Machine Translation Response - Afiq

 I've always been fascinated about machine translation and natural language processing. How is something that cannot actually think the way humans do and feel the way humans feel able to accurately translate sentences with the right nuances across different languages? And the answer up until this point is that they can't, not to the extent a human native in said language would be able to express themselves.

Both the articles "Translation of the Future" and "Finding a Voice" bring up the same problem in which MT faces up until now, and that is to translate based on earlier context. In the former article, it is mentioned that a way to combat improper translations of terms in certain professional fields is by creating a model that trains itself only from professional domains. I believe that translating professional texts is more efficient and doable in this manner as texts of that nature usually impose strict and standard rules of writing in order not to cause any vagueness or misinterpretation of the article in its original language. 

I believe that a lot of the problems arise when MTs are expected to flawlessly translate everyday conversations and creative writing, as standard writing rules are not always followed or sometimes manipulated by the author/speaker. This is why the early rules-based approach mentioned in the "Finding a Voice" article was not entirely successful in translating sentences. The phrase-based approach taken after is a lot more successful as it takes the probability of the word appearing in consideration with the words surrounding it, which does play with the "looking into context" part of a human's speech processing.

I do believe, however, that MT can never be fully independent in translating sentences perfect for its situation. At best, MT can do most of the work for humans, but the final touch-ups in making it coherent and flow best with the surrounding sentences and context must be done by humans, and even then, a lot of people don't get it right. I believe this to be because if even humans themselves can't get whatever they say right in certain contexts, how is MT, built by and learns from the very same humans, ever get it perfect? It is however, an amazing tool for people who wants to simple be able to understand and translate the gist of a certain sentence. 

Japanese Sentence:

私、ちょっと好きだな、この人。バターをざりざりと伸ばしつつ、お天気お姉さんを眺める。雪国めいた肌の白さが、なんとなく北国の出身かなと思わせる。

DeepL Translation:

I kind of like this guy. I looked at the weather girl as I slathered butter on her skin. The whiteness of her skin, which is reminiscent of snow country, makes one wonder if she is from a northern country.

My Translation:

I kinda like her, this weather girl. I pay attention to what she says while spreading the butter. The snow-like tone of her skin makes me think she’s from the North.

Machine Translation - Evan

    Picture a young boy, aged 11 or 12, sitting restlessly in his middle school French class. He had been up late the previous night and had not had the time to properly complete a crucial assignment, so he absentmindedly pulls out his phone, tapped a few keys, and zoom, zip, zop- Google Translate pulls up all the answers he needed. Or, rather, what he thought were the answers he needed. Yet, when looking over his assignment with his teacher later that day, he realizes glaring mistakes. The errors are blatant, like a coffee stain on a white buttondown. 

This was my first realization of how lackluster machine translation is. As it turns out, no translation A.I. can wholly and properly maintain the nuance, connotation, and tone from language to language as well as human translators can. For example, Johnson says, "A novel legal argument, says the translator in Madrid, needs to be almost entirely translated by hand. This is not doable by someone who merely knows Spanish and English and took a few translation classes. The bad news for some translators is that a spigot of repeatable, easy work is being turned off. The good news is that what remains will be brain-challenging stuff for people who have a knowledge of a language and something else,". 

Similarly, another Economist article reads, "Fully automated, high-quality machine translation is still a long way off. For now, several problems remain. All current machine translations proceed sentence by sentence. If the translation of such a sentence depends on the meaning of earlier ones, automated systems will make mistakes. Long sentences, despite tricks like the attention model, can be hard to translate. And neural-net-based systems in particular struggle with rare words."

This is most accessibly clear if one were to copy and paste a large chunk of text into Google Translate. Suddenly, adjectives are misplaced, predicates are missing, adverbs turn to adjectives-- it's a mess. 

However, I think that in recent years, what has been the most egregious example of how machine translation simply cannot compare to human translation, is that of social media websites' translators. For instance, take the Instagram caption below.

The meaning of 現像してきた is not only lost but it is completely mistranslated.

I primarily discussed Google Translate and social media websites' translation tools, however, a much more powerful tool such as DeepL is certainly both promising and scary for the prospects of translation. The fact that it is used so ubiquitously shows its efficacy, however I still believe that human translators will be necessary and used in the near future (and hopefully beyond....!).

