Moore G W, Riede U N, Polacsek R A, Miller R E, Hutchins G M
Am J Med. 1986 Jul;81(1):103-11. doi: 10.1016/0002-9343(86)90190-7.
The feasibility of automated translation of scientific and medical documents remains controversial. This report describes a minicomputer-based German-to-English translation system (TRANSOFT) that employs word order rearrangement followed by word-for-word translation and disambiguation based on context. This translation system was applied to a computer-readable version of Adler's Knochenkrankheiten (Bone Diseases), which contains 118,604 words, with 10,216 distinct words in 7,211 sentences averaging 16.4 words each. The translation required 2,791 word rearrangement formulas, 78 percent of which were first used in the first half of the document. There were 2,392 occurrences of 12 potentially ambiguous terms, of which only 18 (0.8 percent) were not resolvable from the immediate context. As foreign language medical documents become increasingly available in computer-readable form through computerized typesetting, electronic publishing, and improved optical character recognition equipment, automated translation systems may provide a rapid and inexpensive means of obtaining draft translations.
科学和医学文献自动翻译的可行性仍存在争议。本报告描述了一种基于小型计算机的德英翻译系统(TRANSOFT),该系统采用词序重排,然后逐词翻译,并根据上下文消除歧义。该翻译系统应用于阿德勒的《骨病》(Knochenkrankheiten)的计算机可读版本,该版本包含118,604个单词,7211个句子中共有10216个不同的单词,平均每个句子16.4个单词。翻译需要2791个词序重排公式,其中78%首次用于文档的前半部分。有12个潜在歧义术语出现了2392次,其中只有18次(0.8%)无法从直接上下文中解决。随着通过计算机排版、电子出版和改进的光学字符识别设备,越来越多的外语医学文献以计算机可读形式提供,自动翻译系统可能提供一种快速且廉价的获取翻译草稿的方法。