2018-10-17
(last updated 2019-02-22)
I don’t remember the exact sequence of events, but a couple of months ago, I became aware of:
For years, I’ve liked the idea of being able to easily mark passages in books I read and have them trivially resurfaced later, but Kindle’s locked-down ecosystem has made this difficult. When I came across Readwise, and read the other two articles above, I was suddenly motivated to build a better reading workflow of my own.
According to Readwise, there are three fundamental steps in any good reading workflow: Capture, Review, and Integrate. Simon breaks this down differently, but the Readwise categorization resonates better with me. So here’s my reading workflow, phase-by-phase:
Capture: Read in Kindle, Instapaper, dead-tree books, or PDF. In all but the latter case, use Readwise to aggregate. In the case of PDF, a custom script rips annotations from PDFs in a special location in my iCloud Drive.
Review: For physical books, use OCR and paste into Readwise’s manual entry. Mass-export data from Readwise. Import this, as well as the PDF export, into Anki, with a special note type and template. Review these daily.
Integrate: When I encounter a highlight with information that I’d like to remember more explicitly, I use Anki’s “mark” feature. Periodically, I work through Marked items and convert them into fine-grained question-and-answer cards in Anki, then suspend the Highlight note.
In the Capture phase, I’m concerned with capturing highlights with the lowest overhead possible. Source by source:
Kindle: Just highlight.
Instapaper: If I’m reading something on the web longer than a few paragraphs, I generally send it to Instapaper and read it there, highlighting anything that I’d like to be reminded of.
PDF: I save these to a directory named “Highlighted PDFs” in my iCloud Drive, then (on iOS) open them from the “Files” app. Unlike iBooks on iOS, Files supports highlighting and annotating.
Physical Books: Generally I just draw a line in the margin next to the interesting part, then scribble a bit on the lower right corner of the right page. This makes flipping through to find all the marked up pages much easier later on.
The meat of my “Review” phase is simply opening Anki every day and reading a group of highlights, presented at increasing intervals using Spaced Repetition, but there’s some menial labour I have to do when adding new highlights before I can do that. Again, this part is broken down by format.
Kindle: Readwise can import highlights from Kindle,
but it requires me to actually open a browser window so that they can
scrape the Kindle site using a browser plugin. I run this each time I’ve
created new Kindle highlights, then I run a script
“kindle2anki” which downloads all of my Readwise highlights
and imports them into Anki (more detail on this shortly).
Instapaper: Readwise automatically synchronizes Instapaper highlights, and my same script reimports them to Anki.
PDF: I wrote another script that extracts highlights from all of the PDFs in this special iCloud Drive directory that I save them to. The output of this script is also picked up by my Anki import script. I don’t use this as often and I’m still ironing out bugs pretty much every time I use it.
Physical Books: I use the Readwise “Freeform Input” feature, but I sometimes OCR the text rather than typing it, using an App called CamScanner. Once again, this is picked up by my Readwise-to-Anki script. I’m still trying to figure out a good method for this. OCR feels a little better than retyping but it’s still clunky.
Basically, I have two (terrible, terrible quality) things going on:
There’s not much to say about the former script. I extract the document title and author from the PDF, and the text of each annotation (as well as any attached note), and dump them in JSON format.
The latter script uses the very-useful https://readwise.io/munger, which exports all of your highlights in one JSON file. It then dumps these fields in a tab-separated-value file, alongside a digest of a few fields, which I can use as a unique key to prevent duplication in case I change the note later on in Anki. I then import this file into Anki. Any cards with matching digest fields are ignored. If I ever want to “delete” a note, I suspend it instead of deleting it, to prevent my script from re-creating it.
I’ve created a Note type called “Highlight”, with the following fields:
I also have a “Highlights” deck. I enjoy reviewing these separately from the rest of my cards.
I made a few minor changes to the Study options for this deck:
The “Highlight” Note type has one Card type:
<div class="book source">{{Source}}</div>
<div class="book author">{{Author}}</div>
<div class="book">{{Highlight}}</div>
<div class="book note">{{Note}}</div>@import url("_style.anki.main.css");(see _style.anki.main.css)
{{FrontSide}}Now that these scripts are “done”, it only takes a few seconds to vacuum up all my new highlights into Anki (except for physical books). The actual review process is simply reviewing all the due cards each day. I typically have between 1 and 20 to glance through.
As I go through cards each day, I filter them into one of three buckets:
Every week or so, when I’m at a real keyboard, I’ll look at my marked cards, convert them into normal cards (Basic or Cloze), and un-mark them. These become integrated knowledge.
Highlights will tend to stay relatively fresh when reviewed in this way, and knowledge integrated into Q&A-style cards becomes permanent. The underpinning of this system is a commitment to reviewing all due cards in Anki each day, which luckily I do.
There it is in a nutshell. You may notice that this is a kind of incremental reading system, and I’m trying to figure out how much more of that concept to try to bring in. I’ll continue to update this post as I continue to evolve this system.
I spoke with Simon shortly after writing this post and we agreed that
the Zettelkasten
system may be an ideal way to manage this Capture/Review/Integrate
workflow. I tried this out for a while, first using Dynalist and later org-brain. I wrote an
org-brain-to-Anki importer, and reviewed captured knowledge
there on a fairly long interval.
I had a couple of problems with this method:
org-brain) felt unnatural: I spent too much time trying
to find even tangentially-related notes.I like most of the system I have now, but I’d love to have this all implemented as a series of importers in Perkeep rather than routing three quarters of it through readwise, and a script to generate an Anki import file from the contents of my Perkeep instance.