ITERATION, HABITUALITY AND VERB FORM SEMANTICS Frank v~n Eynoe. University oi Leuven Mar=a-Theresiastraat, 21 3000 Leuven Belgium ABSTRACT The veto forms are o4ten claimeo to convey two ;inds o+ information : I. w~et'~er the event Oeecribed in a sentence is present, past or future (= oeictic information 2. whether the event described in a sentence is oresente~ as completed, going on, just starting or being ,inished (= espectual information) [t will be ~emonstrated in this paper that one has tl ado a rhino component to the analysis of verb ~orm meanings, namely w~e~ner or no~ they e>press habltualitv. The 4ramewor~ 04 the analysis is mo~el- theoretic semantics. BACKGROUND The analwls of iteration and ha~ituality in this ~aper is part of a comprehensive semantic ar~Ivsis of temporal expressions in natural kanguage. The research on this topic is carried ob~ in ~he framework of EUROTRA, the MT project o4 the European Community. It is reporteo on e, tensi~,eiv in Van Eynde (lqBT). The original motive for s~arting this research ~as the fact that verbal tenses ann temporal a,:~:ili~ries do not corresponO one-to-one in toe ienguages that EUROTRA has to deal with. Compare for in~taqce ,i EN ne has lived in Copenhagen for 20 years ,Z, Dk nan nan boer i KmOenhavn i 20 ~r ~it~ tnelr equlvaients in the fokiowlng languages • S~ DE er wonnt seit 20 Jahren in Kopenhaoen ~i FR ii haDite ~ Copenhaoue Oepuis 20 ans ~5, NL nij woont sinds twintlg jaar in Kopenhagen When translating from Englieh or Danish to German, ~rench or Dutch the present perfect has to Pe replaceO by a simple present, Di&~ererces like these can be handled In one o; two eaVs either by Oefinlng complex mappings from source language to target language forms in transfer or Oy Oeflnlng mappings Oetween language specific forms and Interlingual meanings in the monolingual components. SL ~orm ) TL form complex mmpc~ngs meanlng ) meanlng ,[ identity I mapping mapping SL ~orm TL form Because c* EUROTR~ s ao~erence to the principle o~ "simPle t~ansfe ~" it was quite OOVlOUS ~rom the start that the interlingual approa~ was the one to opt ~or. It will, hence, be adopted in thls paper a~ well. The paper consists of t~ree parts. In .ths flrst I will present a formalism for the representation of time meanings, together with mooel for the interpr~tatlon o~ those representations. In the seconp this forma|ism wiil be extenOeO so that it can also Pe use~ for the ana]~slS o~ iteration an¢ habitL, alit~. Ann In th~ third pert I ~i~i show how the extendeo formalis~ can be l,~' ~or a~ !n[erllnoua~ a~alvsis O~ the ver~ fo~.S, THE CORE FORMALISM A Temporal Model T~e formalist tha: ~ill oe use: here has oee~ de~ineo e~pilcitiy i~, van Eynce, aes TomDe Q Maes ~5). irk th!s p~per i will on!y give a s~or~ In~ormai present~zion of tDe formailS~ CO%Cemtratlrlg On th~se partS ~ICh will De neeOe~ In the se~onO pert, 270 The model COnSIStS of a set of linearly oroerep irtsrvals. An interval ~s a continuous set of time points on the time axis : I , ) A la limite it might consist of one moment o6 tlme : I For an~ pair of ~ntervaie one can Oeflne tnelr interlection as the set of tlme points which the'/ share: i J Inj Tn~s set m~g~t also be empty, as in I J it Is, furthermore, possible to define some b~narv relations between intervais, such ae l preceoence ' , , ~} I be4ore O ( ~i,J: O J after i ;.~J,i; I ~dent~ty "t I ) I simui .] =(l,a) O I contain , , ' ) I part-of J c~I,J., d 0 contain I ~(J,I; I overlap ' , ~ , ) I leftover J ~<(I,J) O J rightover I >>(O,l) T~ese relations are also useO in Bruce (1972). A Format for Representation For the semantic analysis of the temporal expressions i wili start from tne assumption t~at every sentence can be analvseo in two parts : the temporal informal:on expressa~ by the tenses. auxiliaries anO ao~erbials on t~e one hano. anp ~as~o atempora! proposition on the ot~er hand. (b; the cat sat on t~e mat w~.i. for instance, be analyseo in a basic proposition "the cat sit on the met" and the ~n~ormation