Authorized by AI Technology Camp (ID: rgznai100)
Translation: Shawn
This article contains 10,000+ words, suggested reading time is 10+ minutes. American scholar Douglas Hofstadter points out through his personal experience with Google Translate that machine translation currently lacks thought and is difficult to replace humans.
[ Introduction ] Although machine translation is clearly not capable of handling the much-anticipated long-form content, we must admit that it does provide some convenience in quickly understanding the meanings of words. Strangely, however, both media reports and the industry seem to create an atmosphere that machine translation is about to replace human translators, giving people a false sense of impending reality.
Someone needs to unveil the veil of blind optimism; among them is American scholar Douglas Hofstadter, who won the Pulitzer Prize for his book “Gödel, Escher, Bach.” He points out through his personal experience with Google Translate that translation software currently knows only its form but does not understand its meaning. Further, for machine translation to replace humans, it should not only decode words but more importantly, it needs to possess the same understanding of meaning that humans have. To solve this second problem, machines need to replicate human intelligence, but technical researchers have been avoiding this dilemma.
The following is the translated content:
On Sunday, my friend Frank brought a Danish guest to our weekly salsa dance. Frank’s mother is Danish, and he lived in Denmark when he was young, so he speaks fluent Danish. Frank’s Danish friend can also speak English, quite fluently by Nordic standards. However, during the evening chat, I was surprised to find that this pair of friends relied on Google Translate to communicate via email: Frank would first write the email in English and then use Google Translate to convert it into Danish; conversely, his friend would write the email in Danish and then use Google Translate to convert the Danish into English.
Isn’t that strange? They can clearly understand each other’s spoken words, yet they insist on using Google Translate for emails, which seems redundant, doesn’t it?
Based on my experience with machine translation software, I have always held a high degree of skepticism towards its translation effectiveness, yet these two did not seem to mind. In fact, many intelligent people are advocates of translation software and rarely criticize the shallowness of machine translation, which I find perplexing.
As a language enthusiast, a passionate translator, and a cognitive scientist, I have been fascinated by the intricacies of the human mind throughout my life and have dedicated decades to the mechanization of translation work. My interest in this subject first arose in the mid-1970s when I read a letter from 1947 written by Warren Weaver, an early advocate of machine translation, to Norbert Wiener. In that letter, Weaver proposed a rather famous viewpoint – translation is decoding, and his exact words were:
Whenever I see a Russian article, I say to myself, “This content is actually written in English; it’s just that someone has encrypted it with some strange symbols. Next, I just need to decode it.”
However, years later, he threw out a completely different viewpoint:
“It is clear to any discerning person that machines cannot become Pushkin; machine translation will never convey the elegance and style inherent in the language itself.”
I find Weaver’s later viewpoint resonates with me more, especially after I spent an entire year translating Pushkin’s lengthy narrative poem “Eugene Onegin” into English, where the challenge was how to completely re-create the read Russian content into an English narrative poem.
Weaver’s initial viewpoint merely revealed one side of language being overly simplified. Nevertheless, his 1947 viewpoint of “translation is decoding” has long become an important tenet driving the development of machine translation.
Since then, translation engines have continuously improved, especially with the recent achievements of “deep neural networks” in machine translation, which have led some commentators to begin proclaiming the impending extinction of human translators. For example, Gideon Lewis-Kraus’s article “The Great AI Awakening” published in The New York Times, and Lane Greene’s article “Machine Translation: Beyond Babel” published in The Economist. According to them, human translators will soon be relegated to the roles of quality controllers and proofreaders, no longer the producers of text.
If the translation field truly develops to this extent, my spiritual life will undoubtedly suffer a severe blow. While I can completely understand the allure of trying to maximize the value of machine translation, I do not want to see human translators replaced by emotionless machines. Just thinking about it makes me feel anxious and repulsed.
In my view, translation is an extremely meticulous art that requires the translator to apply rich life experiences and creative imagination in the recreation process. If the somewhat “acceptable” language produced by machine translation leads to human translators becoming outdated “relics,” it will severely undermine my respect for human wisdom, leaving behind endless confusion and sadness.
Whenever I read people claiming that a new technology will soon eliminate human translators, I feel compelled to investigate personally, partly out of fear that the nightmare about machine translation will ultimately come true, and more so to confirm that the article may be exaggerating, which helps alleviate my inner anxiety, as I firmly believe it is crucial to expose the exaggerated AI lies.
