formed via

6
Two reviewers wonder how (23)–(24) can be accommodated under the current assumption. It is worth
noting that a given phrase can indeed contain a non-given phrase (see Selkirk 1995; Schwarzschild 1999
a.o.).

123
J. Y.-K. Lai

parallel movement (i.e. simultaneously). As long as the same syntactic object α is


targeted by two distinct probes, a parallel chain would obtain. In the proposed
derivation (19), there is no reason why G0, which inherently bears a structure-
building feature, should not immediately trigger movement after it is merged into
the structure. As such, the current analysis involves the creation of two parallel Ā-
chains which are formed one after another.
Kandybowicz points out that even in cases like (25a), it is not logically necessary
for the parallel chains (whoi, whok) and (whoj, whok) to be simultaneously formed.
Under the assumption of feature inheritance (Chomsky 2008), the formation of
(whoi, whok) may still precede that of (whoj, whok), given that the latter chain is
created only when T0 inherits the ϕ-probe from the phase head C0 after its merger.

3.4 The nature of the deletion operation

I assume that a DC-specific deletion operation may apply to the TP-copy in Spec-
GP, subject to the following standard condition.

(27) CONDITION ON DELETION


Deletion targets a syntactic constituent.

Given that cauda formation involves deletion, one immediate expectation is that the
cauda need not be a syntactic constituent. This expectation is borne out, as (28)
exemplifies.

(28) keoi5 ting3jat6 wui5 to1dei6 aa3, keoi5 ting3jat6 wui5


3.SG tomorrow will mop-floor SP 3.SG tomorrow will
‘He will mop the floor tomorrow.’

The grammaticality of (28), whose cauda is the remnant of a deletion operation


targeting the VP to1dei6 within the TP-copy in Spec-GP, is expected. Under the
current proposal, no non-constituent extraction need be admitted in (28) and other
similar cases.
There are two further properties of the deletion operation worth mentioning.

3.4.1 Property #1: the deletion operation is not obligatory

The DC-specific deletion operation is optional. In the absence of deletion, a DC


sentence with a full cauda would obtain, as (29) illustrates.

(29) keoi5 zung1ji3 ngo5 aa3, keoi5 zung1ji3 ngo5


3.SG like 1.SG SP 3.SG like 1.SG
‘He likes me.’

Certainly, the sceptic would raise the question of whether sentences like (29) are not
simply two independent sentences in juxtaposition. In fact, such a parse cannot be
excluded, given the existence of a phonologically null SP (hereafter ‘∅SP’) in the

123
Parallel copying in dislocation copying

language, as noted in Matthews and Yip (2011). The question is whether there is any
indication that the DC parse is also available for strings like (29).
It turns out not to be a straightforward matter to tease apart the DC parse and the
two-independent-sentence parse, but there is in fact a way. What we need are cases
in which the two-independent-sentence parse is impossible. A relevant example
would be the following.

(30) keoi5 zung1ji3 ngo5 wo5, keoi5 zung1ji3 ngo5


3.SG like 1.SG SP 3.SG like 1.SG
‘He likes me (I heard).’

The Cantonese ∅SP encodes a more restricted range of meanings as compared to


phonologically overt SPs, and its use often constitutes a blunt way of assertion
(Matthews and Yip 2011). Crucially, ∅SP does not encode evidentiality.

(31) keoi5 zung1ji3 ngo5 ∅SP. #ngo5 m4 zi1 keoi5 zung1-m4-zung1ji3


3.SG like 1.SG 1.SG NEG know 3.SG like-NEG
ngo5
1.SG
‘He likes me. #I don’t know if he likes me.’

The use of ∅SP necessarily commits the speaker to the truth of the proposition, and
hence the continuation in (31) is infelicitous. By contrast, the same continuation is
possible when the same sentence ends in the evidential SP wo5.

(32) keoi5 zung1ji3 ngo5 wo5. ngo5 m4 zi1 keoi5 zung1-m4-zung1ji3


3.SG like 1.SG SP 1.SG NEG know 3.SG like-NEG
ngo5
1.SG
‘He likes me (I heard). I don’t know if he likes me.’

Returning to (30), it is clear that this example cannot have the two-independent-
sentence parse, or else it would result in pragmatic anomaly: the speaker is first non-
committal to the truth of the proposition ‘he likes me’ (the use of wo5 requires that
the speaker not have first-hand information), and then immediately expresses
commitment to the truth of the same proposition (which indicates that the speaker
has sufficient information to assert the proposition). Hence, the felicity of examples
like (30) constitutes evidence that a full cauda is possible with DC sentences, in
perfect line with the current claim about the optional nature of the deletion
operation.
This fact also relates to a reviewer’s question as to the motivation behind positing
an additional functional projection in the DC structure (i.e. GP). Let us suppose that
this assumption was superfluous. The derivation of DC sentences with a full cauda
like (30) would then involve full spell-out of the two TP-copies in (33).

123
J. Y.-K. Lai

However, this would considerably complicate the standard syntax–PF mapping


(Kayne 1994; Nunes 2004) in the sense that Internal Merge does not normally result
in full spell-out of more than one chain link (cf. *That Mrs. Owen will resign is
likely that Mrs. Owen will resign). Moreover, the cause of multiple spelled-out
copies is generally attributed to postsyntactic reasons, such as morphological repair
(van Riemsdijk 1989; Landau 2006) and morphological fusion (Nunes 2004), none
of which appears to be justifiable in the current case. And even if one is willing to
posit such a postsyntactic cause, notice that one must also claim that the
postsyntactic process at issue (e.g. morphological repair or morphological fusion) is
optional, or else the bottom TP-copy in (33) would always be spelt out, leading to
the problem of massive undergeneration.7

3.4.2 Property #2: the deletion operation is DC-specific

One may wonder why the deletion operation is referred to as being ‘DC-specific’. In
fact, there is good reason to consider the current deletion operation as special. The
new observation is that the deletion operation at issue may in fact target material
which otherwise may not be targeted for deletion.

(34) ngo5 jau6 si3haa6 sin1 ngo5 jau6


1.SG again try SP 1.SG again
‘Let me try as well.’
(35) nei5 dou3dai2 heoi3-zo2 bin1dou6 aa3 nei5 dou3dai2?
2.SG DOUDAI go-PFV where Q 2.SG DOUDAI
‘Where the hell did you go?’ (Cheung 2015:258; ex. (104))

Contrast the above DC examples with the canonical examples below, which seem to
indicate that VP-ellipsis may not generally strand such adverbials as jau6 and
dou3dai2.

7
A reviewer notes that Cheng (2007) posits a null element that optionally fuses with a V-copy in her
account of Mandarin verb copying which pursues the option of multiple chain-link spell-out. Crucially,
Cheng demonstrates that the posited null element HAVE is related to you ‘have’ which may immediately
precede durative/frequency expressions (ibid.:165). The postulation of HAVE or any other null element in
the case of Cantonese DC, as far as I can see, would only serve the purpose of rendering two links of a
movement chain distinct. See Sect. 6 for further advantages of postulating GP.

123
Parallel copying in dislocation copying

(36) ngo5 jau6 si3haa6, keoi5 jau6 *(si3haa6)


1.SG again try 3.SG again try
‘Let me try, and let her try.’
(37) Ming4zai2 dou3dai2 heoi3-zo2 bin1dou6? Aa3koeng4 dou3dai2
Ming DOUDAI go-PFV where Koeng DOUDAI
*(heoi3-zo2 bin1dou6)?
go-PFV where
‘Where the hell did Ming go? Where the hell did Koeng go?’

The sharp contrast between (34)–(35) and (36)–(37) therefore indicates that a
distinction must be drawn between DC and non-DC cases, such that the deletion
operation we saw in action in (34)–(35) would not be available otherwise. The
current proposal makes exactly the right kind of distinction without further proviso:
the derivation of examples like (36)–(37) simply cannot involve parallel chain
formation, a natural consequence of the fact that the subjects in each of these
examples are distinct. Because (36)–(37) cannot be derived in the same way as
(34)–(35), there is thus at least a reason why the deletion operation in question
should not be available in the former case. More specifically, I assume that the
functional head G0 is involved in licensing the DC-specific deletion, which is
related to the fact that it is typically given material which can be elided. It therefore
follows that the deletion operation may not operate backwards, as the following
examples demonstrate.