conveyed Dv the past tense. The relation between both is established in two steps : the basic proposition is first relateO to the interval ~or whicn it is said to be true, the socalled time of event (E), and then this interval is related to the time Of speech ~S) : 3 E :,E,S) ~ AT(E,the cat sit on the mat)3 This formula states that "the cat sit on the mat" i~ true at an interval E which precedes the tlme of speech S. Following Reichenbach (1947) I will furthermore assume that the relation between the time of event and the time of speech is mediated by a thiro kind o~ interval, namely the time of reference (R), So, instead of the simple ReI(E,S) we w111 have a composite ReI~E,R) & RefeR,S). Ne.t to thls relational information tn~ tempore: expreeslons can also give specific informatlon about the iocatlon or the length of the reie~ant intervals. This is typically Oone by means o~ t:me aOverbiais, such as "next year", "in the spring':, "for t~o years", "till Christmas", etc. T~is in~ormatlon will be represented bY means o~ one-place preOicates over intervsls : Freo(E) and Pred~), ~n exception ~s tc be ~ade here +or the time o, speec~. ~nose precise location or length is never spec~fleo b, iinQo~stic means, bu[ rather bv pragmatic factors. A possible way to reelect tn~s- In the &oc~,allsm is to trest it as an unbouno variable. In sum, the general format for the representatlon of temporal information looks as follows : 3 R,E [Rei(R.S) ~ Pred~R; & ~eI(E,R) ~ Prep(E) AT(E,p)] where p is a basic atemporal proposition An example ; ~T we will vielt Moscow next year 3 P,E [,~R,S~ & ne~t vear~R) ~ =~E,R) & A ~ E,we visit Moscowi] 271 As it stanos this format is not adequate yet fo ~ the ~epresentatlon of sentences like (8~ last year they played chess every week (e~ he was always late The basic propoe~tions "they play chess" and "he oe late" do not hold for one particular time of event E, but rather for a set of intervals wnicn are spread in time in some way specifieo by "every week" in (8) and "always" in (9). In the following part I will introduce an exter.oeO formalism which can OeaI with these typos 04 iteration. THE EXTENDED FORMALISM Cyclic Iteration Cyclic iteration is marked by aoverpials like "caiiv", "every Monday", "each year", etc. In ~virk e.a. (1972) they are callao periodic frequency adverbials. For the analysis of these adverbials I first IntroOuce the notion Crams time. The frame time ie the interval which contains all the instances of the event describeo in the basic proposition. In (8~ last year they played chess every week t~e ~rame time is last year. In the general forma~ t.e frame time occupies the same place as the time c~ event in non-iterative interpretations (= the E-i~tervai~, ~ext, I de~ine a set of distinct, non- overlapping subintervals ~I~ which are all part o+ the frame time. In (8~, these intervals have a length of one week each. This gives the following ,preliminary) representation : 5 R,E [ (R,S) & last year(R~ & =(E,R) & I [c'i,E) & nI:~ & week,i) x AT~i,they play chess;]] R S similar analysis can be found in Stump (198i where t~e aoverbial frequency ad~ectlvee (P) ere given the following truth condition : F~' is true in a world w at an interval I i~4 ,~m is true in w at non-overlapping subintervals o$ i distriOuteO throughout i ~t perioOs of a speci$ieo length I. " [Stump 1981, 226] 5t~mp s i-interval corresponds to my frame time. and his non-overlapplng subintervals correspond to my I-intervals. As a representation of (B) this formula is not sufficient, though, since the instances of chess pla~ing do not have to take a whole week for (B~ to. be true. A more adequate paraphrase is to say that every week contained at least one subinterval (e~ during which they played chess : ,,o l[c(l,E~ & nl=~ & week!i) > e[c~e,l) & AT(e, they play cness)]] An argument in favor of this refinement is that languages have special means for specifying the e- times. In ~I(' last year she arrived at ~ c clock every da~ the aoverbia2 "at eight o ¢ioc~" denotes the locatlOn 04 t~e e-intervai ; B Notice tha~ the pro~artlee of e are constant within 'the 4tame time : the aoverDial "st eight o clot!" specities t~e time of each o¢ her arrlvals cf last year. The general format for the representation of cyclic iteration is, hence~ 3 R,E [ReI~R,S) & PreO~R~ & Rei(E,Ri & Pred~E} & I [c(l.E~ & ni=O & P(1) e ~:~e~I~ ~ M(e) 2 AT~e,p;]]] where P is replacec ov the head o4 a periooic ~requencv aoverbial, specifying the location or the iengtn o~ I Io -optlona}l~i replaced ov ~ ti~,a advero~6i, sPecifYin~ the length cr the igcatlon C.f e ~n im[,ortar~t property of this format is it ~. chain-like structure : 272 R is oef~neo with respect to S : ReI~R,S~ E as defined with respect to R : ReI(E,R~ I is defineo w~th respect to E : ~(I,E) and e is oefineo with respect to I : c(e.I~ As it stands, the format does not provioe any means for stating a direct relationship between the intervals inside the frame time ~I and e~ ano the intervals outside the frame time (S anO R~. As consequence, the formal~sm predicts that temporal adverbials w~ich are in the scope o~ a frequency adverbial (: the e-specifiers~ cannot refer ba~K to the speech t~me or the eeference time: * Rei(e,S) and * Rel(e,R~, gooo p;ece of evidence for this hypothesis ~s pr~ioed by the WHEN-aoveroiais. In general one can distinguish two kinde of those adverbials : t~e relational ones, which express a relation Oetween the reference time and the speech time, such as "~esterday" a'nd "tomorrow", and the non- relational ones, which identify the location o~ an :nterval without any reference to the speech t~me, suc~ as "between 8 and 9" and "at two o clock". The interesting thing now is t~at only the latter adverbials can occur in the scope of a frequency adverbial. Compare :iI~ she arrived every day between 8 anq 9 e *(12~ she arrived every day yesterday e The fact that the relational WHEN-adverbials cannot occur in the scope of a frequency aoverb~al prcviOes some positive evioence ~or not inciuoln§ direct relations between e ano S in the formal~em. The chaln-like structure of the representation format Is, hence, i~nguistically motivated. Temporal Quantifiers The format Oeveloped for the analysis of cvclic iteratlon can also be useo for the analysis o~ the temporal ~uantifier$, such as "miway~", "scmetlmes", "never", "seldom" ano "often". The ~rmetion they proviOe is less specific than the ona p~ovioed by the period frequency aOverb~ais, ar, d t~s should be refiecteO in their representation. As a starting point I take the general ~ormat ~or the representation o~ sentences w~th a periodic frequency adverbial : ~ i [c(l.E~ & nI=~ & P(li > 3 e [cie,li &Mie) & AT(e,p)]] For a semantlc analysis of the temporal quantiflers this format has to be generalieeo. The most important change is the replacement of the universal ouanti;ier bv a variable : Q I C=(I,E) where Q can be any of the follo~ing quantiflers always 3 eometimes -3 never Few selOom, rarely, now ano then Many o~ten, frequently Most usuallv, mostly, generally .=,is sixfold dzvis:on is taken beer from Lewis ~1975). This analysis account~ for the anomaly of sentences like o ,13} we sometimes played chess every wee~ 3 ? (141 they often met every month Many (15p we always plaveO chess every week 9 These sentences are eemantlceiiy anomalous oecauee t~e sa~e ~ino o* In*ormation. namely the v~iue o~ ~. is epec~lec twice. This leaps to :~cons~etenc~ ~ ~13) and (14} where the Q- ve~ia~ie IB s~l~ to be both universal anO non- ~r;vers~i at tme same time, and it leaos to pleonasm in (15~ where the Q-variable is twice sago to Oe u~,iversal. The ne, t question is whethe," thP temporal quant!