THE TRANSFER OP FINITE VERB FORMS IN A MACHINE TRANSLATION SYSTEM Bente Maegaard Institut for anvendt og matematlsk lingvlstik, K~benhsvns Universltet NJalsgade 96, DK-2300 K~benhavn S, De:m~ark This paper is based on work done Jointly by Hanne Ruus, Ehbe Spang-Hanssen and the author, all of the University of Copen~ hagen. The work was done within the framework of EUROTRA and sponsored by the Commission of the European Community. Work on this paper was begun whilst the author was at ISSCO, Genev a.

Throughout this paper we shall imagine a machine translation system which performs the following: correct analysis of source language, correct structure of target language sentence, correct choice of words in the target language. In this enviro~.ent we shall discuss some aspects of the problem of constructing the correct finite verbal form. The languages involved are the languages of the European Community. It is obvious that it is not possible simply to use the "same" morphological form in the target language as in the source language: verbal tenses are not the same in all languages (not even in languages as closely related as those considered) and even where the same tenses exist they do not necessarily have the same distribution. So a mapping has to be defined.

AS EUROTRA is in its conception multilingual this mapping should not be defined on a binary basis (between pairs of languages), but should rather consist in a definition of some basic and global interpretations of the morphology of verbs in

the languages considered. - 190 -

The study can.not be restricted to morphology proper. Meanings that are expressed morphologically (by auxiliaries or flexives) in one language might be expressed lexically (by the use of particular lexical units) in other languages. As an example~ take the progressive aspect in English: En. they are ~it_~_i~_n~ -~ Fr. ils 4crivent / ils @ont en train d'4crire, --~ --~ Danish de skTiver / de sidder cg skriver / de er i f~rd wed at ~krlve.l.e. the progressive aspect which is expressed in English by morphological mesns, is in French and Danish either not expressed or expressed by certain lexical units.

It is very important to realize the impact of this, There is often a tendency to think that in the analysis of language a borderline can be drawn between morphological information and lexioal information. This is true to s large extent when only one language is considered, but it is certainly not true in a multilingual environment. Roughly speaking, the system should take into account all pieces of information which can be expressed morphologically in at least one language. The propose~ e.yetem for the representation of verbal forms.

The system as such will contain the global information, i.e. the information which has to be transferred to other languages. In this paper only the transfer of active finite verb forms with respect to tense will be considered, i.e. modality, voice, etc. is not taken into account.

Apart from this global information the system can for each language be expanded with all kinds of local information which is useful for analysis or for generation. Some morphological phenomena are only grammatically bound and do not carry any semantic meaning that should be coded for the use of other languages, e.g. the use of the subjonctif in French subordinate clauses following ll faut que. Information about this subjonctif can be regarded as internal to French and should not be expressed in the global system.

The variables proposed to describe the semantic content of tense and aspect are the following (Note that tens_.._ee is - 191 used in a very general meaning. Tense is determined by values of time and aspect. It should also be noted that the use of the concepts aspect and Aktionsart has been de£indd with respect to the languages involved and consequently is not necessarily the traditional one): TENSE ::s (AXIS, PROSPECTIVE, RETROSPECTIVE, ASPECT,

AKTIONSART ) AXIS :'- l~C \[ TU~C PROSPECTIVE z -- PROSPECT- I PROSPECT+ RETROSPECTIVE : := RE~RO- I RETRO+ ~ CONTINUOUS I """ ASPECT : :. AKTIONSART Suggestions AKTIONSART For adverbs a value TIMEWI~lq.

In the paper the use of the variables and their values will be explained. Here only a few remarks can be made.

1. The system is viewed as operating in all languages on two time axes, one for the tenses that are anchored in the present, and a parallel one for the tenses that are anchored in the past (NUNC and TUNC, respectively),

2. The RETROSPECTIVE value CONTINUOUS is used for events that are still going on: F~. il demeure la depui~ un mois end En. He has l~.ved there fo r" a month will get the same values: (I~NC, PROSPECT-, CONTINUOUS, null, DURATIVE).

3. More values for the RETROSPECTIVE variable is to be foreseen.

4. In English it is not possible to have the prefect tense with an adverbial of the type TIMEWHEN. So a rule of the generation ~smmar of English will say that (~UNC, PROSPECT-, RETRO+ ..,..) + TI~WHEH -~ (TUNC, PROSPECT-, RETRO-,..,..)

5. Certain verbs in English tend to prefer the progressive - 192 nul._.~l ~ EMPHATIC PROGRESSIVE ::= DURATIVE ~ NON-DURATIVE of other values (for English, for the examples

below) : ::= DYNAMIC ~ STATIVE or perhaps rather

ACTIVITY VERB ~ PROCESS VERB I TRANSITIONAL

EVENT VERB aspect. Consequently the AKTIONSART must be expanded as to be able to express these facts:

A simulation of the system made on some Community texts shows good results.