(38) *ngo5 jau6 sin1 ngo5 jau6 si3haa6


1.SG again SP 1.SG again try
‘Let me try as well.’
(39) *nei5 dou3dai2 aa3 nei5 dou3dai2 heoi3-zo2 bin1dou6?
2.SG DOUDAI Q 2.SG DOUDAI go-PFV where
‘Where the hell did you go?’

On the other hand, it would be a complete mystery why the deletion operation
available in (34)–(35) is simply unavailable in (36)–(37) if all these examples
underlyingly involved the exact same structure. To correctly distinguish between
(34)–(35) and (36)–(37), the only option left would be to claim that the deletion
operation in question is available only when the subjects are identical. The serious
question that this claim would lead to, however, is why the identity of the subjects
should matter in the licensing of VP-ellipsis. VP-ellipsis succeeds in the following
example, even though the subjects Mary and Susan are clearly distinct (see
Merchant 2019 for an overview of standard approaches towards ellipsis).

(40) Mary will read LGB. Susan will Δ, too. [Δ = read LGB]

Under this alternative view, then, it is clear that the current deletion operation can
only be restricted to DC cases via brute force: through building a mysterious
identical subject requirement into the deletion operation itself.

123
J. Y.-K. Lai

4 Consequences

This section first demonstrates how the core grammatical properties of Cantonese
DC noted in Sect. 2 fall neatly under the proposed account, and then explores other
consequences of the proposal.

4.1 Deriving Cantonese DC properties

Recall the following properties of a Cantonese DC sentence: (i) the cauda may not
contain an overt SP; (ii) the cauda observes a structural object restriction; and (iii)
the SP must be phonologically overt.
Property (i), otherwise puzzling, is straightforward: there is only one C0 in the
proposed Cantonese DC structure. This constitutes an argument in favor of the
current proposal, since any Move-and-Elide approach towards Cantonese DC, with
the necessary concomitant assumption of biclausality, faces the challenge of
explaining why the C0 heading the second CP (which contains the cauda) cannot
host a phonologically overt SP. Under such a view, the impossible occurrence of
overt SPs in the cauda must be stipulated.
Property (ii) is the most challenging to explain. Why should an object phrase not
be able to serve as the cauda? If the cauda were the result of an XP having evacuated
from an ellipsis site, it would not be evident why the object DP may not move, i.e.
[CP DPobj [TP DPsubj V \DPobj[]]. If a Move-and-Elide analysis were correct for
Cantonese DC, some additional machinery would then be necessary for capturing
this curious object restriction. Notice that other constructions shown to fit a Move-
and-Elide analysis do not observe the same structural restriction, such as the
following English examples of sluicing (Merchant 2001).

(41) a. Someone scolded Peter yesterday, but I don’t remember who.


b. Mary scolded someone yesterday, but I don’t remember who.

By contrast, the current structural restriction on the cauda immediately falls under
the current proposal. Because DC formation involves attraction of TP into Spec-GP,
the subject DP and the verb in that TP-copy must have been deleted in order for only
an object phrase to surface as the cauda. This violates the condition in (27), which
requires that deletion target syntactic constituents (see also Sailor and Thoms 2014
who argue that apparent cases of non-constituent ellipsis are illusory).

(42) *[CP TPi [C[•TP•] [GP [TPj \DPsubj V[ DPobj] [G[•XP•] \TPk[]]]]

The current proposal also leads to the expectation that there should exist no
blanket ban on object caudae, which ought to be possible as long as the deletion
condition in (27) is observed. In fact, we have already seen DC examples with an
object cauda in (21)–(22), one of which is reproduced below.

123
Parallel copying in dislocation copying

(43) pin1 man2 ngo5 tai2-zo2 laa3, pin1 man2


CL article 1.SG read-PFV SP CL article
‘The article, I have read.’

The grammaticality of this example straightforwardly follows from the fact that the
TP-copy occupying Spec-GP in the structure for (43) contains a syntactic
constituent which can be targeted for deletion, as illustrated in (44).

(44) [CP TPi [C[•TP•] [GP [TPj pin1 man2 \[TP ngo5 tai2-zo2 pin1 man2][]
[G[•XP•] …

As for the final property (iii), I assume that ∅SP does not bear any structure-
building feature, unlike its overt counterparts, and hence the TP selected by ∅SP
remains in-situ. There exist no known facts against this assumption, and moreover,
the fact that ∅SP is phonologically null means that no detectable effect at PF would
ensue whether or not TP has undergone movement. Accordingly, if ∅SP is merged in
the proposed derivation (19) in lieu of an overt SP, parallel chain formation would
not ensue, and hence sentences like (45) without an overt SP remain underivable
under the current system.

(45) *keoi5 zung1ji3 nei5 keoi5


3.SG like 2.SG 3.SG
‘He likes you.’

What is crucial to note here is that even if the exact same assumption is adopted for
∅SP, the Move-and-Elide analysis still falls short of explaining why Cantonese DC
sentences should require a phonologically overt SP (see Sect. 5.2 for further
discussion).

4.2 The irrelevance of islandhood

Given that the formation of the cauda involves deletion but not direct extraction
under the current proposal, the account predicts that islandhood is not a relevant
factor governing cauda formation. Interestingly, the prediction is borne out.

(46) a. keoi5 zi1dou6 [ngo5 zung1ji3 nei5] me1 {keoi5/ *ngo5}?


3.SG know 1.SG like 2.SG Q 3.SG 1.SG
‘Is it true that he knows that I like you?’
b. keoi5 zi1dou6 [[ngo5 zung1ji3 nei5] ge3 ni1-go3 si6sat6] me1
3.SG know 1.SG like 2.SG GE this-CL fact Q
{keoi5/*ngo5}?
3.SG 1.SG
‘Is it true that he knows the fact that I like you?’

The two examples differ solely in the island status of the complement selected by
the verb zi1dou6. In (46a), the matrix verb selects a clausal complement, which is a

123
J. Y.-K. Lai

non-island, whereas in (46b) the matrix verb selects a complex DP as its


complement, which is an island (Ross 1967). The fact that the embedded subject
ngo5 may not serve as the cauda in both cases is telling: under the Move-and-Elide
analysis, when ngo5 serves as the cauda, one should expect (46a) but not (46b) to be
grammatical.8 Under the current account, the reason why the embedded subject in
both cases cannot serve as the cauda again follows from the fact that non-constituent
deletion is involved, as illustrated in (47) for (46a).

(47) *[CP TPi [C[•TP•] [GP [TPj\keoi5 zi1dou6[ ngo5 \zung1ji3 nei5[] [G[•XP•] …

The judgments reported in (46a–b) are shared by my consultants. In fact, the


literature also reports a similar pattern of judgment. For instance, Cheung (2015)
documents the following example.

(48) Mary jing6wai4 [John hai6 go3 fei1gei1si1] me1, {OKMary/


Mary think John COP CL pilot Q Mary
?
John}?
John
‘Mary thinks that John is a pilot?’ (Cheung 2015:243, ex. (48); tones added)

Cheung remarks that in cases where an embedded subject (e.g. John in (48)) serves
as the cauda, native speakers ‘find them slightly degraded; others, quite marginal’
(ibid.:243). The marginality of such examples, which is unexpected under the
Move-and-Elide analysis per se, thus constitutes another argument for the current
proposal.
Other island examples should be viewed in this new light. Examples involving a
relative clause island and an adjunct island, such as (49) and (50) respectively, are
ungrammatical not because of extraction out of an island, but because of non-
constituent deletion within the TP-copy occupying Spec-GP. These examples no
longer constitute definitive evidence for a movement account.

(49) Relative Clause Island


*ngo5 sik1-saai3 [[RC keoi5 lyun4lok3-gwo3 __ ge3] go2-di1
1.SG know-all 3.SG contact-EXP GE that-CL
gong2ze2] gaa3, keoi5
speaker SP 3.SG
‘I know all the speakers that he contacted.’
(Cheung 2015:234, ex. (20b); tones added)

8
In fact, if repair-by-ellipsis exists (Chomsky 1972; Lasnik 2001; Fox and Lasnik 2003; Merchant 2008
among many others), one should also expect (46b) with ngo5 to be grammatical, contrary to fact. One
must thus assume that the deletion involved in the derivation of (46b), after extraction of ngo5, is one that
does not void island violation. Since an explanation is conceivable here, I will ignore this potential
problem for proponents of the Move-and-Elide analysis.

