Picolibrary: a very small press

Toggle navigation

Occasional Thoughts

Archive
RSS

Picolibrary: a very small pressAugust 31, 2026

I’ve decided to republish a book. Someone on the fediverse – I can’t
remember who – said that Professor Mmaa’s
Lecture was their favorite book. So I
wrote to the rights holder to see if I could get it back into
print. Sorry, I know I’m supposed to be working on the
game, but it’s been a busy summer and
I’ve had a hard time getting a free moment. Also, I first inquired
about these rights a couple of years ago, but just heard back. I’ve
been telling everyone that I bought my first book ever, just to
confuse people.
The agent that I ended up speaking to was a little surprised to hear
from someone who had never published a book before. I can only plead
that I have figured out how to make games, so surely I can figure out
how to make books happen. Also, I happen to know a fair bit about
copyright law since I used to work on legal issues in software
freedom. (Actually, I apparently know more about copyright law than
Amazon, which initially rejected the book because they thought it was
in the public domain. They relented when I explained.) You can just
do things.
What follows are some notes on what the process has been like. To
start with, the contract. I feel confident that I could have written
a book contract, but why bother, because the Authors Guild has a
great model
contract.
I mean, it’s great for authors (as you would expect). But also, it
explains the reasons behind each term in clear English. I didn’t want
to use it verbatim because this is a reprint, so the rights issues are
different (and I disagreeed with some of its terms). Claude made me a
new contract inspired by the Authors Guild one, and I confirmed that
it was what I wanted. I had to remind Claude about the read-aloud
issue;
obviously, I want the book to accessible to blind and other
reading-impaired people! And its internal review of the first draft
surfaced like nine missing clauses. Ask Claude to review its work;
you’ll almost always find something worth fixing.
Then it was time to prepare the text. In the old days, Dover would do
photo facsimilies of public domain works they wanted to republish.
This was not beautiful, but it totally did the job with 20th century
technology. We’re in the 21st century now. You can scan a book and
OCR it. OCR often gives weird scanning mistakes, especially if your
book has non-English text interspersed. Historically, you would clean
them up by hand. Now we have LLMs to do it for us. I should warn that
this is not 100% perfect, especially as regards to
formatting. Professor Mmaa’s Lecture has illustrations, tables,
subscripts, small caps, italics, etc. They all have to be manually
checked. But I have found no errors in the text itself. You might
thing it’s easy to find scannos – just search for non-words. But
this book is full of hapax legomena. “Brillat-Beetonin”, “kcourage”,
“Homomahomet”, “abbovvve”, “Maetermith”, to name just a few. Oh,
also, untranslated French. LLMs do it without breaking a sweat.
Archivists have known about this LLM superpower for a couple of
years.
I had initially thought that I would work off the Internet Archive’s
scan of the book. Unfortunately, their scan is the 1975 edition, and
the rights holder wanted me to use the 1953 edition. There are pretty
major differences – the 1975 edition is actually almost 20%
longer. There are new bits everywhere. Here’s one:

“If we add to this the thesis propounded by the very reverend
Archussher, who, basing his conclusions on right-to-left consumption
of the collection of cellulose which consumed from left-to-right is
known as Genesis v, declared that according to homo itself its
appearance took place at nine in the morning of October 28, 4004
B.C. of the homo calendar&emdashwe shall realize what discrepancies
there are among the various estimates of this mammifer’s age, even
if we agree to call homo what by some scientists is called
’notyethomo,’ and by others: nomorehomo.

