I normally complain about people using features like regex when simpler mechanisms are available. In this case, I think you are over-simplifying, with substr(), when you could batch process. But I see you are collecting the punctuation you see, at the top, although you don't do anything with it ... maybe that's a bit of code you cleared away as not relevant to the problem.
What I would consider is merging the punctuation regex with splitting the line into words, using split to partition on non-word characters ... that is, not alpha, not numeric, not underscore. If that's too generous, you can be more specific.
my @words = split /\W/, $sen;
Also, how many NOUNS are you dealing with? If it's only a few million, I would read it into a hash, and check each word against the hash. Reading the file dozens, hundreds or thousands of times, is ghastly slow. A few megabytes for the hash is not excessively painful. Maybe you can save a copy of nouns.txt split into one word per line ... or save it as a YAML file or some other format that loads quickly as a Perl data structure.
As Occam said: Entia non sunt multiplicanda praeter necessitatem.