ANANKASTIC CONDITIONALS

Transcription

1 ANANKASTIC CONDITIONALS ARNIM VON STECHOW, SVETA KRASIKOVA, & DORIS PENKA Universiä Tübingen 1. Plo Wha is an anankasic condiional? Previous Analyses of Anankasic Condiionals Sæbø s Analysis von Sechow s analysis von Finel and Iaridou s Analysis Huiink s Analysis A Counerfacual Analysis Where we are now Anankasic Condiionals as Lewis-Counerfacuals Anankasic condiionals as condiional speech acs? Ellipsis resoluion: he implici resricion of he anankasic modal Comparing anankasic condiionals and would-condiionals Srenghening he meaning of he anankasic modal? Wan vs. be o Oher ypes of anankasic senences Conraposiion Resricing he Modal Base? Conclusion Appendix: A noe on our semanics...23 References PLOT 1 The objec of our invesigaion is expressing necessary condiion in naural language, paricularly in a cerain kind of condiional senences, he so-called anankasic condiionals (ACs). We will propose a very simple analysis, which is novel o he bes of our knowledge. The firs exensive discussion of he ACs is due o (Sæbø, 1986), who discovered ha hese condiionals are relucan o a sandard modal reamen. A number of recen papers on his opic refreshed linguisic ineres in his phenomenon and moivaed furher research in he semanics of modaliy. To illusrae he problem, le us look a he senences we will focus on: (1) a. You have o ake he A rain if you wan o go o Harlem. b. If you don ake he A rain you can go o Harlem. c. To go o Harlem you have o ake he A rain. This paradigm was brough ino ligh by (Sæbø, 1986), who followed (Bech, 1955/57) in assuming he equivalence of he condiional in (1a) and he infiniival consrucion in (1c). The condiionals in his lis are called anankasic 2, a erm due o (Hare, 1971). They have 1 We wish o hank Orin Percus, Sigrid Beck and Wolfgang Klein for inspiring discussion of he opic. I is obvious ha we owe crucial insighs o he auhors quoed in his sudy; a warm hanks o hem. 2 From Greek anagkê necessiy. 1

2 he special propery ha he ruh of he consequen is he only way ha guaranees he ruh of he aneceden. Or, he consequen is a necessary condiion for he ruh of he aneceden. While Hare had in mind consrucions like (1a), a beer consrucion o make he semanics clear is (1c) wih he infiniival clause you o go o Harlem as aneceden A and he marix clause you ake he A rain as consequen C. The ruh of C is he only way o enail he ruh of A. Deriving he semanics of (1a) in a composiional manner has proved o be a challenging ask, he main difficuly being he reamen of he bouleic modal in he aneceden of he condiional. Mos of he exising approaches are based on he heory of condiionals developed in (Krazer, 1981) ha assumes wo parameers for he inerpreaion of modals a realisic modal base and a no necessarily realisic ordering source (e.g. eleological or deonic). The firs aemp o give an appropriae semanics for condiionals like (1a) is due o (Sæbø, 1986). Sæbø assumes ha he aneceden wihou wan o (called inernal aneceden ) is added o he ordering source, no o he modal base. Normally an aneceden of a condiional is added o he modal base in Krazer s heory. Recenly, (Sechow e al., 2004) and (Finel and Iaridou, 2004) have argued indepenly ha Sæbø s analysis predics wrong ruh condiions for some anankasic condiionals. Boh criiques conain proposals ha are supposed o overcome he shorcomings of Sæbø s proposal. (Finel and Iaridou, 2004) develop an analysis ha reas he inernal aneceden as a designaed goal ha can override any oher goals sanding in is way. (Sechow e al., 2004) solve he problem by leing he inernal aneceden resric he modal base, hus reaing i as a hypoheical fac and ignoring he conribuion of wan. Boh of he laer accouns have been criicised in (Huiink, 2004). They obain a wrong resul if he goal expressed by he aneceden is compaible wih anoher goal in he ordering source. Huiink s proposal is o rea he ordering source as provided by he aneceden only. This rivialises he idea of an ordering source almos compleely. The aneceden worlds are he ideal worlds and he non-aneceden worlds are no ideal. The accoun amouns o a simple sric condiional: he inernal aneceden is added o a conexually given modal base and he consequen has o follow from ha. If we resae he analysis in his way i becomes clear ha he ordering source doesn play any role for he deerminaion of he ruh condiion. The only hing ha maers is he inernal aneceden. Addiional informaion in he orderings source can yield an inuiively wrong ruh condiion. These consideraions lead us o he view ha one should reconsider he enire analysis. We will rea he anankasic condiional as a Lewisonian counerfacual wih he complemen of he wan-proposiion as aneceden. This will accoun for all he problems ha have been discussed in he lieraure. An advanage of he heory will be ha i says precisely wha he modal base of an AC is in each paricular case, namely all facs ha hold in he nex aneceden worlds. The previous modal heories remain silen abou he quesion of wha is in he modal base. Our proposal sars from he observaion by (Finel and Iaridou, 2004) ha anankasic condiionals wih a wan-clause as aneceden are ellipical. More explici versions of he Harlem example are hese: 2

