Giora97 Discourse Coherence

Embed Size (px)

Citation preview

  • 8/12/2019 Giora97 Discourse Coherence

    1/18

    ELSEVIER Journal of Pragmatics 27 (1997) 17-34

    ~mrlmhff

    Discourse coherence and theory of relevance:Stumbling blocks in search of a unified theory

    R a c h e l G i o r a *

    Poetics and Comparative Literature Tel Aviv University Tel Aviv 69978 Israel

    Received January 1995; revised version October 1995

    Abstract

    Contrary to Sperber and Wi lson (1986), I argue that relevance cannot be the only principlethat governs human c ommunicat ion. It can by no means replace current accounts of discoursecoherence (e.g., Grice, 1975; van Dijk, 1977; Giora, 1985a,b, 1988), since it is neither nec-essary nor sufficient for text well-formedness. I show that, although a discourse may be rele-vant to an individual, interacting with her/his set of a ssumptions at a small cost, this discoursemay nevertheless be judged as incoherent by the very same individual. And vice versa: A dis-course may be judged as coherent by an individual and yet be S &W irrelevant to her/him. Ifone of the goals of a pragmatic theory is to account for speakers' intuitions, Sperber andWilson's relevance theory is deficient in this respect. It cannot account for speakers' intu-itions as to coherence and degrees of coherence.

    And to feel that the light is a rabbit-lightIn which everything is meant for youAnd nothing need be explained ...

    The grass is fulland full of yourself. The trees around are for you,The whole of the wideness of night is for you,A self that touches all edges,

    (Wallace Stevens, A rabbit as king of the ghosts)

    I I n t r o d u c t i o n

    There are a n umb er of not ions of re levance prevalent in the l i tera ture: One viewsrelevance as a relation between a set of proposit ions and a discourse-topic (hence-

    I am indebted to Andrew Goatly and Yael Ziv for their critical comments on an earlier version.* E-mail: [email protected]

    0378-2166/96/ 15.00 Copyright 1996 Elsevier Science B.V. All rights reservedSSDI 03782166(95 )00065-8

  • 8/12/2019 Giora97 Discourse Coherence

    2/18

  • 8/12/2019 Giora97 Discourse Coherence

    3/18

    R. Giora / Journal of Pragmatics 27 1997) 17-34 19

    2 . S W ' s n o t i o n o f c o n t e x t

    As a pre l iminary, I would l ike to query the not ion of context proposed by S&W(1986) . S&W's main cont r ibut ion in th i s respec t i s the a l te rna t ive they propose forthe t rad i t iona l v iew o f contex t as g iven pr ior to the ac t of comm unica t ion . Th e keynot ion here is the choic e or se lec t ion of context (s ) . S& W (1986: 137-1 40) assumetha t process ing new inform at ion involves se lec t ing an adequate se t o f backg roundassumpt ions which cons t i tu te the context . For any p iece of new inform at ion manydi ffe ren t se ts of assumpt ions f rom d i ffe ren t sources might be se lec ted as context .Howeve r, t he o rgan iza t ion o f l ong t e rm memory in t e rms o f ' f r ame ' , ' s cena r io ' o r'p ro to type ' o r ien ted ca tegor ies , and the menta l ac t iv i ty the hearer i s cur ren t lyengaged in cons t ra in the choice of context (s ) . What de te rmines the choice of a con-

    tex t is the search for S& W re levance . The context which wi ll mos t e ffe c t ive ly, tha ti s a t the leas t cos t , combine wi th the newly presented assumpt ion to y ie ld maximalcontextual effects , is the appropriate one.

    Howeve r, a cco rd ing to S&W, one does no t come to p roces s ing new in fo rma t ionwi th a 'b lank m ind ' . Rather, one keeps in m in d the assumpt ions s /he has jus t beenprocess ing . At the s ta rt o f each de duct ive process , the mem ory of the deduct ivedevice conta ins an in i t ia l se t o f assumpt ions . This imm edia te ly g iven c ontext can beextended in d i ffe ren t d i rec t ions . In the case of verba l communica t ion , the compre-hender may have to inc lude in the context no t on ly the in te rpre ta t ion of the immedi-a te ly preceding u t te rance , bu t a l so the in te rpre ta t ion o f u t te rances occurr ing ear l ie r in

    the exchange .As m uch as the choice o f context (s ) is an appea l ing idea, the fac t that par t icu larly

    the in it ia l contex t i s no t cons t ra ined i s problemat ic . S ince any th ing can be conta inedin the ini t ia l contex t , and since no part icular or ganiz at ion is requ ired at this s tage, theresu l t o f choos ing the appropr ia te contex t for a new informat ive i tem ( tha t wil l y ie ldcontextua l e ffec ts a t l east cos t ) may be an incoheren t d i scourse .

    An ana lys i s o f S&W's examples b rough t t o bea r on the no t ion o f ' r e l evance incontext ' wil l i l lustrate the point :

    (1) [A ]n assumpt ion i s re levant in a context to the ex ten t tha t i ts contextua l e ffec tsin this context are large.[A]n assumpt ion i s re levant in a context to the ex ten t tha t the e ffor t requi red toproc ess it in this con text is sma ll . (S& W , 1986: 125)

    To ex amine these cond i t i ons fo r r e l evance in a con tex t , S&W examine the amoun tof contextua l impl ica t ions , weighed aga ins t the process ing effor t , a number ofassumptions (e .g. , (3) , (4) , (7) below) recover in a certain context . To do that , theyf i rs t cons t ruc t the context (2) be low (S&W , 1986: 125):

    (2a) Peo ple who are ge t t ing marr ied should consul t a doc tor about poss ib le hered-i tary r isks to their chi ldren.

    (2b) Tw o peop le bo th o f whom have tha l a ssemia shou ld be warned aga ins t hav ingchildren.

  • 8/12/2019 Giora97 Discourse Coherence

    4/18

    20 R. Giora / Journal of Pragmatics 27 1997) 17-34

    (2c) Susan has thalass emia .

    Given the above context , S&W cla im tha t (3) and (4) be low vary in the amount ofcontextual effects they have in this context:

    (3) Susan, wh o has thalassemia, is get t ing married to Bill .(4) Bil l , wh o has thalassemia, is get t ing married to Susan.

    S& W conten d that while both (3) and (4) share the contextual implicat ion in (5) , (4)has an addit ional contextual implicat ion (6):

    (5) Susan and Bil l should consult a doctor about possible hereditary r isks totheir chi ldren.

    (6) Susan and Bil l should be warned against havin g children.

    Were contexts indeed given, this would be an adequate analysis . However, s incecontexts , we are told, are not given, but are rather chosen and determined by thesearch for S& W relevance (S& W, 1986: 141), an al ternat ive analysis is required.Such analysis wil l show that both (3) and (4) are equally S&W relevant .

    To render (3) S&W relevant , the context needed to be searched for is smaller thanthat needed to render (4) S&W relevant . To render (3) S&W relevant , assumption(2b) should not be evoked. (2b) provides information that is of no use, and might

    indeed be a distract ion. Thus, only (2a) and (2c) should be chosen upon processing(3). However, this is not t rue of (4) . On processing (4) , (2b) should be added to thecontext . (4) then, is not more relevant than (3) as S&W tried to show. Though(4) has more contextual implicat ions than (3) , i t is also more effort consuming,necessi tat ing the expansion of the context .

