Professional Documents
Culture Documents
Li Li: Grammar in Cross-Linguistic Perspective
Li Li: Grammar in Cross-Linguistic Perspective
Li Li: Grammar in Cross-Linguistic Perspective
Grammar in
Teruhiro Ishiguro is Professor and Dean of the
Faculty of English and American Language and
Cross-Linguistic
Literature, Tokushima Bunri University, Japan, and
Kang-kwong Luke is Professor of Linguistics,
Nanyang Technological University, Singapore.
Perspective
Both of them specialise in the interfaces be-
tween syntax, semantics, and pragmatics, and
have a common interest in grammatical studies
The Syntax, Semantics, and Pragmatics
of Japanese, Chinese and English from a com- of Japanese and Chinese
parative perspective.
ISBN 978-3-03911-445-0
Grammar in
Teruhiro Ishiguro is Professor and Dean of the
Faculty of English and American Language and
Cross-Linguistic
Literature, Tokushima Bunri University, Japan, and
Kang-kwong Luke is Professor of Linguistics,
Nanyang Technological University, Singapore.
Perspective
Both of them specialise in the interfaces be-
tween syntax, semantics, and pragmatics, and
have a common interest in grammatical studies
The Syntax, Semantics, and Pragmatics
of Japanese, Chinese and English from a com- of Japanese and Chinese
parative perspective.
Peter Lang
Grammar in Cross-Linguistic Perspective
Linguistic Insights
Studies in Language and Communication
Volume 57
Advisory Board
Vijay Bhatia (Hong Kong)
Christopher Candlin (Sydney)
David Crystal (Bangor)
Konrad Ehlich (Berlin / München)
Jan Engberg (Aarhus)
Norman Fairclough (Lancaster)
John Flowerdew (Hong Kong)
Ken Hyland (Hong Kong)
Roger Lass (Cape Town)
Matti Rissanen (Helsinki)
Françoise Salager-Meyer (Mérida, Venezuela)
Srikant Sarangi (Cardiff)
^
Susan Šarcević (Rijeka)
Lawrence Solan (New York)
Peter M. Tiersma (Los Angeles)
PETER LANG
Bern • Berlin • Bruxelles • Frankfurt am Main • New York • Oxford • Wien
Teruhiro Ishiguro & Kang-kwong Luke (eds)
Grammar in
Cross-Linguistic
Perspective
PETER LANG
Bern • Berlin • Bruxelles • Frankfurt am Main • New York • Oxford • Wien
Bibliographic information published by die Deutsche Nationalbibliothek
Die Deutsche Nationalbibliothek lists this publication in the Deutsche National-
bibliografie; detailed bibliographic data is available on the Internet
at ‹http://dnb.d-nb.de›.
ISSN 1424-8689
ISBN 978-3-03911-445-0 pb. ISBN 978-3-0351-0576-6 eBook
Printed in Switzerland
Table of Contents
YOICHIRO HASEBE
A Cognitive Approach to Compound Kango VNPs
in Japanese ........................................................................................ 9
GAO HUA
Tag-Questions in Mandarin Chinese ..............................................43
NORIFUMI ITO
Information in Discourse and Language as Vehicle
of Thought ......................................................................................65
HIDEMATSU MIURA
Focus-driven Semantic Reflexivity in Japanese ............................89
WINNIE CHOR
From Nonsubjective to Subjective to Intersubjective –
a Pathway of Semantic Change in Grammaticalization:
the Case of faan in Cantonese ......................................................125
YUTAKA SHINODA
Japanese Particle Na as a Marker of the Speaker’s
Subjective Judgment ‘Here and Now’ .........................................149
LEUNG WAI-MUN
Evidential Particles in Cantonese: the Case of wo3
and wo5 ........................................................................................175
6 Table of Contents
HAN YANG
Tense, Aspect and Verbal Morphemes in Chinese, Japanese
and English ...................................................................................205
MASANOBU HORIGUCHI
Japanglish in Katakana and a Comparison with Loanwords
in Cantonese .................................................................................243
Preface
Last but not least, the two editors would like to thank the two univer-
sities, particularly President Murasaki of TBU, for their unfailing
support of the exchange programme. They are also grateful to their
able and diligent colleagues and graduate students for their valuable
input into this memorable and meaningful process.
9
YOICHIRO HASEBE
A Cognitive Approach to
Compound Kango VNPs in Japanese
1. Introduction
1.1 Phenomenon
(2) a. Single semantically coherent VNPs are composed of two independent kango
words.
b. The second element of the compound has both nominal and verbal charac-
teristics.
c. The VNP as a whole has not completely lexicalized into one word, pre-
serving syntactic and semantic analyzability.
(3) a. ⩦័
KŠμ‰»(shuukan-ka,
(shuukan-ka,‘conventionalization’)
‘conventionalization’)
b. ᪥ᮏ〇
“ú–{•» (nihon-sei,
(nihon-sei,‘Japan-made’)
‘Japan-made’)
c. ᣮᡓ
Ä’§•í (sai-chousen,
(sai-chousen, ‘re-try’)
‘re-try’)
d. ึ㣕⾜
‰”ò•s (hatsu-hikou,
(hatsu-hikou,‘first-flight’)
‘first-flight’)
(4) a. 㦵ᢡ
œ • Ü (kossetsu,
(kossetsu,‘bone-break’)
‘bone-break’)
b. ධ㝔
“ü‰@ (nyuu-in,
(nyuu-in, ‘enter-hospital,
‘enter-hospital, hospitalized’)
hospitalized’)
c. ᡂ㛗
¬’· (sei-chou,
(sei-chou,‘become-long,
‘become-long,grow’)
grow’)
d. ど⫈
•‹’® (shi-chou,
(shi-chou,‘view
‘viewand
andlisten’)
listen’)
There are a great number of this kind of Japanese words, and they
account for a large part of the overall verbal vocabulary. Many of
these words, however, are lexicalized, and accordingly their compo-
nent morphemes are neither highly independent nor productive. In
fact, these two-character kango VNPs contribute to the formation of
complex VNPs as component structures. Thus, a brief look at the
structure of these simple kango VNPs is made in Section 2.2. The
complex kango VNPs that meet the first condition in (2) examined in
this chapter consist of two parts, each of which is composed of a
smaller unit.
As (2b) stipulates, the second element of complex kango VNPs
functions either verbally or nominally. For instance, shuushuu of
jouhou-shuushuu in (1a) can be used either as an NP or as a semanti-
cally essential part of a VP. Since complex VNPs inherit this charac-
teristic, they can be part of an argument of a VP as in (5a). On the
other hand, they can also combine with the light-verb suru (roughly
meaning ‘do’) as in (5b) to make up a full-fledged VP construction.
(5) a. ᙼ䛻ሗ㞟䜢௵䛫䜛䚹
kare-ni jouhou-shuushuu-o makaseru
he-DAT data-gathering-ACC leave
‘(I) leave data-gathering to him.’
A Cognitive Approach to Compound Kango VNPs in Japanese 11
b. 䜲䞁䝍䞊䝛䝑䝖䛷ሗ㞟䛩䜛䚹
intaanetto-de jouhou-shuushuu-suru
Internet-with data-gathering-suru
‘(I) do data-gathering on the Internet.’
1.2 Framework
1.3 Outline
2 The foundation of LCS theory was laid by Jackendoff (1990). This theory deals
with semantic aspects to language much more than other theories derived from
Generative Grammar. Still, LCS presupposes the existence of so-called se-
mantic primitives and assumes that the meanings of words are generated through
the process of applying semantic functions to those primitives in combination
with other constants and valuables. The latter fact separates the two approaches,
CG and LCS, in a fundamental sense, even though they share a strong affinity
for the semantic aspects of language.
A Cognitive Approach to Compound Kango VNPs in Japanese 13
There are several reasons why kango words have become an essen-
tial part of the Japanese language. Sato (1987) gives three sugges-
tions. First, Chinese words (kango) and their written forms were quite
useful in importing and assimilating new concepts from foreign coun-
tries having different cultures. In the old and medieval ages, those
concepts mainly came from China, and the ways in which they were
expressed were directly adopted from China with some necessary
(mostly phonological) modifications. In the modern period, words
of Chinese-origin and characters were again used in translating the
concepts imported from Western culture.
Second, kango made it possible to concisely express complex
ideas and concepts. Japanese grammar uses post-nominal particles
to represent relationships. In Chinese, on the other hand, there is no
need for such extra grammatical elements; only word order matters.
Moreover, since a semantic unit consists of one character and one
syllable in Chinese, kango words tend to be more succinct as com-
pared to their non-kango counterparts.3
Third, the word formation of kango is morphologically simple,
compared to the original Japanese word formation. Because of the
building-block-like characteristics of kanji, a variety of meanings
can be expressed in a semi-nominal form by constructing a sequence
of various character lengths. Because of this characteristic, kango
can nominalize concepts that are complex enough to be alternatively
expressed in a full sentence.4
3 On the other hand, there is a drawback to adopting and coining many kango
words: it has created a great number of homonyms in Japanese. This is because
the Japanese phonological system is simpler than the Chinese system. Chinese
can place four different tones on a single vowel type, whereas Japanese does
not have such differentiation of vowel tones.
4 One marginal but notable fact is that kango words over a certain length are
often shortened into contracted forms without losing much of the original sym-
bolic relationship between the form and meaning.
A Cognitive Approach to Compound Kango VNPs in Japanese 15
For the purpose of the present research, the last point is quite
significant. Japanese, originally an agglutinative language that needs
particles placed right after each argument nominal, is equipped also
with a characteristic of Chinese, an isolating language, on the level
of word formation. The next section will introduce some patterns of
kango VNP formation.
Simple kango VNs are defined as verbal nouns that are composed of
two kanji and are able to combine with the light verb suru or with the
sequence of accusative case marker o- plus suru. As illustrated in
(6), there are broadly four types of simple kango VNs. Although each
type has its own formal pattern, its instances are rather lexicalized
and thus the patterns are not necessarily highly productive.5
(8) a. ⰼ䛜㛤䛟䚹
hana-ga hiraku
flower-NOM bloom
‘A flower comes into bloom.’
b. 㛤ⰼ (kai-ka, ‘bloom-flower’)
(9) a. 㞵䛜㝆䜛䚹
ame-ga furu
rail-NOM fall
‘Rain falls.’
b. 㝆㞵 (kou-u, ‘fall-rain’)
(10) a. ᬒẼ䛜ᚋ㏥䛧䛶䛔䜛䚹
keiki-ga koutai-siteiru
economy-NOM slowdown-suru-PROG
‘Economy is slowing down.’
b. ᬒẼᚋ㏥ (keiki-koutai, ‘economy-slow down’)
(11) a. ㏻䛜䛧䛶䛔䜛䚹
koutsuu-ga juutai-siteiru
traffic-NOM congestion-suru-PROG
‘Traffic is congested.’
b. ㏻ (koutsuu-juutai, ‘traffic-congestion’)
(12) a. ᙼ䛿⚾䛾䛻ྠព䛧䛯䚹
kare-wa watashi-no an-ni doui-shita
he-TOP I-GEN plan-DAT agree-suru-PAST
‘He agreed to my plan.’
b. 䛣䛾ィ⏬䛾ᐇ⌧䛻䛿䚸ᙼ䛾ྠព䛜ᚲせ䛰䚹
kono-keikaku-no jitsugen-ni-wa kare-no doui-ga hitsuyou-da
this-plan-GEN realization-DAT-TOP he-GEN agree-NOM necessary-AUX
‘His agreement is necessary to carry out this plan.’
(13) a. ୖ₇㛫䜢ၥ䛔ྜ䜟䛫䜛
jouen-jikan-o toiawaseru [verbal]
performance-time-ACC inquire
‘to inquire the performance time (of a play)’.
b. ୖ₇㛫䛾ၥ䛔ྜ䜟䛫
jouen-jikan-no toiawase [nominal]
performance-time-GEN inquiry
‘inquiry of the performance time (of a play)’
(14) a. ㈈ᕸ䜢䛺䛟䛩
saifu-o nakusu [verbal]
wallet-ACC lose
‘to lose (my) wallet’
b. *㈈ᕸ䛾䛺䛟䛧
*saifu-no nakushi [nominal]
wallet-GEN lose
‘loss of (my) wallet’
c. ㈈ᕸ䛾⣮ኻ
saifu-no funshitsu [kango VN]
wallet-GEN disappearance
‘loss of (my) wallet’
Compound kango VNPs are classified into the four types in (15). Note
that the head of the compound is always the second element of the
word sequence. Although the head elements are all verbal in (15), the
subordinate elements (the first elements) have different characteris-
tics in their part of speech (verbal, adverbial, object nominal, or sub-
ject nominal). In this section, I will examine each of the four types of
compound kango VNPs using the theoretical framework of CG.
3.2 V + V
(16) a. ಖᏲ⟶⌮
•Û•çŠÇ—• 䇺maintain-administrate䇻)
(hoshu-kanri,
(hoshu-kanri, ‘maintain-administrate’)
b. ㍺ධ㈍ (yunyuu-hanbai,䇺import-retail䇻)
—A“ü”Ì”„(yunyuu-hanbai, ‘import-retail’)
c. ෭ಖᏑ (reitou-hozon,䇺freeze-keep䇻)
—â“€•Û‘¶(reitou-hozon, ‘freeze-keep’)
d. ᳨ᰝධ㝔
ŒŸ•¸“ü‰@ 䇺checkup-hospitalize䇻)
(kensa-nyuuin,
(kensa-nyuuin, ‘checkup-hospitalize’)
20 Yoichiro Hasebe
Compound VNPs of the V+V type (16) are composed of two kango
words, both having verbal semantic features. There are several cat-
egories among VNPs of this type. First, in VNPs like hoshu-kanri
(‘maintain-administrate’), two similar event concepts are juxtaposed
as equals. Second, in VNPs like yunyuu-hanbai (‘import-retail’), two
related but separate events are arranged in chronological order. Third,
in VNPs like reitou-hozon (‘freeze-keep’), the first element expresses
the way in which the event in the second element is carried out. Fourth,
in VNPs like kensa-nyuuin (‘checkup-hospitalize’) the first element
introduces the purpose of the event expressed in the second element.
Even though the first elements of all of the above subcatego-
ries in this VNP type have some kind of event frame in their concep-
tual structures, the degree of saliency is quite varied. The first ele-
ment of the VNPs in the first type (hoshu-kanri) is relatively high in
its saliency but not as high as that of the second element, which is the
head of the VNP. Still, hoshu in hoshu-kanri is conceptually promi-
nent enough to be regarded appositional to the head kanri. The first
element of a VNP in the second category (yunyuu-hanbai) can repre-
sent an event that is independent of the event depicted by the second
element, as yunyuu (‘import’) can be thought of as a separate event
preceding the next event hanbai (‘retail’). Yet the compound VNP as
a whole does not profile both of the events equally; it is always the
second event that is more salient than the other and thus profiled. For
this reason, yunyuu-hanbai does not refer to a kind of importing but
to a kind of retailing in a default interpretation of the expression.7
The first elements of VNPs in the third (reitou-hozon) and fourth
(kensa-nyuuin) categories are not highly prominent either. Although
events such as reitou (‘freeze’) and kensa (‘checkup’) have concrete
semantic contents, they are regarded only as part of the ground against
which the main events hozon (‘keep’) and nyuuin (‘hospitalize’) are
profiled. Thus, compound kango VNPs of V+V type, which feature
two related events in their conceptual structure, are varied in terms
of the saliency of their first elements. The relationship among the
subtypes is schematically captured in Figure 1.
Figure 1.
3.3 Adv + V
(17) a. 㝔ෆឤᰁ
‰@“àŠ´•õ (innai-kansen,
(innai-kansen, ‘inside
䇺inside hospital-infect’)
hospital-infect䇻)
b. ᪂つཧධ
V‹K•Q“ü (shinki-sannyuu,
(shinki-sannyuu,䇺newly-enter䇻)
‘newly-enter’)
c. 㧗㏿ᅇ㌿ 䇺rapidly-rotate䇻)
(kousoku-kaiten,‘rapidly-rotate’)
‚‘¬‰ñ“] (kousoku-kaiten,
Compound kango VNPs of the Adv+V type (17) are those that are
composed of a head VNP and another preceding kango that modifies
it.8 There are three subtypes in compound VNPs of this type. First,
VNPs like innai-kansen (‘inside-hospital-infect’) have a structure in
which the first element specifies the location in which the event de-
scribed by the second takes place. Second, VNPs like shinki-sannyuu
(‘new-enter [to a business competition]’) have a structure in which
the first element specifies the nature or quality of the event invoked
by the second. Third, VNPs such as kousoku-kaiten (‘rapidly-rotate’)
have a structure in which the first element stipulates the degree of
the speed, rate, duration, etc. of the event described by the second.
In all of these subtypes, the conceptual structure has two com-
ponent structures: one contains an indication of a specific point or
position in a domain or scale of some kind, while the other contains
the event concept that is given the highest degree of prominence
within the overall structure. In innai-kansen, the domain featured is
a spatial/locative one. In shinki-sannyuu, a time scale is featured; in
kousoku-kaiten, the rate/speed scale. The composition process of such
a domain expressed by the first element and an event concept ex-
pressed by the second element is illustrated in Figure 2.
8 Since a kango VNP has both verbal and nominal aspects, it is equally possible
to regard an element modifying it as an adverb or adjective. Since this chapter
focuses on the verbal aspects of kango VNPs, I refer to this type of VNP in
which the first element modifies the second as ‘Adv + V Type’. In a different
context, it would be appropriate to call it ‘Adj+N Type’.
A Cognitive Approach to Compound Kango VNPs in Japanese 23
Figure 2.
In Figure 2, the box on the lower left represents the domain/scale and
a point within it invoked by the first element of the compound. This
abstract structure is merged into a rather concrete event structure in
the box on the right, which is invoked by the second element of the
compound. Finally, the composite structure depicted in the third box
on the top is constructed. Note that the composite structure inherits
the basic configuration of the component structure on the right side,
which is the profile determinant and is thus enclosed by a bold rect-
angle.
3.4 Obj + Vt
CG stipulates that even a case particle or suffix has its own concep-
tual structure, and different markers must be treated as such. Pres-
ently, I will not examine further the unique conceptual structures of
each of the case-marking particles -o and -ni. Compound kango VNPs
of the Obj+V type are depicted in Figure 3.
Figure 3.
3.5 Sbj + Vi
Figure 4.
(24) a. ኴ㑻䛜ᩓṌ䛧䛯䚹
Taro-ga sanpo-shita
Taro-NOM walk-suru-PAST
‘Taro took a walk.’
b. ኴ㑻䛜ᩓṌ䜢䛧䛯䚹
Taro-ga sanpo-o-shita
Taro-NOM walk-ACC-shita
‘Taro took a walk.’
(25) a. ኴ㑻䛜⮬ẅ䛧䛯䚹
Taro-ga jisatsu-shita
Taro-NOM suicide-suru-PAST
‘Taro committed suicide.’
b. ኴ㑻䛜⮬ẅ䜢䛧䛯䚹
Taro-ga jisatsu-o-shita
Taro-NOM suicide-ACC-shita
‘Taro committed suicide.’
28 Yoichiro Hasebe
(26) a. 㛗ዪ䛜ㄌ⏕䛧䛯䚹
Choujo-ga tanjou-shita
first-daughter-NOM born-suru-PAST
‘(Our) first daughter was born.’
b. ? 㛗ዪ䛜ㄌ⏕䜢䛧䛯䚹
? Choujo-ga tanjou-o-shita
first-daughter-NOM born-ACC-suru-PAST
‘(Our) first daughter was born.’
(27) a. ᗣ䜢⥔ᣢ䛩䜛
kenkou-o iji-suru
health-ACC maintain-suru
‘(I) maintain (my) health.’
b. ? ᗣ䛾⥔ᣢ䜢䛩䜛
?kenkou-no iji-o-suru
health-GEN maintain-ACC-suru
‘(I) maintain (my) health.’
(28) a. ᒣ⏣䛿㌴䜢ᡤ᭷䛧䛶䛔䜛
Yamada-wa kuruma-o shoyuu-shiteiru
Yamada-TOP car-ACC own-suru-PROG/STATE
‘Yamada owns a car.’
b. ? ᒣ⏣䛿㌴䛾ᡤ᭷䜢䛧䛶䛔䜛
?Yamada-wa kuruma-o shoyuu-o-shiteiru
Yamada-TOP car-ACC own-ACC-suru-PROG/STATE
‘Yamada owns a car.’
