File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/00/c00-2103_metho.xml
Size: 23,732 bytes
Last Modified: 2025-10-06 14:07:14
<?xml version="1.0" standalone="yes"?> <Paper uid="C00-2103"> <Title>Granularity Effects in Tense Translation</Title> <Section position="3" start_page="0" end_page="713" type="metho"> <SectionTitle> 2 Present or Future? </SectionTitle> <Paragraph position="0"> In contrast to English, the German present tense is commonly used to describe both present and future happenings. One task in translation from German to English is therefore tile dismnbiguation of German/)resent tense to present time or future time.</Paragraph> <Paragraph position="1"> (1) present tense ~ future time -+ present time * This work was fimded by the German Federal Ministry of Education, Science, Research and Technology (BMBF) in the framework of the Verbmobil Project under Grant 01 IV 101 U. Many thanks are due to Profi H. Kamp attd K. Eberle. All errors are my own.</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.1 Temporal Orientation </SectionTitle> <Paragraph position="0"> A prominent factor involved in the choice between present and future time (Butt, 1995) is the temporal orientation of tile time adverbials that modi\[y the tensed verb.</Paragraph> <Paragraph position="1"> Only a limited set of time adverbials can refer to present time. Indeed, the set is so small that it can be enumerated. The adverbials can be fhrther subclassified according to other times they may refer to.</Paragraph> <Paragraph position="2"> * only present (now, at the moment) * also past (just, German eben) * any time (toda35 this week, in the meantime, tbr two weeks) All other time adverbials are incompatible with present time.</Paragraph> <Paragraph position="3"> (2) * On 19th November 2000, I sleep late. * Some adverbials only refer to past time (e.g. ~stm'da35 last week, formerIL recentl), two days ago).</Paragraph> <Paragraph position="4"> (3) * I will be here yesterday.</Paragraph> <Paragraph position="5"> * Others can only be used with filturc time (tomorrow, next week, soon, in PSour d~ys).</Paragraph> </Section> <Section position="2" start_page="0" end_page="712" type="sub_section"> <SectionTitle> 2.2 Verbs Immune to Temporal Orientation Effects </SectionTitle> <Paragraph position="0"> In some cases tile temporal orientation of adverbials sounds a false alarm: Even though an adverbial requiring non-present syntactically modifies the tensed verb, German present tense is translated as present (see examples (4) from the British National Corpus). The effect comes about because semantically the adverbial modifies not the verb's eventuality but one of the verb's objects.</Paragraph> <Paragraph position="1"> (4) a. 'lbmorrow i am ah'eady 1)lamfing a golf trip with the boss. (Verbmobi\] (:orlms) b. And w(&quot; wish trim the very best of lu(:k tomorrow in Birminghmn.</Paragraph> <Paragraph position="2"> (British National Corpus, lINe) ('. Another storm fl'om SE ... is expected here tomorrow. (BNC) The decisive factor seems to be the verb sense involved. Thus, for disambiguation a. list; of such '&quot;il\[llll\[llte&quot; vet'b senses ltlltSt be compiled. Such verbs can be modified by adverbials requiring 1)r(~sent and adverbials requiring non-1)resent at the Sallle time.</Paragraph> <Paragraph position="3"> (5) am Montag haben wit jo%zt noch etwas Zeit 011 ~/\]011(t~l.} r have we i~ow still some time but now we still have some time on Monday.</Paragraph> </Section> <Section position="3" start_page="712" end_page="712" type="sub_section"> <SectionTitle> 2.3 Temt)oral Orientation and Scope </SectionTitle> <Paragraph position="0"> Only the temporal orientation of wide-s(:ol)e adverbials is relevant for tense (lisambiguation (Butt, 1995). lrrequen(:y )tdverbials intercel)t the disambiguating etfe(:t (see (6)). It is (;herefore imi)ortan(; to evaluate only those a(lvcrbials that outs(:ope the fl'eqltelt(;y adverbial with widest scope.