Too Good to Be True-Not so reliable output from Mandarin-English Machine Translation (MT)

Fanglin Lou on June 21 2024 in Swansea of Wales

Key Words: Chinese-English Translation, machine translation, human translation, translation quality , deficiencies of Machine translation systems,preediting,postediting,

Over the last few years, it seems that more and more people in translation industry appear to have entered another period of hype and hyperbole of technological advance of machine translation(MT) and artificial intelligence-someone even claims that machine translation have reached “human parity” in some conditions. And it is more worse that there has been a stern reality that the current MT systems are having a negative impact on the healthy development of translation industry, especially for Chinese-English translation according to a set of statistical data and research from Xu, et al., (2021): 63% of Chinese students from advanced translation colleges, 65% from language colleges, and 87% from foreign language schools of the comprehensive university believe that quality of MT output is average or trustworthy, and 72.9% of Chinese students at the Advanced Translation Institutes and language colleges highly rely on machine translation. In another words it means that the current MT systems has get many translation professional/practitioners in the future (such as the students above) addicted, become highly relying on the MTs and gradually form bad translation habits although the current MTs has little effect on the mastery of students’ translation skills and the improvement of translation competence according to Wei, W. (2019). Meanwhile, according to the recent observation of the author of this essay, more and more Chinese students studying abroad, tends to rely on the various machine translation apps in their mobile phone or translation machine service(and equipment) based on the current MT technology, which does mislead the translation end-user(consumers)- the students does not understand the comments made by their tutors/lecturers according to the output provided by the immature and inherently-defective translation apps or equipment.

That does bring us to the question to be answered today: Is the current machine translation output reliable on the basis of quality?!Is it a product capable of producing acceptable-quality translation for ordinary consumers (or the end-user) of the traditional translation?

In this essay, it discusses the reliability of machine translation (MT) output, particularly focusing on Mandarin-English translation, and explores the possible reasons for the relative untranslatability observed in MT systems ,as well as the potential solutions or future prospects for improvement in MT technology.

Chapter 1 Limited Reliability of the Current Machine Translation Systems

Firstly, let’s have a discussion of the reliability of the current machine translation systems from the various researches of some prominent experts in the fields of machine translation and human translation.

According to Koehn P. (2020), the author of Neural Machine Translation (Cambridge University Press, 2020) and Statistical Machine Translation (Cambridge University Press, 2009) and holder or co-holder of five patents for machine translation ((Philipp Koehn, n.d.), the current MT systems are not capable of producing high-quality translation that requires professional translators who are native speakers of the target language and ideally also experts in the subject matter. Additionally, he also acknowledges that neural machine translation (NMT), the entire research field of machine translation (Page 20), does have the following limitations and weakness: 1)The current Machine Translation system in itself are facing challenges such as poor performance caused by domain mismatch (Page 294), inadequate amount of training data (Page 295), rare/infrequent words (P297), noisy data (i.e. irrelevant information from corpus available, Page 299), ambiguity of natural languages processing that only Human translators can effectively deal with presently (Page 5); 2) Neural machine translation (NMT) may produce bad translation misleading the readers because it sometimes prefers fluency over adequacy to the extent that completely misrepresent the output (Page 20 & Page 295).

Next, let’s get our attention to the specific field we are discussing today: quality of Chinese-English translation by current MT systems. According to Wei, W. (2019), the prominent scholar exploring Chinese translation studies two decades as well as the author of the book An Overview of Chinese Translation Studies at the Beginning of the 21st Century: Past, Present, Future, the current NMT system is only capable of effectively translating daily conversations, news translation, etc. and it could produce output full of errors once the topic area is changed (Page 215). Koehn P. (2020) also acknowledge that machines translation is not a useful tool in translating literature, poetry and marketing message that have to home in on the nuances of the targeted local culture(Page 21).

Meanwhile, it is well-known that translation by human translator, as a mature product and service, has been helping people with different culture and language background to achieve an efficient communication for thousands years. However, according to Koehn P. (2020) the current machine translation cannot compete with professional translators on the basis of quality (Page 2), which cannot achieve a quality that convinces customers to pay a large amount of money for it. In another words, what the current MTs can provide to the human readers, as shown in the following example sentences, is not a mature product that could be called as translation-exactly it is just a semi-product to be post-edited and finalized by a competent human translator because what output the current MT provided, are some unreadable target texts with the various translation errors as shown (the underlined ones) in the following example sentences:

Example 1

ST1:母亲和父亲离婚后，比赛谁先有新家，当然是好看的母亲抢了先，她嫁给了爱情，那个苦苦等她的初恋。小包工头的父亲也不甘落后，母亲新婚的鞭炮还在地上冒烟，父亲已经娶了镇上理发馆的武美媚。

TT from GT (Goggle Translate): After my mother and father divorced, they competed to see who would get a new home first. Of course, the good-looking mother took the lead. She married love, the first love that had been waiting for her so hard. The little contractor's father is not far behind. His mother's wedding firecrackers are still smoking on the ground. His father has already married Wu Meimei, a barber shop owner in the town.

TT from YD (Youdao Translate): After the divorce of mother and father, the competition who has a new home first, of course, is the good-looking mother robbed the first, she married love, the hard to wait for her first love. The father of the small contractor is not far behind, the mother's newly married firecrackers are still smoking on the ground, and the father has married Wu Meimei of the town barber shop.

Corrected TT by human translator: After my mother and father divorced, they competed to see who would get a new home first. Of course, the good-looking mother took the lead-She married love, her first love who had been waiting for her so hard. The father, a labor contractor in my town , is not far behind- while firecrackers celebrating the mother's wedding are still smoking on the ground, he was also getting married Wu Meimei, a barber shop owner in the town.

Example 2

ST2:那一年我十岁，母亲生了弟弟，父亲给我生了个妹妹，有了弟弟母亲忘了前边还有一个女儿，别说管，我来例假吓的哭，奶奶打电话她都让邻居说就当她死了。

TT From GT: I was ten years old that year. My mother gave birth to a younger brother. My father gave birth to a younger sister. After my younger brother was born, my mother forgot that she had a daughter. Don’t worry about it. I was so scared that I cried when I got my period. When my grandma called, she even asked the neighbors. Just treat her as dead.

TT from YD: That year I was ten years old, my mother gave birth to a younger brother, my father gave birth to a sister, with a younger brother, my mother forgot to have a daughter before, let alone, I had my period and cried, grandma called her to let the neighbor say that she was dead.

Corrected TT by human translator: I was ten years old that year when my mother was giving birth to a younger brother and my step-mother a younger sister. After that, it seems that my mother completely forgot that she had a daughter as me in this world-it would be better not to mention about her obligation of looking after me because she did not do her job as my mother at all- I was so scared that I cried when I got my period. Moreover, when my grandma called and told her that, she even asked her neighbor answering my grandma that we can assume she was dead and we’d better not bother her any more.

Example 3

ST3:父亲也一样，武美媚年轻漂亮，生的女儿穿着各式各样的小纱裙，可爱讨喜。

TT from GT: The same goes for the father. Wu Meimei is young and beautiful, and her daughter wears all kinds of gauze skirts, which is cute and lovable.

TT from YD: Father is the same, Wu Meimei young and beautiful, born a daughter wearing a variety of small gauze skirts, lovely.

Corrected TT by a human translator: The same goes for the father. Wu Meimei is young and beautiful, and her daughter in all kinds of gauze skirts, is so cute and lovable.

