You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

464 lines
17 KiB
Python

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

#### PATTERN | IT | INFLECT ########################################################################
# -*- coding: utf-8 -*-
# Copyright (c) 2013 University of Antwerp, Belgium
# Copyright (c) 2013 St. Lucas University College of Art & Design, Antwerp.
# Author: Tom De Smedt <tom@organisms.be>
# License: BSD (see LICENSE.txt for details).
####################################################################################################
# Regular expressions-based rules for Italian word inflection:
# - pluralization and singularization of nouns,
# - conjugation of verbs,
# - predicative adjectives.
# Accuracy:
# 92% for gender()
# 93% for pluralize()
# 84% for singularize()
# 82% for Verbs.find_lemma()
# 90% for Verbs.find_lexeme()
# 88% for predicative()
from __future__ import unicode_literals
from __future__ import division
from builtins import str, bytes, dict, int
from builtins import map, zip, filter
from builtins import object, range
import os
import sys
import re
try:
MODULE = os.path.dirname(os.path.realpath(__file__))
except:
MODULE = ""
sys.path.insert(0, os.path.join(MODULE, "..", "..", "..", ".."))
# Import Verbs base class and verb tenses.
from pattern.text import Verbs as _Verbs
from pattern.text import (
INFINITIVE, PRESENT, PAST, FUTURE, CONDITIONAL,
FIRST, SECOND, THIRD,
SINGULAR, PLURAL, SG, PL,
INDICATIVE, IMPERATIVE, SUBJUNCTIVE,
IMPERFECTIVE, PERFECTIVE, PROGRESSIVE,
IMPERFECT, PRETERITE,
PARTICIPLE, GERUND
)
sys.path.pop(0)
VERB, NOUN, ADJECTIVE, ADVERB = "VB", "NN", "JJ", "RB"
VOWELS = "aeiouy"
re_vowel = re.compile(r"a|e|i|o|u|y", re.I)
is_vowel = lambda ch: ch in VOWELS
#### ARTICLE #######################################################################################
# Inflection gender.
MASCULINE, FEMININE, NEUTER, PLURAL = \
MALE, FEMALE, NEUTRAL, PLURAL = \
M, F, N, PL = "m", "f", "n", "p"
# Word starts with z or s + consonant?
zs = lambda w: w and (w[:1] == "z" or (w[:1] == "s" and not is_vowel(w[1:2])))
def definite_article(word, gender=MALE):
""" Returns the definite article for a given word.
"""
if PLURAL in gender and MALE in gender and (is_vowel(word[:1]) or zs(word)):
return "gli"
if PLURAL not in gender and word and is_vowel(word[:1]):
return "l'"
if PLURAL not in gender and MALE in gender and zs(word):
return "lo"
if MALE in gender:
return PLURAL in gender and "i" or "il"
if FEMALE in gender:
return PLURAL in gender and "le" or "la"
return "il"
def indefinite_article(word, gender=MALE):
""" Returns the indefinite article for a given word.
"""
if MALE in gender and zs(word):
return PLURAL in gender and "degli" or "uno"
if MALE in gender:
return PLURAL in gender and "dei" or "un"
if FEMALE in gender and is_vowel(word[:1]):
return PLURAL in gender and "delle" or "un'"
if FEMALE in gender:
return PLURAL in gender and "delle" or "una"
return "un"
DEFINITE, INDEFINITE = \
"definite", "indefinite"
def article(word, function=INDEFINITE, gender=MALE):
""" Returns the indefinite or definite article for the given word.
"""
return function == DEFINITE \
and definite_article(word, gender) \
or indefinite_article(word, gender)
_article = article
def referenced(word, article=INDEFINITE, gender=MALE):
""" Returns a string with the article + the word.
"""
s = "%s&space;%s" % (_article(word, article, gender), word)
s = s.replace("'&space;", "'")
s = s.replace("&space;", " ")
return s
#### GENDER #########################################################################################
def gender(word):
""" Returns the gender for the given word, either:
MALE, FEMALE, (MALE, FEMALE), (MALE, PLURAL) or (FEMALE, PLURAL).
"""
w = word.lower()
# Adjectives ending in -e: cruciale, difficile, ...
if w.endswith(("ale", "ile", "ese", "nte")):
return (MALE, FEMALE)
# Most nouns ending in -a (-e) are feminine, -o (-i) masculine:
if w.endswith(("ore", "ista", "mma")):
return MALE
if w.endswith(("a", "", "", "ione", "rice")):
return FEMALE
if w.endswith(("e", "oni")):
return (FEMALE, PLURAL)
if w.endswith("i"):
return (MALE, PLURAL)
if w.endswith("o"):
return MALE
return MALE
#### PLURALIZE ######################################################################################
plural_co_chi = set((
"abbaco", "baco", "cuoco", "fungo", "rammarico", "strascio", "valico" # ...
