What to do about the occasional “weird” word?

Ignore it.

Every Semantic Index will contain terms that while mathematically / statistically relevant, may be too difficult to use naturally. It's not necessary to use every term on a Semantic Index. If a term appears in the Semantic Index which isn't easy to integrate into your text, ignore it.

It should still be possible to score a B+ (and often an A+) while still ignoring any Index terms that are statistically relevant, but too contextually awkward to integrate.

Did this knowledgebase entry answer your question?