Multiple Discourse Relations 
on the Sentential Level in Japanese 
Yoshiki Mori* 
Department of Computational Linguistics 
University of the Saarland 
Postfach 151150 
D-66041 Saarbrflcken, Germany 
e-mail: mori~coli.uni-sb.de 
Abstract 
In the spoken language machine trans- 
lation project Verbmobil, the seman- 
tic formalism Language for Underspec- 
ified Discourse representation structures 
(LUD) is used. LUD describes a num- 
ber of DRSs and allows for underspec- 
ification of scopal ambiguities. Dealing 
with Japanese-to-English translation be- 
sides German-to-English poses challeng- 
ing problems. In this paper, a treatment 
of multiple discourse relation construc- 
tions on the sentential level is discussed. 
These are common in Japanese but cause 
a problem for the formalism. It is shown 
that the underspecification is to be rep- 
resented for them, too. Additionally, it 
is possible to state a semantic constraint 
on the resolution of multiple discourse 
relations which seems to prevail over the 
syntactic c-command constraint. 
1 Introduction 
In the Verbmobil project, a spoken language 
machine translation system is being developed. 
Its dialogue domain is restricted to appointment 
scheduling. For the semantic analysis, a version 
of Discourse Representation Theory is used which 
can express underspecification and take composi- 
tionality into account. The semantic construction 
is represented by LUD, Language for Underspec- 
ified Discourse Representation Structures (Bos et 
al., 1996), which takes discourse representation 
This research was funded by the German Federal 
Ministry of Education, Science, Research, and Tech- 
nology (BMBF) under grant number 01 IV 101 R. 
A big bunch of thanks goes to Johan Bos, Bj6rn 
Gamb£ck, Claire Gardent, Christian Lieske, Manfred 
Pinkal and Karsten Worm for their valuable com- 
ments, and to Feiyu Xu and Julia Heine for a kind 
help editing the text. 
structures (henceforth DRSs) as its object lan- 
guage. 
The main focus of the project is on translation 
from German to English, but it also treats that 
from Japanese to English. As for the semantic 
construction, it is aimed at that semantic analy- 
ses of Japanese as well as German should be done 
in the same formalism, which is especially chal- 
lenging, taking differences of the two languages 
into account: compared to languages like German 
and English, peculiarities of Japanese such as the 
absence of definite articles seem to invite common 
semantic analyses based on underspecification. 
For example, in the current LUD-formalism it is 
assumed that a discourse relation has the widest 
scope among the scope-taking elements in a sen- 
tence except for the scope of sentence mood. Thus 
LUD allows for only one discourse relation in 
each sentence. Discourse relations contain not 
only such relations expressed by subordinate con- 
junctions as explanation relations (because), ad- 
verse relations (though) and temporal relations 
(before, after etc.), but also purpose, conditional 
and topic-comment relations. We interpret them 
as relations between two DRSs, consisting of re- 
striction (the antecedent part) and scope (the con- 
clusion part). 
In Japanese, it is possible and even common to 
use a number of discourse relations in one sen- 
tence. Lexical entries which realize discourse rela- 
tions occur in various grammatical positions. Dis- 
course relation elements can be also classified ac- 
cording to the anaphoricity of the elements ex- 
pressing the antecedent part and those expressing 
the conclusion part. In Fig. 1 an explanation re- 
lation in the subordinate conjunction and another 
one in the modality auxiliary are used together 
with a topic relation. 
For this case, the current treatment of LUD im- 
plies that the widest scope should be assigned to 
any discourse relation. This extension of the for- 
malism poses a serious problem: every discourse 
788 
getsuyoubi-wa seminaa-ga haitte 
iru-node zikan-ga na-i noda 
monday-top seminar-nora insert 
asp-pres-conj time-nora fail-prcs aux-pres 
Monday (isn't good) because I don't have any time, 
since some seminars have been inserted (then) 
Figure h Three discourse relations in a sentence 
relation introduces a partition into the antecedent 
and the conclusion part for the sentence in which 
it occurs. 1 
If there are a number of discourse relation ele- 
ments contained in a sentence, the partitions they 
introduce can differ from each other (see Sec. 2). 
