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.
276 lines
8.8 KiB
Python
276 lines
8.8 KiB
Python
#### PATTERN | NL ##################################################################################
|
|
# Copyright (c) 2010 University of Antwerp, Belgium
|
|
# Author: Tom De Smedt <tom@organisms.be>
|
|
# License: BSD (see LICENSE.txt for details).
|
|
# http://www.clips.ua.ac.be/pages/pattern
|
|
|
|
####################################################################################################
|
|
# Dutch linguistical tools using fast regular expressions.
|
|
|
|
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 parser base classes.
|
|
from pattern.text import (
|
|
Lexicon, Model, Morphology, Context, Parser as _Parser, ngrams, pprint, commandline,
|
|
PUNCTUATION
|
|
)
|
|
# Import parser universal tagset.
|
|
from pattern.text import (
|
|
penntreebank2universal,
|
|
PTB, PENN, UNIVERSAL,
|
|
NOUN, VERB, ADJ, ADV, PRON, DET, PREP, ADP, NUM, CONJ, INTJ, PRT, PUNC, X
|
|
)
|
|
# Import parse tree base classes.
|
|
from pattern.text.tree import (
|
|
Tree, Text, Sentence, Slice, Chunk, PNPChunk, Chink, Word, table,
|
|
SLASH, WORD, POS, CHUNK, PNP, REL, ANCHOR, LEMMA, AND, OR
|
|
)
|
|
# Import sentiment analysis base classes.
|
|
from pattern.text import (
|
|
Sentiment as _Sentiment,
|
|
NOUN, VERB, ADJECTIVE, ADVERB,
|
|
MOOD, IRONY
|
|
)
|
|
# Import spelling base class.
|
|
from pattern.text import (
|
|
Spelling
|
|
)
|
|
# Import verb tenses.
|
|
from pattern.text import (
|
|
INFINITIVE, PRESENT, PAST, FUTURE,
|
|
FIRST, SECOND, THIRD,
|
|
SINGULAR, PLURAL, SG, PL,
|
|
PROGRESSIVE,
|
|
PARTICIPLE
|
|
)
|
|
# Import inflection functions.
|
|
from pattern.text.nl.inflect import (
|
|
pluralize, singularize, NOUN, VERB, ADJECTIVE,
|
|
verbs, conjugate, lemma, lexeme, tenses,
|
|
predicative, attributive
|
|
)
|
|
# Import all submodules.
|
|
from pattern.text.nl import inflect
|
|
|
|
sys.path.pop(0)
|
|
|
|
#--- DUTCH PARSER ----------------------------------------------------------------------------------
|
|
# The Dutch parser (accuracy 92%) is based on Jeroen Geertzen's language model:
|
|
# Brill-NL, http://cosmion.net/jeroen/software/brill_pos/
|
|
|
|
# The lexicon uses the WOTAN tagset:
|
|
# http://lands.let.ru.nl/literature/hvh.1999.2.ps
|
|
WOTAN = "wotan"
|
|
wotan = {
|
|
"Adj(": (("vergr", "JJR"), ("overtr", "JJS"), ("", "JJ")),
|
|
"Adv(": (("deel", "RP"), ("", "RB")),
|
|
"Art(": (("", "DT"),),
|
|
"Conj(": (("", "CC"),),
|
|
"Int": (("", "UH"),),
|
|
"Misc": (("symb", "SYM"), ("vreemd", "FW")),
|
|
"N(": (("eigen,ev", "NNP"), ("eigen,mv", "NNPS"), ("ev", "NN"), ("mv", "NNS")),
|
|
"Num(": (("", "CD"),),
|
|
"Prep(": (("inf", "TO"), ("", "IN")),
|
|
"Pron(": (("bez", "PRP$"), ("", "PRP")),
|
|
"Punc(": (("komma", ","), ("open", "("), ("sluit", ")"), ("schuin", "CC"), ("", ".")),
|
|
"V(": (("hulp", "MD"), ("ott,3", "VBZ"), ("ott", "VBP"), ("ovt", "VBD"),
|
|
("verl", "VBN"), ("teg", "VBG"), ("", "VB"))
|
|
}
|
|
|
|
|
|
def wotan2penntreebank(token, tag):
|
|
""" Converts a WOTAN tag to a Penn Treebank II tag.
