From Knowledge Represented in Frame-Based Languages to Declarative Representation and Reasoning via ASP

In this paper we encode some of the reasoning methods used in frame based knowledge representation languages in answer set programming (ASP). In particular, we show how “cloning” and “unification” in frame based systems can be encoded in ASP. We then show how some of the types of queries with respect to a biological knowledge base can be encoded using our methodology. We also provide insight on how the reasoning can be done more efficiently when dealing with a huge knowledge base. Introduction The broader goal of our proposed research is to be able to answer various kinds of questions with respect to knowledge bases constructed by domain experts in particular domains. Towards this effort we noticed that a large body of such knowledge bases have been developed using frame based representations (Fikes and Kehler 1985)1. Examples of such knowledge bases include AURA (Chaudhri et al. 2009), EcoCyc (Karp et al. 2002) and RiboWeb (Altman et al. 1999). A University of Texas site2 links to a large number of these knowledge bases. While the terms such as “object” and “oriented” are reported to be first mentioned in late 1950s and early 1960’s at MIT, and the terms “object” and “instance” were used by Sutherland in 1960-61, the origin of frame based knowledge representation is often attributed to Minsky’s 1975 work (Minsky 1975) . In frame based KR, knowledge is organized in chunks referred to as frames. Frames have slots which can be filled with values or other frames. There are hierarchies of frames and one frame may inherit the properties of another frame based on this hierarchy. Most frame based formalisms include some procedurally expressed knowledge, such as specification regarding how frames should be used and how inheritance should be addressed. Thus, semantics of frames are often specified operationally. Because of inbuilt features, such as inheritance, frames allow concise representation of knowledge. So, despite the statement in Hayes (Hayes 1979) that said “most of frames is just a new syntax for parts of first order logic”, to capture some of the Copyright c © 2012, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. See also http://www.cs.man.ac.uk/∼stevensr/onto/node14.html. http://www.cs.utexas.edu/∼mfkb/related.html semantics of frames, such as inheritance, one needs additional axiomatization. This is also true with respect to Descriptions logics (DLs) (Baader et al. 2003), which although capture many of the declarative aspects of frames do not automatically capture the procedural, operational and inbuilt aspects. They (DLs) do have much richer mechanism to define new classes. There have also been many proposals that combine frame based knowledge representation aspects with rule based knowledge representation aspects. Example of such formalisms are DLV+ (Ricca and Leone 2007) and FAS (Alviano et al. 2008). Some of the operational and inbuilt aspects of frames have been logically formalized. The most common aspect being the formalization of inheritance, logical formalization of which has been given by many (Touretzky 1986; Horty 1994). Of particular note is the work by Brewka (Brewka 1987) that discuses the logic of inheritance in frame systems. However, these formalisms also do not address some of the procedural reasoning mechanisms in many frame based systems. In particular, none of them address the issue of “cloning” and “unification” that we discuss in the following paragraphs. In this paper, while working towards our goal of answering various kinds of questions with respect to knowledge bases constructed by domain experts in particular domains, we first formalize the notion of cloning and unification that were earlier presented in an operational manner, and not precisely enough, in the frame based systems KM (Clark, Porter, and Works 2004)3 and AURA (Chaudhri et al. 2009) 4. However, these notions make it easier for multiple domain experts to work in parallel in representing knowledge using frames. They also make the representation compact. Some basic ideas and motivation behind them is as follows. Suppose a group of domain experts were to describe the knowledge about cells in a knowledge base as they go through the chapters in a book about cells. In a frame based approach each may consider describing a prototype cell with various slots (for example, parts-of) and values. Many of those values, such as the various parts of cells (for example, nucleus, KM has been adopted by various projects, notably Project Halo (Gunning and et 2010). AURA is also developed under Project Halo. 413 Proceedings of the Thirteenth International Conference on Principles of Knowledge Representation and Reasoning