gistfile1.txt

grammar WordPairs {
    rule word-pair { <key=word> <value=word> }
    token word { \S+ }
    rule TOP { <word-pair>* }
}

my %dict;

class WordPairsActions {
    method word-pair($/) { %dict{~$<key>}.push(~$<value>) }
}

my $match = WordPairs.parse(slurp, :actions(WordPairsActions));
say ?$match;

say "The pairs count of the key word \"her\" in wordpairs.txt is {%dict{"her"}.elems}";

One note -- I think you might be able to get better performance feeding in the original text file, instead of the paired version? But I admit I'm not at all certain of that, just suspicious. (ie an entirely new script which takes a complete text file and maps each word to the next word, so the end hash is the same.)