Markov Chain Text Generator in Python

#!/usr/bin/python3

import random
import sys
import re

# This is the length of the "state" (sequence of characters) the next character is predicted from.
STATE_LEN = 4

# This is a maximum for the characters in one sentence.
MAX_CHARACTERS = 1000

# The contents of INPUT_TEXT are used to train the markov model. The longer the provided text is, the better
# or more interesting the results become.
INPUT_TEXT = """
Romanesque architecture is an architectural style of medieval Europe characterized by semi-circular arches. There is no consensus for the beginning date of the Romanesque style, with proposals ranging from the 6th to the 11th century, this later date being the most commonly held. In the 12th century it developed into the Gothic style, marked by pointed arches. Examples of Romanesque architecture can be found across the continent, making it the first pan-European architectural style since Imperial Roman architecture. The Romanesque style in England is traditionally referred to as Norman architecture.
Combining features of ancient Roman and Byzantine buildings and other local traditions, Romanesque architecture is known by its massive quality, thick walls, round arches, sturdy pillars, barrel vaults, large towers and decorative arcading. Each building has clearly defined forms, frequently of very regular, symmetrical plan; the overall appearance is one of simplicity when compared with the Gothic buildings that were to follow. The style can be identified right across Europe, despite regional characteristics and different materials.
Many castles were built during this period, but they are greatly outnumbered by churches. The most significant are the great abbey churches, many of which are still standing, more or less complete and frequently in use. The enormous quantity of churches built in the Romanesque period was succeeded by the still busier period of Gothic architecture, which partly or entirely rebuilt most Romanesque churches in prosperous areas like England and Portugal. The largest groups of Romanesque survivors are in areas that were less prosperous in subsequent periods, including parts of southern France, rural Spain and rural Italy. Survivals of unfortified Romanesque secular houses and palaces, and the domestic quarters of monasteries are far rarer, but these used and adapted the features found in church buildings, on a domestic scale.
"""
# source: https://en.wikipedia.org/wiki/Romanesque_architecture

# learn records in the _model_ an occurrence of the character _next_ after the sequence _state_.
def learn(model, state, next):
    if state not in model:
        model[state] = {}
    if next not in model[state]:
        model[state][next] = 0
    model[state][next] += 1
    return model

# train creates a model based on the input text in _data_.
def train(data):
    model = {}
    # remove everything except letters, numbers and punctuation.
    data = re.sub("[^A-Za-z0-9 .,-]+", "", data)

    state = ""
    skipWhitespace = True
    for i in range(len(data)):
        next = data[i]

        if skipWhitespace and next == ' ':
            continue
        skipWhitespace = False

        model = learn(model, state, next)
        if next == ".":
            state = ""
            skipWhitespace = True
            continue

        state += next
        if len(state) > STATE_LEN:
            state = state[1:]
    return model

# choose the next character based on the possibilities given in _d_ by random weighted choice.
# This is done by adding up all the weights and choosing a number randomly between 0 and the resulting
# sum and using this random number as an index.
def choose(d):
    weights = sum(d.values())
    r = random.randint(0, weights)
    for next,count in d.items():
        r -= count
        if r <= 0:
            return next
    raise "error"

# sentence returns a sentence generated randomly by the model.
def sentence(model):
    state = ""
    text = ""
    for i in range(MAX_CHARACTERS):
        next = choose(model[state])
        text += next
        state += next
        if len(state) > STATE_LEN:
            state = state[1:]
        if next == ".":
            break
    return text


# Train the model based on the provided text, then
# create 3 sentences based on it.
model = train(INPUT_TEXT)
for i in range(3):
    print(sentence(model))