Introduction

This short article describes how I made a python script to generate short text in a chosen language at a particular level. It is based on the OpenAI Cookbook examples and the european language levels Common European Framework of Reference (CEFR) .
At the end of the article I present the prototype python code.

Motivation

One of the things I’ve found fustrating while learning a language on my own is the difficulty of finding reading material that is

on my level
interesting to me.

While playing with the aforementioned OpenAI python examples it occured to me that I could try to to use the european language levels to generate texts in foreign languages at those levels.

Those levels are:

A1 Basic User. Basic phrases and everyday expressions
A2 Basic User. Can understand sentences and frequently used expressions related to personal, family, shopping, geography work
B1 Independent User. Can understand the main points of plain speech/texts
B2 Independent User. Can understand the main ideas of complex text/speech (abstract and concrete)
C1 Proficient User. Can understand a wide range of demanding texts.
C2 Proficient User. Can understand with ease virtually everything heard or read.

Several attempts later. I had a prototype script and prompt that seemed to work reasonably well. I added a bunch of commandline options an defaults and got something halfway decent.

You can give this script the following arguments:

-s or --subject eq. -s "History of Florence" Or -s "History" -s "Florence"
-l or --length number of words eq. 500 (default 500)
-L or --language eq. -L italian (default "Italian")
-a or --level  eq -a A1/A2/B1/B2/C1/C2 (default "B1")
-S or --style eq -S formal/informal/cynical etc etc (default="formal")
-f or --format eq -f article/blog/.. (default="article")
-t or --tone Tone of text to generate. neutral, light, somber, humorous, sarcastic etc (default="neutral")

Caveats

When set to the C2 level the results can be rather ..ahem.. poetic!

When generating a piece of text in a language like latin about a modern subject, the result can be anything from hilarious to absurd.
(I haven’t dared to try Klingon yet)

Running the code

OpenAI account

To use this with OpenAI, I recommend you get a “pay as you go” account. It is the cheapest way to use OpenAI for personal use. Don’t forget to lower the monthly limit from 120 dollars to say 5 dollars. That way you limit the damage from possible runaway programs.

Setup

Generate an openAI api key and store it in the file .env in the same directory as the python file.

Example .env

OPEN_AI_KEY="youropenaikeyhere"

If you use git, don’t forget to add the .env file to the .gitignore file.

Examples

A1 language level

./run.py -s “Amsterdam History” -L English -a A1

**Amsterdam History**

Amsterdam is the capital city of the Netherlands. It has a very interesting history. The city’s history begins in the 12th century. At this time, Amsterdam was a small fishing village. The people in the village lived by the water. They caught fish and traded them with other places. The name "Amsterdam" comes from a dam on the river Amstel. This dam was built to protect the village from the water.

Note the short simple sentences.

C2 language level

./run.py -s “Amsterdam History” -L English -a C2

Amsterdam, the capital city of the Netherlands, boasts a rich and multi-faceted history that dates back to its founding in the late 12th century. Originally established as a modest fishing village along the banks of the Amstel River, the city's name is derived from the word "Amstelredamme," which signifies a dam constructed on the river. This early settlement laid the groundwork for what would become a significant urban center in Europe.

Note the longer more complex sentences and extended vocabulary

Code

#!/usr/bin/env python3

import argparse
from typing import List
from openai import OpenAI
from dotenv import load_dotenv

# load .env with OPEN_API_KEY
load_dotenv()

# OpenAI init
client = OpenAI()

# Set the prompt
prompt_role = '''
You are an expert languages teacher.
Your task is to write text
on the SUBJECT,
in the LANGUAGE,
and the european LANGUAGELEVEL that are given to you.
You should respect the instructions:
the LENGTH, the STYLE, The FORMAT and the TONE
'''


def ask_chatgpt(messages):
    response = client.chat.completions.create(model="gpt-4o-mini",
                                              messages=messages,
                                              max_tokens=800)
    return (response.choices[0].message.content)

# populate the prompt with the given arguments
def format_prompt(
        subjects: List[str],
        language: str,
        level: str,
        length_words: int,
        sformat: str,
        style: str,
        tone: str):
    prompt = f"""{prompt_role}\n
    SUBJECT: {subjects}\n
    LANGUAGE: {language}\n
    LANGUAGELEVEL: {level}\n
    FORMAT: {sformat}\n
    TONE: {tone}\n
    LENGTH: {length_words} words\n
    STYLE: {style}"""
    return ask_chatgpt([{"role": "user", "content": prompt}])

def main() -> None:
    parser = argparse.ArgumentParser()
    parser.add_argument(
            "-l",
            "--length",
            type=int, help="Lenght of article in words",
            default=500)
    parser.add_argument(
            "-L",
            "--language",
            type=str,
            help="Language to generate",
            default="Italian")
    parser.add_argument(
            "-a",
            "--level",
            type=str,
            help="Language level to generate",
            default="B1")
    parser.add_argument(
            "-S",
            "--style",
            type=str,
            help="Style of language to generate. ex formal, informal",
            default="formal")
    parser.add_argument(
            "-f",
            "--format",
            type=str,
            help="Format of text to generate. article, blog etc ",
            default="article")
    parser.add_argument(
            "-t",
            "--tone",
            type=str,
            help="""Tone of text to generate.
            neutral, light, somber, humorous, sarcastic etc""",
            default="neutral")
    requiredNamed = parser.add_argument_group('required named arguments')
    requiredNamed.add_argument(
            "-s",
            "--subject",
            action="append",
            help="Subject/keywords to to write about")

    args = parser.parse_args()

    article = format_prompt(
        args.subject,
        args.language,
        args.level,
        args.length,
        args.format,
        args.style,
        args.tone)
    print(article)
    s = ''.join(str(x) for x in args.subject)
    outfile = open(s + ": " + args.language + " " + args.level + ".md" , 'w+')
    outfile.write(article)
    outfile.close()


if __name__ == "__main__":
    main()

Conclusion

I now use this script to generate daily pieces of text in italian an latin to read during my commute to and from work.

This script ticks all the following boxes for me.

Interesting text to read? Check!
In a language I’m learning? Check
On a level I can manage? Check!
Usable on a phone? Check!
During otherwise boring and nonproductive times? Double Check!

I run it under termux on my Android telephone.

Caveats

When set to the C2 level the results can be rather ..ahem.. poetic!

When generating a piece of text in a language like latin about a modern subject, the result can be anything from hilarious to absurd.

OpenAI And Language Learning

Introduction

Motivation

Caveats

Running the code

OpenAI account

Setup

Example .env

Examples

A1 language level

C2 language level

Code

Conclusion

Caveats