Story-generation-process

Generation pipeline: 19 large language models asked to rewrite 16 classical fairy tales. Each notebook corresponds to one source tale from Project Gutenberg.

The texts produced here feed into LLM_readability_score_analysis for statistical evaluation. Part of a Master's research project at Whitireia New Zealand (2025).

Purpose

Build a controlled corpus of LLM-generated narratives against a fixed human-written baseline, such that readability and accuracy assessments can be run under identical prompts and identical source texts across models.

Design

Each notebook:

1. Loads one classical fairy tale as input
2. Submits the same generation prompt to 19 LLMs from major providers (Anthropic, OpenAI, Google, Meta, Mistral, and others)
3. Collects model outputs into a tabular format
4. Computes textstat readability metrics on each output

Cross-model dispatch was originally done through the Unify.ai routing service; see the historical note below.

Source tales

16 notebooks, one per tale:

Notebook	Source tale
HOW_JACK_WENT_TO_SEEK_HIS_FORTUNE.ipynb	How Jack Went to Seek His Fortune
Henny_Penny.ipynb	Henny Penny
JACK_HANNAFORD.ipynb	Jack Hannaford
JOHNNY_CAKE.ipynb	Johnny Cake
Lazy_Jack.ipynb	Lazy Jack
MASTER_OF_ALL_MASTERS_readability_score.ipynb	Master of All Masters
MR_MIACCA.ipynb	Mr Miacca
MR_VINEGAR.ipynb	Mr Vinegar
TEENY_TINY.ipynb	Teeny Tiny
THE_CAT_AND_THE_MOUSE_fairy_tale.ipynb	The Cat and the Mouse
THE_MAGPIE'S_NEST.ipynb	The Magpie's Nest
THE_MASTER_AND_HIS_PUPIL.ipynb	The Master and His Pupil
THE_OLD_WOMAN_AND_HER_PIG.ipynb	The Old Woman and Her Pig
THE_ROSE_TREE.ipynb	The Rose Tree
THE_STORY_OF_THE_THREE_LITTLE_PIGS.ipynb	The Story of the Three Little Pigs
THE_THREE_SILLIES.ipynb	The Three Sillies

Historical note: Unify.ai

The original pipeline used Unify.ai — a unified-API routing service that consolidated access to LLMs from Anthropic, OpenAI, Google, Meta, Mistral, AWS Bedrock, Fireworks, Together, Anyscale, DeepInfra, and others behind a single SDK (from unify import Unify).

Unify.ai was discontinued in 2024. The UNIFY_KEY variable is preserved in the notebooks as a historical reference but no longer resolves. To re-run the pipeline today, the generation step must be rewritten against each provider's native SDK (openai, anthropic, google-generativeai, and so on).

Reproduce (requires porting)

pip install textstat pandas openpyxl jupyter
# plus the native SDKs for each provider you have access to
jupyter notebook

Replace the unify = Unify(model_name) + unify.generate(prompt, temperature) calls in the third cell of each notebook with equivalent calls against your preferred provider SDK.

Limitations

• Unify.ai decommissioning means strict reproducibility of the original results requires access to the same model snapshots, which is not guaranteed after model versioning updates at each provider
• Only one generation per (model × tale) — no temperature-sweep or multi-seed averaging
• Prompt is held constant; prompt-sensitivity of readability is not studied here

Related

• LLM_readability_score_analysis — statistical analysis of the texts generated here
• Master's research: The comparative analysis of human-written and large language model generated children's tales: Readability and accuracy assessments of LLM-generated texts, Whitireia New Zealand (2025)

License

No license is specified; for commercial use, please open an issue.

---

Language: English · Русский