(29) a. The VNP represents an event that is intentionally carried out by an agent.
b. The VNP represents an event that has specific starting and ending points.
c. The VNP does not represent an event that is emotional/psychological.
(Tanomura 1988, my translation with minor modifications)
A Cognitive Approach to Compound Kango VNPs in Japanese 29
33b). Thus, it can be said that the condition regarding the combina-
tion of simple kango VNs and the -o-suru sequence is also valid when
applied to complex VNPs.
It is now clear that compound kango VNPs combine with -suru
and -o-suru just as simple kango VNPs do, as long as the conditions
in (29) are met. This is quite impressive given that compound VNPs
have rather complex structures both syntactically and semantically.
For a VNP to have -suru (roughly meaning ‘do’) immediately follow-
ing it, the VNP must be describing something that the speaker can
conceive of as one single conceptual unit. Moreover, the particle -o is
reasonably considered to evoke a relationship between the vantage
point of a conceptualizer and a view directed at the target object, re-
quiring the VNP preceding it be a ‘coherent whole’. Thus, it is sug-
gested that a well-formed compound VNP has a conceptual auto-
nomousness that gives it a status as a coherent unit. How can a com-
pound VNP, which sometimes is composed of a structure as complex
as an argument structure, be perceived as a single, coherent concep-
tual unit? This problem is further investigated in Section 5.
(34) a. ᩳ㝧䛾㕲㐨⏘ᴗ䜢ᵓ㐀㌿䛩䜛䚹
shayou-no tetsudou-sangyou-o kouzou-tenkan-suru
declining-GEN railroad-industry-ACC structure-transform-suru
‘(They) restructure the declining railroad industry.’
b. ෑ㢌䛾ᩥ䜢ពゎ㔘䛩䜛䚹
boutou-no bun-o imi-kaishaku-suru
beginning-GEN sentence-ACC meaning-interpret-suru
‘(I) interpret the meaning of the opening sentence.’
with the verbal head of the compound.9 Those external objects pro-
vide a conceptual setting or domain in which the event expressed by
the predicate VNP can be evoked. The conceptual autonomousness
of the compound VNP is preserved because the two objects, which
have syntactically different realizations, are practically identical. This
is illustrated in Figure 5.
Figure 5.
(35) a. *ኻᴗ⪅䜢ᙅ⪅ᩆ῭䛩䜛䚹
*shitsugyousha-o jakusha-kyuusai-suru
the-unemployed (people)-ACC the-weak (people)-assist-suru
‘(They) assist the unemployed – weak – people.
b. ?ኴᖹὒ䛾ᓥ䚻䜢⮬↛◚ቯ䛩䜛䚹
?taiheiyou-no shimajima-o shizen-hakai-suru
Pacific-Ocean-GEN islands-ACC nature-destroy-suru
‘(They) destroy the nature of islands of the Pacific Ocean.’
9 In this regard, the construction with internal and external object NPs is com-
pletely differentiated from syntactically similar-looking phenomena such as
ditransitive constructions. (cf. John gave Mary a flower.)
A Cognitive Approach to Compound Kango VNPs in Japanese 33
On the other hand, compounds of a transitive VN and the subject are not
considered natural, although they almost make perfect sense in terms of
causal structure of events (38a, 39a). Note that if the same events are coded
differently in accordance with the previously described requirements,
then the resulting expressions are considered appropriate (38b, 39b).
34 Yoichiro Hasebe
Many languages are categorized into one of two groups: those that are
based on a nominative/accusative case system and those that are based
on an ergative/absolutive case system. The primary difference be-
tween these two systems is as follows. In languages based on a nomi-
native/accusative system, the subject NPs of an intransitive verb and
a transitive verb is given an identical morphological marking (nomi-
native case, NOM), but a different marking is given to the object NP
of a transitive verb (accusative case, ACC). In languages based on an
ergative/absolutive case system, the subject of a transitive verb is
given a certain morphological marking (ergative case, ERG), but the
subject of an intransitive verb and the object of a transitive verb lack
explicit markings of any kind (absolutive case, ABS). This general
contrast between the two systems can be described in matrices (40)
and (41) below. (The symbol ‘=>’ indicates transitivity of verb.)
not only in English but also in Japanese as well as many other lan-
guages. The causal relationship between participants of an event can
be directly reflected in linguistic forms, a prototypical case of which
is a construction containing transitive verbs with a nominative sub-
ject and an accusative object.
The other natural path corresponds to the conceptual autonomy
of the event. In this natural path, based on what Langacker calls ‘Au-
tonomous/Dependent (A/D) Layering’, the primary concern is the
emergence of an event or the state of a thematic object having gone
through some kind of change, not the causal relationship among the
participants that supposedly played important roles in the event. The
term ‘A/D Layering’ reflects the fact that such an event construal
can have multiple layers in its cognitive structure.
To illustrate how these two natural paths work, let us take an
example from Langacker (1991). He suggests that a single event can
often be linguistically expressed either in a transitive construction or
in an intransitive construction: the former is based on Action Chain,
which is considered the default mode of cognition widely observed
in English; the latter is based on A/D Layering, which is linked to
phenomena that display characteristics of ergative/absolutive linguis-
tic coding. Consider the event “Floyd picks up a hammer, and he
breaks a glass.” English allows one to express this event either as
seen in (42a) or (42b)
(44) a. ᴗ䛾♫ဨເ㞟
kigyou-no shain-boshuu
company-GEN employee-recruit
‘a company’s recruiting of employees’
b. * ♫ဨ䛾ᴗເ㞟
*shain-no kigyou-boshu
employee-GEN company-recruit
‘a company’s recruiting of employees’
c. ᴗ䛜♫ဨ䜢ເ㞟䛧䛶䛔䜛䚹
kigyou-ga shain-o boshuu-shiteiru
company-NOM employee-ACC recruit-suru-PROG/STATE
‘A company is recruiting employees.’
38 Yoichiro Hasebe
(46) a. ᨺⅆ䛻䜘䜛ᐙᒇ↝
houka-niyoru kaoku-zenshou
arson-by house-burn-down
‘burn-down of a house because of an arson attack’
b. * ᐙᒇ䛾ᨺⅆ↝
**kaoku-no
k k houka-zenshou
house-GEN arson-burn-down
‘burn-down of a house because of an arson attack’
c. ᐙᒇ䛜ᨺⅆ䛻䜘䜚↝䛧䛯䚹
kaoku-ga houka-niyori zenshou-shita
house-NOM arson-by totally-burn-suru-PAST
‘A house burnt down because of an arson attack.’
6. Conclusion
References
10 One of the most important areas for future research should be the comparison
between Japanese complex kango formation and Chinese syntax. As is men-
tioned in Section 2.2, many Japanese kango VNPs inherit the structure of their
original Chinese expressions. This being so, it seems essential to investigate
the grammatical similarities or differences between the two languages as well
as the cognitive basis for them.
A Cognitive Approach to Compound Kango VNPs in Japanese 41
1. Introduction
The tag question in Chinese has attracted attention from many gram-
marians and insights have been gained about its forms and uses.
Nonetheless, there are only a small number of studies that are di-
rectly focused on this apparently ‘special’ question type (Shao Jingmin
1996: 17), which mainly include Niu Baoyi (2001, 2002) and Shao
Jingmin (1996: Chap. 8), while in many others it is only touched
upon in discussions on the grammar of questions in general (see e.g.,
Xu Jie / Zhang Linlin 1993; Zhang Bojiang 1997; Xu Shenghuan
1999). In the literature, the interest on this question type is mainly on
its grammatical status in the grammar of questions in Chinese (i.e.,
whether it should be counted as one independent question type) (e.g.,
Lin Yuwen 1985; Shao Jingmin 1996). Relevant to this, attention has
been given to the degree of grammaticalization (䈝⌅ॆ〻ᓖ ) of this
question form (such as when compared with other question forms in
Chinese like the ma-question and the A-not-A question (Zhang
Bojiang 1997)), and its degree of doubt (⯁䰞〻ᓖ ) compared with
other question types (Xu Jie/Zhang Linlin 1993; Xu Shenghuan 1999).
In this study these issues will be re-addressed, and a unified
analysis will be provided of the tag question in Chinese via a cogni-
tive-functional account of the possible motivating factors that are
involved in its formation. Specifically, we try to answer this ques-
tion: if one takes a functional view and considers language as evolved
in response to cognitive-communicative demands (Givón 2001, 2002),
how might the formation of tag questions have been motivated by
these two considerations? It is believed that once an adequate an-
swer to this question is found, a clear picture will come out regarding
the grammatical status of this question type as well as its various
linguistic properties.
(3) ৫ྭнྭ?
ni qu hao bu hao?
2SG go good not good
‘You go, OK?’
(4) ྩᱟнᱟॱҍ?
ta shi bu shi cai shijiu sui?
3F.SG COP not COP only nineteen year
‘Is she only nineteen?’
2 Cheng does not say explicitly that the sentence-final sĶ-m“-sĶ and kám-m“-sĶ
‘isn’t it?’ are tags, probably due to two factors. First, there is presumably no
pause between these constituents and their preceding part (therefore no comma
is used in Cheng’s examples, such as I IJ-hiáu IJng tĶsĶ-μ“-sĶ? ‘Can he use chop-
sticks or not?’). Second, these forms can also appear in split as a higher predi-
cate sĶ ‘be’/kám conjoined with its negative μ“-sĶ ‘not be’. However, consid-
ering the apparent comparability between these forms and Mandarin ᱟнᱟ
(shi-bu-shi ‘Is that right?’) in tag position, there is a strong and good reason
for them to be analyzed as tags when they appear in sentence-final positions.
48 Gao Hua
(5) Ԇ䐏а䎧ᶕⲴੇ?
ta gen ni yiqi lai de ma?
3M.SG with 2SG together come PRT Q
‘Has he come with you?’
(6) Ԇ䐏а䎧ᶕⲴᱟੇ?
ta gen ni yiqi lai de shi ma?
3M.SG with 2SG together come PRT right Q
‘He has come with you, right?’
Such facts indicate that there may still be room for a more in-depth
investigation of the grammatical status of the tag question in Chi-
nese and its relationship with other question forms particularly from
a cognitive-functional viewpoint.
3 These tag particles are not absolutely bound, and there can be a little pause just
before them.
Tag-Questions in Mandarin Chinese 49
(7) ӺཙॱҼਧሩੇ˛
jintian shi’er hao dui ma?
today twelve date right Q
‘Today is the twelfth, right?’
Unfortunately, Niu’s description does not explain why the tag ques-
tion conveys a low degree of doubt. Comparing (7) and (8), and also
the earlier examples of (5) and (6), it can be noticed that the tag
questions express a stronger speaker assumption about the truth or
falsity of the proposition being asked about, and request confirma-
tion or agreement of that assumption.
(8) ӺཙॱҼਧੇ˛
jintian shi’er hao ma?
today twelve date Q
‘Is it the twelfth today?’
3. Cognitive-communicative motivations
of the tag construction
I argue that the combination of the statement stem and the Q-tag may
be regarded as a tag construction which has a constructional mean-
ing that expresses both speaker certainty and speaker uncertainty, in
perception of hearer knowledge or attitude, about the truth or accu-
racy of a statement. On the one hand, the form and meaning of the
tag construction are interdependent; on the other hand, the meaning
cannot be derived directly by adding together the meanings of the
statement and the Q-tag (Croft 2001).
Compared with previous research on tag questions in Chinese
which focused on how the tag leads a statement to a question and
relevant speaker belief to doubt, the present understanding of the tag
construction draws attention to the point that such dynamic episte-
mology about the issue concerned is related to how participants in a
conversation view the knowledge of each other. Put in another way,
the tag not only conveys the speaker’s attitude with respect to the topic
or statement at issue, it also relates to the hearer’s knowledge and/or
attitude (i.e. what the speaker thinks the hearer knows or believes). In
a similar vein, Norrick (1995) has argued for the role of tags such as
the English hunh in coding evidentiality in conversation. Cuenca (1997)
has also found that the tag construction is designed to be an interactive
construction that must be interpreted in a conversational exchange.
The interactive nature of the tag construction may be evidenced
clearly by its placement in conversational sequences like the follow-
ing one:
S
1 Sa Uh, (.) նަԆⲴ, ަԆ㲭❦␈བྷ㣡ഝᡁᜣਟ㜭ਚঐ
Uh, (.) dan qita de ne, qita suran tao da huayuan wo xiang keneng zhi zhan
but other GEN TOP other though Tao Da garden I think maybe only take
2 (1.1) Ⲯ࠶ѻҼॱӄᐖਣ.
baifenzhiershiwu zuoyou.
twenty-five-percent or-so
‘Uh, (.) but how about the others, among others, although victims in Taoda
Garden I guess may only take (1.1) 25 percent or so.’
3 (.)
52 Gao Hua
In this excerpt, Sa was talking to her friend Jay about her estimation
of the percentage of SARS victims in a housing estate. After her
making the statement, there was a noticeable pause (line 3) before
she proceeded to check this information with Jay the hearer, with the
question tag ሩ (dui ba ‘Is that right?’). Interestingly, at the same
time the tag was uttered (line 5), Jay took the floor and made a con-
firmation (line 4). Different from examples given in previous sec-
tions of this chapter, the early ‘arrival’ of the confirmation from the
hearer seemed to truncate the speaker’s tag question into two parts
(lines 1-2 and line 5). The declarative part of the tag question in lines
1-2 was virtually made a declarative question (Gao Hua 2003). And
it was possibly due to the lack of an immediate uptake from Jay, as
indicated by the noticeable pause in line 3, that the tag was used,
which may be seen as a further and a more explicit attempt by Sa to
invite verification from Jay of what had been said earlier.
With an understanding of the interactive nature of the tag con-
struction, some general characteristics about tag questions can be
well accounted for. First, tag questions only appear in conversations
where there are addressees actively present, and never in monologues
(monologues in a strict sense, with no ‘explicit’ or ‘implicit’ hearer).
It may be said that the tag question differs from other question types
in that it is not for asking of oneself. Compare (9) and the impossible
case in (10):
(11) ཙՊл䴘ˈᱟ˛
tian hui xiayu, shi ba?
sky shall rain right PRT
‘It may rain, right?’
(12) ཙՊл䴘DŽ/˛
tian hui xiayu ba./?
sky shall rain PRT
‘It may rain.’ / ‘Is it likely to rain?’
(13) A ≁䰤Ⲵᇩ᱃ˈ
minjian de rongyi,
folk GEN easy ‘Folk (literature studies) are easy (to get published).’
→ B ৽↓ҏ⋑ᴹཊቁӪ൘ᩎˈ
fanzheng ye meiyou duoshao ren zai gao hã
anyway also lack many people ASP do Q-TAG
‘Anyway there are not many people doing (folk literature), right?’
Tag-Questions in Mandarin Chinese 55
A ≁䰤Ⲵ㘼ф≁䰤᮷ᆖ⹄ウˈ
minjian de erqie minjian wenxue yanjiu a,
folk GEN moreover folk literature research TOP
䜭ྭࠐᒤ⋑ᴹᤋࡠᆖ⭏Ҷˈ…
dou hao ji nian meiyou zhaodao xuesheng le
even very a-few year no recruit student ASP
‘Folk, moreover, folk literature research has not recruited new postgrads in
the past few years.’
(17) Il (n’) est (pas) venu avec toi, pas vrai? (French)
3SG (NEG) be+3SG.PRES (NEG) come+PAST.PART with 2SG.ACC not true
‘He has come with you, right?’
(18) Ԇ䐏а䎧ᶕⲴᱟੇ?
ta gen ni yiqi lai de shi ma?
3M.SG with 2SG together come PRT right Q
‘He has come with you, right?’
(19) ᡰԕ⢩ྭਲ਼ˈ˛
suoyi te haoshi, hã?
so very delicious Q-TAG
‘So it’s very delicious, right?’
Tag-Questions in Mandarin Chinese 57
(20) ᓄ䈕Ҡњ˛⊐⭥Ⲵ⭥ݵ
yinggai mai ge chongdian de dianchi hã?
should buy CL recharge GEN battery Q-TAG
‘(We) should have bought a rechargeable battery, right?’
(21) 䘈ⵏн䭉ˈањҏ⋑┿ˈ˛
hai zhen bucuo, yige ye mei lou, hã?
still really not-bad one also not leak Q-TAG
‘(It’s) really not bad, (the dumplings are) still well-stuffed (after the boiling),
right?’
(22) → A 䘉ᣭቹⵏཏᮤ喀Ⲵˈ
ni zhe chouti zhen gou zhengqi de hã,
you this drawer really enough tidy PRT Q-TAG
‘This drawer of yours is really tidy, isn’t it?’
B ᱟᴹ⛩ҡˈ...
shi youdian luan, …
empha a-little disorder
‘(It’s) a little untidy indeed, …’
(23) A …
→ 䛓ཙᡁн䈤䛓њؑᱟੇ.
na tian wo bu gao ni shuo na ge xin shi ma hã.
that day I not tell you say that cl letter right Q Q-TAG
‘Didn’t I tell you that day about that letter?’
[*Ԇ䐏ᡁ䈤ˈ䈤䛓њؑˈᱟ࠶є⅑߉Ⲵˈ...
ta gen wo shuo, shuo na ge xin, shi fen liangci xie de, …
he with I say say that CL letter EMPHA divide two-times write PRT
‘He told me, saying that letter was written twice
B
[.
huh
‘Huh.’
(24) нᱟԆ䛓њ㫉ᗇᴤহᇣˈ(ㅁ)
A: £º²»ÊÇËûÄǸöÃÉμøüÀ÷º¦£¬(Ц)
ⵏᱟ䘉ṧݯ䈤ᴹᰦⲴⵏىᱟˈ䈤䎧ᶕ䈱䈤∄䈱ˈᴹᰦىᜣᜣˈ
B: £ºÕæÊÇÕâÑù¶ùÄãËμÓÐʱºòÕæμÄÊÇ£¬ËμÆðÀ´Ë-Ë
μ±ÈË-£¬ÓÐʱºòÏëÏ룬
䈦ੰᡁ䘈ᜣ∄ྲ䈤Ӻⴻ⋑ݯҖ ˈ
ÚÀѽÎÒ»¹ÏëÄرÈÈçËμ½ñ¶ùû¿´Êé¹þ£¬
ᜣӺⴻ⋑ݯҖӪᇦ⦻㔤ᇦҏн⸕䚃ᒢో 㛟ᇊⴻҖ ˈ
Ïë½ñ¶ùû¿´ÊéÈ˼ÒÍõά¼ÒÒ²²»ÖªμÀ¸ÉÂïÄعþ¿Ï¶¨¿´ÊéÄعþ£¬
а৫ⴻԆˈ࡛䈤Ӻⴻ⋑ݯҖ᰾ݯҏ⋑ⴻਾݯҏ⋑ⴻ㘱нⴻҖ
һȥ¿´Ëû£¬±ðËμ½ñ¶ùû¿´ÊéÃ÷¶ùҲû¿´ºó¶ùҲû¿´Àϲ»¿´Êé
˄ᑖㅁ˅Ӫᇦ䘉њҏ䇨Ӫᇦᓅᆀྭ䈤ਚ㜭䘉Ѹᜣˈ
£¨´øЦ£©È˼ÒÕâ¸öÒ²ÐíÈ˼Òμ××ӺðÉÄãËμÖ»ÄÜÕâôÏ룬
նᱟ䈤ˈࡠҶ∄ྲ䈤㘱ᐸˈ
μ«ÊÇÄãË죬μ½Á˱ÈÈçËμÀÏʦ£¬
䈤ӔъⲴᰦىҶ ˈ
Ëμ½»×÷ÒμμÄʱºòÁ˹þ£¬
Ӕь㾯ˈӪᇦ਼ṧᐳ㖞㔉Ⲵؙь㾯Ӕˈ
½»¶«Î÷£¬È˼ÒͬÑù²¼ÖøøÄãÁ©μĶ«Î÷½»£¬
ૡ䘉њ߉ᗇнྭнྭ৽↓ࡂаㇷӔࠪ৫ҶԆྭˈ
ÔÛÕâ¸öдμò»ºÃ²»ºÃ·´Õý»®À-һƪ½»³öÈ¥ÁËËûºÃ£¬
߉ᗇнྭнྭᴰਾҏӔнࠪ৫
дμò»ºÃ²»ºÃ×îºóÒ²½»²»³öÈ¥
˄ᑖㅁ˅ᩎᗇ㘱ᐸ䜭⋑࣎⌅
£¨´øЦ£©¸ãμÃÀÏʦ¶¼Ã»°ì·¨
ањањᴸӔⲴь㾯Ԇ䶎ᤆؙᴸԘᴸ䘈ӔнҶˈ
60 Gao Hua
A: ᱟੇˈ
B: 䙇ᗇн㹼ˈᴰਾ㾱ᱟ䍘䟿⢩儈ҏ⋑㿱ᗇ
⋑㿱Ԇ∄ૡ䘉њྭˈ
к⅑ᡁԜؙ...ˈ
5. Conclusion
structional meaning that expresses not only speaker certainty but also
speaker uncertainty, in perception of hearer knowledge or attitude,
about the truth or accuracy of a statement. On the basis of this, the
pragmatic characteristics of the tag question in Chinese are explained.