While scopal relations of quantifiers normally can 
be aligned, scopal relations can, but do not have 
to be built between discourse relations, and be- 
tween scope-taking elements in general. Seman- 
tically, this is one of the main reasons that un- 
derspecification should be introduced rigorously. 
Nevertheless, some regular scopal relations may 
be found among discourse relations (and again in 
general among scope-taking elements). These re- 
lations are determined not only syntactically, but 
also by way of semantics and discourse structure. 
The paper outlines a treatment of multiple dis- 
course relations on the sentential level in two as- 
pects. First, it proposes an underspccified treat- 
ment also for these cases along the lines of quanti- 
tiers and other operators. Secondly, it suggests 
some typical orders in which the scopal under- 
specification among discourse relations can be re- 
solved. The paper is organized in tile following 
way. In Section 2, multiple discourse phenomena 
arc presented in terms of an example. In Section 3, 
tile formalism of LUD is introduced. In Section 4, 
a representation for multiple discourse relations 
is proposed. Section 5 discusses possible resolu- 
tions, in which a relationship between semantics 
and discourse structure plays an important role. 
2 Discourse Relations in Japanese 
As mentioned above, it is apparent in Japanese 
that a sentence can include a number of discourse 
relation elements (Fig. 1). Keeping track of the as- 
sumption that all discourse relations in a sentence 
take a wider scope than the other scope-taking el- 
ements in a sentence, we are confronted with the 
l Since the Verbmobil project deals with spoken 
languages, the unit treated is in reality not a sen- 
tence but an utterance which constitutes a turn in 
a dialogue and includes ellipsis mM other typical phe- 
nomena which need special treatments. Here, how- 
ever, the linguistically abstract unit of sentence will 
be presupposed. 
next question which kind of relative scope holds 
among discourse relations. The treatment of dis- 
course relations should thus be modified at least 
in these respects. 
A discourse relation is represented in LUD as a 
predicate with three arguments; the first one is a 
term for the type of the concerning discourse re- 
lation, the second one is an underspecified scope 
domain of the antecedent part, and the last one is 
another underspecified scope domain for the con- 
clusion part. An underspecified scope domain is 
represented by a hole. 
In Japanese sentences, discourse relations occur 
in various grammatical positions. The sentence in 
Fig. 1 contains at least three different discourse 
relations. First, there is a topic relation which is 
expressed by a so-called topic phrase marked by 
wa. It is encoded in tile LUD as in (1) (of. Asher's 
elaboration relation (Asher, 1993)). In Japanese, 
the antecedent part can be syntactically deter- 
mined, so far as the topic phrase is expressed with 
the topic marker. In Fig. 1, getsuyoubi amounts 
to this part. 
(1) 
(2) 
(3) 
12-discrel(topic,hl,h2) 
14-discrel(explanation-noda,hS,h6) 
13-discrel(explanation-node,h3,h4) 
Fig. 1 also contains a discourse relation ex- 
pressed by the auxiliary noda in the modality po- 
sition of the verbal complex of the conclusion part 
of the sentence. Semantically, it is an subordinate 
relation of explanation. It consists of a functional 
noun for the senteutial nominalization no and the 
copula. The use of noda is dif\[crent from the nor- 
real use of the copula in that it takes a temporal- 
ized sentence as a complement and, at the same 
time, lacks the argument of the copular predica- 
tion. It is this lacking argument which makes up 
the conclusion part of the discourse relation (h6 
in (2)). h5 will be bound to a DRS which is con- 
structed out of the sentence subordinated to noda, 
that is, the whole sentence. 
Finally, a discourse relation expressed by a sub- 
ordinate conjunction node can be found in Fig. 1, 
too (3). This form can be seen as a participle form 
(re-form) of noda mentioned above. Semantically, 
the meaning is restricted to explanation. There- 
fore, the term for the discourse relation type is 
basically the same as (2). 
Even taking these pieces of information into ac- 
count, the scope relations both between wa and 
uoda and between wa and node seem to be un- 
derspecified, whereas noda always has scope over 
node. Since every discourse relation has two scope 
789 
domains, this observation leads to the following 
possibilities of scopal relations for Fig. 1.2 These 
scopal relations are at least theoretically able to 
be forced onto the sentence in Fig. 1 (see See. 5). 