ST4:我就成了墙根开花的拉拉草，奶奶眼睛不好被姑姑接走了，我只能趁隔壁三婶去了城里儿子家，摘她家院子里的菜煮自己的面。

TT from GT: I became a flowering grass at the root of the wall. My grandma had bad eyesight and was taken away by my aunt. I could only take advantage of the third aunt next door to go to my son’s house in the city and pick vegetables from her yard to cook my own noodles.

TT from YD: I became the root of the wall flowering Lala grass, grandma eyes are not good aunt picked up, I can only take advantage of the next three aunts went to the city son's home, pick her yard to cook their own noodles.

Corrected TT from a human translator: I have been a weed growing and flowering on my own at the root of the wall because my grandma, with her bad eyesight, was taken away by my aunt from our old house. Since then I could only take advantage of Aunt San of the next door and pick her vegetables in her yard to cook my own noodles while she was living in his son’s house in the city.

In brief, as shown above, what output we can get from the two most prominent Chinese-English machine translation systems (Goggle translate and Youdao Translate),is unreadable and incomprehensible for the readers in case it is not further edited/corrected by an experienced and competent human translator.

So, based on the researches above from those prominent experts in the fields of both Machine translation and human translation, as well as the example sentences above, it could get a conclusion that output of the current MT systems, especially in translation of Chinese to/from English, is not reliable on the basis of quality although it does achieve many successful and useful applications such as helping common people to get the general gist of a document in foreign language and enabling translators become more productive (Page 3, Koehn P., 2020).

Chapter 2 Reasons for the unreliable output of MTs due to the relative untranslatability observed in MT systems

According to Baker, M. (2018) what a human translator must attempt to do is to understand the meanings of words and utterances as precisely as possible and express them into another language by choosing a suitable equivalent in a given context (Page 16)-In another words, translation process by a human translator is essentially composed of understanding the source texts precisely and expressing it in another language adequately, accurately and naturally.

By contrast, according to Koehn P., (2020)what the current machine translation systems are doing during their “ translation” process is just :1)to emulate the natural language processing process of human translators partly by decoding of ST and encoding of TT in such a translation mechanism as: firstly search (or predict) all possible output words (the possible equivalent words) by brute-force computing power from all available resources such as database, corpus and others in the target language (TL),compute a probability over all those words, and pick (or choose) the most probable words to encode and generate the translation output(Page 143);2)make that process more adaptive (Page 22)by learning from the new training material i.e. sentence pair created by human translators (called as “deep-learning). However, due to the finiteness of Chinese data and information that is/will be available and accessible for MTs, it bounds to be efforts without avail to some great extent because there is a cruel fact :in recent years more and more website contents in Chinese are being deleted and/or disappeared and become inaccessible due to economic reasons, strict censorship by government authorities and self-censorship of the publishers and authors according to research by He, J. (May 23, 2024)-it does greatly and negatively affect the accessibility and availability of the various Chinese data and Corpus resources that can be employed to train the MTs models to improve its processing performance in “translating” English into / from Chinese. Meanwhile, even if there are adequate on-line resources in Chinese can be used to search and match during MTs’ translation process, as illustrated from the seven aspects to be stated hereafter, the current MTs do fail to produce an acceptable-quality translation output because of their inability to understand ST precisely and express the meaning of ST adequately, accurately and naturally in the target language due to its rigid, mechanical and full-of-deficiency translation mechanism.

2.1 MTs’ inability to decode ST contextually

In consideration of finiteness and accessibility of the resources available for the MTs, as acknowledged by Koehn P., (2020) it does weaken and strain translatability of the current machine translation systems to the some extent because it only can pursue a research goal of driving down the error rates (Page 19) rather than producing a perfect translation as human translators can achieve by decoding STs including ambiguity of natural language, identifying and resolving the authors’/speakers’ unintentional expression error or misnomer based on a broader contexts and their knowledge.

Meanwhile according to Baker, M. (2018), without making an analysis of the source text contextually it is nearly impossible for human translators to define the basic propositional meaning of a word or utterance with absolute certainty due to the nature of language that, in the majority of cases, words ‘meanings are, to a large extent, negotiable and are only realized in speci?c contexts (Page 16)-In another words ,the current MTs are not capable of decoding and “understanding” the basic propositional meaning of the source texts contextually as human translator are doing so how can we expect of an acceptable-quality output from them?

The limited reliability of translation output from MTs due to its weaken and strained translatability caused by MT’s translation mechanism itself, as well as the translation errors made by MTs due to its inability to decode the STs contextually, could be illustrated by comparing and contrasting translation outputs respectively from a professional human translator( the author of this essay) and the three most prominent Machine Translation Engines for Chinese-English translation: Google Translate (GT), Youdao Translate (YT), and DeepL in the following example sentences:

Example 1

ST: For instance, a roughly 300,000-year-old skull (top) from Jebel Irhoud in Morocco has a relatively long, flat braincase.

TT from GT(Goggle Translate): 例如，来自摩洛哥 Jebel Irhoud 的一个大约 30 万年前的头骨（上）有一个相对较长、平坦的脑壳。(Back translation: For example, a roughly 300,000-year-old skull (top) from Jebel Irhoud, Morocco, had a relatively long, flat braincase.)

TT from DeepL: 例如，摩洛哥杰贝勒伊尔胡德（Jebel Irhoud）出土的距今约 30 万年前的头骨（上图）具有相对较长的扁平脑壳。(Back translation: For example, a skull (pictured above) from around 300,000 years ago unearthed in Jebel Irhoud, Morocco, had a relatively long, flat braincase.)

TT from YT(Youdao Translate): 例如，在摩洛哥Jebel Irhoud发现的一个大约30万年前的头骨(上)有一个相对较长的扁平脑壳。(Back translation: For example, a roughly 300,000-year-old skull (top) found in Jebel Irhoud, Morocco, had a relatively long, flat braincase.)

Corrected TT by human translator: 例如，人们在摩洛哥的Jebel Irhoud考古洞穴遗址发现的一个大约 30 万年前的古人类化石的头骨（如上图中最顶上图所示），就有一个相对较长和平坦的脑腔。（Back translation: For example, the skull of an approximately 300,000-year-old hominid found at the Jebel Irhoud archaeological cave site in Morocco (shown at the top of the image above) had a relatively long and flat brain cavity.）

Explanation: All three MTs fail to add and complement the necessary information that can help the TT readers to understand what “Jebel Irhoud” refer to. Secondly, they also fails to understand what the word top mean in the source text according to the context and just mechanically translate it into a target word “上”， which directly misleads and confuses the readers because this word actually means“as shown in the top one of the three photos above”according to the source texts .

Example 2

ST: That “somehow” became a matter of debate in the 1980s, ’90s and into the 2000s.

TT from GT: 这个“不知怎的”在 20 世纪 80 年代、90 年代和 2000 年代成为了一个争论的话题。

TT from DeepL: 这种 "不知何故 "在 20 世纪 80 年代、90 年代和 2000 年代成为一个争论不休的问题。

TT from YT: 在20世纪80年代、90年代和进入21世纪后，“不知何故”成为了一个争论的问题。

Corrected TT by human translator: 这个上文提到的使得智人（H. sapiens）得以进化形成的“ 未知的原因”，在 20 世纪 80 年代、90 年代和 2000 年代都屡次成为了古人类考古学界人们争论的话题。

Explanation: All the three MTs literally translated the word “somehow” into the target text without considering its exact meaning that could be obtained by reading a sentence “At some point, somehow, H. sapiens arrived and its predecessors vanished” in previous paragraph of ST. But such literal translation creates new difficulties and obstacle for TT readers to understand it. Actually, it seems that there are only two options that MTs can have while translating such a word because it is a polysemous words with only two meanings: A) in a way that is not known or certain; B) for a reason that you do not know or understand-In another words, there should a strong probability that MTs can accurately match the most probable option but in this case it failed to do so. Instead, even if there are two MTs make a right choice but they still fail to achieve the desired readability without adding a modifier “使得智人（H. sapiens）得以进化形成的” as well as a head noun “ 未知的原因”that should be adjusted to form an appropriate tailor-made one that could be comprehended by the readers. Therefore, it is proved that literal translation output from MTs does not work well even if MTs made the best/right match because a readable and accurate translation output does require a tailor-made TT that only human translator is capable of making by adopting some necessary translation techniques.

