In R, hetorical Strue(;ure Theory (RST) the detinitions of some rela|,iotts are rather w~gue because they are given on a pragmal, ic 1)asis. 'Fhis pal)er presents an<>\[her way of seeing the. relations which leads to a more precise specification of the relations. 'l'\]m re.la: \[ions are associated with constrailg;S on the semantic relationships between tile proposit, iomd contents of two clauses, t,heir Modality and Tense/Asl>eet.
The lthetorieal Strucl;ure The(,ry (lIST) by Mann and Thompson \[Manu and Thoml)SOn ,1.<)87\] is a t;hcory of int,er-s<ml,eni,ial (or inter:clausal) 'elationships in a text. All;hough RS'\[' is int;ended to serve both as a framework for taxi, analysis and l,ext generation, it has so flu' been used exclusiw~ly in |;exl, generation \[Hovy et al., 1992} \[Linden et al., 1(,)92\] \[Ri;sner and Stede, i992\]. Several resea.rehers recognise that RST has de%ors as an analyti<:al frame.work. Moore and Pollack \[Moore and Pollack, 19921, for example, claim that the assmul)tion of a single relation bel.ween discourse elCIIICtL|,S is Olle <)\[" I,he reasons why lIST analyses are inherently ambiguous. They also claim that the under: s\[)ecilieil,y of the rhel;orieal relal,ion delinil,ions causes problems,
Our elaim is thai, the main cause ()17 I,he di\[liculties of applying liST t;o t;exl; i)rocessing systems is that SOlile of the relat;ions are delined on the basis of l;he elfeets which they have OIL a reader, This is particularly the case for the relations classitied as prcseulatioual rclalion.s, the relal;ions whose intended etfects are to increase some inclination in a reader.
I),ackgrotmd relatiolh for examph;, is defined as a relation whose Satellite increases the al)ility of a reader tO (;()lLll)reh(~lld all elelnelll, in Nucleus and the reader will not tully comprehend Nucleus before reading the t,ext; of Satellite, This delinition is problematic because there are many ways of increasing the ability of a reader to comprehend Nucleus. More seriously, \[,he delinition *Supporte.d by Old Elcctri(: Industry Co., btd. itself does not predict anything about textual forlus of Nucleus and Satellite.
In order to use RST in actual text processing systems, we have to break down st<('.h detinitions to relate them with I,extual forms. In this paper, we show how the defitfitions can be broken down and be ass()eiated wil;h semaul;ic consl;raints betweell eonstituelg,S (clauses), in order t,o relate them with constraints on surface linguisl,ic forms. Among tile 24 rhetorical relations detined in \[Mann and Thompson, 1987\], we focus on presentational relations (7 relations are elassilied as such) which are the most prol)lematic. The re.suits of al)plying our met;hod t<) leading ari;icles in a aal)anesc newspal)er are also discussed.
Basic Framework In RST, 24 relations are divided into two groups: pres<ntational relations and subject matter rchdious. According to Mmm and Thomps(m \[Malta and Thompson, 1987\], subject matter relations are those whose intended effect is that the reader recogldses the relal;ion in question and presentational relations are those whose intended eft'cot is to increase some inclination in the reader. Moore and Pollack \[Moore and Pollack, 1992\] eotllllletll, i;hai; subject matter relations are infof mational and l)resentat,ional relations are intentional.
Table 1 shows what kind of inclination each presentational relation is inI;ended to increase. One can see that tile detinitions are highly abstract and have not,hing to do with the surface realisations of the relations.
On the other hand, it has been observed that there are wu:ious surface cues in texts which are useful \[or ideld;ifying inl;er-senl, enl;ia\[ (or inter-clausal) units. Ilalliday and llasan \[llalli(lay and llasan, 1985\] identitled a set of linguistic devices for linking one part of a text 1,o another, such as reJcrcnce, subslihtlion and ellipsis, conjuncliou, and \[exical cohesion.
