Query for transitive word translation - sql

Question

I have trouble constructing what I guess should be a CTE query, which would return a word translation.
I have two tables - WORDS:
ID | WORD | LANG
1 'Cat' 'ENG'
2 'Kot' 'POL'
3 'Katze' 'GER'
and CONNECTIONS:
ID | WORD_A | WORD_B
1 1 2
2 1 3
As ENG->POL, and ENG->GER translations already exist, I would like to receive a POL->GER translation.
So for any given word, I have to check it's connected translations, if target language is not there, I should check connected translations of those connected translations etc., up to the point where either translation is found, or nothing is returned;
I have no idea where to start, it wouldn't be a problem if there was a constant number of transitions required, but it might as well require transition like POL->ITA->...->ENG->GER.

Answer 1

This is a complicated graph walking problem. The idea is to set up a connections2 relationship that has words and languages in both directions, and then use this for walking the graph.
To get all German translations for Polish words, you can use:
with words AS (
select 1 as id, 'Cat' as word, 'ENG' as lang union all
select 2, 'Kot', 'POL' union all
select 3, 'Katze', 'GER'
),
connections as (
select 1 as id, 1 as word_a, 2 as word_b union all
select 2 as id, 1 as word_a, 3 as word_b
),
connections2 as (
select c.word_a as id_a, c.word_b as id_b, wa.lang as lang_a, wb.lang as lang_b
from connections c join
words wa
on c.word_a = wa.id join
words wb
on c.word_b = wb.id
union -- remove duplicates
select c.word_b, c.word_a, wb.lang as lang_a, wa.lang as lang_b
from connections c join
words wa
on c.word_a = wa.id join
words wb
on c.word_b = wb.id
),
cte as (
select id as pol_word, id as other_word, 'POL' as other_lang, 1 as lev, ',POL,' as langs
from words
where lang = 'POL'
union all
select cte.pol_word, c2.id_b, c2.lang_b, lev + 1, langs || c2.lang_b || ','
from cte join
connections2 c2
on cte.other_lang = c2.lang_a
where langs not like '%,' || c2.lang_b || ',%' or c2.lang_b = 'GER'
)
select *
from cte
where cte.other_lang = 'GER';
Here is a db<>fiddle.
Presumably, you want the shortest path. For that, you would use this query (after the CTEs):
select *
from cte
where cte.other_lang = 'GER' and
cte.lev = (select min(cte2.lev) from cte cte2 where ct2.pol_word = cte.pol_word);

Answer 2

I recommend changing your table design to something along these lines:
ID | WORD_ID | WORD | LANG
1 | 1 | 'Cat' | 'ENG'
2 | 1 | 'Kot' | 'POL'
3 | 1 | 'Katze' | 'GER'
Now suppose that you want to go from Polish to German, and you are starting off with the Polish word for cat, Kot. Then, you may use the following query:
SELECT t2.WORD
FROM yourTable t1
INNER JOIN yourTable t2
ON t1.WORD_ID = t1.WORD_ID AND t2.LANG = 'GER'
WHERE
t1.WORD = 'Kot' AND t1.LANG = 'POL'
Demo
The idea here is that you want one of your tables to maintain a mapping of every logically distinct word to some common id or reference. Then, you only need an incoming word and language, and a desired output language, to be able to do the mapping.
Note: For simplicity I am only showing a table with a single word across a handful of language, but the logic should still work with multiple words and languages. Your current table may not be normalized, but optimizing that might be another task separate from your immediate question.