Google translate VS DeepL

If you’re relying on it to learn how to form sentences, then yes. Google Translate usually doesn’t produce grammatical errors but it often produces nonsense, really bad prose, awkward word choices, incorrect or extraneous context, and inappropriate politeness level. I wouldn’t really recommend machine translation tools for trying to learn grammar. They excel at getting some kind of rough idea of a text’s meaning when you have absolutely no knowledge of the language. If you’re studying the language, it’s not really going to be very helpful IMO.

Edit:

Also “translating literally” is kind of a weird phrase because it’s really hard to define what exactly that means. If you mean that google translate often misunderstands idioms, yes that does happen at times, and it can be difficult to decipher what the original meaning was.

2 Likes

Thanks so much Phyro. I’ve noticed that my understanding improves when I watch adult Japanese shows versus Japanese Anime (unless the Anime uses advanced concepts). But I’m still pretty confused. How is DeepL better?

Here’s an example from the show “Midnight Diner: Episode 1”
Gay dude is waiting for his Yakusa Crush to show up. He sighs and says…
今日は来るかなと思ってたけどもたフラれちゃったみたい
my mind automatically asks myself… did they mean…
今日は来るかなと思ってたけどもた振られちゃったみたい
and I translate it to mean…
“I was thinking he would come today, but I guess I was flirting.” (as to say it wasn’t anything real)
and Deep L says…
“I was hoping he’d come today, but it looks like he’s been rejected.” — okay, that’s what I thought.
and the subtitles when I switch it over read…
“I thought he’d be here today, but I guess he dumped me.” —Getting closer…
But Google translate says…
“I was wondering if I would come today” —That’s it? for that entire phrase? LMAO? what???
And if I input the way I would translate it in my head… just hearing it…
今日は来るかなと思ってたけどもた振られちゃったみたい
Google translates even further away from reality…
“I was wondering if I would come today, but it seems like I was shaken” — LOL

It’s just when you want to learn grammar and the translations are adding things like “I guess” and “here” when they aren’t even in the sentence, an amateur such as myself can get really confused.

Maybe I shouldn’t be learning my Japanese from a seedy Tokyo Underground, a Gay Club Owner and his soon to be retired Yakusa Man Lover. LOL. But I find the conversations really interesting, all the abbreviations and nuances and such.

I watched the anime Parasyte and absolutely loved it. But there were no Japanese subtitles, so sitting in on a University Lecture on the anthropomorphic relationship between ego (for the sake of survival) and altruism (for the sake of the common good) was lost on me one episode. So sad.

Then I found the show Eraser 僕だけがいない街 which is offered as the anime with English subtitles, but more importantly as the TV show with both English AND Japanese subtitles. When I watch shows in French, Spanish, or Russian, the spoken languages and subtitles don’t always match up. I can only assume this happens with Japanese as well as my Japanese isn’t good enough to catch those slips. I can read the subtitles really well and often ask myself “How come they didn’t just use the kanji for ___? It’s simpler and more elegant.”

I’m a visual learner. (which explains why Rosetta Stone works so well for me). I like to attach as many senses as possible, even feelings/emotions as well to cement a concept. So reading a book of phrases is a real bullet to the head for me, but watching scenarios fold out in front of me over and over in different ways with different people is how we naturally learn. (again…the concept behind Rosetta Stone) But Rosetta Stone doesn’t use Kanji and it sure as heck doesn’t teach past low intermediate level even after you complete the entire program.

When I learn a Kanji now, from WaniKani, Anime, or a TV Show, it’s difficult for me NOT to want to put it in sentences and play around with its usage to cement it to memory. That requires a good translator. Learning HOW to use a phrase properly and in the right context I think is just as important as learning the phrase itself. I remember the first time I learned 勝手…as in… あとは勝手に注文してくれりゃあー It was a totally foreign concpet and made no sense. I was memorizing the definition with zero context. Like a damn robot “as one pleases…” Level 12! I’m almost at Level 52 now!

