Machine Translation Examples

 Inspired by the recent movie "Oppenheimer", I tried translating the famous quote "Now I am become death, destroyer of worlds". I compared three different machine translation tools:

Reverso: 今、私は世界の破壊者である死になっています。

Google Translate: 今、私は世界の破壊者、死となった。

ChatGPT: 今、私は死神となり、世界の破壊者となった。

I think it is interesting how the different translation tools interpret the tense here. This is a tricky quote because it's already oddly phrased in English - we wouldn't usually say "I am become (noun)". If I tried to translate those sentences from Japanese into English, I would get something more like Now, I am the destruction of the world, I became/am becoming death. Since it's such a famous quote, I'm sure it has been translated into Japanese by a human translator , and I'd be curious to learn how it is usually done!

Machine Translation Examples

Example 1: Shakespeare

This first example is from  Romeo and Juliet, and the translation for this is a direct translation from english to Japanese, but this is supposed to be somewhat of an insult in the context of the play which is completely lost in the translation here.

Example 2: Excerpt from a book about entrepreneurship

English to Japanese

Japanese to English

For my second example I wanted to take an excerpt from a business focused book on entrepreneurship, and see how it translated one of the sentences, and see how well the translation back to english would be. I think one thing I noticed is that this sentence is probably one that could be split up into two different sentences instead of one long sentence, and it can be more noticeable that within the translation from Japanese back into english, there is a weird stopping point within the sentence which makes sense for the direct translation, but seems like it would sound awkward. 

It was interesting to see that for something like this, it might be better to actually deviate from the actual literal translation and interpret it in a different way.

Examples of Machine Translation

 

    The true translation is “a frog in a well knows nothing of the sea.” This proverb has its roots in older, more traditional Japanese literature, and it seems that this specific writing style is very difficult for MT to translate. 

       This is obviously taken from the clip we watched in class from "Lost in Translation." The translation is accurate for the most part, except for "say meet your old friend." I think that if the Japanese text was changed from 「会う」to「会った」the translation would be more accurate, but I wanted to use what the director actually said word-for-word.

Machine/AI Translations

 How reliable are machine/AI translations?

(Sampled with idioms)

Example 1: Google Translate

「明日のことを言うと天井のネズミが笑う」translates to "the rats on the ceiling laugh when I mention tomorrow" which on its own is quite nonsensical. Although the translation is accurate to a degree, without prior context, the idiom itself is hard to grasp. This particular saying is similar to the English quote "we make our plans and God laughs," and simply expresses the universal truth of we will never know what the future holds.

Example 2: DeepL

「出る杭は打たれる」is a phrase where its meaning is established with understanding the nuances of Japan's collectivist society. Unlike America, Japanese collectivism places an emphasis on being a part of the group rather than individualism, thus the nail which sticks out gets hammered in. The translation itself is accurate but what was unique here was that DeepL provided an alternative 2nd explanation that incorporated the deeper meaning of the proverb.

Example 3: ChatGPT

「井の中の蛙、大海を知らず」is a phrase similar to the idiom "to be a big fish in a small pond" where both meanings stress the importance of widening a limited perspective. ChatGPT did a great job here, not only translating the sentence accurately but also provided a proper explanation for the idiom without being asked to.

While the translations itself were accurate for all three different programs, I find ChatGPT to offer the most literate explanation and translation but for a simple, direct translation? They seem decent enough with some previous context. 

Examples of machine translation from Japanese

 

In Japanese, sometimes, we abbreviate the pronouns after the sentences. In this sentence, the pronouns for "went to Shibuya to eat" should be we. However, the MT would not know it, and used "I" instead. This is a mistake caused by MT. 

This is a paraphrase that Japanese people use often to begin a conversation respectfully. 

However, if the conversation starts with those translated sentences, readers will be confused. 

MT will still face issues with these idioms, slang, and abbreviations. 

Examples of Colloquial Japanese Phrases Translated

 One criticism of machine translation (MT) is that it fails to capture the cultural context of many phrases. I've decided to illustrate this by entering a few well known Japanese phrases into Google Translate.

1. 「物の哀れ」is a literary idiom used to describe the sadness that one feels towards the transience of physical things. It is a mellow yet profound sorrow that accompanies the realization that nothing is permanent.

Interestingly, the above translation is listed in Wikipedia as the literal translation. While I don't particularly agree with the literal translation, it is more important to see that the nuanced meaning of the phrase is completely lost.

2.「月が綺麗ですね」is a very famous expression attributed to Natsume Soseki, who chose to translate "I love you" as the above, which literally translates to "the moon is beautiful, isn't it?"

As one would expect, the literal meaning is correctly translated, but the literary meaning has been lost.

3. 「草生える」is a commonly used colloquialism in reaction to a humorous or comedic situation. It literally translates to "grass growing."

As in the previous example, the literal translation is correctly shown, but the colloquial meaning is lost.

4. 「狐の嫁入り」is an expression frequently used to refer to beautiful yet eerie events. Most commonly, it is used to refer to sun showers and other meteorological events, such as a rainbow. Its origins come from legends from the Edo period, in which supernatural events were commonly written down. There are also many regional variations of the phrase and its meaning. It literally translates to "the fox's marriage."

Surprisingly, Google did output a non-literal meaning. Yet, much of the cultural and regional nuance has still been lost as it prefers the most commonly used meaning.

Examples of AI Translations from English to Japanese

 Can AI Accurately Translate English to Japanese? - Some Examples

Example 1: A simple sentence from a novel

This quote is from Erin Hunter's Warriors series. All that you need to know here is that the 'three' mentioned are obviously non-human (they are cats). While some portions of the translation are correct, such as the first line and most of the second in Japanese, the AI assumes that the protagonists are human despite the mentioning of paws.

Example 2: Academic studies

This sentence was taken from a literature review of tonic immobility in various shark species. The translation here is done fairly well with no egregious errors. If anyone finds any, feel free to comment on this post!

Example 3: Slang

This translation is hilariously awful due to how unnatural it is. The use of 面白い (interesting) does not suit the meaning of the sentence. To describe something as 'funny' it would be better to use 楽しい. Also, native speakers of Japanese tend to use 草 as an equivalent to 'lol,' and not (笑). A better translation overall would be: それはとても楽しいですね草。

Example 4: An English phrase

This translation was surprisingly well done! 薄幸 (ill-fated) is an equivocal term to 'star-crossed.'

Example 5: A famous quote

The translation of this quote is also well done.

So, can AI be used to accurately translate a piece? That depends. Google Translate was able to translate shorter sentences/terms, as evidenced in example 4. It struggles however when a sentence gets more complex (with the exception of example 2), or incorporates slang terms. From these examples, I believe that AI can be used for short sentences/terms and in academic writing, but it would be difficult to use in longer, more complex sentences or when translating social media posts, which use a lot of slang.