I would clear 1,000 islands of all inhabitants and stick large populations of babies on them with caretakers who don't speak to them and see what kinds of languages they develop. This would be helpful in figuring out how language originated, and might provide interesting insights w.r.t so-called linguistic universals.
The problem with feral children is that they don't interact with other human beings until they are past the so-called "critical period" for language acquisition. The primary purpose of language is communication, and if you don't have anyone to communicate with you're not going to start spontaneously producing language. And then by the time these children are exposed to language, it's too late for them to be able to fully learn it. Feral children are few and far between, but we have a lot of cases of deaf children whose parents don't know sign language, and then they don't learn language until they are much older. Studies show that they can't acquire native-like competence in sign language. Here are some papers on functional and structural brain abnormalities of late learners: 1 2. Here are some papers describing linguistic behavioral deficits of late sign learners: 1 2 Putting a bunch of babies together doesn't have this problem, though. If you're surrounded by other human beings then there are opportunities to communicate, and thus it's possible that language could emerge. It also circumvents the problems of late first-language learners of sign language: deaf babies are surrounded by adults, or children who already know a language, in their formative years. There's evidence that adults are much less innovative than children when it comes to learning and creating language, so we wouldn't expect for a new language to emerge between a deaf child and his/her parents. In short, putting a bunch of babies on an island would be awesome because they don't have another first language to interrupt creation of a new language, they aren't socially isolated so they have motivation to communicate with others, and they haven't reached the "critical period" for learning language yet.
You'd want to make the caretakers as impersonal as possible, maybe even to the point of being robotic. Consider how much of communication is nonverbal. We even have languages that are completely based on hand gestures. Even the slightest reinforcement of certain nonverbal actions by the caretakers might have a profound effect on how language develops. Then again, if you had a thousand islands full of babies, you'd have room to systematically alter variables like level of contact between babies and caretakers and see what differences those alterations made.
Yeah, that's a good point! Deaf kids of hearing parents develop pretty advanced homesigns which might be a result of the building up of communicative skills gained via nonverbal communication. Susan Goldin-Meadow has a case study of a kid whose homesigns even exhibit recursion (although it's still heavily debated since sign data are notoriously hard to analyze). But nonverbal communication is probably a necessary precursor for language...hmm... Revised plan: 2,000 islands! Half with no nonverbal communication, half with haha.