It is suggested that whether the tag question functions to check infor-
mation (by requesting confirmation or agreement) or to serve affec-
tive functions (such as mitigating the otherwise impoliteness in the
previous statement or facilitating discourse management) crucially
depends on the interpersonal perception of participants’ subjective
epistemology with regards to the topic or statement in question.
Moreover, this chapter also suggests an iconic form-function map-
ping for tag questions in Chinese. The mini-question tags are related
to uncertainty about ‘truth’, hence typically used to ask about the
truth or accuracy of a statement; while the particle tags are more
often found with the affective, especially the facilitative, function of
the tag question.
References
References in English
References in Chinese
Within the linguistic research carried out so far, there have been con-
tentious proposals as to the discourse function of information repre-
sented in linguistic expressions and utterances. The unified accounts
of information emerging from the surrounding discourse are heavily
burdened with allocating its complicated character to various presup-
posed domains of linguistic research. That is, discussions of informa-
tion in discourse inevitably give rise to the problem of syntax-prag-
matics and semantics-pragmatics interfaces. At this level of linguistic
investigation, that is, at the level of information in discourse, we are
supposed to make an explanation of the mental state of both addresser
and addressee in discourse in addition to that of the structural aspect
of what is physically written and uttered. The complicated links be-
tween pragmatics and syntax or semantics are also reflected in the
usage of terms for representing information in discourse. In the Prague
School, for instance, the term ‘Functional Sentence Perspective’ is
available. In Halliday’s publications (e.g., 1967) , the term ‘Informa-
tion Structure’ or ‘Theme’ is preferable. Chafe (1976) adopted the
term ‘Information Packaging’ and Vallduvi (1990) used the term
‘Informatics.’ In spite of the diversity of term usage mentioned here,
the expression ‘Information Structure’ or ‘Information Packaging’ is
preferable to others for representing information systematically built
into discourse for the specific purpose of expressing the mental state
of addresser and the way of affecting the mental state of addressee.
As a further step for setting the scene for our argumentation,
we should clarify the definition of information in discourse. Accord-
ing to Chafe (1976), information packaging has to do primarily with
‘how the message is sent.’ It strongly suggests that what information
packaging emerging out of linguistic expressions or utterances has
to do at first is to reflect the discourse strategy of addresser and the
way of affecting the mental state of addressee. After the completion
of the primary function of expressing the way of conveying what the
addresser intends to do, information packaging has to do with the
message itself; in other words, as a secondary function, information
packaging fulfills a role of expressing the propositional meaning.
68 Norifumi Ito
(1) I have been using the term packaging to refer to the kind of phenomena at issue
here, with the idea that they have to do primarily with how the message is sent
and only secondarily with the message itself, just as the packaging of tooth-
paste can affect sales in partial independence of the quality of the toothpaste
inside. (Chafe 1976: 28)
(2) That is, information packaging in natural language reflects the sender’s hypo-
theses about the receiver’s assumptions and beliefs and strategies. (Prince
1981b: 224)
cordance with their beliefs about the hearer: what s/he is thought to know, what
s/he is expected to be thinking about. (Prince 1986: 208)
(5) Let us consider one important use of declarative sentences, namely as means to
influence the addressee’s picture of the world. In such cases, the speaker as-
sumes that the addressee has a certain picture – or model — of the world and he
wants to change this model in some way. We might then identify THE OLD or
THE GIVEN with the model that is taken as a point of departure for the speech
act and THE NEW with the change or addition that is made in this model. OLD
will here be equivalent to PRESUPPOSED in one sense of the term. We can say
that the addressee receives ‘new information’ in the sense that he comes to know
or believe more about the world than he did before. What he believes may be true
or false – the information he gets about the world may be correct or incorrect.
If we accept that last statement, it follows that the object of his belief or the new
information must be something which is capable of being true or false – that is
what is usually called a proposition. Let us therefore call this kind of information
PROPOSITIONAL INFORMATION. (Dahl 1976: 38)
(6) We have complimentary soft drinks, coffee, Sanka, tea, and milk. Also compli-
mentary is red and white wine. We have cocktails available for $2.00. (Birner
1998: 312)
Information in Discourse and Language as the Vehicle of Thought 71
Let us now see inversion (8), in which both the preposed and the
postposed elements are mentioned in the prior discourse. Addition-
ally, we should note that only one phrase in the prior discourse can
evoke both the preposed and postposed constituents in inversion. The
preposed NP ‘authors’ is mentioned in the expression ‘Montagnier
and Barre, whose English was more fluent than Chermann’s,’ that is,
the expression for the larger set of authors. On the contrary, the
postposed NP ‘Barre’ refers to the subset of authors; it is evoked by
the NP ‘Montagnier and Barre, whose English was more fluent than
Chermann’s.’ All in all, the evoking phrase can refer to the superset
directly, while the postposed NP ‘Barre’ can be evoked by a set infer-
ence with the result that the superset is more familiar than the subset:
(8) Over a weekend, Montagnier and Barre, whose English was more fluent than
Chermann’s, hammered out a manuscript reporting the isolation of their new
retrovirus from Frederic Brugiere, who would henceforth be known in the sci-
entific literature as BRU. Listed first among the authors, the position tradition-
ally reserved for the researcher who has made the greatest contribution to the
work, was Francoise Barre. (Birner 1998: 316)
(9) Labor savings are achieved because the crew is put to better use than cleaning
belts manually; also eliminated is the expense of buying costly chemicals. (Birner
1994: 248)
(10) Evoked > Unused > Inferrable > Containing-Inferrable > Brand-New-Anchored
> Brand-New
(11) Enclosed Are the GEICO Home Insurance Rates You Requested. (Birner 1994:
250)
The preposed element ‘the house’ in inversions (13a) and (13b) can-
not play a role as the topic for the following clause, while the
postposed element ‘the rabbits’ becomes the topic of the following
clause. It has to be remembered that canonical sentence (13c) makes
it possible for both NPs to occur as the subject in the following clause.
Rochemont and Culicover (1990) conceive of the difference
of the acceptability of the above sentences as the evidence that the
postposed element in inversion has a special function, that is a pre-
sentational focus. In a similar vein, Bresnan (1994) argues that the
postposed NP plays a role as a presentational focus. The subject NP
is assumed to be inverted to introduce the referent on the scene de-
scribed by the preceding discourse. Here let us see the definition of
the presentational focus by Bresnan (1994).
(14) In presentational focus, a scene is set and a referent is introduced on the scene
to become the new focus of attention. In the core cases, a scene is naturally
expressed as a location, and the referent as something of which location is
Information in Discourse and Language as the Vehicle of Thought 75
predicated – hence, a theme. This imposes a natural selection of the <th loc>
argument structure. (Bresnan 1994: 90)
3. R-dislocations in English
All in all, we can support the view that the operation of left-shifting
constituents engenders the affective element of evaluation in a propo-
sitional meaning. In this Section we set out to analyse the association
of the right-shifting operation with information in a discourse and
additional meanings generated by that operation. There has been a
lot of research on R-dislocation constructions in English, including ,
Lambrecht (1981,1994), Ziv (1994), Ward and Birner (1996), Michae-
lis and Lambrecht (1996), Grosz and Ziv (1998), Gómez-González
(2001), and Gundel and Fretheim (2004). R-dislocation is illustrated
in example (18), not (19). (19) is so-called ‘afterthoughts’ by which
a speaker amends the referential mistake. It has to be noted that R-
dislocation should hold the coreferential relation between the pro-
noun and the right-shifted NP. On the contrary, the coreferential rela-
tion is not necessarily kept in afterthoughts (19).
(19) I met him, your brother, I mean, two weeks ago. (Grosz and Ziv 1998: 296)
(21) He’s terribly confusing, this Chomsky. I can’t figure him out. (Grosz/Ziv 1998:
300)
5 Refer to Ito (2005) for more information as to the Familiarity Scale of Prince
(1981b).
78 Norifumi Ito
(24) I saw Modern Times again yesterday. He is amazing, (this) Charlie Chaplin.
(Grosz/Ziv 1998: 302)
(25) I took my dog to the vet yesterday. He is getting unaffordable, the mangy old
beast. (Grosz/Ziv 1998: 302)
Here we have to assess the conclusion in Grosz and Ziv (1998) that
the R-dislocated NP is used attributively but not merely referentially.
As stated in Jaszczolt (2002), a definite description can generate two
interpretations, that is referential or attributive. For instance, sen-
tence (26) “can mean either that there was a particular person, say
Antonio Gaudi, who was insane, or that whoever designed the church
must have been mad. The first reading is referential, the other at-
tributive. Intuitively, the referential reading seems to be more natu-
ral or even unmarked” (Jaszczolt 2002: 133). According to the clari-
fication of the distinction of referential and attributive interpretations
in Jaszczolt (2002) and the analysis in Grosz and Ziv (1998), we
cannot help concluding that the hearer of utterance (25) may not
recover the exact referent of the R-dislocated NP if this NP is used
attributively. As opposed to this speculation, it is generally assumed
that we are easily capable of recovering the referent of this NP ‘this
mangy old beast.’
(26) The architect of this church was mad. (Jaszczolt 2002: 133)
4. R-dislocations in Japanese
In (27) one sentence is divided into five segments for the conve-
nience of analysis; this is an interview in a TV programme in Japan.
In the immediately preceding utterance, the speaker mentioned that
he had been struck and kicked. So it is argued in Fujii (1995) that the
proposition ‘X of the body was cut’ is given prominence in this dia-
logue with the result that the information concerning which part of
the body, that is the information in segment 5, is R-dislocated to the
sentence end. However, it has to be remembered that the deictic ex-
pression ‘kore’ [this] is used in segment 1. It refers to the sword cut
on the speaker’s face; it is the most accessible in this context. The
hearer is easily and exactly capable of recovering its referent from
the information in the context. We can assume that the segments
from 1 to 4 provide the descriptive proposition about the sword cut
on the face, and as the final process of completing a sentence the
speaker shows the reference of the deictic expression, that is ‘the
sword cut on this face not on other faces.’
Next let us explain (28) which includes the pragmatic
markedness of ‘contrast’ and ‘negation.’
(28) 1. kodomo mo
children also
2. kika nai n desu yo ne,
listen NEG COM COP FP FP
3. naze ka
why Q
4. souiukoto.
that kind of thing (Fujii 1995: 184)
has a general tendency to use zero pronoun when the referent of some-
thing is activated the most in the context. So we restate our earlier
conclusion that the hearer is easily and exactly capable of recovering
the referent of zero pronoun from the information of the context, and
that the segments from 1 to 2 provide the descriptive proposition
about a chat over the old days. And the speaker shows the reference
of zero pronoun, that is ‘a chat over the old days not over anything
else’ as the final process of completing a sentence
In order for our speculation to be confirmed, we will analyze
one more dialogue as follows.
Zero pronoun is used in this dialogue, too. The hearer recovers the
referent of the zero pronoun from all sources of information in the
context. Segments 1 and 2 provide the description about the referent
the hearer recovered from the context, and after that description the
speaker provides the reference of the zero pronoun.
We have examined only three examples of R-dislocation in
Japanese: the affinity between them is that all of them include affec-
tive elements with themselves, for instance ‘zubatto’ [= completely]
in (27), ‘nai’ [= negation marker] in (28), and ‘sugoi’ [very] in (29).
Just as English equivalents do, Japanese R-dislocations convey an
affective meaning which is not so strong as to break a hearer’s ex-
pectation and surprise a hearer. Moreover, R-dislocated elements
exhibit the characteristic of showing the referent of the most acti-
vated elements in the main clause.
84 Norifumi Ito
5. Conclusion
References
HIDEMATSU MIURA
1. Introduction
On the other hand, there are RCs that do not allow such non-reflex-
ive interpretation. Suppose there is a statue of Reagan in a wax mu-
seum. Then compare (2a) and (2b). (2a) never allows such a statue
reading whereas (2b) allows the statue reading as well as the normal
reflexive event reading (Lidz 2000).
Lidz (2000, 2001) claims that there are two types of reflexives: pure-
reflexive and near-reflexive. The former requires the reflexive to be
referentially identical to the antecedent whereas the latter does not. He
claims that the difference comes from lexical reflexivity of the verbs.
90 Hidematsu Miura
That is, the verb of (2a) is lexically reflexive while that of (2b) is not.
He concludes that there is a bidirectional relation between lexical
reflexivity and pure-reflexive interpretation. He calls this principle
‘Condition R’. To my knowledge, Lidz’ theory has not been applied
to Japanese. The relevant Japanese data were examined, but it turned
out that some of the Japanese data do not follow Condition R.
The purpose of this chapter is two-fold. First, it will be dem-
onstrated that Japanese lexical RCs show the behaviors expected from
Condition R. This serves as further confirmation of the universality
of the principle. Second, it will be shown, however, that the behav-
iors of syntactic RCs do not follow from it. It will be argued, follow-
ing Liu (2003), that focus structure, as well as lexical reflexivity, can
also yield the (unambiguous) pure-reflexive interpretation. It will be
claimed that, accordingly, the current bidirectional relation of the
lexicon-semantic correlation in Condition R should be weakened to
unidirectional one.
In Section 2, the two types of reflexives and the principle, Con-
dition R, will be further detailed. In Section 3, the Japanese reflex-
ives will be examined in terms of the principle and it will be shown
that there are unexpected behaviors that do not follow from the prin-
ciple. In Section 4, a structure-based account will be given. In Sec-
tion 4, concluding remarks will be made.
Lidz (1999, 2001) took the Jackendoff’s insight seriously and claimed
that there are referentially two different types of reflexives: those
that require a complete identity with the antecedent and those that do
not. He terms the former ‘pure-reflexive’ and the latter ‘near-reflex-
ive’1. Even when the referent of the reflexive is different from that of
the antecedent, the entity that the near-reflexive refers to must have
some identifiable association with the antecedent (e.g. statue or por-
trait). The difference between pure- and near-reflexives can be rep-
resented as follows:
Thus, incorporating both Jackendoff ’s and R & R’s insights, Lidz pro-
poses his own theory of reflexivity in which semantic reflexivity and
lexical reflexivity have bidirectional relation to each other. The fol-
lowing is the principle he calls Condition R:
The condition on the left is the semantic representation and the one
on the right is the lexical specification. This states that if a predicate
is semantically reflexive (i.e. pure-reflexive), it is lexically reflexive
and vice versa. In other words, what Condition R states is that true
(pure) reflexivity is guaranteed through lexical reflexivity. As a con-
sequence, when this condition is not met, near-reflexive readings
such as statue reading are expected to be available.
Condition R is a semantic condition which does not refer to
any morphosyntactic realization of lexical reflexivity. There are two
realization patterns attested: a morphologically covert type and an
overt type. The Dutch example (6) is an example in which lexical
reflexivity is sublexically expressed (i.e. morphologically covert).
There are languages that have morphologically overt-marking on the
predicate (see Kannada examples below for this morphologically
overt type).
There are two predictions Condition R can make. First, if the
predicate is lexically reflexive, whether it is realized sublexically or
morphologically, only the pure-reflexive interpretation should be avail-
able. This expectation is borne out as in (2), repeated below as (8).
The assumption here is that the lexical entries of the verbs used in
(8a) and (8b) are lexically reflexive and non-reflexive respectively
(the same logic for the Dutch examples 5 and 6 above). As expected,
the sentence with a lexically reflexive predicate, (8a), does not allow
any near-reflexive (i.e. statue) reading while the one with a lexically
non-reflexive predicate, (8b), allows such reading.
Focus-driven Semantic Reflexivity in Japanese 93
2 Lidz (1999, 2001) calls the construction comparative ‘deletion’, but I will call
it comparative ‘ellipsis’ construction following the distinction made in Hoji
(1998). Both ‘deletion’ and ‘ellipsis’ will be used as diagnostic tests below.
The distinction will be further made clear below.
3 Readers should be aware of the circularity in the arguments. Although I do not
pursue this problem any further here, a comment is in order. In order for Con-
dition R to work, we need to know somehow if a predicate is lexically reflexive
or not independently of the two phenomena (the availability of the statue read-
ing and the sloppy reading). When a language has a reflexive morpheme at-
tached on a predicate stem (e.g. Kannada; see below), it is obvious that the
predicate is (lexically) reflexive. But in the case of languages without such a
morpheme (e.g. Dutch), it is hard to see. R&R (1993) suggests a ‘nominalization
test’ for the lexical reflexivity of the Dutch predicates, but the efficacy of the
test is far from convincing. If we use the availability of the statue reading and
the sloppiness as the diagnostic tests for lexical reflexivity, we are led to wrong
conclusions since, as will be discussed later, Chinese and Japanese produce the
unambiguous pure-reflexive interpretation without a lexically reflexive predi-
cate.
94 Hidematsu Miura
(10b) and (10c) show that, without this reflexive affix, the predicate
is non-reflexive and the simple reflexive pronoun does not suffice to
form an acceptable sentence. The predicate in (10c), which is same
as (10b), is not lexically reflexive so that the near-reflexive interpre-
tation is available.
4 Haiman (1983) defines the actions which one generally performs upon one’s
self as ‘introverted’ and the actions which the subject usually performs toward
others as ‘extroverted.’ The following pair represents each type respectively.
Although both of the verbs are transitive, the possibility of the omission of the
reflexive pronoun is an indicator of the distinction.
i. Max washed (himself).
ii. Max kicked himself.
96 Hidematsu Miura
Although the data are somewhat limited, Lidz (2000) has concluded that
Malayalam is the language that does not have any lexical reflexivity.
The following basic assumptions of this chapter were laid out
in this section: the distinction between pure- and near-reflexives,
Condition R, the notion of lexical reflexivity and the cross-linguistic
variation of lexical reflexivity.
3. Reflexives in Japanese
On the other hand, the inalienable type does take an object-like ele-
ment as in (17a). The element is semantically bound by the reflexive
morpheme whose original antecedent is the subject. An entity that
has no semantic/associative relation to the binder (subject) cannot
appear. As in (17b), for example, the crime must be the one commit-
ted by the subject, Taroo, or the one Taroo is somehow involved in.
6 The distinction is not relevant to the current discussion, so it will not be de-
tailed here. See Tsujimura/Aikawa (1999) for their observations for the dis-
tinction.
Focus-driven Semantic Reflexivity in Japanese 99
ing is not possible as in (19a) and (20a). The examples in (19b) and
(20b) further demonstrate that this locality is seen irrespective of the
juncture type. (19a) and (20a) display a clausal juncture while (19b)
and (20b) show a core juncture7.
7 The distinction between clause and core is based on Role and Reference Gram-
mar (RRG; Van Valin and LaPolla 1997, Van Valin 2005). A primary indicator
for the distinction between clause and core is whether the juncture is tensed or
not. In (19a) and (20a), past tense is used so that the juncture is clearly clausal
whereas in (19b) and (20b) past tense cannot be used indicating the juncture is
not clausal (i.e. core).
8 I do not pursue this issue here, but this difference may be indicating a funda-
mental difference in self-forms between the two languages (Dutch and Kannada)
and Japanese.
Focus-driven Semantic Reflexivity in Japanese 101
For (24) suppose that Taroo and Hanako committed different crimes
individually and they know about each other’s crimes. That is, (24)
means that Taroo was less tenacious than Hanako in the police sta-
tion. The crime Taroo confessed must be his and the one Hanako
confessed must be hers. There is no interpretive possibility that
Hanako confessed Taroo’s crime. Thus, the strict interpretation, (24b),
is unavailable. For (25) imagine the situation like Taroo and Hanako,
living in the same area, bought the same item individually from the
same seller and somehow Taroo had to pay more for the carriage due
to the seller’s miscalculation. Again only the sloppy reading is avail-
able10.