123
Parallel copying in dislocation copying

(50) Adjunct Island


*[jyu4gwo2 keoi5 ji5ging1 heoi3-gwo3], ngo5
if 3.SG already go-EXP 1.SG
ting3jat6 zau6 m4 heoi3 laa1, ji5ging1
tomorrow then NEG go SP already
‘If he has already been there, I will not go there tomorrow.’
(ibid., ex. (21b); tones added)

4.3 Apparent counterexamples

There exist DC examples which apparently challenge the claim that only syntactic
constituents may elide within the TP-copy in Spec-GP, however.

(51) a. keoi5dei6 wui5 heoi3 gaa3laa3, wui5


3.PL will go SP will
‘They will go.’
b. nei5 jing1goi1 tung4 jan4 gong2 ge3, jing1goi1
2.SG should with person speak SP should
‘You should speak to someone (about it).’

The caudae of these examples each contain only a modal, which is commonly
analyzed as a head in Chinese (Lin and Tang 1995; Lin 2012). The issue with these
examples is that they apparently involve deletion of both the subject DP and the VP,
which do not form a syntactic constituent. This is not necessarily a problem,
however, if we adopt the standard realizational framework of Distributed
Morphology (Halle and Marantz 1993, 1994), under which syntax does not
manipulate lexical items with phonological content, but rather abstract bundles of
morphosyntactic features. Accordingly, the phonological representations of the
subjects keoi5dei6 and nei5 are not initially present within the TP-copy in Spec-GP.
The fact that the caudae of (51a–b) are ‘subjectless’ can then be linked to the
independent availability of Pro in Chinese languages (Huang 1982, 1984 and
subsequent works). The subject within the TP-copy, as such, does not undergo DC-
specific deletion, as illustrated in (52) after Vocabulary Insertion for (51a).

(52) [CP keoi5dei6 wui5 heoi3 [Cʹ gaa3laa3 [GP [Pro wui5 \[VP heoi3][]
[G[•XP•] …

A further support for the current claim is that the covert subject can in fact be made
overt.

123
J. Y.-K. Lai

(53) a. keoi5dei6 wui5 heoi3 gaa3laa3, keoi5dei6 wui5


3.PL will go SP 3.PL will
b. nei5 jing1goi1 tung4 jan4 gong2 ge3, nei5 jing1goi1
2.SG should with person speak SP 2.SG should

A reviewer moreover notes that in cases where Pro is unavailable, it is indeed the
case that the corresponding DC example with a ‘subjectless’ cauda is ungrammat-
ical.

(54) a. *gok3dak1 keoi5 hou2 gui6 aa3


think 3.SG very tired SP
Intended: ‘(I) think that he is tired.’
b. *ngo5 gok3dak1 keoi5 hou2 gui6 aa3 gok3dak1
1.SG think 3.SG very tired SP think
‘I think he is tired.’

The sceptical reader might wonder whether we now lose our account for the
object restriction on the cauda, since one may assume that in such cases the subject
in the TP-copy occupying Spec-GP is likewise a Pro, and it is only V which
undergoes deletion. Although heads, by definition, are syntactic constituents, I will
assume that only maximal projections may elide, following Johnson (2004), Baltin
(2012) and others. Even for the gapping construction, which apparently involves
deletion of a verb, few contemporary advocates of deletion-based analysis would
analyze gapping as simply involving V-ellipsis (see Coppock 2001 for example,
who analyzes English gapping as involving VP-ellipsis). Nonetheless, the current
line of reasoning is not as strong as the empirical force of the following example.

(55) *ngo5 gin3dou2 keoi5 aa3 ngo5 keoi5


1.SG see 3.SG SP 1.SG 3.SG
Intended: ‘I saw him.’

Sentences like (55) are sharply ungrammatical, in perfect consonance with the
assumption that X0-ellipsis does not exist.
Another potential challenge comes from DC examples with a nonreferential
subject in the host clause and a ‘subjectless’ cauda.

(56) a. bin1go3 wui5 heoi3 waa2 wui5?


who will go SP will
‘Who will go?’
b. mou5jan4 sing1kei4jat6 faan1lai4 zou6je5 gaa4, sing1kei4jat6?
no-person Sunday come_back work SP Sunday
‘Nobody comes back to work on Sunday?’
(Cheung 2015:255, ex. (93); tones added)

Crucially, these examples become degraded when the hypothesized covert subject is
realized, in contrast to the previous case (cf. (53a–b)).

123
Parallel copying in dislocation copying

(57) a. *bin1go3 wui5 heoi3 waa2 keoi5 wui5?


who will go SP 3.SG will
b. *mou5jan4 sing1kei4jat6 faan1lai4 zou6je5 gaa4, keoi5
no-person Sunday come_back work SP 3.SG
sing1kei4jat6?
Sunday

Nevertheless, the ill-formedness of the above examples can be attributed to


Montalbetti’s (1984) Overt Pronoun Constraint (hereafter ‘OPC’), which states that
‘[o]vert pronouns cannot link to formal variables if and only if the alternation overt/
empty obtains’, as formulated in Huang (1991:67). Simply put, the OPC encodes the
fact that there are cases where a pronominal cannot be made overt. (58a–b)
demonstrate that the OPC holds for Cantonese (modelled on Huang’s 1991:68
Mandarin examples).

(58) a. bin1go3i hei1mong6 {*keoi5i/Proi} ho2ji5 tung4 Aa1fan1


who hope 3.SG can with Fan
aak1sau2 waa2?
shake-hand SP
‘Who hopes that he can shake hands with Fan?’
b. mou5jan4i waa6 {*keoi5i/Proi} wui5 lei4
no-person say 3.SG will come
‘Nobody says that he will come.’

Hence, the facts in (57a–b) do not invalidate the claim that apparently ‘subjectless’
caudae are the illusion created by the covert pronominal; their ungrammaticality has
an independent cause. Moreover, there is no issue with identifying the missing
subjects in the caudae of (56a–b) as Pro; these missing subjects will have the same
status as the ones in (58a–b).
The following pair of examples apparently creates problems of a similar kind.

(59) a. keoi5 ting1jat6 wui5 heoi3 gaa3laa3, wui5


3.SG tomorrow will go SP will
‘He will go tomorrow.’
b. bin1go3 gam1jat6 jing1goi1 wan2 nei5 aa3, jing1goi1?
who today should find 2.SG Q should
‘Who should see you today?’

It appears that in these cases, the subject DP, the temporal NP and the VP are elided
altogether within the TP-copy in Spec-GP, which certainly do not form a syntactic
constituent. Instead of relaxing the current constraint on deletion in (27), let us
utilize an operation available in the minimalist parlance known as late merge
(Chomsky 1995; Bhatt and Pancheva 2004; Takahashi and Hulsey 2009; Fox 2014;
Overfelt 2017 among others). Accordingly, the temporal adjuncts ting1jat6 in (59a)
and gam1jat6 in (59b) are not initially present (in the lowermost TP-copy), but are

123
J. Y.-K. Lai

instead introduced counter-cyclically into the TP-copy occupying Spec-CP. Hence,


these adjuncts are not contained in the TP-copy in Spec-GP, and only VP-deletion is
involved in cauda formation.
One of the initial motivations for late merge comes from a well-known
asymmetry between arguments and adjuncts regarding connectivity effects. For
instance, (60a) surprisingly exhibits no Condition C effects, in contrast to (60b). The
standard late-merge account holds that the relative clause in (60a) is merged late
into the structure, whereas the complement clause in (60b), given its argumental
status, cannot be, and hence only the former obviates Condition C effects.

(60) a. Which report that Johni revised did hei submit?


b. *Which report that Johni was incompetent did hei submit?
(Freidin 1986:179, exx. (76a–b))

I will adopt the standard assumption that only adjuncts can be merged late. Non-
adjuncts, including any element which constitutes part of the Hierarchy of
Projections (in the sense of Adger 2003), may not undergo late merge. This explains
why late-merging the experiential aspect marker -gwo3 and the negation m4 in the
TP-copies in Spec-CP in (62a) and (62b) respectively do not lead to convergence.9

(61) Hierarchy of Projections


T 〉 (Neg) 〉 (Perf) 〉 (Prog) 〉 (Pass) 〉 v 〉 V
(Adger 2003:239)
(62) a. *keoi5 hok6-gwo3 gung1fu1 aa3 keoi5 hok gung1fu1
3.SG learn-EXP kung_fu SP 3.SG learn kung_fu
‘She once learnt kung fu.’
b. *nei5 m4 ho2ji5 zau2 gaa3, ho2ji5
2.SG NEG can leave SP can
‘You cannot leave.’

