I'm aware this could be a complex topic, however I'm interested in existing research projects or studies where people are attempting or have succeeded in teaching an AI a foreign language just by training/teaching it from English books. By reading, analysing and understanding, so that it knows the foreign language's rules (such as grammar, spelling, etc.), the same way as a human would learn. The language doesn't have to be Chinese, which is difficult for even humans to learn.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Current approaches for learning a language require having a large corpus of that language; it also doesn't seem reasonable to expect that it will ever be possible to learn about language A by extracting information from a corpus from an unrelated language B.
Even if you want to learn about human languages in general (what sorts of things are true about grammar, vocabulary, and so on), that relies having many languages as training data, so that you can see the different ways of doing things instead of assuming that the way they're done in English is the way they're done in every language.
(There is work in automatic translation that goes from a language to 'concept-space', then goes from that 'concept-space' to another language, so that you can build an English-Chinese translator by building two separate English-Concept and Chinese-Concept translators, instead of ever needing material that directly links English and Chinese. The obvious benefit of this is scalability; in order to make translators for a new language to any other language, you just need to learn that language and the models build themselves.)
