The Utilisation of the Structural Type-Based Routines as a Homograph Resolution Method in the AMPAR System

The homograph resolution algorithm developed in the All-Union Centre for Translation of Scientific and Technical Literature and Documentation provides for analysis of English homographs without any ending or ending in -s. The forms ending in -ing and -ed, were involved only when checks run for those endings produced a reliable result and helped resolve homography or narrow it. What constitutes the basis of this work is the supposition that we need to create, within the framework of working theories like the machine translation by correspondences (MTC) model [1] simulating translation proper for a given language pair as a whole, certain special theories that describe the approach to resolving particular types of ambiguity, and which are based on the contextual determination method. These special theories should lead to identification of diagnosticating structures (structural types) determining the choice of this or that meaning in resolving particular ambiguities. The diagnosticating structures are in essence correspondences in which the interaction of various levels of language in resolving this or that type of ambiguity is reflected. Identification of such correspondences is a unique task whose solution ensures the optimal ordering of language facts in machine translation. Two components are usually identified in the contextual determination theory: 1. The type of text perception, i. e. the context is perceived not only as a certain sequence of symbols with certain dictionary information assigned to each of them, but also as a hierarchically organised combination of information levels. Those levels overlap, the task being to separate them by analysing the elements identifying those levels. 2. Typology of contextual dependences, i. e. the differential approach to the contextual dependence of each unit, identification of units whose meanings depend completely on the context, and also context-independent units,