~iers introduce any extra-conqitions on those Intervals, ouch ms c~l,E), ~I=~ and P~i~. The f~rst of t~ese conditions appears to Pe relevant : the temporal quantifiers are ~ndeeo interpreteO wi~ respect to some given frame time. In ~x he was al~ays late "always" ooesnot oenote AL~ possible intervals. but onl~ all possibie intervals ~n the past. The conoit~on that the subintervals may no~ overlap does not seem to be relevant, though, in (16, quaOratlc equations are aIweye s~mple 273 the Instances for whlon "quadratic equations Pe ~imple" are true are no~ temporally ordereo at all. it, is m~gnt indicate, Ov the way, that the i- objects ~re not necessarily intervals, but rather cases or occasions wnlcn can but need no: be given m temporal interpretation (of. Lewis 1975i. The third conOition concerns the properties of t~e I-objects. In the case of the periodic • ,equency aOverblals the relevant properties concern the location or the length of the interval. In the case of the temporal guantiflers one could think of specifying a relevance conoiticn~ for a sentence like ~ he was always late ones not mea= that he was late at any possible occasion in the past, Put rather that he was late on all occasions on which his being late or timel~ could nave mattered. in Aqv~st, Hoepelman & Rohrer (1980) one can ~ind a proposal to incorporate this information in the semantic representation, but I will not adopt t~is proposal here, since the conditions o~ the ,non)relevance of the occasions are typicaliv determined O~ pragmatic factors, in ~:" he always leaves o~-~ twelve the relevant occasions (1) could just as well oe all occasions on which he leaves as all occasions on Wnlch ne leaves for work as a!i occasions on ~hish he leaves for watching the home game of nls ~avourlte footOaii team. As a result of the foregoing reductions ar~o changes the general format for analysing tempo, al cuantifiers looks as follows : 3 ~,E [ReI(R,S) & Pred(R) & ReI(E,R) & F'reoiE) & Q I [c(l~E) >/& 3 e [c~e,I; & M~ei & AT(e,p!]]] ,here O is replaced by any of {V, 3, "3, Most, Few, Many} M is replaced by some time adverbial which specifies the location or the length of e (if there is anv~ Habituality The sentences oiscusse~ so far all contain an explicit indication of iteration. !he presence of SL~Ch an IndlCatlOn Is, however, not necessary for deriving an iterative interpretation. Take, ÷or instance, (in~ he leaves at twel~e This sentence cannot only mean tnat he will leave at twelve, but also that he has the habit of leaving ~-* twelve. in the representation of interpretation the time adverbial specifiss the t~me of reference : the former "at twelve' 3 ~,E [:(R,S) & at twelve(R) & :(E,R) & AT(E. he leave~] E S R in the representation of tne habitual i~terpretation~ on tne other hand, tne time adverolal shouls be tal~en to specify the multiple e-tlme, for the sentence Ooes no~ report on one o~ his ieavzngs at twelve, out rather on several of socn :ea,es. As a representation of this interpretation I propose : ~,=st ; [ ~I.fJ , _ e Lc~e,I) ,~ at twelve~e & AT~e, he leave, l] R (19~ he leaves at twelve is t-eaten as synonymous with (20, he usually leaves at twelve If this is felt to be undesirable, one cam introGuce a special quantifier for marking habituaiitv, but at this moment ~ do not see an~ reason for SUCh a move. 274 The general format for the representation of habitual ~nterpretat~one Is, hence, 3 R,E [ReI(R,S) ~ Pred(R) & Rel~E.R> ~ Preo(E) Most i [c~I,E) > 3 e [c~e,I) ~ Pred(e) & AT~e,p)]]] The Assignment of Representations to Sentences On t~e basis of the given analyses one O:stinguls~ three kinds of sentence meanings : no iteration no ~ i [ l/periodic cyclic i~eration \ Q I [ ] \indefi-,te can is specified F is not specified Q is any of {~,3, "3,~ost,Manv,Few} The assl~nment of these meanings to particuiar sentences is fairly straightforward when the sentence contains a frequency adverOial or a temporal quantifier, but if there is none o~ those~ then the sentence is amOiguous Oetween a non-lterative and an habitual interpretation ~cf. the two interpretations of "he leaves at tweive"~. It, practice there are some oisambiguatlng ~. I* the basic proposition (p) denotes a state, ~r. er, the sentence can not have an habitual ir~erpreta~ior~ Compare :i;~ ne leaves at twelve ,21 ne is in jei! ~1~ can be interpreted as