From the view point of text processing, these linguist, it devices can be used as cues tor segment.ing a text into structural units (Satellite and Nucleus). However, these cues hardly give any clue about which clause of a unit is Satellite, which clause is Nucleus, and which Table 1: RS relations and their iuclination type
relation ldnd of inclination Background ability of R to comprehend an element 1,hiablement potential ability to perlbrm action iu
N
desire to perforln action in N
belieJ' of N
readiness to accept writer's right in N
positive regard for situation presented
in N
positive regard tbr situation presented
in N Motivation l,;videnee Justify Antithesis Colleessiolt RS relation combines the two clauses into a single unit. For determining these, we have to look for other kinds of surface cues.
Because RS relations are delined pragmatically, their ultimate recognition requires understanding of texts which in turn requires detailed knowledge about the world. Furthermore, the condition that the presentational relations are inherently intcutioual, implies that their recognition requires knowledge about the writer's intention, l)lans, etc Because this kind of information is implicit in texts, its recognition often causes prohlelns.
Ilowew, r, though the writer's intention is implicit, certain linguistic devices give us chtes to infer it. Modality inforntation in a clause, for example, expresses the writer's attitude, toward an event/state described, attd therefore, often gives us clues to recoguise a I{S relation.
Let us consider the following two examples: \[Example 1\] (1) I prepared documents for a meeting. (2) I sent them to the head ol\[ice. \[Example 2\] (1)' 1 am preparing documents tbr a meeting. (2)' 1 have to send them to the head office.
Though these two examples describe pairs of similar events, the relation between (1) and (2) in l~3xample 1 is (temporal) Sequence (a subject matter relation) because they simply describe two events which happened in sequence. On the other hand, in l~3xample 2, (l)' describes an event occurring simultaneously with the utterance, and (2)' concerns what the writer plans to do. While the two events, preparing documents and sending them, may halH)0.n in this sequence, the rela1;ion is not regarded as Sequence but as Background. (2)' gives the reason why the writer is perfomling the action described by (l)'.
This change of I{.S relation occurs due to the difference of ,nodality of (2) and (2)'. Our basic elai,n in that, though they cmmot determine RS relations uniquely, inlbrmation of modality and tense of clauses imposes significant constraints on possible I{S relations, and, being used together with other surface cues like clausal conjunctions, it; Call reasonably restrict a set of possible discourse structures of texts without resorting to detailed knowledge about the world mid the writer's plan.
IIowever, the contribution of modality and tense to the constraints of RS relations is uot straightforward. Both these granunatical \[i~'atures are intertwiued with the propositional content of clauses. There\[ore, in of der to formulate the co~,straints on them properly, we have first to reveal how the intended effects of RS relations can be attained. This leads to our breaking down single RS relations into sets of subschemas, each of which is formulated in terms of the semantic relntionshil)s between propositional eontents of clauses, their modality and temporal relationshil)s.
Like Mann and Thompson, we use clauses as the basic constituents which are related by RS relations, except that clausal subjects and completneuts and restrictive relative clauses are considered parts of their host clause. The constraints which we formulate for each RS relation are exl)ressed in terms of propcrtic,s of clauses. In order to express these eonstra.ints tbrreally, we first introduce the l)asic terms. 3.1 Contents and Modality A clause comprises its Contents and Modality. Modality is the part which expresses the writer's attitude toward the Contents.
While individual languages have their own linguistic devices or grmmnatical forms of modality, what sorts of modality are exl)ressed by such linguistic devices does not vary from one language to another. For example, although the major linguistic device for nlodality are modal auxiliary verbs both in l,'mglish anil in Japanese, some kinds of modality expressed in Japanese by modal auxiliary v('.rt)s are expressed by lexical verbs in English, and vice versa. 1
Furtherntore, we find nla, ny phrasal or quasiq)hrasal expressions which consist of several words, and which collectively express the writer's attitude toward lhe event/state described. In order to treat them, we adopt a semantics-based view tbr the delinition of Modality. That is, we treat expressions which concern the writer's attitude as modal expressions, whichew;r linguistic forms they may take. We \[irst establish a classification schema of Modality based oH semantic considerations (See Section 3.3) and then treat all expressions whose functions can be classified under this schema as modal expressions.
