Recently, an article titled “A Major Breakthrough in the Translation Field! As a Translator, I Now Understand the Concerns and Fears of 18th Century Textile Workers When They First Saw the Steam Engine!” has circulated among friends, causing many translators and foreign language students to express worries about the future of translation, suggesting that machine translation may replace human translators.
This article’s title is indeed quite sensational. Is it calling translators who might lose their jobs to go smash Google headquarters? After all, translation is a form of creative intellectual labor, which is different from purely physical labor (with no intention to belittle physical work). On the contrary, I believe that translators or foreign language students should welcome the arrival of new technologies and actively adapt to new trends, rather than blindly expressing unnecessary worries.
The current written translation market is mixed, with many believing that obtaining a Level 8 certificate is sufficient to become a translator, even offering rates of fifty or sixty yuan per thousand words, which has severely disrupted the order of the translation market. I personally believe that the CATTI Level 2 can serve as a screening tool, given that the pass rate is around 15%. The development of machine translation can positively contribute to the elimination of low-end translators, while the claim that machine translation will completely replace human translation is indeed exaggerated.
James Holmes, who established the independent discipline of Translation Studies, proposed the “Map” of translation studies, defining the scope of research. First, it is broadly divided into “pure theory” and “applied” aspects, with the latter further divided into three branches: “translator training,” “translation aids,” and “translation criticism.” Clearly, “machine translation” (MT) and “computer-aided translation” (CAT) can fall under “translation aids.”1 I have previously used the renowned CAT software Trados, which has powerful functions, allowing translation results to be recorded in a terminology database (MultiTerm) at any time. When the same terminology or similar expressions appear, it provides prompts to ensure consistency in terminology translation throughout the text, thereby reducing the repetitive labor time for translating terminology. Tools like “Google Translate” represent a frontier in translation research—post-editing, which involves modifying and polishing machine-generated translations to make them usable. In summary, “translation aids” can reduce a translator’s large amount of repetitive, meaningless labor, significantly improving translation efficiency, making it a boon for translators. This is also why many translation companies require translators to be proficient in translation aids, which represents a new trend in the translation industry; those who are clueless about translation aid software will inevitably face the risk of being eliminated.
The following discusses the reasons why machine translation cannot completely replace human translation. First, translation aid software is mostly used in scientific translations with a large number of terminologies and has limited applicability in literary translation. Second, machine translation relies on vast corpora and finds it difficult to handle terms that do not have fixed translations, such as China’s foreign publicity translations, which are authoritatively published after collective discussions by experts. Third, fundamentally, machine translation is based on the concept of “equivalence” or similar to the concept of “language pairs” in parallel corpora. Once the concept of “equivalence” is questioned, it fundamentally undermines the theoretical basis of machine translation.
This article points out the latest breakthrough in Google Translate—neural machine translation (GNMT), which has evolved from a phrase-based translation system to a sentence-based translation system, utilizing recurrent neural networks (RNN) to directly learn a mapping from an input sequence (a sentence in one language) to an output sequence (the same sentence in another language). According to the definition of mapping, this is a correspondence that must satisfy a one-to-one or many-to-one relationship, indicating that it essentially relies on the concept of “equivalence.” The Leipzig School’s Kade once explored four types of equivalence: one-to-one, one-to-many, one-to-partial, and one-to-zero, all indicating that “equivalence” requires a determined correspondence.2 Susan Bassnett and André Lefevere, in their seminal work marking the “cultural turn” in translation studies, Constructing Cultures: Essays on Literary Translation, mention machine translation in the preface:
Perhaps the most arresting example of this crumbling of the machine is the long retreat, and final disintegration of the once key concept of equivalence. Twenty years ago those in the field would ask themselves whether equivalence, too, was possible, and whether there was a foolproof way to find it if it were possible. Again, the underlying assumption seemed to be that there could be something like an abstract and universally valid equivalence. Today we know that specific translators decide on the specific degree of equivalence they can realistically aim for in a specific text, and that they decide on that specific degree of equivalence on the basis of considerations that have little to do with the concept as it was used two decades ago.3
By examining the history of translation after World War II, translatability and the development of machines have sparked interest in researching machine translation, while the concept of “equivalence” has made it a reality. However, as translation theory gradually shifts from a linguistic perspective to descriptive translation studies (DTS) and later achieves a cultural turn, the concept of “equivalence” has gradually declined. Now, “equivalence” has shed its early abstract, universally valid characteristics. In specific texts, translators can decide the degree of equivalence based on specific circumstances, but this degree is difficult to quantify, which essentially removes “equivalence” from a static linguistic perspective, allowing for the translator’s creativity.
