Create full text search configuration with two dictionaries - postgresql

Question

I want to perform a full text search on a postgresql column using the english_stem dictionary and the simple dictionary. I can do something like this:
ALTER TEXT SEARCH CONFIGURATION english_simple_conf
ALTER MAPPING FOR asciiword, asciihword, hword_asciipart, word, hword, hword_part
WITH english_stem, simple;
But this checks that the word is in both dictionaries. Is there a way to alter this configuration so the word can be matched with one dictionary OR the other?
Edit:
The reason I think they are not being checked in order is because when searching for a partial word that should be found in the simple dictionary, nothing is returned.
select * from ts_debug('english', 'gutter cleaning services');
alias | description | token | dictionaries | dictionary | lexemes
-----------+-----------------+----------+----------------+--------------+----------
asciiword | Word, all ASCII | gutter | {english_stem} | english_stem | {gutter}
blank | Space symbols | | {} | |
asciiword | Word, all ASCII | cleaning | {english_stem} | english_stem | {clean}
blank | Space symbols | | {} | |
asciiword | Word, all ASCII | services | {english_stem} | english_stem | {servic}
select * from ts_debug('simple', 'gutter cleaning services');
alias | description | token | dictionaries | dictionary | lexemes
-----------+-----------------+----------+--------------+------------+------------
asciiword | Word, all ASCII | gutter | {simple} | simple | {gutter}
blank | Space symbols | | {} | |
asciiword | Word, all ASCII | cleaning | {simple} | simple | {cleaning}
blank | Space symbols | | {} | |
asciiword | Word, all ASCII | services | {simple} | simple | {services}
select name from categories where (to_tsvector('english_simple_conf', name) ## (to_tsquery('english_simple_conf', 'cleani:*')));
name
------
(0 rows)
But searching for a partial in the english dictionary returns as expected.
select name from categories where (to_tsvector('english_simple_conf', name) ## (to_tsquery('english_simple_conf', 'clea:*')));
name
--------------------------
Gutter Cleaning Services

Answer 1

But this checks that the word is in both dictionaries.
That's not correct. As noted in the docs (see the description for the dictionary_name parameter), it checks them in order; it only checks the 2nd dictionary if it did not get a token from the first. You can verify this with ts_debug().
testdb=# ALTER TEXT SEARCH CONFIGURATION english_simple_conf
ALTER MAPPING FOR asciiword, asciihword, hword_asciipart, word, hword, hword_part
WITH simple;
ALTER TEXT SEARCH CONFIGURATION
testdb=# select * from ts_debug('public.english_simple_conf', 'cars boats n0taword');
alias | description | token | dictionaries | dictionary | lexemes
-----------+--------------------------+----------+--------------+------------+------------
asciiword | Word, all ASCII | cars | {simple} | simple | {cars}
blank | Space symbols | | {} | |
asciiword | Word, all ASCII | boats | {simple} | simple | {boats}
blank | Space symbols | | {} | |
numword | Word, letters and digits | n0taword | {simple} | simple | {n0taword}
(5 rows)
testdb=# ALTER TEXT SEARCH CONFIGURATION english_simple_conf
ALTER MAPPING FOR asciiword, asciihword, hword_asciipart, word, hword, hword_part
WITH english_stem, simple;
ALTER TEXT SEARCH CONFIGURATION
testdb=# select * from ts_debug('public.english_simple_conf', 'cars boats n0taword');
alias | description | token | dictionaries | dictionary | lexemes
-----------+--------------------------+----------+-----------------------+--------------+------------
asciiword | Word, all ASCII | cars | {english_stem,simple} | english_stem | {car}
blank | Space symbols | | {} | |
asciiword | Word, all ASCII | boats | {english_stem,simple} | english_stem | {boat}
blank | Space symbols | | {} | |
numword | Word, letters and digits | n0taword | {simple} | simple | {n0taword}
(5 rows)
The reason for the difference in the last two queries is that english_stem stems 'Cleaning' to 'clean', so searching for 'cleani*' will not match. Try adding the to_tsvector and to_tsquery expressions as a column and removing them from the WHERE; you'll see that "Gutter Cleaning Services" is stemmed to 'clean':2 'gutter':1 'servic':3.
testdb=# select to_tsvector('english_simple_conf', name), to_tsquery('english_simple_conf', 'cleani:*'), name from categories;
to_tsvector | to_tsquery | name
---------------------------------+------------+--------------------------
'clean':2 'gutter':1 'servic':3 | 'cleani':* | Gutter Cleaning Services
(1 row)
testdb=# select to_tsvector('english_simple_conf', name), to_tsquery('english_simple_conf', 'cleaning:*'), name from categories;
to_tsvector | to_tsquery | name
---------------------------------+------------+--------------------------
'clean':2 'gutter':1 'servic':3 | 'clean':* | Gutter Cleaning Services
(1 row)
If you change the ts_query to instead search for cleaning:*, that will get stemmed as well and again match. But, english_stem cannot figure out that 'cleani' is meant to stem to 'clean' unless it also sees the 'ng'. So, that falls through to simple, which performs no stemming, and you end up with the mismatch - still a trailing i in the tsquery, but not in the tsvector.
Stemming isn't meant to work on arbitrary prefixes of words, only on whole ones; for prefix matching, you'd use a traditional left-anchored LIKE.