Methods and apparatus for dynamic classification of discourse

Abstract

A dynamic classification system determines content of input discourse. The dynamic classification system includes a dynamic classification system that generates a detailed and comprehensive knowledge catalog based on terminology used in the input discourse. A theme vector processor identifies themes including identifying the relative importance of the themes in the input discourse. The knowledge catalog includes static ontologies arranged in a hierarchical structure, wherein each static ontology contains a plurality of high level knowledge concepts. High level themes extracted from the input discourse are mapped to one or more knowledge concepts in the static ontologies. The dynamic classification system generates one or more dynamic hierarchies, consisting of low level or detailed knowledge concepts, based on themes extracted from the input discourse. The high level themes mapped to the static ontologies are linked to the low level themes in the dynamic hierarchies to generate a world view knowledge catalog. In addition, knowledge concepts in the dynamic hierarchies and static ontologies are cross referenced permitting flexibility to relate one or more groups of knowledge concepts in one or more static and/or dynamic hierarchies. Thus, the knowledge catalog provides both a broad and detailed classification of knowledge for the input discourse.

Claims

What is claimed is:

1. A method for dynamic classification of input discourse, said method comprising the steps of:

storing a plurality of hierarchically arranged categories that form a static hierarchy;

processing said input discourse to identify a plurality of content carrying words;

generating at least one theme for said input discourse;

selecting one of said categories from said static hierarchy to classify said theme;

generating at least one dynamic category based on said content carrying words, wherein said dynamic category defines a greater level of detail than said categories in said static hierarchy; and

generating a link between said dynamic hierarchy and said static hierarchy to form a detailed classification hierarchy, wherein terminology from said input discourse is utilized to classify said input discourse.

2. The method as set forth in claim 1, further comprising the steps of:

generating a plurality of dynamic categories based on said content carrying words; and

generating at least one dynamic hierarchy from said dynamic categories that reflects hierarchical relationships among said dynamic categories.

3. The method as set forth in claim 2, wherein the step of generating at least one dynamic hierarchy comprises the steps of:

identifying a plurality of head words from said content carrying words;

generating, for each head word, contextual relationships based on use of each head word in said input discourse; and

generating a hierarchical relationship among said head words based on common contextual relationships.

4. The method as set forth in claim 2, wherein said hierarchical relationships among said dynamic categories include semantic or linguistic relationships, such that classification of said terminology in dynamic categories reflect semantic or linguistic relationships between terminology of said input discourse.

5. The method as set forth in claim 1, wherein:

the step of storing said static hierarchy comprises the step of storing said categories in a plurality of independent ontologies, wherein an ontology comprises a plurality of related categories; and

the step of classifying terminology of said input discourse comprises the step of referencing dynamic categories from at least two different independent ontologies that relate to said terminology.

6. The method as set forth in claim 1, wherein the step of generating at least one theme for said input discourse comprises the steps of:

storing a thematic profile for said input discourse that includes a plurality of thematic tags for words in said input discourse, wherein said thematic tags indicate applicability of a plurality of thematic constructions that indicate relative thematic importance of words with other words in said input discourse;

storing a lexicon comprising a plurality of words and definitional characteristics for said words; and

generating said at least one theme from words in said input discourse based on existence or non existence of said thematic constructions as indicated by said thematic tags, and based on definitional characteristics of said words as indicated by said lexicon, wherein themes generated identify overall content of said input discourse.

7. A method for dynamically classifying input discourse, said method comprising the steps of:

storing a plurality of hierarchically arranged categories that form a static hierarchy to classify discourse;

generating at least one dynamic category based on terminology from said input discourse, wherein said dynamic category defines a greater level of detail than said categories in said static hierarchy; and

generating a link between said dynamic category and a category of said static hierarchy associated with said dynamic category to form a detailed classification hierarchy, wherein terminology from said input discourse in used to classify said input discourse.

8. The method as set forth in claim 7, further comprising the steps of:

processing said input discourse to identify a plurality of content carrying words;

generating a plurality of dynamic categories based on said content carrying words; and

generating at least one dynamic hierarchy from said dynamic categories that reflects hierarchical relationships among said dynamic categories.

9. The method as set forth in claim 8, wherein the step of generating at least one dynamic hierarchy comprises the steps of:

identifying a plurality of head words from said content carrying words;

generating, for each head word, contextual relationships based on use of each head word in said input discourse; and

generating a hierarchical relationship among said head words based on common contextual relationships.

10. The method as set forth in claim 8, wherein said hierarchical relationships among said dynamic categories include semantic or linguistic relationships, such that classification of said terminology in dynamic categories reflect semantic or linguistic relationships among terminology of said input discourse.

11. A method for dynamically classifying input discourse, said method comprising the steps of:

storing a plurality of hierarchically arranged categories that form a static hierarchy, said categories of said static hierarchy representing high level concepts;

processing said input discourse to identify a plurality of content carrying words;

generating a classification for a portion of said carrying words by mapping applicable content carrying words to categories of said static hierarchy;

generating at least one dynamic category from content carrying words that do not map to said categories of said static hierarchy, wherein said dynamic category defines a greater level of detail than said categories in said static hierarchy; and

generating a link between said dynamic category and a category of said static hierarchy based on contextual use of a respective content carry word in said input discourse.

12. A computer readable medium comprising a set of instructions stored therein, which when executed by a computer, causes the computer to perform the steps of:

storing a plurality of hierarchically arranged categories that form a static hierarchy;

processing said input discourse to identify a plurality of content carrying words;

generating at least one theme for said input discourse;

selecting one of said categories from said static hierarchy to classify said theme;

generating at least one dynamic category based on said content carrying words, wherein said dynamic category defines a greater level of detail than said categories in said static hierarchy; and

generating a link between said dynamic hierarchy and said static hierarchy to form a detailed classification hierarchy, wherein terminology from said input discourse is utilized to classify said input discourse.

