File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/88/c88-2089_metho.xml
Size: 17,445 bytes
Last Modified: 2025-10-06 14:12:07
<?xml version="1.0" standalone="yes"?> <Paper uid="C88-2089"> <Title>AN INTEGRATED MODEL FOR THE TREATMENT OF TIME IN MT- SYSTEMS</Title> <Section position="1" start_page="0" end_page="0" type="metho"> <SectionTitle> AN INTEGRATED MODEL FOR THE TREATMENT OF TIME </SectionTitle> <Paragraph position="0"/> </Section> <Section position="2" start_page="0" end_page="0" type="metho"> <SectionTitle> Abstract </SectionTitle> <Paragraph position="0"> One of the ways to achieve a good translation of verbal forms is the morphosp~tactic approach, which consists in a function pairing the different morphological tenses that occur in a given language with the tenses of the other language. Complicated rules must be established to calculate the right pair for an expression, because of the amount of discrepancies that differmnt languages show with respect to each other.</Paragraph> <Paragraph position="1"> The way we have chosen to deal with this problem is, conversely, the projection of the different values coming from verbs (type, processivity, morftense, morfaspect, moodrequire~nent), from adverbs, prepositional phrases and temporal N~S (deixis, espect, iteration), and from subordinate conjunctions (aspect, moodrequirement).</Paragraph> <Paragraph position="2"> All this information permits to obtain a final value for aspect .~nd tense for the whole sentence, which later on i8 percolated, not only to the verb node, but also to the the rest of elements conveying information.</Paragraph> <Paragraph position="3"> Our proposal relies on the fact that tense/aspect calculation is relevant not only for a good translation of verbs, but also for a good translation of adverbs, PPs, temporal I~.~ and conjunctions, as we have intended to demonstrat~ ~n this paper.</Paragraph> </Section> <Section position="3" start_page="0" end_page="0" type="metho"> <SectionTitle> L Introduction </SectionTitle> <Paragraph position="0"> '1~li~ article deals with a methodology to achieve the right t~s~lsiation of temporal expressions by giving account of the temporal reference and temporal relations in/ between sentences. The task to accomplish is to translate sy,%tactic marks into semantic values that decide/ reflect the aspectnal value of the sentences.</Paragraph> <Paragraph position="1"> Fo3:' o6r treatment of time and aspect we draw on the work of Eamp \[1979\] and Partee \[1984\] who have argued for taking statos and events as primitives and relations of precedanco ~md overlapping between them. The ordering relation ~t~n events is crucial for deciding about the aspect of the sentences involved.</Paragraph> <Paragraph position="2"> ~hn presemt proposal presumes an analysis end a geme~ation t!omponent that deliver a set of S-trees whose leaves cor~:espond to words. The pre-terminals have morpbosyntaC/:tic and ~elational information. As usual, features ar~J percolated and nodes get features assigned.</Paragraph> <Paragraph position="3"> ~he time/ aspectual problem is dealt with under the perspectPS~* of MT with the aim of sketching a system that can be i~lemented independently of the particular formalisms of different MT-systems.</Paragraph> <Paragraph position="4"> To outline a general model for the time/aspect calculation in MT we subsume a system with PSG rules that obtain some sentence structure with no regard to a specific grammar type; it could be an augmented PSG, as in METAL, or some kind of deep syntactic structure, as it is the case in Eurotra.</Paragraph> <Paragraph position="5"> The problem is the well known fact that translations of temporal expressions in NL does not involve a simple mapping of tenses and adverbials. We could just compare Spanish, rich ill aspect and tenses vs. German or English. That is, a MT dealing with Germanic and Romance languages is concerned with different parameters for each language; the whole practice in MT systems is to translate morphological tenses, and syntactical values into reference times that include events or stutes \[Partee 1984\].