Thus, when I read that Google Brain enhanced the old artificial neural network theory with deep learning and achieved revolutionary machine translation results with this technology, I decided to personally try out the latest Google Translate to see if it would truly become a disruptor in the field of machine translation like chess’s “Deep Blue” and Go’s “AlphaGo”?
It is well known that the old version of Google Translate could handle many languages, but the deep learning version of the new Google Translate initially only supported nine languages (Note: currently it supports 96 languages). Therefore, I limited the languages I explored to four: English, French, German, and Chinese.
Before showcasing my findings, I need to point out one thing – the term “(deep)” is somewhat misused as a polysemous word here. When people hear that Google acquired a company that uses “deep learning” to enhance “deep neural networks” (the company is called DeepMind), they instinctively interpret “deep” as “profound,” thereby deducing meanings like “powerful,” “insightful,” and “far-sighted.” However, the real meaning of “deep” here simply refers to the fact that these neural networks have more layers than the old networks that only had 2 to 3 layers (for instance, 12 layers). However, does the extra few layers necessarily mean that the tasks completed by the neural networks are inherently “profound”? That is not guaranteed; it is merely a linguistic usage technique.
I have always held deep skepticism toward Google Translate, after all, the media has overhyped it. But despite my dislike, some capabilities of Google Translate still amaze me. Anyone around the world can use this service for free, and it can translate between about 100 languages. This indeed makes humans feel ashamed. If I dare to call myself a “multilingual person,” then Google Translate is undoubtedly a “polyglot,” as I only know about three languages, and some of those I only understand a little, making my claim to be a “multilingual person” feel rather insecure, but Google Translate’s hundred languages are truly genuine.
In fact, I only need to copy and paste text from Language A into Google Translate’s input box, and it can instantly translate the entire page of content into Language B. Moreover, Google Translate can continuously provide various language translation services to global users.
The practical value of Google Translate is beyond doubt; overall, it is still a decent product. However, the methods it uses have significant flaws, which can be summed up in one word – understanding. After all, the focus of machine translation has never been on understanding language; on the contrary, the research strategy in this field has always been to avoid understanding the content and its meaning, namely, “decoding.” So, is it truly feasible to translate a good article without understanding the content? Can high-quality translations by either humans or machines be accomplished without understanding the meaning of language?
To explore this question, I personally conducted some experiments with Google Translate, which I will explain in detail next.
English-French Translation
To start, I used a short sentence that is simple and easy to understand:
In their house, everything comes in pairs. There’s his car and her car, his towels and her towels, and his library and hers.
(Meaning: In their house, everything comes in pairs. He and she have their own cars, towels, and libraries.)
The translation of this sentence seemed straightforward, but in French and other Romance languages, Google Translate’s handling of the possessive pronouns “his” or “her” does not reflect the change in gender, as it only processes the nouns following the possessive pronouns, resulting in the following:
Dans leur maison, tout vient en paires. Il y a sa voiture et sa voiture, ses serviettes et ses serviettes, sa bibliothèque et les siennes.
As expected, Google Translate fell into my trap; it was completely unable to understand the language like a human: it failed to recognize that this sentence describes a couple, knowing that the sentence emphasizes that for everything the husband has, the wife has a matching one. Instead, the deep learning engine used the same word “sa” to refer to “his car” and “her car,” leaving the reader unable to determine the gender of the car owner. Similarly, it also used the gender-neutral plural form “ses” to refer to “his towels” and “her towels.” As for the last “his library and hers,” the possessive pronoun “hers” with an “s” completely confused Google Translate, making it think that “s” represents a plural form (“les siennes”). In the end, Google Translate completely misunderstood the intended meaning of this sentence.
To convey the original sentence’s intent, I had to translate these short sentences into French myself:
Chez eux, ils ont tout en double. Il y a sa voiture à elle et sa voiture à lui, ses serviettes à elle et ses serviettes à lui, sa bibliothèque à elle et sa bibliothèque à lui.
Here, “sa voiture à elle” refers to “her car,” while “sa voiture à lui” refers to “his car.” After clarifying the expression, I thought it would be a straightforward task to have Google Translate accurately translate my French back into English. But once again, I overestimated it; it made an even more ridiculous mistake:
At home, they have everything in double. There is his own car and his own car, his own towels and his own towels, his own library and his own library.