9
One may wonder whether it is possible to late-merge a topic to the TP-copy in Spec-GP, given that the
current proposal assumes the TP-adjunction view of Chinese topics (Tang 1990).
(i) [CP [TP …] [ C[•TP•] [GP [TP topic\[TP …][] G[•XP•] …
⇑ (late merge)

As a reviewer notes, such a derivation would incorrectly permit examples like (ii).
(ii) *keoi5 zung1ji3 mei5gwok3 caang2 gaa3 caang2
3.SG like America orange SP orange
Intended: ‘Oranges, he likes American oranges.’

This may perhaps suggest that Chinese topics should occupy Spec-TopP (Gasde and Paul 1996; Paul
2014, 2015, 2016), which, as specifiers, cannot undergo late merge. This would entail the trivial
modification that C0 bears the structure-building feature [•FinP•], with TopP being situated below FinP (i.
e. CP[GP[FinP[TopP[TP). We can then suggest that a dangling topic may not have a zero
exponent. Since this has no impact on the core proposal, I will continue to assume that Chinese topics are
adjuncts.

123
Parallel copying in dislocation copying

We have, therefore, successfully defused certain apparent threats of undergener-


ation, and the current assumptions on cauda formation can be maintained as is.

5 The existing accounts

To the best of my knowledge, there exist only two formal accounts of Cantonese DC
in the literature (Tang 2015; Cheung 2015). This section reviews the existing
accounts, with the aim of comparing them with the proposed analysis.

5.1 The generalized schema account (Tang 2015)

Contrary to widespread consensus, Tang (2015) proposes that Chinese SPs are not
X0-elements, but are secondary predicates occupying the structural position labelled
‘XP’ in the generalized syntactic schema illustrated below.

(63) [FP YP [Fʹ F XP]]

Under this view, the Chinese SP projects a maximal projection which occupies the
complement position of F0 in the generalized schema. Moreover, F0 in the schema is
taken to encode a coordinative relation, realizable as the disjunctor or in languages
like English (Tang 2015:22, fn. q). As an illustration, the Cantonese example (64)
would have the structure shown in (65).

(64) keoi5 zau2-zo2 laa3


3.SG leave-PFV SP
‘He has left.’

In later sections of the paper, Tang (2015:16) suggests in passing that the proposed
generalized schema in (63) also extends to Cantonese DC sentences. The DC
example (66), accordingly, would have the schematic structure in (67).

123
J. Y.-K. Lai

(66) ngo5 ting1jat6 gaau1 man2 aa3, ngo5 ting1jat6


1.SG tomorrow submit paper SP 1.SG tomorrow
‘I will submit a paper tomorrow.’
(Tang 2015:16, ex. (64))
(67) [Subject Adverb VP] F [Subject Adverb VP]
(ibid.:16, ex. (66))

Tang essentially assumes the same analysis of Cantonese DC in his subsequent


works (see Tang 2018). Although such a proposal is both simple and elegant, it fails
to capture two of the core properties noted in Sect. 2.
First, in order for the object restriction on cauda formation to fall squarely under
the generalized schema account, it is crucial that structures like (68) be illegitimate:
there, F0 first merges with a TP with an adjoined topic, and the cauda obtains after
deletion of the lower TP-segment. Notice that in order for the current account to
yield ordinary topic sentences without DC, a TP with an adjoined topic must be able
to serve as an argument of F0 (which would occupy the position labelled ‘YP’ in
(63)).

(68) *[TP ngo5 ting1jat6 gaau1 man2 aa3] F [TP man2 \[TP ngo5 ting1jat6
gaau1][]

Importantly, F0 in such structures as (68) does coordinate like categories. In fact,


there is reason to believe that F0 may coordinate unlike categories as well. In the
illustrative structure (65), for example, it is unlikely that YP (which probably is a
TP) and XP (the phrase projected by the SP) are like categories. Moreover, there is
no specific issue with coordinating an ordinary TP with a TP containing an adjoined
topic in Cantonese, as (69) demonstrates.10

(69) jyu4gwo2 [[keoi5 heoi3-gwo3 Faat3gwok3], waak6ze2 [Ngau1zau2


if 3.SG go-EXP France or Europe
kei4taa1 dei6fong1 keoi5 dou1 heoi3-gwo3]], nei5
other place 3.SG all go-EXP 2.SG
wui2-m4-wui2 seon3 aa3?
will-NEG-will believe Q
‘If he has been to France, or other places in Europe, he has all been to,
will you believe it?’

10
The embedding in (69) helps ensure the involvement of coordination, since the use of coordinators
itself does not guarantee that the sentence necessarily involves coordination. This is best seen in English
examples like (i).
(i) A: Mary likes statistics.
B: Or she likes mathematics.

123
Parallel copying in dislocation copying

Hence, although it is desirable for structures like (68) to be ill-formed, it is unclear


why they should be. The main problem is that there is no condition which
guarantees that the coordinates YP and XP in (65) be exactly identical. Such a
condition is also undesirable, in light of (69). By contrast, the same problem does
not arise under the present analysis, since the TP-copy in Spec-CP and that in Spec-
GP are exact copies of the same syntactic object α. Hence, both TP-copies will each
contain a topic only if α contains a topic itself.
A reviewer rightly notes that the source sentence in (68) is rather odd without
deletion. Nevertheless, my consultants unanimously judge (68) with deletion to be
considerably worse. The degree of oddness is therefore different in the two cases.
Moreover, there are cases where the source sentence is as odd as (68), but the
resulting DC sentence is perfect. The DC sentence (70), for example, has (71) as its
source under Tang’s account.

(70) pin1man2 keoi5 tai2-zo2 laa3 keoi5


article 3.SG read-PFV SP 3.SG
‘The article, he has read.’
(71) [TP pin1man2 keoi5 tai2-zo2 laa3] F [TP keoi5 tai2-zo2 pin1man2]

There is thus no clear correlation between the acceptability of the source sentence
without deletion and that of the resulting sentence after deletion. Hence, the object
restriction on cauda formation, under Tang’s account, cannot simply be attributed to
the oddness of the full source sentence in the first place.
In the context of the generalized schema account, the fact that a phonologically
overt SP is obligatory in Cantonese DC formation also means that the bracketed
clause preceding F in (67) must be another full generalized schema which contains
an XP projected by a phonologically overt SP, although not shown there. This
immediately raises the questions of (i) why the complement inside this schema
cannot be one projected by a phonologically null SP, as in (72a); and (ii) why the
F-preceding bracketed clause in (67) cannot contain no SP projection, as in (72b).

If the challenge of barring these structures is not met, the fact that a phonologically
overt SP is obligatory in a Cantonese DC sentence will be left unaccounted for. By
contrast, this fact immediately falls under the proposed account, since a

123
J. Y.-K. Lai

phonologically overt SP is necessary for triggering parallel chain formation, as


discussed in Sect. 4.1.

5.2 The Move-and-Elide account (Cheung 2015)

Cheung’s (2015) four-step derivation of Cantonese DC has the standard flavor of a


Move-and-Elide analysis. Two underlyingly full clauses CP1 and CP2 are involved,
and the cauda, or a larger constituent containing the cauda, evacuates from the
ellipsis site in CP2 prior to deletion. As an illustration, DC examples with a non-
constituent cauda like (73) have the derivation explicated in (74).