Next the comparative deletion construction will be examined.
Hoji (1998: 135) demonstrates that the Japanese comparative dele-
tion construction does not allow the sloppy reading in general and
the strict reading is preferred over the sloppy reading when both are
available. Given the semantic reflexive nature of lexical reflexives,
it is expected that lexical reflexives should produce the sloppy read-
ing despite the general preference for the strict reading. This expec-
tation is borne out. Imagine the same contexts as above for compara-
tive ellipsis construction.
10 It is possible for both (24) and (25) to have the split antecedent interpretation in
which both of them committed the same crime together in (24) or bought the
same item together in (25); however, that is not the reading intended to be
examined here.
104 Hidematsu Miura
The null object construction (NOC) can also be used to show the
reflexivity of zi(ko)-verbs11. (28) is an example of the construction.
As shown in (28), both sloppy and strict readings are available in
some cases (Otani/Whitman 1991).
(28) John wa [zibun no tegami] o suteta.
TOP self GEN letter ACC discarded
‘John discarded self’s letter.’
Mary mo [e] suteta.
also discarded
a. ‘Mary/i also threw out self’s/i letters.’
b. ‘Mary also threw out John’s letters.’
But more generally, the strict reading tends to be preferred over the
sloppy reading just like the comparative deletion construction. Hoji
(1997) points out that in many cases, the NOC allows the strict read-
ing only as in (29b).
Thus NOC provide another robust diagnostic test to show the bound
variable nature of the lexical reflexive construction. (30) and (31)
show that the strict reading is blocked for zi(ko)-verbs, as expected.
11 As Hoji (1998) says, comparative deletion and NOC are expected to show the
same behavior since comparative deletion is a construction which ‘embeds’
NOC in the subordinate (comparative) juncture. In other words, the difference
between comparative deletion and NOC is that the null element of the former
is in intra-sentential position while that of the latter in inter-sentential position.
However, there is a difference also. NOC has a focus particle (mo ‘also’) in the
second sentence which comparative deletion does not have.
Focus-driven Semantic Reflexivity in Japanese 105
12 As in (i) below, the possessor can be specified by the reflexive pronoun, but it
is generally unexpressed and is pragmatically controlled.
(i) Taroo ga zibun(-zisin) no hige o sotta
NOM self(-self) GEN beard ACC shaved
‘Taroo shaved his own beard’
Focus-driven Semantic Reflexivity in Japanese 107
The following, (33), is another example that shows the relevant point.
The verb, kiru ‘cut’, is generally not regarded as an introverted verb.
Given the above observation that the verbs require the same type of
noun as object irrespective of the introvertedness or the transitivity/
reflexivity of the construction, it seems reasonable to assume that
the verbs that do not have any morphological reflexive-marking have
13 The verbs that denote physical activities, especially those that involve change
of state, show this peculiarity in a very consistent manner. See Miura (2002,
2005) for more details on this.
108 Hidematsu Miura
only one lexical entry of the same type (irrespective of the intro-
vertedness or the transitivity/reflexivity). When there is only one lexi-
cal entry, that must be transitive (i.e. lexically non-reflexive) because
it is conceivable to derive reflexives from transitives, but not vice
versa (cf. Sells et al. 1987).
Now we are ready to turn to our primary concern, the behav-
iors of Japanese syntactic reflexives. What we can predict based on
the entire discussion so far is that they should produce both statue
reading and sloppy/strict ambiguity. In the next section, however, it
will be shown this is not borne out.
Suppose the same context as the Ringo sentence. The Prime Minister
Koizumi is wandering in a wax museum and finds his statue. Even
though the reflexive in (36a) could be used to refer to the statue of
the prime minister, the acceptability would be highly marginal and
there is some speaker variation on the judgment. The near-reflexive
(statue) interpretation is just not available for the complex form, (36b).
Generally, the Japanese contact verbs such as sawaru ‘touch’, which
do not denote change of state, can take a reflexive pronoun as object
so that the unacceptability cannot be ascribed to the selectional re-
striction discussed in the previous section (although I admit this is
still an arguable point).
There are two unexpected behaviors here. First, given that
Condition R guarantees the ‘unambiguous’ pure-reflexive reading
only through lexical reflexivity, it is unexpected that only the pure-
reflexive interpretation is ‘unambiguously’ available in (36) despite
the absence of lexically reflexive predicate. Second, the difference
in the acceptability between the two reflexive forms is unexpected
since Condition R does not make any reference to morphology, namely
the morphological complexity of reflexive forms.
More compelling data can be given in the comparative con-
structions. Condition R expects that both strict and sloppy interpre-
tations should be available for the comparative ellipsis construction
because the predicate is not lexically reflexive; however, this expec-
tation is not borne out. Here again, only the sloppy (i.e. pure-reflex-
ive) interpretation is ‘unambiguously’ produced without lexically
reflexive predicates as in (37a) (cf. Sells et al. 1987). This is contrary
to what Condition R predicts. (37b) further shows the sloppy reading
is obtained irrespective of the complexity of reflexive forms. Thus
Condition R clearly fails to account for the behaviors of Japanese
syntactic reflexives in this case.
110 Hidematsu Miura
4. Focus
In the remainder of the chapter, I argue that two types of focus struc-
tures are involved in the phenomena: focus by intensifier and focus
by construction. They will be discussed in turn .
14 Their data have been included in WALS (the World Atlas of Language Struc-
tures; Haspelmath et al. 2005) as “(#47) Intensifiers and Reflexive Pronouns”.
Among 168 languages they investigated, 94 languages use identical forms for
both functions and 74 languages differentiated ones. They also have a web site
and all the information about the languages they investigated is available there.
15 Adnominal and adverbial intensifiers are quite different both syntactically and
semantically. I will limit the discussion to the adnominal type and so in what
follows the term ‘intensifier’ means ‘adnominal intensifier’ unless mentioned
otherwise. König and Siemund’s (2005) discussion is based on the cross-lin-
guistic data (around 100 languages), but for the sake of simplicity I will use
English data unless the English data miss the points of their arguments.
112 Hidematsu Miura
(43) [[ the president himself ]] = ID([[ the president ]]) = [[ the president ]]
The semantic operation above looks trivial, but it is not so. Jackendoff
(1992, 1997) observes that in the Ringo sentence, for example, the
subject Ringo must be the very individual and it must not be some
other entity associated with the person (e.g. portrait or statue). Com-
pare (44a) and (44b). This constraint suggests that the ID function is
a necessary operation independently of focus structure, although it is
beyond the scope of this chapter to explore how the ID function works
in the domains other than focus structure.
The default prosodic pattern also supports the distinction. (48) shows
that the same patterns as (42) are observed also in Japanese. Focus
particles (sae ‘even’ here) generally do not receive prosodic peak
whereas intensifiers generally do.
Before jumping onto the issue of the simple zibun, I would like to intro-
duce Chinese data, argued by Liu (2003), that are highly similar to the
Japanese case we are dealing with. Liu claims that the Chinese reflex-
ive, ziji-benshen, produces only the pure-reflexive interpretation un-
ambiguously in the absence of a lexically reflexive predicate. (51) and
(52) are examples that show the (un)availability of the statue reading
and the sloppy reading in the comparative construction respectively.
Although the Japanese data (36 and 37) and Chinese data (50 and 51)
look alike, there is a crucial difference between them. The difference
118 Hidematsu Miura
is that the Chinese simple reflexive, ziji, produces pure- and near-
reflexive ambiguity unless used with benshen. On the other hand, as
was shown in (36) and (37), the Japanese simple reflexive, zibun,
shows, without zisin, the pure-reflexive interpretation such as the
unavailability of statue reading or producing the sloppy reading in
elliptical constructions. It is reasonable to assume that Japanese has
another system which imposes the identification (ID) function that
Chinese does not have. I argue that the system is a language-specific,
construction-based focus structure.
It has been pointed out in the literature that preverbal position
is the default focus position for SOV languages (e.g. Kim 1988,
Krifka1998). Japanese is one of such SOV languages. The examples
in (53), from Ishihara (2000), show the canonical SOV word order in
Japanese. The focus is indicated by capitals.
the focal NP). In sum, this observation means that the object of the
canonical SOV sentences is always in focus18.
Kiss (1998) argues, discussing Hungarian data, that two focus
structures, identificational focus and information focus, must be dis-
tinguished. The former expresses exhaustive identification and the
latter new information. It is claimed that the focus at preverbal posi-
tion in Hungarian, one of SOV languages, is identificational focus.
(55) is an example.
18 The constructional focus is, however, not so strong as to always decide the
focus structure of the sentence. As noted in Ishihara, the focus can be easily
overridden by prosodic focus. The sentences in (i) below are a possible prosodic
variation of (56a). Virtually, any element can be assigned a stress.
(i) a. Taro ga kyoo HON O katta
b. Taro ga KYOO hon o katta
c. TARO GA kyoo hon o katta
Taro NOM today book ACC bought
‘Taro bought a book today’
120 Hidematsu Miura
19 Sells et al. (1987) also note the same point. Some of their Japanese informants
judged (56) as acceptable.
20 Scrambling does not seem to change the (un)availability of the statue reading.
There seem to be several options to solve this. One is to assume that the
scrambled element is assigned another focus which outweighs the structural
preverbal focus. The other is to assume that zibun has an inherent [+f(ocus)]
feature (cf. Kiss 1998). I leave this issue open.
(i) *?zibun ni Koizumi shushoo ga sawatta
self DAT PM NOM touched
‘The prime minister Koizumi touched himself ’
Focus-driven Semantic Reflexivity in Japanese 121
(i.e. focus assigner), zisin, is scrambled out together with the reflex-
ive. This expectation is also borne out as in (57)21. This observation
suggests that the focus structure based analysis is on the right track.
5. Concluding remarks
Acknowledgement
References
1. Introduction1
3 Cantonese belongs to the Yue dialect group. According to Matthews and Yip,
“Yue dialects are spoken primarily in the southern Chinese provinces of
Guangdong and Guangxi, and in the neighbouring territories of Hong Kong
and Macau” (1994: 2). In this chapter, the term ‘Cantonese’ refers to the vari-
ety of Cantonese that is spoken in Hong Kong.
4 JyutPing (literally ‘Cantonese romanization’, proposed by the Linguistic Soci-
ety of Hong Kong (LSHK) in 1997), is adopted throughout the chapter. (For
details of this romanization scheme, please refer to: <http://cpct92.cityu.edu.hk/
Jyutping/Jyutpin.htm>).
From nonsubjective to subjective to intersubjective 127
1971; Lyons 1982; Langacker 1985, 1990; Traugott 1995), one can-
not possibly say or write anything without expressing some aspect of
personal beliefs or point of view. Natural languages should fulfill the
fundamental need of speakers to convey and assess their feelings,
moods and attitudes. Subjectivity is deeply embedded in our lan-
guage, it is everywhere! Benveniste even raised the question whether
a language could still be called a language without subjectivity as
languages are marked so deeply by the expression of subjectivity. He
remarked that ‘a language without the expression of a person cannot
be imagined’ (1971: 226). It seems obvious that speakers in a dis-
course must take a perspective and encode their point of view to-
wards what they express; this inevitably involves subjectivity.
While subjectivity is pervasive in language use, not all ele-
ments in natural languages serve to express it explicitly. In fact, only
a subset of elements or constructions is used to make subjectivity
explicit. Most of them come to do so through the process of subjecti-
fication, which refers to ‘the structures and strategies that languages
evolve in the linguistic realization of subjectivity or to the relevant
processes of linguistic evolution themselves’ (Finegan 1995: 1).
Subjectification then is a process which involves “speakers recruiting
forms with appropriate meanings to externalize their subjective point
of view” (Traugott 1999: 189).
Traugott (1989) has identified three general tendencies in se-
mantic change. One of these tendencies suggests that meanings tend
to become increasingly situated in the speaker’s subjective belief
state/attitude towards the proposition. It is considered a strong ten-
dency in semantic change and is supported by Traugott’s subsequent
works (Traugott 1995, 1997, 1999, 2003) as well as other studies on
subjectification and grammaticalization (e.g. König 1991; Carey
1995; Brinton 1996; Suzuki 1998).
By treating subjectification as a pragmatic-semantic process
whereby meanings become increasingly based in speakers’ beliefs
about, or attitudes towards, what they are discussing, Traugott (1995)
illustrates how certain expressions that initially articulate concrete,
lexical, and objective meanings have come to serve abstract, prag-
matic, interpersonal, speaker-based functions. She identifies eight
examples of subjectification in English grammaticalization, suggest-
From nonsubjective to subjective to intersubjective 129
(5) Let’s take our pills now, Roger. (to a child or a person in a retirement home)
(6) 彼⯳ẩġ
faan uk-kei
faan home
‘return home’
(7) 崘彼⬠㟉ġ
zau faan hok-haau
run faan school
‘run back to school’ [spatial return]
(8) ⴤ१ԫʕࠃʳ
nam faan jat-di wong-si
think faan some past-events
‘think back over some past events’ [cognitive return (to past)]
(9) ش१ʕء
jung faan go bun syu
use faan that CL book
‘go back to using that book’ [situational return ]
(10) ᝑ१إᇩଡᠲؾ
gong faan zing-waa go tai-muk aa
talk faan previous CL topic PRT
‘return to the previous topic’ [discourse return]
(11) ┊װڣݺ१נᚭΜ
ngo dei heoi tai faan ceot hei laa!
I PL go see faan CL movie PRT
‘(Let’s) go and see a movie!’
Here, the speaker is not suggesting that they should ‘return’ to the
seeing of movie. He is making a suggestion that he thinks it would be
134 Winnie Chor
(12) ݺუᝑ१ࠟΖ
ngo soeng gong faan loeng geoi
I wish say faan two sentence
‘I wish to say a few words (on something good).’ (Gao 1980)
(13) ᐊ१ේනΖ
se faan di je
write faan some thing
‘(I wish) to write something (which is worth writing on).’ (Yuan 1989)
(14) ݺუ၇१ԫଡΖ
ngo soeng maai faan jat go
I wish buy faan one CL
‘I wish to buy one (which is worth buying).’ (Matthews/Yip 1994)
(15) ቮ१Ǝ़ᓳΖ
taan faan haa hung-tiu
enjoy faan while air-conditioning
‘(I wish) to enjoy the air-conditioning for a while.’ (Peng 1999)
From nonsubjective to subjective to intersubjective 135
(16) ݺუ堬१ࣦ࠼Ζ
ngo soeng jam faan bui gaa-fe
I wish drink faan CL coffee
‘I wish to have a (nice) cup of coffee.’ (Tang 2001)
(17) *່ኰএ堬१ࣦ࠼Ζ
zeoi caam hai jam faan bui gaa-fe.
most poor BE drink faan CL coffee
‘The worst thing to do is to have a cup of coffee.’
(18) ൿ༏چΕࢳೋ࿗Δक़ෙࠟଡᤪΜ
sou fuk dei mut zek coeng faa zo loeng go zung
sweep CL floor clean CL window spend PERF two CL hour
‘Swept the floor; cleaned the window – it took (me) two hours!’
(19) ൿ१༏چΕࢳ१ೋ࿗Δ
sou faan
fuk dei mut faan zek coeng
sweep faan
CL floor clean faan CL window
क़ෙࠟଡᤪΜ
faa zo loeng go zung
spend PERF two CL hour
‘Swept the floor; cleaned the window – it (pleasantly) took (me) two hours!’
Without faan, activities like sweeping the floor or cleaning the win-
dow are not necessarily enjoyable. With the use of faan, these other-
wise dull and annoying activities become pleasurable and worth
doing. One can add remarks like ‘but I enjoyed it!’ or ‘it’s very pleas-
urable!’ to (19), but not (18). Faan here can be considered as having
an expressive function, to express the affective and emotional feeling
of the speaker. As for (18), on the other hand, one might expect com-
ments like ‘it’s very tiring!’ or ‘that’s backbreaking!’. Faan in (19) is
136 Winnie Chor
actually expressing some kind of ‘mood’ (yu ï qì). As one can see from
these examples and many others, the original directional meaning of
faan (i.e. ‘go back/return’) is not there anymore. Faan involves a shift
of meaning from objective (sense of direction) to subjective (the ap-
praisal of ‘being good’). It is recruited by speakers of Cantonese to
‘externalize their subjective point of view’ (Traugott 1999: 189), in
this case the speaker’s assessment of ‘being good’.
As Finegan remarks, “in many other languages, including En-
glish, subjectivity is marked in ways sufficiently subtle to be easily
overlooked, and sufficiently complex to prove challenging to expli-
cate” (1995: 3). How can we get to this sense of goodness from the
sense ‘go back’? Why is the shift of perspectives from impersonal to
speaker-oriented possible? These are the interesting issues which need
to be addressed.
In the first instance, how can the semantics of ‘pleasurable’ be
developed from the seemingly unrelated sense of ‘direction’6? An
explanation for this development can be found by looking again at
the core semantics of faan as a verb: back to the source. This ‘source’
is often interpreted as the original point from which the movement
started, as in:
(20) ࡵݺ୮१ᖂீ
ngo ji-gaa faan hok-haau
I now faan school
‘I now go (back) to school.’
Sentence (20) implies that some time in the past the speaker has been
to the school and now he or she is going back to that school. Interest-
ingly, in some situations, faan can be used with a place where the
speaker has never been to:
(21) १խഏ
faan zung-gwok
faan China
‘go (back) to China’
A speaker who has never been to China can still utter (21), but with
the condition that China is his ancestral homeland. If (21) is uttered
by a non-Chinese, faan can only mean a physical return: ‘China’ is a
place which the person left, and is now coming back to. This obser-
vation can be accounted for if we consider again the meaning of
faan: back to source. Metaphorically, this ‘source’ is extended to one’s
native place and can be interpreted as one’s ‘source of origin’ – the
place where one comes from or belongs to. If the speaker of (21) is a
Chinese, then China is considered his ‘source of origin’. Even if he
has never been to China, he can still use faan as he is actually ‘back
to the source’, back to where he belongs to and originates from! Fol-
lowing this direction, we can see why examples (22-24) below are
possible:
(22) ֲᦫݺรԫֲ१ՠΖ
ngo ting-jat dai-jat jat faan gung.
I tomorrow first day faan work
‘Tomorrow is my first day of work.’
(23) ݺଡՖᦫֲၲࡨ१ؔᆇႼΖ
ngo go neoi ting-jat hoi-ci faan jau-zi-jyun.
I POSS daughter tomorrow start faan kindergarten
‘My daughter will start going to the kindergarten tomorrow.’
(24) ݦݺඨڶݺԫֲאױ१֚ഘΖ
ngo hei-mong ngo jau jat jat ho-ji faan tin-tong.
I hope I have one day can faan heaven
‘I hope I can return to heaven one day.’
(25) אݺছ१چᒌΔᦫֲၲࡨ१ֆႼΖ
ngo ji-cin faan dei-pun ting-jat hoi-ci faan gung-jyun
I past faan building-site tomorrowstart faan garden
‘I used to work in a building-site, will start working in a garden tomorrow.’
(26) ᝑ१խഏᖵ…
gong faan zung-gwok lik-si aa
talk faan China history PRT
‘(Let’s) talk about Chinese history…’
By uttering (26), the host is trying to bring the guest back to the topic
of Chinese history, which may or may not have been mentioned or
talked about earlier during their conversation. In cases where the
topic has not been mentioned previously, the use of faan provides an
opportunity for the hearer to make an inference: the speaker is actu-
ally making a suggestion to introduce a topic in a polite way, as if we
are just returning to a previously discussed topic but not really chang-
ing the topic. Faan is used as a conversational strategy – whatever
topic appearing after it is the one which the speaker intends to talk
about. This strategy is also used frequently among friends. If a friend
has been talking about a single topic for a really long time, the other
can probably change the topic in a polite way by making use of the
structure ‘gong faan XX…’ (X = your topic in mind). In these occa-
sions, the verbal particle faan is used as a tone softener for the hearer’s
own comfort, making the utterance sound more friendly and polite.
This politeness strategy is essentially directed to the hearer (hearer-
oriented).
In fact, this intersubjective faan is also used in many other
occasions. Here are some examples which I have come across:
7. Conclusion
References
1. Introduction
1 There are two kinds of na. One is basically a monologic particle and used by
both male and female speakers. The other has the same function as ne, which is
a dialogic particle, and is used only by male speakers (See Miyazaki (2002)).