What is this nonsense? Despite my efforts to clearly highlight the gender information of the owners in the sentences, Google Translate remained oblivious, completely failing to understand the most critical information intended by the sentence and simply converting all personal pronouns to the masculine “his.” Why is this the case?
We humans can understand a variety of abstract concepts such as couples, houses, personal property, self-esteem, competition, jealousy, privacy, etc., as well as more complex situations formed by habits, such as a couple wanting to embroider “his” and “her” on their towels. Google Translate, however, cannot grasp such contexts, or rather, it cannot understand any contextual information. What it is familiar with are merely strings of letters forming words, and words forming strings. Google Translate is only concerned with how to quickly process fragmented text rather than the thoughts, imaginations, memories, or understanding behind the text. It doesn’t even care what the words represent behind them. In principle, computer programs can understand the meanings of language, possess thoughts, memories, and experiences, and utilize them, but this has never been the intent behind Google Translate’s development. Its developers did not even have such ambitions.
Seeing these ridiculous statements from Google Translate, I can’t help but breathe a sigh of relief – machine translation ultimately cannot replace human translators. But I believe I should further test Google Translate more meticulously. After all, one instance does not make a proof; one swallow does not make a summer.
So, what about this sentence “One swallow does not thirst quench” – I created this phrase from the proverb (“One swallow does not a summer make”)? What kind of French sentence would Google Translate produce for it? After trying, Google Translate gave me this result: “Une hirondelle n’aspire pas la soif.” This translation is grammatically correct in French, but it is indeed puzzling.
First, it uses a type of swallow (“une hirondelle”) to refer to the 74 species of birds in the swallow family, and says this bird does not suck (“n’aspire pas”), while the object that the bird sucks is thirst (“la soif”). It is evident that Google Translate completely misunderstood my meaning; it merely re-encoded the sentence into a jumble of nonsensical symbols. As for the sentence “Il sortait simplement avec un tas de taureau,” it translated as “He just went out with a pile of bulls.” Then translating it back into French became “Il vient de sortir avec un tas de taureaux.” Please forgive my poor French; more accurately, it is Google Translate’s pseudo-French.
English-German Translation
Having discussed French, let’s look at German. Recently, I became fascinated with the book “Sie nannten sich der Wiener Kreis” by Austrian mathematician Karl Sigmund, which is titled “They Called Themselves the Vienna Circle” in English. This book discusses a group of Vienna idealist intellectuals from the 1920s to 1930s who had a significant impact on philosophy and science in later generations.
I used a small excerpt from Sigmund’s writing in this book to test Google Translate and see what English it could produce. Let’s take a look, first at the original German text written by Sigmund, then my translation, and finally Google Translate’s result. (By the way, I had two native German speakers check my translation, including Karl Sigmund himself, so you can basically consider my translation accurate.)
Sigmund:
Nach dem verlorenen Krieg sahen es viele deutschnationale Professoren, inzwischen die Mehrheit in der Fakultät, gewissermaßen als ihre Pflicht an, die Hochschulen vor den “Ungeraden” zu bewahren; am schutzlosesten waren junge Wissenschaftler vor ihrer Habilitation. Und Wissenschaftlerinnen kamen sowieso nicht in frage; über wenig war man sich einiger.
Hofstadter’s Translation:
After the defeat, many professors with Pan-Germanistic leanings, who by that time constituted the majority of the faculty, considered it pretty much their duty to protect the institutions of higher learning from “undesirables.” The most likely to be dismissed were young scholars who had not yet earned the right to teach university classes. As for female scholars, well, they had no place in the system at all; nothing was clearer than that.
(Meaning: After the defeat, professors’ political leanings were still predominantly Pan-Germanistic, and they felt responsible for protecting higher learning institutions from the intrusion of “undesirable people.” The most likely to face hostility were young scholars who had not yet earned the right to teach university courses. As for female scholars, they had absolutely no standing; nothing was clearer than that.)
Google Translate:
After the lost war, many German-National professors, meanwhile the majority in the faculty, saw themselves as their duty to keep the universities from the “odd”; Young scientists were most vulnerable before their habilitation. And scientists did not question anyway; There were few of them.