3 (2) a. If you wan o go o Harlem, you have o ake he A rain o do his. b. If you wan o go o Harlem, you have o ake he A rain o go o Harlem. The imporan ask consiss in finding a semanics for he main clause, which is a condiional wih a purpose clause as aneceden and which conains no wan. In a firs sep we propose ha i means: (3) The neares worlds where you go o Harlem, are conained in he worlds where you ake he A rain. The if-clause, conaining wan, is no he aneceden of he condiional. Is role is o add a furher condiion, which does no have much impac on he ruh condiion of he senence, if any impac a all. The analysis given in (3) is presumably no he enire sory for i doesn express he inuiion ha he consequen is a means for achieving he goal expressed in he aneceden. We have ried o srenghen he anankasic relaion, bu we haven succeeded. Bu compeing proposals are no beer in his respec i seems o us. The srucure of he paper is as follows. In secion 2 we will have a closer look a he daa, secion 3 will be a review of wha has been said in he lieraure wih regard o he meaning of anankasic condiionals and in secion 4 we will presen a counerfacual accoun. 2. WHAT IS AN ANANKASTIC CONDITIONAL? Condiionals of he form (1a) are called anankasic condiionals 3. This is a sor of condiionals wih he consequen expressing necessary condiion for achieving he goal or wish expressed by he aneceden. Thus, he if-clause always has a bouleic/eleological modal expression and he marix clause an explici necessiy operaor. Here is an example from (Bech, 1955/57): (4) Wenn Müller mi Schmid verhandeln will/soll, muss er nach Hamburg fahren. If Müller wans/is o negoiae wih Schmid he has o go o Hamburg Senence (4) means ha he only way for Müller o negoiae wih Schmid is o mee him in Hamburg. Noe ha his paraphrase ignores he conribuion of wan o which we will reurn below. This is he mos naural inerpreaion hough no he only one. Such condiionals are no bound o have he anankasic inerpreaion. Compare he following pair of senences from (Hare, 1971): (5) a. If you wan sugar in your soup, you should ask he waier. b. If you wan sugar in your soup, you should ge esed for diabees. Whereas (5a) shows all he sympoms of being an anankasic condiional, (5b) would sound weird on his reading. Raher, i is a normal mus-condiional senence saying ha in view of he medical facs, in any siuaion in which you wan sugar in your soup and respec he speaker s advice you ge esed for diabees. (4) paerns wih (5a) and i would be difficul o make up a scenario in which i could be inerpreed as a normal condiional like (5b). The presence of wan/be o in he aneceden is obligaory if one wans o obain he anankasic reading, bu as (Sæbø, 2001) accuraely observes, he subjec of wan mus corefer wih he subjec of he marix clause for he anankasic reading o obain. In he following group of examples (6b) is no an anankasic condiional in conras o (6a) and (6c). (6) a. If David is o recover we mus find him a maid. 3 The erm is due o (Wrigh, 1963). 3

4 b. If David wans o recover we mus find him a maid. c. If we wan David o recover we mus find him a maid. This requiremen on coreference/disjoin reference suggess ha wan/be o, whaever heir semanic conribuion is, see o i ha he necessary referenial relaions are esablished. So he presence of hese modals is essenial for he reading o be available and heir role has o be explicaed in he analysis. Anoher imporan observaion in (Sæbø, 2001) links anankasic condiionals o various kinds of purpose consrucions. The idea is due o he Danish scholar Gunnar Bech, who uses anankasic condiionals o paraphrase a cerain class of German o-infiniival consrucions he calls deerminaive (Bech, 1955/57:102). Here he o-clause (or he German um-clause) resrics he modaliy; i provides he range of worlds over which he universal quanifier expressed by he modal mus/muss quanifies. Thus, (4), which we repea below in (7a), is a paraphrase of (7b) on is anankasic reading: (7) a. Wenn Müller mi Schmid verhandeln will/soll, muss er nach Hamburg fahren. If Müller wans/has o negoiae wih Schmid he has o go o Hamburg b. Müller muss nach Hamburg fahren, um mi Schmid zu verhandeln. Müller has o go o Hamburg o negoiae wih Schmid Indeed, he senences in (7) seem o express he same meaning, namely, ha he complemen of mus is he necessary condiion for he ruh of he proposiion Müller negoiaes wih Schmid. (Bech, 1955/57) observes ha senence (7b) has a non-anankasic inerpreaion as well. The o/um-clause may express a normal purpose clause, which has a causal meaning. Under his inerpreaion (7b) means somehing like: Müller mus o ravel o Hamburg because he wans/is obliged o negoiae wih Schmid. The purpose inerpreaion has nohing o do wih modaliy. Indeed, we can omi he modal and sill obain a purpose reading: (8) Müller ravels o Hamburg o negoiae wih Schmid. Anankasic condiionals have a conraposed paraphrase: (9) Wenn Müller nich nach Hamburg fähr, kann er nich mi Schmid verhandeln. If Müller doesn go o Hamburg he can negoiae wih Schmid (10) Müller kann nich mi Schmid verhandeln, ohne nach Hamburg zu fahren. Müller can negoiae wih Schmid, wihou going o Hamburg (9) and (10) are equivalen wih (7) or a leas enailed by i. The semanics for anankasic mus/have o should accoun for his meaning relaion. A noe o he erminology. Following common pracice, we conceive of he if-clause of a condiional as he resricion of an over o cover modal. The if-clause is called he aneceden of he condiional and he modalised proposiion or propery is called consequen of he condiional. In (9), he aneceden is if Müller doesn go o Hamburg, and he consequen is he infiniival he no negoiae wih Schmid. The modal can expresses he logical relaion beween he wo. In he example given, he enire consrucion is negaed in addiion. We exend he erminology o modals ha are modified by um/o -clauses. For insance, (7b) is a condiional wih he aneceden o negoiae wih Schmid and he consequen Müller o go o Hamburg. The wo are mediaed by he anankasic modal mus. To summarize his secion. An anankasic condiional consiss of a modal mus/have o, which is eiher resriced by an if-clause conaining an expression of inenion or obligaion or by a o-clause. This resricion is he aneceden of he condiional. The resricion expresses a goal of he subjec. The enire consrucion expresses he idea ha he 4