13. The computer readable medium as set forth in claim 12, further comprising instructions for:

generating a plurality of dynamic categories based on said content carrying words; and

generating at least one dynamic hierarchy from said dynamic categories that reflects hierarchical relationships among said dynamic categories.

14. The computer readable medium as set forth in claim 13, wherein instructions for generating at least one dynamic hierarchy comprise instructions for:

identifying a plurality of head words from said content carrying words;

generating, for each head word, contextual relationships based on use of each head word in said input discourse; and

generating a hierarchical relationship among said head words based on common contextual relationships.

15. The computer readable medium as set forth in claim 13, wherein said hierarchical relationships among said dynamic categories include semantic or linguistic relationships, such that classification of said terminology in dynamic categories reflect semantic or linguistic relationships between terminology of said input discourse.

16. The computer readable medium as set forth in claim 12, wherein:

the instructions for storing said static hierarchy comprise instructions for storing said categories in a plurality of independent ontologies, wherein an ontology comprises a plurality of related categories; and

the instructions for classifying terminology of said input discourse comprise instructions for referencing dynamic categories from at least two different independent ontologies that relate to said terminology.

17. The computer readable medium as set forth in claim 12, wherein the instructions for generating at least one theme for said input discourse comprise instructions for:

storing a thematic profile for said input discourse that includes a plurality of thematic tags for words in said input discourse, wherein said thematic tags indicate applicability of a plurality of thematic constructions that indicate relative thematic importance of words with other words in said input discourse;

storing a lexicon comprising a plurality of words and definitional characteristics for said words; and

generating said at least one theme from words in said input discourse based on existence or non existence of said thematic constructions as indicated by said thematic tags, and based on definitional characteristics of said words as indicated by said lexicon, wherein themes generated identify overall content of said input discourse.

18. A computer readable medium comprising a set of instructions stored therein, which when executed by a computer, causes the computer to perform the steps of:

storing a plurality of hierarchically arranged categories that form a static hierarchy to classify discourse;

generating at least on dynamic category based on terminology from said input discourse, wherein said dynamic category defines a greater level of detail than said categories in said static hierarchy; and

generating a link between said dynamic category and a category of said static hierarchy associated with said dynamic category to form a detailed classification hierarchy, wherein terminology from said input discourse is used to classify said input discourse.

19. The computer readable medium as set forth in claim 18, further comprising instructions for:

processing said input discourse to identify a plurality of content carrying words;

generating a plurality of dynamic categories based on content carrying words; and

generating at least one dynamic hierarchy from said dynamic categories that reflects hierarchical relationships among said dynamic categories.

20. The computer readable medium as set forth in claim 19, wherein instructions for generating at least one dynamic hierarchy comprise instructions for:

identifying a plurality of head words from said content carrying words;

generating, for each head word, contextual relationships based on use of each head word in said input discourse; and

generating a hierarchical relationship among said head words based on common contextual relationships.

21. The computer readable medium as set forth in claim 19, wherein said hierarchical relationships among said dynamic categories include semantic or linguistic relationships, such that classification of said terminology in dynamic categories reflect semantic or linguistic relationships among terminology of said input discourse.

22. An apparatus for dynamically classifying input discourse, said apparatus comprising:

a storage device for storing a plurality of hierarchically arranged categories that form a static hierarchy; and

a content indexing processor coupled to said storage device for generating at least one dynamic category based on terminology from said input discourse, wherein said dynamic category defines a greater level of detail than said categories in said static hierarchy, and for generating a link between said dynamic category and a category of said static hierarchy associated with said dynamic category.

Description

MICROFICHE APPENDICES

Appendix D, entitled "Analysis Documentation" contains four microfiche with a total number of (210) frames.

Appendix A, entitled "Static ontologies" contains one microfiche with a total number of (22) frames.

Appendix F, entitled "Theme Parser Code" contains five microfiche with a total number of two hundred and eighty two (282) frames.

Appendix G, entitled "Code Heading" contains two microfiche with a total number of eighty five (85) frames.

Appendix H, entitled "Oracle Context Linguistics Toolkit Guide Reference" contains four microfiche with a total number of (229) frames.

Appendix I, entitled "Theme Vector Code" contains one microfiche with a total number of sixty three (63) frames.

Appendix J, entitled "Content Indexing Code" contains one microfiche with a total number of seventy one (71) frames.

Appendix K, entitled "Creating a Virtual Bookshelf" contains one microfiche with a total number of (22) frames.

COPYRIGHT NOTICE

Appendices A,B,C,E,F,G,H,I,J and K contain material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by any one of the Appendices as they appear in the United States Patent and Trademark patent files or records, but otherwise reserves all copyright rights whatsoever.

FIELD OF THE INVENTION

The present invention relates to the field of computational linguistics, and more particularly to dynamic classification of input discourse.

BACKGROUND OF THE INVENTION

The ability to accurately classify documents based on the content of the document has a wide range of document search and retrieval applications. For example, the Library of Congress of the United States utilizes a document classification system to store and retrieve documents. Typically, document classification systems, such as the dewy decimal system, are configured in fixed categories. Currently, in order to classify a document, a person familiar with the content of the document manually classifies the document in accordance with the fixed categories. Such a classification system is rigid, and therefore it results in a classification system with limited potential.