|
|
For example: bokkenrijders/N(soort,mv,neut) => bokkenrijders/NNS
|
|
"""
|
|
for k, v in wotan.items():
|
|
if tag.startswith(k):
|
|
for a, b in v:
|
|
if a in tag:
|
|
return (token, b)
|
|
return (token, tag)
|
|
|
|
|
|
def wotan2universal(token, tag):
|
|
""" Converts a WOTAN tag to a universal tag.
|
|
For example: bokkenrijders/N(soort,mv,neut) => bokkenrijders/NOUN
|
|
"""
|
|
if tag.startswith("Adv"):
|
|
return (token, ADV)
|
|
return penntreebank2universal(*wotan2penntreebank(token, tag))
|
|
|
|
ABBREVIATIONS = set((
|
|
"a.d.h.v.", "afb.", "a.u.b.", "bv.", "b.v.", "bijv.", "blz.", "ca.", "cfr.", "dhr.", "dr.",
|
|
"d.m.v.", "d.w.z.", "e.a.", "e.d.", "e.g.", "enz.", "etc.", "e.v.", "evt.", "fig.", "i.e.",
|
|
"i.h.b.", "ir.", "i.p.v.", "i.s.m.", "m.a.w.", "max.", "m.b.t.", "m.b.v.", "mevr.", "min.",
|
|
"n.a.v.", "nl.", "n.o.t.k.", "n.t.b.", "n.v.t.", "o.a.", "ong.", "pag.", "ref.", "t.a.v.",
|
|
"tel.", "zgn."
|
|
))
|
|
|
|
|
|
def find_lemmata(tokens):
|
|
""" Annotates the tokens with lemmata for plural nouns and conjugated verbs,
|
|
where each token is a [word, part-of-speech] list.
|
|
"""
|
|
for token in tokens:
|
|
word, pos, lemma = token[0], token[1], token[0]
|
|
if pos.startswith("JJ") and word.endswith("e"):
|
|
lemma = predicative(word)
|
|
if pos == "NNS":
|
|
lemma = singularize(word)
|
|
if pos.startswith(("VB", "MD")):
|
|
lemma = conjugate(word, INFINITIVE) or word
|
|
token.append(lemma.lower())
|
|
return tokens
|
|
|
|
|
|
class Parser(_Parser):
|
|
|
|
def find_tokens(self, tokens, **kwargs):
|
|
# 's in Dutch preceded by a vowel indicates plural ("auto's"): don't replace.
|
|
kwargs.setdefault("abbreviations", ABBREVIATIONS)
|
|
kwargs.setdefault("replace", {"'n": " 'n"})
|
|
s = _Parser.find_tokens(self, tokens, **kwargs)
|
|
s = [re.sub(r"' s (ochtends|morgens|middags|avonds)", "'s \\1", s) for s in s]
|
|
return s
|
|
|
|
def find_lemmata(self, tokens, **kwargs):
|
|
return find_lemmata(tokens)
|
|
|
|
def find_tags(self, tokens, **kwargs):
|
|
if kwargs.get("tagset") in (PENN, None):
|
|
kwargs.setdefault("map", lambda token, tag: wotan2penntreebank(token, tag))
|
|
if kwargs.get("tagset") == UNIVERSAL:
|
|
kwargs.setdefault("map", lambda token, tag: wotan2universal(token, tag))
|
|
if kwargs.get("tagset") is WOTAN:
|
|
kwargs.setdefault("map", lambda token, tag: (token, tag))
|
|
return _Parser.find_tags(self, tokens, **kwargs)
|
|
|
|
|
|
class Sentiment(_Sentiment):
|
|
|
|
def load(self, path=None):
|
|
_Sentiment.load(self, path)
|
|
# Map "verschrikkelijk" to adverbial "verschrikkelijke" (+1%)
|
|
if not path:
|
|
for w, pos in list(dict.items(self)):
|
|
if "JJ" in pos:
|
|
p, s, i = pos["JJ"]
|
|
self.annotate(attributive(w), "JJ", p, s, i)
|
|
|
|
parser = Parser(
|
|
lexicon = os.path.join(MODULE, "nl-lexicon.txt"),
|
|
frequency = os.path.join(MODULE, "nl-frequency.txt"),
|
|
morphology = os.path.join(MODULE, "nl-morphology.txt"),
|
|
context = os.path.join(MODULE, "nl-context.txt"),
|
|
default = ("N(soort,ev,neut)", "N(eigen,ev)", "Num()"),
|
|
language = "nl"
|
|
)
|
|
|
|
lexicon = parser.lexicon # Expose lexicon.
