The syntax and post-syntax of verb doubling in Mandarin Chinese

Abstract

To appear in Proceedings of WCCFL41

Full text

The text and figures in this PDF are approved by the author(s) for publication. Any mistakes in this PDF
will not be corrected by the publisher. This PDF was created on August 11, 2023.
The Syntax and Post-Syntax of Verb Doubling in
Mandarin Chinese
Tom Meadows & Qiuhao Charles Yan
1. Introduction
In this paper, we argue that post-syntactic processes can give rise to multiple spell-outs on the basis
of verb doubling effects in Mandarin Chinese (cf. Cheng 2007, 2017). In a variety of languages, predicate
fronting can create verb copying constructions (for instance, Russian (Abels 2001), Classical Hebrew
(Harbour 1999), Haitian (Harbour 2008), Modern Hebrew (Landau 2006), among many others), where a
higher VP is pronounced in the clausal left periphery, and another VP is spelled out in a lower position.
In Mandarin Chinese there exists a similar kind of verb copying construction, triggered by VP-internal
particles in the context of VP-fronting. We term such a construction long verb doubling (henceforth LVD)1,
as exemplified in (1). (1-a) is an instance of Manner LVD, where the lower VP is followed by a particle
de and a manner adverbial hˇ
en ku`
ai ‘very quickly’. By comparison, (1-b) instantiates a case of Duration
LVD, in which the lower VP is followed by a tense and aspect marker (TAM) le and a temporal expression
s¯
an ti¯
an ‘three days’.
(1) a. K`
an
read
xiˇ
aoshu¯
o
novel
t¯
a
he
k`
an
read
de
de
hˇ
en
very
ku`
ai.
quickly
‘As for reading novels, he read them very quickly.’ Manner LVD
b. K`
an
watch
di`
anyˇ
ıng
movie
t¯
a
he
k`
an
watch
le
le
s¯
an
three
ti¯
an.
day
‘As for watching movies, he watched them for three days.’ Duration LVD
The key question here is how verb doubling effects are derived in constructions above. We argue that
derivations of both kinds of LVD can be accounted for in a uniform way, under the two PF constraints
proposed by Landau (2006) (i.e., P-Recoverability (PR) and Economy of Pronunciation (EP)). We propose
that all LVD constructions involve the same phrasal movement to the clausal left periphery, where the
higher VP gets completely realised at PF. Rather than being completely deleted as lower copies often are,
in cases of verb doubling the lower VP-copy undergoes partial deletion (cf. Scott 2021, van Urk 2018).
Full deletion of the lower VP-copy is blocked in order to satisfy requirements of both de and le to be
morphosyntactically hosted by the lower V.
Our proposal is fleshed out in the following sections. Section 2 presents evidence for the phrasal
movement of VP to the clausal left periphery. Section 3 focuses on properties of the lower VP as well as
the post-verbal particles de and le. Section 4 outlines the derivation of verb doubling effects. Section 5 is
the conclusion.
*Tom Meadows & Qiuhao Charles Yan, Queen Mary University of London, t.a.s.meadows@qmul.ac.uk;q.
yan@qmul.ac.uk. Thanks to Coppe van Urk and audiences at SICOGG24 and LAGB 2022 for helpful comments on
earlier versions of this work. With respect to the initial discussions of the phenomenon and data, we owe many thanks
to our friends and colleagues, in particular Zhouyi Sun, Xuechun Xiang, and Ka-Fai Yip. Special thanks go to Klaus
Abels, Daniel Harbour, Idan Landau, Zhouyi Sun, Matthew Tyler, Maximilian Wiesner, and members of Syntax &
Semantics Reading Group at Queen Mary University of London, SyntaxLab at University of Cambridge, as well as
Department of Linguistics and Modern Languages at the Chinese University of Hong Kong. All the mistakes are ours.
1The more discussed verb copying constructions in the literature of Mandarin Chinese are what we term short verb
doubling (henceforth SVD), where the higher VP ends up in the clause-medial position between the subject and lower
VP. For reasons of space, SVD is not included in this paper. See relevant discussions in Cheng (2007, 2017).

