NOV 15th 2007

The Curse of Knowledge and the Semantic WebThe Curse of Knowledge: the more you know, the more difficult it is for you to communicate knowledge. When we know something, we can hardly imagine not knowing it. The more we learn about something, the more it becomes even harder for us to think of not knowing it. It is generally difficult for experts (who know much) to explain their expertise to laymen (who know little) because experts have to try hard to imagine the scenario when they were not experts. This is the Curse of Knowledge.

There are many examples of the Curse of Knowledge in the book Made to Stick: Why Some Ideas Survive and Others Die, written by Chip and Dan Heath. One famous example is the experiment of "tappers" and "listeners." A tapper heard a well-known song, such as "Happy Birthday to You" or "The Star-Spangled Banner," and then he tapped out the rhythm to a listener by knocking on a table. After hearing the rhythm being tapped, a listener was then asked to tell the name of the song. According to the book, after repeating this experiment many times the tappers believed that the listeners should have correctly guessed at least 50% of their tapped songs. In reality, however, the listeners only guessed 2.5% songs correctly.

This is a fantastic demonstration of the Curse of Knowledge. The tappers knew the song when they were tapping (so they were like experts), while the listeners knew nothing prior to hearing the tapping (so they were like laymen). The tappers had failed to deliver what they believed to be easy to the listeners. It was the Curse of Knowledge that produced the significant gap between 50% and 2.5%. When we think that we have succeeded in explaining our knowledge, we in fact did not.

The Curse of Knowledge on the Semantic Web

The Semantic Web is engaged with expert-specified knowledge. An interesting question is whether we will suffer the Curse of Knowledge on the Semantic Web. We may have to suffer this curse severely in the early stages of the Semantic Web. With the evolution of the Semantic Web, however, we may gradually defeat this curse.

From the technology point of view, the Curse of Knowledge on the Semantic Web is reflected by the difficulty of ontology mapping. Basically, experts often have ontologies full of professional constructs in their mind about their domain of expertise. These professional ontologies are well sharable among peer experts, but are usually less accessible by laymen. In contrast, laymen also have their unprofessional descriptions of the respective domains, i.e. the layman requests. To construct mappings between expert ontologies and layman requests is the task of ontology mapping. This ontology mapping problem is a long-term difficult problem in the realm of knowledge management.

If ontology mapping is a reflection of the Curse of Knowledge, we may solve the difficult ontology mapping problem by breaking the Curse of Knowledge. The most important way to break the Curse of Knowledge is to present simple ideas. There are several suggestions for being simple in the book Made to Stick. In short, being simple means finding and sharing the core and only the core.

Write down the Commander's Intent instead of the commands

In the real life, we often cannot follow the details of commands due to the change of context. For example, we are asked to borrow the book Made to Stick from the city library, but occasionally all of copies of the book in the library have been checked out, so we are not able to accomplish this command. However, may we still achieve the intent of this command by ignoring the details of the command itself? Certainly we may. For example, if the intent is the book, we can go to a book store to buy a copy. If, however, the intent is to borrow something to read, we can certainly try to borrow another interesting book to read instead of the specified book title.

This solution brings hints for domain experts to build ontologies on the Semantic Web. If only they construct the intent of their domains instead of the details of their domains into their ontologies, we may reduce the difficulty of ontology mapping to the least.

Determine the single most important thing

In real life, we often have multiple purposes (explicit and implicit) in a domain description. Back to our previous example, we may expect to read the book Made to Stick AND borrow the book instead of buying a copy. So we have two intents in one command (one is the book and the other is to borrow), but two intents are confusing. In particular, it causes more difficulty for mapping when the specified context ("in the city library") is not held. With two potential intents, we do not know the priority of the mapping.

This common problem for describing a domain is an evil enemy to ontology mapping. Domain experts often like to produce "complete" ontologies when describing their domains. These ontologies often contain many concepts and, thus, their focuses become fuzzy. Mapping these ontologies then becomes extremely difficult because it requires very complicated concept and relationship set disambiguation algorithms performed in a large scale. Moreover, it also often requires numerous pre-produced domain-specific heuristics to support the mappings.

To solve this problem, the hint from this second solution to the Curse of Knowledge is that we need only small but focused domain ontologies on the Semantic Web. Large ontologies cause more confusion in practice than well organization of knowledge. Every domain ontology should express one and only one intent. In addition to the intent, we may allow users to fill in the details of the intent by specifying personalized epistemological extension of their requests. As the result, the practice of single intent minimizes the burden of ontology mapping by only focusing on the core; and the epistemological extensions preserves the variety of personal preferences.

Don't bury the lead

The lead section of an article is its spirit; and it should always appear at the beginning of the article. One consequence of the Curse of Knowledge is that we often bury this spirit inside the article so that readers are lost. Writers who are domain experts often start to discuss what they believe to be cool and exciting but forget telling the layman readers what the center of the article is. This is called burying of the lead.

Burying the lead is another normal problem for ontology mapping. When we map two ontologies, we need to determine whether they can be mapped, or if they can then which concept is the core. Without explicit specifications of the core of ontologies, it often takes machines much time to figure out answers to these simple and essential questions.

A straightforward method can solve this problem of burying the lead. If every ontology developer explicitly labels the core concept in domain ontologies, the answers to the previous questions become straightforward. This solution is the same as asking writers to explicitly put their lead section always at the beginning. By this simple technique, the process of ontology mapping could be greatly simplified.


In this post, we have introduced a core concept of the Semantic Web - ontology mapping - through a casual discussion of the Curse of Knowledge. Although the Curse of Knowledge is not necessarily the only reason that causes the difficulty of ontology mapping, it's inevitably a major reason. We suffer from the Curse of Knowledge in our everyday life. Almost all the misunderstanding between humans is more or less caused by the Curse of Knowledge: I think of things in this way (because of my background of knowledge), which happens to not be the way you think (because of your background of knowledge). Content on the Web inevitably inherits this knowledge gap and this inheritance causes the severe problem of ontology mapping.

Once we identify the cause, we can design responding methods. Since the Curse of Knowledge is a major reason for ontology mapping on the Semantic Web, we may try to solve this problem by breaking the Curse of Knowledge. By breaking this curse, we may solve the problem of ontology mapping in reality more easily than trying to exploit the complex algorithms of computer science.

For the sake of demonstration, I only listed three methods for breaking the Curse of Knowledge. Readers who are interested in this topic could certainly continue on this path and see how the other methods can help improve the performance of ontology mapping.

About the author

Yihong Ding

I'm currently a Ph.D candidate in Brigham Young University with Prof. David W. Embley in Computer Science Department, Prof. Deryle W. Lonsdale in Linguistic Department, and Prof. Stephen W. Liddle in Marriott School of Management.

Trackback URL for this entry:

Spam protection by Akismet

Comments for this entry:

No one has left a comment for this entry. Be the first!

Post a comment

  1. Spam protection by Akismet