From the NLP side of the data munching isle, it sounds like a lot more complicated variations on language processing problems.

Comparing reads to each other is like approximate string matching (or fuzzy matching) we use to account for spelling errors in words, but genome chunks are longer and you don't have dictionaries to check against.

And finding the best arrangement of those reads is akin to language modeling where we assign probabilities to word sequences which later can be used to predict the most likely sequence. In case of genomes, though, with so little data, no standard "words", and the error rates in reads, it's like trying to put back a shredded book of in illiterate author only having few other books of illiterate authors as reference.