(/<intents <)f a clause is delined as the part which relna.ills aft, el; r(:lllOV0,l of |;be \[llod\[ll ex|)uessiOll. (~olltents COll\[,aiIl exl)ressi(>ns COilcerllillg t.ellSe gild aspect;, which also cont.ril>ute l;o I,he specification of constraints on RS relations. The same discussion as the above can be applied to Tense and Asl)ec|: , so that all expressions whose timer|on is to express tenq)oral aspects of chmses are, regar(llessly of their actual forms, treated in tim same classilication schemas. Tense/Aspect are represented as properties of (k)nl;eni.s (See Section 3.2).
Conl;ents is the tnaitl part of (?l;ntse of which a truth wtlue can be esl, ablished. (1onl,ent.s is characl, erised by l,hree al;l, ril)ut, es: 7!qpc, Time and Quality. (a) Type The truth value of Contents changes according to the time axis. We can think of two time points, 1, and tb, where the Contents C is true during the time interval between t, and t~. l)epending <m the t, eInporal nal;ure, we classify, Conl, elltS into the tbllowing four classes. • Slalic
t, -- undef, tl, - undef, (/(t) = t;rue
(1. < t < / 0 • I)wralivc
1, = de.f, tl, = def, C(I) = t,r,te
(t. < +. < l,,) • Repetitive
1,, = def, tl, =- del!, (/(ti) = true
(t a "< it -( t 2 <~ ..t i < ,.. < I u <~ lb) • No~-rcpetitivc
l,, -- def, tb :- def, CT(ti) :- true
(i = 1; t. < t~ < tb; l,, +- it, ) In the above, l,t/b -= under in SI,al,i(: means that. the i;rul.h wdue of (~olltelll.S does not; change. (b) Time The temporal nature of C, ontenl,s is also classified in terms of the speech time, 7's, as tblh)ws. • HeJbre : 7's < l,, • Simultaneous : t. < J~ < tl, • After : tl, < 7~ We llse the fi.)llowing nol;al;ion 1;o specil~y a temporal relationship hetween two Coul,enl.s (Cl and c~). Ct <2< Cu ... CI occurs before Cu Ct >> (/u '" Ct occurs after C2 ((') (~uality (\]oIll;ellt;S is also class|lied according to whether l;he writer believes it; is good or bad. This classillcw tiou is represented by the attribute Quality (qty) whose vahle is either good or bad. "~
Concerning modality, a number of criteria have been prol>osed. Palmer \[l'ahner, 198(i\] took the same semald;ics-based view of Modality as we discusseJ\[ in Section 3,1, though he hardly extended his analysis to cover l)hrasal <)r quasi--phrasal expressions. We adopt his <:lassitication scltema and modify it. lie class|lied modality into Epistcmic modality and l)c.outic modality. Epistmnic Modality is concerned with language as intbrmation, with the expression of the degree or m~ ture of the writer's commit, ment I;o the truth of whal; s/he says. I)eontic modalii,y is concerned with language as action, mosl;ly wil.h the expression by the. writer of his/her attitude l,owards l>ossil)le ael,iolls hy him~herself and others,
Epistemic modality is class|lied according to l.hc degree of the writer's commil.menl; 1.o I,he I.rul, h of Conl;enl.s, as \[bllows. means that the degre.e of the writer's commil,me:nt o the truth (>f (Jontenl.s C'~; is higher than or equal I.o the degree of the wril,er's commitmenl, to the (.rlll.h of (~ont.enl;s (/.v. l,)vidc nlial ( M-ep,:,,i ) The truth condition of ('(ml,enl;s is based on evidence like sensory evidence or linguistic evidence ConJidenlial (M-ep,:o,J The truth condition of Contents is based on tJm degree of confidence expressed by the writer. lnfercnlial (M-epi,~l) The truth eondil,i(m of (~Olfl,elfl, s is based Oil a I'e;l-soning rnle of the wril.er and inferred from the other \['a (:t.s. A ssu mp tivc (M-cp ....... ) The l;rui,h condition of Cold;cuts is I>ased on some assttml) l.ion. 'Fhe degree of the writer's eolmnitment o the (,ruth 3.3.2 Deontie modality Deontie modality is classified according to the kind of a writer's attitude which s/he expresses. • • • • • Evaluative (M-de+,,a+, M-de ..... ) EvMuative expresses the writer's attitude towards what s/he already accepts as trite in his/her mind. There are two kinds of attitude; positive ('+') and negative ('-'). Volilive (M-de,,ot+, M-de~ot_) Volitive is concerned with a possible action or situation which a writer is hopiug or wishing to occur. There are two kinds of attitude; possible (%') and impossible ('-'). Directive (M-dedir) Directive is concerned with an action which a writer tries to gel others to perform. Though Directive is fiu:ther classified into Permission and Obligation, their distinction is not relevant for our purpose. Commissive (M-de ..... ) Commissive is concerned with an action which a writer commits him/herself to perform or to ellsure that an event takes place. Reques~ (M-der~q) Request is concerned with an action which a writer (:an ask others to do. 3.3.3 Combination of Epistemie and Deontie