|
|
|
|
sentiment = Sentiment(
|
|
path = os.path.join(MODULE, "nl-sentiment.xml"),
|
|
synset = "cornetto_id",
|
|
negations = ("geen", "gene", "ni", "niet", "nooit"),
|
|
modifiers = ("JJ", "RB",),
|
|
modifier = lambda w: w.endswith(("ig", "isch", "lijk")),
|
|
tokenizer = parser.find_tokens,
|
|
language = "nl"
|
|
)
|
|
|
|
spelling = Spelling(
|
|
path = os.path.join(MODULE, "nl-spelling.txt")
|
|
)
|
|
|
|
|
|
def tokenize(s, *args, **kwargs):
|
|
""" Returns a list of sentences, where punctuation marks have been split from words.
|
|
"""
|
|
return parser.find_tokens(s, *args, **kwargs)
|
|
|
|
|
|
def parse(s, *args, **kwargs):
|
|
""" Returns a tagged Unicode string.
|
|
"""
|
|
return parser.parse(s, *args, **kwargs)
|
|
|
|
|
|
def parsetree(s, *args, **kwargs):
|
|
""" Returns a parsed Text from the given string.
|
|
"""
|
|
return Text(parse(s, *args, **kwargs))
|
|
|
|
|
|
def tree(s, token=[WORD, POS, CHUNK, PNP, REL, LEMMA]):
|
|
""" Returns a parsed Text from the given parsed string.
|
|
"""
|
|
return Text(s, token)
|
|
|
|
|
|
def tag(s, tokenize=True, encoding="utf-8", **kwargs):
|
|
""" Returns a list of (token, tag)-tuples from the given string.
|
|
"""
|
|
tags = []
|
|
for sentence in parse(s, tokenize, True, False, False, False, encoding, **kwargs).split():
|
|
for token in sentence:
|
|
tags.append((token[0], token[1]))
|
|
return tags
|
|
|
|
|
|
def keywords(s, top=10, **kwargs):
|
|
""" Returns a sorted list of keywords in the given string.
|
|
"""
|
|
return parser.find_keywords(s, **dict({
|
|
"frequency": parser.frequency,
|
|
"top": top,
|
|
"pos": ("NN",),
|
|
"ignore": ("rt", "mensen")}, **kwargs))
|
|
|
|
|
|
def suggest(w):
|
|
""" Returns a list of (word, confidence)-tuples of spelling corrections.
|
|
"""
|
|
return spelling.suggest(w)
|
|
|
|
|
|
def polarity(s, **kwargs):
|
|
""" Returns the sentence polarity (positive/negative) between -1.0 and 1.0.
|
|
"""
|
|
return sentiment(s, **kwargs)[0]
|
|
|
|
|
|
def subjectivity(s, **kwargs):
|
|
""" Returns the sentence subjectivity (objective/subjective) between 0.0 and 1.0.
|
|
"""
|
|
return sentiment(s, **kwargs)[1]
|
|
|
|
|
|
def positive(s, threshold=0.1, **kwargs):
|
|
""" Returns True if the given sentence has a positive sentiment (polarity >= threshold).
|
|
"""
|
|
return polarity(s, **kwargs) >= threshold
|
|
|
|
split = tree # Backwards compatibility.
|
|
|
|
#---------------------------------------------------------------------------------------------------
|
|
# python -m pattern.nl xml -s "De kat wil wel vis eten maar geen poot nat maken." -OTCL
|
|
|
|
if __name__ == "__main__":
|
|
commandline(parse)
|