Although the vocabulary in Google Translate’s result consists of English words, several words are still inappropriately capitalized for unknown reasons. These words might begin to form a sentence, but as it progresses, you become increasingly lost, and the translation quality is abysmal.
First, let’s look at the quoted term “the ‘odd.” It corresponds to “die ‘Ungeraden” in the German original, meaning “politically undesirable people.” However, Google Translate translates it as “odd” for a reason: statistics. In other words, in the vast bilingual database used by Google Translate, “ungerade” is almost always translated as “odd.”
Although the translation engine does not understand why it makes this conversion, I can tell you the reason. This is because “ungerade” almost always means “odd” (a number that cannot be divided by 2), even though its literal meaning is “not straight” or “not even.” Conversely, I used the term “undesirables” to translate “Ungeraden,” which has nothing to do with the word’s statistical data; it is purely based on my understanding of the context – its meaning is hidden between the lines, and any German dictionary’s definition of “ungerade” does not fit the topic.
Now let’s discuss another German word “Habilitation,” which refers to a university position similar to that of a tenured professor. While there are related terms in English that derive from “Habilitation,” they are extremely rare, and readers are unlikely to associate it with tenured professors. This is why I chose to explain this term rather than use an obscure word; otherwise, ordinary English readers would be utterly confused. Google Translate, of course, cannot do this; it lacks the knowledge model that readers possess.
The last two sentences of the original text illustrate the importance of “understanding” in translation very well. The word “Wissenschaftler,” composed of 15 letters, refers to “scientists” or “scholars” (I believe it refers to the latter, as it pertains to the intellectual class in the original context). Google Translate fails to grasp this subtlety. The last sentence’s “Wissenschaftlerinnen” is the plural form of “Wissenschaftlerin,” which is a feminine German noun. “Wissenschaftler” is grammatically masculine, referring to male scholars; “Wissenschaftlerinnen” is feminine, referring only to female scholars. In my translation, I used “female scholar” to convey this meaning. However, Google Translate does not recognize that the suffix “-in” in “Wissenschaftlerin” is crucial in the last sentence. It does not realize that this term refers only to male scholars, thus using “scientist,” losing the original sentence’s focus on the contrast between male and female.
Aside from this major error, the rest of the translation in the last sentence is a disaster. Let’s look at the first half. Is “scientists did not question anyway” really the correct translation for “Wissenschaftlerinnen kamen sowieso nicht in frage”? The translation and the original meaning are completely mismatched; the words in the translation were generated randomly from the German words, can this be called “translation”?
The second half of the last sentence’s translation is equally poor. The literal meaning of the last six German words is “over little was one more united,” which can be more fluidly expressed as “there was little about which people were more in agreement” (this is almost undisputed), but Google Translate translates this obvious meaning as “There were few of them.” Readers would undoubtedly wonder, “Few of what?” but for the mechanical listener that is Google Translate, this question is meaningless. Google Translate lacks imagination; thus, it cannot answer seemingly simple questions. It does not engage in any significant or minor imagination during translation; it merely combines words randomly, having no concept of the meanings represented by the words.
ELIZA Effect
For those who possess life experience and understanding, and who can skillfully use the different meanings of words, it is still difficult to realize how hollow the translations generated by Google Translate are. People naturally assume that software capable of handling text so fluently must understand the underlying meanings. The classic illusion generated by artificial intelligence projects is referred to as the “ELIZA effect.” In the 1960s, a project named ELIZA could lead users to believe it understood English, when in fact, it had no idea what it was saying. ELIZA simulated a psychotherapist, and many users who “chatted” with it mistakenly believed ELIZA could deeply understand their inner feelings.
For decades, the ELIZA effect has deceived many well-informed individuals, including some artificial intelligence researchers. To help readers avoid this trap, I will quote several sentences from the earlier text: “Google Translate does not understand,” “Google Translate is not aware,” “Google Translate does not have the slightest thought.” These sentences may seem contradictory; on one hand, they state that Google Translate lacks understanding, while on the other hand, they suggest that Google Translate can at least sometimes understand the meanings of individual words, phrases, or sentences. This is not the case; Google Translate is designed to avoid or evade understanding language.
