Perhaps you need a data set of language containing the vocabulary and grammar rules. Next you leap over the rules making into various permutations and combinations to form sentences but the sentences need words. So what to put up first? The vocabulary. Scanning the dictionary for words and picking up the noun, the adjective, the verb and the object, a sentence could be formed. Such formation demands syntax analysis which we have already decided to put through various permutations and combinations.
Next we need a master poet to establish the meaning of the sentences and to enthuse some feeling. The feelings could reside in a database with various types of feelings. Picking up the best feeling you want to convey through the mood could be achieved by the master poet whose work we must emulate into the database prior to making the computer learn to write poems or in other words learning a language.
She bathed in the nectar mixed with milk
All dryads went for Jimmy the pix, our computer model
Well it could recite Milton as well as make bunk notes
For the passengers on the airliners.
Next we wanted pix to be a lover so swoon her
As she would type the lover letters on the pix’s screen
To group similar things together is the key to make a machine learn a language. There has to be some concept that defines words or give them the meaning so similar words having the same meaning fall into a cluster or a group. Do we have to call out for Mr. Sigmund Freud to do the psyche analysis as how the machine develops to have a functioning of making sexual assertions? Of course! Thus lending the machine an intuition and the feel.