2. Evidence for the higher VP-copy movement
We first demonstrate that VP-fronting in Mandarin Chinese is derived via movement (cf. Bartos 2019,
Tang 1990). The first diagnostic comes from island effects. As shown in (2) and (3), neither Manner LVD
nor Duration LVD is possible across a range of island contexts (cf. Huang et al. 2009), which is unexpected
if the fronted VPs are adjuncts base-generated in the left periphery.
(2) Island effects with Manner LVD
a. *K`
an
read
xiˇ
aoshu¯
o
novel
wˇ
o
I
xiˇ
ang
want
zh¯
ıd`
ao
know
[ sh´
enme
what
sh´
ıh`
ou
time
t¯
a
he
k`
an
read
de
de
hˇ
en
very
ku`
ai.
quickly
]
‘I wonder when he reads novels very quickly.’ Wh-Island
b. *K`
an
read
xiˇ
aoshu¯
o
novel
wˇ
o
I
j`uju´
e
reject
[ t¯
a
he
k`
an
read
de
de
hˇ
en
very
ku`
ai
quickly
de
de
shu¯
ofˇ
a.
claim.
]
‘I reject the claim that he reads novels very quickly.’ Complex NP Island
c. *K`
an
read
xiˇ
aoshu¯
o
novel
t¯
a
he
b`u
neg
zh¯
ıd`
ao
know
[ t¯
a
he
k`
an
read
de
de
hˇ
en
very
ku`
ai
quickly
]
make
r`
ang
me
wˇ
o
very
f¯
eich´
ang
surprised
j¯
ıngy`
a.
‘He doesn’t know that he reads novels very quickly makes me so surprised.’
Subject Island
d. *K`
an
read
xiˇ
aoshu¯
o
novel
wˇ
o
I
ch´
ongb`
ai
worship
t¯
a
him
[ y¯
ınw`
ei
because
t¯
a
he
k`
an
read
de
de
hˇ
en
very
ku`
ai.
quickly
]
‘I worship him because he reads novels very quickly.’ Adjunct Island
(3) Island effects with Duration LVD
a. *K`
an
watch
di`
anyˇ
ıng
movie
wˇ
o
I
xiˇ
ang
want
zh¯
ıd`
ao
know
[ w`
eish´
eme
why
t¯
a
he
k`
an
watch
le
le
s¯
an
three
ti¯
an.
day
]
‘I wonder why he watched movies for three days.’ Wh-Island
b. *K`
an
watch
di`
anyˇ
ıng
movie
wˇ
o
I
j`uju´
e
reject
[ t¯
a
he
k`
an
watch
le
le
s¯
an
three
ti¯
an
day
de
de
shu¯
ofˇ
a.
claim.
]
‘I reject the claim that he watched movies for three days.’ Complex NP Island
c. *K`
an
watch
di`
anyˇ
ıng
movie
t¯
a
he
b`u
neg
zh¯
ıd`
ao
know
[ t¯
a
he
k`
an
watch
le
le
s¯
an
three
ti¯
an
day
]
make
r`
ang
me
wˇ
o
very
f¯
eich´
ang
surprised
j¯
ıngy`
a.
‘He doesn’t know that he watched movies for three days makes me so surprised.’
Subject Island
d. *K`
an
watch
di`
anyˇ
ıng
movie
wˇ
o
I
b`u
neg
lˇ
ıjiˇ
e
understand
t¯
a
him
[ y¯
ınw`
ei
because
t¯
a
he
k`
an
watch
le
le
s¯
an
three
ti¯
an.
day
]
‘I don’t understand him because he watched movies for three days.’ Adjunct Island
The second diagnostic relies on binding and reconstruction effects. (4) and (5) are examples of
Principle A reconstruction effects: as shown in (4-a) and (5-a), the anaphor t¯
az`
ıjˇ
ı‘himself’ is able to
precede the subject m´
ei r´
en ‘no one’ linearly and receive the bound interpretation at the same time. The
bound reading between the anaphor and the subject suggests that the higher VPs (which contain anaphors)
need to be reconstructed to a lower position, as if they were obeying Principle A like (4-b) and (5-b).
(4) Principle A reconstruction with Manner LVD
a. Hu`
a
draw
t¯
az`
ıjˇ
ıi/*j
himself
de
de
xi`
aoxi`
ang
portrait
m´
ei
neg
r´
eni
people
hu`
a
draw
de
de
hˇ
en
very
ku`
ai.
quickly
‘No oneidraws the portrait of himselfi/*j very quickly.’
b. M´
ei
neg
r´
eni
people
hu`
a
draw
t¯
az`
ıjˇ
ıi/*j
himself
de
de
xi`
aoxi`
ang
portrait
hu`
a
draw
de
de
hˇ
en
very
ku`
ai.
quickly
‘No oneidraws the portrait of himselfi/*j very quickly.’

