Literary style imitation using Markov chains and the ChatGPT-4 API

Examples

Seeding phrases into the output

One of the most intriguing features is --seed-words, which enables you to seed the output with a phrase that does not exist in the original text.

$ python mimic.py --seed-words "after checking my GPS coordinates"
After checking my GPS coordinates, an indefinable feeling of despair came over me and we stopped nowhere long enough to face the dishonour and emptiness of bereavement.

Comparing the similarity of the output to the original text

Sentiment analysis of the training text using --sentiment

Below is an example of doing sentiment analysis using the TextBlob library.

About

This program can be thought of as a "literary style imitation algorithm" because the primary purpose is to mimic the style and tone of the original text. It creates new content based on the input text rather than directly copying existing content.

The difficult parts, the Markov algorithm code, came from the original repository that this was forked from: Markov-Chain-Sentence-Generator, by GitHub user lachlanpage.

The overall algorithm does this:

1. Input: a text file whose style you want to imitate.
2. Second order Markov prediction (looking backwards by two words at a time) and returning approximately 30 words.
3. Send the 30 words to the ChatGPT-4 API for grammar and punctuation cleanup. For example sometimes there is a missing verb or subject.
4. Output: a sentence imitating the original text's style.

The original intent was to learn more about Markov models and ChatGPT-4 API. Use at your own risk. And don't use it for malicious purposes.

Sample Output from Heart of Darkness

He raided the country perhaps, and at that moment I stood horror-struck at the sight of one of the pilgrims behind the blind whiteness of the immense matted jungle, with the thought that at least this was the reality.

The quote above is an example of how this algorithm, when trained on a text, will generate a convincingly authentic sounding quote that sounds like it came from the original text. Although this quote is not from Heart of Darkness, if you search Google or ChatGPT for the quote they will return results from Heart of Darkness.

Using Markov chains, the result is more convincing than if you simply asked ChatGPT to simulate an imaginary quote from the text; those examples often seem silly or more like broad summarizations of the overall text. Whereas this program's output are often convincingly authentic sounding compared with the original text.

• Literary style imitation using Markov chains and the ChatGPT-4 API
  • Sample Output from Heart of Darkness
• Installation and Usage
• Help output
• Signing up for a ChatGPT-4 API key
• Setting the GPT_API_KEY environment variable
  • Linux, Unix, and Mac OS
    • User level and persistent
    • System level and persistent
  • Windows
    • Windows Powershell
      • User level and persistent
      • Machine level and persistent
    • Windows cmd shell
      • User level and persistent
      • Machine level and persistent
• Files
  • Python source code files
  • Example texts to train the Markov model on

Installation and Usage

You need to be using Python 3.x.

# Clone the repo
git clone https://github.com/tcpiplab/Literary-style-imitation-using-Markov-chains-and-the-ChatGPT-4-API.git
cd Literary-style-imitation-using-Markov-chains-and-the-ChatGPT-4-API

# Install the requirements. You can optionally use a virtual environemt if you know how.
pip install -r requirements.txt

# Add your API key as an environment variable. See instructions below.
echo 'export GPT_API_KEY="xx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"' >> ~/.zshrc
source ~/.zshrc

# Run the program.
python chatGptApiCall.py

Help output

$ python mimic.py -h
usage: mimic.py [-h] [-i INPUT_FILE] [-r] [-sc] [-sw SEED_WORDS] [-v] [-q] [-l LENGTH] [-m MAX_TOKENS] [-st SIMILARITY_THRESHOLD]
                [-w SIMILARITY_WINDOW] [-n NUMBER_OF_RESPONSES] [-temp TEMPERATURE]

A command line tool to generate random phrases that imitate a literary style based on a training text.

