SQL / (Django) : efficient database schema for translations - sql

Question

Situation
I trying to set up a database schema to store translations, between different languages. So far it looks like this (simplyfied):
class Language(models.Model):
tag = models.CharField(max_length=2)
def __unicode__(self):
return self.tag
class Phrase(models.Model):
name = models.TextField()
language = models.ForeignKey(Language)
def __unicode__(self):
return self.name
class Meta:
unique_together = ("name", "language")
index_together = [
["name", "language"]
]
class Translation(models.Model):
phrase1 = models.ForeignKey(Phrase, related_name="translation_as_1")
phrase2 = models.ForeignKey(Phrase, related_name="translation_as_2")
def __unicode__(self):
return self.phrase1.name + " <=> " + self.phrase2.name
class Meta:
unique_together = ("phrase1", "phrase2")
index_together = [
["phrase1", "phrase2"]
]
This database schema seems logical to me. I store phrases in different languages and then have translations that contain exactly two phrases.
Problem
The problem is, that the queries, that result out of this schema, look kind of nasty. For instance:
from django.db.models import Q
name = "my phrase"
translations = Translation.objects.filter(Q(phrase1__name=text)|Q(phrase2__name=text))
translated_names = []
for translation in translations:
name1 = translation.phrase1.name
name2 = translation.phrase2.name
if name1 == name:
translated_names.append(name2)
else:
translated_names.append(name1)
I always have to include the "OR" relationship, to make sure, that I get all the possible translations, since the phrase could be stored as phrase1 or phrase2. On top of that, I have to filter my result afterwards to get the correct translated_name (for loop).
Further Explaination
Before I switched to the described schema, I had the following schema instead (Phrase and Language are the same as before):
class Translation(models.Model):
phrase = models.ForeignKey(Phrase)
name = models.TextField()
def __unicode__(self):
return self.phrase.name + " => " + self.name
class Meta:
unique_together = ("phrase", "name")
index_together = [
["phrase", "name"]
This schema let me make queries like this:
from django.db.models import Q
name = "my phrase"
translations = Translation.objects.filter(phrase__name=text)
translated_names = [t.name for t in translations]
This looks much nicer, and is of course faster. But this schema had the disadvantage, that it presents translations only in one direction, so I moved to the other one, which isn't quite what I want as well, because too slow and too complicated queries.
Question
So is there a good schema for this kind of problem, that I maybe overlook?
Remark
I'm not only interested in Django related answers. A pure SQL schema for this kind of problem would also be interesting for me.

Answer 1

This is the way that I have done it in the past. Adapt it for your naming convention.
Suppose that I had a table with a name and other columns in it like this
Table TR_CLT_clothing_type
clt_id | clt_name | other columns ....
--------------------------------------
1 | T Shirt ...
2 | Pants ...
Now if I decided that it needs translations, first I make a languages table
Table TR_LNG_language
lng_id | lng_name | lng_display
-------------------------------
1 | English | English (NZ)
2 | German | Deutsch
I also need to store the current language in the database (you will see why soon). It will only have one row
Table TA_INF_info
inf_current_lng
---------------
1
Then I drop the clt_name column from my clothing table TR_CLT_clothing_type. Instead I make relation table.
Table TL_CLT_clothing_type
clt_id | lng_id | clt_name
--------------------------
1 | 1 | T Shirt
1 | 2 | (German for T-Shirt)
2 | 1 | Pants
2 | 2 | keuchen (thank you google translate)
Now to get the name, you want to make a stored procedure for it. I have not attempted this in ORM.
CREATE PROCEDURE PS_CLT
#clt_id int
AS
SELECT lng.clt_name, clt.*
FROM TR_CLT_clothing_type clt
JOIN TL_CLT_clothing_type lng
ON lng.clt_id = clt.clt_id
WHERE clt.clt_id = #clt_id AND
lng.lng_id in (SELECT inf_current_lng FROM TA_INF_info)
This stored proc will return the name in the current language and all other columns for a specified language. To set the language, set the clt_current_lng in the TA_INF_info table.
Disclaimer: I don't have anything to check the syntax of what I have typed but it should hopefully be straightforward.
-- EDIT
There was a concern to be able to do "give me all translations for word X in language Y to language Z"
There is a "not so elegant" way to do this with the schema. You can do something like
for each table in database like "TL_%"
SELECT name
FROM table
WHERE id IN ( SELECT id
FROM table
WHERE name = #name
AND lng_id = german
)
AND lng_id = english
Now I would imagine that this would require some auto-generated SQL code but I could pull it off.
I have no idea how you would do this in ORM