Back to results
US 2003/0191627 A1Published application
Topological methods to organize semantic network data flows for conversational applications
AssigneeNone recorded
InventorLAWRENCE AU
Life of this patent
Estimating the term…
Loading patent…
Loading patent…
AssigneeNone recorded
InventorLAWRENCE AU
Estimating the term…
A system and methods for enforcing uniform branching of node-to-node inheritance links within semantic networks, to control data flows to and from such networks in conversational applications. Enforced uniform branching criteria converge the population of directly connected nodes of each node toward a small system-wide constant, and converge each sibling inheritance node to a similar level of abstractness, and are also used to select the best candidate tree from a set of competing representation trees within the semantic network. Uniform branching criteria are applied to competing trees for speech recognition, for object recognition in vision systems, for concept recognition in text scanning systems, and for algorithm definition. For speech recognition, phonemes are identified and matched to dictionary nodes in the semantic network. For visual object recognition, object features are identified and matched. For text scanning, words are identified and matched. For speech, visual and text the sets of competing representation trees are formed from alternative combinations of matched dictionary nodes. For algorithms, competing sets consist of alternative trees of commands. Conversational data flows are directed into active conversations until those conversations comprise a targeted number of inheritor nodes. Active memory space reserves are reclaimed by archiving nodes. Conversational emotional states are categorized by shifts in node topologies of active conversations, to verify that conversations have successfully communicated information. To gather and disseminate information across distributed computer networks, conversational questions are forwarded and conversational answers gathered by moderator computers representing their network-subnet's computers.
1) In a semantic-network data-processing system, a method executed by a data processor for enforcing consistent levels of abstractions for sibling nodes within a semantic network, whereby the average abstractness for each set of sibling nodes is calculated, and abstractness deviations from said average abstractness for each individual sibling node are summed, and alternative inheritance tree representations are proposed, and the alternative tree representation is chosen which has the smallest total sum of abstractness deviation.
5) In a data-processing system, a method executed by a data processor for enforcing uniform branching of node-to-node inheritance links within a semantic network, whereby, for each node having some type or types of inheritance links, the populations of such links are converged to a branching constant, and for each node having such inheritance links, its deviation from said branching constant by its population of such links is summed into a branching deviation sum, and alternative inheritance tree representations are proposed, and the alternative tree representation is chosen which has the smallest total sum of branching deviation over all of its nodes.
15) In a semantic network data-processing system, a method executed by a data processor to identify a preferred path of meaning linking a first node and a second node, where the average abstractness of all nodes in said path is the path abstractness of said path, where said path satisfies three conditions, the first condition being that said path traverses via a common inheritor node which inherits from at least one first peak node belonging to the set of inherited nodes of said first node, the second condition being that said common inheritor node also inherits from at least one second peak node belonging to the set of inherited nodes of said second node, and the third condition being that said path has minimal path abstractness relative to any path simultaneously satisfying both said first condition and said second condition.
27) In a semantic network data-processing system, a method executed by a data processor for calculating the level of abstractness of a conversation, by averaging the abstractness of inherited nodes directly linked to nodes which are a subset of the nodes inheriting from the node symbolic of the conversation.
Dependencies are inferred from the claim text.
Drawings are temporarily unavailable. Open the PDF at USPTO
From the US filing history. Foreign family members aren’t listed.
Backward citations are indexed for grants only.
Handling natural language data: semantic analysis
Using natural language modelling: semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
Computing and data processing
Speech analysis techniques or speech synthesis
Term end not estimated
USPTO file history isn’t available for this patent, so only indexed dates are shown.
See attached FORM PTO-1449.
Not Applicable.
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.
FIG. 1 outlines the use of a semantic network to implement the intention of a function invocation tree, where verbs act as function operators and nouns act as function parameter passing symbols.
FIG. 2 outlines elements of intent in the interactions between two computer systems connected by a conversational interface, in which meaning is transmitted conversationally, similarly to the way people converse in Internet chat sessions.
FIG. 3 outlines the use of a semantic network to implement a general model of conversational intentionality at three levels of abstractions: operator, conversation and motivation; motivations act as an interface level to external data.
In FIG. 1, verbs and nouns of a semantic network are composed into functional lines of intent, so that a semantic network can represent the intention of traditional function-calling hierarchies in computer languages such as C and Pascal. By using noun symbols in place of passed parameter names, the semantic network can pass data inputs from ‘Verb A’ to successively higher verb abstractions, such as ‘Verb A1’ or ‘Verb A2’ until data inputs reach system-supplied verbs, which generate output data. This output data is then directed concreteward back down the semantic network to the originator of the input data. Such a semantic network would thus, in general, transform data by passing new data abstractward for processing, expecting the results of that processing to come back as a return data flow, returning to the originating ‘Verb A’.
In FIG. 2, the semantic data flow architecture of FIG. 1 is implemented inside a conversational computer application, where conversations in the form of suggestions and stories are passed between two computer applications, or a between a person and a computer application. The substantive motivating verb for such an application is the ‘Intent to Converse’ and abstract verbs to which input data is passed are shown as ‘Intent to Learn’, ‘Intent to Refine Memory’ and ‘Intent to Help’. Suggestions and stories, which may take any form, are passed as inputs in contrast to prior art, which uses rigidly structured inputs consisting only of direct imperative commands.
In FIG. 3, the general idea of ‘Intent to Converse’ from FIG. 2 is expanded into a verb abstraction hierarchy consisting of abstract operators, moderately abstract conversations and concrete motivations. The nouns which are required at each verb level are shown. For the level of operators, inputs and outputs are shown. For the level of conversations, analysis, inputs, fluidity, outputs and focus are shown. For the level of motivations, priority, inputs, fluidity, outputs and emotion are shown. At the motivation level, external inputs and outputs are interfaced into the semantic network, so that motivations can prioritize external interface dataflows. Since motivations contain varying priorities, they also control which of the conversations will receive an activating input which in turn would activate particular operators.