modality In l)eontic modality Ewfluative and Volitive are concerned with a writer's attitude toward Contents whic\]t has a truth value. Therefore, clauses with these modalities can also have Epistemic modality. If a clause has any of the other values of Deontie modality like Directive, etc., the Clause has no El)istemie modality as such. floweret, for the sitnplicity of formulation in Seetion 4, we assume their Epistelnic modality value to be Confidential .:3
In this section, we will show how Background, Enablement, Motivation and Evidence of the presentational relations are broken down into subschemas, and give forlnal representations of their constraints. The coltstraints comprise (a) Semantic Relationships between Contents of the
two clauses (b) Constraints on Time A/~ causes a situation change froln b'i to Sj. If a Contents states that Ak causes a situation ,5'j, Si will be omitted. & I = ct~ Clj is held true or acceptable m the environment stated in N. If Clj expresses a situation, this relation is the same as Circumstance. ,5'i ~- CIj Clj is held true or acceptable, if ,5'i is true. If CIj expresses an action caused by El, this relation is the same as Cause aud Result. 5i ~> Aj Si has the possibility to resolve the problem stated in Aj. This relation is the salne as Soluliouhood. ci~cj Ci presents additiouul details about Cj or is inferentially accessible in Cj in one or more ways. This relation is the same as Elaboraliou. (c) Constraints on Modality. (b) and (e) are expressed by using a characterisation of clauses of Section 3. We first show the framework for (a) and then give the actual breakdown of presentational relations. 4.1 Semantic Relations By semantic relationships between Contents we mean the relationships between states/actions/events described by Contents in the extra-linguistic world. 4 As we nee ill Example I and 2, even when two actions seem to stand in the same semantic relationship, they can be used to attain dilferent effects on a reader by adding different expression of a writer's attitude as Modality or putting them in ¢lifferent temporal relationships.
We classify semantic relationships into five categories, four of which also are subject matter relations in RST. That is, if two Contents are presented without any Modality, they stand in the corresponding subject matter relations. We use the following symbols in their definitions.
Cli : Clause i composed of Contents and Modality
C i : Coutents of' Clause i
Si : Contents of Clause i whose Type is Static
Ai : Contents of Clause i whose Type is not Static
Mi : ModMity in Clause i \[Semantic Relations\] pragmatic, ttowever, we adopt a rather narrow definition of the ternt praflmatic and a broad definition of the term semantic. We llSO, tn'agr~atic ollly when it co(lcertls effects Oli. Feadel's or the intention of the writer. The rest, like relationships held in the extra-llnguistic worhl, are called *emantic issues. • • • •
We show breakdowns of follr l;ypical preseid, ai,ional relations inl, o theh: subschenias and stal,e their eonsl;raints lnore |ornially. The subscripl;s of "uu" and "sa" nleans Nucleus alid Sal,ellite, respectiw;ly. 4.2.1 Ba(.kgr(mnd (e) S,<, > S, .... M,, ~ {M-ep,.~d ..... I~,,s'l ...... },
M,,,, ~ {M-ep~,, d .... b,41 ...... } An action in Nuch;us call resolve an uitdesirable situation stated in Sal, ellitc. (I,) &,, << s .... (e) M~<, < {M-cp,,,,il .... Ii,,II ...... } 1. Tiine and st)~ce situal,ions are stated by an action
hi Sai,ellite, ~tlld Ultder l:hese sil, tlal;lons ;ill acti()rl
in Nueleus becomes possit)h,. (e) A,<, >-- d ..... M~. m {M-ep,:vil ..... IinYl ...... }
\[If A,, u beconies possible in the environ.-
lileni; giwm by Asa, then the inodalit,y of A~.