In addition to the new bits, the 1975 version has some minor
rewordings – like, “everyone” gets the
MLP treatment and becomes
“every termite”. It also has American spellings. And it doesn’t have
Bertrand Russell’s preface.
So then I thought I would just photograph each page and have Claude
OCR it. Tedious, but doable. Just to test it out, I went to the Claude
web interface and asked it to do the first page. It refused, citing
copyright law. Buddy, don’t you remember how you, yourself, were
trained? Anyway, it suggested instead that I use a book-scanning
service. Why are you giving me instructions about how to “infringe
copyright” after refusing to do it yourself? (It didn’t believe me
when I told it, honestly, that I had the rights).
I found a place called 1DollarScan. They said to email them if I
wanted the original book back, so I did, and then they quoted a price
that was… not one dollar. Indeed, it was closer to a dollar a page.
So instead, I went with Bound Book
Scanning. Their pricing was much more
reasonable (still not one dollar), and they did a great job.
And Claude Code doesn’t seem to care about copyright. I mean, it
politely asked if I had the rights to Bertrand Russell’s preface
before including it (I didn’t, so I went out and got them). But
otherwise, it’s perfectly happy to clean up scans. It’s not quite the
same job – it’s working from an existing scan rather than doing 100%
of the OCR itself. But actually Claude Code ends up doing a bunch of
OCR anyway.
It’s a little scary to entrust text to a LLM – they are famous for
hallucinating. (I know some people think that hallucinations have been
fixed – they haven’t. I just asked Claude to find me some boxer shorts,
and it confidently, wrongly, asserted that Israel doesn’t have an
underwear industry.) But for OCR, it turns out that Claude can do a
good job even with text that’s somewhat out-of-distribution, like this
novel.
Some decisions need to be made specifically for ebook
publication. Lots of people read books on their phones, which means
that poetry will, unfortunately, have extra line-wraps. Consider:

Here, fore-, mid-, and hind- are all aligned at the dash to indicate
that they all refer to sections of the gut. But on my Pixel 9, with
default Kindle settings, portrait mode, “And my gustatory organ crawls
all along my” takes up a whole line, wrapping fore- onto the next
line. We decided to wrap “my” onto the next line, and keep the dashes
aligned. There’s no perfect solution here, but I think this does the
best job of preserving Themerson’s intent. Claude took a few tries to
get this to look right on various screen sizes, but managaed in the
end. And of course, if you have a larger screen, and in the print
version, we won’t wrap.
That’s an aesthetic challenge; there were also technical challenges.
There’s one bit in the book where “the Detective imitates the sound of
the old Enemy of termites – the Dove.” This is rendered as… well, I
don’t actually have names for these characters. They’re straight lines
and arcs, with accent marks. I almost ended up rendering the arcs as
U+2323 SMILE; the straight lines as em-dashes and vertical bars. But using
combining accents didn’t render right, so Claude suggested using
ruby, which I had
known about (I tried to learn Japanese at one point) but wouldn’t have
ever thought to use. Brilliant idea, except that the Kindle renders
the ruby too far above the smiles, and it doesn’t support the CSS
necessary to fix it (we discovered this after Claude built me a little
tool that would let me interactively adjust the size/height of the
accents to get it just right). So, not actually brilliant. Then I
tried using images of Themerson’s original glyphs instead. But they
were black-on-white, which looks bad in dark mode (and Kindle doesn’t
support the CSS that would let me select different images for dark
mode). The final plan ended up being to trace Themerson’s notation
into a font, and embed the font. Yow.
Anyway, after a bunch of irritating formatting tweaking, the 1953
original text of Professor Mmma’s Lecture was ready to go!
Oh, and I needed a web site,
which meant a domain name, and thus a business name. I do most
projects under my own name, so that everyone knows what they’re going
to get: something weird. But publishers tend to have business names,
and anyway it felt odd to have my name on a book someone else wrote. I
chose the name Picolibrary to emphasize that this was just a small
side project. I know sometimes small projects get big. Bennett Cerf
said, “we just said we were going to publish a few books on the side
at random”, but ended up with Random House. I do not expect this to
happen to me. Still, I have a bunch of spare ISBNs (in the US, if you
want two you might as well buy ten), so anything’s possible.
I asked Claude to put together the site, and it came up with some
hilariously Claude text about what the press is and what its mission
is, all totally made up. It genuinely used the word “genuinely”, in
case any readers were genuinely unsure whether or not it was slop. I
deleted it and used my own voice. (To be clear: this was always the
plan; I don’t use AI prose generation for anything I share with
others, as I find it painful to read).
So now if you want to read one of the weirdest books I’ve ever read
(and I’ve read House of Leaves), now you
can read Professor Mmaa’s Lecture.
It’s about what it’s like to be a termite. Bertrand Russell recommends
it!