Machine Translation Response - Airi Hatori

 It was very interesting to see how machine translation was developed back in the 1950s. Although some flaws in machine translation are still present, it has definitely worked its way up to the human level. DeepL almost scares me because it translates very well, and even my coworker recommended that I use DeepL for my translation tasks. I believe that eventually, machine translation will replace translators' or interpreters' jobs; however, I believe that J-E translation will remain challenging for MT since you have to read into cultural context and different nuances that the Japanese language has. 

I think it would lift us translators' shoulders off some workload, but I still believe that translators are necessary to do a final check on the translated texts just to be sure of the quality of the work. I think that the pros of relying on MT are that it is fast and cost-efficient and that it has the ability to translate large volumes of content quickly and accurately. In terms of cost and time efficiency, I believe that companies like Verticle (Kodansha), where the guest speaker works at, can definitely use MT to speed up the translation process and that they wouldn't need a week to translate one episode of anime. 

Here are the translations of the last stanza of the poem called "雨ニモマケズ " that I was curious to see how the different MT would translate differently: 

Original：

日照りのときは涙を流し 

寒さの夏はオロオロ歩き 

皆にデクノボーと呼ばれ 

誉められもせず苦にもされず 

そういう者に 私はなりたい

Google Translation: 

Shedding tears when the sun is shining

Strolling around in the cold summer

Called Dekunobo by everyone

neither praised nor afflicted

i want to be that person

DeepL: 

When the sun shines, we weep 

In summer when it is cold 

Called dekunobo by everyone 

They don't praise me, they don't bother me 

That's the kind of person I want to be

Roger Pulvers ’s translation : 

He weeps at the time of drought

He plods about at a loss during the cold summer

Everybody calls him ‘Blockhead'

No one sings his praises

Or takes him to heart...

That is the sort of person

I want to be

I personally like Roger Pulver's translation the most because he translated dekunobo as a blockhead, and the last two sentences definitely flow better than the other two. I found a very interesting web page that explains thoroughly about how Pulver translated it and it is very interesting. Here is the link: https://blog.goo.ne.jp/sinanodaimon/e/c5bbbaa8f429ca6b836b1b69deb7989d 

Machine Translation Response

I agree with the general consensus that machine translation is convenient and beneficial for people who don’t know the other language and want to translate it into a familiar language. Jaap van der Meer argued that “for many applications less-than-perfect translation will be good enough.” While I don’t agree with his argument that the future doesn’t need translators, I can agree a bit with the statement that nonperfect translation will be good enough, for certain scenarios. For example, Skeb’s translation from the client’s language (often English) into Japanese is not perfect, but it gets the point across to the artist. However, there are cases where the machine translation completely messes up the content of the message and because of that, the artist draws something completely different than what the client asked for. In this case, neither the client nor the artist can be blamed.  

I don’t think human translators can be replaced and I don’t think the quality of machine translation will improve to the point it can replace human translation. With machine translation, you exchange quality for speed, but if the quality is poor, what is the point of speed? I read translated webnovels and it is very obvious when something has been machine translated – the grammar, vocabulary, and sentence structure are all botched up. The really bad machine translations with little to no human editing is terrible and confusing to read, which makes me wonder why someone would spend the time to release these machine translations.

Occasionally I throw the original webnovel into a translator myself to read it fast, because I can read the translated English faster than reading the original, and interestingly, the machine translations I get is sometimes better than the machine translations released by certain translation groups, but I do still get headaches from trying to “translate” from the machine translation to something understandable. There are machine translations that have been edited translators familiar with both languages, and I think this is fine because it speeds up the translation process, but human translations or human involvement in translations ultimately provide a more enjoyable experience than machine translations.

Machine translation has a lot of benefits! I use Google Translate all the time to translate the tweets of people I follow on Twitter, but I always take it with a grain of salt because the translation is not the best. 

An amusing example of machine translation from English to Japanese with DeepL on Skeb: It is pretty accurate but DeepL included extra emojis, which is funny. 

 

Original Tweet:

毎日投稿100日超えたいと思ってるのでぜひインスタフォローして毎日の楽しみにしてください…

Google Translate:

I want to post more than 100 days every day, so please follow me on Instagram and look forward to every day...

DeepL:

I'm hoping to exceed 100 days of posting every day, so please follow my instagram and look forward to it every day...