))
plural_go_ghi = set((
"albergo", "catalogo", "chirurgo", "dialogo", "manico", "monologo", "stomaco" # ...
))
plural_irregular = {
"braccio": "braccia", # bracci (arms of a lamp or cross)
"budello": "budelli", # budella (intestines)
"camicia": "camicie",
"bue": "buoi",
"dio": "dei",
"dito": "dita",
"doccia": "docce",
"inizio": "inizi",
"labbro": "labbra", # labbri (borders)
"mano": "mani",
"negozio": "negozi",
"osso": "ossa", # ossi (dog bones)
"uomo": "uomini",
"uovo": "uova"
}
def pluralize(word, pos=NOUN, custom={}):
""" Returns the plural of a given word.
"""
if word in custom:
return custom[word]
w = word.lower()
if len(w) < 3:
return w
if w in plural_irregular:
return plural_irregular[w]
# provincia => province (but: socia => socie)
if w.endswith(("cia", "gia")) and len(w) > 4 and not is_vowel(w[-4]):
return w[:-2] + "e"
# amica => amiche
if w.endswith(("ca", "ga")):
return w[:-2] + "he"
# studentessa => studentesse
if w.endswith("a"):
return w[:-1] + "e"
# studente => studenti
if w.endswith("e"):
return w[:-1] + "i"
# viaggio => viaggi (but: leggìo => leggìi)
if w.endswith("io"):
return w[:-2] + "i"
# abbaco => abbachi
if w in plural_co_chi:
return w[:-2] + "chi"
# albergo => alberghi
if w in plural_co_chi:
return w[:-2] + "ghi"
# amico => amici
if w.endswith("o"):
return w[:-1] + "i"
return w
#### SINGULARIZE ###################################################################################
singular_majority_vote = [
("tenti", "tente"), ("anti", "ante"), ( "oni", "one" ), ( "nti", "nto" ),
( "ali", "ale" ), ( "ici", "ico" ), ( "nze", "nza" ), ( "ori", "ore" ),
( "che", "ca" ), ( "ati", "ato" ), ( "ari", "ario"), ( "tti", "tto" ),
( "eri", "ero" ), ( "chi", "co" ), ( "ani", "ano" ), ( "ure", "ura" ),
( "ità", "ità" ), ( "ivi", "ivo" ), ( "ini", "ino" ), ( "iti", "ito" ),
( "emi", "ema" ), ( "ili", "ile" ), ( "oli", "olo" ), ( "esi", "ese" ),
( "ate", "ata" ), ( "ssi", "sso" ), ( "rie", "ria" ), ( "ine", "ina" ),
( "lli", "llo" ), ( "ggi", "ggio"), ( "tri", "tro" ), ( "imi", "imo" )
]
singular_irregular = dict((v, k) for k, v in plural_irregular.items())
def singularize(word, pos=NOUN, custom={}):
""" Returns the singular of a given word.
"""
if word in custom:
return custom[word]
w = word.lower()
# il gatti => il gatto
if pos == "DT":
if w in ("i", "gli"):
return "il"
if w == "el":
return "la"
return w
if len(w) < 3:
return w
if w in singular_irregular:
return singular_irregular[w]
# Ruleset adds 16% accuracy.
for a, b in singular_majority_vote:
if w.endswith(a):
return w[:-len(a)] + b
# Probably an adjective ending in -e: cruciale, difficile, ...
if w.endswith(("ali", "ari", "ili", "esi", "nti")):
return w[:-1] + "e"
# realisti => realista
if w.endswith("isti"):
return w[:-1] + "a"
# amiche => amica
if w.endswith(("che", "ghe")):
return w[:-2] + "a"
# alberghi => albergo
if w.endswith(("chi", "ghi")):
return w[:-2] + "o"
# problemi => problema
if w.endswith("emi"):
return w[:-1] + "a"
# case => case
if w.endswith("e"):
return w[:-1] + "a"
# Ambigious: both -o and -a pluralize to -i.
if w.endswith("i"):
return w[:-1] + "o"