    The same procedure (of extending the context ) may apply to render inappropria tea discourse such as (7) S& W relevant . S& W (1986: 127) claim that (4) is more rel-evant than (7) , s ince they share the same amount of contextual implicat ions in thecontext of (2) , while (7) requires mo re effort , because the extra in form ation in (7) iscomple te ly unre la ted to the g iven context and, con sequent ly, has no contextual e ffec t

    whatsoever :

    (4) Bil l , wh o has thalassemia, is get t ing married to Susan.(7) Bil l , wh o has thalassemia, is get t ing married to Susan, and 1976 was a great

    year for French w ines .

    How ever, s ince contexts are searched for, i t is possible to extend the co ntext so asto render (7) relevant . Suppose the speaker and hearer of (7) have just heard thatthe i r ne ighbor bought them a 1976 bott le of French wine . Ou r ne ighbor bought usa 1976 bottle of French w ine is therefore an accessible assump tion which is con-tained in the ini t ial context and can be effort lessly added to the context . Thisextended context renders (7) equal ly S&W re levant to the context as (4) is . Thoughit requires more processing effort than (4) , i t a lso has more contextual effects . In

  • 8/12/2019 Giora97 Discourse Coherence

    5/18

    R. Giora / Journal of Pragmatics 27 1997) 17-34 21

    sp i te of i ts S& W re levance , (7) i s an incohe rent text . The hearer must b e le f t puzz ledas to how the two proposi t ions in (7) are related to each other (rather than to acontext ) .

    Note tha t the i l l - formedness of (7) has noth ing to do wi th the fac t tha t i t i s madeup of conjo ined u t te rances . The c la im tha t conjunct ions a re presumed to have S&Wre levance ove r and above each o f the con junct s ( s ee B lakemore , 198 7 :1 20) does no thold for jux ta pose d sentences (Cars ton , 1993: 42) . 4 The d iscourse in (7 .1) i s equal lyi l l - formed, though i t i s no t made up of conjo ined sentences :

    (7.1) Bil l , wh o has thalassemia, is get t ing marrie d to Susan. Both he and Susan toldme tha t 1976 was a grea t year for French wines .

    In order to b lock tex ts such as (7) or (7 .1) , S&W have to assume tha t a t l eas t thecontex t - the set o f access ib le assumpt ions - mus t be cons t ra ined by coheren ce con-d i t ions . Thus , ins tead of cons ider ing Relevance re la t ions to a d iscourse- topic in thetex t , S& W wi l l have to assume such re la tedness be tween assumpt ions in the context .This might occas iona l ly a l low for a wel l - formed d iscourse , which should then bev iewed a s t he by -p roduc t o f t he we l l - fo rmedness o f t he con tex t . Howeve r, wh i l ethey admi t such contextua l s t ruc tures (e .g . , in te rms of ' schemas ' o r 'p ro to types ' )only as fa r as long- te rm m em ory i s concerne d (S&W , 1986: 138) , the in i tia l contexti s no t cons idered cons t ra ined by semant ic re la t ions , and r igh t ly so . The resu l t i s ,therefo re, the possibi l i ty of i l l - form ed discourses such as (7) and (7.1) above. 5

    An a l te rna t ive answer to the ques t ion as to how a context i s de te rmined comesf rom Murray (ms . ) . Murray examines what in tu i t ive ly would count as the appropr i -a te context versus what would count as an inappropr ia te context , bu t which wouldnever the less be as h ighly access ib le . She shows tha t what gu ides the choice of theappropr ia te context is what in te r locutors cons ider most ' impo r tan t ' to them - what

    4 TO show that for conjoined, or for that mat ter, for any two juxtaposed ut terances , to have S&W rele-vance over and above each of them individual ly does not guarantee coherence, consider the fo l lowingtext segme nt (d iscussed in Giora , 1985a) . The text i s made up of pai rs of u t terances , each of which has

    S&W relevance over and above each of the individual adjacent u t terances , yet a l together they do notmake up a coherent text :

    (a) Roni t i s never home nowadays because she l ives near school . (b) School , you know, is the centerof the k ids socia l l i fe . (c) Uri has missed school a lo t th is year. (d) He never showed up a t tennis ,either.

    The ad jacen t s egmen t s ( a -b ) , (b -c ) and ( c~ l ) each ach ieve S& W re levance beyond the i r i ndiv idua l u t te r-ances . (a-b) may have 'Ro ni t ' s socia l l i fe i s hect ic ' as a contextual impl ica t ion recovered a t a smal l cost .(b--d) may have Uri ' s socia l li fe i s poo r ' as a contextual impl ica t ion. (c- -d) may have 'Ur i ' s educat ionis def ic ient ' . In Giora (1985a) I argued that coherence is not a l inear proper ty of the text . The exis tenceof coheren ce or even S& W relevance re la t ion betw een adjacent sentences wi l l not necessar i ly resul t ines tabl ishing coherence re la t ion for the whole text ( for a d i fferent v iew see Biakemore , 1988: 241;Jucker, 1993) . Rather, coheren ce re la t ions obta in between a se t of proposi t ions and a governin g

    discourse topic proposi t ion (see Giora , 1985a,b and sect ion 3 below).5 For fur ther d iscuss ion concernin g the s tatus of context conside r Gibb s (1987) and references there in .Gibbs argues that speakers and hearers must coordinate what they mutual ly know. Mutual knowledge isnot only a resul t of comprehension, i .e . , the search for S&W relevance, but a prerequis i te for i t as wel l .

  • 8/12/2019 Giora97 Discourse Coherence

    6/18

    22 R. Giora / Journal of Pragmatics 27 (1997) 17-34

    'mat te rs ' to them, ' in te res t s ' o r ' con cern s ' them. ' Imp or tan ce ' o r ' in te res t ' need notbe cos t e ffec t ive , bu t ra ther Relevant to a d iscourse- topic : [Th e] ques t ion tha t [ thein te r locutors ] care about deeply and in tense ly, even though i t s cogni t ive conse-que nces are small ... is a lmost what e ve ry l ine of their argu men t is relevant to , i t iswhat the whole discourse isabout (ibid.).

    In sum, the search for S&W re levance , for ob ta in ing as grea t a contextua l e ffec tas possible for as small as possible a proces sing eff ort (S& W , 1986: 141 -142 ),ma y not be the only fac tor de te rmining the choice of context. As show n here , i t ma yresul t in render ing a newly in t roduced p iece of in format ion S&W re levant bu t inco-herent . A hearer might have access ib le context (s ) for the process ing of (7) or (7 .1)abov e (which wil l a l low him to obtain as great a contex tual effec t as possible for assmall cost as possible) . H ow eve r, i t seems qu ite safe to con tend that he m ay sti ll f ind(7) and (7.1) undesirable or inappropriate (compared to, e .g. , (4)) .

    The inappropr ia teness , ye t S& W re levance of (7) and (7 .1) sugges ts that , cont ra ryto S&W's assumpt ion , speakers do not have in tu i t ions about S&W re levance , bu tthey do about d iscourse coherence . Fur thermore , such poss ib le d issoc ia t ion be tweencoheren ce and S&W re levance re fu tes S& W 's c la im tha t the not ion of d iscoursecoherence can be d ispensed wi th . S&W's a rgument i s tha t we do not need an inde-pendent no t ion of d iscourse coherence (e .g . , Gr ice ' s d i scourse theory) , s ince every-th ing fo l lows f rom the ques t for S&W re levance . However, the S&W re levance butincohere nce of (7) - (7 .1) sugges ts tha t coherence i s no t a der iva t ive not ion .

