A Lexical Knowledge Representation Model for Natural Language Understanding

Knowledge representation is essential for semantics modeling and intelligent information processing. For decades researchers have proposed many knowledge representation techniques. However, it is a daunting problem how to capture deep semantic information effectively and support the construction of a large-scale knowledge base efficiently. This paper describes a new knowledge representation model, SenseNet, which provides semantic support for commonsense reasoning and natural language processing. SenseNet is formalized with a Hidden Markov Model. An inference algorithm is proposed to simulate human-like natural language understanding procedure. A new measurement, confidence, is introduced to facilitate the natural language understanding. The authors present a detailed case study of applying SenseNet to retrieving compensation information from company proxy filings. DOI: 10.4018/jssci.2009062502 International Journal of Software Science and Computational Intelligence, 1(4), 14-31, October-December 2009 15 Copyright © 2009, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. With one-dimensional natural languages used by human being, in order to understand and describe a highly dimensional environment a series of filtering and transformations are necessary as illustrated in Figure 1. These transformations can be N-dimensional to N-dimensional or one-dimensional to N-dimensional in input process, and N-dimensional to one-dimensional or N-dimensional to N-dimensional in an output process. After these transformations information should be ready to be used by the central processing unit directly. Effectiveness and efficiency of these transformations are very important to knowledge representation and management. A knowledge model describes structure and other properties of a knowledge base which is part of a central processing system. A knowledge representation model is simply a mirror of our world, since one important requirement for a model is its accuracy. In this sense there is hardly any intelligence in a knowledge model or a knowledge base. Instead it is the communication process consisting of filtering and transformations that shows more intelligent behaviors. As expressed by Robert C. Berwick, et al., in a white paper of MIT Genesis project (Berwick, et. al., 2004), “The intelligence is in the I/O”. As shown in Figure 1, a knowledge model may be the easiest component to start since its input has been filtered and transformed tremendously from the original format, and is ready to be stored in the knowledge base directly. On the other hand, a knowledge representation (KR) model plays a central role to any knowledge-based systems, and it eventually decides how far such a system can go. Furthermore, knowledge and experience can make the process of filtering and transformations more efficient and effective. A KR model captures the properties of real world entities and their relationships. Enormous amounts of intervened entities constitute a highly complex multi-dimensional structure. Thus a KR method needs powerful expressiveness to model such information. Many cognitive models of knowledge representation have been proposed in cognitive informatics. Several cognitive models are discussed in (Wang & Wang, 2006). ObjectAttribute-Relation model is proposed to represent the formal information and knowledge structures acquired and learned in the brain (Wang, 2007). This model explores several interesting physical and physiological aspects of brain learning and gives a plausible estimation of human memory capability. The cognitive foundations and processes of consciousness and attention are critical to cognitive informatics. How abstract consciousness is generated by physical and physiological organs are discussed in (Wang & Wang 2008). A nested cognitive model to explain the process of reading Chinese characters is presented in (Zheng, et. al., 2008), which indicates that there are two distinctive pathways in reading Chinese characters, and Figure 1. Communication process for a knowledge-based system 16 International Journal of Software Science and Computational Intelligence, 1(4), 14-31, October-December 2009 Copyright © 2009, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. this can be employed to build reading models. Visual semantic algebra (VSA), a new form of denotational mathematics, is presented for abstract visual object and architecture ma-visual object and architecture manipulation (Wang, 2008). VSA can serve as a powerful man-machine interactive language for representing and manipulating visual geometrical objects in computational intelligence systems. In Artificail Intelligence many KR techniques have been proposed since 1960’s, such as semantic network, frame, scripts, logic rules etc. However, we still know little about how to capture deep semantic information effectively and support the construction of a large-scale commonsense knowledge base efficiently. Previous research focuses more on the expressiveness of KR. Recently there is an emerging interest of how to construct a large-scale knowledge base efficiently. In this paper we present a new KR model, SenseNet, which provides semantic support for commonsense reasoning and natural language understanding.