The following guest post from digital humanities scholar Nichole Misako Nomura is part of our Vanishing Culture series, highlighting the power and importance of preservation in our digital age. Read more essays online or download the full report now.
Punch cards are a fascinating binary data storage format that aren’t just history—they’re still used by knitting machines today! Thanks to the Internet Archive and other collections, we still have access to historic punch cards, but there are some technical challenges to using them in the format they’re stored in. Meet a few folx working on those challenges.
Punch card computation—the good old days, or the bad old days, depending on who you talk to—lives firmly in the land of “the old days” for most—a piece of history, with pedagogical and nostalgic benefit—but it’s alive and well in the textile world.
Histories of computing frequently point to the Jacquard loom as the example of the “first” code, used to create fabric in a variety of patterns—like this 1839 commemorative portrait of the Jacquard loom’s inventor, J.M. Jacquard: These looms use punch cards to lift warp threads above or below the weft, allowing for the mechanized creation of non-repeating patterns across the loom. (
While the Jacquard loom gets all the attention for being the first code, the punch card knitting machine transitioned from being a Jacquard attachment on lace and knitting machines in industrial textile production to the kind of local, DIY code that a lot of people in textiles interacted with—many of whom were women. By the 1970s, they were used by people knitting for themselves and their families, for take-home piece-work, and in textile factory settings. The punch card machine was eventually replaced in commercial and, if you can afford it, home contexts by machines that could control individual needles, instead of depending on a punch card’s repeat—but the machines are still in use in a number of hobbyist workshops (like my own!) and are even still in production (albeit much-reduced).
The knitting machines I own share their punch card dimensions (24 stitches wide) with one of the first punch cards (the Hollerith card, used for the 1890 census, was a 24-column punch card). They’re an important piece of computing history—and crucially, one of the few that isn’t only history because a broad community of people, on- and off-line, are still sharing knowledge on how to hack, restore, and use them.
All punch cards are fundamentally digital, even if we don’t generally think of “digital” as a property physical objects can have. It is only recently that our associations of computing with “the cloud” and other ephemeral metaphors have superseded the fundamentally physical processes that support computation. Working with knitting machine punch cards reminds me that the cloud is a metaphor, and lets me own and manipulate my code in a way I find both challenging and creatively liberating.
The coolest thing about knitting punch cards is that they really are just sequences of “yes” and “no”—and that information is actionable in a wide variety of machines, all of which perform different functions based on that information. Some machines can knit two different colors at once—one color is “yes,” and the other is “no.” Others can skip the stitches marked as “no.” Some machines can make tuck or slip stitches, and others still do something called “weaving,” a variation on the aforementioned two-color knitting. The information encoded by these punch cards, regardless of the actual dimensions of the cards, is interoperable across most machines—and when it is not, it is because the number of holes in the punch card doesn’t permit the same numeric repeat (30 and 24 are divisible by a similar, but not identical, set of numbers).
There are a lot of punch card knitting patterns stored on the internet, found in multi-purpose archives like the Internet Archive and in countless community-hosted Google Drives. Unlike a pattern written for hand-knitting, these punch cards are not, strictly-speaking, usable in the format they are stored in. While I could knit a sweater from a set of directions that look like knit 1, purl 40 from an image, working with images of punch card knitting patterns requires a different workflow—one that, counterintuitively, is challenging because of the digital nature of the punch card itself.
Digitizing the already-digital
Knitting machine punchcards are relatively easy to digitize in a way that preserves the information, but relatively difficult to digitize in a way that makes the transition back from stored-on-the-computer to stored-in-physical-material feasible. It is entirely possible to recreate a punch card using an image—by hand, laboriously, with a physical hole punch. (Image: Usually I work row-by-row, with a ruler across the image, to make sure I’m putting holes in the right spots and chanting things like “3 yes, 1 no, 3 yes, 4 no” in repeating patterns. It is error-prone, but consistent with how generations pre-internet worked with these patterns—translating an image in a book or magazine into binary data of “punch this, not that.”
However, those with more patience for debugging than patience for tedious card-punching have been experimenting with a variety of methods that allow for computer-controlled punching—or, more often, cutting that imitates punching. The Cricut is the standout piece of hardware here, although any machine that can precision cut paper using code will do. These machines, called CNC machines (CNC stands for “computer numerical control”), can have laser or blade attachments, and they work the same way as the massive plasma cutters used for cutting steel. A layer of software, which can be open-source or proprietary, translates an image stored as a SVG (scalable vector graphic) into strings of numbers that control the cutting head.
SVGs aren’t that hard to generate off of images; the challenge here is generating an SVG off an image that actually fits in a punch card knitting machine. There is exactly one spot a hole can go that will work with the dimensions of a knitting machine, and unfortunately, low-quality scans (even pretty-good quality scans) are often too noisy to make it possible to blow up the image and then cut out all the dark spots. I tried, and was rewarded with a punch card that jammed, ripped, and complained loudly for several rows before I gave up. With higher-quality scans, this one-to-one kind of reproduction might work—but only for the machine the punch card was originally designed for. So there’s an incentive to extract the information in those punch cards in a way that is not tied to the specific dimensions of one knitting machine or another. Knitting magazines frequently turned to standardized grid formats for this, preserving the information (“yes, no, yes, yes, no”) but not the specific dimensions of any given punch card.