Example 3

ST: Different human populations may have emerged, with some floating away and petering out and others connecting to varying degrees.

TT from GT: 不同的人类群体可能已经出现，其中一些漂流并逐渐消失，另一些则有不同程度的联系。

TT from DeepL: 可能出现了不同的人类种群，有些种群逐渐消失，有些种群则在不同程度上联系在一起。

TT from YT: 不同的人类种群可能出现了，一些人离开并逐渐消失，而另一些人则在不同程度上联系在一起。

Corrected TT by human translator: 即该地区可能有多个人类种群出现，其中一些种群像小河一样流着流着就逐渐消失了，另一些人类种群则因和其他种群有不同程度的联系交流得以融合壮大进化成更强大族群，像多条小河汇成大江而得以生存下来。

Explanation: All the three MTs fails to analyze the ST contextually, which is necessary and critical to decode and understand the real meaning of this sentence accurately- in the above example sentence the metaphor that human evolution process is compared to a big river merged by “a braided streams “ mentioned in the preceding sentence ,as well as the causal logic concealed in the ST- some human groups/tribes get survived and evolved into a more stronger/bigger groups because they have connections with other human groups and the human groups not connecting with other get extinction/disappeared naturally like the small streams not connecting/merging with others .

But, due to its mechanical and rigid translation mechanism, what MTs do here is just to seek for textual equivalence or the most probable match in the target text as much as they can. So, it seems that it has been going to a dead end no matter how those statistical-based MTs developed and updated to the what extent because there is/are not equivalence at all either in their huge database or from the available online resources such as the several parallel corpus available-what an acceptable-quality target text in Mandarin requires here is to employ some translation techniques to supplement/add necessary information in the target text based on the preceding sentence.

Example 4

ST:教他的时候我才明白，三婶带我干活不是真干活，她是让我吃饭吃的理直气壮，没有寄人篱下的委屈。

TT from GT: When I was teaching him, I realized that my third aunt was not really working when she took me to work. She was letting me eat with confidence, without the grievance of being dependent on others.

TT from YD: When I taught him, I realized that Aunt Three took me to work not really work, she let me eat righteously, without the injustice of being supervised by others.

TT from DeepL: It was only when I was teaching him that I realized that Auntie San wasn't really working when she took me to work, she was letting me eat with a straight face and without the indignity of being a parasite.

Corrected TT: When I was teaching him, I realized that: when my Aunt San had taken me to do farm work before, she didn't really mean that she wanted me to do the farm work for her - she just wanted me to be able to eat her food confidently without the grievance of being dependent on someone else.

Explanation: All three MTs made a same mistake in understanding the meaning of words “干活”in the ST because MTs do lack of capability of decoding ST contextually so what they did here is just find some equivalent word they assume as “most probable” one in the target language ,which do create more barriers for readers/users to understand the TTs.

In brief, what the current MTs have been doing is just a practice of choosing-matching-generating “the most probable equivalence” in the target language rather than a real translation that only a human translator can do, which is composed of two essential steps: firstly decoding and understanding the source texts based on the broader contexts and knowledge being aware and available to the translators, then encoding and producing the natural and idiomatic target text readable to the readers/users. However, as acknowledged by Philipp Koehn (a leading researcher in the field of machine translation and professor of computer science at Johns Hopkins University), the current MTs cannot compete with human translator “on the basis of quality” (Page3, Koehn P., 2020) and MT’s ultimate goal should be to analyze and understand meaning of the source text and then generate meaning-based target text (Page 10, Koehn P., 2020) rather than their current practice of searching, choosing, matching and generating the translation output with most probability.

2.2 MTs’ inability to decode information and meanings that contains in one of language feature of English: cohesion

According to Gleason, et.al (1998) in an English sentence there are several type of cohesion including reference, substitution, ellipsis, conjunction and lexical cohesion, which contains some essential information and meaning that should be rendered in the TTs. Based on my practical analysis and study via the example sentences below, it shows that some essential information and meanings, which are contained and implied in these type of cohesion, cannot be decoded and encoded by MTs, especially the one contained in reference, substitution and ellipsis of a source texts. Now, by the following example sentences being translated from English into Chinese by the three most prominent MTs, let’s have a close examination at those information and meaning that has not been decoded and expressed by MTs as well as the various translation errors that MT made:

2.2.1 Information and meanings that contains in demonstrative reference of an English sentence.

Example 1

ST: You had a job interview the next day and exams the following week, and you still went dancing? -- In my young days we took these things more seriously. We had different ideas then.

TT from GT: 你第二天面试，下周考试，你还去跳舞吗？ ——在我年轻的时候，我们带着这些事情比较严重。 当时我们有不同的想法。(Back translation: You have an interview the next day, an exam next week, will you still go dancing? ——When I was young, we brought these things more seriously. We had different ideas at that time.)

TT from YT: 你第二天要面试，下个星期要考试，你还去跳舞?在我年轻的时候，我们对这些事情更认真。我们当时有不同的想法。(Back translation: You have an interview the next day, an exam the next week, and you still go dancing? When I was young, we were more serious about these things. We had different ideas then.)

TT from DeepL: 第二天要面试，下周要考试，你还去跳舞？-- 在我年轻的时候，我们把这些事情看得更重。那时候我们有不同的想法.(Back translation: You have a job interview the next day and an exam next week, and you still go dancing? -- When I was young, we took these things more seriously. We thought differently then.)

Corrected TT by human translator: 第二天要面试，下周要考试，这种情况下你还去跳舞？- 我们在你现在这个年龄时，有着和你们不一样的看法-我们通常把面试和考试这些事情看得更重，这种情况下，不会还去跳舞。

Translation Error Analysis: Obviously in the example above the MTs fails to analyze the exact meaning of the demonstrative noun these in the source text and what the two CAT tool did to this sentence is just match it with an equivalent word 这些 in the target text, which creates a ambiguity at word level in the target text: What does the pronoun these refer to? Is it the job interview, exams the following week or/and dancing? In case of being translated by a human being translator, it could be analyzed contextually and concluded that the pronoun these in the source text refers to the job interview at the next day and the exams in the following week then the source texts could be comprehended correctly and adequately, which can successfully achieve the first step of an acceptable translation: decoding of the source text.

2.2.2 Information that contains in comparative reference

Example:

ST: A: I got 1000 quid. B: Same here.

TT from GT: A：我有 1000 英镑。 乙：这里也一样。(Back Translation: A: I have 1000 quid. B:Same here.)

TT from YT: A:我有1000英镑。B:我也是。(Back Translation: A: I have 1000 quid. B:I am too)

TT from DeepL: A: 我得到了 1000 英镑。B: 我也一样。(Back Translation A: I got ￡1,000. B: Me too.)

