This is an abridged English translation of my article in Chinese
学术翻译:直译或意译?(Scholarly translation: literal or free translation?).
In 1983, Xu Guozhang (许国璋), a well-known Chinese translator and linguist, described his translation of Bertrand Russell's A History of Western Philosophy as "译文力求醒豁,不按词典译义,而按词的文化史涵义翻译" (The translation strives for clarity. It does not go by the dictionary, but by the sense of the word in its cultural history.) He gave examples of translating "feudal" to "拥据领地(之诸侯)" ((the princes holding) the fiefs) instead of "封建", "anarchy" to "诸侯纷争" (dispute among the princes) instead of "无政府" because "当时无中央政府" (there was no central government back then), "adventure" to either "猎奇于远方" (adventure to places afar) or "探无涯之知" (explore the knowledge domain of no limit) depending on the context instead of "冒险", "antiquity" to "希腊罗马" (Greek and Roman, Greece and Rome) instead of "古代", "fame and beauty" in the context of the Renaissance to "享盛誉于邦国,创文艺之美" (enjoy fame among nations, create beauty of art and literature). He thinks this style of translation achieves the goal that the reader of the translation best understands the text.
No doubt Mr. Xu's translation is aesthetically superb and the sentences flow naturally. But I disagree with him on one point: Scholarly translation should be primarily literal, secondarily free or paraphrasing. Translation strives for fidelity, fluency, and elegance, strictly in that order (see Levels of translation quality proposed by Yan Fu: A small example). But in practice, various factors influence the weight of each of the three standards. Generally, literature demands higher "fluency" and "elegance" and so occasionally lowers the standard for "fidelity". But for a scholarly work, emphasis should be on "fidelity", with "fluency" not below passable readability. As to "elegance", it's something nice to have.
Take "feudal" not being translated as "封建" as an example. Today's consensus among Chinese scholars and dilettantes (or amateurs) is that the mainland China after 1949, actually even a few decades earlier than that, abused and misused the word "封建", which in its true sense only applies to the political and economic system prior to the Qin dynasty (221 to 206 BC), which is close to the feudal system in medieval Europe. So this Chinese word takes two meanings now, the proper one applicable to the pre-Qin China and the medieval Europe, and the misused one "applicable" to almost the entire Chinese history. There seems to be a reason for rejecting the translation of "feudal" to "封建" due to its overwhelming misuse in the past century. However, this "feudal" to "封建" word-mapping has been around for so long that a new invention such as "拥据领地(之诸侯)" would do more harm than good by confusing the reader, who knows exactly the intension and extension of the word "封建" in the context of European history. The same logic goes with translation of "anarchy" to "无政府", "adventure" to "冒险", and others.
Nevertheless, accepting a translation that started out inaccurate but ended with widespread acceptance and correct understanding is limited to ones without blatant errors. When Christopher Columbus arrived in America, he thought it was India and named the people Indians. Several major European languages follow suit, causing confusion between Indians in India and Indians as native Americans. But German is an exception, using two distinct words for these two peoples. So does Chinese ("印度人" and "印第安人", respectively). That's what should be done.
Based on the principle that scholarly translation should be primarily literal, secondarily free, I further propose that the ideal of scholarly translation is for the reader to be able to guess the word in the original language in order to minimize misinterpretation of the original text. This ideal is of course not to be completely realized, because a word in language A may map to multiple words in B, and vice versa. Note that it's not for every reader to be able to guess the word in the original language, but only for those that can read the language. Then, what's the point for a reader who doesn't know the language to guess the original word? And why would one already knowing the language read the translation? If the author in the source language used multiple words with almost the same meaning, it's best to use multiple different words in the target language in translation, so the reader, regardless his knowledge of the source language, is at least given a chance to appreciate the author's choice of words with subtle nuances.
Literal and free translations are two extremes. Literary translation is close to free, scholarly translation close to literal, and translation of science and technology cannot but be literal, or serious consequences might follow. Sometimes a literal translation inevitably leads to difficulty in the target language. A translator's note is to be added. This is a compromise between fluency and translator's responsibility.
Russell's History of Western Philosophy may be considered as a scholarly work. In spite of its wide readership and high popularity, the title does not necessarily reflect its content; it's more like a history of western thought. Most readers are not professional scholars, who would read classics and published research articles to write more research articles or books, but are educated general public, who read it to improve knowledge and cultivate critical thinking. Like any major encyclopedia, this book is in the middle between truly scholarly works and easy readings for laymen. Thus, Mr. Xu's free translation of this specific book is not to be blamed. Scholars or students in academic studies reading Chinese translations, however, would be advised to consult this book in English if they decide to cite references to the book without misinterpretation.