</Paragraph> <Paragraph position="1"> (6) a. John will be here on Monday.</Paragraph> <Paragraph position="2"> b. John is here on Monday every week.</Paragraph> <Paragraph position="3"> (:. John will |)e here in April every week.</Paragraph> <Paragraph position="4"> 3 Simple or Progressive? A distinctive feature of the English tense system th',~t is missing in German is the differentiation be(,ween siml)le and t)rogressive aspect. We t'o(:us here on 1;\]1(: usag(: of aspect in 1)r(:sent time. (7) present time -~ ,~'imple/)re,~'ent -+ l>rcs<mt t)rogres,vi~<'.</Paragraph> </Section> <Section position="4" start_page="712" end_page="712" type="sub_section"> <SectionTitle> 3.1 Linguistic N~etors </SectionTitle> <Paragraph position="0"> The factors involved have been thoroughly studted and classitied in the linguistic literature (Greenbaum and Quirk, 1990). So we confine ourselves to a short re.view here.</Paragraph> <Paragraph position="1"> State Present. Stative verb senses get simple aspect.</Paragraph> <Paragraph position="2"> (8) a. * We are having a house on Oxtbrd Street;.</Paragraph> <Paragraph position="3"> b. We arc having dinner.</Paragraph> <Paragraph position="4"> Habitual Present. A habit is a set of <:vents of the same type. In semantic tern:s, a habit arises from quantiti('ation over events. If the events extend indefinitely into the past and fi:ture., the lmbit is conceived as perutancu, t &quot;and simple aspect is used; if the events occur over ~ limited period of time, the habit is conceived as temporary and progressive aspect is appropriate. A frequency adverbial can be used to speci\[y tit(: (relative) number of occurrences of the event. Ge.ncral .\[acts cml be viewed as a special type of a habit. They arc always expressed in simple asl)ect (see (9)).</Paragraph> <Paragraph position="5"> (9) Because water boils at 100degC, water is boiling at 100degC in the pot.</Paragraph> <Paragraph position="6"> Instantaneous Present. Dynamic verb senst~s that ret'er to a single event with little or no (lur~tion oceun'ing at the Sl)eech time are exl)ressed with simple aspect. This type of \])resent is used in commentm'ies (10a), self commentaries (10b) and with performative verbs (10c) referring to speech acts.</Paragraph> <Paragraph position="7"> (10) a. Joe scores a goal.</Paragraph> <Paragraph position="8"> b. I enclose an apt)lication tbrm.</Paragraph> <Paragraph position="9"> (:. For I)ermission to tmblish this paper, the authors l;tm.nk the l)el):~rtment of Economi(: Develol)ment.</Paragraph> <Paragraph position="10"> Durational Present. Dynamic verl) senses denoting an incomph:te event with dm'ation get progressive aspect.</Paragraph> <Paragraph position="11"> (11.) a. We are looking at; March sixteenth. (Verbmobil corpus) b. This is looking interesting.</Paragraph> <Paragraph position="12"> (Verbmobil corpus)</Paragraph> </Section> <Section position="5" start_page="712" end_page="713" type="sub_section"> <SectionTitle> 3.2 Disambiguation </SectionTitle> <Paragraph position="0"> State Present;. Disambiguation requires definition and classification of all relevant verb senses according to stativity. When in a first ap1)roximation only the most fl'equent verb sense of each verb are considered, a list; of stative verbs can be extracted from a corpus.</Paragraph> <Paragraph position="1"> Habitual l?resent. The presence of a fiequency adverbial points to a reading of Habitual Present. Since every event can be construed as a general fact, general facts arc very difficult to identify and will be disregarded.</Paragraph> <Paragraph position="2"> Instantaneous Present. For disambiguation achievement verbs used in selfcommentm'ies and perfbrmative verbs need to be listed.</Paragraph> <Paragraph position="3"> Durational Present. Present events are usually regarded as having duration, so progressive is the default aspect for dynamic verb senses in the present.