Brief review of the Handbook of Knowledge Representation

The new Handbook of Knowledge Representation edited by Frank van Harmelen, Vladimir Lifschitz and Bruce Porter [1] is an important addition to the body of reference and survey literature. The 25 chapters cover the main areas in Knowledge Representation (KR), ranging from basic KR, such as SAT solvers, Description Logics, Constraint Programming, and Belief Revision, to specific core domains of knowledge, such as Spatial and Temporal KR & R, and Nonmonotonic Reasoning, to shorter ‘application’ chapters that touch upon the Semantic Web, Question Answering, Cognitive Robotics, and Automated Planning.

Each chapter roughly follows the approach of charting the motivation and problems the research area attempts to solve, the major developments in the area over the past 25 years, important achievements in the research, and where there is still work to do. In a way, each chapter is a structured ‘annotated bibliography’—many chapters have about 150-250 references each—that serve as an introduction and a high-level overview. This is useful, for instance, if your specific interests are not covered in a university course but have a thesis student and you would want him to work on that topic, then the appropriate chapter will be informative for the student not only to get an idea about it but also to have an entry point as to which further principal background literature to read; or you are a researcher writing a paper and do not want to put a Wikipedia URL in the references (yes, I’ve seen papers where authors had done that) but a proper reference; or you are, say, well-versed in DL-based reasoners, but come across a paper where one based on constraint programming is proposed and you want to have a quick reference to check what CP is about without ploughing through the handbook on constraint programming. Comparatively with the other topics, anyone interested in ‘something about time’ will be satisfied with the four chapters on temporal KR & R, situation calculus, event calculus, and temporal action logics. Clearly, the chapters in the handbook on KR are not substitutes for the corresponding “handbook on [topic-x]” books, but they do provide a good introduction and overview.

Some chapters are denser in providing a detailed overview than others (e.g., qualitative spatial reasoning vs. CP, respectively), however, and yet other chapters provide a predominantly text-based overview whereas others do include formalisms with precise definitions, other axioms, and theorems (Qualitative Modelling, Physical Reasoning, and Knowledge Engineering vs. most others, respectively). That most chapters do include some logic comes as no surprise for the KR researcher but may be for the novice or a searching ontology engineer. For the latter group, and logic-sceptics in general, there is a juicy section in chapter 1, “General Methods in Knowledge Representation and Reasoning”, called “Suitability of Logic for Knowledge Representation” that takes on the principal anti-logicist arguments and the about 6-page long rebuttal of each complaint. Another section that can be good for heated debates is Guus Schreiber’s (too) brief comment on the difference between “Ontologies and Data Models” (chapter 25), which easily can fill a few pages instead of the now less than half a page used for arguing there is a distinction between the two.

Although I warmly recommend the handbook as addition to the library, there are also a few shortcomings. One may have to do with the space limitations (even though the book is already over 1000 pages), whereas the other one might be due to the characteristics of research in KR & R itself (to some extent at least). They overlap with the kind of shortcomings Erik Sandewall has mentioned in his review of the handbook. Several topics that are grouped under KR are not, or very minimally, dealt with in the book (e.g., uncertainty and ontologies, respectively) or in a fragmented, isolated, way across chapters what perhaps should have been consolidated into a separate chapter (i.e., abstraction, but also ontologies). In addition, within the chapters, it may well occur that some subtopics are perceived to be missing from the overview or mentioned too briefly in passing (e.g., mereology and DL-Lite for scalable reasoning), but this also depends on one’s background. On the other hand, the chapters on Qualitative Modelling and Physical Reasoning could have been merged into one chapter.

The other point concerns the lack of elaboration on real life success stories as significant contribution of that topic that a KR novice or a specialised researcher venturing in another sub-topic may be looking for. However, the handbook charts the research progress in the respective fields, not the knowledge transfer from KR research output to the engineering areas where the theory is put to the test and implementations are tried out. It is a philosophical debate if doing science in KR should include testing one’s theories. To give an idea about this discrepancy, part III of the handbook is called “Knowledge Representation in Applications” (emphasis added), which contains a chapter, among five others, on “The Semantic Web: Webizing Knowledge Representation”. From a user perspective, including software engineers and the most active domain expert adopters (in biology and medicine), the Semantic Web is still largely a vision, but not yet a success story of applications—people experiment with implementations, but the fact that there are people willing to give it a try does not imply it is a success from their point of view. Put differently, it says more about the point of view of KR&R that it is already categorised under applications. True, as the editors note, one needs to build upon advances achieved in the base areas surveyed in parts I and II, but is it really ‘merely’ ‘applying’, or does the required linking of the different KR topics in these application areas bring about new questions and perhaps even solutions to the base KR topics? The six chapters in part III differ in the answer to this question—as in any healthy research field: there are many accomplishments, but much remains to be done.

[1] Frank van Harmelen, Vladimir Lifschitz and Bruce Porter (Eds.). Handbook of Knowledge Representation. Elsevier, 2008, 1034p. ISBN-13: 978-0-444-52211-5 / ISBN-10: 0-444-52211-5.

Advertisements

4 responses to “Brief review of the Handbook of Knowledge Representation

  1. “It is a philosophical debate if doing science in KR should include testing one’s theories” . How on earth could you be doing “science” without bothering to test your theories? I guess it just comes down to a question of what you mean by the word “science”. If you are doing research as a Mathematician – which I suppose is mostly what you are doing when you do research in e.g. logic and reasoning – a completed proof is probably a satisfactory result. But knowledge representation IMHO is too closely tied to squishy human knowledge and to specific purposes for KR research to escape the need for real-world experiments.

    • Hi Ben,

      you’re on the right track with the notion of ‘proofs suffice’, but also, as you say, there are arguments that that does not quite hold. More precisely, I did some additional reading following the post on philosophy and CS and came across “three paradigms of computer science” [1] that discusses this issue at some length, but that I did not want to descend into in the book review. The three paradigms Eden identifies are the rationalist one, technocratic, and scientific, where he argues that the former (common among theoretical computer scientitst) is by now outdated, the latter the only viable one and the technocratic one bad for CS. However, this is one paper, and I need to look into it a bit more–it does not seem that the discussion is closed already.

      [1] Amnon H. Eden. 2007. Three paradigms of computer science. Minds & Machines, 17: 135-167.

      cheers,
      marijke

    • The official answer to Ben’s comment might be that it is an engineering discipline and not a science. This does not necessarily make it squishy. There are reproducible aspects once you agree on certain parameters.

      On the other hand, it is susceptible of a certain amount of scientific verification, and possible to measure progress. For example, Description Logics were a discovery of a certain class of logics for which certain computations became tractable vs. KL1 and had a more precise mode of expression than mere frames. This is just one of the many comparisons one can make between formalisms.

      It’s an exciting field and I thank you for this helpful review!

  2. Keet wrote: does the required linking of the different KR topics in these application areas bring about new questions and perhaps even solutions to the base KR topics?

    Yes.

    Hopefully, we will see new chapters in a few more years 🙂

    Your point in the last paragraph is similar to Sandewal’s ultimate point:
    not enough examples.

    I agree.

    One would think that examples would be the heart of the matter…
    quite apart from the science of logic undergirding it all.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s