How Far Can Machine Translation Go?

Decades ago, machine translation between English and Japanese was a mere joke. In Japan, we used to laugh about how stupid the outcome was. Rendering difficult Japanese people’s name like 西園寺公経 into something like “West Garden Temple Public Sutra,” (translating individual kanji meanings), the program couldn’t even separate proper nouns, let alone correctly read them.

Today, harnessed by a bunch of smart algorithms, mountainous databases, and a bit of AI, machine translation is evolving into something less and less humourous. I would say it can handle short stock sentences often used for manual writing. I’ll show you some examples of Japanese to English translation (using Google translation) so that you could evaluate the outcome:

Source

電源を入れる。
「L」が表示されたら、体温計をわきにはさむ。
温度上昇がほとんどなくなると、「ピピピピ」と３回鳴るので、検温結果を確認してください。
電源を切って、収納ケースに入れる。

Target (Google Translation)

Turn on the power.
When "L" is displayed, put the thermometer aside.
When the temperature rise is almost gone, you will hear a beeping sound three times, so check the temperature measurement result.
Turn off the power and put it in the storage case.

Now, what’s your verdict? Is the result relevant to you?

Well, you may notice something is wrong with the second sentence since you can’t measure your body temperature if you put the thermometer aside. It’s a fatal error caused by the program’s misinterpreting the word “わき.” Not misleading for humans, machines can get it wrong, primarily when written in hiragana (Japanese alphabet) instead of kanji. To test this hypothesis, let me try again, now representing the word in kanji, not hiragana.

Source

「L」が表示されたら、体温計を腋にはさむ。

Target (Google Translation)

When "L" is displayed, put the thermometer in your armpit.

Now You can measure your body temperature! As you see, the Japanese words “aside” and “armpit” are homophones (both sound and write the same in hiragana) using different kanjis (脇 and 腋).

Then, the most reasonable question might be, “Why don’t they use kanji instead of hiragana, which can be ambiguous?” The simple answer is: it has since become a standard practice in communication for the general public. Being reader-friendly in consideration of diverse kanji knowledge levels, many organizations today prefer to avoid complicated kanjis.

However, for translation’s sake, changing hiragana words to kanji letters can enhance the accuracy of machine translation.

But communication in the real world is not that easy; let’s try another, more challenging assignment:

Source

人間と技術の歴史とは――かつては人間がやっていた「すごいこと」が、技術が進歩することで「（技術があれば）誰にでもできる普通のこと」になる――これの繰り返しだと言えます。かつての人間の特殊技能がもう特殊技能ではなくなるわけです。そうなれば、人間はどうすればいいのかというと、さらに上の特殊技能をめざすわけです。「もう我々の出番はない」などと諦観している場合ではなく、次なるステージをめざすのです。それが人間の進化ですね。

Target (Google Translation)

What is the history of human beings and technology? The “great things” that human beings used to do become “ordinary things that anyone can do (with technology)” as technology advances. It can be said that. The former human special skills are no longer special skills. In that case, what humans should do is aim for even higher special skills. We are aiming for the next stage, not when we are resigned to say, “We are no longer in play.” That is human evolution.

As the author of that sample text above, I’m not happy with the output. First of all, I’m not talking about the ‘history’ of human beings and technology. The software can handle simply-structured shorter sentences but seems confused with complicated or unorthodox syntaxes.

Next, I'll rewrite the piece to make it more machine-friendly and see what happens:

Source

かつては人間がやっていた困難なタスクが、技術が進歩することで、技術があれば誰にでもできる普通のことになる。その繰り返しが、人間と技術の歴史と言えます。かつての人間の特殊技能がもう特殊技能ではなくなるわけです。しかしそうなれば、人間はどうすればいいのか？その答えは、さらに上の特殊技能をめざすことです。「もう我々の時代は終わった」と考えて落胆していてはいけません。次なるステージをめざさなければなりません。それが人間の進化です。

Target (Google Translation)

Difficult tasks that were once done by humans can now be done by anyone with technology as technology advances. The repetition can be said to be the history of human beings and technology. The former human special skills are no longer special skills. But if that happens, what should humans do? The answer is to aim for even higher special skills. Don’t be discouraged by thinking, “Our time is over.” We have to aim for the next stage. That is human evolution.

Now it looks better. But please note that the modified version is a bit rusty as a decent Japanese piece.

The conclusion: With appropriate human help, machines can produce more precise English-Japanese translation, and readers can grasp what’s written there with a bit of imagination.

Lastly, let me say that I’m not against the advance of machine translation. On the contrary, I’m even looking forward to it. When it comes to ‘ordinary translation,’ whose purpose is mainly to know the meaning, machines are handier and quicker.

Of course, there’s a possibility that they may snatch human translators’ jobs. But I’m optimistic about that. As I mention in other pages, ordinary translation and creative writing based on the translation are different, and the latter path is always open to aspiring human translators.

Historically, humans are always destined to evolve.