hw4p6.txt

Thinking Questions:

1. The generated word sequences bear very little resemblence to real English. To
	improve their readability, we could use larger n-grams, so that our sequences
	contain larger "correct" sub-sequences. Another possibility would be to use
	part of speech tagging to further weight subsquent word choices such that some
	structure of a grammatically correct sentence is adhered to if possible. This
	would essentially be adding conditions to the conditional probability of an
	n-gram existing after some other n-gram.

2. If my function encounters an unseen context, it terminates the sequences and
	returns the sequence thus far as a sentence. An unseen context could be the
	last token of the corpus, if the very last token of the corpus is unique, 
	and therefore has no "next" words to choose from.