Откривање неконсистентности у обради вишеструке еквиваленције: уводна размишљања

Abstract

У раду се износе уводна размишљања о састојцима система за откривање неконсистентности у обради вишеструке еквиваленције у двојезичним рјечницима. Након што су релевантни појмови објашњени, у раду се идентификују и опримјерују типови неконсистентности у обради вишеструке еквиваленције. У сљедећем дијелу рада дају се начела на основу којих би такав систем за откривање неконсистентности био сачињен, а затим се описују обрасци претраживања рјечничког текста у циљу идентификације неконсистентности, што је заправо срж таквог једног замишљеног система. У завршном дијелу рада изводе се општији закључци на основу свега представљеног и дају назнаке могућег рада на овом замишљеном систему. Овај рад садржи уводна размишљања и замишљен је као први корак у развоју оваквог система подршке лексикографима који раде на двојезичним рјечницима. Нека од рјешења могу се у модификованом облику примијенити и на једнојезичне рјечнике.

The present paper is meant to be the first step in developing the mechanisms for automatic detection of inconsistencies in lexicographic treatment of multiple equivalence, the most prominent form of lexical anisomorphism. In addition to the obvious interest this phenomenon has been commanding in lexicography (e.g., Zgusta 1971, Al-Kasimi 1977, Yong–Peng 2007), important direct and indirect contributions to the study of multiple equivalence can be found in various other intellectual traditions from ordinary language philosophy (most notably in Frege 1982 and Puttnam 1975), to linguistic anthropology (Goddard 2012), translatology (Pym 2009), and second language teaching (Kramsch 1993). Sipka (2015) has identified the following common errors in the treatment of multiple equivalence in bilingual dictionaries: a. strategy mismatch, b. vagueness, c. lack of explanation or specifi cation, d. redundancy, e. underspecification overlaps, f. overspecification gaps, g. mislabeling, h. inadequate exemplification, i. language mismatch, j. inconsistent segregation, k. equivalent omission and l. overinterpretation. Automatic detection of the aforementioned errors in bilingual dictionaries assumes a feasible and reasonably accurate search pattern of some kind. A search pattern, in turn, needs to be efficient and its construction should not be overly time-consuming. More precisely, when deciding if it makes sense to develop a search pattern one should be guided by the following consideration: design feasibility, design resources complexity, comprehensiveness, and exclusiveness. The first parameter of a possible search pattern is a simple check if the error in question lends itself to automatic detection. If the answer to the first question above is positive, i.e., if the development of a search pattern is possible given the type of error, the next question should be what kinds of resources one would need to muster to design that particular search pattern. It may be so that gathering the resources for a search pattern is overly time consuming and that it makes more sense to search for the errors manually, by visually scanning the dictionary database. The next step in building the search pattern is the question about the proportion of the errors in question that will be detected by that particular search pattern. Ideally, the search pattern should detect all instances of treatment errors, but it may make sense to deploy the patterns that find most of them, especially if the design of the pattern is not overly time consuming. In this situation of underdetection, the cases that are not found by the search pattern may be addressed manually, left to be corrected in other editions of the same dictionary, etc. Finally, we want the search pattern to exclude the cases, which are not errors that we are looking for. Ideally, the output of the search patterns should not contain any non-errors (but, realistically, some level of overdetection can be tolerated). Possible search patterns for each of the aforementioned types of multiple equivalence are discussed in the central section of the present paper.