Corrected TT: A:我得了1000英镑。 B：我也得了这么多。

或者： 我有1000英镑。 B：我也有一千英镑。

Translation Error Analysis: The word same in the source text refers to the“1000 quid” in the former sentence but the CAT tool GT also create a ambiguity by conveying it into a “equivalent” words “这里也一样“ in the target text. The YT also make a mistake and translate it literally as “A: I have a 1000 quid. B: I am too”. Comparing to the other two MTs DeepL produced a output with better consistency.

2.2.3Information that contains in verbal substitution

ST: A: Did somebody feed the cat?

B: I did

GT: A：有人喂猫吗？B：我做到了。

(Back translation: Did anyone feed the cat? B: I did it.)

YT: A:有人喂猫了吗?B:是的。

(Back translation: A: Did somebody feed the cat? B: Yes.)

TT from DeepL: 甲：有人喂猫吗？B: 喂了

(Back translation: Did somebody feed the cat? B:(someone) fed it already.

Corrected TT: A: 猫喂了吗？ B：我喂过了.

Translation Error Analysis: Obvious all three MT fails to make an acceptable rendering of ST because they do not function or operate same as human translator by producing output with better consistency between the simple question and answer in the target language.

2.2.4 Information that contains in clausal substitution

Example:

ST: A: Is John married? B: I believe so. C: I think not.

TT from YT: A:约翰结婚了吗? B:我想是的。C:我不这么认为。

(Back translation: A: Is John married? B: I think so. C: I don’t think so.)

TT from GT: A：约翰结婚了吗？乙：我相信是这样。丙：我想不是。

(Back translation: A: Is John married? B: I believe so. C: I think not.)

TT from DeepL: A：约翰结婚了吗？B：我想是的。C：我想没有。

(Back translation: A: Is John married? B: I think so. C:I don’t think so.)

Corrected TT: 约翰结婚了吗？ B：我觉得他结婚了. C: 我觉得他没结婚。

Translation Error Analysis: It seems that MT does not have the capacity of decoding ST as per the context and encoding TT based on the sematic rule of target language- the three sentences in a simple dialogue cannot be encoded appropriately with necessary cohesion and consistency.

2.2.5 Information that contains in Verbal Ellipsis.

Example:

ST: A: Have you been swimming? B: Yes, I have.

TT from GT: A：你游泳了吗？ B：是的，我有。

(Back translation: Did you swim? B: Yes, I have. (The word have means “own (sth) in Mandarin.

TT from YD: A: 你游泳了吗? B:是的，我有。

(Back translation: Did you swim? B: Yes, I have. (The word have means “own (sth) in Mandarin.

TT from DeepL: 甲：你游泳了吗？B：是的，我游过。

(Back translation: Did you swim? B: Yes, I swam.)

Corrected TT: A: 你以前一直都游泳吗？ B：是得，我以前一直都游泳。）

Translation Error Analysis:Based on the example sentences above it could concluded that MT do not analyze the ST syntactically and generate/produce a translation output that retains information of ST such as the time when the event/action take place- in the example sentence the information that contains in the two syntax properties (tense and verbal ellipsis) has been lost in the translation output generated by the MT.

2.3 MT’s inability to generate an acceptable-quality target text based on the syntax rule about word order of a modifier in Mandarin

According to Shei, C. (2014) in Chinese all modifiers including the word-level modifiers, phrasal modifier and clausal modifiers are usually put and stacked to the left of the head noun, whereas in English word-level modifiers are put to the left of the head noun but phrasal and clausal modifiers are put to the right of the head noun in English (P92). But, due to the mechanical and rigid translation mechanism executed by the current MTs, as shown in the example sentences below, MTs do make some obvious translation errors about the word order of a modifier in the target text in Chinese while encoding and generating translation from English to Chinese- it rigidly retains and copy the sentence structure of the ST in English.

Example 1:

ST: We have decided to ask the public and have asked them to suggest ideas to make our products even more environmentally friendly.

TT from GT: 我们决定向公众征求意见，并请他们提出一些想法，使我们的产品更加环保。

(Back translation: We decided to consult the public and ask them to come up with ideas to make our products more environmentally.)

TT from YT: 我们决定向公众征求意见，并请他们提出建议，使我们的产品更加环保。

(Back translation: We have decided to consult the public and ask for their suggestions to make our products more environmentally.)

TT from DeepL: 我们决定向公众征求意见，请他们提出建议，使我们的产品更加环保。

(Back translation: We have decided to consult the public and ask for their suggestions to make our products more environmentally)

Corrected TT: 我们已经决定向公众征求使我们的产品更加环保的建议。

(Back translation: We have decided to ask the public for suggestions to make our products more environmentally.)

Translation error analysis: In the ST “to make our products even more environmentally friendly” is the phrasal modifier of the head noun idea so usually its equivalence in the target language 使我们的产品更加环保的 should be stacked to the left of the head noun idea 建议 in the target text to be produced by human translator or machine translation software. Of the translation output from the three prominent MT engines, at the first sight it seems all of them present a fluent target text containing nearly all adequate information of the ST. But, there are two obvious problem or translation error: 1)the right or appropriate word order of modifier in the target language;2) what the author want to highlight by such a sentence structure “we …asked them to suggest idea….”-it should be the head noun idea being modified rather than the modifier at the right side. So, based on the analysis of sentence structure of the ST as well as intention of the author, it is the best choice to put the modifier at the left side of the head noun in the TT as per the syntax rule of the target language rather than MTs’ practice of generating a TT retaining the original sentence structure of ST.

2.4 MTs’ inability to adjust and generate a sentence structure conforming to the basic syntax rule of the target language

As shown in the following example sentences , the MTs is not able to generate a readable output by correspondingly adjusting and producing a sentence structure conforming to the basic syntax rule of the target language such as Chinese.

Example 1

ST: Liber is the Latin name of Bacchus; and Mulciber of Vulcan.

TT from GT: Liber是巴克斯的拉丁名； 和瓦肯人的马尔西伯。

TT from YT: 利伯是巴克斯的拉丁名;还有火神的穆尔西伯。

TT from DeepL: Liber 是巴克斯的拉丁名；Mulciber 是火神的拉丁名。

Corrected TT: 利贝尔（Liber）是酒神巴克斯的拉丁名, 穆尔塞伯（Mulciber）是火神和金工神伏尔甘的拉丁名。

Explanation: Of the outputs from the three MTs, GT and YT not only fail to decode the ST but also made some TT full of translation errors misleading and confusing the readers because they fails to use a sentence structure conforming to the basic grammar rule of the target language (Chinese), whereas DeepL performs better but it does not adjust the sentence structure of TT correspondingly.

Example 2

ST: The Penates were the gods who were supposed to attend to the welfare and prosperity of the family.

TT from GT: 佩那特斯是负责照顾家庭幸福和繁荣的神。

TT from YT: Penates是负责家庭福利和繁荣的神。

TT from DeepL: Penates是负责家庭幸福和繁荣的神灵。

Corrected TT: 佩纳特斯（Penates）据说是主管家庭幸福和繁荣的多位神明。

Explanation: Based on the above example sentences, output from all three MTs fails to decode and encode the word gods that should refer to more than one so it could be concluded that all MTs do not have the capability to identify and translate these kind of information that contains in the syntax rule of the source texts.

2.5 MTs’ inability to contextually-decode ST with ambiguity of natural language

Example 1

ST: Whenever I visit my uncle and his daughters, I can’t decide who my favorite cousin is.