    3 . D i s c o u r s e w e l l f o r m e d n e s s

    I would l ike now to fur ther subs tan tia te my c la im tha t S& W 's not ion of re levancecannot be an a l te rna t ive to cur ren t d i scourse theor ies , s ince i t cannot account fo r thewel i - forme dness o f tex ts. Spec i f ica l ly, I wi ll show tha t a s tre tch of d i scourse may beS& W re levant to an ind iv idua l bu t incoherent ( s imi la r ly to the examples ab ove) , andvice versa , i t may be S&W i r re levant to an ind iv idua l bu t coherent . To be ab le toweigh S & W 's re levance theory aga inst an a l te rna t ive theory of d i scourse coherence ,cons ider the ca tegor ia l condi t ions for d i scourse wel l - formedness I formula ted in

    previous work:

    (8) An(a)

    in format ive d iscourse i s wel l - formed i f and on ly i f itconfo rms to the Releva nce R equi rem ent in tha t al l it s p ropos i tions a re conce ived o f as re la ted to a d iscourse- topic propos i t ion . Th e d iscourse- topic i s agenera l iza t ion , preferab ly made expl ic i t , and p laced in the beginning of thediscourse . I t func t ions as a re ference poin t re la t ive to which a l l incomingproposi t ions are assessed and stored (as specif ied in Giora, 1985a,b, fol low-ing Grice, 1975),

    and(b) conforms to the Graded Informat iveness Condi t ion which requi res tha t each

    proposi t ion be more (or at least not less) informative than the one that pre-cedes i t in relat ion to the discourse-topic. A message is considered informa-

  • 8/12/2019 Giora97 Discourse Coherence

    7/18

    R. Giora /Journal of Pragmatics 27 1997) 17-34 23

    tive to the extent that i t has propert ies unshared by the previous proposit ion,which, in turn, al low i t to reduce possibi l i t ies by half (as specif ied in Giora,1988, fo l low ing Grice, 1975, an d along the l ines sugges ted by Shann on,1951 ; Attneave, 1959; am ong others) ,

    and(c) marks any devia t ion f rom Relevance and Graded Informat iveness by an

    explici t mark er e .g. , by the w ay , af ter al l (cf. Giora, 1985a; A riel , 1985,1988).

    The co ndi t ions of d iscourse wel l - formedness are der ived f rom m ore genera l con-s t ra in ts on s torage of genera l know ledge in memory . A wel l - formed informat ive textis organized l ike a Roschian, prototype-oriented category. The internal s tructure of a

    prototyp e-oriented cate gory (e.g. , the class of birds) is graded (cf. Rosch, 1973). I t isstored relat ive to the least informative member in the set , and is ordered informa-t ive ly f rom the leas t to the m ost inform at ive member. Likewise , a w el l - formed infor-mat ive text evolves gradual ly f rom the leas t to the most informat ive message (Giora,1985b, 1988). I t begins with a general izat ion and becom es m ore info rmativ e as thetext proceeds.

    I t has been s how n (e.g., Giora, 1985b) that texts which co nfo rm to the above co n-dit ions are the easiest to process. Any deviat ion requires extra processing and isjudged by speakers as less natural (e .g. , Giora, 1988). Along the l ines suggested byGrice (1975), overt violat ions of the requirements are intended to be recognized and

    tr igger the generat ion of implicatures. They are aimed at achieving special effects orprodu cts and are mo re diffic ult to process (Giora, 1990, 1993). In contra st, unin-tended violat ions const i tute anomalies. For an i l lustrat ion of the graded structure ofan informat ive text, cons ider the fo l lowing d iscourse :

    (9) I t has often occurred in the history of science that an important discovery wascome upon by chance . A sc ient i s t looking in to one mat ter unexpectedly cameupon another which was far more impor tant than the one he was looking in to .Penici l l in is a result of such a discovery.

    The text in (9) begins wi th a genera l iza t ion wh ich prese nts the se t of proper tiesshared by al l the proposit ions in the text: Scientif ic ( i ) , chance ( i i ) , discovery ( i i i )of some impor tance ( iv) . The second proposi t ion shares th is se t but adds anotherproperty - the relat ive impo rtance o f the scientif ic chan ce disco very (v) . The thirdproposi t ion repeats a l l the aforement ioned proper t ies whi le adding another one -the d iscov ery of penic i l l in (vi ). The ment ion of the d iscovery o f penici l l in i s a spe-c i f ic ins tance of the d iscovery of impo r tant sc ient if ic chance d iscover ies. This men-t ion el iminates other al ternat ives that could be included in the category at this point .The text thus proceeds a long the informat iveness axis , f rom the leas t to the mostinformat ive message in the se t , conforming to the Informat iveness Requirement . Atthe same t ime, each of i t s propos i t ions repeats informat ion ment ioned in the d is -course- topic presented in the beginning, thereby conforming to the RelevanceRequirement .

  • 8/12/2019 Giora97 Discourse Coherence

    8/18

    24 R. Giora / Journal of Pragmatics 27 (1997) 17-34

    Below I wi l l examine the explanatory power of both S&W re levance theory andthe theory of discourse coherence proposed above.

    3 .1 . S W r e l evance and The Re l evan ce R equ i r e me n t

    3 .1 .1 . Te xt com prehen s ion i s a d i scourse- top ic o r ien ted ac t iv ityIf i t is the case that the quest for S& W relevance is the onl y principle that governs

    hum an com munica t ion , th is sugges ts tha t be ing S &W re levant i s both suff ic ient andnecessary for d iscourse wel l - formedness . C oherence def ined in te rms of re la tednessto a discourse-topic (e.g., van Dijk, 1977; Giora, 1985a,b and (8a) above), must be aderivat ive notion only.

    To test this , consider again occasions where being coherent confl icts with beingS&W re levant . For ins tance , S&W's re levance theory cannot account for the fac tthat discourse-topics are bet ter s tated than assumed, even when they are entirelyuninformative ( i .e . , are S&W irrelevant , s ince they are present in the context and donot affect the strength of the exist ing assumptions) . Note the fol lowing episode:

    At one point , i t was obvious to my students and to me that I would start my lec-ture stat ing that I was leaving the universi ty soon, and hence would propose a make-up- lec ture t imetable. Ho wever, even tho ugh I knew they knew I was going to d iscussthis topic upon the begin ning of the lecture, that is , even th oug h this topic was h ighlyaccessible to them, I could not (I would suggest because of constraints of discoursewell-formedness) s tar t in medias res. I had to start by stat ing my discourse-topic,

    mark ing it as old information, thoug h (as required by (8c) above): A s you wellknow, I am leaving soon, etc. . Only then could I discuss the detai ls that ensue.Though in te rms of S&W re levance theory th is ment ion of the d iscourse- topic i sS& W irrelevant (cf. (12a) below) - i t wou ld not even function as a remind er, beingfresh in the minds o f my audience - i t was nevertheless n ecessary to establish i t forconsiderat ions of coherence. This , then, is an instance that at tests that S&W rele-vance theo ry is insufficient to account for discourse w ell-formed ness: A discoursemay be S&W irrelevant , but coherent and acceptable.

    The coherence of th is d iscourse , however, can be accounted for in te rms of con-forming to the Relevance Requirement ((8a) above). This condit ion requires that the

    discourse-topic be stated and presented in discourse-ini t ial posi t ion, and then elabo-rated on.