To me, the term “translation” exudes a mysterious and enchanting aura. It is a humanistic art form that requires the translator to elegantly express the clear ideas of language A in language B, ensuring clarity while also conveying the original author’s writing style, techniques, and qualities. Before translating, I first carefully read the original text, imprinting its ideas as clearly as possible in my mind, chewing on it over and over. What I chew on are not the words of the original text but the thoughts that evoke various related ideas, and by doing so, I can conceive rich related scenes in my mind. It goes without saying that most of this conceptual process is subconscious. Only when the scenes I construct in my mind are rich enough do I attempt to express them in another language – “to extract them.” During translation, I try to express the scenes in my mind in what I consider a natural way using language B, and these scenes construct the meaning of the original text.
In short, I do not directly translate the words and phrases of language A into the words and phrases of language B. During translation, I subconsciously construct images, scenes, and ideas in my mind, drawing from my past experiences (what I have read, seen in movies, or heard from friends). Only when the non-verbal, imagistic, experiential thoughts form in my mind – only when the ethereal bubbles symbolizing the original meaning float in my mind – do I organize words and phrases in the target language, and then revise them repeatedly. This translation process mediated by textual meaning, while seemingly slow (compared to Google Translate’s speed of translating a page in two or three seconds), is precisely what all serious human translators must accomplish during translation. This is the translation I understand when I hear phrases like “deep mind.”
Chinese-English Translation
Next, I tested Google Translate’s Chinese translation. Compared to French and German, Chinese presents a much greater challenge for deep learning software. I selected a portion from the memoirs of Chinese playwright and translator Yang Jiang (who recently passed away at the age of 104) titled “We Three.” This book recounts the lives of her and her husband Qian Zhongshu (a novelist and translator) and their daughter. Although this book is not particularly obscure, its language is quite exquisite and vivid. I selected a small excerpt for testing Google Translate. Below are the results provided by Google Translate and my own translation (checked by a native Chinese speaker):
Yang Jiang:
锺书到清华工作一年后，调任毛选翻译委员会的工作，住在城里，周末回校。 他仍兼管研究生。
毛选翻译委员会的领导是徐永煐同志。介绍锺书做这份工作的的是清华同学乔冠华同志。
事定之日，晚饭后，有一位旧友特雇黄包车从城里赶来祝贺。客去后，锺书惶恐地对我说：
他以为我要做“南书房行走”了。这件事不是好做的，不求有功，但求无过。
Hofstadter’s Translation:
After Zhongshu had worked at Tsinghua University for a year, he was transferred to the committee that was translating selected works of Chairman Mao. He lived in the city, but each weekend he would return to school. He also was still supervising his graduate students.
The leader of the translation committee of Mao’s works was Comrade Xu Yongying, and the person who had arranged for Zhongshu to do this work was his old Tsinghua schoolmate, Comrade Qiao Guanhua.
On the day this appointment was decided, after dinner, an old friend specially hired a rickshaw and came all the way from the city just to congratulate Zhongshu. After our guest had left, Zhongshu turned to me uneasily and said:
“He thought I was going to become a ‘South Study special aide.’ This kind of work is not easy. You can’t hope for glory; all you can hope for is to do it without errors.”
Google Translate:
After a year of work at Tsinghua, he was transferred to the Mao Translating Committee to live in the city and back to school on weekends. He is still a graduate student.
The leadership of the Mao Tse Translation Committee is Comrade Xu Yongjian. Introduction to the book to do this work is Tsinghua students Qiao Guanhua comrades.
On the day of the event, after dinner, an old friend hired a rickshaw from the city to congratulate. Guest to go, the book of fear in the book said to me:
He thought I had to do “South study walking.” This is not a good thing to do, not for meritorious service, but for nothing.
Below I will point out several oddities.
-
First, although the name “Zhongshu” appears three times in the original text, Google Translate fails to express it as a proper name. In the first instance, Google Translate uses the pronoun “he”; in the second instance, it translates “Zhongshu” as “the book”; in the third instance, it translates “Zhongshu” as “the book of fear in the book.” Just look at this translation!
-
The second oddity is that the original text states that Zhongshu supervises graduate students, while Google Translate claims he himself is a graduate student.
-
The third oddity lies in the phrase “毛选翻译委员会” (Mao Tse Translation Committee), where Google Translate omits the character “泽” from Chairman Mao’s name.