In modern society, there is an increasing demand for use of vast amounts of information covering a wide range of topics. In order to best utilize the vast amounts of information, an accurate and detailed classification system is desirable. For example, a person may wish to locate documents on a very specific topic within a more general topic. Without a detailed classification system, the person would be required to read through all documents on the more general topic to locate the more specific topic desired. Therefore, the increased amounts of information in a wide variety of topics or subject matter only increases the problems of classification in fixed categories because greater levels of detail or categories are required to accurately classify documents.

Because of the vast amounts of information, it is impossible to generate a fixed classification system that includes low level detailed topics for all general topics. Therefore, it is desirable to generate a dynamic classification system that includes detailed classification categories based on the content of the document. Furthermore, it is desirable to accurately classify documents without requiring a human to first read the document, and then categorize it based on the readers interpretation of the content of the document.

As is explained below, a dynamic classification system of the present invention generates a detailed and comprehensive hierarchical classification based on, in part, the specific terminology contained in the document. In addition, the dynamic classification system does not require human intervention to read the document to determine a classification category.

SUMMARY OF THE INVENTION

A dynamic classification system generates a detailed and comprehensive knowledge catalog based on terminology used in the input discourse. The dynamic classification system includes static ontologies, wherein each static ontology contains pre-determined high level knowledge concepts arranged hierarchically. The dynamic classification system receives a plurality of themes extracted from the input discourse. The high level themes are mapped to one or more of the high level knowledge concepts in the static ontologies to classify the high level themes. The dynamic classification system generates one or more dynamic hierarchies based on themes extracted from the input discourse. Specifically, the dynamic hierarchies consist of low level or detailed knowledge concepts derived from the extracted themes.

The high level themes mapped to the static ontologies are linked to the low level themes in the dynamic hierarchies to generate a world view knowledge catalog. In addition, knowledge concepts in the dynamic hierarchies and static ontologies are cross referenced permitting flexibility to relate concepts in one or more static and/or dynamic hierarchies. The knowledge catalog provides both a broad and detailed classification of knowledge for the input discourse. The knowledge catalog has application for classifying input discourse based on content.

In one embodiment, the dynamic classification system generates a plurality of head words from the input discourse, wherein the head words represent the lower level themes extracted from the input discourse. For certain head words, the dynamic classification system generates contextual relationships based on use of the head word in the input discourse. The dynamic hierarchies are then generated based on the head words and their contextual relationship with other head words.

Other features and advantages of the present invention will be apparent from the accompanying drawings, and from the detailed description that follows below.

BRIEF DESCRIPTION OF THE DRAWINGS

The features and advantages of the present invention will be apparent from the following detailed description of the preferred embodiment of the invention with references to the following drawings.

FIG. 1a illustrates several example static ontologies.

FIG. 1b illustrates a high level conceptualization for a set of static ontologies.

FIG. 2 is a block diagram illustrating a dynamic classification system for input discourse.

FIG. 3 is a flow diagram illustrating the high level methodology for content processing.

FIG. 4 conceptually illustrates an example of theme vector processing.

FIG. 5 illustrates an example of expanding the static ontologies into the dynamic level of the knowledge catalog.

FIG. 6a illustrates a portion of an ontology for "economics."

FIG. 6b illustrates a portion of an ontology for "business and industry."

FIG. 7 illustrates a high level methodology for dynamic content indexing.

FIG. 8 illustrates example static ontologies for classifying theme concepts.

FIG. 9 illustrates a dynamic level hierarchy for an example input discourse.

FIG. 10 illustrates a high level block diagram of a general purpose computer system in which the dynamic classification system of the present invention may be implemented.

DETAILED DESCRIPTION

KNOWLEDGE CATALOG

A knowledge catalog of the present invention provides a unique infrastructure to accurately represent concepts that define knowledge. The knowledge catalog includes a set of static ontologies to define knowledge. The knowledge catalog may be characterized as including several static ontologies because the knowledge catalog provides multiple views, characterizations, and organizations of concepts. As is explained below, the knowledge catalog contains a broad coverage of concepts to provide a world view of knowledge. The static ontologies are configured as hierarchical structures such that knowledge concepts are organized from high level generalized concepts down to more specific concepts. The static ontologies are parallel and independent of each other.

FIG. 1a illustrates several example static ontologies configured in accordance with one embodiment of the present invention. Specifically, FIG. 1a illustrates hierarchical structures for the "business and industry", "family", "education", "metrics", and "economics" static ontologies. As shown in FIG. 1a, each static ontology contains a plurality of concepts, illustrated as nodes in the hierarchical structures for the respective static ontologies. For purposes of simplicity, the specific "concepts" are illustrated only as nodes. However, the specific concepts contained within each static ontology hierarchical structure are set forth in Appendix A. Appendix A includes a plurality of static ontologies configured in accordance with the preferred embodiment of the present invention. Although Appendix A provides a set of static ontologies for the preferred embodiment, modifications to the classification or concepts may be made without deviating from the spirit or scope of the invention.

Each ontology shown in FIG. 1a contains a plurality of levels that form the hierarchical structure. For example, the "business and industry" static ontology contains three levels of concept classifications under the highest level concept, and the "economics" contain four levels of concept classifications. Generally, the static ontologies of the present invention includes no more than six levels.

The actual configuration, structure and orientation of a particular ontology is dependent upon the subject matter or field of the ontology. Therefore, each ontology in the set of static ontologies of the present invention contain a different point of view. The different points of view for the static ontologies result in different approaches or different ways of viewing the knowledge concepts in the different static ontologies. Consequently, the organization of the knowledge concepts in each ontology is not consistent. As is explained below, the different points of view for the ontologies permit combining "like" forms of knowledge with "unlike" forms of knowledge through linking of one or more static ontologies.

