Let's suppose the machines are spying on you and you want to learn a secret language of humans, a language with the highest ratio of human speakers to NLP technology. Which language should it be?
I think it would be a language spoken in populous but under-served areas, where English is a widely-used second language, so those who do use computers are content to use them in English. But not spoken by so many people that you could expect people to be prioritizing it for NLP anyway.
Tamil and Hausa would be my guesses. 70 million speakers each, largely in India and Nigeria respectively, both awash in English as a second language.
I think American Sign Language would probably be an okay choice, given the extra overhead of image and video processing, and the added requirement of line of sight possibly disrupting partial conversations. It would also force a strict interpretation of multiple channels, sight and sound, in case information gets dropped.
Anyway, recontextualizing existing languages often works well against adversarial humans, and probably would find some effectiveness against computers, since computers are rarely permitted to deviate from commonly understood norms, because when they do it is often an error.
Consider humor in foreign languages. Puns and punchlines often just disappear when you cross the language barrier. People wind up laughing at jokes that only a fluent speaker acclimated to local norms could ever understand, and you often find yourself reduced to elementary school conversations, unable to deal in the sophisticated sense of humor adults develop as part of culture. It takes a special person to immerse and bridge cultural divides like that.
3 comments
[ 3.1 ms ] story [ 21.0 ms ] threadI think it would be a language spoken in populous but under-served areas, where English is a widely-used second language, so those who do use computers are content to use them in English. But not spoken by so many people that you could expect people to be prioritizing it for NLP anyway.
Tamil and Hausa would be my guesses. 70 million speakers each, largely in India and Nigeria respectively, both awash in English as a second language.
Anyway, recontextualizing existing languages often works well against adversarial humans, and probably would find some effectiveness against computers, since computers are rarely permitted to deviate from commonly understood norms, because when they do it is often an error.
Consider humor in foreign languages. Puns and punchlines often just disappear when you cross the language barrier. People wind up laughing at jokes that only a fluent speaker acclimated to local norms could ever understand, and you often find yourself reduced to elementary school conversations, unable to deal in the sophisticated sense of humor adults develop as part of culture. It takes a special person to immerse and bridge cultural divides like that.