Might be a bit overkill:
If you want to find paragraphs which might be incorrectly split, here's what I've come up with - it needs a little tweak sometimes, but generally rather good. I wouldn't recommend replacing everything, unless you grep first for results (think I have an alternative with span/[bsiu]'s ignored somewhere... mmm).
Code:
(?smi)(?<=[^[:punct:]])</p>\s*<p[^<>]*>(?=[\.-?])|</p>\s*<p[^<>]*>(?!\s*(<[sbui]>|[[:punct:]\s])+[[:upper:]])(?=[[:punct:]\s]+[[:lower:]])|</p>\s*<p[^<>]*>((?=[ \.>]{2,}([[:punct:]]|[[:lower:]]))|(?=[[:lower:]]))|(?<=,)</p>\s*<p[^<>]*>
Replace with a space character, else it will join the end words.