Sign in

Ben Brumfield

@benwbrum.bsky.social
1.4K followers 1.7K following 249 posts

Open Source #DigitalHumanities software engineer. Founder of FromThePage.com, a platform for collaborative #manuscript #transcription to engage the public in #archives and create digital scholarly editions.

PostsRepliesMedia
Reposted by Ben Brumfield
Laura Morreale @lauramorreale.bsky.social · 09/10/2026
Excited to report on some work I did with @sims-mss.bsky.social & some tough-to-read but very rewarding historical sources. Come hear me talk about transcription tools now at our disposal, & what is required to use them (hint: it's not computer fairy dust...) www.library.upenn.edu/events/schoe...
library.upenn.edu
Using Handwritten Text Recognition for Rich and Abundant Historical Source Collections: The Case of Medieval Italian Financial Records
Laural Morreale (Independent Scholar) will describe how to train Handwritten Text Recognition models to read difficult sources like the Italian financial records and will outline the costs, potential ...
03212
Ben Brumfield @benwbrum.bsky.social · 06/10/2026
Do you have any ideas on that? I'm no expert on LLMs, but this really goes against everything I thought I knew.
110
Ben Brumfield @benwbrum.bsky.social · 06/10/2026
The thing that really surprised me about your findings was how closely the LLM followed the letterforms, even when the reading made no sense. Surely the Vulgate Matthew is well represented in its training set, so producing "cum" for "eum" goes against the "autocomplete on steroids" analogy.
210
Ben Brumfield @benwbrum.bsky.social · 06/10/2026
One question I had about your method is whether you're keeping all parameters the same between the five runs, or changing the random seed each time?
100
Ben Brumfield @benwbrum.bsky.social · 06/10/2026
More recently, @wjbmattingly.bsky.social has developed a technique that identifies probable hallucinations generated by LLM transcription: wjbmattingly.com/blog/where-t... This seems very promising, but is not perfect: the reading ¨suscitaverunt cum¨ is wrong but not a lectio facilior, so ???
wjbmattingly.com
Where the Readings Disagree: Finding HTR Errors by Asking a Model Five Times — William Mattingly, PhD
A small local app that runs a handwritten text recognition model several times over the same manuscript page and uses the disagreement between runs to show, word by word and letter by letter, where th...
182
Ben Brumfield @benwbrum.bsky.social · 06/10/2026
And that was the state of the art up until November, when Gemini 3 was released. I really don´t understand how they did it, but they managed to solve the hallucination problem, at least for straightforward transcription of monolingual texts in well attested languages.
151
Ben Brumfield @benwbrum.bsky.social · 06/10/2026
pattern is poisonous -- the errors are more reasonable than the original, making it impossible to spot-check the transcription. And a transcription of a unique archival document that has plausibly seductive errors is worse than no transcription at all.
161
Ben Brumfield @benwbrum.bsky.social · 06/10/2026
This was always my criticism of people using multi-modal LLMs to transcribe (non-ancient) handwritten material: in the case of ambiguity, the statistical model would always favor a reading that "makes sense" -- the lectio facilior in textual criticism terms. For documentary editing, this error
121
Reposted by Ben Brumfield
Joel @sentantiq.bsky.social · 02/10/2026
the fundamental question I have had for these applications is how can LLM or statistical based methods support the "lectio difficilior?
36919
Ben Brumfield @benwbrum.bsky.social · 02/10/2026
She was very impressed to see three reenactors playing Nationalist Chinese parts.
000
Ben Brumfield @benwbrum.bsky.social · 02/10/2026
Her caption: Rockford Illinois WW2 days reenactment: A Japanese soldier (bottom) and a member of the Republic of China’s “X force” (top) after retraining and re-equipment by the Americans and British. Circa Ledo road operations and later.
100
Ben Brumfield @benwbrum.bsky.social · 02/10/2026
I got a report from my daughter on last weekend's event. Apparently she and her boyfriend joined up with a Polish bicycle troop (?), shopped for records (pictured), and also got to hang out around some kind of glider-borne tank (pictured)
Young woman in period dress, in front of a small tank.A young man and woman observe another woman leafing through a binder of 78rpm records.
100
Reposted by Ben Brumfield
Gabriel Bodard @palaeofuturist.bsky.social · 02/10/2026
#TEIFriday Can anyone give examples of magical texts (spells/curses...) encoded in TEI XML? Any language, date, etc. #digitalhumanities #ancientmagic
21011
Ben Brumfield @benwbrum.bsky.social · 28/09/2026
At my undergrad institution in the 90s, 50% of engineering majors had a cross-disciplinary second major as well. "Why not both?"