</Paragraph> </Section> </Section> <Section position="4" start_page="713" end_page="713" type="metho"> <SectionTitle> 4 Perfect or Not? </SectionTitle> <Paragraph position="0"> In a special case, Germm~ present tense can be rendered as English present perfect: In English, perfect is used to describe periods that begin in the past and lend up to the present; German uses a non-perf~ctive tense in this situation.</Paragraph> <Paragraph position="1"> (1.2) Wir leben schon fiinf Jahre in Amsterdam.</Paragraph> <Paragraph position="2"> We live ah'eady five years in Amsterdam.</Paragraph> <Paragraph position="3"> We have lived in Amsterdam for five years.</Paragraph> <Paragraph position="4"> Whenever a period is described that begins before and still holds at speedl time, a limitative time adverbial i is used. (This term is due to Bras (1.990).) This peculiarity makes disambiguation very easy.</Paragraph> <Paragraph position="5"> (13) a. Er ist erst zwei Wochen hier.</Paragraph> <Paragraph position="6"> lie is onlyt two weeks here.</Paragraph> <Paragraph position="7"> He has only been here for two weeks.</Paragraph> <Paragraph position="8"> b. Er ist nur zwei Wochen hier.</Paragraph> <Paragraph position="9"> He is only two weeks hcxe.</Paragraph> <Paragraph position="10"> He is here for only two weeks.</Paragraph> </Section> <Section position="5" start_page="713" end_page="715" type="metho"> <SectionTitle> 5 Definition of Granularity </SectionTitle> <Paragraph position="0"> For our purposes we model the time axis as the set of rational 2 numbers Q. An interval is then a pair of rational nmnbers <s, c), such that s < e. The d'uration of an interval is the distance between start and end of the interval (d~Lr({s, c)) = e - s). On the interval structure we define the relations of inclusion (143) and overlap (14t)).</Paragraph> <Paragraph position="1"> (1.4) a. (81,c1} C (82, c2) +-} 82 ~.~ s1 A ~1 ~ (22 b. @1,C1) 0 (82,('2) ~ 81 ~ (32 A 82 ~__ e 1 1Limigal;ive adverbials go with the prepositions since mid for plus temporal measure nouns in English; in German they occur with the preposition seit and as duration adverbials modified by schon and erst.</Paragraph> <Paragraph position="2"> '~Although natural mmlbers could have been used to(), rational mm~bers are convenient since they allow free choice of the unit.</Paragraph> <Section position="1" start_page="713" end_page="713" type="sub_section"> <SectionTitle> 5.1 Temporal nouns </SectionTitle> <Paragraph position="0"> A temporal noun denotes a set; of intervals. We define the granularity of a temporal noun formally as a pair of numbers specifying the minireal and maximal duration of its intervals (e.g.</Paragraph> <Paragraph position="2"> unit is a dw). The following relation is used to compare granularity values.</Paragraph> <Paragraph position="4"> If 7t 1 has coarser grmmlarity than rt2, then an interval of rt:l cannot be included in aD. interval of' ~7,2.</Paragraph> <Paragraph position="5"> (16) ,,2: > -+ Vii ~ r~q,i,e c n2 : il g i2) Consider the fbllowing definition. (17) A telnporal noun n has t;11o property of disjoh~tness iff Vi~, i2 E n : ~il O i9 Every calendar measure noun defines a partition on the time axis and has therefore the t)roperty of disjointness (e.g. hour, day, week, etelniti)O. Nouns functionally dependent on calendar mea-Sln'e l/ouns inherit the property (e.g. Monda35 ChristlnaS, holida:B 6:45, ...). Event and state nouns have the disjointness property, if the described intervals fimctiolmlly depend on one of the particil)ants and the pm'ticil)allt is definite. Thus, example (18) is &wiant.</Paragraph> <Paragraph position="6"> (18) * Tomorrow at 6:30 Jones will give a talk at every conference.</Paragraph> <Paragraph position="7"> \]if ~zl has finer granularity than a noml ~z~ which has the property of disjointness, then every interval in r~,l overlaps with at most one interval in n2 (we disregard the case where il joins two intervals in n2).