    That discourse-topics are bet ter s tated than not has been established empirical ly.Dool ing and Lac hma n (1971) and Garrod and Sanford (1977) showed tha t provid inginformation about the discourse-topic at the beginning of the text act ivates there lated schem a or genera l know ledge w hich a l lows incom ing informat ion to be in te-grated more easi ly. Giora (1985b) showed that discourses with a discourse-topicproposit ion in ini t ial posi t ion are read signif icantly faster than identical discourseswith discourse-topic mention in f inal posi t ion. Bransford and Johnson (1972)showed that in certain cases, comprehension of texts is impossible without dis-course-topic mention in ini t ial posi t ion (e.g. , in the t i t le) . George et al . (1994) toodemonstrated that discourse comprehension is a discourse-topic oriented act ivi ty.Subjects recal led t i t led discourses bet ter than unti t led ones (see also Bransford and

  • 8/12/2019 Giora97 Discourse Coherence

    9/18

    R. Giora /Journal of Pragmatics 27 1997) 17-34 25

    Johnson, 1972) , and they searched for the d iscourse- topic propos i t ion in the begin-n ing of the tex t . They fur ther found tha t words in the unt i t led paragraphs e l ic i tedgrea ter N400 ampl i tude than words in the t i t l ed paragraphs . N400 i s the componentof event re la ted bra in poten t ia l . I t increases in response to unexpected words in asentence (e .g. , Kutas and Hil lyard, 1982). Here i t was shown to be sensi t ive todisco urse in coh eren ce as well . Absen ce o f a discou rse-top ic (e .g. , a ti t le) affectsN400.

    Fur ther suppor t for the hypothes is tha t d i scourse comprehens ion i s a d i scourse-topic or ien ted ac t iv i ty comes f rom s tudies by Gernsbacher and her assoc ia tes(Gernsbach er, 1985, 1990; G ernsb acher e t a l. , 1989, and Gernsb acher and Harg-reaves , 1988) . Accord ing to Gernsbacher and her co l leagues , the pr imary goa l oflanguage c om preh ens io n i s the bui ld ing of cogni t ive s t ruc tures - a coherent menta lrepresenta t ion of the informat ion be ing processed . Gernsbacher and her co l leaguesshow that bui lding a coherent , mental s t ructure s tar ts with laying a foundat ion fortha t menta l s t ruc ture . Lay ing the foun dat ion , bo th w i th in d iscourses and w i th in sen-tences , i s e ffor t consuming. The f i r s t word , p ic ture , o r sen tence takes longer toprocess than subsequent in format ion . However, f i r s t ment ion i s ra ther advantageous .After comprehension, the f i rs t i tem, in our case, the f i rs t c lause or sentence in a dis-course , i s cons iderably eas ie r to access than subsequent in format ion . This advanta-geous cogni t ive s ta tus a l lows subsequent in format ion to be mapped onto the f i r s tc lause . Indeed , a f te r comprehenders lay the foundat ion , they develop the i r menta ls t ruc tures by mapping subsequent in format ion onto the f i r s t c lause . I f incoming

    informat ion does not cohere , comprehenders sh i f t and in i t ia te a new subs t ruc ture .That d i scourse- topics p lay a c ruc ia l ro le in tex t comprehens ion has a l so been

    shown by Hough (1990) and Schneiderman e t a l . (1992) . Hough demonst ra ted tha tr igh t hemisphere damaged pa t ien ts have more d i ff icu l t ies than o ther pa t ien ts andnormal subjec ts comprehending nar ra t ives whose d iscourse- topic sen tences a resh i f ted to the end o f the nar rat ive . Schn eiderm an and her co l leagues have es tab l ishedtha t the presence of a d iscourse- topic propos i t ion fac i l i t a tes tex t comprehens ion forle f t hem isphere dam aged and non-hem isphere dam aged indiv iduals . Mo reove r, r igh themisphere damaged pa t ien ts do not benef i t f rom i t s presence .

    Schn eiderm an and her co l leagues sugges t tha t th i s def ic i t (and o thers ) 6 of r igh themisphere damaged indiv idua ls does not resu l t f rom an inabi l i ty to render informa-t ion S W relev ant , so to speak. I t has been show n that such individuals can, in fact ,in tegra te newly presented informat ion wi th the i r se t o f ex is t ing assumpt ions . Forexam ple , in re te ll ing s tories, they use inform at ion presented in the nar ra t ive and in te-gra te i t in to the i r own personal exp er ienc e and wor ld know ledge . Schn eiderm an e t a l

    There is evidence fro m aphasia literature which suggests that discourse comprehension depends onthe righ t hemisphere (e.g., Joanette et al.~ 1990; Molloy et al., 1990; Chiarello, 1988). While left hemi-sphere damaged aphasics with intact hemisphere functioning are able to benefit from thematic informa-tion in the discou rse topic (e.g., Engel-Ortelieb, 1981; Cannito et a l., 1986; Huber, 1989), right hemi-sphere damaged individuals are impaired in their ability to u tilize such information(e.g., Brownell et al.,1986; Beeman, 1993; Hough, 1990). They are unable to distinguish between salient and trivial informa-tion, between normal and deviant discourses, or see the main po int or moral of a story.

  • 8/12/2019 Giora97 Discourse Coherence

    10/18

    26 R. Giora / Journal ~fPragmatics 27 1997) 17-34

    prefer to in te rpre t th i s def ic i t as s temming f rom a more genera l impai rment in for-mula t ing macros t ruc ture .

    This d i ffe rence be twe en r igh t brain dama ged and le f t and non-bra in dam aged indi-viduals at tests to the independence of various discourse funct ions. In part icular, thephys io logica l , o r ra ther, the neuro logica l au tonom y of Relev ance to a d iscourse- topicand S W re levanc e suppor ts the v iew pro posed here tha t the func t ions of d i s -course- topics and d iscourse coherence a re not depend ent on or der ived f rom S Wrelevance .

    3 .1.2 . De g re e s o f cohe rence a nd t he Re l e vance r equ i r em en tTo fu r the r examine cohe rence -based ve rsus S W re l evance -based app roaches,

    cons ider S W re levant , bu t incoherent d i scourses (and see aga in (7) , and (7 .1)above) . Exam ples (10) and (11) show that S W 's re levance theory cannot pred ic tthe i l l - formedness of the (a ) vers ions as opposed to the be t te r- formedness of the (b)versions. (10a) is S W r elevan t in one context , i.e ., the one containing the set ofassumpt ions the reader must have abou t Ida ' s husband, s ince i t may y ie ld contextua limplicat ions at a small cost . Nev erthele ss i t is incoh erent (ev en if we ignore the rep-e t i t ions and incons is tency which may affec t coherence) . Al te rna t ive ly, i t i s easy toimagine a complementary tex t (e .g . , (10b)) which cons t i tu tes a minimal pa i r wi th i tand is as S W re levant , bu t mo re coherent :

    (10a)

    (lOb)

    This f i r s t t ime she was marr ied her husband came f rom Montana . He was the

    kind tha t when he was not a lone he would look thought fu l . He was the k indtha t knew tha t in Montana there a re mounta ins and mounta ins have snow onthem. He had not l ived in Montana . He wo uld leave Montana . H e had to marryIda and he was thought fu l . (Taken f romIda by Ger t rude S te in)This f i r s t t ime she was marr ied her husband came f rom Montana . He was thekind who loved to be a lone and thought fu l . He was the k ind who loved moun-ta ins , and wanted to l ive on them. He loved M ontana . But he had to marry Idaand leave Montana .

