PostgreSQL string character replacement

I'm trying to write a lexical database for storing words comprised of roots and patterns, and I was wondering how I could create a column that will combine the root and pattern for me, while ignor开发者_Python百科ing rows which don't have both columns of the SELECT query populated.

Basically, I have this output from a PostgreSQL DB:

SELECT root, root_i FROM tbl_roots NATURAL JOIN tbl_patterns NATURAL JOIN tbl_patterns_triliteral;

  root   | root_i
---------+--------
 {s,ş,m} | 1u2u3a
 {p,l,t} | 1u2u3a
 {t,m,s} | 1u2u3a
 {n,t,l} | 1u2u3a
 {s,ş,m} | 1a2oi3
 {p,l,t} | 1a2oi3
 {t,m,s} | 1a2oi3
 {n,t,l} | 1a2oi3
 {s,ş,m} | 1o2i3
 {p,l,t} | 1o2i3
 {t,m,s} | 1o2i3
 {n,t,l} | 1o2i3
 {s,ş,m} | a12e3
 {p,l,t} | a12e3
 {t,m,s} | a12e3
 {n,t,l} | a12e3
 {s,ş,m} | 1u2á3
 {p,l,t} | 1u2á3
 {t,m,s} | 1u2á3
 {n,t,l} | 1u2á3
 {s,ş,m} |
 {p,l,t} |
 {t,m,s} |
 {n,t,l} |
 {s,ş,m} | 1e2é3
 {p,l,t} | 1e2é3
 {t,m,s} | 1e2é3
 {n,t,l} | 1e2é3
 {s,ş,m} |
 {p,l,t} |
 {t,m,s} |
 {n,t,l} |
 {s,ş,m} |
 {p,l,t} |
 {t,m,s} |
 {n,t,l} |
 {s,ş,m} |
 {p,l,t} |
 {t,m,s} |
 {n,t,l} |

And I want to convert it on-the-fly into something resembling this:

  root   | root_i | word_i
---------+--------+--------
 {s,ş,m} | 1u2u3a | suşuma
 {p,l,t} | 1u2u3a | puluta
 {t,m,s} | 1u2u3a | tumusa
 {n,t,l} | 1u2u3a | nutula
 {s,ş,m} | 1a2oi3 | saşoim
 {p,l,t} | 1a2oi3 | paloit
 {t,m,s} | 1a2oi3 | tamois
 {n,t,l} | 1a2oi3 | natoil
 {s,ş,m} | 1o2i3  | soşim
 {p,l,t} | 1o2i3  | polit
 {t,m,s} | 1o2i3  | tomis
 {n,t,l} | 1o2i3  | notil
 {s,ş,m} | a12e3  | asşem
 {p,l,t} | a12e3  | aplet
 {t,m,s} | a12e3  | atmes
 {n,t,l} | a12e3  | antel
 {s,ş,m} | 1u2á3  | suşám
 {p,l,t} | 1u2á3  | pulát
 {t,m,s} | 1u2á3  | tumás
 {n,t,l} | 1u2á3  | nutál
 {s,ş,m} | 1e2é3  | seşém
 {p,l,t} | 1e2é3  | pelét
 {t,m,s} | 1e2é3  | temés
 {n,t,l} | 1e2é3  | neşél

Where the word column is dynamically generated by replacing the digits in the root_i column with the character in that digit's index in the root column. I also need to drop queried rows that don't have an entry in both columns to reduce clutter in my output.

Can anyone help me devise a postgres function that will do the merging of character[] and text strings? The little bit of regex I need shouldn't be complicated, but I have no idea how to mix this with a query, or better yet, turn it into a function.

select
  root,
  root_i,
  translate(root_i, "123", array_to_string(root,'')) as word_i
NATURAL JOIN tbl_patterns
NATURAL JOIN tbl_patterns_triliteral
where root is not null and root_i is not null;

I must admit to not liking to do much string manipulation in sql/plpgsql functions. Perl has an operator for replacing regexp matches with generated replacements, which works fairly nicely:

create or replace function splice_to_word(root text, root_i text)
  returns text strict immutable language plperl as $$
  my $roots = shift;
  my $template = shift;
  $template =~ s{(\d+)}{substr($roots,$1-1,1)}ge;
  return $template;
$$;

There is some nastiness in that postgresql arrays don't seem to be translated into Perl lists, so I've assumed the roots are passed in as a string, e.g.:

select root, root_i, splice_to_word(array_to_string(root, ''), root_i) from data