-
The fourth oddity is that “after our guest had left” is reduced to “guest to go.”
-
The fifth oddity is that the final sentence of Google Translate’s output makes no sense at all.
These five errors are enough to leave Google Translate utterly embarrassed, but let’s not pursue further. Now, let’s look at a puzzling phrase: the content in quotes in the last paragraph (“南书房行走”). This phrase can literally be translated as “south book room go walk,” which clearly makes no sense; it serves as a noun in the original text, while Google Translate’s “South study walking” is completely incorrect.
I admit I do not understand what this Chinese phrase means either. Although it literally refers to walking around in a study room located on the south side of a building, I know that cannot be right; it makes no sense in context. To translate this phrase, I needed to clarify if there was some Chinese culture I was unaware of. Who could I ask for help? Google! (But not Google Translate). I entered “南书房行走” into Google search, along with the quotation marks, and the search engine quickly displayed a large number of Chinese web pages. I then painstakingly read the opening paragraphs of the first few pages to try to understand what this phrase means.
I found that this phrase can be traced back to the Qing Dynasty (1644–1911), referring to the emperor’s writing assistant in the south study. The assistant’s job was to help the emperor write edicts in the “南书房” of the Forbidden City. “行走” literally means “go walk,” but in this context, it refers to an assistant. Based on the information provided by Google search, I translated this phrase as “South Study special aide.”
Isn’t it a pity that Google Translate cannot utilize Google search like I did? While Google Translate can translate web pages in the blink of an eye, it cannot understand the content of those web pages. Can it? Below is the output text I received after submitting the content of the web pages I searched to Google Translate:
“South study walking” is not an official position, before the Qing era this is just a “messenger,” generally by the then imperial intellectuals Hanlin to serve as. South study in the Hanlin officials in the “select chencai only goods and excellent” into the value, called “South study walking.” Because of the close to the emperor, the emperor’s decision to have a certain influence. Yongzheng later set up “military aircraft,” the Minister of the military machine, full-time, although the study is still Hanlin into the value, but has no participation in government affairs. Scholars in the Qing Dynasty into the value of the South study proud. Many scholars and scholars in the early Qing Dynasty into the south through the study.
Is this really English? Of course, we know it is – to be precise, most of it consists of English words, but is this a coherent English article? In my view, this content has no meaning; thus, it cannot be considered English. It is merely a jumble of random English words or a logically incoherent mix of words.
If you are interested, below is my translation of this web content (which took me several hours):
The nan-shufang-xingzou (“South Study special aide”) was not an official position, but in the early Qing Dynasty it was a special role generally filled by whoever was the emperor’s current intellectual academician. The group of academicians who worked in the imperial palace’s south study would choose, among themselves, someone of great talent and good character to serve as ghostwriter for the emperor, and always to be at the emperor’s beck and call; that is why this role was called “South Study special aide.” The South Study aide, being so close to the emperor, was clearly in a position to influence the latter’s policy decisions. However, after Emperor Yongzheng established an official military ministry with a minister and various lower positions, the South Study aide, despite still being in the service of the emperor, no longer played a major role in governmental decision-making. Nonetheless, Qing Dynasty scholars were eager for the glory of working in the emperor’s south study, and during the early part of that dynasty, quite a few famous scholars served the emperor as South Study special aides.
Some readers may suspect that I deliberately selected poorly translated segments to criticize Google Translate, believing that the results of Google Translate for the vast majority of paragraphs are much better. This may sound somewhat reasonable, but that is not the case. Every segment I input from the book I was reading into Google Translate resulted in various major and minor errors, including nonsensical and incomprehensible sentences like the ones above.
Of course, I acknowledge that sometimes the translations provided by Google Translate do indeed look decent (even if they may be ambiguous or completely incorrect). A whole paragraph or several sentences may be translated quite well, which might lead people to mistakenly believe that Google Translate knows what it is doing, that it knows what “reading” is. In such cases, Google Translate seems to perform remarkably well – almost like a human! This is undoubtedly thanks to its creators and their efforts. But at the same time, let’s not forget Google Translate’s terrible performance when translating the two previous Chinese and the French and German articles.