My translation:

I want to exceed the 100day challenge, so definitely follow my insta for daily posts! please look forward to it~  

I took a lot of creative liberty with this to make it more internet friendly(?) and casual. 毎日投稿100日 refers to the 100 day challenge for posting art daily for 100 days, so I wanted to reference that as the “100 day challenge” instead of a literal translation.

Tiffany

Machine Translation Comments - Kadin

Even with the development of more advanced AIs, I still think that human translation is necessary in proofreading the text. Particularly with Japanese-English Translation, most sentences are hard to translate word for word without context, which is hard for a machine to do. Other types of texts, such as a poem or a comedy skit, would also require incorporating puns or thematic ideas connotated with specific word usage, and I think that an AI won't be able to reproduce something to a similar effect no matter how sophisticated it may be.

I still think that the use of machine translation to help a translator cross reference their work would significantly relieve their workload. Though, I think that it is important for translators to not grow to be too dependent on AI. More reliance on machine translation would lead to more occurrences of mistakes that could have been avoided.

Original Text:

オリヴィが軽く会釈をする。二人はこうして村から村へ旅をする生活を送っていた。困っている人がいれば力を尽くし、解決すれば次の村に行く。その繰り返しだ。

DeepL Translation:

Olivi bows lightly. They lived their lives traveling from village to village in this way. If someone was in need, they would do their best to help, and once the problem was solved, they would go to the next village. It was a repetitive cycle.

My Translation:

Olivi gave a light bow. The two has spent their lives traveling from village to village. If someone had a problem, they would do their best to help, and after the issue was resolved, they would go on to the next village. Rinse and repeat.

Machine Translation Post - Suis

I appreciate Johnson’s view on machine translation because he suggests coexistence between humans and technology. Much of the responses I’ve heard from fellow translators on the topic of machine translation replacing human translators has been negative. Translators often feel threatened by MT. Johnson notes that although MT is improving as AI (deep learning) improves, translations are still inadequate for translating more than basic translations. Johnson suggests that the use of machine translation is efficient and it leaves the “brain-challenging stuff for people who have a knowledge of a language and something else.” Specialized knowledge can be knowledge in classical languages (such as Latin or Classical Japanese) which machine translation is not great for translating (yet). Furthermore, I think about translating literature as something that fits under “specialized knowledge.” When translating literature, one must consider the context, history, and world of the entire text. Word choice is deliberate and important in the interpretation of literature. More importantly, I don’t think MT is good enough to translate poetry.

In the “Beyond Babel” article, it was reassuring to see that the most accurate translations are done by humans and that “computers, no matter how sophisticated they have become, cannot yet truly grasp what a text means.” I feel this is especially the case for poetry. Even if the machine has the “context,” which should be the “entire poem” here, it fails to recognize its entirety.

Original text (poetry):

陽射しに閉ざされた輪郭が

こんなにも　ましろく

君を　漂白する（ケシテシマウ）

私は

君の残像を　なぞり　ながめ　ながら　あるく

Machine Translation (DeepL):

The contours closed by the sun

So pale

I bleach you.

I trace and ponder

While I trace and ponder your afterimage.

My Translation:

The contours of your body, enclosed by the sunlight

Is so very   white

This  bleaches and erases  you

I

Trace and  gaze at  your afterimage  and  walk

Machine Translation

 I see a lot of people online use machine translation to communicate with people of different languages, so machine translation is definitely useful—however, it is generally very easy to tell if someone is using translation software or not, especially if the text being translated is from a non-Roman language to English and vice versa. I highly disagree that translators will not be needed in the future or that translators of the future will be like “quality-control experts.” Words translated by translators will always sound more natural, and while translators can interpret meaning in text, machines can’t. Translators aren’t just looking at quality—that’s editors—they’re also creating the things that need the quality checks. Language is always evolving too. Words used today may be used differently in the future; new words may pop up; machine translation software will always be playing catch-up in addition to trying to train more on the current languages.   

I have read (web)novels translated by groups who only use machine translation, groups who use machine translation and then manually edit, and groups who manually translate and edit from the source novel. And the quality is in that order too, from worst to best. The groups who manually translate and edit have the best translated novels; they try to make the puns and riddles work in English and it’s fun. Overall, the reading experience is smooth and enjoyable. The groups who machine translate and then manually edit have decent quality; some puns, jokes, names, culture specific things don’t get across and it’s ok since I usually know what they’re referring to since I know the source language. The reading experience is decent but lacking. The groups who machine translate—I don’t even finish. Sometimes I’d start reading and think everything is ok, decent enough, and then all of a sudden, the names or pronouns change, and the grammar becomes horrid and nothing makes sense anymore.