010
Reposted by Ben Brumfield
Princeton University Press @princetonupress.bsky.social · 24/09/2026
"A triumph of ancient Mediterranean archaeology & of comparative history."—Josephine Quinn, author of How the World Made the West Slaving States by Elizabeth Fentress & Adam Rabinowitz is out now (20 Oct UK pub). Explore a free preview: buff.ly/MmGkTnn #AncientHistory #History #Archaeology
Slaving States: From West Africa to the Ancient Mediterranean by Elizabeth Fentress and Adam Rabinowitz. A pathbreaking new perspective on the ways ancient societies were shaped and transformed by slave trading.
022
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
(I've been asking questions about KMT vs. CCP contributions to the war effort in WW2, or US foreign policy in 1942-1947, and it's not blatant; always very reasonable.)
000
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
Well, I don't want anything that I can't verify, so perhaps I should be looking for KMT publications in English that try to sway US opinion against the CCP? Amnesty International letters about Tienanmen Square? I'm sure that it would transcribe the Declaration fine; the bias is subtle.
100
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
I'm trying to test out a Chinese LLM for potential bias. Using a chat interface, I have detected at least some, however my goal is to use this for HTR. What are some handwritten primary sources that a CCP-friendly LLM might refuse to transcribe or would edit to be pro-CCP?
111
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
Do I get bragging rights from being shushed by the Chief Information Officer of the Library of Congress? Because that happened last week.
110
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
A response from my daughter:
020
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
To your point, the proportion of axis uniforms on the website do not match the proportion in the photos my daughter sent me. (Likely also unrepresentative, however.)
100
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
110
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
They sent us a great photo of themselves in front of a 48-star flag in period (civilian) costumes—made me want to buy war bonds!
100
Ben Brumfield @benwbrum.bsky.social · 22/09/2026
My daughter and her (Polish) boyfriend attended that one a year or two ago! Much higher representation from Polish reenactors at that one than most, thanks to the Chicago community.
100
Reposted by Ben Brumfield
Dr. Emily Friedman @ecfriedman.com · 10/09/2026
Intro to Secretary Hand today. Our room full of 18th century scholars collectively saying “thank God they stopped doing this in our period”
Letterforms from secretary hand. Only some of them are recognizable.
1011212
Ben Brumfield @benwbrum.bsky.social · 10/09/2026
Honestly, it sounds a bit smoother than most of the talks I give, although I still have plenty of filler words and pauses while I point to problems in the MSS on the slide. Much better than being Judith Martin-inflected! (Also, keep on being awesome!)
020
Ben Brumfield @benwbrum.bsky.social · 10/09/2026
I just watched a webinar I recorded after binge-listening to @bretdevereaux.bsky.social interviews, and am embarrassed by how much of his oral style I copied. Reminds me of the time my college friend told me he could tell I'd been reading Miss Manners anthologies again by how I talked.
130
Ben Brumfield @benwbrum.bsky.social · 09/09/2026
Recording of my webinar on how I use (or avoid) AI to edit historic account books. www.youtube.com/watch?v=jtZZ...
youtube.com
Editing Historic Account Books: When and Whether to Use AI
YouTube video by fromthepage
031
Ben Brumfield @benwbrum.bsky.social · 30/08/2026
We have been on the receiving end of sales campaigns so customized, well researched, and well executed that you'd expect them from a thousand-employee company selling to a ten-thousand employee company. We're a mom-and-pop, and I'm not sure the aspirant sellers were more than one dude.
000
Ben Brumfield @benwbrum.bsky.social · 29/08/2026
say. Would you add pointing to a scholarly edition? How about to a transcription designed to support full-text search? What would work better for a researcher using a screen reader?
000
Ben Brumfield @benwbrum.bsky.social · 29/08/2026
The transcription looked good, but when I compared it against one produced by a volunteer, there were massive differences. Looking more closely, Gemini had added pointing to the text, totally absent from the original. This feels wrong to me, but I'm not sure what someone with more Hebrew would
Diff view of transcript of Hebrew text, showing only a handful of matching letters.
100
Ben Brumfield @benwbrum.bsky.social · 29/08/2026
I noticed something interesting in the #HTR that #Gemini produced from this #Hebrew letter today: עֵת נִפְרְדָה פְּרַאנְסִיס מֵאֲהָבֶיהָ הוּסַר מִבֵּית כֹּהֵן כְּלִיל תִּפְאֶרֶת עָלֶיהָ בָּכוּ עֲנִיֵּי עַם יַעַן נֶעֱדֶרֶת זוֹרַעַת חֶסֶד בִּתְבוּנוֹת כַּפֶּיהָ
Hebrew text in careful, outlined script.
100
Reposted by Ben Brumfield
Nelson Rare Books @nelsonrarebooks.com · 26/08/2026
Do they eat feral hogs in Texas?
321
Ben Brumfield @benwbrum.bsky.social · 26/08/2026
but it was too gamey. Then I used my mother-in-law's recipe to make four gallons of green pork chili for a potluck the next week. The acid and spice worked. One of my coworkers said, "I wish I had a swimming pool of this." I love running my own #digitalhumanities company, but man, I miss potlucks.