Although the set of static ontologies are broad to cover a wide range of concepts, they are not extremely detailed. In addition, the set of static ontologies do not contain concepts that are extremely volatile. For an example ontology, a classification for "computers", may contain a sub classification for "operating systems." The sub classification "operating systems" may further include the sub-classification "Windows." However, the particular versions or attributes of the Windows operating system would not be contained in the example computer ontology. Because the static ontologies define a broad range of concepts but not in great detail, they remain relatively stable over time. Therefore, the static ontologies do not require frequent updates.

The ontologies contained within the knowledge catalog of the present invention are organized such that the concepts are independent of language and culture. Therefore, the set of static ontologies in the knowledge catalog of the present invention provide a world view of knowledge. The knowledge concepts in the static ontologies are applicable to all cultures and languages. For example, in the "business and industry" ontology, the concept "commerce and trade" is a sub classification of the top level "business and industry" concept. Under the concept "commerce and trade", the ontology includes the concepts "marketing", "retail trade industry", and "wholesale trade industry." For this example, all cultures engaging in commerce and trade have concepts in marketing, retail trade industry, and wholesale trade industry. Although different words in different languages are used to describe the concepts of "marketing", "retail trade industry", and "wholesale trade industry", these concepts exist in all cultures. Therefore, the "business and industry" ontology, similar to all of the static ontologies, is not dependent on language or culture.

The "business and industry" ontology hierarchy shown in FIG. 1a contains three levels of concepts in the hierarchical structure. With only three levels, the "business and industry" ontology does not contain too much detail by including low level concepts or words that are specific to the business and industry field. For example, an additional category for "Wallmart" may be desired to more specifically classify knowledge beyond the concept of "convenience stores." However, the concept of "Wallmart" is too detailed for the business and industry static ontology because "Wallmart" does not translate across cultures and languages. For this example, in order to classify "Wallmart", the concept is placed in a dynamic level as is explained below.

The static ontologies in the knowledge catalog are in essence a new type of book. For example, a thesaurus provides words with their linguistic connotations. However, a thesaurus does not provide real world culturally independent knowledge concepts, including both semantic and linguistic derivatives of the knowledge concepts. Similarly, a dictionary, which provides definitions and origins of words, does not provide such an organization of knowledge concepts. Also, an encyclopedia, which provides detailed information on a variety of subject matters, does not provide a world view of knowledge as the knowledge catalog of the present invention.

A high level conceptualization for a set of static ontologies are shown in FIG. 1b. For this embodiment, a set of static ontologies 105 are organized in an industry domain 110, a concrete/generic domain 120, and an abstract domain 130. The industry domain 110 includes a plurality of the static ontologies to represent words that define concrete terminology specific to one or more industries or fields of study. There are different types of industry domain 110 ontologies. For example, a classification for "electronics" includes sub-classifications, for "computers" and "micro electronics." Other classifications, such as "geography", are separated into two high level classifications, "political" and "physical" geography. The industry domain 110 separates the political and physical geography classifications because the concepts of political and physical geography are very different, and each classification includes different sub-classifications beneath them.

Some of the set ontologies in the industry domain 110 are not as much industries but rather concrete concepts. For example, "clothing" represents a specific concrete ontology. In the hierarchical structure, some sub-classifications provide factual information for the higher level or parent classification, while other sub-classifications provide more richness to the parent classification. For example, the term "computer program" provides more richness to the parent classification "computer industry." However the term "California avocados" provides factual information for the parent classification "avocados" by representing that the avocados are from California.

As shown in FIG. 1b, for the preferred embodiment, the static ontologies 105 further include the concrete/generic domain 120 and abstract domain 130. The concrete/generic domain 120 contains static ontologies to represent knowledge concepts that have concrete definitions, but may be applicable to one or more specific industries or categories (e.g. the terms are generic). For example, the word "plant", when used to describe a manufacturing plant, is a concrete term that applies to many industries. Therefore, a classification for "plant" is located in the concrete/generic domain 120. The abstract domain 130 contains representations of words that are not concrete, but rather describe abstract concepts. For example, the concepts of "knowledge" and "methods" do not describe material items, but are "abstract" concepts. Although the present invention is described in conjunction with a specific domains (e.g. industry, concrete/generic, and abstract domains), other organizations may be used to define the static ontologies without deviating from the spirit and scope of the invention.

The static ontologies of the present invention are relational such that the linking of one or more static ontologies, or portions thereof, result in a very detailed organization of knowledge concepts. As described above, each static ontology provides a high level view of a particular subject; however, linking or cross referencing among two or more static ontologies results in many combinations of knowledge hierarchical structures. For example, for the static ontologies shown in FIG. 1a, if a particular classification included concepts in "business and industry" and "economics", then the combination of the "business and industry" and "economics" ontologies provides seven hierarchical levels for that classification. Therefore, the parallel and independent nature of static ontologies, when linked or cross referenced, provides an extremely detailed and comprehensive coverage of knowledge concepts. In order to provide a complete classification hierarchy without cross referencing the independent and parallel ontologies of the present invention, it would require developing a hierarchy that anticipated every combination of knowledge concepts. As can be realized, if this approach is taken for each subject, then setting forth every conceivable breakdown becomes impossible. The groupings of independent and parallel ontologies provide a relational knowledge catalog.