Any advice is much appreciated.

1 Like

I like ichi.moe for this kind of thing: https://ichi.moe/

It does extremely literal word-by-word translation. Basically what you would get by looking up each word in a dictionary, only faster. (Strip furigana first, or you’ll get each word twice.)

I like it precisely because it doesn’t attempt to parse the grammar. That’s left up to me.

4 Likes

Good luck with that. Anglophones aren’t even willing to embrace the metric system.

4 Likes

I think this is a great example of what we’ve been talking about, because the most “literal” translation here is the English subtitles - “I thought he’d be here today, but I guess he dumped me.”

To my eyes that’s a very direct rendition of the original. To try to show the original structure a little more, here’s an even more direct (less correct in English but trying to mimic the original structure as much as possible) translation from me, with implied subjects in parentheses:
"today, (I) wondered if (he) would come, but (I’ve) been dumped it seems.

DeepL works hard to make correct looking English, and does a pretty good job, but it can’t fill in the full context, so it’s only possible for it to guess as to the subjects, so the shift to the speaker as the subject for フラれちゃった isn’t caught - and also, although it happens to be correct here, it goes without saying that all the gendered pronouns here are 100% guesswork on DeepL’s part.

Google gets the subjects even more wrong (not figuring out that if you wonder if someone would come today, you’re probably not thinking about yourself), and completely whiffs the slang-ish meaning of 振られる.
In retrospect, it’s impressive that Deepl managed either of those things - that’s the quality people talk about as Deepl being an improvement over Google Translate.

But - for your goal of understanding the structure of the sentence, I feel like both were misleading and unhelpful. From your original translation, you got the idea, but from my interpretation your main stumbling block was not being sure exactly how 振られる is used, and like DeepL, not being sure on the subject in the second part of the sentence.
You got the idea of the situation, that he met this person, thought it was something, and now he’s wondering if maybe it wasn’t a thing after all, and DeepL got closer to the sense and usage of 振られちゃった with “been rejected,” but because DeepL gets the subject of it wrong, I think structure-wise it may just muddy the waters unless you’re really confident with knowing the subjects in Japanese sentences (and I don’t think very many native English-speaking learners are, since it’s one of the main difficulties between the two languages)

That’s why personally I would recommend just sticking with the Japanese original as much as possible and puzzling it through - and if you need to check yourself, reach for a human help or an official translation before machine translation.
You can try to break it down yourself slotting in English directly into the Japanese structure as something like :

talking about today, come maybe? thought, but got dumped seems like

And in my opinion that’s a more helpful “translation” (especially if you do it yourself) since even though it’s clearly much worse English, it’s closer to the Japanese structure so there’s less opportunity for guesses to impact further guesses. If I tried to start inserting subjects to make better English - I might be wrong, and that might lead to more mistakes.

And that’s basically what DeepL does all the time, because it super prioritizes “the output should look good.” There just is never a way to ultra-directly translate Japanese to English or vice versa that looks like good English - they’re different languages. So it will always introduce inferences that will be harder for learners to detect and account for the newer they are to it.
So especially if unsure about exactly who’s doing what in a sentence, I think it’s better to revert to the original and try to break it down (or ask someone) rather than reach for Deepl or google translate.

Hope that helps!

7 Likes

you sound like straight Yoda and that’s precisely what I’m hoping for. I want to be able to break down a sentence, understand each part, and apply it to other “example sentences” I can create myself utilizing some form of machine translation or auto-fillin to help guide me. If I was explaining the translation to someone…sure… but I want my little pretty head to Yoda the heck out of it so I can drill it and drill it, so it will start making sense, the same way multiple encounters in real life can actually help you remember a grammar point organically.

ie:
緒:しょ together (*when I learned this with wanikani it had zero context other than the mnemonic)
一緒に+verb= let’s do ____ together!
hmmm…
一緒に行こう
一緒に頑張ろう
ご飯一緒に食べるべ
今度はさ一緒作るべ

so there’s just this…natural progression of usage that only came from “hearing” these phrases in multiple episodes of the Japanese TV Show Eraser, based on the 漫画とアニメ, 僕だけがいない街。

but if I could just type in…
一緒に

and get a list of example sentences that don’t suck… (Google gave me only one of those… 一緒に頑張ろう) that would be freaking stellar.

