Best practise for storing multilingual strings

First make sure that the database locale can deal with different languages. Use an UTF-8 server-encoding. Optionally set LC_COLLATE = 'C' to be on neutral ground or use a collation for your first language to have a default sort order. Start by reading the chapter Collation Support in the manual.

I would strongly suggest that you use the latest version of PostgreSQL (9.1 at time of writing) because it has superior collation support.

As for the table structure: keep it simple. It sounds like there is a low, fixed number of languages to deal with. You could just have a column for each language then:

CREATE TABLE txt (  txt_id serial PRIMARY KEY ,txt    text NOT NULL -- master language NOT NULL? ,txt_fr text -- others can be NULL? ,txt_es text ,txt_de text);

This is pretty effective, even with many languages. NULL storage is very cheap.
If you have a varying number of languages to deal with, a separate table might be the better solution. This solution assumes that you have a "master language", where the string is always present:

CREATE TABLE txt (  txt_id serial PRIMARY KEY ,txt    text NOT NULL -- master language NOT NULL?);CREATE TABLE lang (  lang_abbr text PRIMARY KEY -- de, es, fr, ... ,lang      text NOT NULL ,note      text);

Or, if a (two-letter) abbreviation is enough, just create a enum type to identify the language.

CREATE TABLE txt_trans (  txt_id    int REFERENCES txt(txt_id) ON UPDATE CASCADE ON DELETE CASCADE ,lang_abbr text REFERENCES lang(lang_abbr) ON UPDATE CASCADE ,txt       text NOT NULL -- master language NOT NULL? ,CONSTRAINT txt_trans_pkey PRIMARY KEY (txt_id, lang_abbr));

Not treating the master language special and keeping all language variants in the same table might make handling in your app simpler. But it really depends on your requirements.