The parallel and independent ontologies of the present invention allow for the greatest amount of flexibility in developing a detailed classification breakdown. For example, a first ontology may include a concept for "geography", and a second ontology may include a concept for "computers." Geography and computers are very different topics, and seemingly unrelated. The "geography" and the "computers" concepts are located in different ontologies due to the very different nature of the subject matter. However, the cross referencing of the "geography" and the "computers" ontologies permits linking the two different subject matters. For example, for the document classification application, a document may be about how different computers are used in different geographical areas.

The hierarchically structured concepts contained within the static ontologies may be semantic derivatives and/or linguistic derivatives of the high level concept. For example, a classification hierarchy may associate a "house" with articles placed within a house, such as dining room tables, chairs, bedroom furniture, etc. For this example, a classification hierarchy that includes semantic derivative concepts based on the "house" concept is desirable. However, for a purely linguistic classification hierarchy, the house "concept" may include types of houses, such as a ranch style house, a contemporary house, etc. The knowledge catalog of the present invention provides the framework or structure to generate hierarchical classifications including both semantic derivatives and/or linguistic derivatives of high level concepts. Utilizing the static ontologies of the present invention, a knowledge map may be generated that links both concepts representing articles placed within a house and concepts representing types of houses to the higher level "house" concept. Consequently, the flexibility provided by the structure of the knowledge catalog permits generation of a true knowledge map.

DYNAMIC CLASSIFICATION SYSTEM

FIG. 2 is a block diagram illustrating a dynamic classification system of the present invention. In general, a dynamic classification system 200 receives input discourse, and generates, as output, a dynamic classification index 265. The dynamic classification index 265 contains hierarchical knowledge representations of the input discourse integrated to the static ontologies. The dynamic classification system 200 contains a morphology section 210 coupled to a linguistic engine 220. The linguistic engine 220 receives the input discourse for processing. In general, the linguistic engine 220 processes the input discourse by analyzing the grammatical or contextual aspects, as well as the stylistic and thematic attributes of the input discourse. Specifically, the linguistic engine 220 generates contextual, thematic, and stylistic tags that characterize the input discourse. As shown in FIG. 2, the linguistic engine 220 generates a structured output 230 containing, in part, the contextual, thematic, and stylistic tags. Furthermore, the linguistic engine 220 extracts the topic and content for each sentence in the input discourse.

The morphology 210 includes a lexicon 212 and the knowledge catalog 100 of the present invention. In general, the lexicon 212 contains definitional characteristics for a plurality of words and terms. For example, the lexicon 212 defines whether a particular word is a noun, a verb, an adjective, etc. The linguistic engine 220 utilizes the definitional characteristics stored in the lexicon 212 to generate the contextual, thematic, and stylistic tags in the structured output 230. The generation of thematic tags is described more fully below. Appendix B, entitled "Lexicon Documentation", provides a description of the definitional characteristics in accordance with one embodiment.

The dynamic classification system 200 further includes a knowledge catalog processor 240, theme vector processor 250, and content indexing processor 260. As shown in FIG. 2, the knowledge catalog processor 240, theme vector processor 250 and content indexing processor 260 are coupled to the lexicon 212 and knowledge catalog 100. In general, the knowledge catalog processor 240 receives, as inputs, the structured output 230 including the extracted content, the knowledge catalog 100, and the lexicon 212, and processes the extracted content for use with the knowledge catalog 100. The theme vector processor 250 receives, as inputs, the structured output 230, the processed output from the knowledge catalog processor 240, the knowledge catalog 100, and the lexicon 212. The theme vector processor 250 identifies the most important themes in the input discourse, and maps these themes into the knowledge catalog 100.

The content indexing processor 260 receives as inputs, the definitional characteristics in the lexicon 212, the structured output 230, and theme information from the theme vector processor 250. The content indexing processor 260, in accordance with the dynamic classification system of the present invention, extends the static ontologies 105 into the dynamic level 135 based on the specific terminology contained in the input discourse. Furthermore, the content indexing processor 260 attaches the dynamic level 135 terminology into the concepts in the static ontologies 105 that were mapped by the theme vector processor 250.

As described above, the knowledge catalog 100 describes all senses for a particular word or term. In a preferred embodiment, the content and topic of sentences is extracted in the linguistic engine 220, and is input to the knowledge catalog processor 240 in the form of content carrying words. The dynamic classification system 200 only operates on the content carrying words for which the sense has been determined with a large degree of certainty (e.g. the non ambiguous content carrying words). After determining the sense of content carrying words, the dynamic classification system 200 determines the expressions of content for the input discourse.

The dynamic classification system 200 may be implemented in either hardware or software. For the software implementation, the dynamic classification system is software that includes a plurality of computer executable instructions for implementation on a general purpose computer system. Prior to loading into a general purpose computer system, the dynamic classification system software may reside as encoded information on a computer readable medium, such as a magnetic floppy disk, magnetic tape, and compact disc read only memory (CD-ROM). In one hardware implementation, the linguistic engine 220, the knowledge catalog processor 240, the theme vector processor 250 and the content indexing processor 260 may each comprise dedicated processors including processor instructions for performing the functions described herein. Circuits may also be developed to perform the functions described herein. The lexicon 212 and the knowledge catalog 100 may be implemented as databases stored in memory for use by the dynamic classification system 200.

FIG. 3 is a flow diagram illustrating the high level methodology for content processing. The ontologies, such as the static ontologies 105, are provided as shown in block 300. As shown in block 310, the sense of content carrying words in the input discourse are determined. For all words in which the sense has been determined, the relative importance of these words, with respect to each other, are determined as shown in block 320. Based on the relative importance, the words are mapped to the static ontologies to reflect the high level or general concepts disclosed in the input discourse as shown in block 330. As shown in block 340, based on words contained in the input discourse, the static ontologies are expanded to include dynamic ontologies that contains more detailed concepts. The static ontologies and the dynamic ontologies provide a detailed world view of knowledge. The input discourse is dynamically classified using the static and dynamic ontologies, as shown in block 350.