(4) wa (monday ,noda (node (h3 ,h4) , anaphoric) ) 
(5) noda (wa(monday,node (h3 ,h4) ) , anaphoric) 
(6) noda(node (wa(monday,h2) ,h4) ), anaphoric) 
(7) noda (node (h3, wa (monday ,h2) ) ), anaphoric) 
3 Theoretical Framework: 
DRT and LUD 
Since the Verbmobil domain is spoken dialogues 
rather than isolated sentences, it is natural to 
choose a variant of Discourse Representation The- 
ory, DRT (Kamp and Reyle, 1993), as the frame- 
work of its semantic formalism. To treat scope 
ambiguities and other underspecification phenom- 
ena adequately, we have, however, needed to ex- 
tend the formalism to one which suits for repre- 
senting underspecified structures (Bos, 1995). As 
further described in (Bos et al., 1996), LUD is 
a declarative description language for underspeei- 
fled DRSs. The basic idea is that natural language 
expressions are not directly translated into DRSs, 
but into a representation that describes a number 
of DRSs. It is different from UDRS (Reyle, 1993) 
in that not only DRSs, but all predicates and dis- 
course markers are labeled. Moreover, holes for 
scope domains are discerned from other labels. 
A LUD-representation U is a triple U =< 
Hu, Lu, Cv >, where Hu is a set of holes (vari- 
ables over labels), Lu is a set of labeled conditions, 
and C~r a set of constraints. Holes are special la- 
bels for the slot of an operator's scope domain. A 
hole will be bound by means of a plugging func- 
tion to a standard label which stands for a DRS 
of a certain element. 
The set of constraints is divided into alfa con- 
ditions and leq (less-or-equal) conditions, alfa 
conditions define presuppositions and anaphorie 
relations. They stipulate relations of those DRSs 
which do not come into scope relations to those 
DRSs which do. leq conditions, on the other 
hand, define partial order constraints between 
holes and labels which give a semi-lattice struc- 
ture on Hv UCcr with a hole at the top (top 
hole). They should be maintained in the defini- 
tion of a consistent subordination relation. The 
latter, called a possible plugging, fully specifies 
2In tl~s example, each discourse relation element 
is taken as a predicate with the antecedent and the 
conclusion part as its arguments. 
the relations of holes to labels by way of an injec- 
tive plugging function from holes to labels, which 
determines which hole is instantiat'ed into by (or 
is bound to) which label. The interpretation of 
a possible plugging at the top hole is the inter- 
pretation of the matrix DRS. In this way, a LUD- 
representation describes a set of possible pluggings 
at once. 
There are two main exceptions to this charac- 
terization of LUD. First, modifiers share its in- 
stance with the modified DRS and show no differ- 
ent scopal behavior. Secondly, DRSs for discourse 
relations are assumed to always instantiate into 
the top hole. In the current version, the top hole 
is simply assumed to be the hole argument of the 
sentence mood predicate of the main clause. 
4 Representations for multiple 
discourse relations 
In the Verbmobil semantic construction, Japanese 
dialogues are analysed within the same theoreti- 
cal framework and with largely identical semantic 
macros as German ones. In order to apply the 
theory and implementation of LUD to Japanese, 
some modifications are needed. As for discourse 
relations, a major source of complication comes 
from the assumption that predicates for discourse 
relations have two holes as their arguments. The 
first problem lies in the fact that everything that 
goes into a leq relation to one hole cannot be in 
a leq relation to the other hole of the same dis- 
course relation predicate because of its partition- 
ing character. Another problem is the treatment 
of multiple occurrences of discourse relations in a 
sentence. We will be concentrated on the latter 
problem in the following sections. 
For the problem of processing multiple dis- 
course dependencies there are a few approaches 
(Mann et al., 1992; Kurohashi and Nagao, 1994). 
(Gardent, 1994) uses Tree Inserting Grammar 
based on the feature-based Tree Adjoining Gram- 
mar (Vijay-Shanker and Joshi, 1988) to develop 
a formal theory about a discourse semantic repre- 
sentation. This paper is distinguished from these 
works in two perspectives: First, it concentrates 
on the sentential level and offers a treatment of 
multiple discourse relations in terms of a formal- 
ism for underspecified structures of DRSs. Sec- 
ondly, it does not concern multi-functions of one 
discourse relation element, but multiple occur- 
rences of various discourse relation elements. 
