One more way.

#!/usr/bin/perl

use warnings;
use strict;

my %huge_trans = (
    zero  => 0,
    one   => 1,
    ones  => 11,
    two   => 2,
    three => 3,
#    ...
    million => 1_000_000,
    );

my $rx_word = qr/\b\w+\b/i;

print <> =~ s!($rx_word)! $huge_trans{lc $1} // $1 !ger;
[download]

Imagine you have huge dictionary for translation. Then this way should work faster.
Previously suggested way with alternations should also work fast, because Trie-optimization should kick-in.

Upd. BTW, it is not about "combining regex'es", it's only a use of a hash. All words, i.e. hash keys, must be simple words, not regexes.