In general, the concepts mapped to the static and dynamic ontologies capture the overall world view content classification of the input discourse. However, through discourse of a document, a combination of the abstract expressions of the knowledge catalog elements represent the actual content of the particular document. Therefore, the concepts from the static ontologies are an abstract representation of input discourse themes in time. The combination of these themes in time represent the actual content of the input discourse.

The knowledge catalog processor 240 processes content carrying words for use by the theme vector processor 250. Specifically, the knowledge catalog processor 240 generates the noun or nominal forms of the content carrying words in the input discourse. The knowledge catalog processor 240 parses the knowledge catalog 100 to compare the content carrying words with the knowledge concepts stored in the static ontologies. If the content carrying word is not contained in at least one domain in the static ontologies, then the word is noted as being not yet understood. If the word is stored more than once in the static ontologies, then the word is classified as ambiguous. If the word is stored only once in the static ontologies, then the word is classified as non ambiguous. If a word is classified as non ambiguous, then the theme vector processor 250 determines the concept of the word from the static ontologies. In a preferred embodiment, a minimum of 50% of the content carrying words must be recognized as non ambiguous.

As discussed above, the knowledge catalog processor 240 generates noun and nominal forms of content carrying words for look-up in the knowledge catalog 100. In order to accomplish this, the knowledge catalog processor 240 reads the grammatical tags in the structured output 230 to determine which words are the content carrying words. The contextual tags determine whether a word carries content. For example, the word "will" could be an auxiliary verb or it could be a noun (e.g. a legal term). As an auxiliary verb, the word "will" carries no content. Using the content carrying words, the knowledge catalog processor 240 separates nouns and non nouns. The non noun content carrying words, such as adjectives, adverbs, and verbs, etc., are converted into nouns or nominal forms. For example, the verb "fishing" is converted to "to fish." The content carrying nouns are further analyzed to determine whether the word has a mass sense and/or a count sense. For example, the word "baseball", which is the mass sense of the word, has a completely different meaning than the count sense "baseballs."

In the preferred embodiment, because the noun sense of a word may be different than the nominal sense, and the mass sense may be different than the count sense, a separate classification or tree attachment exists for nominal and noun forms as well as for mass and count senses. Therefore, the knowledge catalog 100 stores all possible senses of a content carrying word. For example, the noun "a fish" has a different meaning than the verb "fishing" which was nominalized to "to fish." Based upon the above criteria, every content carrying word contains up to four potential pointers into one or more categories in the static ontologies 105. Specifically, each content carrying word may contain a pointer to the static ontologies 105 for the noun form, the nominal form, the mass sense, and the count sense.

In addition to nouns and regular content carrying words discussed above, words in the input discourse may also be characterized as special words. These special words include orienters and global classifiers. Orienter words guide the reader to a field of study regardless of the use of the word. A global classifier word carries the essence of a particular category, wherein the category may include sub categories beneath the category defined by the global classifier. In generating the pointers to the static ontologies, if there is not a concept for the nominal, then the concept of the noun is attached to the concept of the nominal. The special words may be mapped directly into the static ontologies.

After the sense of a majority of the content carrying words has been determined, the dynamic classification system 200 maps key content carrying words into the static ontologies 105. In order to accomplish this, the theme vector processor 250 determines the relative importance of content carrying words in relationship to other content carrying words. The most important content carrying words are mapped to the knowledge catalog elements in the static ontologies 105.

FIG. 4 conceptually illustrates an example of theme vector processing. For purposes of explanation, the static ontologies 105 of the knowledge catalog 100 is shown as having three classification hierarchies 510, 520 and 530. Specifically, the database element "A" represents the highest level for the classification hierarchy 510, and the database element "B" and "C" are sub classifications of "A." Also, the classification hierarchy 510 includes sub classifications "D" and "E" linked to the database element "C." Furthermore, as shown in FIG. 4, the classification hierarchies 520 and 530 contain the database elements "U" and "L", respectively, as the high level classifications.

The theme vector processor 250 receives, as an input, the structured output 230 that contains contextual and thematic tags. In addition, the theme vector processor 250 receives the definitional characteristics from the lexicon 212. With this information, the theme vector processor 250 executes a plurality of heuristic routines to generate the relative importance of content carrying words in the input discourse. The theme vector processor 250 only operates on non ambiguous content carrying words or ambiguous words that have been disambiguated. In a preferred embodiment, the theme vector processor 250 generates a theme strength based on the output of the heuristic routines. The theme strength of a particular content carrying word is then used in conjunction with the theme strength of another word to generate a ratio of relative thematic importance between all words.

A theme vector output, labeled 500 in FIG. 4, illustrates several relative theme strengths for a number of words in an example input discourse. As shown in FIG. 4, the ninth word "W9" received a relative weight of 35, and the first word "W1" received a relative weight of 27. Therefore, "W9" has a relative importance of 35/27 in relationship to "W1." The theme vector processor 250 links the most important content carrying words into the ontologies of the static ontologies 105. As discussed above, the theme vector processor 250 only operates on non ambiguous content carrying words. Therefore, the content carrying words are directly mapped into the database elements of the static ontologies 105. For the example shown in FIG. 4, the most important word, "W9", has the "D" concept, and therefore is mapped to the "D" knowledge concept in the static ontologies 105. Note that for this example, the three most important words tie into the ontologies in three separate classification hierarchies. Accordingly, the theme vector processor 250 determines that the three most important concepts in the input discourse are "D", "W", and "N".

