Translation Accuracy and Translation Efficiency

[MT Summit III, Washington, DC, USA, July 1991] Translation Accuracy and Translation Efficiency Wanying Jin Computing Research Laboratory New Mexico State University P. O. Box 30001/3CRL Las Cruces, NM 88003 wanying@nmsu.edu April 22, 1991 Abstract ULTRA (Universal Language Translator) is a multi-lingua] bidirectional translation system between English, Spanish, German, Japanese and Chinese. It employs an interlingua] struc- ture to translate among these five languages. An interlingual representation is used as a deep structure through which any pair of these lan- guages can be translated in either direction. This paper describes some techniques used in the Chinese system to solve problems in word or- dering, language equivalency, Chinese verb con- stituent and prepositional phrase attachment. By means of these techniques translation qual- ity has been significantly improved. Heuristic search, which results in translation efficiency, is also discussed, 1 Introduction of the ULTRA system: ULTRA (Universal Language Translator) is a multi-lingual bidirectional machine translation system developed at Computing Research Labo- ratory, New Mexico State University. It includes English, Spanish, German, Japanese and Chi- nese. The system employs an interlingual struc- ture, i.e. an intermediate representation (IR) is used as a pivot through which any pair of these languages can be translated in either direction. Each language component works independently to convert its own language string from/into the intermediate representation regardless of the lin- guistic expressions in any other languages. Sym- metric grammar rules are employed in all lan- guages to perform bidirectional translation. The vocabulary consists of 10,000 word senses for each language. In addition, common linguistic information is encoded into the interlingual lex- icon specifications shared by all languages. The intermediate representation (IR) of UL- TRA developed by Farwell [Farwell & Wilks, 1990] is a data structure produced dynamically during the parsing of an input string. It con- tains syntactic, semantic and pragmatic infor- mation to represent what is said; how it is said; and why it is said during communication. The intermediate representation is hierarchically or- ganized. IR tokens which refer to definitions in Longman Dictionary of Contemporary English [Procter et al, 1978] are encoded for word sense disambiguation. An intermediate representation for a particular sentence: The printer is placed to the left side of the display. will be represented as an IR in the following form: [prdctn,[type,indpnt] , [class,dcl], 85 [form,fin], [prop, [type , indpnt], [class,dcl], [pred,[tense,prs], [aspect,simp], [mood,indic] , [voice,pasv], [pol,pos], [rel,[type,dyn], [s_case,agnt] , [o_case,pat] , [io_case,none], [s_class,human], [o_class,p_obj] , [io_class,none], [r_desc,place2_l]]], [arg,[g_rel,subj], [k_rel,agnt], [t_rel,top], [ent,[type,nrm] , [class,p_obj] , [agree,ts] , [det,spin], [quant,unq], [e_desc,printer1_3]]], [p_mod,[g_rel,oo], [k_rel,loc], [t_rel,none] , [ent,[type,nrm], [class,p_prp], [agree,tu] , [det,spin], [quant,unq], [e_desc,left_side_x] , [e_mod,[type,oo], [class,posr] , [ent,[type,nrm], [class,p_obj], [agree,ts], [det,spin], [quant,unq], [e_desc,display_x]]]]]]]. where the predication (prdctn) at the top level, basically, represents the sentential infor- mation and the syntactic structure, the proposi- tion (prop) represents the sentential information and the syntactic structure at the clause level, the predicate (pred) represents all functional in- formation of the verb, the argument (arg) rep- resents the information relating to a subject, di- rect object or indirect object, accordingly, …

Content truncated to 500 words