contextlab/melville-corpus

ContextLab Herman Melville Corpus

Dataset Description

This dataset contains works of Herman Melville (1819-1891), preprocessed for computational stylometry research. The texts were sourced from Project Gutenberg and cleaned for use in the paper "A Stylometric Application of Large Language Models" (Stropkay et al., 2025).

The corpus includes 10 books by Herman Melville, including Moby-Dick, Bartleby the Scrivener, and Typee. All text has been converted to lowercase and cleaned of Project Gutenberg headers, footers, and chapter headings to focus on the author's prose style.

Quick Stats

• Books: 10
• Total characters: 5,257,881
• Total words: 912,881 (approximate)
• Average book length: 525,788 characters
• Format: Plain text (.txt files)
• Language: English (lowercase)

Dataset Structure

Books Included

Each .txt file contains the complete text of one book:

File	Title
10712.txt	White Jacket; Or, The World on a Man-of-War
11231.txt	Bartleby, the Scrivener: A Story of Wall-Street
13720.txt	Mardi, and a voyage thither, Vol. 1 (of 2)
13721.txt	Mardi, and a voyage thither, Vol. 2 (of 2)
15.txt	Moby-Dick; or, The Whale
15422.txt	Israel Potter: His Fifty Years of Exile
21816.txt	The Confidence-Man: His Masquerade
2694.txt	I and My Chimney
28656.txt	Typee
4045.txt	Omoo: Adventures in the South Seas

Data Fields

• text: Complete book text (lowercase, cleaned)
• filename: Project Gutenberg ID

Data Format

All files are plain UTF-8 text:

• Lowercase characters only
• Punctuation and structure preserved
• Paragraph breaks maintained
• No chapter headings or non-narrative text

Usage

Load with datasets library

from datasets import load_dataset

# Load entire corpus
corpus = load_dataset("contextlab/melville-corpus")

# Iterate through books
for book in corpus['train']:
    print(f"Book length: {len(book['text']):,} characters")
    print(book['text'][:200])  # First 200 characters
    print()

Load specific file

# Load single book by filename
dataset = load_dataset(
    "contextlab/melville-corpus",
    data_files="54.txt"  # Specific Gutenberg ID
)

text = dataset['train'][0]['text']
print(f"Loaded {len(text):,} characters")

Download files directly

from huggingface_hub import hf_hub_download

# Download one book
file_path = hf_hub_download(
    repo_id="contextlab/melville-corpus",
    filename="54.txt",
    repo_type="dataset"
)

with open(file_path, 'r') as f:
    text = f.read()

Use for training language models

from datasets import load_dataset
from transformers import GPT2Tokenizer, GPT2LMHeadModel, Trainer, TrainingArguments

# Load corpus
corpus = load_dataset("contextlab/melville-corpus")

# Combine all books into single text
full_text = " ".join([book['text'] for book in corpus['train']])

# Tokenize
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")

def tokenize_function(examples):
    return tokenizer(examples['text'], truncation=True, max_length=1024)

tokenized = corpus.map(tokenize_function, batched=True, remove_columns=['text'])

# Initialize model
model = GPT2LMHeadModel.from_pretrained("gpt2")

# Set up training
training_args = TrainingArguments(
    output_dir="./results",
    num_train_epochs=10,
    per_device_train_batch_size=8,
    save_steps=1000,
)

# Train
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized['train']
)

trainer.train()

Analyze text statistics

from datasets import load_dataset
import numpy as np

corpus = load_dataset("contextlab/melville-corpus")

# Calculate statistics
lengths = [len(book['text']) for book in corpus['train']]

print(f"Books: {len(lengths)}")
print(f"Total characters: {sum(lengths):,}")
print(f"Mean length: {np.mean(lengths):,.0f} characters")
print(f"Std length: {np.std(lengths):,.0f} characters")
print(f"Min length: {min(lengths):,} characters")
print(f"Max length: {max(lengths):,} characters")

Dataset Creation

Source Data

All texts sourced from Project Gutenberg, a library of over 70,000 free eBooks in the public domain.

Project Gutenberg Links:

• Books identified by Gutenberg ID numbers (filenames)
• Example: 54.txt corresponds to https://www.gutenberg.org/ebooks/54
• All works are in the public domain

Preprocessing Pipeline

The raw Project Gutenberg texts underwent the following preprocessing:

1. Header/footer removal: Project Gutenberg license text and metadata removed
2. Lowercase conversion: All text converted to lowercase for stylometry
3. Chapter heading removal: Chapter titles and numbering removed
4. Non-narrative text removal: Tables of contents, dedications, etc. removed
5. Encoding normalization: Converted to UTF-8
6. Structure preservation: Paragraph breaks and punctuation maintained

Why lowercase? Stylometric analysis focuses on word choice, syntax, and style rather than capitalization patterns. Lowercase normalization removes this variable.

Preprocessing code: Available at https://github.com/ContextLab/llm-stylometry

Considerations for Using This Dataset

Known Limitations

• Historical language: Reflects 19th-century America vocabulary, grammar, and cultural context
• Lowercase only: All text converted to lowercase (not suitable for case-sensitive analysis)
• Incomplete corpus: May not include all of Herman Melville's writings (only public domain works on Gutenberg)
• Cleaning artifacts: Some formatting irregularities may remain from Gutenberg source
• Public domain only: Limited to works published before copyright restrictions

Intended Use Cases

• Stylometry research: Authorship attribution, style analysis
• Language modeling: Training author-specific models
• Literary analysis: Computational study of Herman Melville's writing
• Historical NLP: 19th-century America language patterns
• Educational: Teaching computational text analysis

Out-of-Scope Uses

• Case-sensitive text analysis
• Modern language applications
• Factual information retrieval
• Complete scholarly editions (use academic sources)

Citation

If you use this dataset in your research, please cite:

@article{StroEtal25,
  title={A Stylometric Application of Large Language Models},
  author={Stropkay, Harrison F. and Chen, Jiayi and Jabelli, Mohammad J. L. and Rockmore, Daniel N. and Manning, Jeremy R.},
  journal={arXiv preprint arXiv:2510.21958},
  year={2025}
}

Additional Information

Dataset Curator

ContextLab, Dartmouth College

Licensing

MIT License - Free to use with attribution

Contact

• Paper & Code: https://github.com/ContextLab/llm-stylometry
• Issues: https://github.com/ContextLab/llm-stylometry/issues
• Contact: Jeremy R. Manning ([email protected])

Related Resources

Explore datasets for all 8 authors in the study:

• Jane Austen
• L. Frank Baum
• Charles Dickens
• F. Scott Fitzgerald
• Herman Melville
• Ruth Plumly Thompson
• Mark Twain
• H.G. Wells