meaning that he has the naPlt of leaving at twelve, bu~ (21i cannot Oe interpreted ms meaning that he has the habit of bel=g in jail. ~, Certain verb forms can biock the Oerivation o~ one of t~s two possiole interpretations. Compare ~2~ he is drinking coffee 12]) he drinks coffee (22, can Oenote a single instance of drinking as wei" as a recent habit of him to drink:: coffee ~cf. in the sense of "he is. drinklng coffee nowadays"). (2;,, on the other hand, can only denote a habit; it cmnnot be used to report on a single instance o~ drinking. This demonstrates the need to distingulsn oi4ferent types of verb forms : the ones that will aiways elicit an habitual interpretation, the ones that block the derivation o~ an habitual interpretation, and the ones that admit both kinds of interpretmtions. The firs~ are unequivocall~ [+habitual], the second C-habitual[ and the last will be given the feature [+/-habitual]. THE INTERLINSUAL ANALYSIS OF THE VERB FORMS The Meanings of the Verb Forme In the previous parts i have presente¢ a formaliem for the representation of temporal information in sentences. This formallsm is especially deeigned for the anaiyeis of natural language, but not for the analysis o~ any particular natural language, such ae English, Dutch or Kiswahili. Its mmin purpose is to provide a conceptuall~ well-defined language for de;ining and comparln~ the ~eanings of te~poral expressions in different natural l~nguagee. In order to serve this purpose it is not s~fficlent ~o have a formalism, ~nouon. What is also needed is a general specification o4 now the semmntic representations relate to tnelr imnguage specific co~nterpmrts, i.e. the tenses, the temporal auxli:ries and t~e time aoveroials. The ÷orme~ two wiil furcner de caileO veto forms, For c {'is~ ~n~, those verb forms are summec up in the followlng rL~ie : Vero form ~. [+/-F'ast] (wi11+ir.f) (have+EP) ({be+iNS to+frill) ~e going T, hi_'¢ rule ylelds 24 (=2x2x2x3) 'verb forms. Their role in the semantic interpretation of sentences .:an easily de expressed in terms of the given formalism. They specify i. the relation Petween reference time anO speech time : ~eI(R,S) (= oeictic information) 2. the relation between event time and reference time : ReI,E,R) (= aspectual information~ 5. whether the sentence has an habitual and!or ; non-iterative ~nterpretaZlon 275 The meaning of a verb form can, hence, be representeO as a triple ~x,y,z> where x and v are substi~uteO for one of the possible dinar, -elations oe~ween intervals, and where z is one of the three poesible habituali~y values. The aame verb ~orm can, of course, have oifferent meanings and will, hence, Oe assoclateO ~th a set of such triples. The details o~ this association have been discL:ssed elsewhere~ at ]east for the x ann ¥ values ~cf. Van Eynde, des Tombe & Maes 1985i. In tnls paper I will only discuss the z values in some detail. The Mabituality Value A good starting point for demonstrating the relevance of the habituality value is provided by the following iist of sentences. They are taken from hess (1985). ~) a text editor makes modifications to a text file ~25) a text editor is makin~ modifications to a text file ~26) a text editor made mooiflcatione to a text file • 27~ a text editor has made modifications to a text file In L24) it is said "that a text editor ma~es modifications to a text file in general, almost by Oefinition. We might read this sentence in a re~erence manual" (Hess 1985, 10). In (25-27), on the other hand, it is said "that there is, or was, a case of a text editor mankind modifications to a text file. These remarks might ~e made by a system operator, watcnlng ~is screen' (lb.). Hess concludes from these observations that the quantifier of the subject is universal in (24) and e~:isten~ial in (25-27), However~ this conclusion does not foliow automatically. In terms of the formalism presented in this paper one could sa~ that (24) has an habitual interpretation, whereas the other sentences have a non-iterative