计算文本中的单词数

这个网站上有一个任务：统计单词数。单词被认为是英文字母字符的连续序列（从 A 到 z）。
以下是示例：

Hello there, little user5453 374 ())$. I’d been using my sphere as a stool. Slow-moving target 839342 was hit by OMGd-63 or K4mp. 
contains "words" 
['Hello', 'there', 'little', 'user', 'I', 'd', 'been', 'using', 'my','sphere', 'as', 'a', 'stool', 'Slow', 'moving', 'target', 'was', 'hit', 'by', 'OMGd', 'or', 'K', 'mp']

有些词应该被排除。这些话是：

"a", "the", "on", "at", "of", "upon", "in" and "as", case-insensitive.

我的解决方案是这样的：
我们用文字来分解它——我们删除所有不必要的东西。在循环中，我们检查该单词是否被禁止或为空字符串并增加计数器

import codewars_test as test
import re

except_word = ["a", "the", "on", "at", "of", "upon", "in", "as"]


def word_count(s):
    all_word = re.sub(r'([^A-Za-z]+)', r' ', s).split(' ')
    print(all_word)
    cnt = 0
    for word in all_word:
        if word in except_word or word == '':
            continue
        else:
            cnt += 1
    return cnt

if __name__ == '__main__':
    test.assert_equals(word_count("hello there"), 2)
    test.assert_equals(word_count("hello there and a hi"), 4)
    test.assert_equals(word_count("I'd like to say goodbye"), 6)
    test.assert_equals(word_count("Slow-moving user6463 has been here"), 6)
    test.assert_equals(word_count("%^&abc!@# wer45tre"), 3)
    test.assert_equals(word_count("abc123abc123abc"), 3)
    test.assert_equals(word_count("Really2374239847 long ^&#$&(*@# sequence"), 3)

    long_text = r"""
            I’d been using my sphere as a stool. I traced counterclockwise circles on it with my fingertips and it shrank until I could palm it. My bolt had shifted while I’d been sitting. I pulled it up and yanked the pleats straight as I careered around tables, chairs, globes, and slow-moving fraas. I passed under a stone arch into the Scriptorium. The place smelled richly of ink. Maybe it was because an ancient fraa and his two fids were copying out books there. But I wondered how long it would take to stop smelling that way if no one ever used it at all; a lot of ink had been spent there, and the wet smell of it must be deep into everything.
            """

    test.assert_equals(word_count(long_text), 112)

基本测试：

        test.assert_equals(word_count("hello there"), 2)
        test.assert_equals(word_count("hello there and a hi"), 4)
        test.assert_equals(word_count("I'd like to say goodbye"), 6)
        test.assert_equals(word_count("Slow-moving user6463 has been here"), 6)
        test.assert_equals(word_count("%^&abc!@# wer45tre"), 3)
        test.assert_equals(word_count("abc123abc123abc"), 3)
        test.assert_equals(word_count("Really2374239847 long ^&#$&(*@# sequence"), 3)

该函数通过了所有基本测试，但不会在长文本上产生正确的答案，
这是一个基本示例：

Example Input 2

I’d been using my sphere as a stool. I traced counterclockwise circles on it with my fingertips and it shrank until I could palm it. My bolt had shifted while I’d been sitting. I pulled it up and yanked the pleats straight as I careered around tables, chairs, globes, and slow-moving fraas. I passed under a stone arch into the Scriptorium. The place smelled richly of ink. Maybe it was because an ancient fraa and his two fids were copying out books there. But I wondered how long it would take to stop smelling that way if no one ever used it at all; a lot of ink had been spent there, and the wet smell of it must be deep into everything.

Example Output 2

112

我的函数返回 113 也许我对任务的理解不正确（翻译自 Google）？在服务器上的随机测试中，我的解决方案总是产生 1 到 3 个额外的单词。

请解释一下，这是什么原因？就好像我忘记删除某些东西（我不明白是什么）。