A text is shown in the appednix An French (source) and in English. Both versions are official Community texts. For each verbal form which is finite in French and has a finite equivalent in English is shown how the system works. a adopt ~ (NUNC, PROSPECT-, RETRO+, null, NON-DURATIVE) +

. ~u~.WH~N --@ a ~t___~ (~NUNC, PROSPECT-, KETRO+, null, DURATIVE) +

+ TIMEWHEN ---~ wa.~s

--~ ros_._e a au~nent~

(NB. The TIMEWHEN value from the previous sentence must

be used, otherwise we get the prefect tense) a @t.___~ -~ wa_~s (because of TIMEWHEN) ont ~t6 -@ have been ont baiss~ (NUNC, PROSPECT-, RETRO+, ..., DURATIVE) ÷

+ TRANSITIONAL EVENT VERB -9 have been fallin~ ~'est d~t~rior@e as above, with the AKTIONSART being PROCESS

VERB --> has been deterioratin~ ~ont tombte a. (NUNC, PROSPECT-, RETRO+, ..., NCN-DURATiVE) -9

-~ have dropp@d to

(NB. the English version in fact has dropped to, but as

there is no TIMEWHEN this is not possible to obtain

with the system. However it is possible to choose

both translations from the French original.

Note furthermore, that tombe_.__~r in the meaning "decrease"

is DURATIVE, while tomber~a is NON-DURATIVE).

To our knowledge no other system has been proposed which works for more than one language. Reichenbach's system is made for English and shows deficiences even for English. Bruce "s system repairs these deficiences and works for English, but relies heavily on the structure of English verbal forms and cannot easily be brought to work for other languages as well. - 193 The system which we have proposed seems promising, but still a good deal of work is needed to define adequate values for the different variables. Reference s Bruce: A Model for Temporal References and Its Application

in a Question Answering Program. Artificial Intelligence

1972, vol 3, PP 1-25. Greenbaum, Quirk, Leech, Svartvik: A Qramm~r of Contemporary

E~lish. Longman. London 1976. Maegaard, Ruus, Spang-Hanssen: Investigation of the Verbal

S2stemfor.EUROTRA, Report for the EEC, 1981. Reichenbach: Elements of S~mbolic LoKic. Macmillan. New York

1947, pp 287-298. - 194 APPENDIX Charbon 2oi.96. La Commission a adopt6 le 19 avril 1979, apres consultation du Comit4 consul%atif OEOA (Bull. CE 3-1979, point 2.3.75), le rapport annuel sum le march~ charbonnier de la Commumaut4 en 1978 et see perspectives pour 1979 (JO C 120 du 14.5.1979). Avec quelque 287 millions de tonnes, la consommation de charbon a ~t~ environ la m~me en 1978 q~'en 1977, oependant q~e. la oo~atlon du charbon destin~ a la production d 61ectricit6 a au~nent~ de 5~ environ pour atteindre 159 millions de tonnes environ en 1978 contre 151 millions de tonnes 1 annie pr~c~dente. Avec 238 millions de tonnes environ la production d~ charbon de la Communaut~ en 1978 a 4t6 inf~rieume de l~ a cells de 1977; elle ee caract~rise surtout par une 16g~re augmentation de la production au Royaume-Uni et par de l~ger, reculs ailleurs. Les hausses de prlx du charbon de la Communaut~ ont ~t~ modifies. Toutefois, comme les prix du march~ mo~xprlm~s dans la p~upart des monnaies de la Communaut~ ont balsa6 du falt de 1 affaiblissement du dollar am~rioqin, la position conourrentielle du charbon de la Communaut~ s est d~t~rior6e oonsid~rablement. N~amnoins, les importatfons de charbon en provenance de pays tiers ont dimlnu~ de 1,5~ et sont tomb~e~ a 45 millions de tonnes, dont plus de la moiti~ de charb6n vapeur. Coal 2.1.96. ~ (Bull.

After consulting the ECSO Consultative Committee EC 3-1979, point 2.3.75), on 19 April the Commission

the annual report on the Communi~ coal market forecasts for 1979 (OJ C 120 of 14.5.1979). At around 287 million tonnes coal consumption was about same in 1978 as in 1977, while consumption for-6Tectrlcity generation rose by about 58 to reach roughly 159 million 1978 as oom~-~d with 151 million t the previous year. ity coal production in 1978"I~ma was about 238 million t or 18 lower than in 1977, the features being a slight in production in the UK and slight drops elsewhere. Price increases of Community coal have he qn moderate. However, as world market prices expressed ih most Community cttrrencles have been fallin~ because of the weakening of the US dollar, the competitiveness of Community coal has been deterioratin~ considerably. Nevertheless, coal imports from third countries ~o bY about 1.5% to 45 million t, of which over half was alo - 195 in 1978 the

t in Communsome

rise