In a preferred embodiment, the theme vector processor 250 uses the sixteen most important words of each sentence to map the concepts into the static ontologies 105. If the input discourse consists of a single paragraph, the sixteen concepts are more than enough to capture the content of the document. For a larger document consisting of many paragraphs, the sixteen most important concepts captures the verbal content, and extending beyond the sixteen concepts typically only reinforces the high level concepts that were deemed most important.

After the most important concepts are mapped into the static ontologies 105, the content indexing processor 260 expands additional concepts into the dynamic level 135 of the knowledge catalog 100. In general, the content indexing processor 260 learns how to use content for tie-ins. Terms that fall within the concrete/generic domain 120 and abstract domain 130 are not learned, only terms falling within the industry domain 110 are learned. For example, in a document containing the word "bear", the content indexing processor 260 determines that the sense of bear was used as an animal. Accordingly, the term is attached from the dynamic level 135 up into the animal category in the static ontologies 105.

FIG. 5 illustrates an example of expanding the static ontologies 105 into the dynamic level 135. The content indexing processor 260 receives, as inputs, contextual and thematic tags from the structured output 230 as well as the output of the theme vector processor 250. From this information, the content indexing processor 260 generates the dynamic classification hierarchies. The example illustrated in FIG. 5 extends the conceptualize example static ontologies illustrated in FIG. 4. The knowledge concept "D" is expanded into the sub classifications D.sub.1, D.sub.2, D.sub.3, D.sub.4, D.sub.5, and D.sub.6. Similarly, for the knowledge concepts "W" and "N", the content indexing processor 260 generates additional sub classifications.

THEME PARSING OVERVIEW

As discussed above, the dynamic classification system 200 utilizes the contextual tags stored in the structures output 230. In one embodiment, the linguistic engine 220 generates the contextual tags via a chaos loop process. Appendix C, entitled "Chaos Processor for Text", contains an explanation for generating contextual or granrnatical tags. All words in a text have varying degrees of importance in the text, some carrying grammatical information, and others carrying the meaning and content of the text. In general, the chaos loop process identifies, for words and phrases in the input discourse, grammatical aspects of the input discourse including identifying the various parts of speech. In order to accomplish this, the chaos loop process ascertains how the words, clauses and phrases in a sentence relate to each other. By identifying the various parts of speech for words, clauses, and phrases for each sentence in the input discourse, the context of the input discourse is defined. The chaos loop process stores information in the form of the contextual tags. For a detailed explanation of the contextual or grammatical tags, see Appendix D, entitled "Analysis Documentation."

The theme vector processor 250 also utilizes the thematic tags stored in the structures output 230. Each word carries thematic information that conveys the importance of the meaning and content of the input discourse. In general, the thematic tags identify thematic context of the input discourse, such as text. Each word is discriminated in the text, identifying the type of importance or meaning, the impact on different parts of the text, and the overall contribution to the content of the text. The thematic context of the text is determined in accordance with predetermined theme assessment criteria that is a function of the strategic importance of the discriminated words. The predetermined thematic assessment criteria define which of the discriminated words are to be selected for each thematic analysis unit. The text is then output in a predetermined thematic format.

In addition to the grammatical tags and thematic tags, the structured output 230 also includes outputs for extracted topics, kernel sentences, and extracted content. The knowledge catalog processor 240 utilizes these output areas to generate the content carrying words found in the input discourse. The linguistic engine 220 further includes a topic extractor, that identifies the topics of the input discourse, a kernel generator, that generates summarized versions of the input discourse, and a content extractor that identifies key content of the input discourse. For a complete description of theme parsing to generate the thematic tags, see Appendix E, entitled "Text Theme Parser for Text." In addition, Appendix F, entitled "Theme Parser Code" contains a source code listing for implementing the Theme Parser. Appendix G, entitled "Code Headings", contains a list of defined source code headers corresponding to the theme parser source code of Appendix F. Also, for a further explanation of the thematic output, see Appendix H, entitled "Oracle ConText.TM. Linguistics Toolkit, Guide and Reference, Release 1.1."

THEME VECTOR PROCESSING

As discussed above, the theme vector processor 250 identifies the most important theme terms and theme concepts for the input discourse for use by the dynamic classification system of the present invention. In order to generate the theme vector output 200, the theme vector processor 250 receives thematic tags and contextual tags from the structured output 230 (FIG. 2). In addition, the theme vector 250 receives the content carrying words from the knowledge catalog processor 240. The content carrying words may include single words or phrases. Note that the content carrying words output from the knowledge catalog processor 240 are converted to the noun or nominal form. In turn, the theme vector processor 250 generates a theme vector output 500 (FIG. 4). In a preferred embodiment, the theme vector output 500 contains, for each theme term, an importance number (#), a. theme strength, and a theme concept.

The theme vector output 500 presents a thematic profile of the contents of input discourse (e.g. a sentence, paragraph, or document). The theme vector output 500 contains a list of nominalized words, labeled "Theme Term" in FIG. 4, that represent the major themes in the input discourse. In a preferred embodiment, the theme vector output includes up to sixteen of the most important themes for each sentence, paragraph and document. A theme term in a theme vector is typically based on an actual word or phrase that is contained in the text of the input discourse, or a nominal or plural form of a word. However, the word defining the theme concept may not appear in the input discourse. Instead, the theme concept is conceptualized from the theme term as set forth in a particular ontology stored in the knowledge catalog 100.

