The structure can be defined as

Tree : Node*

Node : Term Tree
     | Term
[download]

It's not the ideal structure since a subtree needs to be distinguishable from a term and we need to look ahead, but it makes the tree more compact.

sub flatten {
   my ($tree) = @_;
   my @results;
   for (my $i=0; $i<@$tree; ) {
      my $has_subtree = ref($tree->[$i+1]);
      if ($has_subtree) {
         my $term    = $tree->[$i+0];
         my $subtree = $tree->[$i+1];
         push @results, map { [ $term, @$_ ] } @{ flatten($subtree) };
         $i += 2;
      } else {
         my $term = $tree->[$i+0];
         push @results, [ $term ];
         $i += 1;
      }
   }
   return \@results;
}

print("@$_\n") for @{ flatten($data_struct) };
[download]

Update: Small code reformat to resemble the tree structure description.