The target of our investigation is the former.
150 Yutaka Shinoda
2. Previous studies
has the same knowledge as she has.2 Thus, the speaker of (3) as-
sumes that the hearer also thinks that it is fine today, and she seeks
for his agreement. Similarly, the speaker of (4) assumes that the hearer
is Mr. Yamada, and she seeks confirmation from him.
2 Throughout this chapter, I will refer to the speaker with he, him and his, and the
hearer, she and her, unless otherwise stated.
152 Yutaka Shinoda
Note that the shared knowledge theory does not explain this example.
Here the speaker (= B) cannot assume that the hearer (= A) has the
same knowledge as she, that is, the knowledge that it is 3 o’clock
now. It is because the hearer has no such knowledge that he asks the
question.
The internal verification theory would argue that the speaker
of (5) is in the verification process of the proposition in question, just
as in (4). The difference is that in example (4), the speaker needs the
hearer’s response to complete her verification procedure, while in
(5), she verifies her assumption by looking at her own wrist watch.
Takubo and Kinsui (1997, 2000) calls the latter type of confirmation
‘self-confirmation’.
Compare (5) with the following:
Kamio (2002) reports that many informants feel that the answer
250 en desu ne would be unnatural if the speaker uttered it in an
Japanese Particle Na 153
instant and definite way. The speaker of (Bb) knows in advance that
the meat costs 250 yen, and this information is already activated on
her mind. Consequently, she needs no mental computation for veri-
fication.
Although the internal verification theory is very persuasive con-
cerning such examples as (5)-(7), it does not seem to account for
more common examples like (3) above. When we say Ii tenki desu
ne (‘It’s a fine day, isn’t it?’) in a normal situation, it should be obvi-
ous to us that it is fine. We can see the clear sky with our own eyes at
the very moment of the utterance. The proposition is so obvious that
it does not need to be verified. Then why does (3) have ne? (3) would
be less natural without ne.
Takubo and Kinsui (2000) claim that even (3) involves a trivial
mental computation. But how trivial is that? I think that it is as trivial
as a computation needed when we identify our own names. The in-
ternal verification theory cannot differentiate (3) from (6). For the
speaker, both her own name and the weather when and where the
utterance is made are too manifest. It is not necessary for her to verify
them in any sense.
Takubo and Kinsui (2000) state that ne indicates a verification
procedure for some judgment. I suggest that ne does not mark a veri-
fication of a judgment but a judgment itself. (Probably the function
of ne extends way beyond that, however. See Section 4.) Suffice it to
say here that their theory cannot deal with such a typical example as
(3) satisfactorily.
As for na, Takubo and Kinsui (2000) mention it only briefly;
they assume that it has almost the same meaning as ne, except that
na can be used in monologic discourse. This means that na also marks
a verification process.
Now let us move to review some of the previous studies on na.
One of the very important studies is Miyazaki (2002). As we
have seen in Section 1, na is often referred to as an exclamatory
particle. For instance, the speaker of (8) is impressed by the fine
weather. The sentence form itself is not an exclamatory sentence, but
its meaning is close to ‘How fine the weather is today!’.
154 Yutaka Shinoda
In this example, the speaker simply shows the hearer what he has
just felt. He does not necessarily seek for the hearer’s agreement or
Japanese Particle Na 155
3 The iconic principle of quantity, which states that the more form, the more
meaning, accounts for this distinction. The more prolonged pronunciation the
particle has, the more emotion it involves. That is, naa, compared to na, con-
veys a stronger emotion.
156 Yutaka Shinoda
This utterance is less natural than the same one without na. NKK
(2003) states that na cannot be used in a situation where the speaker
makes a discovery without a prior recognition. (11) might be accept-
able in the situation that the speaker has expected a wallet to be on
the ground and confirms the expectation. (See subsection 3.3 for fur-
ther discussion.)
Returning to Miyazaki’s (2002) analysis, it would predict that
(11) is natural because this example, if it were acceptable, would be
an expression of what the speaker has recognized on the spot. As
I said before, therefore, Miyazaki’s definition of na as signaling the
speaker’s ‘recognition’ on the spot, is too broad. It cannot rule out
unacceptable sentences like (11).
Lastly, I would like to mention Suzuki (1998). He also makes
an important observation. He states that na basically expresses the
speaker’s subjectivity, and that consequently it cannot be used to
state as a fact what is self-evident, the speaker’s experience, and so
on. In my opinion, this is the reason why (11) is odd. (11) expresses
a mere fact which the speaker has just discovered. This observation
by Suzuki that na marks the speaker’s subjectivity is essential for the
discussion later.
Suzuki (1998) goes on to say that na is used not to express
facts or the speaker’s experiences just as they are, but to state the
speaker’s feelings about them. However, I doubt that “to state the
speaker’s feelings” is the right characterization. Recall that ne might
be used when the speaker takes some time to answer a question (see
(5) and (7)). It seems that na could be used in a similar but monologic
discourse:
Japanese Particle Na 157
3. The function of na
Miyazaki’s (2002) example (9) belongs to the same type. Let me call
this type the ‘recognition’ type, for lack of a better term. I argue that
this type is the most fundamental. See subsection 3.2.1 for discussion.
A second type is the ‘exclamation’ type. A typical example
is (8), which is cited again below as (14):
158 Yutaka Shinoda
(15) (When the speaker makes sure that the windows are locked before going out)
Yoshi, shimatteru na.
OK is locked SFP.
‘OK, (the windows) are locked.’
In this type of examples, the speaker has some prior expectation. The
speaker of (15) has expected that the windows are locked, and is
confirming it. The sentence would sound bizarre if it were uttered
out of the blue, without any prior expectation.
An example of the last type is (12), which we observed at the
end of the last section. Let me call this type the ‘self-confirmation’
type, following Takubo and Kinsui (1997, 2000), to whose theory
I do not necessarily commit myself, however. This kind of na could
be regarded as a marker which indicates that the speaker has found
the answer to a question or a problem. (Actually, this type may be
dissolved into other types. See Section 3.2.1)
Now that we have reviewed the classification of na, we will
consider what is the feature common to all the four types.
(16a) may describe a fact, but (16b) can be interpreted only as judg-
mental. The latter means that ‘Because the light is on in Taro’s room,
I judge that he is in his room.’
Next, let us examine the ‘exclamation’ type. Consider (14)
again. The sentence may seem to describe a mere fact. Note, how-
ever, that the speaker makes an evaluation of the weather, and that
evaluation is a kind of judgment. The speaker judges whether the
weather is fine or not, and how fine it is. When evaluation provokes
a relatively strong emotion, it is expressed as exclamation.
Compare the following three sentences:
brightly white, or even too white. On this reading, the speaker of (17b)
judges that the wall is very white, and feels a rather strong emotion.
She might be surprised, impressed, or disappointed at the whiteness
of the wall.
In contrast, shiro-kuro [black and white] is not a gradable ad-
jective, so (17c) is interpreted as a mere fact, not an evaluation. Hence,
the utterance sounds odd. Na, which is a judgment marker, cannot
occur with sentences which describe facts.
Here some comments may be needed on the distinction be-
tween the ‘exclamation’ type and the ‘recognition’ type. As far as
evaluation is a kind of judgment, every ‘exclamation’ type of na-
sentence is of the ‘recognition’ type at the same time. The former
may be regarded as a special type of the latter. In this sense, the
‘recognition’ type is the most basic of all the usages of na; it is a
direct realization of the function of na. When the speaker’s judgment
accompanies a strong emotion, it is realized as exclamation. In fact,
contrary to the popular view that na expresses exclamation above
all, in actual conversations we find as many examples with the ‘rec-
ognition’ na as those with the exclamatory na, and there are a large
number of ambiguous (or rather vague) examples, too.
Incidentally, what exactly is the difference between na and naa?
(19) (When the speaker makes sure that the windows are locked before going out)
Yoshi, shimatteru na.
OK is locked SFP.
‘OK, (the windows) are locked.’
The speaker, who has expected that the windows are locked, is con-
firming the expectation, uttering this sentence. Notice that confirma-
tion is a kind of judgment, just as evaluation is. When we confirm
something, we judge whether the expectation has been met or real-
ized, or not.
Lastly, we will consider the ‘self-confirmation’ type. See the
following example (= (12)):
The speaker of (20) asks herself what time it is now, and needless to
say, does not know the answer in advance. She has to find the an-
swer. The second utterance shows that she has found it. How can it
be judgmental? I would claim that there are two cases where this
utterance becomes judgmental.
One case is as follows: when the speaker woke up, she thought
perhaps vaguely that it was approximately 5 o’clock then, and now
by looking at the clock she has judged that her guess is correct or at
least not incorrect. Na marks this judgment. In this case, therefore,
the utterance is actually of the ‘confirmation’ type.
The alternative possibility is that the utterance involves an im-
plicit judgment. That is to say, when the speaker knew the time, she
implicitly judged, for instance, that because it was already 5 o’clock,
she had better get up then, or that because it was only 5 o’clock, she
still could sleep for one more hour. In short, the utterance conveys a
judgment as an implicature, as it were, though it is not made in a con-
versation. In this second case, the utterance might be considered to be
a special kind of the ‘recognition’ na-sentence. (See Subsection 3.3)
162 Yutaka Shinoda
If the actual answer is far from her speculation (in the first
case above) or the speaker has no particular implicit judgment in
mind (in the second case), she will use ka rather than na, as in (21).
Although ka is usually a question marker pronounced with a rising
intonation, in (21) it is pronounced with a falling intonation and func-
tions as a marker which signals that the speaker has accepted a piece
of new information.
4 My intuition is not clear as to whether the same can be said on the correspond-
ing ‘self-confirmation’ type of ne.
Japanese Particle Na 163
Both (22a) and (23a) are acceptable. On the other hand, though (22b)
is perfect, (23b) is problematic. Why? The reason is that whereas na
is a marker of the speaker’s judgment, tenki yohoo dewa [according
to the weather forecast] shows that the judgment in question is at-
tributed to the weather forecast. This contradiction makes the utter-
ance unacceptable. It may be possible to say that na functions as a
kind of evidential.
These sentences show the same pattern as seen in (22) and (23). Why
is (25b) unnatural? It is because (24a) is a subjective judgment while
(25a) is an objective judgment in that everyone will acknowledge it
as true. The subjectivity of na and the objective nature of (25a) are
incompatible with each other. Hence, (25b), which is (25a) plus na,
is unacceptable.
As Nakau (1994) points out, the English sentence I think that Tom is
a spy is ambiguous because it offers two interpretations. In Japanese,
the two meanings are represented by distinct expressions, that is,
(26a) and (27a). (26a) has the instantaneous present meaning; it ex-
presses the speaker’s judgment at the utterance time. (27a) has the
continuous (or ‘unrestrictive’ in Leech’s (1987) term) present mean-
ing; it means that the speaker has had the idea for some period of time.
The meaning of the latter is not in harmony with the function of na as
a marker of the judgment ‘here and now’. Hence, (27b) is unnatural.5
5 (27a) and (27b) have no difference in cognitive meaning. Both express the
speaker’s judgment at the utterance time. The difference is that whereas (27a)
can be used in either spoken language or written language, (28b) is used only
in speech.
Japanese Particle Na 165
Lastly but not least importantly, na-sentences may not express judg-
ments explicitly but involve them implicitly. (Recall the discussion
in 3.2.1)
To illustrate, consider the next sentences:
Imagine that you are driving with no purpose. Suddenly you see sea
gulls flying. Then you are likely to say (35a) but not (35b). In this
situation, (35b) would be unnatural because neither the ‘exclama-
tion’ nor ‘recognition’ reading is easy to obtain. Moreover, since you
have no prior expectation about sea gulls, the ‘confirmation’ reading
is not possible, either.
However, this does not mean that the ‘exclamation’ and ‘rec-
ognition’ readings are impossible. They will become possible if the
fact that sea gulls are flying leads you to make another judgment. For
example, by seeing sea gulls, you may judge that you are approach-
ing the sea. In that case, you can say (35b) (the ‘recognition’ reading)
with the implicit meaning ‘I am approaching the sea’. Or suppose
that you happen to dislike noisy sea gulls. Then you might say (35b)
(the ‘exclamation’ reading), with the intended meaning ‘How noisy
sea gulls are! I hate them.’ Or if you are driving in the mountains,
you might say the sentence, meaning ‘How strange! There can’t be
sea gulls around here.’ (the ‘recognition’ and ‘exclamation’ readings).
There will be other possible candidates for the interpretation of (35b).
The above discussion shows that na-sentences may have impli-
cit judgments with them. In the literal sense they may describe mere
facts, not judgments, but they can be accompanied by implicit judg-
ments, which might be communicated as implicatures in conversations.
4. The function of ne
Here the speaker utters simply what he has thought, a judgment about
a glass of gimlet. He obviously directs his words to the hearer; nev-
ertheless, he does not necessarily seek the hearer’s response. Even if
he had used na instead of ne, there would have been little difference
in meaning. In examples like this, therefore, we may say that ne, as
well as na, expresses the speaker’s subjective judgment ‘here and
now’. And that can be regarded as the core function of ne.6
So far, so good. But in fact, there are various examples where
ne diverges from this core meaning.
To begin with, as Miyazaki (2002) goes on to argue, ne does
not show what he calls “the nature of the recognition being made on
the spot”, in its original form, when it is used in the ‘confirmation-
seeking’ sense (and perhaps in the ‘agreement-seeking’ sense, too).
It follows that in the case of ne, the speaker’s judgment is not always
made ‘here and now’.
Secondly and more importantly, in the ‘agreement-seeking’ use
and the ‘confirmation-seeking’ use, the speaker assumes that the
hearer shares the same judgment with her. For instance, consider
(37) (= (3)) again:
Here both the speaker and the hearer see the same sky. The speaker
sees the clear sky and judges that the weather is very fine. She knows
that the hearer sees the same sky, and assumes that he judges the
same. Ne indicates that the speaker assumes that the hearer’s subjec-
tive judgment is the same as her subjective judgment. To put it an-
other way, the speaker’s judgment is not merely subjective, but rather
intersubjective. Hence, ne is not a marker of the speaker’s subjective
judgment; it can express the speaker’s intersubjective judgment.
6 From this it follows that at its core ne has at least three types of usages, just like
na: the ‘recognition’, ‘exclamation’, and ‘confirmation’ types.
Japanese Particle Na 169
Here the speaker has found a wallet on the ground by chance and
talks to the hearer, assuming that he has also recognized the wallet at
the same time. The only difference between (33) and (38) is whether
the sentence-final particle used is na or ne. Nevertheless, it seems
that (38) is more natural than (33). If, as was discussed earlier, the
unacceptability of (33) can be attributed to the fact that the sentence
describes a mere fact whereas na is a judgment marker, then the only
explanation for the difference between these two sentences is that
unlike na, ne is not a judgment marker.
In addition, ne can be used with more kinds of sentence than
na. Na occurs with statements only. Since na does not presuppose
the hearer, naturally it is not used with utterances which require the
interaction with the hearer. On the other hand, ne can occur with
offers, requests, and even polite orders. But can we say that offers,
requests and orders are judgments? If not, it means again that what
ne indicates is not judgment. One might claim that judgments under-
lie these kinds of utterances. For instance, the request Please come
here presupposes the speaker’s judgment that the hearer should come.
But this is not the whole story.
Ne can be attached to not only sentences but phrases and words;
that is, it can be used sentence-internally:
(39) Ano ne, Taroo ga ne, issyoni iki tai tte.
filler SIP Taro Nom. SIP together go want Quote
‘Uh, Taro says he wants to go with us.’
7 In Shinoda (2005), I argued that since the speaker’s use of ne tends to invite the
hearer’s backchannels, there is a significant correlation between ne and
backchannels. I proposed that ne has interpersonal functions corresponding to
the functions of backchannels. Brunner (1979: 733) suggests that backchannels
give the speaker feedback on the following three levels:
a. At the first level, backchannels signal the auditor’s involvement and par-
ticipation in the interaction. They indicate that the auditor is attending to
what the speaker says and that a conversation, not a monologue, is occur-
ring.
b. At the second level, backchannels provide information about the auditor’s
level of understanding, allowing the speaker to adjust his or her communi-
cative endeavor so as to get the ideas across efficiently.
c. At the third level, backchannels can signal the auditor’s personal response
to what the speaker has just said. This might mean agreement or disagree-
ment, shock, amusement, scorn, or any number of other reactions.
Japanese Particle Na 171
5. Conclusion
The present study has been concerned with the function of the Japa-
nese sentence-final particle na. Na functions as a marker of the
speaker’s subjective judgment ‘here and now’. The speaker’s judg-
ments expressed by na-sentences may be explicit or implicit.
Na can be used in monolog and inner speech, unlike its closely
related particle ne, which presupposes the hearer. Although ne, being
hearer-directed, extends its function in various ways, it seems to in-
herit its core meaning from na. In that sense, explicating the function
of na will contribute to exploring the function of ne. Uncovering
how ne develops various functions from its core meaning will be left
to future research.
In the introduction, I stated that an analysis of na would re-
quire native speakers’ inner reflection. This does not mean, however,
that it is unimportant to observe actual conversations. Rather, the
present study should be supplemented with research from conversa-
tion analysis or corpus linguistics. Further studies will be necessary
to depict a comprehensive grammar of na.
Brunner (1979) says that “In general, if an action functions on a higher level it
also functions on the ones below it.”
In a similar way, I proposed, ne has the following three-level functions:
a. At the first level, ne signals that the speaker is attending to the existence of
the hearer, urging him to attend to what she says and to participate in the
conversation.
b. At the second level, ne urges the hearer to understand what the speaker
says.
c. At the third level, ne urges the hearer to give his responses, such as agree-
ment and confirmation, to what the speaker says.
Just as in the case of backchannels, if an utterance of ne functions on a higher
level, it also functions on the levels lower than it.
This characterization of ne focuses on the speaker’s interaction with the hearer,
so it takes the same direction with analyses in terms of joint attention.
172 Yutaka Shinoda
References
LEUNG WAI-MUN
1. Introduction
When people communicate with each other, they often provide the
source of information and express their own standpoint and attitude.
Evidentiality thus refers to how people indicate the source of know-
ledge, attitude towards knowledge and commitment to knowledge.
Evidentiality is a very important linguistic phenomenon, which is
not only concerned with the source of information, whether directly
seen or heard, indirectly inferred, or obtained from other people, but
also reflects how much the speaker is willing to be responsible for
the information he or she is providing, which is related to whether
the information is reliable or not.
Evidentiality generally appears in every human language as a
linguistic phenomenon (Plungian 2001), but it is expressed by differ-
ent grammatical structures. The grammatical forms of evidentiality
are called ‘evidentials’ or ‘evidential markers’, which include two
types: one is inflectional in terms of prefixes or suffixes, and the
other is lexical, involving modal verbs, adverbs and adjectives etc.
Evidentiality basically adheres to sentence propositions, that is, when
the evidentials or evidential markers in a sentence are removed, the
sentence proposition is not affected.
It is rare that people speak completely objectively without sub-
jective feelings. Whatever degree of certainty that the speaker has
about the credibility of the source of information, it is inevitable that
subjective expressions are involved, conveying the speaker’s stand-
points, emotions and attitudes (Shen 2001). These expressions are
referred to as ‘subjectivity’, a concept which is opposite to objectiv-
176 Leung Wai-mun
2.1.1 Evidentiality
Jakobson (1957) was among the earliest scholars to propose the con-
cept of evidentiality. He was also the first to separate evidentiality
from mood and modality clearly and categorized evidence into four
types: quotative evidence, revelative evidence, presumptive evidence
and memory evidence. Based on Jakobson’s observations, Chafe and
Nichols (1986) further divided evidentiality into five kinds: degree
of reliability, belief, inference, hearsay and general expectation. Their
treating of evidentiality as a separate area of study made a great im-
pact on the research that followed. Aikhenvald and Dixon (2003)
Evidential Particles in Cantonese: the case of wo3 and wo5 177
3.1 Evidentiality
3.2 Mirativity
Matthews (1998) claims that ‘it is the third parameter, whereby know-
ledge is matched against expectations, that mirativity encodes.’ (3) is
a kind of miratives which indicates new and unexpected information
for which the speaker lacks psychological preparation. DeLancey
(1997: 36) remarks:
[The category mirative] marks both statements based on inference and state-
ments based on direct experience for which the speaker had no psychological
preparation, and in some languages hearsay data as well. What these appar-
ently disparate data sources have in common […] is that the proposition is one
which is new to the speaker, not yet integrated into his overall picture of the
world.