interpretation, In the former case the existential quantifier of the subject will be in the scope o~ the Most-quantifier, whereas in the latter case it wlii not be in the scope of any non-existential quantifier, and this accounts for the difference in interpretation without havinq to postulate two possiole meanings for the indefinite article. Hess s examples are useful in this context, t~ough, because they clearly illustrate the roie of the vend for~ in the interpretation. Since it is the only variable part in the sentences, the ~ifferences in interpretation can only be ascribeo to them, more specifically to their habltuaiity value. ;or the assignment of an habltuality value to a given verb form one has to test whether it can or cannot elicit an habitual interpretation in some given context. In testing this one should i. always use sentences with a non-stative basic proposition, for i~ the latter is stative the sentence can never be habitual (of. supra) ; 2. pay attention to the other interilngual values of the verb form. The English simple present. for instance, is uneouivocally [+habitual] in its sim~Itaneoue meaning, but in its posterior meaning it can be [-naoitual] too (of. the non-iterative interpretation of "he leaves at twelve"~. The relevance of the [+/-Habitualitv]- distinction has so far only been demonstrated from a monolln~ual semantic point of view. It is, however, possible to give some translational evidence for this oletinctlo= as well. The relevant cas~s are tne ones where the corresponding verb forms have Oi~ferent habi:~allty values. A good example of this is the translation of the Dutch simple present in En~ilsh. The Dutch simple present can be both habitua} and ~on-hacitual in It~ simultaneous meaning : 28; hi~ o,'inxt aileen whisky <simui,y,~haOitual~ "he drinks only whisky' 29, Liji~, hij dr!nit k~4ie .,simul,/,-habltuai> "look, he Orinks co,fee" The English simple present, on the other hand, s always habitual in its simultaneous meaning unless in sentences Oee:ribing states, of course (~0~ he only drinks whisky <slmui,y,+habitua~. *~31) iooi:, he drinks ~o~fee <simul,y,-ha~itual Pot the expression of slmul~aneous non- iterativity one has to use She progressive : 32) look, De is crinking coffee As a conseoue~ce. ~e mapping of (29) to ~32) in~ol~es a non-~riviai tense replacement, and it il o~e of the merits o~ the given formaliem that it car handle this ir an lnteriingual way. 276 REFERENCES ~qviet Lennart, Hoepelman Jaap & Rohre? Ch~-istiah (19BO~, "Adverbs of frequency:, in Rohr~r (ed.), Time~ tense and quantifiere. Niemever. T~oingen, 1-17. Bruce Bertram (1972), "A model for temporal reference and its application in a question- answering program", in Artificial Intelligence 3, 1-25. Hess M~chael (I~B5), "How does natural language quantify ?". in Proceedings of the Secono Cmnferenc~ of the European Cnapter of the ACL, Geneva, B-15. Lemis David ~1975~ "Adverbs of ouantification", in Keenan (ed.), Formal semantics of natural language. Cambrioge University Press, Cambridge, ~-15. ~u!rk Randolph, Greenbaum Sioney, Leech Geoffrey Svartvik Jan (1972J, A grammar of contemporar~ English. Longman~ London. Relcnenoach Hans (1947~ Elements of symDollc logic. University of California Press, Berkeley. Stump Gregor. ~19BI~, "The Interpretation of frequenc~ ~ adjectives", In L~nguistics ant ~nilosophy 4. 221-257. Van Eynde Frank~ des Tombe Louis & Maes Fons ~1985)~ "The specification of time meaning ~or machine translation", In Proceedings of the Second Conference of the European chapter of the ACL, Geneva, 35-40. Van Eynde Frank (1987), Time. A unified theory of tense, asoect and Aktionsart, An internal Eurotra Ooeument (78 pages). Leuven. 277 . ITERATION, HABITUALITY AND VERB FORM SEMANTICS Frank v~n Eynoe. University oi Leuven Mar=a-Theresiastraat, 21 3000 Leuven Belgium ABSTRACT The veto forms. C-habitual[ and the last will be given the feature [+/-habitual]. THE INTERLINSUAL ANALYSIS OF THE VERB FORMS The Meanings of the Verb Forme In the