(5) Principle A reconstruction with Duration LVD
a. Hu`
a
draw
t¯
az`
ıjˇ
ıi/*j
himself
de
de
xi`
aoxi`
ang
portrait
m´
ei
neg
r´
eni
people
hu`
a
draw
le
le
s¯
an
three
ti¯
an.
day
‘No oneidrew the portrait of himselfi/*j for three days.’
b. M´
ei
neg
r´
eni
people
hu`
a
draw
t¯
az`
ıjˇ
ıi/*j
himself
de
de
xi`
aoxi`
ang
portrait
hu`
a
draw
le
le
s¯
an
three
ti¯
an.
day
‘No oneidrew the portrait of himselfi/*j for three days.’
(6) and (7) are another set of examples showing Principle C reconstruction effects: as seen in (6-a)
and (7-a), the referential object Piet cannot be coreferential with the subject t¯
a‘he’. The fronted object
(which is contained by the VP) behaves c-commanded by the subject as in (6-b) and (7-b), demonstrating
the Principle C effects. Again, the similarities suggest that the higher VPs in (6-a) and (7-a) undergo
reconstruction.
(6) Principle C reconstruction with Manner LVD
a. *Hu`
a
draw
Pieti
Piet
de
de
xi`
aoxi`
ang
portrait
t¯
ai
he
hu`
a
draw
de
de
hˇ
en
very
ku`
ai.
quickly
*‘Heidraws the portrait of Pietivery quickly.’
b. *T¯
ai
he
hu`
a
draw
Pieti
Piet
de
de
xi`
aoxi`
ang
portrait
hˇ
en
very
ku`
ai.
quickly
*‘Heidraws the portrait of Pietivery quickly.’
(7) Principle C reconstruction with Duration LVD
a. *Hu`
a
draw
Pieti
Piet
de
de
xi`
aoxi`
ang
portrait
t¯
ai
he
hu`
a
draw
le
le
s¯
an
three
ti¯
an.
day
*‘Heidrew the portrait of Pietifor three days.’
b. *T¯
ai
he
hu`
a
draw
le
le
Pieti
Piet
de
de
xi`
aoxi`
ang
portrait
s¯
an
three
ti¯
an.
day
*‘Heidrew the portrait of Pietifor three days.’
We conclude, based on the diagnostics presented above, that the higher VPs in LVD are derived via
phrasal movement.
3. Properties of the lower VP-copy and its interaction with post-verbal particles
There are several similarities and distinctions between Manner LVD and Duration LVD in terms of
their basic properties, which can be attributed to the syntactic structures of their own VP-lower copy
interacting with different post-verbal particles. Thus, we adopt two different structures for manner de and
duration le.
To start off, (8-a) and (9-a) show that the particle de and the TAM le obligatorily occur in the LVD
constructions. Furthermore, (8-b) and (9-b) reveal that the post-verbal arguments are unable to co-occur
with the post-verbal particles in the lower VPs. In other words, only particles are allowed to occupy the
post-verbal position within the lower VPs.
(8) a. K`
an
read
xiˇ
aoshu¯
o
novel
t¯
a
he
k`
an
read
*(de)
de
hˇ
en
very
ku`
ai.
quickly
‘As for reading novels, he read them very quickly.’
b. K`
an
read
xiˇ
aoshu¯
o
novel
t¯
a
he
k`
an
read
de
de
(*xiˇ
aoshu¯
o)
novel
hˇ
en
very
ku`
ai.
quickly
‘As for reading novels, he read them very quickly.’