As suggested above, discourse relation elements 
have the following characteristic in LUD. The two 
holes which are contained in each of them parti- 
tion the sentence in which the element occurs into 
790 
two parts, whereas it will be subordinated to an- 
other hole by way of a leq constraint as a "unit". 
This has lead to the decision that a discourse re- 
lation element should be directly subordinated to 
the top hole. Other labels for DRSs should be 
subordinated to the discourse relation element in 
the way in which each of them is unambiguously 
subordinated to one of its two holes. The first 
problem mentioned at the beginning of this sec- 
tion can be dealt with in this manner if only one 
discourse relation element occurs in a sentence. 
At least two problems remain when there are 
a number of discourse relation elements in a sen- 
tence. First, if we keep the solution above, dis- 
course relation elements in the sentence are all 
candidates for the directly subordinated position 
to the top hole in a semi-lattice structure. Sec- 
ondly, each discourse relation element introduces 
a different partition of the given sentence. 
For a general solution, the paper proposes a de- 
vice to introduce a special kind of predicate mode 
which has a hole as the only argument for the bot- 
tom of a lattice structure which is built by the top 
hole and discourse relation elements. This enables 
us to keep the decision, on the one hand, that dis- 
course relation elements are in a next-to-top posi- 
tion in a possible plugging and to keep DRSs for 
other parts of the sentence underneath the mode 
predicate, on the other. Every discourse relation 
is situated above any other scope-taking element. 
This proposal crucially relies on the fact that for 
every discourse relation element which occurs in 
a sentence, one of its two holes can be plugged 
by a DRS in a lexically determined way. Addi- 
tionally, it is assumed that we have a syntactic 
strategy in which the topic phrase is dealt with 
as an adjunct modification which should be in- 
terpreted in the discourse structure with respect 
to the main predicate of a sentence. Therefore, 
what is subordinated to the hole introduced by 
the mode predicate amounts to the matrix clause 
of the given sentence. In this way, an ordinary 
underspecification treatment of multiple discourse 
relations among each other gets possible. 
For the sentence in Fig. 1, the LUD- 
representation can be implemented like in 
(8). Labels are represented under lud preds. 
lud_grouping and lud_mota show among others 
which labels are to be treated together to con- 
struct DRSs. Under lud_scoping, alfa and loq 
conditions are found. The labels 112 and 113 are 
presuppositions of 18 and 111. leq relations read 
that labels are always less or equal to labels in the 
given order. Fig. 2 is a graphical representation 
of the loq constraints of (8). Discourse relations 
and discourse markers are abbreviated to discrol 
and din, respectively. 
(8) index: (i8,118,h0) 
lud_preds: ll-mood (deal ,hO) 
12-discrel (topic, hl, h2) 
13-discrel (node ,h3 ,h4) 
14-discrel (noda ,h5 ,h6) 
16-dm(iD 
17-predicate (get suyoub i, i I ) 
19-dm(i2) 
1 lO-predicate (halite, i2) 
11 O-role ( i2, art3, i3) 
lll-role(i2,tloc,i4) 
112-dm(i5) 
114-dm(i6) 
115-predicat e (seminaa, i6) 
116-dm(i~) 
117-mo de (h7) 
119-dm(iS) 
120-predicat e (zikan, i8) 
122-dm(i9) 
ll3-neg (i9,h8) 
lud_grouping : 15-inc ( \[16,17\] ) 
18-inc (\[19,110\]) 
113-inc ( \[114,115\] ) 
ll8-inc ( \[119,120\] ) 
121-inc ( \[122,123\] ) 
lud_meta: modifies(18,111) 
lud_scoping : alfa(i6,udef, 18,113) 
alfa(iS,pron,lll,112) 
leq(12 ,hO) 
leq(13 ,hO) 
leq(14,hO) 
leq(15,hl) 
leq(18,h3) 
leq(ll6,h6) 
leq(llZ,h2) 
leq(117,h4) 
leq(llT,h5) 
leq(ll8,hT) 
leq(118,h8) 
leq(121 ,h7) 
The mode predicate can be seen as a secondary 
sentence mood predicate. For example, it serves 
in a similar way to the predicate used for the intro- 
duction of a propositional complement of propo- 
sitional attitude verbs. This kind of use of the 
mode predicate does not seem to be restricted to 
discourse relations. For example, multiple oc- 
curences of modal expressions show a concerted 
behavior as regards scopal relations as in "we can 
perhaps meet there". The mode predicate is ap- 
plicable when multiple occurrences of predicates 
in one semantic class take a scope over any other 
scope-taking elements together but the scope re- 
lations among each other are underspecified. 
