h/plugins/translate.py

189 lines
4.9 KiB
Python

import htmlentitydefs
import re
from util import hook, http
########### from http://effbot.org/zone/re-sub.htm#unescape-html #############
def unescape(text):
def fixup(m):
text = m.group(0)
if text[:2] == "&#":
# character reference
try:
if text[:3] == "&#x":
return unichr(int(text[3:-1], 16))
else:
return unichr(int(text[2:-1]))
except ValueError:
pass
else:
# named entity
try:
text = unichr(htmlentitydefs.name2codepoint[text[1:-1]])
except KeyError:
pass
return text # leave as is
return re.sub("&#?\w+;", fixup, text)
##############################################################################
def goog_trans(text, slang, tlang):
url = 'http://ajax.googleapis.com/ajax/services/language/translate?v=1.0'
parsed = http.get_json(url, q=text, langpair=(slang + '|' + tlang))
if not 200 <= parsed['responseStatus'] < 300:
raise IOError('error with the translation server: %d: %s' % (
parsed['responseStatus'], parsed['responseDetails']))
if not slang:
return unescape('(%(detectedSourceLanguage)s) %(translatedText)s' %
(parsed['responseData']))
return unescape(parsed['responseData']['translatedText'])
def match_language(fragment):
fragment = fragment.lower()
for short, _ in lang_pairs:
if fragment in short.lower().split():
return short.split()[0]
for short, full in lang_pairs:
if fragment in full.lower():
return short.split()[0]
return None
@hook.command
def translate(inp):
'.translate [source language [target language]] <sentence> -- translates' \
' <sentence> from source language (default autodetect) to target' \
' language (default English) using Google Translate'
args = inp.split(' ', 2)
try:
if len(args) >= 2:
sl = match_language(args[0])
if not sl:
return goog_trans(inp, '', 'en')
if len(args) >= 3:
tl = match_language(args[1])
if not tl:
if sl == 'en':
return 'unable to determine desired target language'
return goog_trans(args[1] + ' ' + args[2], sl, 'en')
return goog_trans(args[2], sl, tl)
return goog_trans(inp, '', 'en')
except IOError, e:
return e
languages = 'ja fr de ko ru zh'.split()
language_pairs = zip(languages[:-1], languages[1:])
def babel_gen(inp):
for language in languages:
inp = inp.encode('utf8')
trans = goog_trans(inp, 'en', language).encode('utf8')
inp = goog_trans(trans, language, 'en')
yield language, trans, inp
@hook.command
def babel(inp):
".babel <sentence> -- translates <sentence> through multiple languages"
try:
return list(babel_gen(inp))[-1][2]
except IOError, e:
return e
@hook.command
def babelext(inp):
".babelext <sentence> -- like .babel, but with more detailed output"
try:
babels = list(babel_gen(inp))
except IOError, e:
return e
out = u''
for lang, trans, text in babels:
out += '%s:"%s", ' % (lang, text.decode('utf8'))
out += 'en:"' + babels[-1][2].decode('utf8') + '"'
if len(out) > 300:
out = out[:150] + ' ... ' + out[-150:]
return out
lang_pairs = [
("no", "Norwegian"),
("it", "Italian"),
("ht", "Haitian Creole"),
("af", "Afrikaans"),
("sq", "Albanian"),
("ar", "Arabic"),
("hy", "Armenian"),
("az", "Azerbaijani"),
("eu", "Basque"),
("be", "Belarusian"),
("bg", "Bulgarian"),
("ca", "Catalan"),
("zh-CN zh", "Chinese"),
("hr", "Croatian"),
("cs", "Czech"),
("da", "Danish"),
("nl", "Dutch"),
("en", "English"),
("et", "Estonian"),
("tl", "Filipino"),
("fi", "Finnish"),
("fr", "French"),
("gl", "Galician"),
("ka", "Georgian"),
("de", "German"),
("el", "Greek"),
("ht", "Haitian Creole"),
("iw", "Hebrew"),
("hi", "Hindi"),
("hu", "Hungarian"),
("is", "Icelandic"),
("id", "Indonesian"),
("ga", "Irish"),
("it", "Italian"),
("ja jp jpn", "Japanese"),
("ko", "Korean"),
("lv", "Latvian"),
("lt", "Lithuanian"),
("mk", "Macedonian"),
("ms", "Malay"),
("mt", "Maltese"),
("no", "Norwegian"),
("fa", "Persian"),
("pl", "Polish"),
("pt", "Portuguese"),
("ro", "Romanian"),
("ru", "Russian"),
("sr", "Serbian"),
("sk", "Slovak"),
("sl", "Slovenian"),
("es", "Spanish"),
("sw", "Swahili"),
("sv", "Swedish"),
("th", "Thai"),
("tr", "Turkish"),
("uk", "Ukrainian"),
("ur", "Urdu"),
("vi", "Vietnamese"),
("cy", "Welsh"),
("yi", "Yiddish")
]