(73) keoi5 ting1jat6 wui5 heoi3 Sau2ji5 aa3, keoi5 ting1jat6


3.SG tomorrow will go Seoul SP 3.SG tomorrow
wui5
will
‘She is going to Seoul tomorrow.’
(74) Step 1: Juxtapose two parallel CPs
[CP1 aa3 keoi5 ting1jat6 wui5 heoi3 Sau2ji5] [CP2 aa3 keoi5 ting1jat6 wui5
heoi3 Sau2ji5]
Step 2: Apply αP-ellipsis in CP2, where αP={VP/ModP/TP/CP}
[in this case, αP=VP]

Step 3: Move δP (containing the αP) to the left periphery of CP2


[in this case, δP=TP]

Step 4: Elide the (lower) CP in CP2

123
Parallel copying in dislocation copying

It is the precise move from Step 3 to Step 4 in (74) which instantiates Move-and-
Elide: TP first undergoes Internal Merge to a higher structural position, followed by
deletion which applies to a constituent containing its lower copy.
The Move-and-Elide analysis is appealing, given that dislocation phenomena in
other languages have been proposed to involve a similar derivation (Tanaka 2001;
Ott 2014; Ko 2016; Ott and de Vries 2016).11 However, since a central aim of the
present paper is to exactly expose the inadequacy of a Move-and-Elide approach
towards Cantonese DC, the current account inevitably inherits at least some of the
problems raised earlier.
The Move-and-Elide analysis essentially claims that a Cantonese DC sentence
underlyingly comprises two independent sentences. In order to capture the
impossible occurrence of phonologically overt SPs in the cauda, the current
account resorts to the CP-deletion operation shown in Step 4. Notice that the
deletion operation must be obligatory, or else the current core property would be
lost, but this immediately raises the question of why the operation should be
obligatory. Moreover, notice that Cantonese otherwise readily allows for juxtapo-
sition of two identical sentences with the same phonologically overt SP, as
illustrated in (75).

(75) [Context: The speaker is excited about the fact that her friend is travelling
tomorrow.]
Keoi5 ting1jat6 wui5 heoi3 Sau2ji5 aa3. Keoi5 ting1jat6 wui5
3.SG tomorrow will go Seoul SP 3.SG tomorrow will
heoi3 Sau2ji5 aa3!
go Seoul SP
‘She is going to Seoul tomorrow. She is going to Seoul tomorrow!’

11
Ko (2016), for instance, examines gapless right-dislocation examples like (i), which resembles the
Korean counterpart of Cantonese DC. Importantly, (i) illustrates that the object DP sakwa-lul can be
repeated.
(i) Cheli-ka sakwa-lul mek-ess-e sakwa-lul Korean
Cheli-NOM apple-ACC eat-PST-DECL apple-ACC
‘Cheli ate an apple.’ (Ko 2016:4, ex. (2b))

123
J. Y.-K. Lai

As such, the obligatory nature of CP-deletion in Step 4 essentially amounts to a


stipulation that is unwarranted. While one might instead claim that the CP2 built in
Step 1 must contain a phonologically null SP, which would then dispense with
obligatory CP-deletion, the question remains as to why the SP contained CP2 is
necessarily phonologically covert in the first place.
Second, Cheung’s account also importantly assumes the αP-ellipsis operation in
Step 2, which intends to capture the object restriction on cauda formation. The
reasoning is that once a phrasal projection containing the object argument has
undergone ellipsis, the object argument itself will then be inaccessible to the
subsequent step ‘Move δP’ in Step 3. Although this might sound reasonable,
positing the αP-ellipsis operation in fact raises various non-trivial issues. Notice that
in order for Cheung’s account to successfully capture the object restriction, αP-
ellipsis must again be obligatory, and it also has to minimally target a VP (as noted
in Cheung 2015). There is no option for it not to apply, or else movement of a sole
object argument to the left periphery would be possible in Step 3. This raises the
same conceptual question which the CP-deletion operation in Step 4 raises, namely
why αP-ellipsis should be obligatory, as well as the important question as to why the
obligatory αP-ellipsis operation must target at least a VP. And even if there turns out
to be a way of motivating the obligatory nature and the requisite size of αP-ellipsis,
it is still crucial that the partial derivation as sketched in (76) be blocked under the
current account.

(76) Step 1.5: Move the object DP to the left periphery in CP2
[CP1 …] [CP2 [DP Obj] C [CP SP Subj V tDP]]
Step 2: Apply αP-ellipsis in CP2, where αP=VP
[CP1 …] [CP2 [DP Obj] C [CP SP Subj [VP V tDP]]]

In this derivation, an object first moves to the left periphery of CP2 before the
obligatory application of αP-ellipsis, which violates no obvious principles of
grammar. In order for Cheung’s account to really bear fruit, αP-ellipsis in Step 2
must precede any application of Internal Merge. Such an approach therefore
(implicitly) relies on an ordering of operations that is extrinsically imposed,
rendering its explanation of the object restriction unsatisfying.
Another issue with assuming αP-ellipsis to capture the object restriction is that, as
we saw earlier in Sect. 3.4.1, there are DC examples involving a full cauda like (30),
reproduced below.

(77) keoi5 zung1ji3 ngo5 wo5, keoi5 zung1ji3 ngo5


3.SG like 1.SG SP 3.SG like 1.SG
‘He likes me (I heard).’

As such, the posited αP-ellipsis operation cannot be obligatory, or else the Move-
and-Elide account undergenerates. However, once αP-ellipsis is assumed not to be
obligatory, the explanation for the object restriction on cauda formation is then
immediately lost. A virtue of the account proposed in the present paper is thus that it

123
Parallel copying in dislocation copying

naturally derives the object restriction on cauda formation without positing any
obligatory deletion operation, while at the same time being more empirically
adequate in that Cantonese DC caudae can, as we have seen, be full TPs.
Lastly, the Move-and-Elide analysis offers no ready account for the requirement
that Cantonese DC sentences must contain a phonologically overt SP. This is
because, given the existence of ∅SP in the language, there is no reason why CP1
may not contain such a null SP. More specifically, the current Move-and-Elide
analysis only requires that two parallel CPs be generated and juxtaposed, which
should therefore permit the (partial) derivation in (78).

(78) Step 1: Juxtapose two parallel CPs


*[CP1 ∅SP [TP …]] [CP2 ∅SP [TP …]]

Crucially, even if we adopt the same assumption as in the parallel chain account that
∅SP does not bear any structure-building feature, there is still no immediate reason
why the derivation in (78) should not lead to convergence. Hence, under the Move-
and-Elide analysis, it is again a mystery why the phonologically overt SP
requirement should hold of Cantonese DC.

5.3 Section summary

From the foregoing discussion, it is clear that the only two existing accounts of
Cantonese DC do not adequately capture most of its core properties. In comparison,
the account proposed here not only captures all of the DC properties noted in
Sect. 2, but crucially also does so without the aid of any unnatural machinery, and
is, therefore, more explanatory (see Sect. 4.1 for details).

6 Beyond Cantonese DC

I will now demonstrate another virtue of the proposed analysis of Cantonese DC as


compared to the two competing analyses reviewed in the previous section. Thus far,
we have considered the case where the functional head G0 attracts a TP. Given that
C0 would also attract the same TP, parallel chain formation ensues. In fact, G0
(which bears the general structure-building structure [•XP•] under the present
proposal) may in principle attract any XP to its specifier. Suppose that it attracts a
non-TP syntactic object, as schematized in (79).

123
J. Y.-K. Lai

Unlike the case where G0 attracts a TP syntactic object, the current derivation in
(79) involves no parallel chain formation, and will necessarily output sentences
where the pre-SP clause contains a gap, instantiating remnant movement (see
Müller 1998; Abels 2007; Thiersch 2017 among others). The current analysis,
proposed originally for Cantonese DC, therefore also predicts dislocation structures
of the following kind to exist in the language.

(80) a. Aa3koeng4 sik6-zo2 laa3, wun2 zuk1


Koeng eat-PFV SP CL congee
‘Koeng ate the bowl of congee.’
b. keoi5 wui5 hou2
3.SG will very
siu2sam1-gam2 tai2 gaa3, go2-bun2 syu1
careful-way read SP that-CL book
‘He will read that book very carefully.’