(9) a. K`
an
watch
di`
anyˇ
ıng
movie
t¯
a
he
k`
an
watch
*(le)
le
s¯
an
three
ti¯
an.
day
‘As for watching movies, he watched them for three days.’
b. K`
an
watch
di`
anyˇ
ıng
movie
t¯
a
he
k`
an
watch
le
le
(*di`
anyˇ
ıng)
movie
s¯
an
three
ti¯
an.
day
‘As for watching movies, he watched them for three days.’
The main difference between Manner LVD and Duration LVD lies in the non-verb doubling construc-
tions, for example, (10) and (11). (10) shows that either the particle de or the internal argument can occupy
the post-verbal position preceding the adverbial, indicating that de and the argument are in a comple-
mentary distribution. By contrast, (11) demonstrates that the TAM le obligatorily occurs in constructions
where no verb doubling effects arise, regardless of the occurrence of the internal argument.
(10) a. T¯
a
he
k`
an
read
(*de)
de
xiˇ
aoshu¯
o
novel
hˇ
en
very
ku`
ai.
quickly
‘He read novels very quickly.’
b. T¯
a
he
k`
an
read
de
de
(*xiˇ
aoshu¯
o)
novel
hˇ
en
very
ku`
ai.
quickly
‘He read very quickly.’
(11) a. T¯
a
he
k`
an
watch
*(le)
le
t`
aitˇ
ann´
ık`
eh`
ao
Titanic
s¯
an
three
ti¯
an.
day
‘He watched Titanic for three days.’
b. T¯
a
he
k`
an
watch
le
le
(t`
aitˇ
ann´
ık`
eh`
ao)
Titanic
s¯
an
three
ti¯
an.
day
‘He watched (Titanic) for three days.’
To capture the similarities and distinctions that we have seen above at the same time, we postulate two
slightly different VP structures respectively for Manner LVD (12) and Duration LVD (13). To be precise,
in both structures, we adopt the assumption that there exists an external structure out of the core VP, i.e.,
the VoiceP (Kratzer 1996). The VoiceP takes the Mod(ifier)P as its right-adjoined adjunct, whose location
is outside the core VP. The particle de is assumed to realise the head of the ModP optionally, taking
adverbials as its complement (Sybesma 1999). However, the TAM le is assumed to be base-generated
within the verbal extended projections, always leaving covert the head of the ModP, which still takes
adverbials as its complement.
Let us briefly summarise. The different distributions of the particle de and the TAM le can be supported
empirically. In general, le can show up in cases where there is no post-verbal modifier. But the occurrence
of de is more constrained: in principle, it only co-occurs with post-verbal modifiers. We will argue in
the next section that the ways of de and le being adjacent to the V head are realised by distinct PF processes.
(12) The lower VP in Manner LVD
VoiceP
Voice’
Voice VP
ModP
Mod
de
AdvP
(13) The lower VP in Duration LVD
VoiceP
Voice’
Voice AspP
Asp
le
VP
ModP
Mod AdvP

