Up until yesterday I thought the Japanese I spoke was fairly close to the way a native speaker would speak. I was wrong.
Yesterday my girlfriend showed me the speech recognition feature on her iTouch. She demonstrated by speaking a few long sentences that were subsequently translated into garbled English. I understood exactly what she was saying so I thought the machine was broken. I then read a long passage and, lo and behold, the machine translated it perfectly! I was shocked.
We then did the reverse: I spoke in Japanese and so did she. I said my favorite phrase: child cat (kodomo neko) and was shocked when the machine didn’t translate it perfectly. I tried it twelve times – each time the machine gave a different answer – in a row until I could finally get the machine to understand me. My girlfriend then proceeded to try: First she tried to say it correctly, and then she repeated the word again incorrectly to make fun of my pronunciation. The machine recognized both versions perfectly. I was shocked, again.
I practiced those two words all night and, finally, in the morning, I was able to reproduce them accurately enough for the machine to understand three times in a row. It was a humbling reality check.