5 consequen has o be rue if he goal is o be rue. The consrucion has a can-varian. Then i means ha he goal canno be rue wihou he consequen being rue. The consrucion mus no be confused wih normal purpose consrucions, which have a differen meaning. 3. PREVIOUS ANALYSES OF ANANKASTIC CONDITIONALS In his secion we give an overview of approaches o analysing anankasic condiionals. Since (Sæbø, 1986) he goal has been o develop an adequae analysis for his curious kind of condiionals wihin he heoreical framework of (Krazer, 1981), he mos popular heory for modals and condiionals. (Finel and Iaridou, 2004) and (Sechow e al., 2004) refued Sæbø s analysis by independenly showing ha i fails in he face of inconsisen goals. Their soluions have been criicized in a more recen accoun in (Huiink, 2004) who discusses cerain scenarios wih consisen goals and considers hem problemaic for he former wo accouns. By doing his she brings up an imporan propery of anankasic consrucions a resriced noion of necessary condiion. Le us look a hese analyses in deail and see wha lessons we can draw from hem Sæbø s Analysis Sæbø was he firs o idenify anankasic condiionals as problemaic daa for (Krazer, 1981). According o Krazer, he wo conexual parameers for he inerpreaion of modals are he realisic modal base and he ordering source, f and g respecively. f and g are defined as ses of proposiions 4. f deermines accessible worlds ha are furher resriced by g. g(w) conains goals, wishes, regulaions ha are used o order accessible worlds according o how many of hose hey make rue. (11) Ordering relaions 5 : Le g be an ordering source and le u,v,w be worlds: v < g(w) u iff {p g(w) : p(u)} {p g(w) : p(v)} The if-clause of indicaive condiionals resrics he modal base in he sense ha H(w) is inerseced wih he proposiion i expresses. If we apply his semanics o he anankasic condiional in (12), we predic ha i is rue in w wih respec o f and g iff he condiion in (13) holds: (12) If you wan o go o Harlem you have o ake he A rain. (13) ( w f(w)) you wan o go o Harlem in w & ( w f(w)) you go o Harlem in w & w < g(w) w ) you ake he A rain in w where f(w) conains relevan facs, e.g. rain schedules, and g(w) is a se of your goals/wishes in w. Condiion (13) saes ha he senence is rue iff in all accessible worlds which make as many of your goals rue as possible and in which you wan o go o Harlem, you ake he A rain. This fails o capure he inuiive meaning of (12), as Sæbø correcly observes. This is so for he following reason: I is a fac ha you go o Harlem only if you ake he A rain. I is no a fac ha you wan o go o Harlem only if you ake he A rain. 4 The proposiions in f mus be consisen, herefore i is possible o rea f(w) as a se of worlds. We will follow his pracice in he res of he paper. 5 We assume a simplified definiion of he ordering source due o Irene Heim (lecure noes) ha presupposes he Limi Assumpion. 5

6 To remedy Krazer s analysis Sæbø suggess ha i is he ordering source ha grows as a resul of processing he if-clause, no he modal base. Wan in he if-clause is a kind of indicaor ha he inernal aneceden (he complemen of wan) is added o he ordering source and he modal base is no exended by a hypoheical fac, as is he case wih ordinary condiionals. Thus, Sæbø s revised semanics for condiionals consiss of wo clauses - he firs one applies o normal condiionals, he second akes care of he anankasic ones: (14) (Sæbø, 2001:442) The idea behind he qualificaion (i) is ha v F(v) is supposed o be, he exernal aneceden, and his proposiion is added o he modal base f(w) for non-anankasic condiionals. 6 The qualificaion (ii) for anankasic condiionals is bes undersood by considering he sandard example (1). G is he informaion you wan. v G (v) he se of proposiions ha are rue in every world where you wan o go o Harlem. Presumably, his se conains he proposiion you go o Harlem and every proposiion enailed by i. This se is added o he ordering source g(w). The firs problem of arising wih his analysis is ha i is no composiional. There is no sysemaic procedure he informaion G is obained from he aneceden. The second problem is ha Sæbø underesimaes he complexiy of he ordering source and he role of he inernal aneceden as a hypoheical fac. To show why i is enough o consruc a scenario where he goal expressed in he aneceden is in conflic wih he real goals of he subjec. In his case, he condiional comes ou false under Sæbø s accoun, which should no be he case. One such scenario for senence (12) is discussed in (Finel and Iaridou, 2004) under he ile The Hoboken Problem : (15) The Hoboken Scenario a. You wan o go o Hoboken. b. Harlem and Hoboken are conflicing goals, e.g. for ime reasons you can visi boh places on one day. c. The PATH rain goes o Hoboken. d. The A rain goes o Harlem. F&I correcly observe ha if (12) is uered in a siuaion like (15) Sæbø s analysis fails. According o Sæbø s accoun, he senence is rue iff in all he bes worlds you ake he A rain. The bes worlds are he Harlem worlds and he Hoboken worlds. Bu i does no follow from he relevan facs ha you ake he A rain if you go o Hoboken. So you don ake he A rain in all he bes worlds. There are some worlds in which you ake he PATH 6 We don quie undersand where he F comes from. A more ransparen way would been o say ha is added o he modal base f(w). 6

7 rain, viz. he Hoboken worlds. Therefore he condiional is false, which is a wrong predicion.(sechow e al., 2004) brough up he same poin independenly by discussing he following senence from (Krazer, 1981: 315): (16) If you wan o become he mayor, you mus go o he pub regularly. (17) The mayor scenario a. You wan o become he mayor. b. You don wan o go o he pub regularly. c. You become he mayor only if you go o he pub regularly. If we follow Sæbø and add he proposiion you become he mayor o g(w) wihou any resricion of he modal base, we ge he same problem again. The senence is incorrecly prediced false, because here are bes world in which he consequen does no hold, viz. he worlds where (17b) is fulfilled von Sechow s analysis In his lecure noes from 2003 proposes ha he wan in he aneceden is empy a he logical form. The aneceden is added o he modal base, which is a circumsanial one, i.e., i expresses facs of he acual world. So he inernal aneceden plays he role of a hypoheical fac. Since anankasic condiionals have he form of indicaive condiionals, he aneceden has o be consisen wih he modal base. The analysis can cope wih boh he Hoboken problem and he Mayor problem. I is no enirely composiional, because he conribuion of wan in he aneceden remains unclear. If he second parameer would be a bouleic ordering source, we would obain Huiink s problem as well von Finel and Iaridou s Analysis (Finel and Iaridou, 2004) discuss hree differen soluions of he Hoboken problem. Their final choice is he designaed goals analysis, which we will discuss in he following (for he criical review of he remaining wo alernaives see (Sechow e al., 2004)). von Finel and Iaridou are guided by he inuiion ha he aneceden of an anankasic condiional conribues a designaed goal o he semanics. Crucial for hem is ha here should be a mechanism ha makes he proposiion expressed by he complemen of wan or by he o-clause in he infiniival consrucion override any oher goals in he ordering source. Following his idea hey sugges ha eleological modals are resriced by he designaed goal argumen. Their proposal is his: (18) von Finel & Iaridou s analysis The goal expressed by he if-clause of anankasic condiionals conribues he designaed goal by filling he relevan argumen slo, since i is conexually salien. If we consider he Harlem senence, he designaed goals analysis correcly predics ha in all circumsanially accessible worlds, in which your goal o go o Harlem is achieved, you ake he A rain, which is a welcome predicion. 7

