A Polish-To-English Text-To-Text Translation System Based On An Electronic Dictionary

The process of creating a Polish-to-English electronic dictionary destined to be used in computerised text translation was performed out in the following steps: I. The preparatory phase. In this phase classification files of the inflection of Polish words as well as the coding of Polish and English inflection paradigms were prepared. 2. The phase of creating the dictionary of canonical forms. This phase was carried out by lexicographers aimed by an interactive computer application. Each entry in the dictionary is supplied with inflection codes of its Polish and English parts as well as other syntactic-semantic information. The inflection code of the Polish part of an entry is a reference to a set of inflection endings stored in one of classification files prepared in phase 1. The format of the inflection code of the English part is "selfconstructive", i.e. it enables the generation of appropriate inflected forms from a canonical form without the necessity of a time-consuming look up of any classification file. (Designing a "self-constructive" code for a highly flexional language like Polish would have been a complex task). 3. The phase of generating the SGML.type dictionary of inflected forms. This phase is executed automatically. The inflected forms are generated on the basis of Polish inflection codes attached to all entries in the dictionary of canonical forms. Only inflected forms of Polish words (phrases) are created in this phase. Each form inherits the syntactic-semantic information from its canonical form. English equivalents of Polish inflected forms are not derived. The derivation of an appropriate English form is left to the morphological synthesis in the translation process. Storing the dictionary as an SGML-type document aims at comfortable browsing of its contents as well as facilitating its use in an application other than the POLENG translation algorithm. 4. Converting the dictionary into two modified finite-state automata. This phase is executed automatically in order to optimise the access time. The first automaton stores single words. Its alphabet coincides with the Polish orthographic alphabet. Reaching a terminal state of the automaton is equivalent to finding a Polish inflected form in the dictionary. Whenever a finite state is reached, references to the table of morphological features and the table of canonical forms are obtained. Due to the references, all morphological and syntactic-semantic information as well as the canonical forms of the English equivalents of the found word are achieved. The second automaton stores the lexical phrases. The alphabet of the automaton is the set of identifiers of the words stored in the dictionary of single words. This means that the process of searching a phrase is executed "word by word" (in contrast to the "letter by letter" search in the automaton of single words).