Example #1
0
f = open(r"sample_desire.tsv", "w", encoding='utf-8')

filtered_lines = []
for s in range(1, len(lineList)):
    my_split = lineList[s].split('\t')
    my_tokens = my_split[7].split(' ')
    tagged_tokens = nltk.pos_tag(my_tokens)
    formlist = ['desire', 'desired']

    if not (fl.isGerund(my_split) or fl.hasGerundAfter(my_split, my_tokens)
            or fl.isIntransitive(my_split, tagged_tokens)
            or fl.hasToAfterVerb(my_split, my_tokens) or fl.isAdjectiveOrNoun(
                my_split, my_tokens, tagged_tokens, formlist)
            or tagged_tokens[int(my_split[5]) + 1][1].startswith('VB') or
            (my_tokens[int(my_split[3]) + 1].lower() == 'to'
             and tagged_tokens[int(my_split[3]) + 2][1] == 'VB')):

        filtered_lines.append(lineList[s])

n = 0
sampled_numbers = []
while n < 100:
    s = random.randint(1, len(filtered_lines))
    if not s in sampled_numbers:
        sampled_numbers.append(s)
        my_split = filtered_lines[s].split('\t')
        if not fl.verbHasXcomp(my_split):
            f.write(filtered_lines[s])
            n = n + 1

f.close()
Example #2
0
            or fl.hasGerundAfter(my_split, my_tokens)
            or beConsidered(my_split, my_tokens) or
            (my_tokens[int(my_split[5]) + 1].lower() in ['that', 'what']
             and tagged_tokens[int(my_split[5]) + 1][1] == 'IN')
            or fl.isAdjective(my_split, my_tokens, tagged_tokens, formlist)
            or my_split[2] in ['Considered']
            or tagged_tokens[int(my_split[3]) + 1][1].startswith(
                ('JJ', 'DT', 'VBN')) or
            (tagged_tokens[int(my_split[3]) + 1][1].startswith('RB')
             and tagged_tokens[int(my_split[3]) + 2][1].startswith(
                 ('JJ', 'VBN'))) or
            (not tagged_tokens[int(my_split[3])][1].startswith('NN'))
            or my_split[5] == "0" or my_split[1].lower()
            in ['matter', 'fact', 'possibility', 'issue', 'idea'] or
            tagged_tokens[int(my_split[3]) + 1][1] in ['VBP', 'VBP', 'VBZ']):

        filtered_lines.append(lineList[s])

n = 0
sampled_numbers = []
while n < 100:
    s = random.randint(1, len(filtered_lines))
    if not s in sampled_numbers:
        sampled_numbers.append(s)
        my_split = filtered_lines[s].split('\t')
        if not (fl.verbHasXcomp(my_split) or fl.verbHasWrongIobj(my_split)):
            f.write(filtered_lines[s])
            n = n + 1

f.close()