4. PF-sensitivity and partial deletion
4.1. Regulating copy deletion with the PF algorithm
In this section we set out our proposal that post-syntactic processes can drive multiple spell-outs.
Below is the basic intuition we would like to implement.
(14) The PF-sensitivity Intuition
Copy deletion cannot completely apply if certain post-syntactic processes would otherwise apply
to part of the lower copy.
We assume that there exists a subset of post-syntactic processes which copy deletion is obliged to
allow to apply. Landau (2006) effectively argues that the hosting of inflectional morphology in Hebrew is
just such a process. As exemplified in (15), verbs in Hebrew are normally inflected for tense or agreement.
When a VP is fronted into the left periphery (i.e., Spec,CP), the higher copy of the verb cannot display
such an inflection. However, the inflection cannot go unrealised either. Copy deletion, thus, is obliged to
not block the realisation of inflectional morphology. It follows that enough of the lower VP copy avoids
deletion to host tense or agreement.
(15) [CP [VP liknot
buy.inf
et
acc
ha-praxim]
det-flower.plu
[TP hi
she
[VoiceP kanta
buy.pst.3sg.fem
et ha praxim]]]
‘As for buying flowers, she bought.’ Landau (2006: ex. 8a)
Clearly, the size of copy deletion is not random. Instead, copy deletion is regulated by the PF algorithm
(16) proposed by Landau (2006). P-Recoverability (PR) guarantees that the spell-outs of copies that satisfy
certain post-syntactic requirements are obligatory, since these requirements can provide copies with novel
PF content which, in theory, enables them to differ from other copies (with established PF content).
Economy of Pronunciation (EP), by contrast, is an economy constraint, which in principle maximises
deletion at PF. As we will see later, certain post-syntactic requirements can be linked to part of copies
only instead of the entire ones. In this sense, copies are able to be deleted or spelled out partially at PF.
We apply these two constraints in the derivation of Manner and Duration LVD in the following analyses.
(16) a. P-Recoverability (PR)
Copies associated with certain PF content must be pronounced.
b. Economy of Pronunciation (EP)
Delete all copies not associated with certain PF content.
4.2. Derivation of Manner LVD
The entire syntactic structure of Manner LVD (1-a) is represented as in (17), and the post-syntactic
process required by the particle de is schematised in (18). To start off, the higher VP-copy is merged in
Spec,CP due to phrasal movement. Such a structural position assigns to the higher VP a contrastively
topicalised interpretation, which is considered as certain PF content when the VP-copy is sent to be spelled
out. Therefore, the higher VP-copy must be pronounced completely according to PR.
Contrary to the higher VP-copy, the lower one should in theory be totally deleted at PF, subject to EP,
since it is not associated with specific PF content (e.g., contrastively topicalised interpretation as shown
above). However, the particle de is postulated to have the hosting requirement of local dislocation (Embick
2007, Embick & Noyer 2001). As we have seen previously, de, as the head of ModP, must form a complex
head with V under linear adjacency. Such a requirement of de supplies novel certain PF content to the
lower VP. Correspondingly, deletions that preserve this PF content are prioritised by PR, i.e., by licensing
the V head to escape deletion, thereby feeding local dislocation of de. Partial deletion (cf. Scott 2021,
van Urk 2018), which targets the internal argument within the lower VP-copy, is selected as the prioritised
PF-operation. That is, only the argument undergoes deletion according to EP, which in turn feeds the PF
process of de. As a result, the ‘lower V + de’ sequence is spelled out, and verb doubling effect arises.

(17) CP
VP
V
k`
an
DP
xiˇ
aoshu¯
o
C’
C TP
DP
t¯
a
T’
T VoiceP
Voice’
Voice VP
V
k`
an
DP
xiˇ
aoshu¯
o
ModP
Mod
de
AdvP
hˇ
en ku`
ai
(18) . . .
. . . VoiceP
Voice’
Voice VP
V
V
k`
an
‘read’
Mod
de
DP
ModP
Mod AdvP
hˇ
en ku`
ai
‘very quickly’
Local Dislocation: Cannot apply
with DP linearly intervening
4.3. Derivation of Duration LVD
The syntactic structure of Duration LVD (1-b) is represented as in (19), and the post-syntactic process
required by the TAM le is schematised in (20). As with the case in Manner LVD, the higher VP-copy here
is also merged in Spec,CP due to the identical phrasal movement. Still, the higher VP-copy receives a
contrastively topicalised interpretation in such a position, which means, certain PF content is determined
by the clausal structure. Thus, the higher VP-copy is spelled out as a whole again according to PR.
Similarly, the TAM le is posited to have some specific PF process in order to escape EP. Unlike
the local dislocation requirement owned by de,le, which realises an Aspect head in the verbal extended
projections, must appear right adjacent to the V head (without necessarily forming a complex head). Such

a requirement can be implemented by several methods, including morphological lowering,PF movement,
etc. Remaining agnostic towards the nature of the PF process required by le, the lower VP can certainly be
supplied with novel PF content as well. Once more, this implies that deletions preserving the PF content
are prioritised by PR. In such a situation, partial deletion is again picked out as the prioritised PF-operation,
from which spell-out of the ‘lower V + le’ sequence results, thereby giving rise to verb doubling effect.
(19) CP
VP
V
k`
an
DP
di`
anyˇ
ıng
C’
C TP
DP
t¯
a
T’
T VoiceP
Voice’
Voice AspP
Asp
-le
VP
V
k`
an
DP
di`
anyˇ
ıng
ModP
s¯
an ti¯
an
(20) . . .
. . . VoiceP
Voice’
Voice AspP
Asp
-le
VP
V
k`
an
‘watch’
DP
di`
anyˇ
ıng
‘movies’
ModP
s¯
an ti¯
an
‘three days’
Lowering
PF movement
...

