LiquidText

By Craig Tashman.

I’ve found most academic HCI work on augmented reading pretty disappointing: a lot of “minimum viable paper” type stuff, often attempting to play out one idea, with just enough fidelity to do a user evaluation for a paper, but Academic software interfaces generally aren’t suitable for serious use|not enough for serious use.

Reflections on LiquidText

As part of Constructing richer “readings”, I tried using LiquidText to actively read two papers, and to construct “posters” in the workspace representing my read of those texts. Immediately before this exercise, I performed a similar task by hand on printed versions of those papers, using ink, scissors and tape. See 2026-09-15 - Making a poster for Creative Reading with LiquidText and 2026-09-17 Voice Memo - Active reading and its discontents - LiquidText poster for full voice notes; summary and synthesis here.

• The biggest problem I experienced was a pervasive flattening of visual texture.
  • Text extracts are re-rendered so that they can be resized. We lose cues from the paper’s typography and line-level layout.
  • This also means that if you excerpt from two different papers into the same workspace, you don’t get visual cues to distinguish the papers—instead, you get a (non-customizable and ugly?) textual footer label.
  • Ink intersecting text extracts is left behind. In my paper posters, I used multi-level annotation to e.g. highlight and underline parts of a phrase, then extracted a broader segment. Those multi-level cues make the resulting layout much easier to parse visually.
  • It’s possible to add ink to an extract in the workspace, but the ink attaches to the extract rectangle, not the text, so it’ll lose its meaning if the extract is resized.
  • If part of the extract has a text highlight, that doesn’t come along quite as-is; instead, the extract gets a very bright border treatment. This feels lossy, and the visual hierarchy inappropriate.
  • If I’d written a comment on the text being excerpted, that comment is also left behind. This surprised me! It made excerpting very fiddly.
  • By default, my own text comments use identical typography to extracts from the paper’s. So there’s no visual distinction between my words and the authors’.
  • You can change this, but the controls are fiddly, require many taps, and don’t express built-in opinions about these distinctions.
  • If I draw an arrow connecting two excerpts, the arrowhead is removed: the model can only represent lines.
  • Likewise, on paper I’d sometimes write a comment, then draw an arrow from it to part of an extract (which I’d underline). No way to do this in LiquidText.
  • And on the paper I sometimes drew an arrow across several excerpts, linking them all together, as pointing to one destination. Can’t represent that either.
  • Ink on the workspace itself feels second-class. I tried labeling clusters of excerpts with handwritten headings, but there’s no way to make those headings “group with” the clusters. Text box headings can be created and grouped with the clusters, but it’s so fiddly that I found myself not wanting to bother.
  • Likewise, highlighter ink and “highlighted text” are treated differently in the document itself. The former doesn’t show up in the mini map, and doesn’t affect the edge banding of excerpts.
  • The result of all this is that the resulting “poster” created in LT feels much less expressive, is much harder to scan and read hierarchically, etc. This makes it less practically useful, but it also makes me feel less emotionally connected to the output, and feel less satisfied with my work.
• The base aesthetic experience of highlighting text and handwriting remarks is much, much worse digitally, even on an iPad Pro 13” with an Apple Pencil Pro. Lots of palm rejection issues; the screen is weirdly a little warm on my palm; my handwriting ends up messier. The paper experience was much more physically and viscerally enjoyable.
• I found myself distracted by a lot of basic issues with craft and execution: long hangs, a crash, interactions and controls that don’t follow the standard behaviors of the platform, etc. I found myself thinking: “This is just not software I would want to rely on.”
• LT was significantly better than the paper poster process in several ways:
  • Excerpting passages is way, way faster than with scissors. I particularly like that (via AutoExcerpt) I can create an excerpt on the workspace by direct interaction with part of the paper. (Also a big principle in Latticework)
  • But AutoExcerpt just accretes vertically in a 1D column, which quickly falls of the bottom of the screen and doesn’t make good use of the two-dimensionality of the workspace. When I switched from “accretion mode” to “sensemaking mode”, I felt a big ugh around dealing with this big 1D column, which forced me to scroll and zoom around unpleasantly.
  • Entering text would have been much faster than handwriting (if I hadn’t encountered lots of weird keyboard bugs).
  • Excerpts have backlinks! That’s a categorical difference, not just an efficiency.
  • I can continue to manipulate and iterate on the workspace, unlike my posters, which are made of tape and ink.
  • Convenient storage and retrieval: I can put this work away without dismantling it.
  • I can more easily share this work with others.
• The 13” screen did not feel as restrictive as I’d expected. For processing 8-10 page papers, it seemed to be fine.
  • Working with multiple papers simultaneously feels quite cumbersome and constrained to me, though.
• I notice (per Nio Ono) that I find myself increasingly uninterested in using software that I can’t reshape as I please with coding agents.

If I ignore my complaints about the base aesthetic experience of working digitally, I think LiquidText delivers a lot of the foundational elements I find myself wanting in Constructing richer “readings”. One angle would be to try to make a richer, more expressive, and much more malleable version of this system. But that still assumes a pure-digital workflow, which I’m not ready to sign up for.

Maybe it’s better to think about a system like LT as an output for something like BookBridge: you do your primary reading and sensemaking with physical materials, and the digital canvas exists for retrieval and follow-up—a secondary surface, rather than a primary one. But I don’t necessarily want to think about this as a one-way operation: ideally I could move between physical and digital surfaces repeatedly without costly overhead.