    The coherence d i ffe rence be tween (10a) and (10b) i s no t accountable in te rms of

    number of contextua l e ffec ts weighed aga ins t p rocess ing effor t . Whi le both d is -courses may be equal ly S W re levant , they never the less d i ffe r dras t ica l ly in te rmsof coheren ce : (10b) i s more co herent than (10a) . Note fur ther, tha t, though (10a) i sthe less coherent vers ion , it can be cons idered a lo t mo re S W re levant than (10b) ,because i t i s mo re poe t ic and c onsequ ent ly has contextua l ( i. e. , poe t ic ) e ffec ts . How-ever, i f a comp rehe nde r f inds (10a) more d i ff icu l t to process than (10b) , the aes the ticeffects (e .g. , s tyl is t ic novel ty) offset the extra effort invested, and turn (10a) and(10b) s imilar in terms of S W relevan ce, but s ti ll dis t inct in terms o f coh erenc e.

    S W re levance , then , cannot account for the cohere nce d i ffe rence be tween (10a)and (10b) . H owe ver, i t i s eas i ly expla inable in te rms of the d iscourse cohere nce the-ory proposed above . (10b) i s more coherent because i t conforms to the RelevanceRequirement ((8a) above) . Unlike (10a) , a l l i ts proposi t ions are related to a general-izat ion that governs the given set . They are interpretable as being about a cer tain dis-

  • 8/12/2019 Giora97 Discourse Coherence

    11/18

    R. Giora / Journal o f Pragma tics 27 1997) 17-34 27

    course- topic , e .g . , 'W hat Ida ' s husban d had to g ive up upon marry ing her ' . 7 (10a)how ever, does not lend i t se l f to such a summ ar iza t ion tha t wi l l subsume a ll o r mostof i t s p ropos i t ions (c f . Giora , 1985b, and Giora and Shen , 1994, on how discourse-topics a re der ived) .

    At th i s s tage , S&W might a rgue tha t process ing a re la t ive ly incoherent tex t i smore e ffor t consu ming than process ing a coheren t one. But th is on i ts own ne ed notrender the tex t less S&W re levant , s ince such ex t ra process ing may be offse t by agrea ter amount of contextua l e ffec ts (e .g . , (10a) ) . The not ion of S&W re levance oni t s own, def ined in te rms of the number of contextua l e ffec ts and the process ingeffor t requi red in obta in ing them, can not then accou nt for d iscourse cohere nce .

    Cons ider, fur ther, anoth er ins tance of coh erenc e d i ffe rence (be twe en (1 la ) and( lb) be low) w hich cannot be accounted for in te rms of S& W re levance . The seg-men t in (1 lb) was judg ed as more c ohere nt or appropr ia te than (1 la ) b y 20 speak-ers . St i l l , both share the same contextual implicat ions and require the same process-ing effort . Ho we ver , while in (1 lb) the last prop osi t ion s tarts a new paragra ph, s incei t introdu ces a new discou rse-top ic ( the c han ce disc ove ry o f penici l l in) , in (I l a) ,which is not segmented, this proposi t ion is not Relevant (cf l (8a) above): i t is notre la ted to the g iven d iscourse- topic ( 'The impor tance of sc ien t i f ic chance d iscover-ies ' ) in t roduced a t the beginning of the paragraph:

    (1 la ) I t has of ten o ccurred in the h is tory of sc ience tha t an im por tan t d i scov ery wascome upon by chance . A sc ien t i s t looking in to one mat te r, unexpected ly came

    upon another which was fa r more impor tan t than the one he was looking in to .Penici l l in is a resul t of such a discovery. Penici l l in was accidental ly discov-ered by Fleming in 1928 . . .

    (1 lb) I t has of ten oc curred in the h is tory o f sc ience tha t an impor tan t d i sco very w ascome upon by chance . A sc ien t i s t looking in to one mat te r, unexpected ly cameupon another which was fa r more impor tan t than the one he was looking in to .Penici l l in is a resul t of such a discovery.

    Penic i l lin was acc iden ta l ly d iscovere d by Flemin g in 1928 ...

    Thus , whi le S& W canno t accou nt for the low er coheren ce of (1 la ) , o r for themot iva t ion for the segmen ta t ion w hich renders i t mo re c oheren t in (1 lb) , these a reexpla inable in te rms of the d iscourse theory proposed here . 'The acc identa l d i scov-ery of penic i ll in ' cons t i tu tes a new discourse- topic , which mot iva tes forma l segmen -tat ion. Though i t shares some propert ies with the discourse-topic of the f i rs t seg-ment , i t never the less adds too much new informat ion to be inc luded in tha t segment(see also Giora, 1983a,b, 1988). Thus, apart f rom the quest for information, there istha t requi rement for re la tedness ( te rmed Relevance to a d iscourse- topic , c f . (8a)above) , w hich cu rbs the search for as grea t a contextua l e ffec t as poss ib le ( see Gr ice ,

    7 Not e hat the discourse topic need not be made explicit in the text. It is enough that it can be gener-ated (c f. (8a) above).

  • 8/12/2019 Giora97 Discourse Coherence

    12/18

    28 R. Giora / Journal of Pragmatics 27 (1997) 17-34

    1975, andGiora , 1988, on how the Relevance Requi rement cons t ra ins the Informa-t iveness Requi rement ) .

    On var ious occas ions , p roponents of S&W re levance theory (e .g . , Blakemore ,1987; Blass , 1990) mad e the c la im tha t the f requent occurr ence of ( re la tive ly) inco-herent bu t S&W re levant d i scourses a t tes t s tha t speakers and hearers a re not neces-sar i ly cons t ra ined by the search for coherence . This c la im i s no t problemat ic for acohe rence -based theo ry. A cohe rence -based theo ry endeavor s t o accoun t fo r andmake expl ic i t speakers ' in tu i t ions as to the wel l - formedness of d i scourses . I t doesnot assume tha t coherence i s the only pr inc ip le tha t governs human communica t ion .

    3.2. S W relevance and The Informativeness Requirement

    Given S&W's t heo ry o f r e l evance , two types o f i n fo rma t ion a r e r ende red S&Wirrelevant:

    (12a) New ly p resented informat ion , which a l ready forms par t o f the context anddoes not a ffec t even the s trength of ex is t ing assumpt ions . Such inform at ion i stherefore en t i re ly uninform at ive and a for t ior i , i r re levant . (S&W , 1986: 121)

    (12b) Ent i re ly new informat ion tha t does not in te rac t wi th ex is t ing assumpt ions toyield contextual effects and wil l only be cost ly upon processing. ( ibid. : 48)

    Note tha t such a quant i ta t ive not ion i s ak in to the Gr icean requi rement for in forma-

    t iveness (for a s imilar view see Gazd ar and Good , 1982; Ziv, 1988; B erg, 1991). Ihave proposed to account for in format iveness in te rms of the probabi l i ty of anassumption (see Giora, 1988, and (8b) above) . An assumption is informative relat iveto the nu mb er o f op t ions i t reduces . The grea te r the num ber o f possib i li t ies reduce d ,the mo re improb able and hence , in form at ive the assumpt ion . To be cons idered infor-mative, an assumption must at least reduce opt ions by half . This suggests that twoassumpt ions , which might have the same number of (un)cos t ly contextua l e ffec ts ,may never the less d i ffe r in amount of in format iveness . Var ious contextua l impl ica-t ions may reduce a different amount of possibi l i t ies . As an i l lustrat ion, consider(13a) and i ts counterpart (13b):

    (13a)

    (13b)

    She was gay exac t ly the same way. She was ne ver t i red of be ing gay tha t way.She had learned many l i t t l e ways to use in be ing gay. Very many were te l l ingabout us ing o ther ways in be ing gay. She was gay enough, she was a lways g ayexact ly the same way, she was always learning things to use in being gay, shewas tel l ing about using other ways in being gay, she was tel l ing about learningother ways in be ing gay, she was learn ing o ther ways in be ing gay, she w ouldbe us ing o ther ways in be ing gay, she would a lways be gay in the same way,when Georgine Skeene was there not so long each day as when GeorgineSkeene was away. (Taken f rom Miss Furr and Miss Skeene by Ger t rude S te in)She thought she would a lways be gay in the same way, when Georgine Skeenewas there as wel l as when Georgine Skeene was away. But when GeorgineSkeene was away for as long as a whole day, she could no longer be gay. She

  • 8/12/2019 Giora97 Discourse Coherence

    13/18

    R. Giora / Journal of Pragmatics 27 1997) 17-34 29

    would pre tend being gay, she would t ry be ing gay, but a t the end of the dayshe was no lon ger gay. She was wai t ing for Georgine Skeene to be back a t theend o f the day so tha t she would be gay once again .

