The thing is, those Mention rules are almost the same:

Mention helping_humanity_is_important: whether helping humanity is important.
Mention money_is_important: whether money is important.
Mention status_is_important: whether status is important.

Rather than treating all these separately, we can unify them all by having a slightly more complex representation of beliefs: when we want to talk about the belief that something is important, we’ll write that in the code as [important something], where something is the thing that’s supposed to be important.  Then we can write John and Richard’s beliefs as:

[predicate] [randomly]
Believes john god_exists.
Believes john [important money].
Believes richard god_exists.
Believes richard [important helping_people].

By bracketing together the word important with the thing that’s important, we tell the system that they come as a unit, and they’re one parameter.  Now we can collapse the different Mention rules down to just:

Mention [important ?thing]: whether ?thing is important.

Now, as we add more character beliefs about what's important, we don't have to keep adding new Mention rules.

These bracketed values are called tuples and they let us represent complex data relatively compactly.  The pattern matching process treats tuples largely the way it treats other data objects: if you match a tuple against a variable with no value, the variable is given the tuple as its value.  If you match a tuple against something that isn’t a tuple, that just fails, since they’re different.  The only thing that’s slightly complicated is that when you match tuples with one another: then they must have the same number of elements and those elements need to match.  So we can’t match [important money] to [important], but we can match [important money] to [important ?thing] by setting ?thing = money.