TAKE one baby. Install a network of microphones and video cameras to monitor him 14 hours a day, 365 days a year until he's 3 years old. No, it's not the latest reality TV series, but a project designed to provide a unique insight into the role environment plays in the seemingly miraculous process by which children acquire language.

Deb Roy of the MIT Media Lab in Cambridge, Massachusetts, has so much faith in the project that he volunteered his own family as guinea pigs. Since his newborn son returned from hospital nine months ago, 14 microphones and 11 1-megapixel "fish-eye" video cameras, attached to the ceilings of each room, have been capturing a continuous stream of experiences, such as the mother using a melodious voice or frequently repeating a certain task, which the researchers hope to match to changes in the baby's linguistic performance.

How babies go from scarcely being able to gurgle at birth to chattering fluently by the age of three is still hotly disputed. Most psycholinguists agree that just listening to speech is not enough for a child to learn the basic rules of a language, but they still argue about how much of the extra information comes from specific "language genes" and how much from other stimuli in the environment apart from words.

In the past researchers have recorded mothers and babies playing in the lab or gone into their homes to observe them. But following babies in such an unnatural atmosphere does not reflect their normal experiences, and makes it impossible to tell whether changes in a child's speech are sudden or merely appear that way because of gaps in the recording.

Roy hopes his "speechome" project, as it has been dubbed, will fill in some of these gaps. "It allows us to put a microscope on the day-by-day and hour-by-hour changes that go into learning a language," says Steven Pinker, a psycholinguist at Harvard University, who is an adviser to the project. "Nothing remotely on this scale has ever been done."

The cameras are switched on between 8 am and 10 pm each day and will capture 85 per cent of the baby's waking hours up to his third birthday. Roy and his wife can switch them off when they choose using wall-mounted touch display, or press an "oops" button to delete recordings. Every week the data is transported to a vast petabyte storage sink at MIT, where it is downloaded and processed to extract meaningful results from the recordings (see "MICS, Camera, Action" and Graphic).

Whether Roy will be able to make sense of the data remains an open question, says Linda Smith of Indiana University in Bloomington. But if he is successful, Roy says the project could lead to better strategies for diagnosing and treating language disorders and a even a new breed of robots that learn to speak without being programmed with language.

As for the ethics of subjecting his son to such surveillance, Roy believes he is providing him with an incredible gift. "He might be the first person to have a memory that goes back to birth," he says.