Sentences like (80a–b), as it turns out, exactly exemplify the kind of dislocation
structure analyzed in Lee (2017), which he refers to as a ‘subtype of right
dislocation’ in Cantonese. I will refer to this type of dislocation construction as
‘RDO’, given that the pre-SP clause contains an object gap.
The prediction of the current proposal, therefore, is borne out; sentences such as
(80a–b) do exist in Cantonese.12 It is immediately worth noting that the existing
accounts of Cantonese DC (Tang 2015; Cheung 2015), which analyze the DC
construction very differently, do not appear to have the same potential as the current
proposal in establishing potential connections with other types of dislocation that
exist in the language. While Cantonese DC, with a gapless host clause, may appear
to be the ‘odd’ one out, it is actually not.
In this connection, a reviewer offers the interesting suggestion that the object
restriction which holds of Cantonese DC can be explained if extraction of an XP in

12
In fact, details aside, Lee (2017) independently arrives at a very similar analysis for Cantonese RDO.

123
Parallel copying in dislocation copying

(79) necessarily silences the XP-copy at the origin site. It then follows that SUBJ V SP
OBJ is grammatical whereas *SUBJ V OBJ SP OBJ is not, and there is no need for parallel
chains. While this explanation indeed works for object cases, it will not work for
subject cases, for SUBJ V OBJ SP SUBJ is grammatical. Although one may suggest that
Cantonese DC examples with a subject cauda be derived differently, an important
virtue of the proposed parallel chain account is that it straightforwardly offers a
unified account for both cases.

6.1 Understanding the differences between RDO and DC

The current proposal not only yields the correct linear order for RDO, it also
explains several differences between RDO and DC in terms of their grammatical
properties. The first interesting difference is that RDO does not observe any object
restriction, unlike DC. Compare the RDO example (80b) with the DC example (13),
repeated below.

(81) keoi5 wui5 hou2 siu2sam1-gam2 tai2 go2-bun2 syu1 gaa3


3.SG will very careful-way read that-CL book SP
{keoi5/ *go2-bun2 syu1}
3.SG that-CL book
‘He will read that book very carefully.’
(Cheung 2015:248, ex. (70); tones added)

This difference immediately follows from the current proposal: in order to yield an
object DP as the cauda in (81), non-constituent deletion would be involved within
the TP-copy in Spec-GP. By contrast, (80b) is what would obtain when G0 directly
attracts an object DP to its specifier, with no non-constituent deletion involved.
A second difference concerns the fact that wh-phrases may not serve as the post-
SP material in RDO, although they may in DC.13

(82) a. *Aa3koeng4 zung1ji3 aa3 bin1go3? [RDO]


Koeng like Q who
‘Who does Koeng like?’
b. bin1go3 zung1ji3 Ming4zai2 aa3 bing1go3? [DC]
who like Ming Q who
‘Who likes Ming?’

This contrast follows from the fact that the wh-phrase bin1go3 in (82a) directly
enters into a Spec-Head relation with G0, which instructs the semantic component to
interpret its specifier as given information. Given that the wh-phrase is clearly

13
Due to the object restriction, the cauda of (82b) involves a subject wh-phrase.

123
J. Y.-K. Lai

incompatible with such an interpretive property, the derivation eventually crashes.14


On the other hand, the same wh-phrase in the cauda of (82b) does not enter into any
direct Spec-Head relation with G0, but rather the TP-copy containing the wh-phrase
does. The resulting derivation is therefore convergent. Notice that the above contrast
also constitutes another piece of evidence for the claim that DC formation with a
non-full cauda involves deletion rather than direct extraction.
Moreover, the fact that the functional projection GP is at stake also captures
further properties of RDO. Compare (83a) and (83b).

(83) a. keoi5 maai5-zo2 bou6 din6nou5 aa3 [canonical sentence]


3.SG buy-PFV CL computer SP
‘He bought {a computer/the computer}.’
b. keoi5 maai5-zo2 aa3 bou6 din6nou5 [RDO]
3.SG buy-PFV SP CL computer
‘He bought {*a computer/the computer}.’

While the object DP bou6 din6nou5 in (83a) may be interpreted as a nonspecific


indefinite or a definite (Leung 1980; Cheng and Sybesma 1999), the same DP in
(83b) may only be interpreted as a definite. This follows from the given
interpretation which G0 assigns to its specifier, which excludes the nonspecific
indefinite reading.
Another interesting fact is that a phrase serving as the answer to a wh-phrase may
not undergo RDO (see also Lee 2017:67).

(84) A: keoi5 maai5-zo2 mat1je5 aa3?


3.SG buy-PFV what Q
‘What did he buy?’
B: keoi5 maai5-zo2 bou6 din6nou5 aa3 [canonical sentence]
3.SG buy-PFV CL computer SP
‘He bought {a computer/the computer}.’
Bʹ: #keoi5 maai5-zo2 aa3 bou6 din6nou5 [RDO]
3.SG buy-PFV SP CL computer
Intended: ‘He bought the computer.’

Crucially, response (84Bʹ) is infelicitous even when the object DP is interpreted as


definite. This is because the object DP, as an answer, encodes information focus (Kiss
1998), in conflict with the given interpretation assigned by G0.

14
I assume that an antecedent must have a non-empty phonological matrix. A general problem with
remnant movement is that a moved XP does not c-command its occurrence within (the higher copy of) a
remnant YP, raising the question about spell-out of copies. Collins and Sabel’s (2015) suggestion is that
once an XP undergoes movement, the phonological features of its lower copy are immediately deleted.
One way of executing this idea in the current system is that the value of the (inherent) [P] feature (à la
Arregi and Pietraszko 2019) on the lower copy of an XP undergoing Internal Merge is immediately
reversed to [−P], which is interpreted as non-insertion of Vocabulary Items at the phonological
component. Accordingly, the XP-copy contained within the higher remnant TP-copy in (79), having the
[−P] feature, does not count as an antecedent for the XP-copy in Spec-GP.

123
Parallel copying in dislocation copying

Based on the discussion so far, there is good reason to suppose that the derivation
of an RDO sentence utilizes the exact same functional projection assumed to be
involved in the derivation of a DC sentence. The interpretive properties of the post-
SP element of an RDO sentence also mesh well with the semantics which GP
contributes. The postulation of GP, now not being specific to the DC structure,
receives independent support.

6.2 A brief note on RDS

We have seen cases where the functional head G0 in (79) attracts an object DP to its
specifier. As a matter of fact, if G0 attracts a subject DP, we would then seem to
yield another kind of dislocation structure in Cantonese (referred to here as ‘RDS’),
as exemplified in (85a–b).

(85) a. sik6-zo2 wun2 zuk1 laa3, Aa3koeng4 [RDS]


eat-PFV CL congee SP Koeng
‘Koeng ate the bowl of congee.’
b. wui5 hou2 siu2sam1-gam2 tai2 go2-bun2 syu1 gaa3,
will very careful-way read that-CL book SP
keoi5 [RDS]
3.SG
‘He will read that book very carefully.’

These examples are what Cheung (2009) calls the ‘Dislocation Focus Construction’.
Notice that Cheung proposes a rather different analysis of RDS, as illustrated in (86)
with (85a).

The above analysis involves no remnant movement; the post-SP material of an RDS
sentence is accordingly stranded material. Hence, before one can reach the
conclusion that the present proposal successfully unifies both gapless (i.e. DC) and
all kinds of gapped right-dislocation (i.e. RDS and RDO) in Cantonese, it is crucial
that Cheung’s analysis be thoroughly compared with the current analysis when
extended to RDS. Since it is not the primary goal of this paper to propose a general
theory of Cantonese dislocation, I will leave their potential unification along those
lines for future investigation.

123
J. Y.-K. Lai

7 Conclusion

This paper offered detailed argumentation that, contrary to appearance, the Move-
and-Elide analysis is ill-suited for the phenomenon of dislocation copying attested
in Cantonese. As such, it has uncovered a type of dislocation which the familiar
Move-and-Elide analysis does not readily accommodate (cf. Tanaka 2001; Ott 2014;
Ko 2016; Ott and de Vries 2016). Instead, it was proposed that the derivation of DC
involves the formation of two parallel Ā-chains anchored to the same foot. The only
two existing analyses of Cantonese DC (Tang 2015; Cheung 2015) were also
evaluated in detail, and were shown to be less adequate than the proposed analysis.
To the extent that it is on the right track, the present proposal has at least three
important implications. First, there exists a dislocation-specific projection (identi-
fied here as ‘GivenP’) in the left periphery of the Cantonese clausal spine, as
illustrated below.

(87) CSPP[(…)[GivenP[(…)[TP

Second, apart from spell-out of multiple chain links (see Cheng and Vicente 2013
who utilize this option in deriving Mandarin V-doubling sentences), parallel chain
formation should also be recognized as an option available in the grammar, which
may give rise to repetition phenomena not obviously conditioned by any
morphophonological factor (Kandybowicz 2008).
Lastly, the present proposal successfully unifies dislocation copying and another
type of dislocation in Cantonese, with their syntactic and interpretive differences
falling directly under the proposal. It may thus be the case that the apparent
differences between different types of Cantonese dislocation simply reduce to the
size of the syntactic object which the functional head G0 attracts. Although there
remain a number of open questions, the intriguing possibility that a general theory
of Cantonese dislocation might materialize is now visibly on the horizon.