8 This analysis is successful in solving he problem wih inconsisencies in he ordering source. In fac, he ordering source doesn play any role for he analysis. However, he claim ha he analysis is composiional is no jusified. I works no beer in his respec han (Sechow e al., 2004). Wan does no conribue o he meaning of he senence. A he end of heir paper, von Finel and Iaridou speculae on how he exernal aneceden could be inegraed ino he ruh condiion. One possibiliy considered is he inroducion of a second silen episemic modal, somehing like If you wan o go o Harlem MUST [o go o Harlem, mus you o ake he A rain]. Or he if-clause somehow expresses an addiional modificaion of he circumsanial modal base. None of hese proposals solves he composiional problem, because wan is ignored for he essenial par of he ruh condiion, viz. (18) [= (34)]. Noe ha we canno see any difference beween [34a] and he analysis proposed in (Sechow e al., 2004), excep for wording. The designaed goal is reaed precisely as if i where he aneceden of a condiional wih circumsanial modal base and eleological ordering source, i.e., i plays he role of a hypoheical fac. von Finel and Iaridou make a furher sep o make sure ha he aneceden is reaed composiionally. They allow i o resric he modal base of he eleological modal. One of he advanages of von Finel and Iaridou s proposal is ha i disinguishes beween mus-condiionals and weaker ough-condiionals. The difference is ha he laer are evaluaed wih respec o he eleological ordering source, whereas for he former he ordering source can be empy. There is one paricular feaure in he analysis ha srikes us as being correc: he auhors assume ha he main clause of he anankasic condiional is ellipic: he resricion for he modal is deermined by he conex and perhaps he if-clause iself. We will sick o ha idea in he following Huiink s Analysis (Huiink, 2004) is he mos recen aemp o solve he puzzle of anankasic condiionals. Huiink argues ha if here are several non-conflicing goals a sake and several ways o achieve he goal in he aneceden, he anankasic reading canno obain. So anankasic senences are false in such cases. However, hey can be prediced rue under he analysis in (Finel and Iaridou, 2004). The scenario ha should make he argumen clear is he following: (19) The Ruud von Niselrooy scenario a. To go o Harlem, you can ake he A rain or he B rain. b. You wan o go o Harlem. c. You wan o kiss Ruud van Niselrooy (Duch soccer sar). d. Ruud van Niselrooy is on he A rain. The designaed goal analysis would predic ha he Harlem senence is rue a leas in is ough-version: (20) If you wan o go o Harlem you ough o ake he A rain. Wha we ge is ha in he bes Harlem worlds, i.e. he worlds in which you kiss Ruud van Niselrooy, you ake he A rain. This follows from he facs in he described scenario. So he senence is rue bu i shouldn, because he A rain is no he necessary condiion for going o Harlem in (19). (Huiink, 2004) raises he same objecion agains he accoun in (Sechow e al., 2004). 8

9 Here, he same resul obains for (20) as wih he designaed goal analysis if we admi a non-empy ordering source. Huiink follows Sæbø in assuming ha he inernal aneceden of he condiional is added o he ordering source. In view of he possible inconsisencies, she has o make an imporan modificaion. She assumes ha he aneceden alone consiues he ordering source. This draws on he idea ha he ordering source in conras o he modal base mus be explicily saed. Discussion. There are wo possible problems wih Huiink s analysis. The fis is ha he inernal aneceden has o be consisen wih he modal base. Oherwise rue anankasic condiionals would prediced o be false. The compaibiliy requiremen does no follow from he archiecure of Krazer s semanics for modaliy. The orderings source ypically conains proposiions ha are no compaible wih he ordering source. The second problem is he quesion of which relevan facs are in he ordering source. If his quesion is no answered, he heory is virually empirically empy. Problem 1: The inernal aneceden has o be compaible wih he ordering source. Assume a siuaion w in which he proposiion p = his waer doesn boil is rue. Suppose he modal base includes his fac. The following anankasic condiion is false in w: (21) If his waer is o boil, is emperaure ough o be Celsius. The worlds quanified over by he modal are all p-worlds. The goal p canno be added because i is inconsisen wih he modal base. Therefore he ruh condiion of he anankasic condiional is he following saemen, which is false: (22) In every world where his waer doesn boil and..., is emperaure is Celsius. I has o be explained why i is no possible ha a goal is in conflic wih his very salien fac and he modal base in general. Problem 2: Which are he relevan facs in he modal base? Consider (20) again. There are many ways o go o Harlem. You can ake he A rain, you can ake a axi, you can ask someone o give you a ride, you can walk all he way hrough Manhaan, you can preend o be an emergency and call for he ambulance and so on. In oher words, i is no only a fac ha you will reach Harlem, if you ake he A rain. I is a fac as well ha you will reach Harlem if you walk hrough Manhaan and so on. If all of hese where equally relevan for he evaluaion of he ruh of (20), he anankasic condiional would be prediced o be false under Huiink s analysis. So how do we know ha of all hese facs he only hing ha maers is he fac ha you will reach Harlem if you ake he A rain? We hink he answer is ha aking he A rain is he easies way of going o Harlem. The oher ways menioned are more remoe possibiliies. So we really need a heory ha chooses he leas remoe possibiliy among several possibiliies. We hink ha Lewis s heory of counerfacuals does precisely his. Huiink s conclusions are imporan in one more respec. She insiss on he purely anankasic reading which is no available in scenarios like (19). Bu anankasic senences are no always false in such scenarios. Quie ofen we acually have o deal wih pseudo anankasic readings in he sense ha we resric he domain in which he necessary condiion holds. We will discuss ha poin in secion We can now look a our ask ou of a new perspecive. 9