shouhl be more cei't, aiii I,han l,hal; of A,.,.) (c) S~,, 7- A .... M.<,, 6 {M-.cp,,,,i I ...... I/nil ...... } (..) c~<,-~ cD,,, (b) .o (c) M,,a C {M.-¢F,,~,,il ...... Ii,,fl ...... },
M .... ~ {M-el.',:,,i I..... Ii,, s'l ....... }
(Both (~iauses will be understood ;in t, ruc., so
they haw~ to haw'~ l,ruth wdue.) 00 ,% -( .1,~ <, ) ~ ,S'I \['#:) :#'<'(4, A .... ~ ,<>', (b) A~<, << A .... (c) M~,< ~ {M-cv,:,.i .... }
(Asa is ;~,11 event,, which has occurred or is oc-
curring, or a writer is eonfident about the
ew~nl,. A writer inl,ends t,o do A,~u t,o resolve
~t l)robhmi ea~use(l Asa.) 0') ,S'.,,\[qt:/: I><,< 4 t~ ,S',,,,\[qty :~'<'<4 (I,) .%,, << ,s',., 7. An action in Nucleus is caused by a situation in
Satellite. (~) ,%. I A .... (I,) ,%. << el ..... (c) ,%. _>. A ..... M~. < {M-ep~:,,q ..... ti,~/t ...... },
M,~,, C {M-tp<,.,i I...... \[i.fl ....... } 8. Based on *t sitm~t, iou which in caused by an actkm
ill Satellite, a writer's a.t,l, il;ude sl,~ted hi Nueleus
is a.ceepta,lde. (,~) ,%.'..,<, V: '::t .... o,: S.,, l= CZ ..... 0,) no (,:) ,%,, >_ s,,., M.. < {M-cp.,q .... t~,,Sl ..... ),
A4nu C {M-tic ...... i,,ollm,. I...... i,.<:q} (~) Cl~,< l- C/ ..... <)r (:,'l~<, t- c;I .... (i)) no (,:) :VL<, < {M-d<: ..... },
A'I,., ~ {M-de. ..... i~,<>tl,Url ...... I,'~v} 4.2.2 l'\]nal flenmnt 1. Nucleus stalx~s an aetion which will I)e perforined
by a reade, r, and the action becomes possible by
presenting the situation in Satellite. (a) ,%,, l- A .... (b) .%. << A...
(When ,%a is presenl,ed, An. becomes possi-
ble. So, l, he time of ,%. is before ,4...) (e) M,. ¢i {M-cp,.,,i I....... }, M,,. G {M-dc,m.I,.~,~}
(,%. already exists or will exist, so 3',,a has
the possibility to ha, ve tru0~ vahle. If I%, is
l, rue, A .... becomes possible~ ~qo, ,%. should
be more cerl, ain titan A ...... ) 2. Nucleus sI,~i,es an action which will be performed
by a reader, ~md l;he acl;ion becomes possibh~ by
presenting the sit,tmtion which is caused by an ac-
tion in Satellite. (.) .%-~&<,)~ &, & F A .... (b) A,,, << A.., (c) M,. e {M-cv,;,, d .... }, M.,, e {M-dc,u,,I,,,:~/) 4.2.3 Motivation 1. An action stated in Nucleus catlses a good situa-
tion stated in Satellite. It is considered that tile
situation motivates the reader to l)erform the ac-
tion. (a) So-(&,,)~ &,,, ,%,\['~0 :good\],
actor(A ..... Reader) (l/) &,, << ,%,, (~) m.,, e {~-eV~.o,,,~,,~l ..... }, 2. An action stated in Nucleus causes a bad situa-
tion stated in Satellite. It is considered that the
situation motiwttes the reader not to perform the
action.