</Paragraph> <Paragraph position="6"> Once the set of factors and values relevant for the specification oi! time, aspect, and mood has been defined, a calculation or unification in the form of rules fixes the values that must be generated in the target language. Such a system must &quot;preserve&quot; the same &quot;semantic interpretation&quot; for different language realisations, in order to be able to generate the right morphological tenses and periphrasis from language to language.</Paragraph> </Section> <Section position="4" start_page="0" end_page="437" type="metho"> <SectionTitle> 2. Aktionsart </SectionTitle> <Paragraph position="0"> Vendler's definition of Aktionsart \[1967\] relies on a classification of verbs into sentences. However, we start from verb primitives that are categorized (Verb Type) into: - stative : to be in love, stand, know..</Paragraph> <Paragraph position="1"> - dynamic : w~ite, arrive, love,..</Paragraph> <Paragraph position="2"> and we take an additional feature &quot;Processive&quot; to mark the differences between activities vs. accomplishment, achPSevement or quality/state vs. stance (R. Quirk 1985). The inherent type of these verb primitives is overridden during, the analysis by the morphological tense and by the syntactic functions, so that we finally obtain a value for the Aktionsart of the whole phrase.</Paragraph> <Paragraph position="3"> - write (ancribir): VT = activity, Proc --yes (a durative verb) - send (enviar) : ~ : activity, Proc --no (a punctual verb) The original aktionsart of the verb, durative activit,y becomes an accomplishment in the sentence &quot;I wrote a letter&quot;, In Spanish this information must be marked in order to choose the right past form.</Paragraph> <Paragraph position="5"/> </Section> <Section position="5" start_page="437" end_page="439" type="metho"> <SectionTitle> 3. Deixis </SectionTitle> <Paragraph position="0"> The calculation of the time value, which is always expressed as a deictic value relating the speech time to the reference time, is obtained out of the deictic values of the morphological tenses together with the adverbials present in the sentence. For instance, the Spanish present tense has the value \[simultaneous and posterior\]. If it al~ears with a posterior adverbial, the final value is posterior.</Paragraph> <Paragraph position="1"> The deixis assigment in Spanish is reflected in table i. The left side is the assigmant in contexts with no temporal transposition; the right side shows the assigmant for tense transposition. The context for tense transposition is given for knowledge and diction verbs.</Paragraph> <Paragraph position="2"> Table 1 : Deictic assigmcnt to the tenses in Spanish No Tense Transposition I Tense T~ansposition .................................. I ........................ Tense Deixis i Deixia .................................. I ....................... Present simul, post ~ simul, post factors: a\] the demostrativas end b) the temporal type of the singular noun ~'s when temporal desiguatora have the Deixia \[ant,post\]. When they appea~ with a plural definite a~ticle, this is an operator that overrides the original deixi8 values end yields the iterativa aspect.</Paragraph> <Paragraph position="3"> i.e. : &quot;el lunes &quot;\[past, ante\] (=&quot;on monday&quot;\]ve. &quot;los lunes&quot; liter |(on mondays) The delxis feature is the dianri~Linatlng factor to get the right transfer for &quot;lunes&quot;.</Paragraph> <Paragraph position="4"> 4 - Subordinate clauses : the deictic value of this adverbials is a calculus of the deictic value of the conjunction (if present), and that of the tense and of the mood, if present.</Paragraph> <Paragraph position="5"> There are two possible cases depending on the source language ( Germanic or Romance). For the first case, the axpectual value of the conjunction end the temporal value of the main clause select the mood of the subordinate clause. The deini8 value of the main clause i8 copied onto the subordinate conjunction; this value decides about the 8ubjuntive ~squir~ent of the conjunction. Let's take the conjunction &quot;till&quot; (Eng.), &quot;his&quot; (Ge~.), &quot;bests que&quot; (Spa,.) that has no deictic value. If the main clause is in future, &quot;till&quot; gets the deixis &quot;post&quot;, if it iS past than it' gets accordingly the deixis &quot;ante&quot;.