10 4. A COUNTERFACTUAL ANALYSIS 4.1. Where we are now In order o develop an adequae analysis of anankasic condiionals, le us summarize wha he previous accouns had o sruggle agains: The hypoheical fac in he aneceden is he mos decisive facor in he evaluaion. Bu i is embedded under he bouleic modal, so no direcly accessible. If one reas i as a wish/goal, bu no as a relevan fac, one has o give i precedence over whaever else is in he eleological ordering source or one has o define he ordering source in a compleely new way. (F&I and we conra Sæbø) Ranking accessible worlds on he basis of he eleological ordering source is problemaic. Wha is bes wih respec o your goals is no always wha is necessary. (Huiink) In he nex secion, we will aack he problem by looking more carefully a he synacic consrucion of anankasic condiionals wih if-clauses ha conain wan. This closer invesigaion will reveal ha he if-clause forms no par of he anankasic condiional a all. The consequence will be ha we need no find a locus for wan in he composiion of he ruh condiion Anankasic Condiionals as Lewis-Counerfacuals The idea is simple. Le us assume ha we are dealing wih a differen source of ordering in he case of anankasic condiionals. We rank worlds on he basis of comparaive similariy o he acual world, in he sense of (Lewis, 1973b). As wih counerfacuals, we resric accessible worlds o hose ha are closes o he acual world, i.e. make as many of is facs rue as possible. To keep pace wih he preceding discussion, we remain in Krazer s framework, where a counerfacual is formalized as a modaliy wih an empy modal base and a oally realisic ordering source. f is an empy modal base if f(w) is he necessary proposiion W for any world w, and g is oally realisic if g(w) = {w} for any w. Formally, he ruh condiion looks exacly as he semanics ha von Finel and Iaridou sae for anankasic ough in (18), which is repeaed for convenience): (23) Anankasic necessiy (o be revised) o p, ough o/have o q is rue in w wih respec o modal base f(w) and ordering source g(w) iff all he g(w)-bes worlds in f(w) where p is rue are q-worlds (iff all he g(w)-bes worlds where p is rue are q-words) The g(w)-bes p-worlds are he p-worlds ha are as similar o w as hey can be. This semanics is a reformulaion of Lewis (1973b) semanics for he counerfacual operaor in Krazer s erms. 7 The definiion neglecs he qualificaion for vacuous ruh and assumes Salnaker s limi assumpion. For he ime being we assume ha wan does no play any role a all. This assumpion will be revised. The ruh condiion for he anankasic condiional (1) reads as follows hen: (24) To go o Harlem you have o ake he A rain is rue in w wih respec o g iff 7 For a comparison beween Krazer s and Lewis semanics for counerfacuals, see (Lewis, 1981). 10

11 you ake he A-rain in every g(w)-bes world where you go o Harlem. The reader may wonder a his sage where he you wan par of he aneceden figures in he ruh condiion. We will reurn o ha quesion in secion 4.3. Noe ha he problem discussed in (Huiink, 2004) does no arise under his accoun. Recall he scenario in (19). If here are wo rains going o Harlem, whaever your preferences are, he senence is false in his siuaion. We are considering he nex Harlem worlds. No in all such worlds you ake he A rain, in some of hem you ake he B rain. Consider now he following senence: (25) If you wan o go o Harlem and kiss Ruud van Niselroy, you have o ake he B rain. In he given scenario, (25) is prediced rue, which is correc. Ineresingly, his shows ha aneceden srenghening is no a valid inference for anankasic condiionals, a well-known fac abou counerfacuals. This analysis srikes us as aracive. We don need o inven a new semanics for necessary condiion senences we rea hem as insances of counerfacuals Anankasic condiionals as condiional speech acs? In his secion we will give argumens for he claim ha he if-clause in anankasic condiionals figures as he aneceden of wha has been called a condiional speech ac in he lieraure. Two kinds such condiionals have been sudied in he lieraure: RELEVANCE condiionals and FACTUAL condiional. 8 Unforunaely, ACs are differen from eiher of hem. The funcion of he if-clause is similar o ha of a relevance condiional, he synax is more like ha of a facual condiional. The following examples illusrae he relevance condiional: (26) a. If I may be hones, you are no looking good. (Bha and Pancheva, 2004: [95]) = To be hones, you are no looking good. b. If you wan o know, 4 isn a prime number. c. If you are hirsy, here is beer in he fridge. Bha & Pancheva characerize his use of condiionals as follows: The if-clause in Relevance condiionals specifies he circumsances in which he consequen is discourserelevan, no he circumsances in which i is rue. The following example is a facual condiional: (27) a. My friend Joe, whom you haven me, is very smar.(bha and Pancheva, 2004: [97]) b. Oh yeah? If he s so smar why isn he rich? This use can be paraphrased as: If i is rue ha he is so smar, why isn he rich? The quesion is only acivaed in conexs where i is aken for graned ha he aneceden is rue. Noe incidenally ha he following conras ha has been observed by Ernes Adam and is quoed in(lewis, 1973b: p. 3): (28) a. If Oswald did no kill Kennedy, hen someone else did. b. If Oswald had no killed Kennedy, hen someone else would have. Lewis commens on hese ha he firs one is presumably rue while he second is false. He 8 The erms facual condiional and relevance condiional are aken from (Bha and Pancheva, 2004: secion 5.2). Presumably he erms go back a leas o (Iaridou, 1991). 11

