This is the third in a series of articles introducing Kards. The first uncovered what we believe is the fundamental problem with current note-taking apps. The second explained our approach by telling the story of how Carl Linnaeus invented the index card.
The most critical aspect of Kards' interface is the knowledge card, or kard for short. The quickest way to explain a kard is to say that it is the digital version of an index card. There are many different ways to use an index card, but the breakthrough performance achievable with Kards depends on using them like Linnaeus did.
Just as knowledge cards represent the most important aspect of Kards, labels represent the most important aspect of a kard. Most of Kards' innovations — and there are plenty — can be understood as revolving around kards and their labels.
Each kard is labeled in plain English. Kards then applies natural language analysis to these labels in order to integrate the contents of your knowledge cards into a browsable book. However, any approach that would result in a nonzero probability of miscategorizing an item isn't acceptable. And as we explained in the previous post, natural language analysis isn't 100% reliable in general. There are two things we did to get around this problem:
- Kards does not try to understand the contents of a knowledge card. Natural language analysis is only applied to the label. So we reduced the problem of applying natural language analysis from understanding an entire text to the understanding of only a fragment of a sentence.
- We managed to single out a subset of the English language that Kards can reliably make sense of.
Here is the basic idea. Consider the following example:
Sentences like this are known as is-a statements. A thing named “Invictus” is defined by what comes after “is a”. If you created a knowledge card using Kards, “wonderful history movie by Clint Eastwood recommended by Jürgen” would be the label of the card and “Invictus” would go into the body. What comes after “is a” is called a noun phrase by some linguists. Relax — you don't really need to know that in order to use Kards. We simply call it the natural language label.
Our example sentence pretty much explains what kind of expression Kards expects in the label field. Let's walk through it word by word. Using color helps to better see what's going on:
wonderful history movie by Clint Eastwood recommended by Jürgen
The item described here is categorized as a movie. It is not just any movie, it is a history movie. We refer to “history” in this context as an area because it denotes an area of knowledge rather than a category of things. For example, another knowledge card could talk about a history book. Kards organizes your things by category, e.g., movie or book, as well as by area, history in our example.
Our history movie is also described as wonderful. This assigns a particular attribute to our movie. Currently we only support attributes meant to articulate liking or disliking and associate them with ratings. Attributes provide yet another way to browse your knowledge. For example, if somebody asks for recommendations you can easily call up lists of things you liked.
There are two more aspects that deserve our attention. It is claimed that the movie was directed by Clint Eastwood and that a certain person whose name is Jürgen seems to have recommended it. Note that the label didn't mention the word “director” nor did it say that Jürgen is a person. Kards infers both facts using a feature which we call background knowledge. We'll explore the specifics of background knowledge in a future article.
Kards uses the director and recommender properties to provide you with a wealth of browsing options. Want to watch one of those entertaining but somewhat serious movies Jürgen recommends? Simply call up the list of “movies recommended by Jürgen”. Somebody asks you for a movie recommendation? Go to the list of “great movies watched” and make your choice. You could narrow down this list even further should your friend prefer certain genres, directors, actors, or subjects. Wait — we haven't yet discussed how to specify actors or subjects? Can you come up with a natural language label that would?
Let's see what a completely filled out kard looks like. Let's assume you looked up Invictus on the Internet Movie Database. Using our browser extensions it takes only one click to create the following kard:
What you see here has been automatically created by Kards analyzing the movie page for Invictus. Kards grabbed the movie poster, identified the title of the movie, retrieved a summary, and even suggested a label for the kard. The URL is also preserved so you can easily go to the original page where all this info was extracted from. You can edit any of this information before saving the kard. I often edit at least the label to make sure it accurately reflects why the knowledge item is interesting to me.
This already makes Kards a great movie app, even though that's not what it was specifically designed to be. Everything I've explained so far only makes use of Kards' general ability to deal with knowledge. It works just as well with books, quotes, recipes and web bookmarks, and those are just the most obvious applications that come to mind. Give it some thought, and the possibilities seem endless.
This article introduced a key idea of Kards by discussing a single example only. But there is much more to say in order for the whole depth of our new approach to become apparent. Stay tuned!Table of Contents