Multilingual

Document processing apparatus having an authoring capability for describing a document structure

6981218

Abstract

An apparatus and method are disclosed for easily generating document data (tag file) in a form that makes it possible to perform various processes upon the document data. An original document (plain text) is divided into morphological elements, and morphological information is added thereto. Information representing the hierarchical document structures is also added. Furthermore information indicating referential relations between portions in the original document is also added.


Claims

1. A document processing apparatus comprising:

automatic analysis means for automatically analyzing an electronic document and attaching hierarchical structure information representing a document structure to said electronic document in accordance with the result of said automatic analysis, said automatic analysis means automatically analyzes the document structure of said electronic document in the order from a lowest hierarchical level to a highest hierarchical level;

information presenting means for presenting information about the electronic document including said structure information at each hierarchical level so that a user may correct internal information associated with said electronic document on the basis of said information displayed on a display; and

correction means for correcting said internal information associated with said electronic document in response to an operation performed by the user in accordance with the internal information displayed on the display, said correction means corrects the internal structure of said electronic document by adding, removing, or modifying internal information in the order from the lowest hierarchical level to the highest hierarchical level.

2. A document processing apparatus according to claim 1, wherein said automatic analysis means comprises morpheme dividing means for dividing said electronic document into morphemes and morphological information attaching means for attaching morphological information to each said morpheme.

3. A document processing apparatus according to claim 2, wherein when candidates of internal information are attached by said automatic analysis means to an electronic document, said information presenting means presents information for prompting a user to select one of said candidates of internal information.

4. A document processing apparatus according to claim 3, wherein said candidates of internal information represent different manners in which said electronic document is divided into morphemes.

5. A document processing apparatus according to claim 3, wherein said candidates of internal information represent different document structures.

6. A document processing apparatus according to claim 3, wherein said candidates of internal information represent different referential relations between portions of said electronic document.

7. A document processing method comprising the steps of:

attaching structure information representing a document structure to an electronic document by automatically analyzing said electronic document in the order from a lowest level to a highest level of the hierarchy of the document structure;

presenting information about the electronic document including said structure information at each level of the hierarchy so that a user may correct internal information associated with said electronic document on the basis of said information displayed on a display; and

correcting said internal information associated with said electronic document at each level of the hierarchy of the document structure in response to an operation performed by the user in accordance with the internal information displayed on the display, wherein said correction step corrects the internal information associated with said electronic document by adding, removing, or modifying internal information.

8. A document processing method according to claim 7, wherein said step of attaching structure information includes the steps of dividing said electronic document into morphemes and attaching morphological information to the respective morphemes.

9. A document processing method according to claim 8, wherein if candidates of internal information are attached in said step of attaching structure information, said step of presenting information presents information so as to prompt a user to select one of said candidates of internal information.

10. A document processing method according to claim 9, wherein said candidates of internal information represent different manners in which said electronic document is divided into morphemes.

11. A document processing method according to claim 9, wherein said candidates of internal information represent different document structures.

12. A document processing method according to claim 9, wherein said candidates of internal information represent different referential relations between portions of said electronic document.

13. A storage medium including a computer-controllable program stored thereon, said program comprising the steps of:

automatically analyzing an electronic document and attaching hierarchical structure information representing a document structure to an electronic document in accordance with the result of said automatic analysis, wherein said hierarchical structure comprises an order from a lowest hierarchical level to a highest hierarchical level;

presenting information about the electronic document including said structure information at each hierarchical level so that a user may correct internal information associated with said electronic document on the basis of said information displayed on a display; and

correcting said internal information associated with said electronic document by adding, removing, or modifying internal information in the order from the lowest hierarchical level to the highest hierarchical level in response to an operation performed by the user in accordance with the internal information displayed on the display.

14. A storage medium including a computer-controllable program stored thereon, according to claim 13, wherein said step of attaching structure information includes the steps of dividing said electronic document into morphemes and attaching morphological information to the respective morphemes.

15. A storage medium including a computer-controllable program stored thereon, according to claim 14, wherein if candidates of internal information are attached in said step of attaching structure information, said step of presenting information presents information so as to prompt a user to select one of said candidates of internal information.

16. A storage medium including a computer-controllable program stored thereon, according to claim 15, wherein said candidates of internal information represent different manners in which said electronic document is divided into morphemes.

17. A storage medium including a computer-controllable program stored thereon, according to claim 15, wherein said candidates of internal information represent different document structures.

18. A storage medium including a computer-controllable program stored thereon, according to claim 15, wherein said candidates of internal information represent different referential relations between portions of said electronic document.


Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to a document data generating apparatus for generating/processing electronic document data.

2. Description of the Related Art

WWW (World Wide Web) is widely used to supply hypertext information via the Internet.