</Paragraph> <Paragraph position="6"> The interesting fact is that, according to the obtained deictic value the conjunctim~, selects the subjunctive requirement. When the conduction has assigned in the lexicon the com~itunt of subjunctive requirement, this value is the most relevant. For example: transfer of the conjunction. Here some examples for Spanish end German: .,..deg .................. .,,deg,0,...degdeg ...... cuando \[enb. ~\] ...... als cuando \[pest\] ..... werm ~ando \[8imul\] ....... wnnn relent ras \[8imul, ante\] -- wahrend mientras \[post\] ..... solange seit \[ante\] ..... dase hace~ 8eit \[post\] ...... desd~ hacxa .... . .... o.o ............. . ..............</Paragraph> <Paragraph position="7"> ~x: &quot;mientza8 el trabaja, ?o t~no el sol&quot; &quot;waehrend er arheitet, aonne\[Pres\] ich reich&quot; * ~.</Paragraph> <Paragraph position="8"> &quot;mient~u el t~abaje\[au~j.\], yo sere / \[Fur\] loll=.&quot; &quot;anlange er arbsitet, bin \[Prns\] ich qlue~lPSch&quot; 4~ Aspect The a~pect.al calculus we propose relies on: - 'A) le~ical t~ of the verb pr~itives ~!) mo~phologxcal tenses of the verb forms .- :~) the presence of t~pornl adverbials.</Paragraph> <Paragraph position="9"> ~', d) the de~ee of definiteness of the sentence constituents Tht~ possible as~etual values are: \[inchoactive, dnrativo, tex~ative, perfective, retrospective, p~OSpecti~i\].</Paragraph> <Paragraph position="10"> T~ lexical tipoloT\] was already mentioned: the Verb '~oe (~))?astute, and the Proceseive Feature.</Paragraph> <Section position="1" start_page="438" end_page="438" type="sub_section"> <SectionTitle> 4.1 Calculus of the Aspect of the verb group </SectionTitle> <Paragraph position="0"> ~n order to cover all the aspectual meanings in Spanish ~e keep p~ogz~essivity as a value out of the aspectuel feature he,muse it does not appear in alternance with the othe~ valu~ss. ~ ~y to proceed in the calculation oPS the aspectual ~alue is the following: -t'irst: ~apture all the enpentual information coaain~ from the verb: - ~rb type & processivity feature -. verb fo~ (simple or with auxiliaries) -. second: the aspectual in formation of the adverbials.</Paragraph> <Paragraph position="1"> ~n~ aspectual assig~ent to the tenses in Spanish is shown in table 2.</Paragraph> <Paragraph position="2"> F.r arbeitet gerede \[prnx. \] The relevance of the presence/absence of the progresaivity can be shown when we translate into Greek. In this language the same aspect value generates different tenses according to presence or absence of progressivity. For instance, if aspect i8 retrospective and there is progressivity then the corresponding tense is Imperfect, hut if there is no presence of progressivity the corresponding tense for this aspect is Present perfect.</Paragraph> <Paragraph position="3"> The morphological tenses also imply monolingnal aspectual assignments. Here we will follow the proposal of F. van E~de. According to it, each language has a different distribution of aspectual features in tenses. However, here again we prefer to split up the mo~hological values according to two attributes : form end pro~fesivily form.</Paragraph> </Section> <Section position="2" start_page="438" end_page="439" type="sub_section"> <SectionTitle> 4.2 Aspeetual value of adverbials </SectionTitle> <Paragraph position="0"> Adverbs as well as prepositions and conjunotPSons have in the lexicon an aspectual value assig~ent. Temporal NP's as adverbials get the aspectual assigment from their dete~inents or quantifiers. Bare plurals or distributive quantifiera assign the value \[ iterativity\], while All-cuantifier8 furnish the NP with durativity, expressed in some languages as progressivity.</Paragraph> <Paragraph position="1"> &quot;EE a~beitet den ganzen Tag ..... Esta trabajando todo el dla For instance, iterativity must be delivered from the source language in order to generate the right tense even for punctual verbs, e.g. : &quot;un d~a I envio flores a Carmen&quot; = &quot;he once sent flowers to Carmen&quot; vs.</Paragraph> <Paragraph position="2"> &quot;envlaba cede d~a flore8 a Carmen&quot;= &quot;he sent to Carmen flowers every day&quot; Deterniners/quentifiers have aspentual values as: un \[parr\]; cede \[ iter\]; todaa \[iter); todo/a \[dur\] So the transfer of &quot;d~a&quot; is &quot;once&quot; if we are concerned with a perfeative HP-adverbial. However the transfer of &quot;d/a&quot; e8 'day' when the aspeetual calculus has furnished the NP with the feature \[dnr\] or liter\].