Machine translation isn’t bad though. It makes language more accessible, and I think more sophisticated machine translation software may be helpful to translators the way premade assets and models are to digital artists--simplifying the process. However, I think machine translation is better for more casual translations like short messages/grammatically correct sentences.

In addition, sometimes when people see the (machine) translated text, they will/may have preconceived notions of how the text should be translated and that way might not be the best way. So maybe machine translation will be a tool to aid translators or maybe machine translation will be a rock that trips up translators.

 Original text:

#今年も残りわずかなのでお気に入りの4枚を貼る

 

My translation:

The year is almost over so I’ll post my 4 favorite pieces

Maybe #2022Favorite4Artworks will work for this

Google Translate:

There are only a few left this year, so I'll put my favorite 4

 

DeepL:

I'll put up four of my favorites as there are only a few days left in the year

Both machine translations get the meaning across, though the Google Translate one may be confusing to readers; however the machine translated ones don’t translate well to an actual hashtag in English.

 

-Lesley

Machine Translation: A Response - Reshma

     I found the topic of the paper interesting, especially given the sudden increase in my friends using fairly sophisticated AI in the last one week. I'm not sure how I feel about this topic entirely. AIs definitely won't be able to translate things to the extent a human can with all the nuance and emotion, but I'm willing to say that they could make it easier on translators. I have a friend who plays a Japanese mobile game and doesn't know the language. She uses an MTL program to understand what's going on, and as far as I can make out, it gets the point across, even if the translation isn't great on its own. With more sophisticated AIs, it could be feasible to use AIs and edit their output using proofreaders. 

    I think this could definitely help with reducing the workload of translators, and more importantly, help people share their work despite language barriers, not having to hope to one day be translated. That said, a lot of the calls we make in class– formatting, use of addresses, word choice– these questions remain even without the use of a machine. I don't think the role of translators will end up dying, they might just find themselves specializing a bit more. 

Original text: 

未だ書かぬ自分の作品の計画を語る場合に於いては、作者はたいていこのやうにあどけない法螺を吹くものである。そんなに、うまくは行きませぬて。）まあさ、とにかく、まあ、聞き給へ。どうせ、気焔だがね。とにかく、ひやかさずに聞いてくれ給へ。

Machine translation (DeepL):

(Authors usually blurt out this kind of bluff when discussing plans for their own works that have yet to be written. It's not going to work that well.) Anyway, anyway, listen to me. It's a flame, anyway. Anyway, please listen to me without any irritation.

My translation:

(Authors often tend to make naive exaggerations when talking about works of theirs that are yet to be written. Everyone knows that it doesn’t usually go that well. But do me a favor and just listen. It’s all just hot air anyway. Bite back your snarky comments and hear me out.)

Machine Translation Response - Elliot

 We've seen significant strudes in AI technology in recent times, and I'm interested in seeing how its place in the translation world develops. While AI obviously won't be able to spot the more subtle aspects of something like literature translation, I agree it has value in reducing the amount of time spent on the more mundane sections.

I consider myself apprehensive towards AI in general, but I'm not especially pessimistic in terms of how it will affect the industry. While corporations will certainly prefer automation for its price, human translation staff will remain indispensable not just to catch errors, but to translate things that AI can't handle. In the end, corporations do have to bend to demand, and I think the perception of MTL being objectively inferior to a human translation will remain, even if it somehow becomes indistinguishable. MTL is viewed as compromising on quality, so I think demand will remain for professional translators.

Upon examining a machine translation of my project, I'm even less optimistic about it. It falls into the pitfalls of Japanese-English translation even harder than humans, since Japanese requires a lot of context awareness. It usually can't figure out what the subject is, and it can't really make long sentences with lots of modifiers sound good in English. I can certainly see it working better for translating between romance languages, though.

Original text:

男が喚き散らす。​ 

そのたびに、男に繋がった縄が左右に引っ張られる。​ 

元気なものだ。​ 

DeepL translation:

The man rants and raves.

Each time this happens, the rope connected to the man is pulled from side to side.

It's a cheerful thing.

My translation:

The man screams wildly.

With each yell, he strains against the ropes binding him.

He's a lively one.