TT from GT: 每当我去看望叔叔和他的女儿们时，我都无法决定谁是我最喜欢的表弟。（Back translation: Whenever I visit my uncle and his daughters, I can't decide who is my favorite cousin.）

TT from YT: 每当我去拜访叔叔和他的女儿们时，我都无法决定谁是我最喜欢的堂兄。（Back translation: Whenever I visit my uncle and his daughters, I can't decide who my favorite cousin is.）

TT from DeepL: 每次我去看望舅舅和他的女儿们时，我都无法决定谁是我最喜欢的表妹。（Back translation: Every time I visit my uncle and his daughters, I can't decide who my favorite cousin is.）

Corrected TT: 每次我去看望舅舅和他的女儿们时，我都分不清哪一个才是我最喜欢的那个表妹。

Translation Error Analysis: Without knowledge about facts of family relationships there should be ambiguity that cannot be decoded in the ST even if it is translated by a human translator because the word cousin can refer to male or female in the source language (English). However, based on the outputs above, both a human translator and DeepL can avoid the translation error made by GT and YT, who fails to achieve consistency of the main clause and subordinate clause in the sex of the object and translate ST into two separate TT without connection. So, based on the example sentence, it shows that these two MTs are not capable of identify and decode sentence structure contextually and strains translatability of the ST, which finally results in such translation error. Meanwhile, even if DeepL performs better as mentioned above but what output this MT offers is not an idiomatic TT that a human translator can easily deal with by encoding a TT with a perfect consistency.

2.6 MTs’ inability to correctly decode ST in case of authors’ unintentional expression errors or inadequate expression, misnomer, unintentionally-deleted contents.

As a professional translators/interpreters or/and translation Practitioners in reality it is inevitable and very common to face many unexpected challenges and “surprises” that was/were (and will be ) unintentionally created by the authors or/and translation agencies ,such as ST authors’ unintentional expression errors or inadequate expression, misnomer and/or unintentionally-deleted contents, which could be identified and recognized by an experienced human translators. Under this circumstance as a responsible translator it should be our responsibility to make a timely-remedy by employing the various translation strategies, methods and/or technique in order to achieve a successful and efficient communication between the authors and the readers. However, under this circumstance all current MTs surely have been failing to produce an output that is/are capable of competing with the one from a human translators because they never have the essential sense of responsibility, professional ethic and subjectivity as a human. Following are some example sentence showing the translation errors made by MTs while doing “translation “under this circumstance:

Example 1

ST: The more you, the PA, has a general overview of the situation and things that need to be made, some of them obvious, others less, the more you manage to lighten the burden on the shoulders of your boss or your boss in as part of their daily work, and you can prepare this must be done properly - or you take care it yourself, in part.

TT from GT: 作为 PA，你对情况和需要做的事情了解得越多，其中一些是显而易见的，另一些则不太明显，你就越能减轻你老板或你老板肩上的负担。 他们的日常工作，你可以准备这必须正确完成 - 或者你自己处理一部分。

(Back translation: As a PA, the more you understand about the situation and what needs to be done, some of which are obvious and some of which are less obvious, the more you can take the burden off the shoulders of your boss or your boss. You can prepare them for their daily routine which must be done correctly - or handle part of it yourself.)

TT from YT: 作为私人助理，你对情况和需要做的事情有越多的总体了解，其中一些是显而易见的，另一些则不那么明显，你就越能减轻你的老板或你的老板日常工作中的负担，你就能准备好这件事必须妥善完成——或者你自己照顾它，在某种程度上。

(Back translation: The more you have an overall understanding of the situation and what needs to be done as a PA, some of which are obvious and others not so obvious, the more you can lighten the load on your boss or your boss's daily routine, and you'll be prepared for the fact that this thing has to be done properly - or that you take care of it yourself, to some extent.)

TT from DeepL: 你，PA，对情况和需要做的事情有一个总体的了解，其中一些是显而易见的，另一些则较少，你就越能减轻你的上司或你的老板肩上的负担，作为他们日常工作的一部分，你可以准备这必须做得很好 - 或者你自己照顾它，部分。

Back translation: The more you, PA, have a general understanding of the situation and what needs to be done, some of which is obvious and others less so, the more you can take the burden off the shoulders of your supervisor or your boss as part of their daily routine, and the more you can either prepare for this having to be done well - or you take care of it yourself, in part.

Corrected TT: 作为公司董事的秘书，你对需要做的事情及该事情发展变化情况了解得越多(一些事情及其发展变化情况是显而易见的，另一些则不太明显)，你就越能减轻你老板肩上的负担，或者减轻他们的日常工作所带来的部分负担。对于你老板必须正确完成的那部分工作，你可以准备好大量的资料信息备她/他查阅- 或者你帮他/她处理其中一部分工作。

Translation Error analysis: Output from all three MTs shows that MTs are not capable of contextually decoding the ST with expressing error or misnomer unintentionally made by the author of the ST-what MT can do in this case is just to retain the sentence structure of ST, search/predict the equivalent words from the available online or offline resources in the target language and finally match with the most probable ones to generate a “fluent” but inadequate/inaccurate translation output. MTs’ translation mechanism itself leads to a failure of dealing with ST under such circumstance because MT do not have the subjectivity and flexibility as a human translator ,who can comprehensively decode ST and encode TT based on the contexts, relevant knowledge and reasoning.

Example 2

ST: The earliest purported hominins (purple) show some signs of upright walking, which became more routine with the rise of (green).

TT from GT: 据称最早的古人类（紫色）显示出一些直立行走的迹象，随着（绿色）的兴起，这种直立行走变得更加常见。(Back translation: The earliest hominins (purple) are said to show some signs of upright walking, which became more common with the rise (green).)

TT from YT: 最早的人类(紫色)显示出直立行走的一些迹象，随着(绿色)的兴起，直立行走变得更加常规。(Back translation: The earliest humans (purple) showed some signs of upright walking, which became more routine with the rise of (green).)

TT from DeepL: 据说最早的类人猿（紫色）有一些直立行走的迹象，随着类人猿（绿色）的兴起，直立行走变得更加常见。(Back translation: The earliest hominids (purple) are said to have shown some signs of upright walking, which became more common with the rise of hominids (green).

Corrected TT: 传说中的最早的古人类（如下图中紫色线段所示）已显示出一些直立行走的迹象，随着南方古猿（如下图中绿色线段所示）的兴起，这种现象变得更加常见。(Back translation: The legendary earliest ancient humans (shown by the purple line segment in the figure below) already showed some signs of upright walking, which became more common with the rise of the southern archaeopteryx (shown by the green line segment in the figure below).

Translation Error Analysis:All the three MTs fails to generate an output that clearly explain what the word purple and green refer to because the ST from the customer is not a complete one-the illustration/figure in the ST had been deleted unintentionally by the customer. So, under this circumstance, it cannot be smoothly and accurately decoded even if it is translated by a human translator, not mentioned about any MT. However, as a responsible professional translator I immediately contacted the customer and asked her to confirm if there is/are some contents deleted because I cannot understand and decode it. As you can image it was solved after my call because it is translated by a human being with adequate responsibility, subjectivity and professional ethic rather than a machine. In brief the current MT cannot deal with such situation because it is just an indifferent machine who never doubt if there will be any problem with the ST-Actually , even if the illustration is there ,the current MT also have no idea of how to decode it contextually because its rigid and mechanical translation mechanism.