120
Ben Brumfield @benwbrum.bsky.social · 26/08/2026
person I can call about this. I shot two feral sows and field dressed them, but don't have enough room in my deep freeze for all the pork. Would you be able to use a sow's leg if I brought it to work on Monday?" The answer was, of course, "yes". We tried baking part of it in a Dr. Pepper sauce,
120
Ben Brumfield @benwbrum.bsky.social · 26/08/2026
About 20 years ago, I worked for a software company in Austin. One Sunday evening, I got a call from the Director of Engineering, my boss's boss. This was very strange--I remembered weekend calls from the start-up days, but this guy has fixed all that chaos when he came in. "Ben, you're the only+
120
Ben Brumfield @benwbrum.bsky.social · 20/08/2026
I'm getting really excited about tomorrow's webinar. Not because I think it'll be our best--editing historic account books is pretty niche, and the presenter (me) is way out of his depth--but because it's on my own historical research. content.fromthepage.com/aug-2026-web...
content.fromthepage.com
Editing Historic Account Books: When and Whether to Use AI (August 20, 2026) - FromThePage Blog
091
Reposted by Ben Brumfield
sam blickhan @snblickhan.bsky.social · 19/08/2026
my god, we’ve been going about this all wrong 🙄
011
Ben Brumfield @benwbrum.bsky.social · 18/08/2026
According to a different set of spammers, I should be thinking of the capital, since "We've raised millions of dollars in financing for crowdsourced manuscript transcription businesses in the last year!" Sure you have...
120
Ben Brumfield @benwbrum.bsky.social · 18/08/2026
I'm getting very tired of LLM-based spam. No, I don't think that you've "booked 100 calls" with companies buying manuscript transcription in the last month, Caleb.
160
Ben Brumfield @benwbrum.bsky.social · 18/08/2026
We've also seen an institutional account funded by researcher A for two years turn the bill over to B in year 3 -- totally different projects, but shared infrastructure. (One university is on its third funder, with 4+ different researchers using the software across three departments.)
021
Ben Brumfield @benwbrum.bsky.social · 18/08/2026
Great (and sometimes painful) question, Alex. We've seen this work best when researchers partner with the library to provide the software a a campus-wide resource. But libraries are under budget constraints too!
121
Ben Brumfield @benwbrum.bsky.social · 18/08/2026
Join me this Thursday to walk through my process transcribing 19th-century account books in a quest to memorialize people buried in a Virginia slave cemetery. Editing Historic Account Books: When and Whether to Use AI (August 20, 2026) content.fromthepage.com/aug-2026-web...
Two facing pages listing the debit and credit entries for the account of Daniel & William King with Jeremiah White Graves
030
Ben Brumfield @benwbrum.bsky.social · 18/08/2026
It's so much nicer than the 19th-century account books I'm currently fighting with. Go for it!
Opening showing debit and credit sides of accounts between Daniel & William King and Jeremiah White Graves
020
Reposted by Ben Brumfield
Prof Katherine Schofield 🇬🇧🇦🇺🇪🇺 @katherineschof8.bsky.social · 14/08/2026
Oh god. I've just heard. The poor man. And in the final reckoning, all of us who participated in the furore on social media — even out of the best of motives and with care — were ultimately part of the mob. I am sorry. I am so sorry to his family and loved ones. The poor man.
3643
Ben Brumfield @benwbrum.bsky.social · 13/08/2026
Aside from new efforts like community gardens (which are typically a single 1m x 2m raised bed), I don't think the settlement and landownership patterns for most of the US would have supported them. If anywhere, maybe in New England?
010
Reposted by Ben Brumfield
California State Library @castatelibrary.bsky.social · 11/05/2026
Dust off your cursive skills by transcribing handwritten letters from authors like influential journalist Ambrose Bierce or native Californian Gertrude Atherton. Head over to From the Page and check out our collections recently uploaded to fromthepage.com/castatelibra...
letter written in cursive by Ambrose Bierce postcard with a waterfall and trees and the written message part on the right.
001
Reposted by Ben Brumfield
Jordan S. Carroll 🌹 @jordanscarroll.bsky.social · 26/05/2026
Early modern recipe books are like
Entries such as “For sore eyes, the evil, to preserve eyes, elder wine”
1296
Ben Brumfield @benwbrum.bsky.social · 30/07/2026
If you go back another generation or two, the answer doesn't change, though the work gets increasingly harder without guano or artificial fertilizer. (I have my great-grandfather's 2-page memoir, where he talks about the work of "rolling logs": slash-and-burn agriculture was awful in Virginia.)
000
Ben Brumfield @benwbrum.bsky.social · 30/07/2026
100 years puts me at my great-grandparents generation, but I know what they were doing: Tobacco farmer Tobacco farmer Tobacco farmer Tobacco farmer My grandparents would have been approximately 10 years old, but they would have been pulling leaves or tying hands with the adults, depending on gender.
120