</Paragraph> <Paragraph position="8"> (19) V/;1 E ')'1: I{ C/=: i', A ,:1 o ,:2 }1 -< 1</Paragraph> </Section> <Section position="2" start_page="713" end_page="714" type="sub_section"> <SectionTitle> 5.2 Functions on Granularity </SectionTitle> <Paragraph position="0"> Relational Nouns. Among the temporal relational nouns we distinguish nouns describing periods of definite length (e.g. quarter, t;hird, hall) f'ronl those that refer to periods of indefinite length (e.g. beginning, middle, end).</Paragraph> <Paragraph position="1"> In grmmlarity calculation, relational nouns contribute a factor. With relational nouns referring 1. Some 1)rcl)osil;ions do not alter gramfl~ril;y (e.g. /;Cml)oral location on, in, at, duration during, throughout,/br).</Paragraph> <Paragraph position="2"> 2. So:::(: pr(;posiCions des(:ril)c an oi)en-(md('d int;(;rval (e.g. bcibrc, at'tot, ti'om-ml, raM1). 3. S()ln(; preposil;ions requir(; a. d(;t;ailed del:cr null d(,,signal;e a long period more (tel;nil is given in the NP a rguln(,nl;. TIHI,% l;h(', rcl(,.vanI; l;(;mporal l)r('posil;ions map l;hc. n(mn gra.mflarii;y 1;o th('. n(',xl; higher level ()n a ccrI;~till Sca.lc (day nlonl;h year el;t,rnil;y).</Paragraph> <Paragraph position="3"> (':0) 1,efo,.(,, 6:; 0 1)(Zore Monday (< inOlli;h or y('.a.r) b(~fore Monday, lsl: of ()(:l;ol)(,w (< year) 1)('.for('. Monday, lst; of ()(:l;ol)cr 1998 (infinite) Detcrmin(trs. Two (:lass(;s of dci;(;rmin('xs (:alJ t)e disl;inguis\]md: Sl)(;(-ifi(: ((lolinil;(; or ind(:linil;(;) ~md qua,nl;iti(:a.l;ional (Ka.m t) and l/(;yh',, 1993). Sp(x'ific d(;l;(;rmin(;rs (to noI: (:hang(: gra:mla.rit;y. Q tmnl;ifi(;rs, how('ver, ('.xt('.n(l gramila.ril;y if il; is ('lem&quot; l;\]mt, ovc.ry inl:(;rval (l(:noi;c.d 1)y a. tX'ml)oral noun occurs only on(:(' wil;hin a. (:(;rl;ain l)eriod. (21) Mo,,d., y (,v(;(,,k) (;v('a'y beginning of a (;olff(~'r(!nc(; ((:onf:'renc('O O,V(:l'y bimont;hly m(;cl;ing (two monl;hs) BUT: (;very qum't;e,r of a y(;ax&quot; (ttu'c(: months, nol; year) Appositions. If temporal nouns form ~t (:onsl;ituent ((;.g. yo.vtcrday nt'tcrl~oo.u), the gram:\]a:dl;y of the. head noun is (:hosc, n (tyl)i(:ally l;his is l;h(, filw.r gra.nularity). Sill(:(; ini;(;rwds arc usually descril)oxt Oll S()llrC(. ~ }/,lid (;}/,l'g(;l; side, granularii;y iI~\[brInal;ion of \])oth source and l;arg(;t tOtal)oral nouns can 1)(, exl)loil;cd to achieve higher precision (e.g. Vormittag. morning).</Paragraph> <Paragraph position="4"> ll; can b(; shown thai; if an itc'ral;ive qua nl;ifier qua.ni;iii('.~ over i;h(' range of a funcl;iona\] concepi;, il; musl; oul;sCOl)e. I;he dom a.in quantifier (e.g. m .... c W t?~thcr oul;s('opt',s a. stude,ut; in r'veW ti~thc,': ()f a .'-,'tlldcut) (Moran and lk,r(;ira, 1992).</Paragraph> <Paragraph position="5"> 'li;mporal ov(',rlap is a fllncl;ional conc(;pl; if the donmin inl;('.rva\] has Iin(,.r gramlla\]'il;y t;h;m l:he rm, ge inlx;,'va.l (see (19)). Ih'.ncc', if l;wo time a dvert,ials 'n,i (on Monday in (22)) and n2 (c'veW u'ec'k in (22)) modify the same ev(!nl;, nl has liner gramflm'il;y l;han '/~,2, and r~,2 is il;(;ral;iv('ly (tuanl;ificd, i;h(m ~*,2 Illll,ql; outs(:op(; hi.</Paragraph> <Paragraph position="6"> (22) .lotto visil;ed .lane every w(!('k on Monday.</Paragraph> </Section> <Section position="3" start_page="714" end_page="715" type="sub_section"> <SectionTitle> 6.2 Temporal Quantification </SectionTitle> <Paragraph position="0"> Temporal quanl;ificatioll has a curious i)roi)(;rty.</Paragraph> <Paragraph position="1"> Prcl)o,~il;iona\] t)hrases are gonel'ally l;real;(',d as inWrsc('l;ive modifiers to the head insl;m:c('. (AIsh~uvi, 1.992). II:' we analyse sent;on(:(; (23~) i.</Paragraph> <Paragraph position="2"> l;his v(:in, w(; gel; (231)) as h)gical r(;1)rescntat;ion: The des('ribcd evenl;s arc situal;ed in May and in ()V(}l'y x, VO, Ok.</Paragraph> <Paragraph position="3"> (23) a. In May John visited ,Jmm (:very wed(.