The WWW is a system that allows electronic document to be treated in a new manner, that is, generated, processed, disclosed, and used in common. However, from the point of view of practically using documents, the WWW has a limitation in the capability of processing documents. Thus, there is a need for a higher-level document processing technique such as categorization or summarization of documents. In order to realize such high-level document processing, it is necessary to automatically process the contents of documents.

However, such automatic processing of the contents of documents has difficulties as described below.

Firstly, the HTML (Electronic Markup Language) prescribes the manner of representing documents, but does not prescribe the contents of the documents. Secondly, it is not necessarily easy for users to understand the contents of documents that are linked to one another via a hypertext network. Thirdly, authors usually write documents without bearing in mind the convenience of readers, and no adjustment is made as to the difference in convenience between authors and readers.

Although the WWW is a new electronic documentation system having various advantages, the WWW is not capable of performing high-level document processing which needs additional automatic processing. In other words, in order to realize the high-level document processing, it is required to automatically process documents.

To the above end, systems for assisting in automatically processing a document has been developed on the basis of natural language processing technology. One such method is to automatically process a document according to tags which have been attached, by the author of the document or other person, to the document so as to represent attribute information about the internal structure of the document.

In recent years, computers have become increasingly popular, and many computers are connected to one another via a network. As a result, there occurs a need for a higher-level document processing technique to perform generation of a text document, labeling, and a modification of a text document, in accordance with an index depending upon the content of a document. More specifically, there is a need for a technique to summarize or categorize a document in response to a request issued by a user.

To the above end, document data or a document file supplied to a user should include information required to process the document data. Thus, there is a need for an authoring technique for generating document data including such information.

The authoring technique should be easily used not only by users having high-level knowledge but also general users who do not have high-level knowledge.

SUMMARY OF THE INVENTION

Thus, it is an object of the present invention to provide an apparatus and method for processing a document in a manner that satisfies the above-described requirements. According to an aspect of the present invention, to achieve the above object, there is provided a document processing apparatus comprising: automatic analysis means for automatically analyzing an electronic document and attaching structure information representing a document structure to the electronic document in accordance with the result of the automatic analysis; information presenting means for presenting information about the electronic document including the structure information so that a user may correct internal information associated with the electronic document on the basis of the information displayed on a display; and correction means for correcting the internal information associated with the electronic document in response to an operation performed by the user in accordance with the internal information displayed on the display.

In this document processing apparatus according to the present invention, the automatic analysis means preferably comprises morpheme dividing means for dividing the electronic document into morphemes and morphological information attaching means for attaching morphological information to each morpheme.

When candidates of internal information are attached by the automatic analysis means to an electronic document, the information presenting means may present information for prompting a user to select one of the candidates of internal information.

The candidates of internal information may represent different manners in which the electronic document is divided into morphemes.

The candidates of internal information may represent different document structures.

The candidates of internal information may represent different referential relations between portions of the electronic document.

The correction means may correct the internal information associated with the electronic document by adding, removing, or modifying internal information.

Preferably, the automatic analysis means automatically analyzes the document structure of the electronic document in the order from the lowest hierarchical level to the highest hierarchical level, and the correction means corrects the internal structure of the electronic document in the order from the lowest hierarchical level to the highest hierarchical level.

According to another aspect of the present invention, there is provided a document processing method comprising the steps of: attaching structure information representing a document structure to the electronic document by automatically analyzing the electronic document; presenting information about the electronic document including the structure information so that a user may correct internal information associated with the electronic document on the basis of the information displayed on a display; and correcting the internal information associated with the electronic document in response to an operation performed by the user in accordance with the internal information displayed on the display.

In this document processing method according to the present invention, the step of attaching structure information preferably includes the steps of dividing the electronic document into morphemes and attaching morphological information to the respective morphemes.

If candidates of internal information are attached in the step of attaching structure information, the step of presenting information may present information so as to prompt a user to select one of the internal information.

The candidates of internal information may represent different manners in which the electronic document is divided into morphemes.

The candidates of internal information may represent different document structures.

The candidates of internal information may represent different referential relations between portions of the electronic document.

The correction step may correct the internal information associated with the electronic document by adding, removing, or modifying internal information.

Preferably, the step of attaching structure information automatically analyses the electronic document as to the document structure in the order from the lowest level to the highest level of the hierarchy of the document structure, and the correction step corrects the internal structure of the electronic document in the order from the lowest level to the highest level of the hierarchy of the document structure.

According to still another aspect of the present invention, there is provided a storage medium including a computer-controllable program stored thereon, the program comprising the steps of: automatically analyzing an electronic document and attaching structure information representing a document structure to the electronic document in accordance with the result of the automatic analysis; presenting information about the electronic document including the structure information so that a user may correct internal information associated with the electronic document on the basis of the information displayed on a display; and correcting the internal information associated with the electronic document in response to an operation performed by the user in accordance with the internal information displayed on the display.

