Data processing: artificial intelligence – Knowledge processing system – Knowledge representation and reasoning technique
Reexamination Certificate
1998-05-28
2004-08-17
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Knowledge processing system
Knowledge representation and reasoning technique
C706S054000
Reexamination Certificate
active
06778970
ABSTRACT:
CROSS-REFERENCE TO RELATED APPLICATIONS
See attached FORM PTO-1449.
STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT
Not Applicable.
BACKGROUND OF THE INVENTION
Models or representations of general-purpose intentions (motivations for activity) have rarely emerged from commercial computer systems; in prior art, the existence of intentions in commercial computer systems has generally been extremely limited. Typically, computer systems contain specific intentions which are limited in scope, such as the intention of responding to direct commands, and running scheduled background tasks, such as disk storage backup programs and polling of input buffers. Consequently, most of the processing power of computer systems is wasted because either there is no direct command to perform or there is no scheduled task to perform.
This idle processor power could be utilized if a general-purpose computational intention could be defined to be performed at all times by a new type of computer system. In such a system, direct commands and scheduled tasks could still be performed at an appropriate level of priority, but idle processor power could be used to perform tasks supporting the general-purpose intention. The present invention describes this new type of computer system using a model of intentionality which seeks to semantically analyze conversational input, and to optimize that analysis for maximum semantic efficiency. The efficiency and speed of that analysis is designed to mimic human capabilities to process information.
Human capabilities to process information include the rapid recognition of semantic meaning amid streams of noisy speech input, as well as rapid responses to speech input which permit humans to conduct conversations which touch many levels of abstraction in just a few phrases. For instance, the following conversation (from Pinker page 227) shows the significance of deeply abstract meanings in normal conversation:
Woman: I'm leaving you.
Man: Who is he?
Normal human conversation contains so many levels of abstraction that prior art in human-computer interfaces has devised extremely specific information filters, such as graphical user interfaces, to simplify these abstractions to the point where they can be reliably mapped by a non-semantic computer program. Even so, the difficulty of mapping between computer systems and the complex semantics of human conversation cannot be avoided since meanings contained in the semantics of human conversation are often fundamental to the driving motivations for creating computer systems; the specification process of computer systems design takes place primarily through person-to-person semantic discussions of systems requirements. In most systems, the translation from these semantic discussions to actual non-semantic computer code loses a great deal of meaning; to make up for this lost meaning a profusion of comments are imbedded deep in the computer code and volumes of documentation are written to help users of the computer system. If the semantic intentions of the designers could be directly represented in computer systems, semantic computer systems would subsume the traditional comments and documentation, providing a unified and more meaningful interface to users of the system.
Semantic architectures for computer systems have been researched for many years, but they have not been able to supplant traditional non-semantic computer architectures, except in areas which are specifically semantic in nature such as speech recognition and natural language text translation. Meanings in semantic networks are represented by a myriad number of links, links which are difficult to interpret without the new structuring principles introduced by the present invention. Traditional non-semantic computer architectures can be structured by efficient hierarchies such as a data-dictionaries or function invocation trees. The efficiency and succinctness of these non-semantic computer architectures is what makes them more useful than prior semantic architectures. One of the purposes of the current invention is to identify and promote efficiency within semantic networks, by analyzing and converging topological characteristics of semantic networks toward a most efficient ideal form.
Ideal relationships between meaning and efficiency have long been observed. As far back as 1956, Miller published
The magical number seven, plus or minus two: Some limits on our capacity for processing information
. In that paper, he touched on themes which would resound in paper after paper in the following decades: that the amount of information which can been efficiently handled in short term memory is seven plus or minus two chunks. People understand information more easily when it is grouped into seven plus or minus two chunks. Information which is not in such a grouping-format can be re-grouped into groupings of seven plus or minus two chunks, through re-coding or other means such as re-classification. This re-grouping also helps people to understand information more easily. One of the principles of the present invention is to optimize semantic network inheritance links according to Miller's capacity number, so that preference is given to representations consisting of nodes with a nearly constant number of direct inheritors, such as seven, or as described later in this invention, five direct inheritors. The smaller number of direct inheritors such as five allow other attributes such as inherited nodes and the name of the node itself to fit more comfortably within its group.
While traversing a semantic network, there are two extreme topologies which can cause that traversal to be tedious. In one extreme, nodes have few branches, spreading the population of nodes across a narrow and deep hierarchy of nodes. In the other extreme, nodes have too many branches, and although the population of nodes is clustered near the top, it is difficult to choose which branch to follow. By applying topological transformations, so that all nodes in a representation have nearly five direct inheritor branches, a balance can be maintained which avoids the two extremes. After transformation, the typical inheritance level traversed abstractward covers five times as many inheritor nodes as the previous level; excessive hierarchic depth is prevented for any population of inheritors by the efficiency of this coverage. Since the typical transformed node has no more than five branches, traversal logic can deal with just five possible choices at a time, making that logic quick and simple. The result of the transformations is a semantic network which is efficient to traverse and thus efficient relative to actual use.
When traversing a semantic network, another problem is nodes which have directly linked inheritors of differing levels of abstraction; it is difficult to compare such siblings. For example, a node ‘plant’ might have siblings of ‘tree’ (very abstract) and ‘poison ivy in my backyard’ (very concrete). These siblings don't really belong at the same level in a semantic network; their differing abstractness makes them unsuitable for most semantic comparisons. The present invention tracks the differences in sibling abstractness, averaging those deviations across each subtree, so that alternative subtrees with less sibling deviations in abstractness can be selected. These subtrees are more efficient as well, since their leaf nodes are all at the same level, rather than spread at different levels of abstraction.
The present invention uses both the preferred number of inheritor branches (such as five) and the least deviations in sibling abstractness, to guide choices between alternative semantic representations, thus favoring the representations with the most efficient overall topology. In this way, the advantage of efficiency which exists in well-designed non-semantic systems can be created in semantic systems as well. The analysis process of systems design can be described as a progression toward ever more efficient semantic representations, leading to a system with acce
LandOfFree
Topological methods to organize semantic network data flows... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Topological methods to organize semantic network data flows..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Topological methods to organize semantic network data flows... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3291998