Do we need linguistic knowledge to build NLP systems? One argument for it is that even humans seem to have some form of innate linguistic knowledge: Children learn a new language faster and from fewer examples than what seems to be possible if they learned from exposure alone. How could this be reflected in NLP?

You are sold to the new paradigm of neural NLP. Which statement is a good fit for your views?

Many NLP problems can be described with the help of this formula:

$\hat{y} = \mathop{\text{argmax}}_y \text{score}(x, y; \theta)$

You want to build an NLP system that generates the most probable headline for a news article. What is $y$ in this case?

You have a sentence consisting of 18 words. Suppose that each word can have two possible part-of-speech tags. (For example, the word ‘want’ can be both a noun or a verb, depending on context.) What is the size of the search space of all possible tag sequences that an NLP system would have to consider?