Not totally sure what you're asking here. What don't you get?

The PunktTokenizer is just a tokenizer that has been trained to detect the "part of speech" for a given word in a sentence. We want to know the part of speech, since it has a significant impact on the actual meaning itself, and if a machine knows it, it can better help the machine to understand the meaning of sentence.

If you just want the verbs or adjectives from a sentence, or maybe just the nouns, you can also use it to filter.