2.7 MTs’ Inability to employ the various indispensable translation methods and translation technique

According to Koehn P. (2020) words in languages have not only an explicit meaning but also an implications that often does not have any equivalent in another language and another culture. But, essentially speaking, what the current MTs have been doing is : search (they calls it as “predict””) a set of equivalents (or candidates) to words in ST, then make a computation/calculation of the probability over those equivalents, finally choose the most probable one to generate the output. But, here we have another question (also can be called a problem): How the MT will handle it or proceed on in the event that there isn’t any equivalent in the target language as mentioned above. That brings us to another serious and deadly deficiency that MTs have: the current MTs’ inability of employing translation strategies, translation methods and translation technique that effectively help human translators to expand translatability of language and get the untranslatable items into translatable one. According to Shei, C., & GAO, Z.-M. (2017) there are seven most important Chinese-English translation techniques: addition, omission, division, combination, shift, pragmatic translation and cross- cultural translation (Page 48), which cannot be employed by MTs to generate an acceptable-quality translation output presently. Now let’s have a close examination of MTs’ poor performance by the following example sentences:

Example 1.

ST: Scientists were also skeptical of Dart. While a student in London, he had earned a reputation as a “scientific heretic, given to sweeping claims,” according to a paper coauthored by a colleague.

TT from GT: 科学家们也对 Dart 持怀疑态度。 据一位同事合着的一篇论文称，当他在伦敦上学时，他就赢得了“科学异端，热衷于横行霸道的主张”的名声。(Back translation: Scientists are also skeptical of Dart. While he was at school in London, he earned a reputation as a "scientific heretic, fond of overbearing claims," according to a paper co-authored by a colleague.)

TT from DeepL: 科学家们对达特也持怀疑态度。他在伦敦读书时，就赢得了 "科学异端，喜欢夸夸其谈 "的名声。(Back translation: Scientists were also skeptical of Dart. While a student in London, he earned a reputation as a "scientific heretic with a penchant for rhetoric.")

TT from YT: 科学家们也对达特持怀疑态度。据他的一位同事与他人合著的一篇论文称，在伦敦读书时，他就赢得了“科学异端，喜欢一马当先”的名声。(Back translation: Scientists were also skeptical of Dart. According to a paper co-authored by one of his colleagues, while a student in London he earned a reputation as a "scientific heretic with a penchant for being ahead of the curve.")

TT Corrected: 达特还在伦敦上学时其治学态度就遭人诟病，一位曾与其合著论文的同事评价说达特是一位 “ 科学异端分子，热衷于发表过分笼统一概而论的主张 ”。

Explanation: Based on the output from the three MTs, it shows that MT prefers fluency over adequacy by choosing some four-words idioms such as 横行霸道,夸夸其谈, 一马当先 in the target language, which inevitably leads to translation error that mislead the readers. Meanwhile, MTs also fails to employ the necessary translation technique “addition “to improve the readability and accuracy of the target text by adding modifiers including “其治学态度” and “曾与其合著”.

Example 2

ST: The four men huddled there and said nothing. They dared not smoke. They would not move.

TT from GT: 四个人挤在一起，一言不发。 他们不敢抽烟。 他们不会动。（Back translation: The four people huddled together and said nothing. They dare not smoke. They won't move.）

TT from DeepL: 四个人挤在那里一言不发。他们不敢抽烟。他们一动不动。（Back translation: The four people huddled there without saying a word. They dare not smoke. They were motionless.）

TT from YT: 那四个人挤在一起，什么也没说。他们不敢抽烟。他们不愿动。（Back translation: The four people huddled together and said nothing. They dare not smoke. They didn't want to move.）

Corrected TT: 那四个人在那儿挤在一起，不说话，不敢抽烟，也不敢走开。(Back translation: Those four men were huddled together there, not talking, not daring to smoke or walk away.)

Explanation: Obviously output from all three MTs do lack necessary adhesion between sentences ,which gives the readers an impression that there is not any logical connection as a result the translation output from MTs is not readable, fluent or/and accurate to some great extent. MTs’ mechanical and literal translation without adjustment of structure of ST does contribute to the obvious translation errors above. Instead ,what the human translator does here is to employ the translation technique “combination”,i.e. several sentences in ST are translated into one sentence in TT, which effectively solves the problem of readability and adhesion of TT in Chinese.

Meanwhile, human translators’ choice of suitable equivalent depends on a wide variety of factors including a range of restrictions that may operate in a given environment such as censorship and various types of intervention by parties other than the translator, author and reader according to research from Baker,M.(2018). So in consideration of more and more strict censorship and the other restrictions in China that might be caused by the nature of the regime there, it might have a negative impact on the accessibility and availability of on-line Chinese data and resources that can be employed by MTs, which also inevitably leads to MTs’ inapplicability to the language pair to some extent in the future.

Chapter 3 Potential solutions or future prospects for improvement in MT technology

According to the research of Shei & Zhaoming Gao (2018) translatability in many ways relies on the availability of equivalence, whose absence or notable lack probably results in untranslatability (Page 103) even if the source text is being translated by a competent and experienced human translator. Obviously there might be worse extent untranslatability caused by such an absence or lack of equivalence while a text is processed and translated by a machine translation system, which are more widely available and proven by the example sentences in this essay. However, according to Shei & Zhaoming Gao (2018), a certain level of tolerance of partial translation is acceptable in consideration of the varieties of interaction between two languages and cultures. Therefore, as mentioned in the preface of this essay, as a professional translator and a linguist, in order to maximize the advantage of MTs in improving productivity of a professional human translator, I just suggest and recommend an alternative solution composed of the following steps:

1) with the consent of the authors, a competent human translator ,whose native language is the source language, could be assigned to pre-edit the source texts a standard sentence structure such as subject–verb–object，subject-verb or other sentence structure that could be recognized and understand by the readers of the translation output;

2) The pre-edited ST could be reviewed by the authors (if possible) to ensure that it does retain the core meanings and connotations, the author’s intension as well as some necessary/essential information that cannot be lost/removed.

3) Then, the source text that has been edited by the translator (and reviewed by the author/s if necessary and possible) could be processed and “translated” by a MT.

4) Finally output from the MT shall be reviewed and finalized by a human translator whose native language is the target language.

Regarding to the productivity and efficiency desired/expected by the multiple parties including authors, translator and readers, it could be demonstrated by the following example sentences I just tried in my assignment before:

Example 1

ST: 那一阵风过处，只听得乱树皆落黄叶，刷刷的响，扑地一声，跳出一只吊

睛白额斑烂猛虎来，犹如牛来大。

TT from DeepL based on the original ST: That a gust of wind over the place, only to hear the chaotic trees are falling yellow leaves, brush sound, pouncing, jumped out of a hanging A fierce tiger with white eyes and a spotted forehead came, as big as a cow.

Pre-edited ST: 在那一阵风所过之处，听到黄叶皆从树上刷刷响的落下时，武松看到一只白色额头上有斑斓图案,如公牛般大小的吊睛猛虎突然间扑地一声，从树后跳了出来。

TT from DeepL based on Pre-edited ST: In that a gust of wind passed by, heard the yellow leaves are from the tree brush sound fall, Wu Sung saw a white forehead has a colorful pattern, such as the size of a bull hanging eyes fierce tiger suddenly pounced, jumped out from behind the tree.

TT finalize/post-edited by a human translator: Where that gust of wind passed, while seeing that all the yellow leaves are falling from the trees with a brush, suddenly with a pouncing sound, Wusong saw a fierce tiger with hanging eyes and a spotted pattern on its white forehead ,as big as a bull, jumping out from behind the trees.

