目前正在读机器翻译文章，其中一小段一直读不懂，恳请版上前辈给予指教。 Error classification and annotation is carried out when the focus is on the understanding the types of errors produced by an MT system and their frequency. An example of an error typology for the evaluation of MT is proposed by Vilar, Xu, d'Haro, and Ney. This form of evaluation was particularly useful when the dominant MT paradigm was rule based; that is, it was possible to "code" linguistic rules for the transfer of words, phrases, and grammatical structures from one source language into a target language. The use of error typology for the more recent data-driven or statistical machine translation is more limited because, in this case, the nature and volume of the data, as opposed to formal lingusitic rules, dictate the output to a large extent. error classification指将机器译错的部分分类，例如missing word, incorrect word order... 等。这一套做法不适用於statistical machine translation (SMT有学习能力，给予几组翻译译文对照，机器进行分析，得到某种规则或公式，接下 遇到类似的翻译，就有能力翻得出来。 我看不太懂的地方在the nature and volume of the data dictate the output to a large extent。依照SMT操作模式，译文产出与否，DATA量做为规则归纳来源， 所以很重要，但和nature (资料种类?)什麽关系？ --

※ 发信站: 批踢踢实业坊(ptt.cc), 来自: 36.235.13.246 ※ 文章网址: https://webptt.com/cn.aspx?n=bbs/Linguistics/M.1553264337.A.AE3.html