Previous post: Is it a joke?

Hello

This is a blog by David Turner.

Search

Recent Posts

Picolibrary: a very small press
Is it a joke?
Blue Prince (1989)
Responses to the Hacker News comment on Six Mistakes
Six mistakes I made and one I avoided
Building a game with the Real Engine
Middles updates, design, and prior art
Middles: A Daily Word Game
A review of one puzzle from Quern: Undying Thoughts
Turds

The process of republishing a book, specifically Professor Mmaa’s Lecture, involved navigating complex issues related to rights acquisition, contract negotiation, textual preparation through modern technology, and final aesthetic formatting. The author initially sought to secure the rights to reprint the work after hearing it was a favored text, requiring engagement with rights holders despite being a first-time publisher. This process necessitated leveraging knowledge of copyright law, as the author possessed background in software freedom and legal issues, which proved helpful when addressing initial rejections by distributors who initially considered the work to be in the public domain.

The contractual phase involved drafting an agreement; while the author felt confident about writing a book contract, they opted to utilize a model provided by the Authors Guild, adapting it for their specific situation due to the nature of a reprint and disagreeing with certain terms. Using a large language model assisted in creating a revised contract inspired by the Authors Guild framework, which was subsequently refined to ensure accessibility for blind and reading-impaired individuals through read-aloud considerations. Further review of the draft revealed missing clauses, underscoring the iterative nature of legal document creation.

Text preparation presented significant technical challenges, especially concerning the transition from historical methods to modern optical character recognition (OCR). While historical methods involved manual photofacsimilation, the contemporary approach involves scanning and OCR, which introduces errors, particularly when dealing with complex layouts containing illustrations, tables, subscripts, and different typographical styles. Although LLMs are capable of performing this cleanup effectively, the text required careful manual verification to ensure perfect fidelity regarding formatting elements like mathematical notation or specific character rendering. Furthermore, the source material presented textual challenges such as hapax legomena and untranslated French, which LLMs handled with relative ease.

A complication arose when determining the correct edition; the rights holder insisted on using a 1953 edition over the Internet Archive’s digitized 1975 version, resulting in differences in length and content. This necessitated a manual decision regarding source material integrity. The process of obtaining the final text involved experimenting with external services; initial attempts to use direct LLM processing for OCR were met with copyright refusals, prompting the author to explore commercial book scanning services like Bound Book Scanning, which proved more practical.

The most intricate hurdles emerged during ebook formatting, balancing aesthetic intent with technical constraints across various screen sizes. Layout decisions, such as managing line wrapping for poetic structures or aligning typographical symbols, required sophisticated algorithmic adjustments executed by the LLM and custom tools. This included complex rendering issues concerning specialized notation—where characters like accents were rendered incorrectly by standard systems—leading to an exploratory process involving font embedding and interactive adjustment tools to achieve the desired visual outcome, ultimately resulting in tracing original glyphs into a custom font for final publication.

Finally, establishing the publishing identity required selecting a business name that reflected the project's scope without implying association with established publishers; Picolibrary was chosen to denote a small side project. While the author utilized the LLM for prose generation, they consciously chose to replace its generated content with their own voice to maintain authenticity, recognizing the inherent risk of hallucinations in reliance on such models when dealing with sensitive material. The entire endeavor culminated in making Professor Mmaa’s Lecture available through this small press operation.