Method to quantify abstraction within semantic networks

• US 5,937,400 A
• Filed: 03/19/1997
• Issued: 08/10/1999
• Est. Priority Date: 03/19/1997
• Status: Expired due to Fees

• Current Assignee: Hanger Solutions, LLC (IP Investments Group LLC)
• Original Assignee: Lawrence Au
• Inventors: Au, Lawrence
• Primary Examiner(s): Downs, Robert W.
• Application Number:
  US08/825,552
• Time in Patent Office: 874 Days
• Field of Search: 706/10, 706/55, 706/50
• US Class Current: 706/55
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation

Abstract

A computer implemented conversational method to parse meaning from input sequences of symbols, using a large dictionary stored in a semantic network, where thousands of dictionary entries for symbols are stored as nodes connected by inheritance links. Input symbols are matched to dictionary entries, with priority given to entries which cover a range of concreteness similar to the range covered by the dictionary. Ranges of concreteness which are missing are identified and reported back to the user. Concreteness of each entry is quantified from topology of inheritance links; quantification is done automatically when defining meanings by inheritance. As network topology shifts due to creation and removal inheritance links, quantified concreteness of affected dictionary entries are updated, automatically maintaining a consistent concreteness numbers. New inheritor links cause entries to become more concrete. When they become too concrete to be inherited from, indirect links via entries of lesser concreteness are substituted for the direct links, thus maintaining a uniform and consistent inheritance topology within the semantic network. Variations in dictionary semantics emerging between authors or evolving over time are automatically cohered into a single consistent inheritance topology, by shifting the inheritance links so that concrete entries always inherit from more abstract entries. When parsing text, the accuracy of the method of Most Specific Common Abstraction is improved by the consistency and uniformity of inheritance enforced by the shifting of inheritance links to conform to a consistent topology of inheritance.

Claims

1. A computer implemented method for natural language processing, parsing meanings from conversational input symbol sequences of using a semantic inheritance network which stores semantic inheritance links describing dictionary entry senses of meaning, each dictionary entry assigned a numeric concreteness, the various senses of meaning comprising a range of concreteness, each dictionary entry inheriting only from dictionary entries having lesser concreteness of meaning, where the method of parsing meanings from sequences of symbols comprises the following steps:

a) parsing said input symbol sequences,

b) matching the symbols of said sequence to identical symbols in said dictionary entries, to identify the possible meanings for each symbol which matches,

c) from said possible meanings, choosing a subset of meanings having a range of concreteness similar to said semantic inheritance network,

d) identifying ranges of concreteness in said semantic network which are missing from the range of concreteness covered by said subset of meanings,

e) producing conversational output in the form of messages describing senses of meaning from said dictionary in the missing range or ranges of concreteness.

2. Apparatus for processing natural language input symbol sequences, comprising

means for storing dictionary entry senses of meaning within a semantic inheritance network;

means for assigning to each dictionary entry a number describing concreteness of meaning;

means for parsing natural language symbol sequences, matching symbols to said dictionary entries to identify possible meanings for each input symbol;

means for identifying dictionary numeric ranges of concreteness covered by said dictionary entry senses of meaning;

means for identifying input numeric ranges of concreteness covered by said input symbol sequences;

means for identifying missing dictionary numeric ranges of concreteness not covered by said input numeric ranges of concreteness;

means for producing output describing missing dictionary numeric ranges of concreteness.

3. Apparatus as defined in claim 2 further comprising means for increasing said assigned concreteness of meaning of an entry when adding a symbolic inheritor link from that entry to an inherited entry.

4. Apparatus as defined in claim 2 further comprising means for decreasing said assigned concreteness of meaning of an entry when removing a symbolic inheritor link from that entry to an inherited entry.

5. Apparatus as defined in claim 2, further comprising means for assigning concreteness of meaning to a first entry when creating a new inheritor link, designated as a tethering link, from said first entry to a second entry representing inheritance of meaning from said second entry, wherein a concreteness of meaning is assigned to said first entry which is a predefined number greater than the concreteness of said second entry.

6. Apparatus as defined in claim 5, further comprising means for propagating changes in concreteness from inherited entries to corresponding inheritor entries directly connected to said inherited entries by tether links, wherein adjustments to the concreteness of inheritor entries are the same or proportional to the amount of change in concreteness of inherited entries.

7. Apparatus as defined in claim 2, further comprising

means for normalizing the concreteness of non-normal entries inherited by inheritor entries of lesser concreteness;

means for identifying non-normal inheritor entries whose concreteness is greater than their corresponding inheritor entries;

means for creating intermediary entries between said non-normal inherited entries and corresponding inheritor entries, by replacing the direct links between non-normal inheritor entries and corresponding inheritor entries by indirect links through the intermediary entries, thereby reducing the population of said direct links so that the concreteness of said non-normal inheritor entries is reduced to less than concreteness of said inheritor entries.

8. Apparatus as defined in claim 2, further comprising procedural input and output values passed solely along inheritance hierarchies.

9. Apparatus as defined in claim 8, further comprising procedural operators of logic, arithmetic, parsing, concatenation and dataflow meanings defined within said semantic network to support actions described in claim 2.