Acknowledgements I am grateful to Karlos Arregi, Lawrence Cheung, Toru Ishii, Paul Law, Stephen
Matthews, Jason Merchant, Hooi Ling Soh, Vassilios Spyropoulos, Masaya Yoshida and Erik Zyman for
helpful discussion, as well as the three anonymous reviewers for their constructive comments. For their
judgments, I thank Kathleen Ng, Szeto Pui Yiu and Alan Yu. Earlier versions of this paper were presented
at the 20th Seoul International Conference on Generative Grammar (SICOGG 20) and the 49th Annual
Meeting of the North East Linguistic Society (NELS 49), and thanks are due to the audiences at the two
conferences for their feedback. Special thanks go to Yenan Sun and the JEAL editors for their
encouragement. The usual disclaimers apply.

References

Abels, Klaus. 2007. Towards a restrictive theory of (remnant) movement. Linguistic Variation Yearbook
7: 52–120.
Aboh, Enoch Oladé, and Marina Dyakonova. 2009. Predicate doubling and parallel chains. Lingua 119:
1035–1065.
Adger, David. 2003. Core syntax. Oxford: Oxford University Press.
Arregi, Karlos. 2010. Ellipsis in split questions. Natural Language & Linguistic Theory 28: 539–592.
Arregi, Karlos, and Asia Pietraszko. 2019. The ups and downs of head displacement. Manuscript.
Available online at: http://ling.auf.net/lingbuzz/004096. Accessed 3 May 2019.

123
Parallel copying in dislocation copying

Baltin, Mark. 2012. Deletion versus pro-forms: An overly simple dichotomy? Natural Language &
Linguistic Theory 30: 381–423.
Beaver, David I., and Brady Z. Clark. 2008. Sense and sensitivity: How focus determines meaning.
Oxford: Blackwell.
Bhatt, Rajesh, and Roumyana Pancheva. 2004. Late merger of degree clauses. Linguistic Inquiry 35: 1–
45.
Chan, Brian Hok-Shing. 2013. Sentence-final particles, complementizers, antisymmetry, and the final-
over-final constraint. In Theoretical approaches to disharmonic word order, ed. Theresa Biberauer
and Michelle Sheehan, 445–468. Oxford: Oxford University Press.
Cheng, Lisa Lai-Shen. 1991. On the typology of wh-questions. Ph.D. dissertation, MIT, Cambridge, MA.
Cheng, Lisa Lai-Shen. 2007. Verb copying in Mandarin Chinese. In The copy theory of movement, ed.
Norbert Corver and Jairo Nunes, 151–174. Amsterdam: John Benjamins.
Cheng, Lisa Lai-Shen, and Rint Sybesma. 1999. Bare and not-so-bare nouns and the structure of NP.
Linguistic Inquiry 30: 509–542.
Cheng, Lisa Lai-Shen, and Luis Vicente. 2013. Verb doubling in Mandarin Chinese. Journal of East
Asian Linguistics 22: 1–37.
Cheung, Lawrence Yam Leung. 2009. Dislocation focus construction in Chinese. Journal of East Asian
Linguistics 18: 197–232.
Cheung, Lawrence Yam Leung. 2015. Bi-clausal sluicing approach to dislocation copying in Cantonese.
International Journal of Chinese Linguistics 2: 227–272.
Cheung, Lawrence Yam Leung. 1997. A study of right dislocation in Cantonese. M.Phil. thesis, The
Chinese University of Hong Kong.
Chomsky, Noam. 1972. Some empirical issues in the theory of transformational grammar. In The goals of
linguistic theory, ed. Stanley Peters, 63–130. Englewood Cliffs, NJ: Prentice-Hall.
Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.
Chomsky, Noam. 2008. On phases. In Foundational issues in linguistic theory, ed. Robert Freidin, Carlos
P. Otero, and Maria Luisa Zubizarreta, 133–166. Cambridge, MA: MIT Press.
Collins, Chris, and Joachim Sabel. 2015. A C/I-interface constraint on remnant movement. In Remnant
movement, ed. Günther Grewendorf, 93–131. Berlin: Mouton de Gruyter.
Coppock, Elizabeth. 2001. Gapping: In defense of deletion. Proceedings of the Chicago Linguistic
Society 37: 133–148.
Duncan, Philip T. 2016. Parallel chain formation in Ibibio contrastive focus. In Proceedings of the forty-
second annual meeting of the Berkeley Linguistics Society, ed. Emily Clem et al., 87–106.
Fox, Danny. 2014. Extraposition and scope: Evidence for embedded Late Merge. Paper presented at the
Second graduate student workshop in memory of Tanya Reinhart, Tel Aviv University, April.
Fox, Danny, and Howard Lasnik. 2003. Successive-cyclic movement and island repair: The difference
between sluicing and VP-ellipsis. Linguistic Inquiry 34: 143–154.
Freidin, Robert. 1986. Fundamental issues in the theory of binding. In Studies in the acquisition of
anaphora, vol. I, ed. Barbara Lust, 151–188. Dordrecht: Reidel.
Gallego, Ángel J. 2018. Freezing effects in a free-Merge system: A configurational approach. In
Freezing: Theoretical approaches and empirical domains, ed. Jutta M. Hartmann et al., 66–104.
Berlin: Mouton de Gruyter.
Gasde, Horst-Dieter, and Waltraud Paul. 1996. Functional categories, topic prominence, and complex
sentences in Mandarin Chinese. Linguistics 34: 263–294.
Goldsmith, John. 1976. Autosegmental phonology. Ph.D. dissertation, MIT, Cambridge, MA.
Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In The view
from building 20: Essays in honor of Sylvain Bromberger, ed. Kenneth Hale and Samuel Jay Keyser,
111–176. Cambridge, MA: MIT Press.
Halle, Morris, and Alec Marantz. 1994. Some key features of distributed morphology. In Papers on
phonology and morphology, ed. Andrew Carnie, Heidi Harley, and Tony Bures, 275–288.
Cambridge, MA: MIT Press.
Heck, Fabian, and Gereon Müller. 2007. Extremely local optimization. In Proceedings of the 34th
Western Conference on Formal Linguistics, ed. Erin Bainbridge and Brian Agbayani, 170–182.
Department of Linguistics, University of Fresno, CA.
Huang, C.-T.James. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15:
531–574.
Huang, C.-T.James. 1991. Remarks on the status of the null object. In Principles and parameters in
comparative grammar, ed. Robert Freidin, 56–76. Cambridge, MA: MIT Press.

