Skip to content

Translation and cache

Translate between Odia and other languages. English to Odia tries a bundled offline dictionary first and only then calls Google Translate. Results that come from Google are kept in a cache.

Examples that need the network have no output shown and are not run by the test suite, because Google can rate limit or change its answers. Everything else on this page is run by the suite.

Functions

Function Direction Default
other_lang_to_odia(text, source_language_code="en") any language to Odia English source
odia_to_other_lang(text, dest_language_code="en") Odia to any language English target
universal_translation(text, source_language_code="en", dest_language_code="or") any to any English to Odia

Language codes are the ones Google Translate uses, for example hi, de, or.

from openodia import other_lang_to_odia, odia_to_other_lang, universal_translation

other_lang_to_odia("शेयर बाज़ार", source_language_code="hi")
odia_to_other_lang("ନମସ୍କାର", dest_language_code="de")
universal_translation("Guten Morgen", source_language_code="de", dest_language_code="en")

The offline dictionary

When the source is "en", other_lang_to_odia looks the text up in an English to Odia dictionary that ships with the package. The lookup is case-insensitive and needs no network.

>>> from openodia import other_lang_to_odia
>>> other_lang_to_odia("good morning")
'ଶୁଭ ସକାଳ'
>>> other_lang_to_odia("Good Morning")
'ଶୁଭ ସକାଳ'
>>> from openodia.corpus.dictionary import get_dictionary
>>> dictionary = get_dictionary()
>>> len(dictionary) > 100_000
True
>>> dictionary["a bad apple"]
'ଏକ ଖରାପ ଆପଲ୍'

The dictionary is loaded once per process, so the first call pays for reading the JSON file and later calls reuse it.

Gotchas

  • The key is lowercased but not stripped. "good morning " with a trailing space misses the dictionary and goes to Google.
  • A non-English source never uses the dictionary, even for a word that is in it.
  • odia_to_other_lang and universal_translation always go to Google, never to the dictionary.

The translation cache

Every Google answer is stored under the key (text, source_code, destination_code) in an in-memory LRU cache of 10,000 entries. cache.stats() reports hits, misses and sizes, and the same call repeated later is served from memory.

You can seed the cache yourself, which is also how to see a hit without the network:

>>> from openodia import cache, odia_to_other_lang
>>> cache.clear()
>>> cache.get_cache().set(("ନମସ୍କାର", "or", "en"), "Hello")
>>> odia_to_other_lang("ନମସ୍କାର")
'Hello'
>>> cache.stats()
{'hits': 1, 'misses': 0, 'size': 1, 'max_size': 10000, 'disk_size': 0}

cache.get_cache().get(key) returns None on a miss and counts it:

>>> cache.get_cache().get(("ଅଜଣା", "or", "en")) is None
True
>>> cache.stats()["misses"]
1

Size and eviction

cache.configure(max_size=n) sets the capacity. The least recently used entry is evicted first, and reading an entry makes it the most recent.

>>> _ = cache.configure(max_size=2)
>>> c = cache.get_cache()
>>> c.set("a", "1"); c.set("b", "2"); c.set("c", "3")
>>> c.get("a") is None, c.get("c")
(True, '3')
>>> cache.stats()["size"]
2
>>> cache.configure(max_size=0)
Traceback (most recent call last):
  ...
ValueError: max_size must be >= 1, got 0

Gotcha: configure starts a new cache

configure installs a fresh cache. Entries and hit counters from the old one are gone, and disk_path is dropped unless you pass it again.

Persist across runs

Install the extra and give configure a directory:

pip install "openodia[cache]"

from openodia import cache

cache.configure(max_size=50_000, disk_path="~/.cache/openodia/translate")

Reads check memory first, then disk. You can also set the defaults before the process starts with the environment variables OPENODIA_CACHE_MAX_SIZE and OPENODIA_CACHE_DISK. An invalid size raises ValueError the first time the cache is used.

Next