In actual translation operations, many “inequivalents” or even “untranslatable” situations arise. For example, the translation of “culturally-loaded words,” such as “Jiangnan,” which carries specific cultural meanings, is difficult to explain even in Chinese, let alone find an equivalent in English. English readers cannot directly experience the beautiful imagery implied in phrases like “the grass grows in Jiangnan, and wildflowers bloom among the trees.” Often, transliteration with annotations is used, and transliteration is essentially “not translating.” Another common example is Eugene Nida’s translation of the Bible, where he pointed out that if “Lamb of God” were to be translated into Eskimo, it should be “Seal of God,” as they have never seen a lamb, thus achieving the same effect as in English. Therefore, translation sometimes requires considering cultural backgrounds for appropriate adjustments, which fixed correspondences cannot accomplish; it is a creative activity.
To achieve complete machine translation, one must overcome the hurdle of literary translation. To resolve this issue, there are two possibilities: first, machine translation develops to the point where it can perform literary translation; second, literature disappears, and naturally, literary translation ceases to exist.
Regarding the first possibility, literary translation is a creative activity and an art. Although there are corpus tools available to study an author’s style through word frequency analysis, there remains controversy over whether “style” can be translated. Mo Yan’s Nobel Prize in Literature can be attributed in part to the efforts of his translator, Howard Goldblatt, who considered the acceptability of Western readers and made significant cuts in the original text. Sometimes, translation requires creative rebellion, as French translator Gilles Ménage once said about the “unfaithful beauty.” As for Joyce’s Finnegans Wake, it is almost a display of writing skills, filled with fragments and deconstructed languages. Various translations have taken decades and still yield unsatisfactory results, which machine translation clearly cannot handle. Many times, the brilliance of literary translation lies in successfully preserving the multiple interpretations of the original text, which can be termed “many-to-many,” and this is precisely what machine translation cannot achieve. As for the second possibility, as Professor Wang Ning mentioned in a lecture, wherever there are people, there will be literature. Although literature, as a traditional elite culture, is currently under the impact of popular culture due to market forces, it does not mean that literature will disappear, and the boundaries between classical and non-classical works are not fixed. Unless, as described in the dystopian novel 1984, all past literature is destroyed, and literary works are produced using a novel-writing machine.
New perspectives in translation studies have far exceeded the capabilities of machine translation. For example, according to deconstruction theory, translation cannot completely replicate the meaning of the original text; excellent translations can give a “second life” to the original work, allowing it to continuously gain vitality through dissemination, traces, dislocating, and deviations. As for feminist theory, which emphasizes women’s subjectivity, strategies such as supplementing, prefacing and footnoting, and hijacking can be employed—for instance, deliberately changing originally non-feminine terms to feminine ones, thereby endowing the translated text with feminist consciousness. Additionally, post-colonial translation theory views translation as a tool to escape colonial discourse and reshape national history from political and power perspectives. There are also ethical issues in translation, such as whether users of machine translation possess the rights to attribution and copyright. If machine translation could completely replace human translation, even illiterates could become translators merely by clicking a mouse to operate translation software. From the perspective of English-Chinese comparative studies, Chinese is a paratactic language, where words or clauses do not require linguistic formal means to connect, and grammatical meaning and logical relationships are expressed through the meanings of words or clauses.4 Even if Google Translate can translate at the sentence level now, it still cannot determine the logical relationships between sentences. Thus, when translating the original text into the hypotactic language of English, it cannot automatically add necessary conjunctions.
In summary, while the development of machine translation will eliminate low-end translators, completely replacing human translators is almost impossible, unless humans can completely unravel the mysteries of the brain. Translators and foreign language students need not worry excessively; instead, they should actively learn new technologies, keep up with new trends, and strive to enhance their translation skills.
1. Jiemili Mandi, “Introduction to Translation Studies: Theory and Application,” Beijing: Foreign Language Teaching and Research Press, 2014, p. 13.
2. Liao Qiyi, “Contemporary Western Translation Exploration,” Nanjing: Yilin Press, 2000, p. 45.
3. Susan Bassnett & André Lefevere, Constructing Cultures: Essays on Literary Translation (Shanghai: Shanghai Foreign Language Education Press, 2001), 1.
4. Lian Shuneng, “Comparative Studies of English and Chinese,” Beijing: Higher Education Press, 2010, p. 73.
(The original title of the article was “Rebuttal Against Machine Translation Replacing Human Translation,” authored by Gu Zeqing from the University of International Relations.)