In a preferred embodiment, each theme in the theme vector output 500 is assigned a normalized or relative theme strength. The normalized theme strength is calculated using the theme strengths of each theme term in the text that contribute to the particular theme. At the sentence level, the theme vector processor 250 calculates a theme strength for each word in the sentence. The theme strength of the word measures the thematic weight of the word within the context of the sentence in which it appears. Prior to processing, each word receives a default theme strength that varies depending upon the word. As the theme vector processor 250 analyzes the surrounding text in the sentence to determine the thematic role of the word, it increases or decreases the default theme strength for the word. For example, a theme grading determines how each word in a sentence is functioning within the context and structure of the sentence. The theme grading assigned to a word affects the theme strength for the word. For a description of theme gradings, see Appendix H, entitled "Oracle ConText.TM. Linguistics Toolkit, Guide and Reference, Release 1.1", pages 2-8 to 2-13.

In the preferred embodiment, the theme vector processor 250 executes a plurality of heuristic routines to generate the theme strengths for each theme. Appendix I, entitled "Theme Vector Code", contains a source code listing for implementing theme vector processing in accordance with one embodiment. Appendix I contains source code for generating the theme strengths in accordance with one embodiment. The source code disclosed in Appendix I includes detailed comments to explain the operation of the code. Although the generation of the theme strengths is disclosed as computer source code, the theme strengths may be generated by any means, such as through a hardware implementation, without deviating from the spirit or scope of the invention. Thus, a theme vector identifies the 16 strongest themes in a sentence, paragraph or document in proportion to the sentence, paragraph or document for which the theme vector output is generated.

The theme vector processor 250 returns the concept, if one exists, for each theme term in the theme vector. Similar to the theme terms, the theme concepts in the theme vector output 730 contain a theme strength. As the theme vector processor 250 identifies themes that have the same concept in the input discourse, the theme strength for the theme concept increases. This indicates that the theme concept is significantly developed in the input discourse. As a result, the theme vector processor 250 may promote the theme concept to a separate theme in the theme vector output 730 for the corresponding input discourse. For example, if the theme vector processor 250 determines that "printing press" is one of the theme terms in a paragraph, then the theme concept "printing" is assigned to the theme term "printing press." In addition, the theme vector processor 250 assigns a theme strength to the theme concept. The theme vector processor 250 increases the theme strength for "printing" if the paragraph contains other words such as "bold face" and "typesetting", that conceptualize to "printing." The theme concept "printing" may then appear as a theme term for the paragraph, along with its own theme concept, "publishing industry." In a preferred embodiment, the theme vector processor 250 includes input adjustments (not shown) to prevent promoting theme concepts in a theme vector output 730 to additional theme terms.

The following example illustrates a theme vector output generated by the theme vector processor 250 for input discourse, such as a paragraph. The example paragraph is from a document about commercial banks lowering interest rates based on changes to the Federal funds rate.

Treasury bill yields dropped substantially Friday morning in anticipation of further reductions in the Federal funds rate by the Fed, market watchers said. The 3-month bill fell 8 basis points to a discount equivalent rate of 7.70%, while the 1-year bill was down 12 basis points to 7.30%.

Table 1 illustrates the theme vector output for the example paragraph. As shown in Table 1, for each theme term, the theme vector output 30 includes a theme strength and a theme concept.

                  TABLE 1
    ______________________________________
    #   Theme Strength
                     Theme Terms   Theme Concept
    ______________________________________
     1  43           banking       finance and investment
     2  25           basis points  stocks, bonds, and
                                   commodities
     3  24           treasury bill yields
                                   banking
     4  22           stocks, bonds, and
                                   finance and investment
                     commodities
     5  22           points        stocks, bonds, and
                                   commodities
     6  21           yields        banking
     7  17           bills         bills
     8  12           federal funds rates
                                   banking
     9  11           reductions    banking
    10  10           rates         banking
    11   9           discount equivalent
                                   commerce and trade
                     rates
    12   9           three-month   three-month
    13   8           1-year        1-year
    14   8           rates         commerce and trade
    15   7           discounts     commerce and trade
    16   7           equivalents   equivalencies
    ______________________________________

                  TABLE 2
    ______________________________________
    Dynamic Classification Index
    applications   operating systems
    for OS/2 software
                   installation
    Windows        OS/2
    methods of converting
                   proprietary
    for use        ›see also! applications
    ›see also! operating systems
                   ›see also! computer software industry
                   ›see also! softwares
                   ›see also! systems
    applications programs
                   OS/2 operating systems
    applications softwares
                   OS/2 software
                   applications for
    commercial values
                   patents
                   software
                   ›see also! legal customs and formalities
    computer hardware industry
                   platforms
    ›see also! platforms
                   computer system
                   ›see also! computer hardware industry
    computer industries
                   programs
                   applications
                   computer software
                   ›see also! computer software industry
    computer industry
                   proprietary operating systems
    ›see also! computers
    computer software industry
                   software
    ›see also! operating systems
                   OS/2
    ›see also! programs
                   ›see also! computer software industry
    ›see also! software
    ›see also! systems
    computer software programs
                   software patents
    computer system platforms
                   software programs
                   computer
    computers      softwares
    examples of    Windows applications
    ›see also! computer industry
                   amounts of
                   ›see also! operating systems
    converting     system platforms
    methods of     computer
                   varieties of
    electronics    systems
    ›see also! industries
                   operating
                   ›see also! computer software industry
                   ›see also! operating systems
    examples       UNIX, Windows and OS/2
    of computers   inclusion
                   ›see also! operating systems
    industries     Windows applications
    computer       methods of converting
    ›see also! electronics
                   for use
    legal customs and formalities
    ›see also! patents
    ______________________________________