Evidential Particles in Cantonese: the case of wo3 and wo5 183
(2) 䇠䇠
ª¨ª¨ ¥s §A⎓§Y¨è ¥hἈ Ø{ ⌛⇣ ⍣ ┶
baa4baa1
baa4baa1 giu3 nei5
nei5zik1 zik1
hak1hak1
heoi3 wo3
heoi3 wo3
father
father askask
2SG immediate
2SG go SFP
immediate go SFP
‘Father
‘Father asked you to go immediately!’
immediately!’
(3) 䇠䇠
ª¨ª¨ ¥s §A⎓§Y¨è ¥h Ἀ? ⌛⇣ ⍣ ┅
baa4baa1 giu3 nei5 zik1hak1
nei5 heoi3 wo5 heoi3
zik1hak1 wo5
father askask
2SG immediate
2SG go SFP
immediate go SFP
‘Father said to me:
me: ‘Ask
‘Ask him/her
him/hertotogo
goimmediately.’
immediately.’‘ ‘
3.3 Subjectivity
(2) 䇠䇠
ª¨ª¨ ¥s §A⎓§Y¨è ¥hἈ Ø{ ⌛⇣ ⍣ ┶
baa4baa1 giu3 nei5
nei5zik1 zik1
hak1hak1
heoi3 wo3
heoi3 wo3
father askask 2SG
2SG immediate immediate
go SFP go SFP
‘Father asked you to go immediately!’
immediately!’
(3) 䇠䇠
ª¨ª¨ ¥s §A⎓§Y¨è ¥h Ἀ? ⌛⇣ ⍣ ┅
baa4baa1 giu3
giu3 nei5 zik1hak1
nei5 heoi3 wo5 heoi3
zik1hak1 wo5
father askask
2SG immediate
2SG go SFP
immediate go SFP
‘Father said to
to me:
me: ‘Ask
‘Ask him/her
him/hertotogo
goimmediately.’
immediately.’‘ ‘
(7) (HKUCC)
␊ ??┚◊
©O Ê\ -ø Ἂ
ª¾ ? Ⓖ 䞍 ┅
nei1di1 je5 keoi5 m4 zi1zi1wo5wo5
DEM things 3SG
3SG NEG
NEG know
knowSFPSFP
‘he said he does not know
know about
about these
thesethings.’
things.’
(10) (HKUCC)
◑炰 ¯u«Y
¼M¡I 䛇Ὢ ´X ®Þ¤Ò ⸦ Ø{¡I㡛⣓ ┶炰
waa1 zan1hai6
waa1 zan1hai6gei2gei2 so1fu4
so1fu4 wo3 wo3
INJ really
INJ really quite
quite wonderful wonderful
SFP SFP
‘Wow, it’s really wonderful!’
‘Wow, wonderful!’
4.2.1 Realization
In Example 12, the speaker takes some bones to feed a dog, and then
remembers that in the backyard there is a cat, which might be hungry
then, and so he takes some food to feed the cat also. The one who
‘realizes’ that something has to be done is the speaker himself.
4.2.2 Reminding
(13) (HKUCC)
A: ¡G¶¼
梚 ¥¤⤞¯ù ¦P勞 ¶¼ ¯ù
⎴³£ 梚 -ø ¦P Ø{¡A-øª¾
勞 悥 Ⓖ ÂI ¸Ñ ⎴ ┶炻Ⓖ䞍 溆 妋
jam2 naai5caa4
naai5caa4tung4
tung4jam2
jam2caa4
caa4dou1
dou1m4
m4tung4 wo3,
tung4 m4m4
wo3, zi1zi1
dim2gaai2
dim2gaai2
drink milk
milktea
teaand drink tea ADV
and drink NEGNEG
tea ADV same SFP,
sameNEG knowknow
SFP, NEG why why
¶¼ ¥¤⤞
梚 ¯ù ·| 勞
«d ?㚫?¡A ¦]⇲¬° Ê\ ┚ ▭炻 ⚈ 䁢 Ἂ
naai5caa4 wui6
jam2 naai5caa4 wui6 soek3
soek3 di1 ge2, jan1wai6
di1 ge2, keoi5
jan1wai6 keoi5
drink milk tea AUX
AUX weaken-stomach
weaken-stomach ASP ASPSFP,
SFP,because
becauseit it
㽫 ±o º¢ §r¡H
¿@ ⼿ 㺗 ⏨烎
nung4
nung4 dak1 zai6 dak1aa4 zai6 aa4
concentrated COM excessive SFP?
‘Milk tea and plain tea taste different. I don’t know why milk tea weakens
the stomach. Is it because it is too concentrated?’
B: Ὢ ⓲ 㽫 ┚ˤ
hai6 aa3 nung4 di1
COP SFP concentrated ASP.
‘Yes, it’s more concentrated.’
(14) (HKUCC)
§A ¤d¬è
Ἀ ⋫䣰-ø ¦n Ⓖ ݯ ⤥§Ú °Ú¡A
㏝ §Ú ㆹ ¤w¸g ⓲炻 ¤E ㆹ -Ó project
䴻 °Ú¡A ḅ ᾳ project ⓲炻
nei5 cin1kei4
nei5 cin1kei4m4m4hou2
hou2wan2
wan2ngo5
ngo5 aa3,
aa3, ngo5ngo5 ji5ging1
ji5ging1 gau2
gau2 go3go3project
project aa3,
aa3,
2SG ADV
2SG ADV NEG NEGgood
good find
find 1SG SFP, SFP,1SG
1SGalready
already nine
nine CLCLproject SFP, SFP,
project
§A ¦Û¤v
Ἀ 冒¥h ÁÙ ⍣ °Õ¡A§A
怬 ┎炻Ἀ ¦Û¤v ¶¤ ®I 冒¥h Ø{ 昲 ❳ ⍣ ┶
nei5 zi6gei2
nei5 zi6gei2heoi3
heoi3waan4
waan4laa1,
laa1,nei5
nei5 zi6gei2
zi6gei2 deoi2maai4
deoi2 maai4 heoi3
heoi3 wo3
wo3
2SG self
2SG self go return SFP, SFP,
go return 2SG2SGself handselfin ASP
handgo in SFP.
ASP go SFP.
‘Don’t ask me to do it as I have nine projects at hand already. Why don’t you go
yourself? You just need to drop it.’
⍰ ℵ 敾䤵 ⓲
mai5 jau6 zoi3 cong2wo6 aa1
NEG again again make trouble SFP
‘Remember, follow me closely and don’t make trouble again.’
4.2.3 Contrasting
In this example, Speaker A hears some noises in the car park and
feels worried, as there are valuables in the car. He then asks Speaker
B whether everything is fine. Speaker B replies that there is no prob-
lem with the car. In Speaker B’s reply, wo3 appears at the end of the
utterance and conveys the meaning ‘in spite of appearances or what
192 Leung Wai-mun
you think, the car is okay’. Thus, the particle shows that the message
contains an element of ‘contrast’.
4.2.4 Hearsay
19th century to the early 20th century, wo3 was mainly used to report
other’s speech, thus having a quotative function, which can be seen
in the following two examples:
The two examples above clearly show that wo3 is used for reporting
other’s speech. The English translations were provided by the text-
book authors, and from the translations ‘he said’ and ‘he says’, we
can judge that they are examples of indirect speech. In Example 18,
speaker A asks speaker B what time ‘he’ will come, and speaker B
uses the structure ‘ Ἂ娙…┶ ’ [He said…wo3] in his reply. In Ex-
ample 19, in speaker B’s reply to speaker A’s question ‘What did he
say?’, the clause ‘ Ἂ娙…’ [He said] is omitted as it is known from
the context, and the particle wo3 alone signals that the answer is not
speaker B’s but a report of the speech of that person they are talking
about. In terms of sentence structure, there have not been any major
changes in the past one hundred years; both ‘ Ἂ娙…┶’ [He/She
said…wo3] and ‘…┶ ’ […wo3] are acceptable.
194 Leung Wai-mun
(21) (HKUCC)
ㆹ °Ý⓷
§Ú 旧䓚 Ãä-Ó
ªü¥Ò ©O¡A ␊炻 ª± 怲ᾳ 䍑
±o ³Ì Åö ©O¡A ⼿ 㚨 䘚 ␊炻
ngo5 man6 aa3gaap3 ne1, bin1go3 waan2dak1
bin1go3 waan2 dak1 zeoi3
zeoi3 din1
din1 ne1, ne1,
1SG askask PRE-A
PRE-A SFP, play
SFP, who who play wildly
COM most COM SFP,
most wildly SFP,
Ἂ ¸Ü Ê\娙Ø{¡C
Ê\ Ἂ ┶ˤ
keoi5wo3 wo3
keoi5 waa6 keoi5
said3SG3SG
3SG said SFP. SFP.
‘I asked Mr. A who played
played the
the most
most wildly,
wildly,and
andhe
hesaid
saidMr.
Mr.BBdid.’
did.’
4.3 Discussions
of Chafe and Nichols (1986) (see 3.2). We see that both wo5 and wo3
can be called ‘evidential particles’. With the advancement of evi-
dentiality studies, what we now know about the features of wo5 and
wo3 is much more complicated than previously recognized. These
particles do not simply show the moods of the speaker, but reflect
how they recognize the external world as well.
The fourth feature of wo3 is hearsay which is considered to be
more complex to account for. It is said not to be easy to differentiate
wo3 and wo5 in some situations because wo3 overlaps with wo5 in
hearsay evidentiality. Both are often used in reported speech. Sub-
jectivization, however, gives us a very good explanation of the two
particles. We can prove it from the following minimal pair:
evidence for his or her statement with his or her own attitudes. The two
examples above show the different attitudes and stances of the speak-
ers. Thus, wo5 and wo3 obviously have different social functions from
the pragmatic point of view.
Moreover, the quotative evidential wo5 used in indirect speech
shows that the speaker is not willing to take the responsibility of the
quoted content, and the SFP wo3 explains others’ motivations be-
hind his/her behavior by the speaker’s own speculation and emotion,
which connotes the speaker’s assessment of the evidence for his or
her statement. From daily conversations, it is found that the hearsay
evidential wo5 is used when the speaker is not involved in the re-
ported content so as to speak objectively, irrespective of personal
opinion and subjective speculation on the issue. Aikhenvald (2004)
points out that evidentiality is neutral as to the speaker’s commit-
ment to the reported information. It can be seen that the speaker does
not show his stance when wo5 is employed, just merely reports other
people’s wording unchangeably and objectively, or explains other
people’s intention. However, wo3 is employed when the speaker is
involved in the reported content, or shows the quotation based on
his/her own speculation or intention. The apparent difference can be
seen in the example below:
Example 24 is the only example that the mirative wo3 and the quota-
tive wo5 appear in the same utterance in our data. Wo3 and wo5 are
certainly not interchangeable in this example, otherwise the mean-
ings conveyed will be totally different. The sentence with the SFP
wo5 is used to show the speaker does not stand for the opinion of the
reported speech, which is ‘Madam committed adultery’. On the con-
trary, what we discover from wo3 used in the last sentence is when
wo3 is used; the speaker shows his stance explicitly. The reported
content and the speaker are closely related, as the speaker interprets
others’ behavior subjectively. Specification associated with wo3 may
have subsequently become reanalyzed and absorbed directly into the
element wo3 as an inherent restriction on its use. From such a con-
trast, we can say that SFP wo5 focuses on the source of information
and wo3 focuses on the speaker’s reactions (surprise and unexpect-
edness). wo3 can also be used for reporting others’ speech, where the
speaker’s stance is still clearly shown, implying that the speaker does
not report exactly what others mean, or just reports others’ messages
by his/her own understanding, or shows how true the reported speech
is (to share the responsibility with the person being cited):
reveal what ‘Daiso’s sister’s mother’s brother’s son’’ says, but also
the speaker’s own viewpoint, i. e., he does think ‘it did have’ some-
thing. In addition to this, one of the situations we should be aware of
is, when SFPs wo3 and wo5 are used to interpret others’ motivations
behind their behavior, wo3 is preferred once the behavior is gener-
ally accepted and obligatory, and wo5 is preferred once the inten-
tions of the behavior are unclear.
Based on this difference, wo5 has developed another usage, and
the two cannot be substituted by each other. That they are not interchange-
able is shown in the following minimal pairs, where in Example 26
wo3 shows reminder, and in Example 28 wo5 shows that the informa-
tion gathered is not correct and the tone-of-voice is disapproving:
(27) *¥X
*↢ ? ¦æ ♇ Á¿ 埴 ? ¹D¸q ¦n 嫃? ⎻ 忻佑 ⤥ ┅
ceot1 lai4
lai4haang4 gong2gong2
haang4 haa5 dou6ji6
haa5 hou2 wo5dou6ji6 hou2 wo5
out comecome
walkwalk
talk ADVtalkvirtue ADV
good SFP virtue good SFP
‘Being in the society,
society, won’t
won’t you
you try
try and
and be
be kind
kindand
andloyal?’
loyal?’
(29) *¥H¬° Ê\ Ἂ
*ẍ䁢 ¯q §A䙲Ø{¡A Ἀ §Q¥Î §A ┶炻 ©Q¡I ⇑䓐 Ἀ 炰
keoi5 jik1
ji5wai4 keoi5 jik1nei5 wo3,
nei5 lei6jung6
wo3, nei5 zaa3
lei6jung6 nei5 zaa3
think 3SG benefit
3SG 2SG 2SG
benefit SFP take advantage
SFP of 2SG SFP
take advantage of 2SG SFP
will benefit
‘Don’t think he will benefit you,
you, he
heisis just
justtaking
takingadvantage
advantageofofyou!’
you!’
One more point to add, from the pragmatic point of view, is that the
elements in the language that are being used to achieve communica-
tive purposes are worth studying since the combined meaning of the
individual words in a sentence may differ from what the speaker or
writer intends to mean. In pragmatics, a speech act is divided into
three parts: locutionary act, illocutionary act, and perlocutionary act.
Of these, the illocutionary act is the core of the speech act. A locution-
Evidential Particles in Cantonese: the case of wo3 and wo5 199
5. Conclusions
References
English
Chinese
HAN YANG
1. Introduction
Table 1.
It is evident from Table 1 that the time reference of the events may
be lexicalized by means of various time adverbials without even con-
sidering how tense and aspect might be grammaticalised in the verb
phrases in these three languages.
Given that both the so-called ‘tensed’ (Japanese and English)
‘tenseless’ (Chinese) languages employ the lexicalization of the time
reference of the events, it is clear that lexicalization, rather than ver-
bal particles or inflections is more basic to the expression of time
reference. Verbal particles and inflections are on the other hand sec-
ondary modifications.
Tense, Aspect and verbal morphemes in Chinese, Japanese and English 207
Japanese English
Past (1) Kare wa kyonen sotsugyoo shita (3) He graduated last year.
Non-past (2) Kare wa rainen sotsugyoo suru (4) He is graduating next year.
Table 2.
The deictic time kyonen and the perfective marker -ta in Japanese
(1) indicate the past time reference, and such sentences are generally
regarded as having ‘past tense’ in Japanese. Likewise, the morpheme
-ed and the deictic time adverbial last year in English (3) indicate
past time reference, and such sentences are said to have ‘past tense’
in English.
In the case of Japanese (2), the verb ending -ru, together with
the deictic time adverbial rainen gives the indication that this sen-
tence is of non-past reference. In the same way as Japanese (2), the
copula is and morpheme -ing in English (4), together with the deictic
time adverbial next year indicate that the time reference is one of
non-past.
Propositions equivalent to those expressed by (1)-(4) in Table
2 may be realised in Chinese as follows:
(5) Past:
Ta qunian biye.
He last year graduate
(He graduated last year.)
208 Han Yang
and
(6) Non-past:
Ta mingnian biye.
He next year graduate
(He will graduate / is graduating next year.)
From the above it is clear that Chinese does not have obligatory tense
marking as in Japanese and English.
This can be further evidenced by the following unacceptability
caused by the conflict between the time adverbials and tense mark-
ers in Japanese and English (examples (13)-(16)), as well as the ac-
ceptability of all time adverbials by an identical verb in Chinese with-
out any verbal particles (examples (17)-(18)):
The ungrammaticality of (13) and (14) below is caused by the
conflict between future time reference and morphemes that signal
non-past time reference:
(13) *Kare wa lainen sotsugyo shita.
He topic-prt next-year graduate past-marker
By contrast, Chinese does not exhibit this type of conflict at the gram-
matical level:
(17) Future time reference:
Ta mingnian jiu biye le.
He tomorrow then graduate le
210 Han Yang
2 Discussions in this and the following section are based on Unit 3 ‘Tense, as-
pect, reference’ in the course material entitled Introduction to Semantics and
Pragmatics, pp. 21-26 written by the present author and published by the Open
University of Hong Kong (1999) and (2001). Reprinted (2002).
Tense, Aspect and verbal morphemes in Chinese, Japanese and English 211
This indicates that in Japanese the focus of the verbs such as ‘finish’
may well be on the B (beginning) point. Therefore a marker that
specifically indicates one event that precedes another is needed to
indicate the relative order of events.
212 Han Yang
The substitution of wan [finish] with le, same as the presence of the
past tense marker -ed in English (22) above, causes ungrammaticality:
However, this does not imply that the le particle does the same thing
as the English -ed to mark the past tense. This is supported by the
fact that le can actually co-occur with lexical verbs that signal comple-
tion, as in:
Neither stative nor activity verbs have an end point F, and so they are
called atelic verbs. Whereas the notion of atelic denotes an event
without an F point, telic concentrates on the end result of an event.
Achievement and accomplishment verbs belong to this category. The
Tense, Aspect and verbal morphemes in Chinese, Japanese and English 213
5.1 State
3 Readers interested in this aspect of the study can refer to Hofmann’s discus-
sion on Limits to Events (1993: 139-158) with special reference to aspect of
events and verb classes.
214 Han Yang
For example:
(27) *I am loving you.
(29) Wo ai ni.
I love you
(I love you.)
4 There is a classical study on verb classification in Japanese using a -te iru test
by Kindaich (1955), where this verbal suffix is characterised as expressing
whether progressive or perfect meaning depending on the lexical meaning of
the main verb to which is it attached.
Tense, Aspect and verbal morphemes in Chinese, Japanese and English 215
Even when the sentence does not express the present state, this class
of verbs in Japanese still requires progressive marking as in:
5.2 Activity
(41) English: John read War and Peace for three hours.
All of the above three examples from Chinese, Japanese and English
do not tell us if someone has finished reading the book; it simply
indicates that the activity of reading took place, i.e. someone did
some reading, and that reading lasted for three hours.
216 Han Yang
All activity verbs share the same characteristic, that is, with-
out implication of an end point. It is therefore only natural that all
three languages should accept the progressive marking in the case of
activity verbs. Below are some examples:
5.3 Achievement
AIRPORT
Figure 1.
This sentence shows that the person has not yet arrived at the airport,
where mei is a negative item signalling the non-realisation of an event
in Chinese.
The sentence below is of interest: the le particle in post-verbal
position is optional, whereas the le particle is necessarily required at
the sentence-final position.
Tense, Aspect and verbal morphemes in Chinese, Japanese and English 219
It may thus be said that the le particle here behaves in a similar man-
ner as the progressive marker -te iru in Japanese.
This can be seen clearly by utilising the same Figure 1 above
with the addition of the Chinese equivalents as shown below:
AIRPORT
Figure 2.
5.4 Accomplishment
Learning a skill and teaching a language both require time and ef-
fort, and in this sense, the accomplishment verbs are durative as op-
posed to the punctive characteristic of the achievement verbs.
Similarly, verbs such as narau [learn] in Japanese do not have
the implication of attainment. In order to achieve the attainment mean-
ing of ‘learned how to …’, specific verb dekiru with -ta ending is
employed, as in:
Other examples are zhaodao [to find = look for + have found], shasi
[to murder = kill + have died], tingdong [to understand = hear + have
understood], and so on.
