Lucky me, that is more correct than the "first code". If you
replace $key with $word, youŽll see the reason why:

Every entry in the hash is a potentional prefix (or suffix
for that matter) of $word. Thus yes, the blazingly fast(tm)
code does the right thing.(tm)

I encountered a second problem, that is, that simple prefix
suffix consideration is far away from being enough (no worry
- it alway was meant as first approximation). But there are
in various languages all kinds of morphemes, suffixes, prefixes,
infixes, reduplication, elipses etc. In fact, if you look
at a given word of n chars, the FULL segmentation would
require you to produce 2^(n-1) alternatives.

Try that with a 39char word (Eisenbahnerwohnungsnutzungsverordnungen)

So I thought we reduce it to already present entries of a lexicon.

Ciao