(a) So-(A,.,)~ S. a, S,,,\[¢,y: b,,al,
actor(d ..... Reader)
(b) A .... << ,%.
(c) M~a d {M-ep~o,qinjl ...... },
M,., G {M~de~.~.looqai,. I ... . l,'~q} 3. Satellite states some attributive information re-
lated to an action in Nucleus, and the information
may be desirable for Reader. (a) &,,\[qO : qoo,q ~ A .... (b) no (c) Ms. ~ {M-ev,:< .... I~,,Vl ....... } 4.2.4 Eviden('e (a) c.., I- c,., (h) c., < c;,,, (~) M,,, e {M-ev.,d, M.,, e {rv~-eV~o,,>,~} 5 Examples We, will show au example of a text structure analysis. Figure 1 shows a sample text from a leading article in a Japanese newsl)a.t)er '5~ and 'l)able 2 shows the attributes of each sentence. The discourse structure of the sample text is shown in Figure 2.
In this example, the following relations are analysed as l/resentational relations. The mm~ber attached to a relat;ion name shows the sul)schema number of the relation. • Background(8) between '1-2' and '3' Sentence 3 has Evaluative modality al)out the situation '3' (economic crisis) and ii; is based on the situation of 'l--2' (drop of dollar). These satisfy the eonstraints of the 8th subschema of Background.
(A world-wide confusion in stock market afDcted
the foreign exchange rate,) 2 g~'b'~Nl'~Jfi,~ 1 \]',1 , (and exchange rates tbr the
dollar dropped sharply.) regarded that the aspect of eeonomic crisis has been worsened.)
@72 ~ ~ o (Disagreenmnt of iuternatioital policy is
making the market unsettle.) 5 t<,t/~llJj~j~-,5/c.g)~_ (to protect tile dollar) is high time that US should show a resolute attitude.)
lar is the key currency,) 8 'g U~t~q~ ~d:tct;~'. (it' US comes to a dead-
lock,) 9 ,~,D~-at~\[~. ~*~E~o (a bad economical influence.
afDets all the world.) b/)~ 8/3~l"aVClg~ " &/~Cb a° (Other countries should re-solidify their cooperatiou taking this point into consideration.)
(@1.987 Asahi Shimbun) Figure 1: Sample text
Modality M-epe~i M-ep~,,i M-de,t,,,M-ep(.o,~ M-epco, ~ M-ep~o, M-de,u,. M-ep~o~ M-epas m M-ep~.o,~ M-der,. q • Background(6) between '4' and '5-6' Tim semantic relation is that a bad situation in sentence '4' (unsettle market) will be resolved by pertbrming an actiou in '5-6' (show a resohtl, eattitude). Sentence '5-6' has 1)irective modality. These satisfy the consl;raints of the 6th subschema of llaekground. • Background(g) between '7' and '8-.9' The situation '7' (dollar is a key) is held true, so Contents '8-9' (effect of bad inlhmnce) is true. These satisfy the constraints of the 7th subsche.ma of llackgrou,M. No. 1 2 3 4 5 6 7 8 9 10 Table 2: Attributes of sanq)le selltences
Type Time l)urative Belbre Non-repetitive Belbre Static Simult. Static Simult. Durative After Non-repetitive After Static Simult. Non-repetitive After Durative After I)urative After
1-10 (-~, II~i"kg"°und l-3 4-ill defiuing all t.he relations, wc will apply tlmm Io amdyse a. full range of I;exl;.
7-10 M ol.ival,il//n
(--_.~ 8-9 References Ilalliday, M. A. K. and llasCm, IL (1985) La,g~u~flc.
