the practice (possibly apocryphal, but I wouldn't be surprised were it true) of dictionary editors inventing non-existant words
Gabriel, google "esquivalience" for at least one non-apocryphal, intentional example of this, and "dord" for an accidental instance.
I've asked a US English dictionary editor about Dot's question, and she replied:
"[L]emmata are not copyrightable, unless they do not exist other places (like the copyright trap with esquivalience). As long as they can show that they know of the words' existence in another text, I'd say they were fine."
Another existing example that his would not be a problem, is the case of Scrabble wordlists. Merriam-Webster's Official Scrabble Players Dictionary is itself a compilation of the wordlists of 15 US college dictionaries, and it is surely protected by all sort of nasty copyright. But on the internet there are readily available lists of legal words from all Scrabble jurisdictions.
Your use would be quite transformative, educational, and a factual statement in the vein of a concordance or index.
Or do I really need to talk to a lawyer?
It wouldn't hurt, as long as you don't take "no" for an answer. Cfr. http://chillingeffects.org/
O.