123
J. Y.-K. Lai

Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. Ph.D. dissertation,
MIT, Cambridge, MA.
İnce, Atakan. 2012. Fragment answers and islands. Syntax 15: 181–214.
Johnson, Kyle. 2004. How to be quiet. In Proceedings of the Chicago Linguistic Society, 1–20. Chicago,
IL: University of Chicago.
Kandybowicz, Jason. 2008. The grammar of repetition: Nupe grammar at the syntax-phonology interface.
Amsterdam: John Benjamins.
Kayne, Richard S. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press.
Kiss, Katalin É. 1998. Identificational focus versus information focus. Language 74: 245–273.
Ko, Heejeong. 2016. Gapless right-dislocation: The role of overt correlates. Language Research 52: 3–32.
Kučerová, Ivona. 2012. Grammatical marking of givenness. Natural Language Semantics 20: 1–30.
Landau, Idan. 2006. Chain resolution in Hebrew V(P) fronting. Syntax 9: 32–66.
Lasnik, Howard. 2001. When can you save a structure by destroying it? In Proceedings of NELS 31, Vol.
2, ed. Minjoo Kim and Uri Strauss, 301–320. Amherst: GLSA, University of Massachusetts.
Lasnik, Howard, and Mamoru Saito. 1984. On the nature of proper government. Linguistic Inquiry 15:
235–289.
Lasnik, Howard, and Mamoru Saito. 1992. Move α: Conditions on its application and output. Cambridge,
MA: MIT Press.
Leben, William. 1973. Suprasegmental phonology. Ph.D. dissertation, MIT, Cambridge, MA.
Lee, Tsz Ming. 2017. Defocalization in Cantonese right dislocation. Gengo Kenkyu 152: 59–87.
Leung, Chung-Sum. 1980. Guo-Yue Yu bijao yufa zhaji—dingzhi zhishici [Notes on comparative
grammar of Mandarin and Cantonese: Markers of definiteness]. Zhong-Ying Yuwen Xueyan
[Language Learning and Research] 1: 15–28.
Lin, T.-H.Jonah. 2010. Structures and functional categories of Mandarin sentences. UST Working Papers
in Linguistics 6: 41–79.
Lin, T.-H.Jonah. 2012. Multiple-modal constructions in Mandarin Chinese and their finiteness properties.
Journal of Linguistics 48: 151–186.
Lin, Jo-wang, and C.-C.Jane Tang. 1995. Modals as verbs in Chinese: A GB perspective. Bulletin of the
Institute of History and Philology Academia Sinica 66: 53–105.
Lohndal, Terje. 2010. Medial-wh phenomena, parallel movement, and parameters. Linguistic Analysis 34:
215–244.
Matthews, Stephen, and Virginia Yip. 2011. Cantonese: A comprehensive grammar, 2nd ed. London:
Routledge.
Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford: Oxford
University Press.
Merchant, Jason. 2004. Fragments and ellipsis. Linguistics and Philosophy 27: 661–738.
Merchant, Jason. 2008. Variable island repair under ellipsis. In Topics in ellipsis, ed. Kyle Johnson, 132–
153. Cambridge: Cambridge University Press.
Merchant, Jason. 2019. Ellipsis: A survey of analytical approaches. In The Oxford handbook of ellipsis,
ed. Jeroen van Craenenbroeck and Tanja Temmerman, 19–45. Oxford: Oxford University Press.
Montalbetti, Mario M. 1984. After binding: On the interpretation of pronouns. Ph.D. dissertation, MIT,
Cambridge, MA.
Müller, Gereon. 1998. Incomplete category fronting: A derivational approach to remnant movement in
German. Dordrecht: Kluwer Academic Publishers.
Müller, Gereon. 2010. On deriving CED effects from the PIC. Linguistic Inquiry 41: 35–82.
Nunes, Jairo. 2004. Linearization of chains and sideward movement. Cambridge, MA: MIT Press.
Ott, Dennis. 2014. An ellipsis approach to contrastive left-dislocation. Linguistic Inquiry 45: 269–303.
Ott, Dennis, and Mark de Vries. 2016. Right-dislocation as deletion. Natural Language & Linguistic
Theory 34: 641–690.
Overfelt, Jason. 2017. Extraposition, polarity, and Late Merge. In A schrift to fest Kyle Johnson, ed.
Nicholas LaCara, Keir Moulton, and Anne-Michelle Tessier, 305–312. Amherst, MA: Linguistics
Open Access Publications.
Paul, Waltraud. 2014. Why particles are not particular: Sentence-final particles in Chinese as heads of a
split CP. Studia Linguistica 68: 77–115.
Paul, Waltraud. 2015. New perspectives on Chinese syntax. Berlin: Mouton de Gruyter.
Paul, Waltraud. 2016. Where “complex” sentences are not complex and “subordinate” clauses not
subordinate: The case of Mandarin Chinese. In Coordination and subordination: Form and meaning

123
Parallel copying in dislocation copying

—Selected papers from CSI Lisbon 2014, ed. Fernanda Pratas, Sandra Pereira, and Clara Pinto, 185–
208. Newcastle: Cambridge Scholars Publishing.
Potsdam, Eric. 2017. Exceptives and ellipsis. Poster presented at NELS 48.
Rizzi, Luigi. 2006. On the form of chains: Criterial positions and ECP effects. In Wh-movement: Moving
on, ed. Lisa Lai-Shen Cheng and Norbert Corver, 97–133. Cambridge, MA: MIT Press.
Rizzi, Luigi. 2010. On some properties of criterial freezing. In The complementizer phase: Subjects and
operators, ed. E. Phoevos Panagiotidis, 17–32. Oxford: Oxford University Press.
Rochemont, Michael. 1986. Focus in generative grammar. Amsterdam: John Benjamins.
Rochemont, Michael. 2013a. Discourse new, focused and given. In Approaches to Hungarian 13, ed.
Johan Brandtler, Valéria Molnár, and Christer Platzack, 199–228. Amsterdam: John Benjamins.
Rochemont, Michael. 2013b. Discourse new, F-marking and normal stress. Lingua 136: 38–62.
Rochemont, Michael. 2016. Givenness. In The Oxford handbook of information structure, ed. Caroline
Féry and Shinichiro Ishihara, 41–63. Oxford: Oxford University Press.
Ross, John Robert. 1967. Constraints on variables in syntax. Ph.D. dissertation, MIT, Cambridge, MA.
Sailor, Craig, and Gary Thoms. 2014. On the non-existence of non-constituent coordination and non-
constituent ellipsis. In Proceedings of the 31st West Coast Conference on Formal Linguistics, ed.
Robert E. Santana-LaBarge, 361–370. Somerville, MA: Cascadilla Proceedings Project.
Sauerland, Uli. 2005. Don’t interpret focus! Why a presuppositional account of focus fails and how a
presuppositional account of givenness works. In Proceedings of Sinn und Bedeutung 9, ed. Emar
Maier, Corien Bary, and Janneke Huitink, 370–384. Nijmegen: Nijmegen Centre of Semantics.
Schwarzschild, Roger. 1999. GIVENness, AvoidF, and other constraints on the placement of accents.
Natural Language Semantics 7: 141–177.
Selkirk, Elisabeth. 1995. Sentence prosody: Intonation, stress, and phrasing. In The handbook of
phonological theory, ed. John A. Goldsmith, 550–569. Oxford: Blackwell.
Simpson, Andrew. 2014. Sentence-final particles. In The handbook of Chinese linguistics, ed. C.-T. James
Huang, Y.-H. Audrey Li, and Andrew Simpson, 156–179. Oxford: Blackwell.
Takahashi, Shoichi, and Sarah Hulsey. 2009. Wholesale late merger: Beyond the A/Ā distinction.
Linguistic Inquiry 40: 387–426.
Tanaka, Hidekazu. 2001. Right-dislocation as scrambling. Journal of Linguistics 37: 551–579.
Tang, Sze-wing. 2000. Identity avoidance and constraint interaction: The case of Cantonese. Linguistics
38: 33–61.
Tang, Sze-wing. 2015. A generalized syntactic schema for utterance particles in Chinese. Lingua Sinica
1: 3.
Tang, Sze-wing. 2018. Yanshenju de jufa fenxi [A syntactic analysis of incremental sentences]. Yuyan
Jiaoxue yu Yanjiu [Language Teaching and Linguistic Studies] 3: 48–57.
Tang, C.-C. Jane. 1990. Chinese phrase structure and the extended Xʹ-theory. Ph.D. dissertation, Cornell
University.
Tang, Sze-wing. 1998. Parameterization of features in syntax. Ph.D. dissertation, University of
California, Irvine.
Thiersch, Craig. 2017. Remnant movement. In The Wiley Blackwell Companion to Syntax, 2nd ed, ed.
Martin Everaert and Henk C. Riemsdijk. Hoboken: Wiley-Blackwell.
Van Craenenbroeck, Jeroen, and Anikó Lipták. 2006. The crosslinguistic syntax of sluicing: Evidence
from Hungarian relatives. Syntax 9: 248–274.
Van Riemsdijk, Henk. 1989. Movement and regeneration. In Dialectal variation and the theory of
grammar, ed. Paola Benincá, 105–136. Dordrecht: Foris.
Wagner, Michael. 2006. Givenness and locality. In Proceedings of Semantics and Linguistic Theory XVI,
ed. Masayuki Gibson and Jonathan Howell, 295–312. Ithaca, NY: CLC Publications.
Wei, Ting-Chi. 2018. Fragment questions in Mandarin Chinese: Topic movement and pied-piping.
Language and Linguistics 19: 266–305.
Williams, Edwin. 1997. Blocking and anaphora. Linguistic Inquiry 28: 577–628.
Yim, Changguk. 2012. Fragment answers containing -yo in Korean: New evidence for the PF deletion
theory of ellipsis. Linguistic Inquiry 43: 514–518.

Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

123

You might also like