Example 2

ST: 这武松被那一惊，把肚中酒都变做冷汗出了。说时迟，那时快。武松见大虫扑来，只一闪，闪在大虫背后。原来猛虎项短，回头看人教难，便把前爪搭在地下，把腰跨一伸，掀将起来；武松只一躲，躲在侧边。大虫见掀他不著，吼了一声，把山岗也振动。武松却又闪过一边。

TT from GT: Wu Song was so shocked that the wine in his stomach turned into cold sweat. It's too late, but it's soon. When Wu Song saw the big insect coming towards him, he just dodged and got behind the big insect. It turned out that the tiger had a short neck, so when he looked back and saw that it was difficult for him to teach him, he put his front paws on the ground, stretched his waist, and lifted it up. Wu Song just ducked and hid on the side. Seeing that he couldn't catch him, the big insect roared and shook the hills. Wu Song ducked aside again.

ST Pre-edited by a human translator: 被那声虎啸一惊，武松顿时惊出一身冷汗，刚才喝酒带来的醉意也没了！刹那间，见大虫扑来，武松只轻轻一闪，便闪躲在了大虫背后。由于颈项短，回头看人较难，那老虎此时便把前爪搭在地下，把腰跨一伸，想将眼前的猎物掀翻；武松只作了一闪躲，便躲在了老虎的侧边。那老虎见掀翻不了他 ，便又发出一身振动山岗的虎啸 ，武松此时又趁机闪躲到老虎的另一边。

TT from GT based on the Pre-edited ST: Startled by the roar of the tiger, Wu Song suddenly broke into a cold sweat, and the drunkenness caused by drinking just now was gone! In an instant, when Wu Song saw the big insect coming towards him, he just dodged slightly and hid behind the big insect. Due to its short neck, it was difficult to look back at people. The tiger put its front paws on the ground and stretched its waist, hoping to overturn the prey in front of it. Wu Song only dodged and hid on the side of the tiger. Seeing that the tiger could not overturn him, the tiger let out another roar that shook the mountain. Wu Song took the opportunity to dodge to the other side of the tiger.

TT finalized by a Human Translator: Startled by the roar of the tiger, Wu Song suddenly broke into a cold sweat, and his drunkenness caused by drinking was gone! In an instant, when Wu Song saw the tiger coming up towards him, he just dodged slightly and hid behind the tiger. Due to its short neck, it was difficult to look back to identify where the prey is, so the tiger put its front paws on the ground and stretched its waist, hoping to overturn the prey. Wu Song only dodged and hid on the side of the tiger. Seeing that failure of overturning the prey, the tiger was angry and let out another roar that shook the mountain. Wu Song took the opportunity to dodge to the other side of the tiger.

Example 3

ST: 原来虎伤人，只是一扑，一掀，一剪，三般捉不著时，气力已自没了一半。武松见虎没力，翻身回来，双手轮起稍棒，尽平生气力，只一棒，只听得一声响，簌簌地将那树枝带叶打将下来。原来不曾打著大虫，正打在树枝上，磕磕把那条棒折做两截，只拏一半在手里。

TT from GT: It turns out that a tiger can only hurt a person with one pounce, one lift, and one scissor. When the tiger fails to catch it, half of its strength has been lost. Wu Song saw that the tiger was weak, so he turned back, raised a stick with both hands, and used all his strength to strike down the branch with leaves with only one blow. There was only a sound. It turned out that the big insect had not been hit before, but it was hitting the branch. He broke the stick into two pieces and only held half of it in his hand.

Pre-edited ST: 老虎猎杀人类的方式，通常都是先进行一次正面扑杀，如果不成功则用身躯掀fan猎物，再不成功则挥动其铁棒般的虎尾拍击猎物。如果这三种猎杀方式都不奏效，老虎此时就几乎耗去了自身一半的力气。此时见那老虎没了力气，武松则立刻翻身回来，双手举起哨棒，用尽平生气力，将手中棒砸向那老虎-他此时只想凭这一棒之力击杀老虎，却不曾想一声响后，他见那哨棒只是簌簌地打落了带叶的树枝，不曾打著那老虎！更糟糕的是，打在树枝上的哨棒也折断成了两截，只剩一半握在他的手里。

TT from GT Based on Pre-edited ST:

The way tigers hunt humans is usually to kill them head-on. If unsuccessful, they will lift the prey with their body, and if unsuccessful, they will slap the prey with their iron rod-like tail. If these three hunting methods fail, the tiger will have expended almost half of its strength at this time. Seeing that the tiger had no strength at this time, Wu Song immediately turned back, raised the whistle stick with both hands, and used all his strength to smash the stick in his hand at the tiger. He just wanted to kill the tiger with the power of this stick, but he couldn't. Unexpectedly, after the sound, he saw that the whistle stick only knocked down the leafy branches, but did not hit the tiger! To make matters worse, the whistle stick that hit the branch also broke into two pieces, with only half left in his hand.

TT finalized by Human Translator:

Regarding to the way that tigers use to hunt humans it is to kill them head-on firstly. If unsuccessful, they will throw the preys on the ground with their body, and if unsuccessful still, they will slap and brandish the prey with their iron rod-like tail. If these three hunting methods fail, it could be supposed that the tiger probably have expended almost half of its strength at this time. Seeing that the tiger had no enough strength to launch an immediate attack, Wu Song immediately turned back, raised the stick with both hands, and used all his strength to smash the stick at the tiger-At that moment he just wanted to kill the tiger with the power of this stick! But, unexpectedly, after hearing a snap, he saw that the stick only knocked down the leafy branches rather than hit the tiger! Even worse, at this moment the stick that hit the branch also broke into two pieces, with only half being held in his hands!

Example 4

ST: 这武松心中，也有几分慌了；那虎便咆哮性发，剪尾弄风起来，向武松又只一扑，扑将来。武松一跳，却跳回十步远。那大虫扑不著武松，把前爪搭在武松面前，武松将半截棒丢在一边，乘势向前，两只手挝在大虫顶花皮，使力只一按，那虎急要挣扎，早没了气力。武松尽力挝定那虎，那里肯放松。

TT from GT: Wu Song was a little panicked in his heart; the tiger roared, cut its tail to make the wind blow, and pounced on Wu Song again. Wu Song jumped, but jumped back ten steps. The big insect couldn't pounce on Wu Song, so he put his front paws in front of Wu Song. Wu Song threw aside half of the stick and took advantage of the momentum. He placed both hands on the flowery skin of the big insect and pressed hard. The tiger was about to struggle. No energy left. Wu Song tried his best to pinpoint the tiger, but he was willing to relax.

ST Pre-edited by human translator: 此时武松的心里也有几分慌了。见此情形，那虎便又发出一声咆哮，挥动其铁棒般的虎尾，带着风声，再次向武松扑过来。武松只作了一跳，便跳回到离老虎十步远之地。那老虎见扑不着武松，便把前爪搭在武松面前暂时蓄力，想要再次扑向那猎物。这时武松将半截棒丢在一边，趁机向前，两只手抓住那老虎头上虎皮往下一按，便将老虎按在地上！此时那虎虽然急着要挣扎起身，却因早没了气力，只得任由武松按在地上动弹不得。此时的武松也尽全力将那老虎死死地按在地上，那里肯放它起身伤人。

TT from GT based on Pre-edited ST: At this time, Wu Song felt a little panicked. Seeing this, the tiger roared again, waved its iron rod-like tail, and rushed towards Wu Song again with the sound of wind. Wu Song only made one jump and then jumped back to a place ten steps away from the tiger. Seeing that the tiger couldn't pounce on Wu Song, the tiger put its front paws in front of Wu Song to temporarily accumulate strength, hoping to pounce on the prey again. At this time, Wu Song threw half of the stick aside, took the opportunity to move forward, grabbed the tiger's skin on the head with both hands, and pressed down, knocking the tiger to the ground! At this time, although the tiger was eager to struggle to get up, because he had long since lost his strength, he had no choice but to let Wu Song pin him to the ground and be unable to move. At this time, Wu Song also tried his best to pin the tiger to the ground, willing to let it get up and hurt people.