    (13a) seems almost uninformative. However, a close inspection wil l reveal that eachassumpt ion adds a cer tain amo unt o f informat ion , and therefore extends the contextset of assump tions. For instance, the firs t sentence open s up possibi l i t ies such as ' thec i rcumstances under which the protagonis t would not be as gay ' , or 'unpredic tablec i rcumstances under which she would never the less go on be ing gay ' , e tc . The nextsentence, however, does not reduce any such possibi l i t ies . Rather, i t s trengthens anal ready g iven assum pt ion by repeating i t, adding only a smal l amount o f new infor-mat ion: 'never t i red of be ing gay ' . S imi lar ly, the fo l lowing sentence , whi le repeat-

    ing g iven assumpt ions , adds some new informat ion ( ' l earned many l i t t le ways inbeing g ay ' ) which never theless does not reduce poss ib le opt ions and so for th . (Onlythe last sentence seems to contr ibute a greater amount of information, which reducesopt ions by ha l f ) . The addi t ion of such smal l amounts of informat ion may resul t incontextual effects . However, i t wil l not render the text informative enough. In thissense of inform ativeness , but not in S W relevance, this passage differs from themore coherent text in (13b).

    In (13b), each sen tence reduce s by h alf the possibi l i t ies given r ise to by the previ-ous sentence. While the f irs t sentence suggests that the protagonist would ei ther goon being gay or s top being gay, the second sentence reduces these possibi l i t ies by

    hal f by implem ent ing one . H aving read the second sentence , there i s a chance tha tsomething might change, as wel l as the poss ib i l i ty tha t noth ing wi l l change . Theimplementa t ion of one of these poss ib i l i t ies by the fo l lowing sentence reduces thepossibi l i ty of the other, thereby rendering the text informative, and so forth. Unlike(13a), (13b) evolves gradually along the informativeness axis , reducing possibi l i t iesby hal f , thereby obey ing the Graded Informat iveness Requiremen t (cf . (8b) above) .I ts rela t ive wel l - formedness , then , does not fo l low f rom S W re levance . Nei therdoes the relat ive i l l -formedn ess of (13a). W hile (13a) does no t differ from (13b) innumber of contextual e ffec ts (weighed agains t process ing effor t ) , the amount ofinformat iveness of each assu mpt ion i s d i fferent . In (13a), each assum pt ion i s h ighlyprobable, and does not reduce possibi l i t ies by half . By contrast , in (13b) eachassum ption is inform ative, s ince i t reduces possibi l i t ies by half . S W 's releva ncetheory, then, cannot predict or account for the difference between (13a) and (13b)i

    I ronica l ly, according to S W, (13a) may be cons idered even more re levant than(13b). Unlike (13b), i t yields extra contextual effects . I ts repeti t iveness renders i tpoet ic ( for how repet i t ions achieve S W re levance see S W, 1986: 2 20-221 ;Blake mo re, 1992; Jucker, 19948). So, while (13a) is more S W relevant than (13b),i t i s much less coherent . The very device (breach of the Graded Informat ivenessRequ iremen t) which renders (13b) relat ively incoherent , renders i t more S W rele-

    On how poetic devices such as repetitions, analogiesand irony impair discourse comprehensioncon-sider Giora (1990, 1993, 1995), Giora et al. (1996).

  • 8/12/2019 Giora97 Discourse Coherence

    14/18

    30 R. Giora /Journal qf Pragmatics 27 1997) 17-34

    vant . The S& W re levanc e of the above d i scourses , then, mask s d i ffe rences inamou n t o f i n fo rma t ivenes s and cons equen t l y, i n deg r ee o f coh e r e nce . I t c anno taccount fo r our in tu i t ions concern ing the coherence d i ffe rence be tween the two ( sup-pos e d ly equ a l l y S &W re l evan t ) d i s cou r ses .

    The exam i na t i on o f t he d i scou r se s i n (10 ) - ( 13 ) s ug ges ts t ha t t he mechan i s m o fobta in ing as g rea t a con tex tua l e ffec t fo r as smal l a p rocess ing e ffor t does no t t akein to account the re la t ive cohere nce o f the var ious d i scourse s t ruc tures . W hi le in(10a) the poe t ic e ffec t resu l t s f rom the v io la t ion of the Relevan ce Re qui rem ent , in(13a) i t i s the breach of the Graded Inform at iven ess Requi re men t tha t accounts fo ri t s i nco he re n ce . Howeve r, because bo th d i s cou r s e s a ch i e ve S& W re l evan ce , t he i rre la t ive incoherence or the type of incoherence i s masked and fos te rs an i l lus ion ofequiva lence .

    4 . S W r e l e v a n c e a n d a t t e n t i o n

    Reca l l tha t fo r S&W, re levance i s a quant i t a t ive no t ion . Comprehenders wi l l paya t ten t ion to in format ion tha t wi l l y ie ld as many contex tua l e ffec t s as poss ib le a t assm a l l a co s t a s pos s ib l e . Howeve r, so me be hav i o r c a nn o t be a cco un t e d fo r i n t e rm sof e ffec t versus cos t on ly. Co ns ider an occas ion on w hich heare rs wi l l mos t p roba-b ly s t o p p ay i ng a t t en t ion t o h ighly S & W r e l e van t i n fo rm a t ion because a mo r e i mpor-tan t , even though less S&W re levant message may a t t rac t the i r a t t en t ion . An e labo-

    ra t ion of an ep isode c i ted in Smi th and Wi lso n ' s (1992: 5 ) wi ll se rve to i llus tra te thepoin t :

    [S ]upp ose t ha t someone wa lk s i n t o a n i mp or t an t l e c t u r e and s ays :(14) Ladies and g en t lemen, I have to te l l you tha t the bu i l d in g ' s on f i re .