return w
#### VERB CONJUGATION ##############################################################################
# The verb table was trained on Wiktionary and contains the top 1,250 frequent verbs.
verb_majority_vote = [
("iresti", "ire" ), ("ireste", "ire" ), ("iremmo", "ire" ), ("irebbe", "ire" ),
("iranno", "ire" ), ( "ssero", "re" ), ( "ssimo", "re" ), ( "ivate", "ire" ),
( "ivamo", "ire" ), ( "irete", "ire" ), ( "iremo", "ire" ), ( "irono", "ire" ),
( "scano", "re" ), ( "hiamo", "are" ), ( "scono", "re" ), ( "hiate", "are" ),
( "vano", "re" ), ( "vate", "re" ), ( "vamo", "re" ), ( "simo", "e" ),
( "rono", "re" ), ( "isse", "ire" ), ( "isti", "ire" ), ( "tino", "tare"),
( "tato", "tare"), ( "irai", "ire" ), ( "tavo", "tare"), ( "tavi", "tare"),
( "tava", "tare"), ( "tate", "tare"), ( "iste", "ire" ), ( "irei", "ire" ),
( "immo", "ire" ), ( "rerò", "rare"), ( "rerà", "rare"), ( "iavo", "iare"),
( "iavi", "iare"), ( "iava", "iare"), ( "iato", "iare"), ( "iare", "iare"),
( "hino", "are" ), ( "ssi", "re" ), ( "sse", "re" ), ( "ndo", "re" ),
( "irò", "ire" ), ( "tai", "tare"), ( "ite", "ire" ), ( "irà", "ire" ),
( "sco", "re" ), ( "sca", "re" ), ( "iai", "iare"), ( "ii", "ire" ),
( "hi", "are" )
]
class Verbs(_Verbs):
def __init__(self):
_Verbs.__init__(self, os.path.join(MODULE, "it-verbs.txt"),
language = "it",
default = {},
format = [
0, 1, 2, 3, 4, 5, 6, 8, # indicativo presente
34, 35, 36, 37, 38, 39, 24, # indicativo passato remoto
17, 18, 19, 20, 21, 22, # indicativo imperfetto
40, 41, 42, 43, 44, 45, # indicativo futuro semplice
46, 47, 48, 49, 50, 51, # condizionale presente
52, 521, 53, 54, 541, # imperativo
55, 56, 57, 58, 59, 60, # congiuntivo presente
67, 68, 69, 70, 71, 72 # congiontive imperfetto
])
def find_lemma(self, verb):
""" Returns the base form of the given inflected verb, using a rule-based approach.
"""
v = verb.lower()
# Probably infinitive if ends in -are, -ere, -ire or reflexive -rsi.
if v.endswith(("are", "ere", "ire", "rsi")):
return v
# Ruleset adds 3% accuracy.
for a, b in verb_majority_vote:
if v.endswith(a):
return v[:-len(a)] + b
v = v.replace("cha", "ca")
v = v.replace("che", "ce")
v = v.replace("gha", "ga")
v = v.replace("ghe", "ge")
v = v.replace("ghi", "gi")
v = v.replace("gge", "ggie")
# Many verbs end in -ire and have a regular inflection:
for x in ((
"irò", "irai", "irà", "iremo", "irete", "iranno", # future
"irei", "iresti", "irebbe", "iremmo", "ireste", "irebbero", # conditional
"ascano", # subjunctive I
"issi", "isse", "issimo", "iste", "issero", # subjunctive II
"ivo", "ivi", "iva", "ivamo", "ivate", "ivano", # past imperfective
"isti", "immo", "iste", "irono", "ito", # past perfective
"isco", "isci", "isce", "ite", "iscono", "indo")): # present
if v.endswith(x):
return v[:-len(x)] + "ire"
# Many verbs end in -are and have a regular inflection:
for x in ((
"erò", "erai", "erà", "eremo", "erete", "eranno", # future
"erei", "eresti", "erebbe", "eremmo", "ereste", "erebbero", # conditional
"iamo", "iate", "ino", # subjunctive I
"assi", "asse", "assimo", "aste", "assero", # subjunctive II
"avo", "avi", "ava", "avamo", "avate", "avano", # past imperfective
"ai", "asti", "ò", "ammo", "aste", "arono", "ato", # past perfective
"iamo", "ate", "ano", "ando")): # present
if v.endswith(x):
return v[:-len(x)] + "are"
# Many verbs end in -ere and have a regular inflection:
for x in ((
"essi", "esse", "essimo", "este", "essero", # subjunctive II
"evo", "evi", "eva", "evamo", "evate", "evano", # past imperfective
"ei", "esti", "è", "emmo", "este", "erono", "eto", # past perfective
"ete", "ono", "endo")): # present
if v.endswith(x):
return v[:-len(x)] + "ere"
if v.endswith("à"):
return v[:-1] + "e"
if v.endswith("ì"):
return v[:-1] + "ire"
if v.endswith("e"):
return v[:-1] + "ere"
if v.endswith(("a", "i", "o")):
return v[:-1] + "are"
return v
def find_lexeme(self, verb):
""" For a regular verb (base form), returns the forms using a rule-based approach.