(:onlezl~ and Ic,~:l: aspccls oJ langu.agc in a social-
scmiolic perspective. Ox\[ord /Jniw!rsity Press. Ih)vy, E., l,avid, J., Maier, E., Mil,l,al, V., and Paris,
C. (1(.)92). Employing knowledge resom'ces in a new
l;~xl, lelamu!r architecl, m:c In Proc. of 61h \]nlc~n.a-
lio~lal Workshop on NLC, pp. 57 72. I,imlen, IC V., (hunming, S., and Mart.m, .l. (19.92).
Using syst, em m~l, works t.o huihl rhetorical st, rucLLn,'es
In Proc. of 61h \[nlcrnali+mal Worl:shop o~t NL(/, pp.
183 198. Mmm, W. C. ~md Thompsoli, S. A. (1.987). Hhctori
cal si, rucl;ure l,hcory: A theory o\[ text organizal.ion
USC/ISI Iteprint .Series I{S-87-190. Moore, .1. 1). and I'ollack, M. E. (1992). A probhm,
\[br I/ST: The need Ibr lmdt, Mcwq disccmrs¢~ nlmlysis.
(/o'mpulalio, al \],i,guisli~s, 18(4), pp. 5:17 544. l~ahncr, F. R. (\[986). Mood and modMily. (:aml~ridge
University Press. I{.i;sner, I). and St,ede, M. (1992) ('usl.o,,fizing I/ST
lot the ~tltOlll~tl;ic productio,, of I;el:ludc;d IllillllHllS
In Proc= of 6lh hltcrualional Workshop o, NL(/, 1)17 .
199 214
1-2 3 4-6 Non- \] Ilaclc~rotln(\]\[ voliti(\]nM ~ 5-6 Figure 2: I)iscourse structure of the Saml)h~ text * MotiwLt, ion(2) bel.wee,i '7-9' and ' 10' ,¢'Jelltctlce '7-9' states a bad situation (ell'eel of bad inlluence), aml the acl;ioI/ ill '10' (re-solidi\[}/ their coopcral, ion) has l,hc lmssibilil.y t,o clmnge Ihc situation. The writer iv requcst.ing the other countries to take this actitm. These sa£isl'y the coustraints of the 2ml sul~schema of Motiwltiou. • 13aclcgromM(9) I,ctwcen '1-3' and '4-10' The request in '4-t0' (re-solidify their cooperi~lion) iv based on the .}tldgcllleltl; eft'l-3' (~ wi'il;er's ewdual.ion o15 I, he ecotloluic crisis). T\[lcse satisl~z the const,rainl,s of the 91,h subschcn,a o1" \[lackgr<mnd. 6 In this paper, we ln:OpOSC further a breakdown of the im~sentational relations in liST into their subschenms. The subschenias represeut strategies by which two states/actions/events which stand in certaill senmn tic relationships cau be used t.cJ attniu inte,ldcd eFli~cts on readers By associating the definitions of the relations with formally stated const\]'ainl,s, these subschemas help hUllUtll analysts to recognise them in texts, and thus improw: lIST as an analytical tool, Moreover, because characterisation o\[ clatlSeS ill So(> lion 3, especially Modality and Tense/Aspect, are accompanied hy their actual linguistic rcalisal;ions, some parts of the consl, raints staled ill SccLion 4 can readily I)e associated with texl;ual \['orms and be used \[or \[,ext processing systems. Although constraints on semantic relM;ions bel;wecn (~ont,ents can only be evahml, ed by refin'ence to a knowledge base, we expect that, ewm without constraints on semailtic rel~d, ions, the other constraints can be used I;o restrict a set of l>ossible inter-clausal sl,rltcl,/lres of texts. Conclusion
We imve detlncd four presenLat.ioual relat, ions in RS'I? lnore formally a.nd analysed a sample I;exl using these definitions. But the delinitions do not cover all the r<> lations iu ItST ~l, ld \[lil, vc IlOl, bcA;ll widely tested. After