Artificial neural networks do both supervised and unsupervised learning. While this is a very rough generalization, unsupervised learning is good for building models and encodings of data, while supervised learning is good for minimizing the error rate when answering questions about data. The state of the art for training neural networks is to "initialize" the network with unsupervised learning, then "tune" it with supervised learning.
To me, a child learning from listening to the TV is like a form unsupervised learning. It probably helps them build an internal representation of language structure, but doesn't teach them much if anything about proper use of language. If I had to guess, I'd say it is probably mildly helpful when the child is very young (especially if the alternative is silence), but it probably stops providing any value fairly quickly (by about a year of age for the typical child would be my very uneducated guess). At this point, I'm guessing improvements in language facilities probably require focused interaction (supervised learning).
To me, a child learning from listening to the TV is like a form unsupervised learning. It probably helps them build an internal representation of language structure, but doesn't teach them much if anything about proper use of language. If I had to guess, I'd say it is probably mildly helpful when the child is very young (especially if the alternative is silence), but it probably stops providing any value fairly quickly (by about a year of age for the typical child would be my very uneducated guess). At this point, I'm guessing improvements in language facilities probably require focused interaction (supervised learning).