12 observes ha he wo canno be disinguished simply in erms of presupposiions. He says ha his heory of condiionals doesn cover he firs one. We hink ha i is a facual condiional, i.e., he senence means ha someone differen from Oswald killed Kennedy and i is appropriae only in a conex where someone differen from he speaker believes ha Oswald did no kill Kennedy. This facual condiional is roughly equivalen o a maerial condiional. Anankasic condiionals probably aren relevance condiionals, because hey do no saisfy any of he sandard ess known from he lieraure. 9 The consrucion is no easily classified as a facual condiional eiher, because a facual condiional implicaes ha he speaker is no convinced of he ruh of he if-clause. We leave he precise saus of he consrucion open bu assume ha he consrucion is a hypoheical speech ac. For our purposes he following crude rule of use is sufficien. (29) The appropriaeness condiion Le c be a conex of use. Then [[ if ]] c is only defined if c [[ ]]. If defined, [[ if ]] c =[[ ]]. This rivial semanics makes i possible o accoun for he role of you wan in he consrucion. Consider example (2). is he clause if you wan o go o Harlem. So he enire senence can be used only in conexs, where you wan o go o Harlem. I follows ha he inner aneceden expresses a goal of he subjec of wan. This explains he oddness of he following senence: (30) # If you don wan o go o Harlem, you have o ake he A rain o go o Harlem. The aneceden ells us ha he conex of use implies he proposiion ha he addressee 9 Among oher hings, (Bha and Pancheva, 2004) menion he following crieria of relevance condiionals, which are mosly no me by anankasic condiionals. (i) hen should no be possible in he consequen; (ii) in V2-languages such as German, he if-clause should be generaed ouside CP; (iii) relevance condiional can be embedded only under speech ac verbs, no under normal aiudes. (A) a. #If I may be hones, hen you are no looking good. b. If you wan o go o Harlem, hen you have o ake he A rain. In anankasic condiionals, Then inserion seem marginally possible: (B) a. Wenn es dich ineressier, 4 is keine Primzahl. if i you ineress, 4 is no prime b. *Wenn es dich ineresssier, is 4 keine Primzahl. (C) a. *Wenn du nach Harlem fahren wills, du muss den Zug nehmen. b. Wenn du nach Harlem fahren wills, muss du den A-Zug nehmen. The conras beween (B) and (C) shows ha he if-clause of he anankasic consrucion occupies SpecCP. (D) a. John said ha if you are hirsy here is beer in he fridge. b. *John believes ha if you are hirsy here is beer in he fridge. c. John believes ha you have o ake he A rain if you wan o go o Harlem. In (c) he embedded clause has an anankasic reading. So none of he sandard es for relevance condiionals applies. Therefore, he anankasic if-consrucion isn a relevance condiional of he usual kind. 12

13 doesn wan o go o Harlem. I would be impolie or poinless o uer he consequen, which conradics he inenion of he addressee. The senence would be ok under a differen consrual, namely: (31) If you don wan o go o Harlem, you have o ake he A rain for no going here. Though appropriae, such a saemen would be a lie under he acual circumsances Ellipsis resoluion: he implici resricion of he anankasic modal As von Finel and Iaridou have observed, anankasic wan-condiionals are ellipical. An explici saemen of he Harlem example was given above and is repeaed here: (32) If you wan o go o Harlem, you have o ake he A rain o do ha. = If you wan o go o Harlem, you have o ake he A rain o go o Harlem. The ruh condiion we have given in secion 4.2 was inended for he main clause, which doesn conain wan. One funcion of he if-clause is ha i delivers he aneceden for he ellipsis. So a closer inspecion reveals ha he Harlem-senence doesn express a condiional wih one aneceden, bu a condiional wih wo anecedens, one of which is he if-clause of he condiional speech ac and he oher is he resricion of he modal. The LF of he senence is somehing like his: (33) If you wan o go o Harlem [you have [ o go o Harlem] o ake he A rain] We claim ha i is he o-clause ha funcions as he resricion of have o, no he ifclause. The complex main clause alone expresses he following anankasic condiional, and ha is all wha we need: (34) In all he worlds where he goal ha you go o Harlem is achieved and which make as many of he facs rue as possible, you ake he A rain. I seems o us ha his is precisely wha von Finel & Iaridou and Huiink have in mind. Bu we have said more han hem, namely wha he relevan facs in he modal base are, namely hose ha are coenable wih he aneceden. 10 Turning back o anankasic condiionals, we sugges ha he funcion of he if-clause is o reaffirm ha he conex is appropriae for he following ellipical anankasic saemen For hese paricular condiionals, i has o follow from he conex ha he aneceden is somebody s goal or wish for he anankasic saemen o be felicious. The eleological aspec is inheren o all kinds of anankasic senences and was responsible for assuming he corresponding kind of ordering source in he previous analyses. We believe ha his meaning componen is due o he presupposiion associaed wih every 10 For he noion of coenabily, see (Lewis, 1973b: 2.6). is coenable wih premise in world w if eiher (1) = W or (2) holds hroughou some -permiing sphere around w. A counerfacual is rue in world i iff here is an auxiliary premise coenable wih in world i such ha and ogeher logically imply. may be regarded as he circumsanial modal ha is needed for he modal analysis of Huiink o work. Each counerfacual can be reformulaed as a sric condiional along hese lines, bu we have a price o pay: he coenable premise depends on he aneceden of he condiional. Differen condionals require a differen coenable premise. In he same way he circumsanial modal base canno be provided by he conex: i depends on he aneceden. Inuiively, i is unclear wha he coenable premise should be. There is no way of making he premise explici by means of one or more paraphrases. 13