</Paragraph> <Paragraph position="4"> 1,. ?,,,: a v,,,, : ,,,(,(,k(,,,) visit(c, John, .lmm) A c C. 'm, A (' C 'u; On closer insl)ecl;ion wc s(;(; thal; I;h(; repr(;s(;ni;al;ion is (:Ollla'adictory. Take an arbil;rary week, say in April. Tlw.n I;he fornmla ass(;rts that t, her(; is a visit; in thi,q w(,ek (i.(;. in April) and ill May, which is inconsistent. Thus, the formula should not quantify over weeks in general but over weeks in May: An inclusion restriction to the wide-scope adverbial is needed in the narrow-scope adverbial (Kamp and Reyle, 1993).</Paragraph> <Paragraph position="5"> (24) ~m: May(rr~) A Vw: week(w) A w C rn, ~e : visit(e, John, Jane) A e C m A e C_ w Since temporal quantification requires that a narrow-scope iterative adverbial be included in the wide-scope adverbial, configurations are excluded where by conceptual knowledge the narrow-scope adverbial al cannot be included in the wide-scope adverbial a2. By theorem (16) this is the case if a,i has coarser granularity than a2.</Paragraph> </Section> <Section position="4" start_page="715" end_page="715" type="sub_section"> <SectionTitle> 6.3 Granularity and Scope </SectionTitle> <Paragraph position="0"> We have now seen two nlotivations for the prin- null ciple (25).</Paragraph> <Paragraph position="1"> (25) If ~,~ has finer granularity than r~2 and rt,2 is iteratively quantified, ~,1 camot have scope over ~t2.</Paragraph> <Paragraph position="2"> Let us now consider the tbllowing t)rinciple: (26) If 7/'1 has finer granulm'ity than 7~,,) and r~,l is iteratiwfly quantified, ~zl cmmot have scope over ?~,2.</Paragraph> <Paragraph position="3"> We are not in a position to tbrmally explain the principle. It, holds for at least all nouns with the disjointness property. Although in example (27a) the vq reading would make perfect sense (Jones is always on holiday), it is excluded. In contrast, example (27b) allows this reading (.Jones shuttles between conference sites). The principle is not restricted to temporal grmmlarity: In (27c) the V~ reading is excluded, too. (27) a. Every Monday, Jones was here in a month in which he was on holiday.</Paragraph> <Paragraph position="4"> b. Every afternoon, Jones gave a talk at a conference.</Paragraph> <Paragraph position="5"> c. On every page, 1%und something interesting in a paper I read.</Paragraph> <Paragraph position="6"> Taken together, the two principles assert that the grmmlarity ranking determines the scope order. In the dismnbiguation algorithm presented in section 7 we are mainly interested in the position of the highest (iterative) quantifier. So if every pair of time adverbials can be compared in terms of granularity, we have a procedure to compute this position. Comparison of granularity (1.5) is not defined if the granularity values overlap or m'e equal. Equal granularity is only possible with specific time adverbials.</Paragraph> <Paragraph position="7"> (28) * John came from every Wednesday to every weekend Other heuristics will have to come into play in case of overlapping granularity (see section 8 for further discussion).</Paragraph> </Section> <Section position="5" start_page="715" end_page="715" type="sub_section"> <SectionTitle> 6.4 Deictic Adverbials </SectionTitle> <Paragraph position="0"> Another factor for determining scope order is deixis. Some adverbials ~tre connected in their interpretation to the time of speedl (now, at the moment, next week, last week). Since time of speech is deictic (it depends on the context of utterance), hence defilfite, every function on it will also be definite. Deictic adverbials always get wide-scope position.</Paragraph> <Paragraph position="1"> Sentences in which deixis conflicts with granularity m'e deviant.</Paragraph> <Paragraph position="2"> (29) * Next month, I will be here every year.