Machine Transcoding

There are grounds for questioning whether translation done by computer should resemble closely, either in process or in product, translation done manually. Several workers in machine translation have proposed radical departures from the naively adopted goal of simulating manual translation mechanically. Because they saw a similarity between their mechanically feasible output and the pidgin languages that have arisen naturally in some parts of the world, they dubbed their proposals 'pidgin translation'. This paper explores a more rigorous approach, which may be described as follows: texts in one language are encoded into words and other morphs taken from a second language, but this is done without disturbing the syntax of the source language. The term 'transcoding' is coined to denote that process. Transcoding results in a product in many ways similar to earlier types of pidgin translation (and incidentally more like natural pidgin languages than the latter were), but is founded on the principle that no information-bearing elements at all may be lost in the process. This principle is embodied in two strict guidelines: I. Each and every grammatical morph in the original must be reproduced in the transcoded text by a code morph; the linear order of elements is not to be changed. 2. The transcoding dictionary (the codebook) must be so constructed as to preserve all the lexical contrasts inherent in the source vocabulary. These guidelines are investigated and supplemented, and their effects are predicted with regard to (1) the output, (2) the user's ability to interpret transcoded material and to learn to interpret it better, (5) the possibility-of employing transcoding to make rough drafts for the human professionals in translation agencies to revise. Sample passages of transcoded Arabic and transcoded Chinese are presented, following which some comparisons are made between transcoding and pidgin translation, machine translation as practised or envisioned, and manual translation.