When the achievement verb in Chinese is a simple verb, such
as cheng [become], si [die], the le particle is necessarily required,
as in:
(63) Ta si le
he die le
(He died)
6. The le particle5
Le is often treated as an aspect marker together with guo, zai, zhe, and
sometimes, ne, within the study of the temporal system of Chinese
(e.g. Wang 1965, Chao 1968, Rohsenhow 1977, Li/Thompson 1981,
Li/Thompson /Thompson 1982, Chu /Chang 1987, Huang 1988,
Huang/Davis 1989, Smith 1991, He 1992, Campbell 1992, Mangione/
Li 1993, et al.). Le is also treated by some scholars as having the
function of marking both tense and aspect (Li 1998), and certain
occurrences of le (e.g. as in You xiwang le.) have even been excluded
from the examination of the temporal system altogether for being a
‘mood’ particle (Li 1998). Also when le comes after a verb at the end
of the sentence, it is claimed to be “difficult to determine whether it
is the perfective verb suffix -le, or the CRS [currently relevant state]
sentence-final particle le”. (Li/Thompson 1981: 296). Nevertheless,
most scholars agree that whenever there is an involvement of le in a
sentence, a sense of sequential ordering of events is detected.
Some scholars have in the past suggested that the le particle, when
occurring in post-verbal position, signals the past tense (Gao 1970),
but this position was challenged by many who claim that the post-
verbal le marks perfective aspect. (e.g. Wang 1965, Chao 1968,
Rohsenhow 1977, Li/Thompson 1981, Li/Thompson/Thompson 1982,
Chu/Chang 1987, Huang 1988, Huang/Davis 1989, Smith 1991, He
1992, Campbell 1992, Mangione/Li 1993, et al.).
It is indeed the case that Chinese does not obligatorily relate
the time of situation being described to the time of utterance by any
systematic variation in the sentence structure (cf. Lyons 1977), and
the le particle can and does occur in sentences where non-past time
reference is clearly indicated:
and
and provide explanations to the effect that the function of le, when
occurring in this position also signifies a transition in border-cross-
224 Han Yang
ing journey irrespective of the relation between the speech time and
the reference time, i.e. the relative tense.
Location
Direction
Movement
Border-crossing
A B
A and B are contiguous locations, and the arrow (à) is the component
of directed movement, the process of which is an essential compo-
nent of journey events.
Expressions focussing on only one stage or phase of a journey such
as Egor left France (i.e. X left location A) or Egor reached Italy (X reached
location B) may be represented by Figure 4 and Figure 5 below:
A B
A B
a. Post-verbal position:
b. Sentence-final position:
This indicates that the event described must have taken place at some
point before the speech time and that the stated event no longer ex-
ists at the time of utterance.
The post-verbal le construction can however allow another
clause to follow:
There are two things that are apparent: (a) Xiao Wang ate some food
and (b) Xiao Wang is no longer eating.
The post-verbal le construction can therefore fit into Jessen’s
three-state-journey (Figures 4 and 5) in the following manner:
E [~E]
E [~E]
This implies that the event stated in the le clause Xiao Wang chi fan
le has not come to an end; therefore, the inception of another event
as stated in the second clause cannot take place. The speaker of (68)
is then saying two things: (a) Xiao Wang began eating, and (b) Xiao
Wang has been eating since, with a presupposed meaning of Xiao
Wang stopped being in the state of not eating.
The acceptance of non-past reference time by the sentence-
final le construction as exemplified by (78) below allows us to inter-
pret that this sentence has a meaning of Xiao Wang will begin eating
with a presupposed meaning of Xiao Wang will stop being in the
state of not eating:
[~E] E
(82) Xiao Wang chi fan le, xianzai hai zai chi ne.
Little Wang eat food le, now still at eat ne
(Xiao Wang started eating, and he is now still eating.)
The meaning of (69) therefore has two parts: (a) Xiao Wang
ate something, as indicated by the past-verbal le; and (b) the state of
Xiao Wang’s having stopped eating has begun, as indicated by the
sentence-final le, as represented below:
E [~E]
[~E] E
6.9 Summary
It has long been accepted that when le occurs in a position that is both
sentence-final and post-verbal, the interpretation of the resulting sen-
tence is ambiguous, as observed by Li and Thompson (1981: 296):
The achievement verb zou [to leave], in combination with the func-
tion of the ‘order of events’ of the le particle indicates that a transi-
tion (i.e. border-crossing) from Xiao Wang’s being at a place to his
not being at that place has taken place at some point prior to the ST.
In other words, up to the point of transition, Xiao Wang was at loca-
tion E, at some point in time prior to the ST a transition took place
and that transition is no longer in the process of happening, and that
resulted in him being not in location E; so he is not there in location E
at the ST.
Graphically, this transition in relation to ST can be represented
as follows:
ST
--- --- --- --- --- --- ---
time
E [~E]
This type of le construction allows both the future and past time ref-
erences as exemplified by (87) and (88), and can be represented by
Figure 13 and Figure 14 below:
Past:
(87) Xiao Wang zuotian zou le.
Little Wang yesterday leave le
(Xiao Wang left yesterday.)
Future:
(88) Xiao Wang mingtian zou le.
Little Wand tomorrow leave le
(Xiao Wang is leaving tomorrow.)
Tense, Aspect and verbal morphemes in Chinese, Japanese and English 235
Yesterday Today
time ST
E [~E]
Figure 13. Xiao Wang zuotian zou le and speech time (ST).
Today Tomorrow
time ST
E [~E]
Figure 14. Xiao Wang mingtian zou le and speech time (ST).
Does this mean that this is ambiguous? Indeed it has any of the three
possibilities:
The third possibility exists as Chinese syntax does not allow the se-
quence of two identical particles in a sentence.
This long accepted syntactic ambiguity is in fact invari-
ably dissolved by a given context as demonstrated below:
7. Conclusion
References
This shows that English words and phrases are not only always used
with their original meanings but are often modified by Japanese speak-
ers depending on the situations in which they are used. It has been said
1 In this chapter, I represent Japanese Katakana letters (i.e., Japanese phonograms)
by the combination of consonant and vowel such as CV, V, CCV. Each set
represents a mora, or phonologically a half-sized unit of foot in English terms.
Japanese morae cannot end with consonant because they should be open syl-
lables, but the exceptions are those morae ending with syllabic nasal (i.e., N)
and the front half of consonant gemination (various consonants). Borders be-
tween morae are represented by blanks, so that the number of morae in (a) is 6,
and that of (b) is 4 while the number of syllables in (a) and (b) are 3 and 2.
(a) KI K KO O MA N (Kikkoman) (b) KYA N DI I (candy) [Katakana representation]
1 2 3 4 5 6 1 2 3 4 [the number of morae]
KIK KOO MAN KYAN DII [syllable representation]
1 2 3 1 2 [the number of syllables]
244 Masanobu Horiguchi
1. Loanwords in Cantonese
(5) a. ⓗኈ for taxi,ᕮኈ for bus,㜿㔛ᕮᕮ for Ali Baba,ဋ for card
b. ᮒྂຊ
•éŒÃ—Íforfor chocolate,ᚷᗮ
chocolate, Šõ—õfor for
cream,ྍཱྀྍᵹ for Cocafor
cream, ‰ÂŒû‰Â•Ù Cola,
Coca Cola,
⡀
?? for for
curry, •z™² for for
curry,ᕸ⏩ pudding, •U—–’n forfor
pudding,ᣆ⹒ᆅ brandy, ˆÐ•mŠõ for
brandy,ጾኈᚷ forwhisky
whisky
c. ព for Italy,௨Ⰽิ for Israel
d. ㏔ኈᑽ for Disney,ዲⴊሳ for Hollywood,⡿ወ for Mickey,
㧗㐔 for Gundam (Japanese robot cartoon)
246 Masanobu Horiguchi
In the following examples, letters from the alphabet are used as well
as Chinese characters as phonograms:
The above (7b) FANCL is an ‘acronym,’ which means that this word
is pronounced as [fænkl], not as [ef ei en si: el]. On the other hand,
in the following examples of ‘initialisms,’words are pronounced let-
ter by letter (Tabata 2004 for the distinction between acronym and
initialism):
For example, (8a) DHC is pronounced as [di: eit5 si:]. Words in this
category come with more ‘foreign’ impressions.
All in all, as can be seen from (4), the Cantonese language
adopts fewer phonograms in loanword formation than Japanese, and
Japanglish in Katakana and a Comparison with Loanwords in Cantonese 247
As is well known, the Japanese writing system has three sets of sym-
bols: Chinese characters (ideogram); Hiragana letters (phonogram);
and Katakana letters (phonogram). Combinations of Chinese char-
acters and Hiragana letters are thought to be basic Japanese sentence
structures by Japanese speakers while Katakana letter representa-
tions of loanwords, proper nouns, and onomatopoeia or mimesis are
considered somewhat marked words or phrases (i.e., italicized phrases
in English terms). According to Bunkacho (1997: 38), the usage of
Katakana letters for transcribing loanwords dates back to almost 1500
years ago, when Japanese people tried to import Buddhism. By cit-
ing the examples from Bunkacho (1997: 13), the following Buddhist
terminology was first transcribed by Chinese characters phonogrami-
cally, like in the case of Cantonese in (5), and, then, Katakana letter
transcriptions were added alongside these Chinese character words
because Chinese characters can be pronounced in various ways:
(10) ྜ⩚(KA
a. ‡‰H (KA P PA from
from Portuguese
Portuguese‘capa,’
‘capa,’meaning
meaning‘raincoat’)
‘raincoat’)
ኳ㯑⨶(TE
“Vên—… NN
(TE PUPU RARAfrom Portuguese
from ‘têmporas,’
Portuguese or tempura)
‘têmporas,’ or tempura)
↮ⲡ(TA
‰Œ‘• BA
(TA KO
BA KO from
fromPortuguese ‘tabaco,’
Portuguese or or
‘tabaco,’ tobacco)
tobacco)
• 㦵∩(KA RU TA from Portuguese ‘carta,’ meaning ‘Japanese-style card
game’)
⌆⍄KO
b. àÛàè (KO OOHI HIIIfrom
fromDutch
Dutch ‘koffie,’
‘koffie,’ or
or coffee)
coffee)
㘚ຊ(BU
èF—Í (BU RI KI from
from Dutch
Dutch ‘blik,’
‘blik,’meaning
meaning‘tin’),
‘tin’),
㯏㓇(BI
”••ð (BI I IRU
RUfrom
fromDutch
Dutch ‘bier,’
‘bier,’ or
or beer)
beer)
ὒⅉ(RA
—m“” (RANNPUPUfrom
fromDutch
Dutch‘lamp,’
‘lamp,’ororlamp)
lamp)
It was not until 1902 that general guidelines for adopting Katakana
letters in transcribing loanwords were established by the government’s
language research committee (Bunkacho 1997: 38), but Katakana
letters were considered loanword-specific from much earlier times.
Compound word formation involves not only English but also other
foreign languages:
There are also suffix examples such as ‘in,’ ‘up,’ or ‘down’ as follows2:
b. something + A P PU (up):
I ME E JI A P PU (image + up): improve one’s reputation
RE BE RU A P PU (level + up): raise the level of something
c. something + DA U N (down):
I ME E JI DA U N (image + down): damage one’s reputation
RE BE RU DA U N (level + up): lower the level of something
Japanese products dominated the world market much before the era
of the ‘bubble economy,’ and 60 percent of them consist of Katakana
naming, which may make these products sound suitable for export
(if we count Katakana compounds with Chinese characters, the sum
will rise to 74.4 percent, according to Abe 1990: 191). This is in stark
contrast with the small usage of loanwords in newspapers (i.e., 12
percent according to Ishiwata 1985: 17). In the following examples
(from Kimura 1998, Akiba and Chikuma 2000, and Yasuda 2003) of
Katakana brand names, Katakana phonograms are used as if they
played the role of masking the original meanings of companies or
products (Morioka 1993: 1909):
In any of the above examples, brand names are too familiar to Japa-
nese people to take trouble to analyze and reconsider the origins of
the words, and most Japanese people do not know what the above
brand names are made up of. In (24a), Japanese speakers can intu-
itively break down the compound into each part, which corresponds
to the sequences of morae, or Japanese minimum phonological units
(Vance 1987 and Kubozono 1989). In (24b), Japanese speakers may
have some difficulty in reaching the complete analyses of these com-
pounds, which entail the concept of syllables in English terms (Kahn
1976 and Selkirk 1982), not morae: for example, E ME RO N should
be divided into E ME R + O N, and the division of one mora ‘RO’
into two subcomponents ‘R + O’ cannot intuitively be possible in
Japanese. In (24c), the examples are too complex to be analyzed in-
tuitively: assimilation or nasalization of [p] into [m] in Jeemny [Jeep
+ -ny] cannot easily be realized in Japanese; and DE MI O (de: of +
mio: mine) sounds like one word, not a compound for ordinary Japa-
nese speakers (because both components are function words, so that
they rarely make sense to ordinary Japanese speakers, except for those
who know Spanish.
Representing Chinese company names or Chinese product
names by Katakana letters dates back to as early as the pre-Second
World War period:
I think most Japanese speakers regard the above maker names (27a)
and product names (27b) as originally Katakana letters or even alpha-
bet letters.
More complex cases of making Japanese phrases or even sen-
tences sound like Western languages can be found in the examples
by Gennai Hiraga (1728-1779), a natural scientist and drama writer
in the Edo period (Bunkacho 1998: 35):
(29) a. Umakacchan: a brand name for instant noodle (in 1979: House Food Corp.),
meaning ‘tasty’ in Japanese Kyushu dialect
b. Donto: a brand name for a pocket body warmer (in 1982: 㔠㫽Corp.),
meaning ‘at a burst’
c. Ken’onkun: a brand name for electric thermometer (in 1983: OMRON Corp.),
meaning ‘Mr. Measurement Man’
d. Otakkusu: a brand name for fax machine (in 1991: Matsushita Electronics
Corp.), meaning ‘home (Otaku) fax’
(30) a. Panasonic wide-screen TV set ‘ ⏬⋤’ (in the 1990s: meaning ‘screen king’)
→ Panasonic thin flat panel TV set ‘TA U’ (Tau) (in the 2000s: Greek
alphabet)
b. HITACHI wide-screen TV set ‘ 㠉ඣ’ (in the 1990s: meaning ‘revolutio-
nary boy’)
→ HITACHI plasma TV set ‘U U’ (Wooo) (in the 2000s: onomatopoeia)
258 Masanobu Horiguchi
They are no longer ‘Katakana’ words since they are all composed of
alphabet letters, and many of them are found among computer item
brand names, so that they will create a more indifferent and artificial
image than Katakana brand names. In the case of acronyms such as
(32a), however, the way Japanese speakers pronounce them reveals
that these acronyms keep the most stable phonological length as Japa-
nese (i.e., three or four morae), which corresponds to three or four
Katakana letters.
young Japanese speakers. As far as I know, there are many similar examples
such as follows:
(f) KI S SU (kiss) → KI SU (g) FI RI P PI N (Philippine) → FI RI PI N
Taking all the above factors into consideration, it may safely be said that the
occurrence of consonant gemination in loanword formation is not caused purely
by the phonetic impression of original English words. Instead, insertion of
geminated consonants will create, in such loanwords, a special impression such
as ‘swiftness’ or ‘speediness,’ which older Japanese speakers recognize as more
foreign and consider to be found also in onomatopoeia and mimesis.
Japanglish in Katakana and a Comparison with Loanwords in Cantonese 261
At about the same time, boys’ hero programs (Saito 1998: 16) tended
to take more rigid forms as ‘Katakana robot or battleship names with
a few Chinese characters’ as follows:
Japanglish in Katakana and a Comparison with Loanwords in Cantonese 263
What is unique to girls’ programs is that Hiragana letters are used for
some heroines’ names (cf. the nature of Hiragana letters as familiar-
ity to women and children in Yasuda 2003: 195-196). As pointed out
by Saito (1998), it is true that the border between hero and heroine
cultures is disappearing, and that the stories or settings of young-
sters’ TV programs are getting more and more complicated, but, pos-
sibly due to the influence of parents now in their 30s, 40s, and 50s,
the new generation of children will keep the hero and heroine speech
style, not to mention a lot of Japanglish used in these programs.
Japanglish in Katakana and a Comparison with Loanwords in Cantonese 265
6. Concluding Remarks
References
1. Introduction
There were a few attempts to sense tag the Sinica treebank and the
Penn Chinese Treebank (for examples, cf. Dang, Chia, Palmer and
Chiou (2002) and Ker and Chen (2004)), but such resources are not
publicly available for research. Neither is the largest sense-tagged
Chinese corpus (one million words) tagged by Li et al. (2003). The
IR-lab of Harbin Institute of Technology (2005) has released a small
corpus of 1,000 sentences (around 21,000 words) to the public in
2005. Like Li et al. (2003), the sense inventory in this study is the
Tongyici Cilin (Thesaurus) (Mei 1983), which uses sense ID num-
bers to represent the hierarchical classification of concepts. The dis-
advantage is that human beings cannot interpret the meanings from
the ID numbers directly, but must check against the matching tables.
Semantic Annotation of Chinese Texts 273
1 <http://godel.iis.sinica.edu.tw/CKIP/hk/index.html>.
274 Wong Ping Wai
NO.=040263
W_C=姀侭
G_C=N
E_C=
W_E=journalist
G_E=N
E_E=
DEF=human|Ṣ,#occupation|借ỵ,*gather|㍉普,*compile|䶐廗,#news|㕘倆
The first line is the record number. The Chinese word is shown in
“W_C”, with the POS and example shown in “G_C” and “E_C”. The
English word and the corresponding POS and example are shown in
“W_E”, “G_E” and “E_E”. The concept definition is shown in “DEF”.
The approach to meaning representation is based on compo-
nential analysis. The sememe appearing in the first position of “DEF”
– ‘human| Ṣ ’ is called the categorical attribute. It names the hyper-
nym or the superordinate term, which gives a general classification
of the concept. The sememes appearing in other positions: ‘occupa-
tion| 借ỵ ’, ‘gather| ㍉普 ’, ‘compile| 䶐廗 , ‘news| 㕘倆’ are additional
attributes, which provide more specific, distinguishing features.
The definition tells us that journalist is a kind of human, who
is the agent of gathering and compiling news. This concept is related
to occupation.
There are four main categories of content words in the HowNet
dictionary. The example mentioned above is an Entity concept, which
mainly consists of nouns.
Another category is Event, which mainly contains verbs, e.g.,
“ ⬀㫦 ” (savings).
The third category is Attribute, which contains nouns referring
to attributes, e.g., “Ṗ⹎ ” (brightness). Quantity is a subcategory of
Attribute, e.g., “ ” (times).
276 Wong Ping Wai
(1) read
object
subject auxiliary
I will book
determiner
the
X → Y means (e.g. X=“read”, Y=“I”):
• “Y depends on X” or “X governs Y”, e.g. “I” depends on “read” or “read”
governs “I”
• X - the governor/head of Y, e.g. “read” is the governor of “I”
• Y - the dependent of X, e.g. “I” is the dependent of “read”
The relation is asymmetric such that the label, e.g. “subject” and “auxil-
iary” indicates the relation of dependent to the head. For example, the
Semantic Annotation of Chinese Texts 277
This chapter adopts the conventions described in Gan and Wong (2000)
for the representation of dependency relations, which are based on,
but also a modification of, Mel’cuk (1988), as shown in the following:
(2)
C2 C2
R1 , R2
C1 C1
In the figure above, C2 is located higher than C1. Between them are
the dependency relations R1, R2 or both. C2 is the governor which
governs the dependent C1. R1 is on the same level as C1 whereas R2
is located on the same level as C2.
ˤ炻烊烉烎炰
ˣ( )˪˫ˬ˭
ˮ˯“ ”‘ ’炽ʇʇ……
Semantic Annotation of Chinese Texts 279
This is the term used in Gan and Wong (2000) and Wang (2002). It is
similar to word segmentation. This process is assumed to be mainly
solved in the Sinica Corpus, in which each word is already segmented
and tagged with POS.
However, there are some differences in the criteria of segmen-
tation between the Sinica Corpus and HowNet. This chapter aims to
find out the conversion rules to make adjustments before the step of
sense tagging.
The following tables (person names and place names) show
some examples of the adjustments made by this chapter to the subset
of the Sinica corpus according to HowNet.