5 Possible Resolutions 
It is sometimes possible to resolve scopal un- 
derspecifications of discourse relations on several 
grounds. Actually, there seems to be only one 
791 
12-discrel(wa, hi, h2) 13-discrel(node,h3,h4 14-discrel(noda,hS,h6) 
15-inc\[16,17\] ~06) ........ / 1-~7_mLoderh7 ' unbound 
16-dm(il) 115-pred(~emin~,i6) / H/-: ~ 
17-pred (getsugoobi ,il) - k, / 
/ 112-inc\[122,123\] alfa(i6,udef,18,113) '~ / 
ll2-dm(i5) " ' " ', / 122-dm(t9) 
"". al fa(i5,pron,111,112) '/ 123.:.aeg(i9,h8) 
"'- 18- inc\[19,110\] 
111-role(i2,tloc,i4) ~ 19 -dm 02) 118-inc\[119,120\] 
modifies ll0 -pred(haitte, i2) ll9-dm(i8) 
110 - role(i2,arg3,i3) 120-pred(zikan,i8) 
Figure 2: A graphical representation of the sentence in Figure 1 
plausible resolution possibility for the sentence of 
Fig. 1. This resolution possibility corresponds to 
(5). The plugging function for this case is as fol- 
lows (9). It should be read such that a label is 
bound to (plugged into) a hole. 
(9) plug_into (14,hO) 
plug_into (12,h5) plug_into (13,h2) 
Confinement of resolution possibilities depends 
on various factors. One of the most important fac- 
tors is lexical determination of the scope domains 
of the antecedent part or the conclusion part of a 
discourse relation. Especially when one of the two 
is determined as anaphoric, that is, sentence ex- 
ternal, the scope of this discourse relation seems 
to be wider than the others, noda in Fig. 1 is an 
example for this. In the same vein, the scope of 
noda supercedes that of a conditional discourse re- 
lation nara in Fig. 3. The latter's scope domains 
of the antecedent as well as the conclusion part 
are sentence internal. 
gogo-nara yamada-ga i-ru noda 
afternoon-cond PN-nom be-pres aux-pres 
(If you mean) the afternoon, Yamada will be here 
Figure 3: Discourse relations with and without 
anaphoric force 
Among discourse relations with sentence exter- 
nal anaphoric binding there are two types: those 
whose antecedent part is bound sentence exter- 
nally and those whose conclusion part is bound 
sentence externally. Discourse relation parti- 
cles like dakara (therefore) belong to the former 
(Fig. 4), subordinate explanation relations like 
noda belong to the latter. 
dakara getsuyoubi-de daijoubu-des-u 
therefore monday-oblwith okay-cop-pres 
(I) am therefore ready for monday 
Figure 4: A relation with auaphoric antecedent 
Though the semantics of so-called topic phrases 
marked by wa goes beyond the scope of this paper, 
we assume that their discourse relations belongs to 
those whose antecedent part and conclusion part 
are both plugged sentence internally. This pre- 
dicts a narrower scope than that of the subordi- 
nate relation noda. This not only corresponds to 
the intuition in (9), but is also the case in Fig. 5. 
gogo-wa yamada-ga i-ru noda 
afternoon-top PN-nom be-pres aux-pres 
(as \]or) the afternoon, Yamada will be here 
Figure 5: A topic relation getting narrow scope 
792 
On the other hand, scope underspecification 
among discourse relations cannot be disam- 
biguated straightforwardly if they are of the same 
type according to the above classification. They 
can all be of the type whose antecedent and con- 
clusion part are both bound sentence internally. 