To understand such poor performance, we must always keep the ELIZA effect in mind. While Google Translate can translate over a hundred languages, it cannot read – in the human-defined sense of “reading.” It simply processes text, and the symbols it processes are disconnected from real-life experiences. Google Translate lacks the memory and understanding capabilities to draw upon; its rapid-fire translations sometimes carry no meaning whatsoever.
How Far Are Machines from Truly Understanding Language?
A friend once asked me whether the level of translation by Google Translate is merely a function of the program’s database. His point was that if the database size were expanded a millionfold or billionfold, Google Translate would eventually be able to translate any language perfectly. I do not believe so. No amount of “big data” can grant machines understanding capabilities, as understanding fundamentally relies on having thoughts. The lack of thought is the root of all current problems with machine translation. Therefore, I believe that a larger database – or even an ultra-large database – cannot solve the problem.
Another common question is whether the use of neural network technology can help machines achieve true language understanding capabilities. At first, this sounds reasonable, but current attempts are all limited to the translation of words and phrases. While neural networks can utilize various statistical facts about large databases, these statistical facts merely link words to other words, not words to the meanings of thoughts. No one has yet attempted to create internal structures that can realize thoughts, imaginations, memories, or experiences. At this point, achieving such super-intelligent technology through computation is merely a pipe dream; hence, people have turned to fast and mature statistical word clustering algorithms. However, the results produced by these algorithms are far from satisfactory and cannot be compared to the thought processes involved in human reading, understanding, creating, modifying, and evaluating an article.
Although I hold a negative attitude, many people still highly praise the services provided by Google Translate: it quickly converts meaningful text written in language A into text in language B that is not obviously meaningful. As long as the text in language B is somewhat understandable, many people feel very satisfied with it. If they can “roughly understand” articles written in languages they do not know, they feel content. Personally, I do not believe this is the definition of “translation,” but some people consider Google Translate a good service, and the results it provides can be seen as translation. Well, I understand their needs, and I understand why they feel satisfied. For them, this is a stroke of luck.
Recently, I saw some bar charts made by new technology enthusiasts representing the quality of human versus computer translations, indicating that the latest translation engines are very close to human levels of translation. In my view, quantifying immeasurable pseudoscience is merely a technical fanatic’s attempt to use mathematics to solve something intangible, subtle, and artistic that they do not understand. In my opinion, the results produced by Google Translate are sometimes excellent, sometimes ridiculous, but I cannot quantify my feelings about these translations. In the first example I provided, the thoughtless Google Translate almost translated every word correctly, but it completely failed to convey the original meaning. In such cases, can the quality of translation be quantified? Using seemingly scientific bar charts to represent translation quality is merely dressing pseudoscience in a “scientific” guise.
When discussing the unfortunate future of human translators, who will soon be surpassed and eliminated by machines, gradually becoming quality controllers and proofreaders, this is the best outcome for mediocre translators. But true artists will not stoop to handling translations filled with errors, generating elegant artistic works through endless modifications. This is not the essence of art, and translation is indeed an art.
Throughout my years of writing, I have always believed that the human brain is a machine – a very complex machine. I strongly oppose the notion that machines themselves cannot process meaning. There is a philosophical school that even claims computers can never “grasp semantics” because their “construction” (silicon) is wrong. In my view, this is complete nonsense. I won’t delve into this topic in this article, but I do not want readers to think that I believe machines can never achieve intelligence and understanding. If I have given readers that impression in this article, it is because the technology I discussed does not attempt to replicate human intelligence. Instead, it seeks to bypass human intelligence in an indirect manner, and the translations provided above clearly reveal the flaws of this technology.
I believe there is no substantial theoretical basis to prove that, in principle, machines cannot think, create, joke, reminisce, feel excited, be afraid, experience ecstasy, obey, or be hopeful, nor that they cannot perfectly translate languages. There is also no substantial theoretical basis to prove that machines cannot perfectly translate jokes, puns, scripts, novels, poetry, or articles like this one. But such dreams can only be realized when machines possess thoughts, emotions, and experiences like humans. I believe that day is still far from us, and as a deep admirer of human intellect, this is what I fervently hope for.
If one day translation engines can produce clever, engaging, and catchy poetic novels in English, with rhyme in iambic pentameter, then it will be time for me to retire from the literary scene.
Original article link:
https://www.theatlantic.com/technology/archive/2018/01/the-shallowness-of-google-translate/551570/