5. Conclusion
In this paper we propose that LVD constructions in Mandarin Chinese involve the interaction of
(i) phrasal movement of VP to the clausal left periphery, (ii) different kinds of PF-processes underlying
the distribution of post-verbal particles, and (iii) partial copy deletion (cf. Scott 2021, van Urk 2018)
constrained by Landau (2006)’s PF algorithm. We apply this analysis to LVD cases with two different
kinds of particles, manner de and duration le. The morphosyntactic hosting requirements of both particles,
though differing from each other in certain respects, prevent full deletion of a lower VP-copy. The result
of this partial deletion is doubling of the lexical verb. We suggest that other kinds of LVD, not examined
so far, can be handled under the present analysis. One kind is Resultative LVD (21-a), which also contains
a particle de and can be analysed as with Manner LVD. The other kind is Frequency LVD (21-b), which
contains another TAM gu`
oand can be analysed as with Duration LVD. Our proposal provides novel
support for the view that copy deletion is sensitive to post-syntactic processes.
(21) a. K`
an
read
z`
ım`u
subtitles
t¯
a
he
k`
an
read
de
de
hˇ
en
very
p´
ıb`
ei.
exhausted
‘As for reading subtitles, he felt exhausted to read them.’ Resultative LVD
b. K`
an
see
y´
ay¯
ı
dentist
t¯
a
he
k`
an
see
gu`
o
guo
s¯
an
three
c`
ı.
time
‘As for going to the dentist, he went there three times.’ Frequency LVD
References
Abels, Klaus. 2001. The predicate cleft construction in Russian. In Steven Franks & Michael Yadroff (eds.), Formal
Approaches to Slavic Linguistics 9, 1–19. Bloomington, IN: Michigan Slavic Publications.
Bartos, Huba. 2019. The V-copy construction in Mandarin: A case temporarily reopened. In Jianhua Hu & Haihua
Pan (eds.), Interface in Grammar, 167–205. Amsterdam/Philadelphia: John Benjamins.
Cheng, Lisa L.-S. 2007. Verb copying in Mandarin Chinese. In Norbert Corver & Jairo Nunes (eds.), The copy theory
of movement, 151–174. Amsterdam/Philadelphia: John Benjamins.
Cheng, Lisa L.-S. 2017. Verb copying. In Rint Sybesma, Wolfgang Behr, Yueguo Gu, Zev Handel, C.-T. James
Huang & James Myers (eds.), Encyclopedia of Chinese Language and Linguistics, Vol. 4, 488–490. Leiden:
Brill.
Embick, David. 2007. Linearization and Local Dislocation: derivational mechanics and interactions. Linguistic
Analysis 25. 303–336.
Embick, David & Rolf Noyer. 2001. Movement operations after syntax. Linguistic Inquiry 32. 555–595.
Harbour, Daniel. 1999. The two types of predicate clefts: classical Hebrew and beyond. MIT Working Papers in
Linguistics 34. 859–175.
Harbour, Daniel. 2008. Kilvaj predika, or predicate clefts in Haitian. Lingua 118. 853–871.
Huang, C.-T. James, Y.-H. Audrey Li & Yafei Li. 2009. The syntax of Chinese. Cambridge: Cambridge University
Press.
Kratzer, Angelika. 1996. Severing the external argument from its verb. In Johan Rooryck & Laurie Zaring (eds.),
Phrase structure and the lexicon, 109–137. Dordrecht: Kluwer.
Landau, Idan. 2006. Chain resolution in Hebrew VP-fronting. Syntax 9. 32–66.
Scott, Tessa. 2021. Two types of resumptive pronouns in Swahili. Linguistic Inquiry 52. 812–833.
Sybesma, Rint. 1999. The Mandarin VP. Dordrecht: Kluwer.
Tang, C.-C. Jane. 1990. Chinese phrase structure and the extended X’-theory. Ithaca, NY: Cornell University disser-
tation.
van Urk, Coppe. 2018. Pronoun copying in Dinka Bor and the Copy Theory of Movement. Natural Language and
Linguistic Theory 36. 937–990.