Segram - a program visualizing stages of grammatical competence
Bengt Sigurd, Gisela Håkansson, Johan Dahl · Lund University Publications (Lund University) · 2005
This paper will report on four machine translation projects dealing with restricted domains and controlled or semi-controlled language. The computer program Seatra translates marine forecasts between Swedish and English. It is in daily use by the Swedish meteorological agency (SMHI) and generally needs no postediting. The same basic program but with different lexicons is used in the program Maltra which can translate official Malaysian marine forecasts into English and Chinese. The program Reptra dealing with quarterly company reports and Receptra dealing with recipes use the same platform. All these programs (which can be used bidirectionally) will be presented and commented on, above all from a linguistic point of view. It is suggested that no more than 3 mistakes per 100 words be accepted if the postediting should not be too heavy, and this quality can be obtained in restricted-domain systems. 1 Seatra The Canadian program METRO (Kitti-edge et al. 1973; Goldberg, Kittredge & Polguere 1988) is well known and the task to translate marine forecasts from Swedish into English suggested by the Swedish meteorological agency SMHI didn't seem too hard. Early work is described in Sigurd, Lastow & Vavargard 1996. The program Seatra was developed in Prolog (LPAProlog http://www.lpa.co.uk), using a technique which is based on lexicons containing single words, multi-word phrases, discontinuous phrases and some grammatical phrases such as noun phrases, prepositional phrases and adverbial phrases. The program is designed to produce simple and fast translation in a restricted domain without offering deep syntactic and