TT finalized by Human Translator: At this time, Wu Song felt a little panicked. Seeing this, the tiger roared again, waved its iron rod-like tail, and rushed towards Wu Song again. Wu Song only made one jump and then jumped back to a place ten steps away from the tiger. Seeing that failure of pouncing on the prey, the tiger put its front paws in front of Wu Song to temporarily accumulate strength, hoping to pounce on the prey again. At this time, while throwing half of the stick aside, Wu Song took the opportunity to move forward, grabbed the tiger's thick mass of hair that grows on its head with both his hands, and pressed it down, pining the tiger to the ground! At this time, although the tiger was eager to struggle to get up, because he had used up his strength, he had no choice but to let Wu Song pin him to the ground and be unable to move. At this time, Wu Song also tried his best to push the tiger to the ground, not willing to let it get up and hurt people.

Example 5

ST: 一面把只脚望虎面上眼睛里，只顾乱踢；那虎咆哮，把身底下，扒起两堆黄泥，做了一个土坑里。武松按在坑里，腾出右手，提起拳头来，只顾狠打，尽平生气力。不消半歇儿时辰，把那大虫打死。躺卧著，却似一个绵布袋，动不得了。

TT from GT: While looking into the tiger's face and eyes with one foot, it just kicked around; the tiger roared, picked up two piles of yellow mud under its body, and made a pit. Wu Song pressed down on the pit, freed his right hand, raised his fist, and just hit hard, using all his strength. In less than a moment, the big insect was killed. Lying down, it looks like a cotton bag, unable to move.

Pre-edited ST by human translator: 双手将虎头死死按在地上的同时，武松将一只脚朝老虎的眼睛乱踢。那虎痛得发出一阵阵的咆哮，但还是被武松按得起不了身，只得拼命用爪子在自己身下扒坑-但没想到扒出的两堆黄泥，又形成了一个让老虎更加动不了的较大的土坑。这时，将那虎死死地按在那坑里的同时，武松终于腾出右手，提起拳头来，用尽平生气力不停地狠打那虎。接下来武松用了不到一个小时，就把那老虎给打死了-此前要吃人的老虎，此刻已经成了一具躺卧著的一动不动的虎尸，看起来好似一个有着漂亮图案的绵布袋。

TT from GT based on Pre-edited ST: While pressing the tiger's head to the ground with both hands, Wu Song kicked one foot into the tiger's eyes. The tiger roared in pain, but Wu Song couldn't hold him up, so he had to use his claws to dig holes under his body - but he didn't expect that the two piles of yellow mud he dug out formed a trap that made the tiger move even more. Large pits that cannot be reached. At this time, while pressing the tiger firmly into the pit, Wu Song finally freed his right hand, raised his fist, and kept hitting the tiger with all his strength. It took Wu Song less than an hour to kill the tiger. The tiger that had been trying to eat people was now a lying, motionless corpse, looking like a tiger with beautiful patterns. Cotton bag.

TT finalized/post-edited by Human translator: While pressing the tiger's head to the ground with both hands, Wu Song kicked one foot into the tiger's eyes. While making roars in pain, but couldn't get up due to being pressed and pinned on the ground by Wu Song, the tiger had to use his claws to dig holes under his body to escape - but he didn't expect that: the two piles of yellow soils he dug out formed a trap, a large pit that does confine his more movement. At this time, while pressing the tiger firmly into the pit, Wu Song freed his right hand, raised his fist, and kept hitting the tiger with all his strength. Finally it took Wu Song less than an hour to kill the tiger-The tiger that had been trying to kill people was now a lying, motionless corpse, looking like a bag with beautiful patterns on the ground.

Summary and Conclusion

In this paper, by quoting statements and claims from some prominent experts and scholars in the area of Chinese-English translation and Machine Translation, plus a close analysis and examination of example sentences being translated by both MTs and an experienced and professional human translator, firstly it discuss and confirms that quality of translation output from the current MT systems is not as reliable as boasted and exaggerated in the popular press ,then it makes a study and exploration of the possible causes of MTs’ unsatisfactory translation performance, and then concluded that the translation mechanism employed by current machine translation systems contributes to their unsatisfactory performance in translation quality, finally in the last chapter of this essay a potential alternative solution is presented as a further collaboration between human translators and machine i.e. pre-editing ST by human translator, translating the pre-edited ST by MT, and finalizing/post-editing the MT output by human translator.

In brief, MTs’ translation processing mechanism of “searching and matching the equivalents”, same as (a man) looking for a satisfactory life partner in this world with a population of 7 billion, may not produce a satisfactory translation output because the satisfactory life partner (the equivalent in the target language and culture) might not exist at all there, so how can you find a perfect and satisfactory one (a solution that only can be tailor-made by human translator presently)? –It is partly because there are ‘culture-specific concepts’ where the source-language word may express a concept which is totally unknown in the target culture’ (p. 18) and there are instances where the source-language concept is not lexicalized in the target language’ according to Wei, W. (2019, Page 18),which does require intervention of a competent human translator by employing translation strategies and techniques such as cultural substitution, i.e.to translate using a load word plus an explanation, or by paraphrasing and so on(Page 150).

However, along with the advancement and maturity of artificial intelligence and machine deep learning technologies in recent years, with the common and joint efforts of linguists and the scientists dedicating in this field of AI and MT it can be expected that untranslatability of a language and cultures such as Chinese could be reduced to some extent in the coming decades although there are a great deal of challenges and difficulties due to its particularity in language/culture features for translation of this language pair ,as well as the regional political and society system there.

Reference:

1. Koehn P. (2020).Neural Machine Translation. Cambridge University Press.

2. Shei, C. (2014). Understanding the Chinese language: a comprehensive linguistic introduction. Taylor and Francis.

3. Shei, C., & GAO, Z.-M. (2017).The Routledge Handbook of Chinese Translation. Routledge.

4. Xu, M., Xin, H., & Liu, J. (2021, January 2). 许明 | 外语专业学生机器翻译使用现状调研 | 项目研究成果_崔启亮. Www.sohu.com . https://www.sohu.com/a/441980995_312708

5. Wei, W. (2019). An Overview of Chinese Translation Studies at the Beginning of the 21st Century. Routledge.

6. Gleason, J. Berko., & Ratner, N. Bernstein. (1998). Psycholinguistics (2nd edition). Harcourt Brace College Publishers.

7. Baker, M. (2018). In other words?: a course book on translation (Third edition.). Routledge

8. Philipp Koehn. (n.d.). Johns Hopkins Whiting School of Engineering. https://engineering.jhu.edu/faculty/philipp-koehn/

9. B Hatim, & Munday, J. (2019). Translation: an advanced resource book. Routledge.

10. He, J. (2024, May 23). 中文互联网正在加速崩塌. Cnbeta.com.tw . https://www.cnbeta.com.tw/articles/tech/1431972.htm

Afterwords and Acknowledgment

I am so grateful of the help and guidance from my tutor Dr. Chris Shei during my writting of this paper.Great thanks again for his contribution to this as well as his understanding, supports, help and encouragement upon me !.