14 anankasic condiional, namely, ha he resricion of he anankasic modal is in he se of goals aribued o he logical subjec of he marix clause or he if-clause in he evaluaion world. This leads o a more accurae formulaion of he meaning of he anankasic mus/have o: (35) Enry for he anankasic have o 11 [[have o ac ]] f,g,c (w) is only defined if c is a Salnaker common ground, f(w) = W and g(w) is oally realisic. When defined, [[have o ac ]] f,g,c (w) = p s.c goal(p): q s.( w f(w))(p(w ) & ( w (w < g(w) w & w p) w q) Given ha f(w) = W, he ruh condiion boils down o p s.c goal(p): q s.( w p) ( w (w < g(w) w & w p) w q), i.e., q is rue in he neares p-worlds. The resricion on he firs argumen makes i o one of he salien goals in he evaluaion world. (goal(p) is he proposiion ha p is goal, i.e. p is waned.) This condiion has o be me for he uerance o be appropriae in he conex. By ascribing he goal saus o he aneceden of anankasic condiionals we explain why anankasic condiionals are alking abou goals and means for achieving hem. This is wha makes anankasic condiionals disinc from oher kinds of condiional saemens, in paricular subjuncive condiionals. Noe ha so far he disincion does no concern he ruh condiions bu is on he pragmaic level Comparing anankasic condiionals and would-condiionals Hearing he erm counerfacual analysis, one hinks immediaely of would/couldcondiionals, and one would hink ha every anankasic condiional is expressible as a would-condiional. This, however, is no so, and his fac has been objeced o our proposal. Recall, however, ha he Salnaker/Lewis semanics for counerfacuals has nohing o do wih wording. The counerfacual semanics covers boh subjuncive and indicaive condiionals, hough hese seem o have quie differen meanings. The difference in meaning is explained by a difference in feliciy condiions for use: he aneceden of an indicaive condiional mus be compaible wih he common ground. 12 Counerfacual carry some sor of presupposiion ha he aneceden is false (Lewis, 1973b: p. 3), i.e., he aneceden of a subjuncive condiional mus be incompaible wih he common ground. The anankasic condiionals considered so far are indicaive condiionals and are herefore used in differen conex from would-counerfacuals. A furher difference is ha he resricion of have o is an infiniival. Infiniivals under modals are fuure oriened if hey express a non-saive Akionsar, and o-infiniivals are seem o be fuure oriened quie generally. n approximae paraphrase of our anankasic condiional as a normal condiional should be somehing like his: (36) You have o ake he A rain o go o Harlem 11 Henceforh we use he convenion for formulaing presupposiions adoped in (Heim and Krazer, 1998). Our definiion mus be read as follows: [[ have o]] f,g,c (p)(q) is defined iff c goal(p). If defined, [[ have o]] f,g,c (p)(q) = 1, iff ( w p) w (w < g(w) w & w p) w q 12 Here is he relevan quoe from (Salnaker, 1975) making he poin clear: I is appropriae o use an indicaive saemen or supposiion only in a conex which is compaible wih he aneceden. 12 Subjuncive condiionals, on he oher hand, are argued o presuppose ha heir aneceden is false. 14

15 =? If you will go o Harlem, you mus ake he nex A rain. The ranslaion is no perfec, because he informaion ha he aneceden is a goal of he subjec is no expressed. If we ake accoun of fuure orienaion and add he informaion ha he aneceden is a goal here is some plausibiliy ha he former o senences have he same ruh condiions. Oher cases, however, seem o speak agains his idenificaion. Consider he following classical example from (Lewis, 1973b): (37) a. If Oo behaved himself, he would be ignored. (Lewis) b. For Oo o behave himself, he has o be ignored. These srike us as differen in meaning. Le us look wha our accoun can say abou he conras. The firs difference comes from he difference in emporal orienaion. The firs senence speaks abou counerfacual behaviour now and a subsequen ignoring. In he second senence, he ignoring comes firs and he well behaving laer. So he wo canno mean he same. Bu even if so, here is anoher difference: he second senence gives he informaion ha a possible fuure well behaving of Oo has o be caused by he ignoring. There is no such relaion of causaion in he firs senence. The meaning rule given in (35) doesn express his causaion. I is perhaps oo weak as i sands. The impression ha here is a real difference in ruh condiions beween wouldcondiionals and anankasic ones grows sronger when we look a Lewis sandard example: (38) a. If kangaroos had no ails, hey would opple over. (Lewis) b. For kangaroos o have no ails, hey have o opple over. Here boh he aneceden and he consequen are saes, and sressing our inuiions somewha, he second senence may have a simulaneous inerpreaion, even a enseless one. Again we have he feeling ha he wo mean somehing differen. The second senence suggess ha he habi of oppling over somehow causes he sae of having no ails. This is an absurd idea. Therefore our reacion is ha we don undersand (b) or ha i is false. (a) on he oher hand is rue. Again our meaning rule (35) canno explain he difference. The following lis conains examples wih decreasing anankasiciy: (39) a. In order o go o Harlem, you have o ake he A rain. b. In order o go o Harlem, you have o ake a rain. c. In order o go o Harlem, you have o wai. d. In order o go o Harlem, you have o breah. e. In order o go o Harlem, you have o be a person. The inuiion seems o be ha he consequen can be regarded less and less as a means for achieving he ruh of he aneceden. I would seem hen ha anankasic necessiy is a sronger logical relaion han counerfacual necessiy. So i was perhaps premaure o idenify he wo relaions. Le us inroduce a noaion ha disinguishes hem. To faciliae he reading, le us use Lewis noaion for counerfacuals insead of Krazer s: (40) is rue in world w wih respec o he ordering relaion iff holds in every -nex -world. is rue in world w wih respec o he ordering relaion iff holds in some -nex -world. Recall ha he ordering relaion can be defined via an ordering source g. Le us denoe he anankasic relaion by means of he symbol. In he previous secion we assumed he ideniy of and. The consideraions of his secion sugges ha is a sronger 15