Persons Places
Patterns Sinica HowNet Patterns Sinica HowNet
~⩴ (suspect) 㜿⩴ 㜿 ⩴ ~ⶪ (city) ⎘ᷕⶪ ⎘ᷕ ⶪ
~㝸 (someone) 䌳㝸 䌳 㝸 ~㷗 (harbour) 檀晬㷗 檀晬 㷗
~⤛ (girl) 惕⤛ 惕 ⤛ ~䷋ (county) ⼘⊾䷋ ⼘⊾ ䷋
~⨎ (woman) 湫⨎ 湫 ⨎ ~悱 (village) 䤷冰悱 䤷冰 悱
~⥣ (wife) 㜿⥣ 㜿 ⥣ ~捖 (town) 㔿⋿捖 㔿⋿ 捖
~㭵 (mother) 昛㭵 昛 㭵 ~⋨ (district) ⇵慹⋨ ⇵慹 ⋨
~䇞 (father) 昛䇞 昛 䇞 ~埿 (street) ⃩㔯埿 ⃩㔯 埿
~⭞ (family) 昛⭞ 昛 ⭞ ~慴 (lane) ⸛慴 ⸛ 慴
Table 1. Different Segmentation Criteria between the Sinica corpus and HowNet.
A word may contain more than one sense. It is necessary to find out
the most likely senses and prune out the unlikely ones during the
process of sense tagging (both manual and automatic). It is achieved
by refrence to relevant clues in the context as well as the hand-crafted
knowledge base of HowNet.
280 Wong Ping Wai
After each word is tagged with HowNet senses, the next step is to
identify message structures. Message structures extend the research
on the HowNet knowledge base to the research on a specific lan-
guage – Chinese (Z. Dong, 2000). Different languages can have dif-
ferent message structures based on the same knowledge base. Mes-
sage structures are formed by words or phrases (two or more), which
are syntactically and semantically sound and convey a message.
A message structure is specified by the following format:
It is noted that the patterns found in word formation are also found in
phrase level and sentence level in Chinese. Take for example, the
pattern “(time| 㗪攻) [time] ← (event| ḳẞ)” can be found at vari-
ous levels (Gan and Wong, 2000):
(3)
word level: “㘐[time]ɤ忳” (exercise in the morning)
phrase level: “ℓ⸜[time]ɤ㈿㇘” (war against aggression for eight years)
sentence level: “4 㚰 6 㖍㗇㛇㖍[time]ɤ㓦`” (It is a holiday on Sunday,
the 6th of April)
(4) 㞍㋽ 徫䉗 ⼴ ⌛ 㓄ⷞ 嬎㥵
Catch escaped convict after immediately carry gun
“Carried the gun after catching the escaped convict.”
Semantic Annotation of Chinese Texts 281
(5) 炻
kernel
㓄ⷞ!
time patient
manner
⼴! ⌛! 嬎㥵!
,time
㞍㋽!
content
徫䉗!
282 Wong Ping Wai
Each text in the Sinica Corpus has the text information at the begin-
ning:
%% 㔯栆=⟙⮶
%% 㔯橼=姀㔀
%% 婆⺷=written
%% ᷣ柴=䉗伒
%% ⨺橼=⟙䳁
%% ⥻⎵=
%% ⿏⇍=䓟⤛
%% ⚳䯵=ᷕ厗㮹⚳
%% 㭵婆=ᷕ㔯
%% ↢䇰╖ỵ=冒䓙㗪⟙
%% ↢䇰⛘=冢䀋
%% ↢䇰㖍㛇=199012
%% 䇰㫉=
Gan and Tham (1999) have chosen the text with the theme “crime”
“ ᷣ柴=䉗伒 ”, and manually annotated 103 texts. I checked the Sinica
corpus again and found that there should be 193 texts in total, 90
texts of which are not tagged yet.
Manual tagging by Gan and Tham (1999) is time-consuming.
Thus, this study has processed the remaining part with the help of an
automatic parser, followed by human proofreading.
Semantic Annotation of Chinese Texts 283
(6)
<2051>
<syntax class="PERIODCATEGORY"><semantic class="{㧁溆},"><relation head="0" name="duplicate">ˤ
</relation></semantic></syntax>
<syntax class="VJ"><semantic class="䘤䓇,"><relation head="4" name="restrictive">䘤䓇
</relation></semantic></syntax>
<syntax class="Na"><semantic class="ḳね,午,#㭢⭛,#㬣,伒,"><relation head="2" name="existent">┳埨㟰
</relation></semantic></syntax>
<syntax class="DE"><semantic class="{㥳≑},"><relation head="8" name="dummy">䘬
</relation></semantic></syntax>
<syntax class="Nb"><semantic class="㔯⫿,⥻⎵,⮰,"><relation head="8" name="restrictive">㕘ᶾ凒
</relation></semantic></syntax>
<syntax class="Neu"><semantic class="㔠慷ῤ,㫉⸷,䫔,"><relation head="7" name="restrictive">⺧ᶨ
</relation></semantic></syntax>
<syntax class="Nf"><semantic class="䫎嘇,"><relation head="8" name="restrictive">嘇
</relation></semantic></syntax>
<syntax class="Na"><semantic class="凡,*㋱ỷ,#欂,彚,"><relation head="9" name="dummy">㺩凡
</relation></semantic></syntax>
<syntax class="COMMACATEGORY"><semantic class="{㧁溆},"><relation head="0" name="EOS">炻
</relation></semantic></syntax>
</2051>
Each word in a sentence has an index number. In the excerpt of the file
“crime.f80014c.30.xml” above, there are nine words in the sentence
(line no. 2051). Each word is represented by an index, beginning with 1.
This hidden information is listed explicitly in the following table.
284 Wong Ping Wai
Each sentence ends with a punctuation mark defined in 5.1. The top-
most node is the punctuation mark at the end of the sentence, as
shown in the following. In the XML format, as it is the topmost one,
it does not have a head (indicated by head=“0”), and it marks the
“EOS” – “end of sentence”.
ˤ 炻 烊 烉 烎 炰
(7) ⛐ ᶨ 䇯 びや 倚 ᷕ 炻
PREP one CL pity sound LOC
“In the midst of pity”
炻
kernel ⛐
scope ᷕ
scope 倚
content びや
quantity ᶨ
quantity 䇯
(8) 溻⊝ ⼺䛦 崟ḳ炻(f79109.4)
encourage people uprise
“Encourage people to uprise”.
炻
kernel 溻⊝
patient ⼺䛦
ResultEvent 崟ḳ
7.1.3 Unresolved
The sense tagging process of Gan and Tham (1999) was done on a
running text, without a concordance tool for further proofreading.
There might be some inconsistencies of sense tags. I have made some
corrections after the examination of word senses with a concordance
tool.
The annotation of word senses by Gan and Tham (1999) could only
rely on HowNet 1999, because HowNet 2000 was not available at
that time. It is not clear about the content of HowNet 1999, as it was
no longer available for download since the release of HowNet 2000
in October 2000. I have extracted all the senses found in the corpus
and checked against the HowNet 2000 dictionary. The following table
shows the statistics:
288 Wong Ping Wai
There are 2,447 sense definitions in the corpus, 1,976 of which match
the sense definitions in HowNet 2000, whereas 471 do not. Out of
471, there are 219 new entries that are not found in the HowNet 2000
dictionary. They were added by Gan and Tham (1999). I have exam-
ined each of them carefully, and added them to HowNet 2000 (cus-
tomized dictionary). As for the remaining 252 sense definitions, they
constitute the discrepancies between the HowNet 2000 dictionary
and the corpus. In most cases, the definitions would follow HowNet
2000.
For example, “ 㔁ⷓ ” and “ 侩ⷓ ” are translated to “teachers”
in English, but their usage is different. The former is often associated
with an occupation whereas the latter is used as an address term to a
teacher. The difference is reflected in the sememe ‘#occupation| 借ỵ ’
of sense definitions in HowNet 2000. Perhaps there was not a dis-
tinction in HowNet 1999, and neither was there in the corpus of Gan
and Tham (1999). I accepted the changes of sense definition of
HowNet 2000, and made the changes to the corpus accordingly.
Table 5. Difference of sense definition between Gan and Tham’s corpus and HowNet
2000.
Semantic Annotation of Chinese Texts 289
Two words in the corpus were not sense tagged. I thus manually
tagged them according to the knowledge acquired from the corpus.
Before using the tagger to tag the new corpus, I have conducted an
experiment on Gan and Tham’s (1999) corpus. 2,542 sentences (24,780
words) are allocated as the training corpus, with 636 sentences (6,196
words) as the testing corpus.
The following table shows the preliminary test result:
2 <ftp://ftp.cis.upenn.edu/pub/adwait/jmx/jmx.tar.gz>.
Semantic Annotation of Chinese Texts 291
The MaxEnt tagger has a default cut-off value for rare words. For
example, words with frequency lower than five would be regarded
as unknown words. For this small corpus, the data sparseness prob-
lems result in too many unknown words, which have reduced the
accuracy in the preliminary test. As it is not possible to adjust the
default value of this tagger, this chapter follows Wong and Yang
(2002), to concatenate three copies of the training corpus. The size
thus increases to 7,626 sentences (74,340 words). Thus, more rare
words are coped with by the parser, resulting in a better test result:
Tom Lai, Changning Huang and Ming Zhou have conducted research
on parsing Chinese texts using syntactic dependency parsers from
1994 to 2000 (Lai/Huang 1994, 1998, 1999, 2000, Zhou/Huang 1994
and Zhou 2000). The parser applied by Lai and Huang is unification-
based, but it is time consuming to write the lexical rules and gram-
matical rules. They have built some small corpora for testing with-
out public releasing, so such valuable resources are not available for
research in the current study.
This chapter applies the deterministic data-driven parser called
Malt Parser to dependency parsing. The parser (version 0.23) applies
the Memory-Based Learning (MBL) algorithm, which simply stores
experiences in memory, and solves problems by reusing solutions
from similar previously solved problems (Nivre/Scholz 2004).
The parser was designed to analyze the syntactic dependency
between POS tags. This study applies the parser to semantic depen-
dency which requires sense tags. The list of POS is much shorter and
exhaustive, but the list of sense tags is very long. For example, there
are 46 POS tags but 2,447 sense tags in the corpus of Gan and Tham
(1999). The semantic tags are too fine-grained. They need to be sim-
plified (become coarse-grained) so that dependency parsing could
be carried out.
The simplification process is based on the sememes involved
in the Message Structures base. For example, all sense tags of At-
tribute concepts are reduced to categorical attribute (the first sememe)
‘Attribute| Ⱄ⿏ ’. Most concepts of attribute value are simplified to
the categorical attribute ‘aValue| Ⱄ⿏ ’, except some special cases
that requires finer-grained senses. The following table shows some
of these examples.
Sense tags Examples Dependency relations
aValue|Ⱄ⿏ῤ,range|ⷭ⹎ 悥, ℐ Range
aValue|Ⱄ⿏ῤ,degree|䦳⹎ 朆ⷠ, ⼰ Degree
aValue|Ⱄ⿏ῤ,type|栆✳,special|䈡 恋, 忁 Restrictive
Accuracy m2 m3 m4
Head word 74.46% 76.98% 79.09%
Semantic Dependency 67.11% 71.28% 72.54%
8. Conclusion
There are two broad types of semantic annotation: (i) Semantic fea-
tures of words – annotation of word senses. (ii) Semantic relations of
items in a sentence – annotation of event roles of participants in-
volved in events. Previous works usually either annotated word senses
or semantic relations in a Chinese corpus. Li et al. (2003) annotated
both, but applying two different knowledge bases that hardly seemed
to be integrated. This chapter adopts HowNet’s approach to incorpo-
rate both types of semantic information when annotating a Chinese
corpus.
This chapter, based on the semantic features revealed by the
sense tags, has annotated Chinese texts with message structures. A
message structure base mainly contains examples at word and phrase
levels. The specialty of Chinese lies in the consistent patterns at word,
phrase and sentence levels. In this study, message structures are used
for annotating semantic dependency relations between the words in
a sentence, using the information of semantic features revealed by
sememes in sense definitions. It is a good proof of the robustness of
HowNet.
This study has applied automatic annotation to help annotate
Chinese texts with semantic information based on HowNet’s frame-
work. It can save much time and effort required in manual annota-
296 Wong Ping Wai
tion as employed in previous studies, e.g. Gan and Tham (1999) and
Gan and Wong (2000). As far as the author is aware, this chapter is
the first study that uses a data-driven parser to annotate semantic
dependency in the corpus manually tagged by Gan and Wong (2000).
Sense tags are made more coarse-grained, so that they can be adapted
to automatic semantic dependency parsing.
One of the limitations is the sense inventory. This chapter uses the
old version HowNet 2000, which is a free online resource, instead of
the latest version that requires a high licensing fee – HowNet 2005.
This is because the only available corpus tagged with HowNet’s word
senses is Gan and Tham (1999), which applies the HowNet 1999 and
is also publicly available for free. Having updated the sense defini-
tion to HowNet 2000 in the corpus, this chapter continues their work,
aiming to enrich this indispensable resource with more semantic an-
notation. In future work, I will work on an efficient way to update
the word senses of HowNet 2000 to the latest version (HowNet 2005)
in the annotated corpus.
The corpus in the current study breaks a sentence with a comma,
following the convention of the Sinica Corpus. This results in in-
complete sentences. The unlinked elements are tagged as ‘unresolved’
in the first parse, which is to be resolved in the second parse in future
work, which deals with event relations and event role shifting in a
sentence group.
The automatic parser applied in this study has encountered
problems of data sparseness due to small corpus size. The knowl-
edge bases of HowNet (such as dictionary and Message Structures
base) were not fully utilized yet. It is a direction of further research
to incorporate the advantages of both the knowledge base and the
annotated corpus in the parsing task.
Semantic Annotation of Chinese Texts 297
References
Baker, C.F. / Fillmore, C.J. / Lowe, J.B. 1998. The Berkeley FrameNet
Project. Paper presented at the COLING/ACL, Montreal,
Canada.
C.K.I.P. 1995. ㈨埻⟙⏲ 95-02:ᷕ⣖䞼䨞昊⸛堉婆㕁⹓䘬ℏ⭡冯婒㖶
[The Content and Illustration of Sinica Corpus of Academia
Sinica, Technical Report no. 95-02]: Institute of Information
Science, Academia Sinica.
Chen, K.-J. / Huang, C.-R. / Chen, F.-Y. / Luo, C.-C. / Chang, M.-C. /
Chen, C.-J. 2004. Sinica Treebank: Design Criteria, Represen-
tational Issues and Implementation. In A. Abeille (Ed.), Build-
ing and Using Parsed Corpora: Kluwer.
Dang, Hoa Trang / Chia, Ching-Yi / Palmer, Martha / Chiou, Fu-
Dong 2002. Simple Features for Chinese Word Sense Disam-
biguation. The 19 International Conference on Computational
Linguistics (COLING 2002), Taipei, Taiwan, August 24 - Sep-
tember 1, 2002.
Dong, Z. 2000. ᷕ㔯ᾉ〗䳸㥳㧉⺷ [The Pattern of Chinese Mes-
sage Structure]. Retrieved December 2, 2004, from <http://
www.keenage.com>.
Dong, Z. / Dong, Q. 1999. HowNet. from <http://www.keenage.com>.
Gan, K.W. / Tham, W.M. (1999). ➢㕤䞍䵚䘬ⷠ嬀䞍嬀㧁姣 [Gen-
eral Knowledge Annotation Based on HowNet]. Computational
Linguistics and Chinese Language Processing. 4/2, 39-86.
Gan, K.W. / Wong, P.W. 2000. Annotating information structures in
Chinese text using HowNet. Proceedings of the 2nd Chinese
language processing workshop, Association for Computational
Linguistics, 85-92.
Guo, J. 1993. PH – A Chinese Corpus. Communications of COLIPS.
3, 45-48.
Huang, C.-R. / Chen, F.-Y. / Chen, K.-J. / Gao, Z.-M. / Chen, K.-Y.
2000. Sinica Treebank: Design Criteria, Annotation Guidelines,
and On-line Interface. Paper presented at the 2nd Chinese Lan-
guage Processing Workshop, ACL, Hong Kong, October 7,
2000.
298 Wong Ping Wai
with which they can more easily and effectively draw words and
expressions in spontaneous speech and writing in English. (E-mail:
yhasebe@mail.doshisha.ac.jp)
This series aims to promote specialist language studies in the fields of linguistic
theory and applied linguistics, by publishing volumes that focus on specific
aspects of language use in one or several languages and provide valuable
insights into language and communication research. A cross-disciplinary ap-
proach is favoured and most European languages are accepted.
Vol. 4 Maurizio Gotti, Marina Dossena, Richard Dury, Roberta Facchinetti & Maria Lima
Variation in Central Modals. A Repertoire of Forms and Types of Usage
in Middle English and Early Modern English.
364 pages. 2002. ISBN 3-906769-84-4 · US-ISBN 0-8204-5898-8
Editorial address:
Prof. Maurizio Gotti Università di Bergamo, Facoltà di Lingue e Letterature Straniere,
Via Salvecchio 19, 24129 Bergamo, Italy
Fax: 0039 035 2052789, E-Mail: m.gotti@unibg.it
Vol. 5 Stefania Nuccorini (ed.)
Phrases and Phraseology. Data and Descriptions.
187 pages. 2002. ISBN 3-906770-08-7 · US-ISBN 0-8204-5933-X
Vol. 15 Gabriella Del Lungo Camiciotti & Elena Tognini Bonelli (eds)
Academic Discourse. New Insights into Evaluation.
234 pages. 2004. ISBN 3-03910-353-9 · US-ISBN 0-8204-7016-3
Vol. 23 Vijay K. Bhatia, Maurizio Gotti, Jan Engberg & Dorothee Heller (eds)
Vagueness in Normative Texts.
474 pages. 2005. ISBN 3-03910-653-8 · US-ISBN 0-8204-7169-0
Vol. 25 Ana María Hornero, María José Luzón & Silvia Murillo (eds)
Corpus Linguistics. Applications for the Study of English.
2nd printing. 526 pages. 2006, 2008. ISBN 978-3-03911-726-0
Vol. 27 Debbie G. E. Ho
Classroom Talk. Exploring the Sociocultural Structure of Formal ESL Learning.
2nd edition. 254 pages. 2006, 2007. ISBN 978-3-03911-434-4
Vol. 55 Jan Engberg, Marianne Grove Ditlevsen, Peter Kastberg & Martin Stegu (eds)
New Directions in LSP Teaching.
331 pages. 2007. ISBN 978-3-03911-433-7
Vol. 64 Vijay K. Bhatia, Christopher N. Candlin & Paola Evangelisti Allori (eds)
Language, Culture and the Law.
The Formulation of Legal Concepts across Systems and Cultures.
342 pages. 2008. ISBN 978-3-03911-470-2
Vol. 77 Ingrid Tieken-Boon van Ostade & Wim van der Wurff (eds)
Current Issues in Late Modern English.
436 pages. 2009. ISBN 978-3-03911-660-7
Vol. 79 Qing Ma
Second Language Vocabulary Acquisition.
333 pages. 2009. ISBN 978-3-03911-666-9
Vol. 123 Vijay K. Bhatia, Christopher N. Candlin & Maurizio Gotti (eds)
The Discourses of Dispute Resolution.
290 pages. 2010. ISBN 978-3-0343-0476-4
Vol. 132 Leyre Ruiz de Zarobe & Yolanda Ruiz de Zarobe (eds)
Speech Acts and Politeness across Languages and Cultures.
402 pages. 2012. 978-3-0343-0611-9
Vol. 140 Sandra Campagna, Giuliana Garzone, Cornelia Ilie & Elizabeth Rowley-Jolivet (eds)
Evolving Genres in Web-mediated Communication.
337 pages. 2012. ISBN 978-3-0343-1013-0
Vol. 145 David Tizón-Couto, Beatriz Tizón-Couto, Iria Pastor-Gómez & Paula Rodríguez-Puente (eds)
New Trends and Methodologies in Applied English Language Research II.
Studies in Language Variation, Meaning and Learning.
283 pages. 2012. ISBN 978-3-0343-1061-1
Vol. 151 Priscilla Heynderickx, Sylvain Dieltjens, Geert Jacobs, Paul Gillaerts &
Elizabeth de Groot (eds)
The Language Factor in International Business.
New Perspectives on Research, Teaching and Practice.
320 pages. 2012. ISBN 978-3-0343-1090-1
Vol. 152 Paul Gillaerts, Elizabeth de Groot, Sylvain Dieltjens, Priscilla Heynderickx &
Geert Jacobs (eds)
Researching Discourse in Business Genres. Cases and Corpora.
215 pages. 2012. ISBN 978-3-0343-1092-5
Vol. 160 Carol Berkenkotter, Vijay K. Bhatia & Maurizio Gotti (eds)
Insights into Academic Genres.
468 pages. 2012. ISBN 978-3-0343-1211-0