No sooner did I post this…did I find this other post on an app called yomichan? is that a thing? here’s the thread.

and…this… https://ichi.moe

フラれる can be written like this, but yeah it means to get rejected or dumped or things like that. I would translate it as “I thought he would come today, but it looks like he ditched me.” The reason the machine translators are introducing things like “I guess” is because you can translate みたい in different ways. “It looks like” matches the Japanese wording the most, but it’s not necessarily the best way to translate it in every context.

Also I don’t think DeepL is better, I personally don’t think machine translation tools are very useful outside of using them for entirely unknown languages. In that case, it’s “better than nothing” as it were. My view is a bit extreme so feel free to disagree with it; I know some people find it useful for getting a gist. I often find what it spits out to be more confusing than helpful, and even when it doesn’t make any mistakes, it incentivizes me to give up on sentences and rely on English instead of trying to understand them in Japanese.

This happened because using the kanji gives the impression that you intend the literal meaning of “shake” instead of the metaphorical meaning. A native speaker would probably understand from context what you meant, but the machine translators probably rely on whether something is in kanji or not to help them try and guess the context. I think that writing it in katakana here makes it more clear that it’s being used in its figurative meaning of “getting dumped/ditched/rejected.”

1 Like

Yeah, and I think unfortunately it’s that level of precision and correctness that machine translation unfortunately won’t be able to really help with. It’s fine for “any port in a storm,” “I just need a quick guess at the overall picture of what this means,” but sadly it’s not gonna be “drill it and drill it” quality. I think in that kind of context you definitely want to avoid introducing errors if at all possible, and machine translation can’t guarantee that.

Seeking out resources with lots of example sentences is a good move!
I’ve generally just let usage subtleties bleed in slowly over time from reading (and I haven’t focused on production), so I can’t help seek those out, but I can at least attest that stuff does get a lot clearer over time. I can remember learning フラれる and a time when things like the られちゃった would’ve come to me a lot less naturally… Which is just to say time and persistence are major helpful factors here too.

2 Likes

precisely what I’m talking about.
みたい is such a common ending and yet the meaning varies like crazy! I’d love to be able to type in みたい and find “common phrases” based on google posts… it would be great if there was a ranking right? of how many times a phrase is used and it what way… like 52% of the time it’s used this way… and…2% of the time its misued used that way.

Hmmm…does this exist? Because if it doesn’t… go go computational linguistics! I want to code the heck out of that!! it would be sooooo useful for language learnings of all languages…

sure it would be bad in terms of conmen who want to “pretend” they are native speakers. one of my ways of testing the validity of people is throwing a mess of idioms at them.

(Russians have the best idioms in my opinion with people ranging from Turkey to Kazahkstan claiming to be living in Moscow).

But I digress…

please check out my reply to the following thread.

This is kind of intense (and I don’t actually know much about this resource so I’m not sure, for example, how the corpus was sourced), but you might be interested in https://tsukubawebcorpus.jp/

e.g. 読み込み中…┃NINJAL-LWP for TWC
gives more-or-less what you’re describing.
So I can see for example, that みたい + に is waaay more common than other particles, which makes sense:
image

It’s just a lot of intimidating data and all in Japanese!
So it can be more exhausting to use than the text you were hoping to learn about…

I’ve definitely gotten a bit of use out of it though sometimes for picking through how words might be commonly used.

(Pulling example sentences from it is tempting sometimes but I would caution against it, as they’re generally snippets from intense texts and long passages, not full complete sentences meant for learners.)

8 Likes

I used to think they same, then I realized it’d turn the American, British, Australian, etc varieties of English into entirely different written languages, and I doubt anyone wants that to happen.