options:
  -h, --help            show this help message and exit
  -i INPUT_FILE, --input-file INPUT_FILE
                        Path to the input file a.k.a the training text (optional)
  -r, --raw-markov      Print the raw Markov result (optional)
  -sc, --similarity-check
                        Quantify how similar the output is to the original text (optional)
  -sw SEED_WORDS, --seed-words SEED_WORDS
                        Word(s) to seed the Markov search. If not found in the original text, it will be prepended to the output. (optional)
  -v, --verbose         Enable verbose mode
  -q, --quiet           Disable logging completely
  -l LENGTH, --length LENGTH
                        Approximate length of the output (optional)
  -m MAX_TOKENS, --max-tokens MAX_TOKENS
                        Maximum number of tokens to generate. If not specified, it increases automatically if you specify length. (optional)
  -st SIMILARITY_THRESHOLD, --similarity-threshold SIMILARITY_THRESHOLD
                        Floating point similarity threshold for the similarity check (optional)
  -w SIMILARITY_WINDOW, --similarity-window SIMILARITY_WINDOW
                        Number of consecutive words in the sliding window used for the similarity check (optional)
  -n NUMBER_OF_RESPONSES, --number_of_responses NUMBER_OF_RESPONSES
                        Number of responses to generate. Higher number also increases temperature and increases likelihood of
                        repetition(optional)
  -temp TEMPERATURE, --temperature TEMPERATURE
                        Specify the AI temperature (creativity). Float between 0 and 2.0.
  --sentiment           Perform sentiment analysis on input data.
  -nc, --no-chat-gpt    Do not call the ChatGPT API. Print the raw Markov result instead. (optional)
  -t, --test            Test the API call

Signing up for a ChatGPT-4 API key

You can sign up for a ChatGPT-4 API key here. It will cost you a little bit of money each time you use it. At the time of this writing it is pretty cheap. For example, so far it has cost me less than $3.00 to develop and test all the AI features of this program. And they allow you to set soft and hard limits so that you don't accidentally spend too much money.

Setting the GPT_API_KEY environment variable

This is mandatory in order to make calls to the ChatGPT-4 API.

Linux, Unix, and Mac OS

The examples below are for if you're using the zsh shell. If you're not using the zsh shell, the ~/.zshrc file should be replaced with whichever file is appropriate for the shell you're using and how your shell is set up. If you're using the zsh shell, then just follow the example below. If you're using the bash shell then here's the rule of thumb:

• If you want your changes to be available in all shell sessions (both login and non-login), place them in ~/.bashrc and make sure that your ~/.bash_profile sources your ~/.bashrc file. This is usually already set up for you by the OS install scripts.
• If you want your changes to only apply to login shells (like when you ssh into your machine), put them in ~/.bash_profile.

If you're using another shell like ksh, tcsh, csh, or sh, then I'm sure you know exactly what you're doing and can figure out how to set environment variables with no problem.

User level and persistent

echo 'export GPT_API_KEY="xx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"' >> ~/.zshrc
source ~/.zshrc

System level and persistent

You'll have to do this as root or use sudo. Either way, I assume that you know what you're doing if you have this level of authorization.

sudo echo 'export GPT_API_KEY="xx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"' > /etc/profile.d/myenvvars.sh

Windows

Windows Powershell

User level and persistent

To set a permanent environment variable that persists across sessions and reboots, you can use the [System.Environment]::SetEnvironmentVariable() method. For example, to set a user-level environment variable use the following command but replace the xx-xxxxxxxx... key value with your own key value. Leave the User field just like you see below:

[System.Environment]::SetEnvironmentVariable("GPT_API_KEY", "xx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx", "User")

After setting this permanent environment variable, you need to restart your PowerShell session for the changes to take effect.

Machine level and persistent

To set a machine-level (system-wide) environment variable you need to run PowerShell with administrative privileges. Leave the Machine field just like you see below:

[System.Environment]::SetEnvironmentVariable("GPT_API_KEY", "xx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx", "Machine")

Now restart your computer for the changes to take effect.

Windows cmd shell

User level and persistent

setx GPT_API_KEY "xx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

Now close your cmd window and open another one.

Machine level and persistent

To set a system-wide environment variable (as opposed to a user-specific one), you need to run the cmd shell as an administrator and use the /M switch:

setx /M GPT_API_KEY "xx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

Now close your cmd window and open another one. For other users to use the environment variable you may have to reboot.

Files

Python source code files

• n_order_markov.py
• chatGptApiCall.py is the file you need to run. It calls n_order_markov.py and does the call to the ChatGPT-4 API
• main.py is single order Markov Model - not currently used (included from the original repository this was forked from)
• second_order.py is second order Markov Model - not currently used (included from the original repository this was forked from)

Example texts to train the Markov model on

• bleak-house.txt is the full text of Charles Dickens' Bleak House
• book.txt is a Harry Potter test file
• sampletext.txt is a collection of Donald Trump tweets
• heartOfDarkness.txt is the full text of Joseph Conrad's Heart of Darkness