Greetings,
Here is how I understand it.
in your code

my %res = map {
    $_ => do {
        my $re;
        $re .= $translations{$_} for split //;
        $re;
    } if !$res{$_};
  } @words;
[download]

you are refering to the list being created, in this case the %res hash, from within the block doing the assignment, the  map {$_ => do {#code} if !$res{$_};}@words; line. If you take out the  if !$res{$_}; part your resulting %res hash will have the value of the last of any possible identical values in the @words array. My gut feeling is that its more the !$res{$_} part than it is the use of the %res hash on both sides... but then there is your second example of

  my @b = (7,5,5,6);
  my @a;
  @a = map { $_ > 5 ? $_ : $a[0] } @b;
[download]

Which again makes me think it's the use of %res on both sides of the = sign. So a basic rule of thumb to use might be that map only works on the contents of @b and never questions @a, only gives (given your example above).
...that made more sense when I started.
Anyone else with a more concise explanation?

-InjunJoel

Update!
Perhaps its the getkey() that is making it unclear? Have a look at perldoc question (how to get info on getkey). The getkey() call was tripping me up in the map documentation when I read it. So assuming that getkey() is only working on $_ from @array and not anything related to %hash^{(for it may only giveth)} it all makes sense...