16 relaion. Or is i perhaps a logically independen relaion? We don hink so because in cerain respecs behaves like : neiher of he wo relaions allows for srenghening of he aneceden, for ransiiviy or for conraposiion. (41) No srenghening of he aneceden =/=> & An example of case was Huiink s (25), which shows ha he following argumen is invalid. (42) If you wan o go o Harlem, you have o ake he A rain. If you wan o go o Harlem and see Ruud van Niselroy, you have o ake he A rain. As for counerfacual, ransiiviy does no hold for. We can show ha by using Lewis examples. (43) Failure of ransiiviy & =/=> For Oo o go o he pary, Anna has o go. For Anna o go o he pary, Waldo has o go. For Oo o go o he pary, Waldo has o go. The argumen is invalid similar reasons as hose given in (Lewis, 1973b: p. 33). Conraposiion canno hold eiher as an adapaion of an example of Lewis shows: (44) Failure of conraposiion =/=> For Boris o go o he pary, Olga has o go. For Olga no o go o he pary, Boris has no o go. Le us summarize he discussion in his secion. Wih respec o srenghening of he aneceden, ransiiviy and conraposiion, he anankasic relaion and he counerfacual relaion behave alike. An idenificaion of he wo relaions could explain he behaviour. On he oher hand, a plain modal accoun would have o say somehing o he quesion of why hese logical properies don hold. We observe, however, ha would-condiionals canno be paraphrased as ACs and we have o explain why his should be so. Differences in meaning migh come from he differen feliciy condiions for he use of he wo consrucions and he difference in emporal orienaion. So far we have been assuming ha ACs are indicaive condiionals and herefore subjec o Salnaker s resricion. Bu i is no clear ha indicaiviy is an essenial resricion for ACs can be pu ino he subjuncive: (45) To go o Harlem, you would have o ake he A rain. The requiremen ha he aneceden has o be goal doesn help o disinguish he wo consrucions eiher. I migh be a goal ha kangaroos have no ails, bu sill senence (38b) is very odd. Our semanics doesn express he idea ha he proposiion ha kangaroos opple over describes he mos appropriae means for making sure ha hey have no ails. The only imporan difference ha migh yield an explanaion is he difference in emporal orienaion. If he aneceden and he consequen of a would-ccondiional are boh in he presen ense, he aneceden will ypically be rue before he consequen. For an AC, he ypical emporal relaion is exacly he oher way round. The aneceden will be rue afer he consequen. This fac suffices o guaranee a difference in meaning in many cases. This, however, canno explain all weird examples. The kangaroo-example canno be 16

17 blocked because he senence expresses a sor of law and is herefore imeless. The same can be said for oher laws: (46) If his waer is o boil, is emperaure mus be Celsius. This is an AC. Thus i seems ha he definiion of he anankasic relaion should be somehing like his: (47) : & R, where R means is a means for achieving The quesion is of course, wha R could be. This is he opic of he nex secion Srenghening he meaning of he anankasic modal? The mos obvious way of expressing he idea ha he consequen somehow causes he ruh of he aneceden is o build in conraposiion in he semanics, because his is he essenial ingredien of defining causaion. 13 (48) : &? In oher words, R would be he conraposiion of. We know, however, ha conraposiion doesn hold of. So his definiion is no an opion. In addiion, he definiion would no block any of he problemaic examples of he las secion. For insance, he anankasic condiional (38b) would sill be rue. Anoher failing aemp consiss in adding he inuiively correc condiion underlying he following version of an AC: (49) You can go o Harlem wihou aking he A rain. In oher words le R be he saemen ( ). (50) : & ( )? This doesn buy anyhing because he second conjunc means precisely he same as he firs one. In a previous version of his paper we ried o use a differen semanics for making (38b) false. (51) is rue in w : Sim w ( ) Sim w ( )? Here, Sim w ( ) are he -worlds ha are mos similar o w. This semanics would make (38b) false and could herefore explain is oddness. (Huiink, 2004) has objeced ha his proposal would make anankasic saemens wih a false aneceden and a rue consequen false. (In addiion, he anankasic relaion would be ransiive.) Huiink gives he following example: (52) If combusion is o occur, oxygen mus be presen.(huiink) She assumes a scenario where no combusion occurs bu where oxygen is presen. The senence is law of naure and should no depend on he acual ruh of he aneceden or consequen. So his is anoher failed aemp. We have no beer idea of how o define he relaion R. So eiher you ake he relaion is a means for achieving as a primiive, or you sick o he meaning rule in (35). 13 See (Lewis, 1973a). 17

18 4.7. Wan vs. be o As a final sep in our analysis we will explicae anoher, purely synacic, funcion of ifclauses. This secion will offer a way o accoun for he disribuion of wan/be o in he ifclauses observed by Sæbø. The ruh condiion for anankasic condiionals saed in (24) reas he o-aneceden of he main clause as if i were a proposiion. Bu he acual synax is no exacly like ha. Afer resoluion of he ellipsis, he aneceden of he condiional is a o-clause. The missing subjec is redeermined by lexical conrol of he modal have o. The complemen of wan is a consiuen ha has precisely he correc logical ype for serving as an argumen of a conrol modal have. (53) Enry for he anankasic conrol have o 14 : [[ have o C ]] = P D s(e). q D s(e). x D e : w.p w (x) is a goal of x in w. w.p w (x) is rue in g(w)-neares w.p w (x)-worlds. The index C sands for conrol here. This have C requires a o-infiniive as he firs argumen. P and Q are properies of individuals, i.e., hey have he ype s(e). We have ignored he modal base, because i plays no role for he evaluaion of he counerfacual. g(w) is a oally realisic ordering source. Le us apply his semanics o he Harlem senence. Afer ellipsis resoluion, i is (54) and is LF is (55). (54) If you wan o go o Harlem you have o ake he A rain. (55) e you e e o ake he A rain ep e e o go o Harlem ep ep e have The ype mismach requires he applicaion of Inensional Funcional Applicaion wo imes. 15 The analysis predics ha we have subjec ideniy in anankasic condiionals ha are inroduced by an if-clause conaining wan. This predicion seems o be born ou by he facs. The precise formulaion of he ruh condiion for (54) is given in (24). Now le us look a he srucure wih be o: (56) If David is o recover we mus/have o find him a maid. Again, we have o make he ellipic main clause explici. A possible ellipsis resoluion migh be his. (57) For David o recover, we mus/have o find him a maid. This ime, he anankasic modal mus be a raising verb. The LF for he consrucion would 14 Here and in he res of he definiions we ignore he assignmen funcion. 15 See (Heim and Krazer, 1993: chaper 12) 18