</Paragraph> <Paragraph position="3"> In PP'a the final aspectual value is the result of the calculus of the aspect value of the preposition together with the daixis of the adverbial. For instance, the |~undary preposition &quot;deeds&quot; {since} with anterior adverbials yields a final terminative aspect, but with a posterior adverbial, or even with a anterior adverbial but together with another boundary expression (the final Point of the temporal} PSt yields a perfective aspect. Examples: &quot;dasda e.yer\[ente.\] t rabaja aq.~ ..... > teminati~ &quot;dearie a~ora traDajard en EUROTRA&quot; .... > perfective &quot;trabajo agu\[ desde el lunes hasta el jueves .... >perfective \[perf\] \[limitl--term\] \[limit 2--perf\] The resulting aspect value of this calculus must he attached to the preposition or adverb (overriding) the old values in order to get the right transfer.</Paragraph> <Paragraph position="4"> .... ......</Paragraph> <Paragraph position="5"> en \[perf.\] --> in for\[perf.\] --> en for \[dur. \] --> durante In subordinate clauses the process is the same. As we have seen above the deixis of the calculus of the main clause must be attached to the conjunction. Thus, the calculus of the aspect of the main clause must be furnished to the conjunction in order to be able to get the right transfer of the conjunction and also for generating the right tense in the subordinate clause.</Paragraph> <Paragraph position="6"> Another important factor in the calculus of the aspect is the syntactic type of the constituents in the sentences. These determine the aspectual properties (Dowry). To illustrate it we take the examples of Dowry(1986) &quot;John walked&quot; -- Juan paseaba \[activity\] &quot;John walked to the station&quot; -- rue * la esteciotn \[achievment\] &quot;John walked a mile&quot; -- anduvo una mille \[accomplishment\] As we can appreciate, the Spanish sentences have a different lexical verb form for each of these aspectual differences. Up to now METAL have different transfers according to the different verb frames. Eurotra-Germany proposes to disambiguate in the Interface Structure by means of semantic feature bundles. Besides the constituency factors of the syntax, another important fact is the presence/absence of bare plurals for the objects: &quot;lleva escribiando novelas tres a~os&quot; -&quot;she has bean writing novels for 3 Years&quot; &quot;sie schreibt Romane seit 3 Jahren&quot; vs.</Paragraph> <Paragraph position="7"> &quot;He wrote a novel&quot; -- &quot;escribi~ una novela&quot; \[Past~ per feotive\] &quot;He wrote novels&quot; --- &quot;escribla novelas&quot; \[Past, durative\] That means that the original verb value for prooessivity is cverriden according to the NP's that go with the verb. NP's with quantifiers assign the aspantual values so that punctual verbs express processes, or viceversa.</Paragraph> </Section> </Section> <Section position="6" start_page="439" end_page="439" type="metho"> <SectionTitle> 5. Process for the calculus </SectionTitle> <Paragraph position="0"> The calculus of time and aspect can be implemented either as: - a unification of features together with filters that catch or kill complex cases. (EUROTRA) - an assigment table gathering the possible TI~ and ASPECT configurations, and another table yielding the TENSE corresponding to each of the TIME/ASPECT configurations for the right target language generation.</Paragraph> <Paragraph position="1"> The calculation process delivers a final value as a result which must be percolated to the S node. This process has the characteristic of a functional composition, provided that the different values can be considered as operators applied in a hierarchical order. The dominant value is the adverbial.</Paragraph> <Paragraph position="2"> \[Adv. (post) \[V (simul,post) \] \] ---> S (post) \[Adv. (term) \[V (perf, dur) \] \] ---> S (term) The calculus provides a value for S, and S-subclauses, but also for all the temporals implied so that they can be properly translated into morphological values, or into lexicalized forms.</Paragraph> </Section> class="xml-element"></Paper>