#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import argparse
import codecs
import json
import logging
import os
import sys
import time
logger = logging.getLogger(__name__)
[docs]class Nefnir(object):
"""
A rule-based lemmatizer
"""
def __init__(self):
"""
Initialize an instance of the Nefnir lemmatizer.
"""
nefnir_dir = os.path.dirname(sys.argv[0])
# Load rules
rules_path = os.path.join(nefnir_dir, 'rules.json')
with open(rules_path, encoding='utf-8') as f:
self.rules = json.load(f)
# Load tagset
tag_path = os.path.join(nefnir_dir, 'tags.json')
with open(tag_path, encoding='utf-8') as f:
self.tagmap = json.load(f)
self.proper = {t for t in self.tagmap if t[0] == 'n' and t[-1] in {'m', 'ö', 's'}}
self.unanalyzed = {t for t in self.tagmap if t[:2] == 'nx'} | {'x', 'e', 'as'}
[docs] def lemmatize(self, form, tag):
"""
Lemmatize a word form given its part-of-speech tag.
:param form: A word form.
:param tag: The word form's part-of-speech tag.
:return: The word form's lemma.
"""
try:
ntag = self.tagmap[tag]
except KeyError:
if any((c for c in tag if c.isalpha())):
logger.warning("Unknown tag: {}".format((form, tag)))
return form
# Websites and interjections
if tag in {'v', 'au'}:
return form.lower()
# Unanalyzed words
if tag in self.unanalyzed:
return form
# Words that end with a hyphen
if form[-1] == '-':
if tag in self.proper:
return self.recase(form, tag, form)
return form.lower()
# Words that end with a punctuation mark
if not form[-1].isalpha():
return form
form_lower = form.lower()
if ntag not in self.rules:
logger.debug("No rules for this tag: {} {} {}".format(form, tag, ntag))
return self.recase(form, tag, form)
if form_lower in self.rules[ntag]['form']:
suffix_from, suffix_to = self.rules[ntag]['form'][form_lower]
else:
suffixes = get_suffixes(form_lower)
try:
target = next(s for s in suffixes if s in self.rules[ntag]['suffix'])
suffix_from, suffix_to = self.rules[ntag]['suffix'][target]
except StopIteration:
logger.debug("No rules for this word form: {} {} {}".format(form, tag, ntag))
return self.recase(form, tag, form)
form_prefix = form_lower[:-len(suffix_from)] if suffix_from else form_lower
lemma = form_prefix + suffix_to
if not lemma:
logger.warning("Rule produced an empty lemma: ({}, {}, {}) ('{}' -> '{}')".format(form, tag, ntag,
suffix_from, suffix_to))
lemma = form_lower
return self.recase(form, tag, lemma)
[docs] def recase(self, form, tag, lemma):
"""
Determine how to properly case a lemma given the word form and part of speech tag it was derived from.
Nefnir transforms words into lowercase prior to lemmatization. Some words, such as proper nouns, abbreviations
and foreign words therefore need to be re-capitalized or changed back into uppercase.
:param form: A word form, cased as it was written.
:param tag: The word form's part-of-speech tag.
:param lemma: The word form's lemma, in lowercase.
:return: A properly cased lemma.
"""
# Hyphenated words: try to maintain original casing in every part
# 1) (DNA-þræðinum, nþeþg) -> dna-þráður -> DNA-þráður
# 2) (Vestur-Íslendingum, nkfþ-s) -> vestur-íslendingur -> Vestur-Íslendingur
# 3) (Stoke-on-Trent, e) -> stoke-on-trent -> Stoke-on-Trent
if '-' in form[1:-1]:
fparts = form.split('-')
lparts = lemma.split('-')
result = []
for fpart, lpart in zip(fparts, lparts):
if fpart.lower() == lpart.lower():
# part was not transformed by lemmatization
result.append(fpart)
elif fpart.isupper():
# part was transformed and was uppercase
result.append(lpart.upper())
elif fpart.istitle():
# part was transformed and was capitalized
result.append(lpart.title())
else:
# part was transformed and not uppercase or capitalized
result.append(lpart.lower())
if tag in self.proper and not result[0].isupper():
result[0] = result[0].title()
return "-".join(result)
# Proper nouns: capitalize the lemma
# 1) (Halldórs, nken-s) -> halldór -> Halldór
# 2) (HALLDÓRS, nken-s) -> halldór -> Halldór
if tag in self.proper:
# if len(form) > 1 and form.isupper():
# return lemma.upper()
return lemma.title()
# If none of the above applies, return lemma in lowercase
return lemma
[docs]def get_suffixes(s):
"""
Return an iterator yielding a string's suffixes, from the largest to the smallest.
:param s: A text string.
:return: An iterator for the string's suffixes.
"""
return (s[pos:] for pos in range(len(s) + 1))
[docs]def main():
# Command line interface
parser = argparse.ArgumentParser()
parser.add_argument("-i", "--input-file", help="read input from specified file", required=True)
parser.add_argument("-o", "--output-file", help="write output to specified file", required=True)
parser.add_argument("-f", "--from-encoding", help="character encoding of input file (default: utf-8)",
default="utf-8")
parser.add_argument("-t", "--to-encoding", help="character encoding of output file (default: utf-8)",
default="utf-8")
parser.add_argument("-s", "--separator", help="the string separating word forms, tags and lemmas (default: \\t)",
default='\t')
args = parser.parse_args()
args.separator = codecs.decode(args.separator, 'unicode_escape')
# Lemmatize input
time_start = time.time()
nefnir = Nefnir()
logger.info("Reading input from {} ({})".format(args.input_file, args.from_encoding))
logger.info("Separator set to {}".format(repr(args.separator)))
with open(args.input_file, encoding=args.from_encoding) as f:
lines = f.read().splitlines()
num_lines = len(lines)
lines = {l: None for l in lines}
for line in lines:
try:
form, tag = line.split(args.separator)
if form:
lemma = nefnir.lemmatize(form, tag)
lines[line] = args.separator.join((form, tag, lemma))
except ValueError:
if line.strip():
logger.warning('Ignoring line: {}'.format(line))
# Stats
time_elapsed = time.time() - time_start
lines_per_second = num_lines / time_elapsed
stats = "{:,} lines processed in {:.2f} s ({:,.1f} lines/s)".format(num_lines, time_elapsed, lines_per_second)
logger.info(stats)
# Write output
logger.info("Writing output to {} ({})".format(args.output_file, args.to_encoding))
with open(args.input_file, encoding=args.from_encoding) as f_in:
with open(args.output_file, 'w', encoding=args.to_encoding) as f_out:
for line in f_in:
line = line.rstrip('\n')
output = lines[line] or ''
f_out.write(output + '\n')
if __name__ == '__main__':
FORMAT = '%(asctime)s - %(levelname)s %(message)s'
logging.basicConfig(level=logging.INFO, format=FORMAT)
main()