    This in te r ru p t ion ( s ic p . 5 ) i s obv ious ly S& W re levan t to the audience , s ince i t hascontex tua l impl ica t ions in the g iven contex t . However, i t i s no t necessar i ly moreS &W re l eva n t t han t he l e c tu r e r ' s cu r r e n t con t r i bu t i on . G iv en S& W 's qua n t i t a t i v enot ion of re levance , i t i s qu i te poss ib le tha t the lec ture r ' s cur ren t u t te rance , in the

    contex t o f her ta lk , ma y ach iev e a g rea te r amoun t o f S& W re levance than the u t te r-ance in (14) . Never the less , there i s no doubt tha t th i s more S&W re levant con t r ibu-t ion wi l l be abandoned for the sake of the new and less S&W re levant in format ion .Obv ious ly , the a t trac t iveness o f th i s u t te rance has no th ing to do wi th i t s amo unt o funcos t ly con tex tua l e ffec t s , bu t wi th i t s spec i f ic con ten t , impor tance , o r qua l i ty. Acontex tua l impl ica t ion tha t has to do wi th l i fe and dea th i s wor th a thousand , l essfa ta l ly laden ones . W hat th i s examp le sugg es t s i s tha t a quant i t a t ive no t ion of re le -van c e w i l l no t suf fi c e. W e need ano the r exp l ana ti on fo r why peop l e w o u ld p a y a t t en-t ion to on e typ e o f s t imulus ra ther than to ano ther in t e rms of i t s con ten t. 9 Or, t akefor ins tance , S& W 's exa mp le (15), c i t ed in Blass (1990: 22). The u t te rance in (15b)

    For a discussion of a similar example see Jucker (1992: 83).

  • 8/12/2019 Giora97 Discourse Coherence

    15/18

    R. Giora / Journal of Pragmatics 27 1997) 17-34 31

    lends i tself to two interpretat ions; both, acco rdin g to Blass , are consis ten t with thepr inc ip le of S W re levance , though one ( the men t ion of Susan ' s asser t ion) is mo recoherent than the o ther ( the speaker ' s asser t ion) . Yet the incoherence of the la t te rdoes not rule out i ts use:

    15a) What d id Susan say?(15b) You have d ropped your pu r se .

    Blass concludes tha t speakers and hearers a re not cons t ra ined by the search forcoherenc e . H ow ever, a c lose inspec t ion o f the less coherent in te rpre ta t ion wil l revea ltha t ne i ther a re in te r locutors cons t ra ined by the search for S W re levance . The lesscohe rent in te rpre ta t ion is a l so less S W re levant . The chang e of d iscourse- topicwhich renders i t l ess coherent a l so requi res ex t ra process ing (e .g . , Gemsbacher,1990), such that is not necessa ri ly offset by a r ich array of con textua l effects . Bo thin terpre ta t ions may be equal ly r ich in contextua l e ffec ts . The cos t , however, mayvary. Would i t be safe to assume tha t the hearer wi l l never the less recover the lessre levant in te rpre ta t ion and bend down to p ick up her purse , a t the cos t of ignor ingthe more S W re levant in te rpre ta t ion , or even a t the cos t of making a fool ofherse l f , expos ing her misunders tanding?

    Speakers and hearers ma y hav e o ther goa ls than jus t enr ich ing the i r cogni t iveenvi ro nm ent or that of the i r addressees . Saving one ' s l i fe or re t r iev ing one ' s los tpossess ion seem mo re impor tan t than be ing S W re levan t ( see a l so Murray, ms . ).

    5. Conclusions

    I have sho wn tha t S W re levance cannot be the only pr inc ip le tha t governshuman communica t ion . Speakers and hearers a re not cons t ra ined only by the searchfor re levance . In addi t ion , coherence cons idera t ions cons t ra in communica t ion andplay a majo r ro le in d iscourse s t ruc turing and unders tanding . I f one of the goa ls of apragm at ic theory is to account for speakers ' in tu it ions, S W 's re levance theory fa il sto account for speakers ' in tu i t ions as to coherence . On cer ta in occas ions , i t cannotd i st i ngu i sh cohe rence f rom incohe rence o r deg rees o f cohe rence .

    S W 's assum pt ion that cohe rence i s der ived f rom S W re levance i s un tenable .Al though a d iscourse may be S W re levant to an ind iv idua l in te rac t ing wi th her /h is se t o f assumpt ions a t a small cos t , th i s d i scourse ma y n ever the less be judg ed asincoheren t by the same indiv idual . And v ice versa : A d iscourse may be judg ed ascohe rent by an ind iv idua l and ye t be S W i r re levant to her /h im. S W 's n ot ion ofre levance cannot account for d i scourse coherence because , be ing a theory of a t ten-t ion , i t a l lows for bo th informat ion tha t i s essent ia l for a coherent d i scourse andinform at ion tha t in te r fe res wi th coheren t d i scourse to be def ined as S W re levant -as wor th the bearer ' s a t ten t ion . S W 's re levance theory i s therefore not an a l te rna-t ive to cur ren t d i scourse theor ies . Clear ly, i t cannot d ispense wi th e i ther Relevance(e.g. , Grice, 1975; Giora, 1985b) or Informativeness (e .g. , Grice, 1975; Giora, 1988)Requi rements .

  • 8/12/2019 Giora97 Discourse Coherence

    16/18

    32 R. Giora / Journal of Pragmatics 27 1997) 17-34

    e f e r ences

    Ariel, Mira, 1985. The discourse function of given information. Theoretical Linguistics 12 2/3): 99-113.

    Ariel, Mira, 1988. Retrieving propositions from context: Why and how. Journal of Pragmatics 12 5/6):567-600 . [Reprinted in Asa Kasher, ed., Cognitive aspects of language use. Amsterdam : North- Hol-land, 1989.1

    Attneave, Fred, 1959. Applications of information theory to psychology. New York: Holt, Rinehart andWinston.

    Beeman, Mark, 1993. Sem antic processing in the right hemisphere m ay contribute to drawing inferencesfrom discourse. Brain and Language 44: 80-120.

    Berg, Jonathan, 1991. The relevant relevance. Journal of Pragmatics 16 5): 41 14 25 .Blakem ore, Diane, 1987. Sem antic constr aints on relevance. Oxfor d: B lackwell.Blakem ore, Diane, 1988. The organiz ation of discourse. In: F.J. Newm eyer, ed., Linguistics: The Cam -

    bridge survey, Vol. 4, 229-250. Cambridge: Cambridge University Press.Blakem ore, Diane, 1992. Und ersta nding utterances. Oxford: Blackwe ll.

    Blass, Regina, 1990. Relevance relations in discourse: A study with special reference to Sissala. Cam-bridge: C ambridge U niversity Press.

    Bransford, John D. and Marcia K. Johnson, 1972. Contextual prerequisites for understanding: Someinvestigations of comprehension and recall . Journal of Verbal Learning and Verbal Behavior 11:717-726.

    Brown, Gillian and George Yule, 1983. Discourse analysis. Cambridge: Cambridge University Press.Brownell, H. Hiram, Heather H. Potter, Amy M. Bihrle and Howard Gardner, 1986. Inference deficits in

    r ight hemisphere damaged pat ients . Brain and Language 27:3 10- 321 .Cannito, M., J. Jarecki and R.S. Pierce, 1986. Effects of thematic structure on syntactic comprehension

    in aphasia. Brain and Language 27: 38-49.Carston, Robyn, 1993. Conjunction, explanation and relevance. Lingua 90: 27-48.Chiarello, Christine, 1988. Right hemisphere contributions to lexical semantics. Heidelberg: Springer.Dascal, Marcelo, 1977. Conversational relevance. Journal of Pragmatics 1 309-328.van Dijk, Teun A., 1977. Text and context. London: Longman.van Dijk, Teun A., 1979. Relevance assignment in discourse comprehension. Discourse Processes 2:

    113-126.Dooling, D. James and Roy Lachman, 1971. Effects of com prehension on retention of prose. Journal of

    Experimental Psychology 88: 216-222.Engel-Ortelieb, D., 1981. Discourse processing in aphasics. Text 1: 361-383.Garrod, Simon an d Antho ny Sanford, 1977. Interpreting anaphoric relations: The integration of seman-

    tic information while reading. Journal of Verbal Learning and V erbal Behav ior 16: 77 -90.Gazdar, Gerald and David Good, 1982. On a notion of relevance. In: N.V. Smith, ed., Mutual knowl-

    edge, 88-100. London: Academic Press.