"""
v = verb.lower()
v = re.sub(r"rci$", "re", v)
v = re.sub(r"rsi$", "re", v)
v = re.sub(r"rre$", "re", v)
b = v[:-3]
if verb.endswith(("care", "gare")):
b += "h" # moltiplicare => tu moltiplichi
if verb.endswith(("ciare", "giare")):
b = b[:-1] # cominciare => tu cominci
if v.endswith("are"):
# -are = 1st conjugation
a1, a2, a3, a4, a5, a6, a7 = "a", "a", "ò", "a", "i", "e", "a"
elif v.endswith("ere"):
# -ere = 2nd conjugation
a1, a2, a3, a4, a5, a6, a7 = "e", "o", "è", "i", "a", "e", "e"
elif v.endswith("ire"):
# -ire = 3rd conjugation
a1, a2, a3, a4, a5, a6, a7 = "i", "o", "i", "i", "a", "i", "e"
else:
# -orre, -urre = use 2nd conjugation
a1, a2, a3, a4, a5, a6, a7 = "e", "o", "è", "i", "a", "e", "e"
if verb.lower().endswith("ire"):
# ire verbs can add -isc between the root and declination.
isc = "isc"
else:
isc = ""
v = [verb.lower(),
b + isc + "o", b + isc + "i", b + isc + a7, b + "iamo", b + a1 + "te", b + isc + a2 + "no", b + a1 + "ndo",
b + a1 + "i", b + a1 + "sti", b + a3, b + a1 + "mmo", b + a1 + "ste", b + a1 + "rono", b + a1 + "to",
b + a1 + "vo", b + a1 + "vi", b + a1 + "va", b + a1 + "vamo", b + a1 + "vate", b + a1 + "vano",
b + a6 + "", b + a6 + "rai", b + a6 + "", b + a6 + "remo", b + a6 + "rete", b + a6 + "ranno",
b + a6 + "rei", b + a6 + "resti", b + a6 + "rebbe", b + a6 + "remmo", b + a6 + "reste", b + a6 + "rebbero",
b + isc + a4, b + isc + a5, b + "iamo", b + a1 + "te", b + isc + a5 + "no",
b + isc + a5, b + isc + a5, b + isc + a5, b + "iamo", b + "iate", b + isc + a5 + "no",
b + a1 + "ssi", b + a1 + "ssi", b + a1 + "sse", b + a1 + "ssimo", b + a1 + "ste", b + a1 + "ssero"
]
for i, x in enumerate(v):
x = x.replace("ii" , "i")
x = x.replace("cha", "ca")
x = x.replace("gha", "ga")
x = x.replace("gga", "ggia")
x = x.replace("cho", "co")
x = x.replace("chò", "")
v[i] = x
return v
verbs = Verbs()
conjugate, lemma, lexeme, tenses = \
verbs.conjugate, verbs.lemma, verbs.lexeme, verbs.tenses
#### ATTRIBUTIVE & PREDICATIVE #####################################################################
adjective_predicative = {
"bei": "bello",
"bel": "bello",
"bell'": "bello",
"begli": "bello",
"buon": "buono",
"buon'": "buona",
"gran": "grande",
"grand'": "grande",
"grandi": "grande",
"san": "santo",
"sant'": "santa"
}
def attributive(adjective):
""" For a predicative adjective, returns the attributive form.
"""
# Must deal with feminine and plural.
raise NotImplementedError
def predicative(adjective):
""" Returns the predicative adjective.
"""
w = adjective.lower()
if w in adjective_predicative:
return adjective_predicative[w]
if w.endswith("ari"):
return w + "o"
if w.endswith(("ali", "ili", "esi", "nti", "ori")):
return w[:-1] + "e"
if w.endswith("isti"):
return w[:-1] + "a"
if w.endswith(("che", "ghe")):
return w[:-2] + "a"
if w.endswith(("chi", "ghi")):
return w[:-2] + "o"
if w.endswith("i"):
return w[:-1] + "o"
if w.endswith("e"):
return w[:-1] + "a"
return adjective