I work with punch cards in my home workshop for fun, but I’m also fortunate enough to work with them at Stanford’s Textile Makerspace, where Quinn Dombrowski has been teaching data visualization using textiles on an assortment of knitting machines, looms, and sewing machines. Quinn’s colleague Simon Wiles, a Digital Scholarship Research Developer at Stanford’s Center for Interdisciplinary Digital Research, has worked on a computer-vision approach for converting images of punch cards into data that could be used to generate new physical punch cards. He previously worked on an incredible digitization effort on behalf of the Stanford Libraries to digitize their player piano rolls, which posed related technical challenges, so knitting-machine punch cards seemed like a challenge right up his alley.
When I asked Simon to describe his ideal digitization and preservation workflow for knitting machine punch cards, he said something that surprised me—that the encoded information preserved in magazines and books might be a better starting place than the punch cards themselves, depending on the goals of the project. It’s really hard to scan a punch card well. He pointed out that all sorts of things happen to physical punch cards that make them harder to digitize—they get bent or torn (and in the case of the player piano rolls he’s worked on, people repair and modify them in a variety of ways)—all of which are interesting material information about use, but which pose challenges for computer vision. The question of what to do with a hole that has been taped over is not only a creative decision, but also a technical one: will the scan be able to capture that? Do we introduce a new character to represent the tape in the encoding? Not that magazines are foolproof, he stresses—there are plenty of challenges in digitizing shiny paper, especially if one is trying to do it quickly or automatically.
Regardless of source material, Simon stresses the importance of high-quality scans: “From the point of view of posterity: the scan quality is really important—preserve it the best you can: things that are difficult to parse now will only get easier to parse in the future.”
Punch Card Encoding
Storing the parse—and circulating that information without having to repeat the process of either manual or computer-vision-assisted encoding—relies, at the moment, on community-supported infrastructure.
The format accepted by Brenda A. Bell’s generator, which generates SVGs for a given punch card style based on a user’s plain text file, has become one of the de facto encodings for this information as a .txt file encoded in ASCII—a way to archive and share punch cards that skips over the limitations of image-based archiving, even as it requires more upfront investment in labor. See image below for an example of what this looks like.
Text files are a lot smaller than images, and can be stored easily on both personal hard drives and cloud storage. There are many community-run Google Drives that act as repositories for these punch cards. As far as storing and circulating go, the ASCII format accepted by Bell’s generator offers a lot in terms of flexibility—allowing us to quickly remix, edit, and modify punch card patterns using lightweight, open-source software, even if the current format decontextualizes the information from its original conditions of use. Simon pointed out that a standardized metadata structure could do a lot there—maybe a standardized plain-text header—and I imagine what I could do with a corpus of punch card encoding linked to metadata about its provenance and digitization and to source images stored somewhere like the Internet Archive. What would we learn about knitting and textile history? What creative remixes would be possible?
Punch cards preserve the past and future
Knitting punch cards are an important part of any feminist computing history, and surprisingly resilient. They’re interoperable across machines with the same repeat, can be stored as physical (but still fundamentally digital) copies without worrying about hard drives going bad or requiring ongoing power consumption, and are also, in the age of seemingly-endless proprietary software and terms and conditions, refreshingly punk, in a minimal computing, open-source sort of way. How many people actually read the source code of the open-source software they use? Punch cards are the source, in something so fundamentally binary that fluency is not hard to come by. (Fluency in binary for almost all other tasks is nearly impossible.) I can repeat a row as many times as I wish. I can change whether my machine ignores the 1s, knits the 1s, purls the 1s, etc. I can perform subsequent operations on the punch card’s outputs with manual manipulation. And I own it. I own my knitting machine, can take it apart and repair it without violating some terms of service, and can hack and modify it and my punch cards to my heart’s content.
In a dream world, we’d have naming conventions or databases that let us link the .txt files to their corresponding stored images, in a system that balances the practicalities of storage and future use with the incredibly rich history available to us in the images. Punch card archiving supports an active, developing space where folx continue to develop computational and coding expertise in a variety of formats and ways—from working with mathematical modeling software to generate new punch cards to working out new designs with a hole punch and the memory cartridges at their machine. Our digitization and archiving practices can help us better understand the history of computing at the same time as they support an ongoing community working in creative computation. The Internet Archive and other community archives—which Simon says “are our best hope against enclosure”—don’t only preserve history, they enable communities to continue using and developing our technological resources.
About the author
Nichole Misako Nomura has a PhD from Stanford in English and an MA in Education, and studies digital humanities pedagogy. She’s currently an Associate Director at the Stanford Literary Lab, a digital-humanities research collective, and a lecturer in the Stanford Department of English.