    George, Mary St. , Suzanne Manne s and James E. Hoffman, 1994. Global semantic expectancy and lan-guage comprehension. Journal of Cognitive Neuroscience 6 1): 70-83.

    Gernsbacher, M orton Ann, 1985. S urface information loss in comprehension. Cognitive Psychology 17:324-363.

    Gernsbacher, Morton Ann, 1990. Language comprehension as structure building. Hillsdale, NJ:Erlbaum.

    Gernsbacher, M orton Ann and David J. Hargreaves, 1988. Access ing sentence participants: The advan-tage of mention. Journal of Memory and Language 27: 699-717.

    Gernsbacher, Morton Ann, David J. Hargreaves and Mark Beeman, 1989. Building and accessing clausalrepresentations: The advantage of first mention versus the advantage of clause recency. Journal ofMemory and Language 28: 735-755.

    Gibbs, W. R aymo nd Jr., 1987. M utual knowledge a nd the psychology of conversational inferences, Jour-

    nal of Pragmatics 11 5): 561-588.Giora, Rachel, 1983a. Segmentation and segment cohesion: On the thematic organization of the text.

    Text 3 2): 155-182.

  • 8/12/2019 Giora97 Discourse Coherence

    17/18

    R. Giora / Journal of Pragm atics 27 1997) 17-34 33

    Giora, Rachel, 1983b. Functional paragraph perspective. In: Janos S. Pet6fi and Em el Stizer, eds., Microand macro connexity of texts, 153-182. Hamburg: Buske.

    Giora, Rachel, 1985a. Toward s a theory of coherence. Poetics Today 6(4): 69 716.

    Giora, Rachel, 1985b. A text-based A nalysis of non-narrative discourse. Theoretical Linguistics 12(2/3)115-135.Giora, Rachel, 1985c. W hat ' s a cohere nt text? In: Emel SiSzer, ed., Text connexity, text coherence:

    Aspects, methods, results, 16-35. Hamburg: Buske.Giora, Rachel, 1988. On the informativeness requirement. Journal of Pragmatics 12(5/6): 547-5 65.

    [Reprinted in Asa Kasher, ed., Cogniti~ce aspects of language use. Amsterdam: North-Holland, 1989,63-96.]

    Giora, Rachel, 1990. On the so-called evalu ative material in inform ative texts. Tex t 10(4): 29 320 .Giora, Rachel, 1993. On the function of analogies in informative texts. Discourse Processes 16:

    591-611.Giora. Rachel, 1995. On irony and negation. Discourse Processes 19: 239-264.Giora, Rachel, Nachshon Meiran and Paz Oref, 1996. Identification of written discourse-topics by

    structure coherence and analogy strategies: General aspects and individual differences. Journal ofPragmat ics 26: 4 554 74 .

    Giora, Rachel and Yeshayahu Shen, 1994. Degrees of narrativity and strategies of semantic reduction.Poet ics 22: 447458.

    Gorayska, Barbara and Roger Lindsay, 1993. The roots of relevance. Journal of Pragmatics 19(4):301-323.

    Grice, H. Paul, 1975. Logic and conversation. In: P. Cole and J. Morgan, eds., Syntax and Semantics,Vol. 3: Speech acts, 41-58. New York: Academic Press.

    Hough, M.S., 1990. Narrative comprehension in adults with right and left hemisphere-damage: Themeorganization. Brain and Language 38: 253-277.

    Huber, W., 1989. Text com prehen sion and production in aphasia: Analysis in terms of micro and macro-processing. In: Y. Joanette and H.H. Brownell, eds., 1989.

    Joanette, Yves and Hiram H. Brownell eds., 1989. Discourse ability and hem isphere damage: Theoreti-cal and empirical perspectives. New york: Springer.

    Joanette, Yves, Pierre Goulet and D. Hannequin, 1990. Right hemisphere and verbal communication.New york: Springer.

    Jucker, Andreas H., 1992. Conversation: Structure or process? In: J.R. Searle et al., (On) Searle onconversation, 77-90. Amsterdam: Benjamins.

    Jucker, Andreas H., 1993. The structure and coherence of discourse. In: H. Luffler, ed., DialoganalyseIV. Referate der 4. Arbeitstagung, Basel 1992, 71-78. Tubingen: Niemeyer.

    Jucker, Andreas H., 1994. Irrelevant repetitions: A challenge to relevance theory. In: A. Fischer, ed.,Repetitions (Swiss Papers in English Language and Literature 7), 47-60. Tubingen: Narr.

    Kasher, Asa, 1976. C onversational maxim s and rationality. In: A. Kasher, ed., Language in focus: Foun-

    dations, methods and systems, 197-216. Dordrecht: Reidel.Kasher, Asa, 1982. Gr icean inference reconsidered. Philosophica (Gent) 29: 25 -44.Kutas, Marta and S.A. Hillyard, 1982. The lateral distribution of event-related potentials during sentence

    processing. Neuropsychologia 20(5): 579-590.Levinson, Stephen C., 1989. A review of Relevance. Journal of Linguistics 25: 455-4 72.Mey, Jacob L. and Mary Talbot, 1988. Computation and the soul: A propos Dan Sperber and Deirdre

    Wilson's Relevance. Journal of Pragmatics 12(5/6): 743-791.Molloy, R., Hiram H. Brownell and Howard Gardner, 1990. Discourse comprehension by right hemi-

    sphere stroke patients: Deficits of prediction and revision. In: Y. Joanette et al., 1990, 113-130.Murray, Dinah K.C., unpublished ms. Talking, thinking and wanting.Reinhart, Tanya, 1980. Conditions for text coherence. Poetics Today 1(4): 16t-180.Reinh an, Tanya, 1981. Pr agmatics and linguistics: An analysis of sentence topics. Philosophica (27)1:

    53-94.Rosch, Elinor, 1973. On the internal structure of perceptual and semantic categories. In: T.E. Moore, ed.,

    Cognitive development and the acquisition of language, 11-144. New York: Academic Press.

  • 8/12/2019 Giora97 Discourse Coherence

    18/18

    34 R. Giora / Journal of Pragmatics 27 1997) 17-34

    Sadock, Jerrold M., 1986. Remarks on the paper by Deirdre Wilson and Dan Sperber. In: A .M. Farley,P.T. Farley and K.-E. McCullou gh, eds., Papers fr om the Parassession on Pragmatics and Gram mati -cal Theory. CLS 22, 85-90. Chicago, IL: Chicago Linguistic Society.

    Schneiderman, Eta I., Kuniko G. Murasugi and Douglas J. Saddy, 1992. Story arrangement ability inright hem isphere -dam aged patients. Brain and Languag e 43( I): 107-120.

    Smith , Neil and Deirdre Wilson, 1992. Introduction. Lingu a 87: 1-10.Sperber, Dan and Deirdre Wilson, 1986. Relevance. Oxford: Blackwell.Strawson, P.F., 1 964. Identif ying reference and truth values. Theor ia 30. [Reprinted in D. Steinbe rg and

    L. Jakobovits, eds., 1971, Semantics, 74-81. Cambridge: Cambridge University Press.IZiv, Yael, 1988. On the rationality of 'relev ance ' and the relevance of rationality'. Journal of Pragmat-

    ics 12(5/6): 535-545.