My alternative solution: replace the Latin alphabet with just kanji.

1 Like

There’s no need to encode regional accents into a standard language. German is (mostly) spelled phonetically, but people still keep their dialects. Just pick the predominant accent and use that.

Good luck getting every country to agree to that :grinning_face_with_smiling_eyes:

1 Like

yeah, I am. it’s proof that what I’m looking for doesn’t actually exist. as a native speaker of a language, sure this would be interesting, but for a language learner, you’re right, it’s pretty intense.

this isn’t how people “learn a language.”

It goes back to childhood development, creating social interactions in our mind even when they are not available to us. “Playing House” and so forth. We create scenarios and we work out word usage and phrases, repeating things we heard even if we haven’t fully grasped their meanings. The more often we hear the phrase, the more we can learn the proper usage of that phrase. But just like a person who mispronounces a word or uses a phrase out of context, this is oftentimes due to learning this from reading text instead of actually experiencing the phrase or word through cultural immersion. How can we recreate that process? Well we do have some interaction when we watch t.v. shows and film, but its still very passive learning.

I think I’m on to something. I’ll have to noodle on this a bit.

I think an obstacle for a bilingual learner’s version of a database like that is it’s going to be a lot harder to figure out what to measure.

As an example, the english word “like” used as “I like pickles” is certainly a different meaning/usage than “it’s like an accordion,” but are “it’s, like, an accordion” or “I like you” also separate senses? The line between homophones and just a word that encompasses a lot can get blurred pretty fast, especially if you have to figure out how to express those in a second language, where one word in the original might correspond to multiple even in just one sense. The monolingual one can get around all that by focusing just on what words it shows up next to rather than the exact meaning.

I think maybe the closest thing to what you’re looking for would be grammar patterns, in a resource like the grammar dictionaries, (or some other alternatives). In an entry they show enough of how a common structure is formed, the distinctions of what it means, and plenty of examples, I think well enough to start playing around with the phrase yourself. That might be something else to look into, if you haven’t already.

1 Like

What are we? Living in the dark ages? We are living in the age of machine learning, algorithms, and hyptertext. There’s no need for such archaic practices as “grammar book dictionaries.”

Think of it this way… a long long time ago…in a far away land called San Francisco…the world wide web was nothing more than a bunch of random disconnected websites that you only could check out if you had the precise web address. One letter off and you were screwed. Then people started to catalogue and group their favorite sites together, posting their own conpilations online. These places became hubs for users, a go-to for resources, taking countless manhours to keep up to date. It was an archaic process, but it was the best we had. Then the search engine was born, and you know what? it sucked. It was terribly innacurate. In fact it was innaccurate the same way as the precise example you gave! Like: “I like pickles”…“it’s like an accordion,” Not only could it not differentiate, there was no telling what crap would pop up! And the more sites that appeared, the worse it got.

But theres this wonderful thing called users…and they are self-correcting. We could collectively make such a search engine accurate, over time. The more native speakers used the software, the more accurate the results would eventually be.

But text is boring. When it comes to language learning, it’s environment and behaviorial social cues that help guide meaning and memory. First in a rather binary fashion and then later with greater complexity and nuance. We see this with every kanji we learn with wanikani, these mnemonics suggestions that wield our imaginations when we have a society overflowing with its own imagery to boot.

Right now language learning online is much like the internet in its nascent, a one way conversation, disconnected, ineffective, and requires extensive personal fortitude, sifting through useless data for little gems of goodness. The same algorithm that is utilized to figure out which YouTube Recommendation could be used to create such a global language learning database. I think this is what the field of computational linguistics was made for. I need to noodle on this a bit.

It would be better if DeepL would write something like (he/she/it) instead of “he” all the time, so people could know when they have to infere the context.

If language would be taught like that I would never be able to say what I want. I always only find music I love and books I love with a research that guides me along my interests (and that includes also conversations with strangers who point out interesting aspects). No algorithm could ever replace that.