If you go back far enough, what language didn't? All writing systems are derived from or borrowed from a script that in its ancient past was pictographic. Braille is the only exception I can think of.
Give emoji a few thousand years, maybe less, and it could become a proper language. I think it's already starting to evolve. The eggplant symbol has been repurposed to mean "penis". (So I've heard.)
But I think the main roadblock to the machine translator described in the article is that emoji are currently used as an adjunct to peoples' native language. Does a Japanese speaker use emoji the same way as a Russian speaker? Or German or Farsi? I suspect the answer is they don't. So trying to train an AI using texts from all these different speakers will be trying to build a Tower of Babel.
To write fish in chinese you write a pictogram. That pictogram is an image of a fish, except the image has evolved over the millennia to be unrecognisable.
The English word "fish" on the other hand, is not a pictogram, and its evolution has not been primarily visual. Instead it has undergone evolution in pronunciation, which has indirectly affected how the written word "looks".
No, to write the word for fish, you write the character pronounced "Yú" which happens to be 魚. Chinese speakers don't see that and think "that looks like a fish". They think "that sounds like Yú". Just as when you see the letter F you think of the sound, not that it looks like a hook: Phoenician 𐤅 - wāw. (Or if you know that the Latin root is piscis: 𐤐 - pē, that means mouth.) And before you point out that a Chinese speaker would describe 魚 as "the character for fish" remember that an English speaker when asked to spell a word might say "F as in fish."
Like I said, all written scripts if you go back far enough started with pictograms. But that was many millennia ago. Their evolved descendants are divided into alphabetic or syllabic scripts, both using the symbols to represent sounds but with a different granularity.
And really, Firefox spell-check, you're not going to recognize "millennia"? What's wrong with you?
Comments
Written Chinese started as an image-based language.
If you go back far enough, what language didn't? All writing systems are derived from or borrowed from a script that in its ancient past was pictographic. Braille is the only exception I can think of.
Give emoji a few thousand years, maybe less, and it could become a proper language. I think it's already starting to evolve. The eggplant symbol has been repurposed to mean "penis". (So I've heard.)
But I think the main roadblock to the machine translator described in the article is that emoji are currently used as an adjunct to peoples' native language. Does a Japanese speaker use emoji the same way as a Russian speaker? Or German or Farsi? I suspect the answer is they don't. So trying to train an AI using texts from all these different speakers will be trying to build a Tower of Babel.
To write fish in chinese you write a pictogram. That pictogram is an image of a fish, except the image has evolved over the millennia to be unrecognisable.
The English word "fish" on the other hand, is not a pictogram, and its evolution has not been primarily visual. Instead it has undergone evolution in pronunciation, which has indirectly affected how the written word "looks".
No, to write the word for fish, you write the character pronounced "Yú" which happens to be 魚. Chinese speakers don't see that and think "that looks like a fish". They think "that sounds like Yú". Just as when you see the letter F you think of the sound, not that it looks like a hook: Phoenician 𐤅 - wāw. (Or if you know that the Latin root is piscis: 𐤐 - pē, that means mouth.) And before you point out that a Chinese speaker would describe 魚 as "the character for fish" remember that an English speaker when asked to spell a word might say "F as in fish."
Like I said, all written scripts if you go back far enough started with pictograms. But that was many millennia ago. Their evolved descendants are divided into alphabetic or syllabic scripts, both using the symbols to represent sounds but with a different granularity.
And really, Firefox spell-check, you're not going to recognize "millennia"? What's wrong with you?