In this case, the resolution seems to depend on the 
syntactic c-command information. This explains 
the stipulated scope relation between the topic wa 
and the explanative node in (9). (In (9), the scope 
relation is also influenced by antecedent resolution 
of the temporal-local modification which is needed 
from the syntactic information.) The same ex- 
planation holds for the scope difference which is 
observable between the two sentences in Fig. 6. 
getsuyoubi-wa gogo-nara daijoubu-da 
monday-top afternoon-cond okay-coppres 
As \]or Monday, it is ok i\] it is in the a\[ternoon 
gogo-nara getsuyoubi-wa daijoubu-da 
aftcrnoon-cond monday-top okay-cop-prcs 
I\] it is in the afternoon, the Monday is okay 
Figure 6: Topic and conditional relations 
Discourse relations can, in contrast, all be of the 
type whose antecedent part or conclusion part is 
bound sentence externally. This can be observed 
in Fig. 7. Not only the syntactic modality auxil- 
iary noda, but also the discourse particle dakara 
includes a part which is bound sentence externally. 
To the extent that the c-command relation is un- 
clear between them, the resolution remains un- 
clear here. 
dakara ike-na-i nodes-u 
therefore gomid-auxneg-pres aux-pres 
(It is since) (0 could ,,ot go because of it 
Figure 7: Two relations with anaphoric force 
6 Conclusions 
The LUD formalism that describes DRSs in an un- 
derspecified way also pertains to dealing with mul- 
tiple discourse relation constructions, which are 
common in Japanese. The problem is to distin- 
guish the discourse relations which take the wide 
scope relative to other scope-taking elements on 
the one hand and to have them underspecified 
among each other, on the other. The solution has 
a general character; several scope-taking elements 
can go into scope relations collectively if they be- 
long to the same semantic class. The scope among 
them is underspecified again. This treatment re- 
flects the fact that each element can introduce a 
different partition of the same sentence. 
We have also stated an interesting semantic con- 
straint on the resolution of multiple discourse re- 
lations which seems to prevail over the syntactic 
c-command constraint: discourse relations should 
be scopally compared with each other on the crite- 
ria whether the restriction (antecedent part) or to 
the scope (conclusion part) of a discourse relation 
has an anaphoric force. 
References 
Nicholas Asher. 1993. Reference to Abstract Ob- 
jects in Discourse. Kluwer, Dordrecht, Holland. 
Johan Bos, Bj6rn GambKck, Christian Lieske, 
Yoshiki Mori, Manfred Pinkal, and Karsten 
Worm. 1996. Compositional semantics in Verb- 
mobil. In Proceedings of the 16th Interna- 
tional Conference on Computational Linguis- 
tics, Koebcnhavn, Denmark, August. ACL. 
Johan Bos. 1995. Predicate logic unplugged. 
In Proceedings of the lOth Amsterdam Collo- 
quium, University of Amsterdam, Amsterdam, 
tIolland. 
Claire Gardent. 1994. Discourse multiple de- 
pendencies. ILLC-report, University of Ams- 
terdam, Amsterdam, Holland. 
Hans Kamp and Uwe Reyle. 1993. From Dis- 
course to Logic: An Introduction to Modelthe- 
oretic Semantics of Natural Language, Formal 
Logic and DRT. Kluwer, Dordrecht, Holland. 
Sadao Kurohashi and Makoto Nagao. 1994. 
Automatic detection of discourse structure by 
checking surface information in sentences. In 
Proceedings of the 15th International Confer- 
ence on Computational Linguistics, pages 1123- 
1127, Kyoto, Japan, August. ACL. 
William C. Mann, Christian M.I.M. Mathiessen, 
and Sandra Thompson. 1992. Rhetorical struc- 
ture theory and text analysis. In W.C. Mann 
and S.A. Thompson, editors, Discourse De- 
scription. John Benjamins, Amsterdam, Hol- 
land. 
Uwe Reyle. 1993. Dealing with ambiguities by un- 
derspecification: Construction, representation 
and deduction. Journal of Semantics, 10:123- 
179. 
K. Vijay-Shanker and A.K. Joshi. 1988. Feature 
structure based tree adjoining grammars. In 
Proceedings of the 12th International Confer- 
ence on Computational Linguistics, volume 2, 
pages 714-719, Budapest, Hungary, August. 
ACL. 
793 