</Paragraph> </Section> </Section> <Section position="6" start_page="715" end_page="716" type="metho"> <SectionTitle> 7 The Disambiguation Algorithm </SectionTitle> <Paragraph position="0"> In the implementation, an underspecified senlant;ie representation formalism is used to encode the source analysis and the transfer result for the target (the Verbmobil Interface Term (FIT) formalism (Bos el; al., 1998), which is based on the theory of Underspecified Discourse Representation Structures (Ileyle, 1995)). The disambiguation heuristics of the system completely rely on local context. The most important features in local context are source tense, the predicate names of the tensed verb in source mid target, and the time adverbials inodifying the verb in source and target (Sdfiehlen, 1998).</Paragraph> <Paragraph position="1"> In a first step, the source and target representations are converted into an abstract representation, using the VIT transfer forlnalism (Dorna and Emele, 1996). All information irrelevant to tense resolution is removed. Since no full-fledged tense logic is implemented, information about temporal reference is discarded as well.</Paragraph> <Paragraph position="2"> Temporal adverbs are decomposed into prepositional phrases (e.g. .yesterday -+ on a past, day). In particular the following featm'es are extracted: (30) a. multiple classification of verb senses * stative or dynamic (for English only, cf. section 3.2) * 1)otentially pertbrmative/selfcolnmentary or not (for English only, cf. section 3.2) * &quot;immune&quot; or not (cf. section 2.2) b. temporal relation expressed by preI)ositions null e. temt)oral orientation of pret)ositions (on Monday), adjectives (the previous Monda:v) and determiners (this Monday) (of. section 2.1) d. granularity of nouns and adverbs (e.g.</Paragraph> <Paragraph position="3"> Cerma,) g ,,, ti gig (,,,ho>-(l<,j)) e. classification of determiners (quantiticational/specitic/alnbiguous) null f. classification of adverbs (frequency adverb or not) We now give an outline of the disambiguatixm algorithm for translating German t)resent. It is (:lear l;hat the algorithln is \]mm'isti(&quot; in lllglJxy Hst)ects, but in the absence of Colltext alld (lel;aile, d discourse anMysis it does not seem possible to do much better.</Paragraph> <Paragraph position="4"> 1. For every tensed verb, determine all time adverbials modii~ying it and collect them in the set TA.</Paragraph> <Paragraph position="5"> 2. Order the tinle adverbials in TA according to scope (of. section 6).</Paragraph> <Paragraph position="6"> 3. Let STA be the set; of all specitic time ad- null verbials in 5/'A (i.e. adverbials denoting a specific interval) not in the scope of a quantificational adverbial or frequency adverb (cf. section 2.3).</Paragraph> <Paragraph position="7"> 4. Perfect or Not? (cf. section 4) If one of the time adverbials in ETA is connected to the tensed verb over a linfitative relation, choose perfect; else choose nont)erf'ective tense.</Paragraph> <Paragraph position="8"> 5. Present or Future? (cf. section 2) Choose present if the verb is &quot;immune&quot; to temi)oral orientation (cf. section 2.2). Else unify the temporal orientation contributed by the time adverbials in ETA. In case the result excludes present time, choose fltture; else choose present (cf. section 2.1).</Paragraph> <Paragraph position="9"> 6. Simple or Progressive Aspect? (of. section 3) (a) If the tensed verb has a stative sense, choose simple (State Present).</Paragraph> <Paragraph position="10"> (b) If the tensed verb is used in a selfcommentary or perfonnatively, choose simple (Ilxstantaneous Present).</Paragraph> <Paragraph position="11"> (c) If there is an adverbial ill STA, choose progressive (Temporary Habit and Durational Present).</Paragraph> <Paragraph position="12"> (d) If there is a quantificational adverbial or flequency adverb in TA, choose simple (Permmmnt Habit).</Paragraph> <Paragraph position="13"> (e) Else choose progressive (Durational Present).</Paragraph> </Section> class="xml-element"></Paper>