In this storage medium according to the present invention, the step of attaching structure information preferably includes the steps of dividing the electronic document into morphemes and attaching morphological information to the respective morphemes.

If candidates of internal information are attached in the step of attaching structure information, the step of presenting information may present information so as to prompt a user to select one of the candidates of internal information.

The candidates of internal information may represent different manners in which the electronic document is divided into morphemes.

The candidates of internal information may represent different document structures.

The candidates of internal information may represent different referential relations between portions of the electronic document.

The correction step may correct the internal information associated with the electronic document by adding, removing, or modifying internal information.

Preferably, the step of attaching structure information automatically analyses the electronic document as to the document structure in the order from the lowest level to the highest level of the hierarchy of the document structure.

According to still another aspect of the present invention, there is provided a signal carrying a computer-controllable program, the program comprising the steps of: automatically analyzing an electronic document and attaching structure information representing a document structure to the electronic document in accordance with the result of the automatic analysis; presenting information about the electronic document including the structure information so that a user may correct internal information associated with the electronic document on the basis of the information displayed on a display; and correcting the internal information associated with the electronic document in response to an operation performed by the user in accordance with the internal information displayed on the display.

In this signal according to the present invention, the step of attaching structure information preferably includes the steps of dividing the electronic document into morphemes and attaching morphological information to the respective morphemes.

If candidates of internal information are attached in the step of attaching structure information, the step of presenting information may present information so as to prompt a user to select one of the candidates of internal information.

The candidates of internal information may represent different manners in which the electronic document is divided into morphemes.

The candidates of internal information may represent different document structures.

The candidates of internal information may represent different referential relations between portions of the electronic document.

The correction step may correct the internal information associated with the electronic document by adding, removing, or modifying internal information.

Preferably, the step of attaching structure information automatically analyses the electronic document as to the document structure in the order from the lowest level to the highest level of the hierarchy of the document structure.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram illustrating an embodiment of a document data providing system according to the present invention;

FIG. 2 is a block diagram illustrating an embodiment of a document processing apparatus according to the present invention;

FIG. 3 is a schematic diagram illustrating a document structure employed in the embodiment of the invention;

FIG. 4 is a schematic diagram illustrating a window for displaying a sentence structure according to the embodiment of the invention;

FIG. 5 is a flow chart illustrating a manual categorization process according to the embodiment of the invention;

FIG. 6 is a flow chart illustrating an indexing process according to the embodiment of the invention;

FIG. 7 is a schematic diagram illustrating activation values of elements used in the embodiment of the invention;

FIG. 8 is a flow chart illustrating an activation value spreading process according to the embodiment of the invention;

FIG. 9 is a flow chart illustrating a process of updating an activation value according to the embodiment of the invention;

FIG. 10 is a schematic diagram illustrating a categorization window according to the embodiment of the invention;

FIG. 11 is a schematic diagram illustrating a browser window according to the embodiment of the invention;

FIGS. 12A and 12B are tables illustrating a categorization model according to the embodiment of the invention;

FIG. 13 is a flow chart illustrating an automatic categorization process according to the embodiment of the invention;

FIG. 14 is a flow chart illustrating an automatic categorization step according to the embodiment of the invention;

FIG. 15 is a flow chart illustrating a process of calculating a word sense relevance values according to the embodiment of the invention;

FIG. 16 is a table illustrating word sense relevance values according to the embodiment of the invention;

FIG. 17 is a schematic diagram illustrating an example of a browser window according to the embodiment of the invention;

FIG. 18 is a schematic diagram illustrating an example of a browser window in which a summary is displayed, according to the embodiment of the invention;

FIG. 19 is a flow chart illustrating a process of generating a summary according to the embodiment of the invention;

FIG. 20 is a flow chart illustrating a process of reading aloud a document according to the embodiment of the invention;

FIG. 21 is a flow chart illustrating a process of generating a read-out file according to the embodiment of the invention;

FIGS. 22A and 22B are schematic diagrams illustrating another example of a tag file according to the embodiment of the invention;

FIG. 23 is a schematic diagram illustrating another example of a tag file according to the embodiment of the invention;

FIGS. 24A and 24B are schematic diagrams illustrating another example of a read-out file according to the embodiment of the invention;

FIG. 25 is a schematic diagram illustrating another example of a read-out file according to the embodiment of the invention;

FIG. 26 is a schematic diagram illustrating a read-out window according to the embodiment of the invention;

FIG. 27 is a block diagram illustrating an embodiment of an authoring apparatus according to the present invention;

FIG. 28 is a flow chart illustrating an authoring process according to an embodiment of the invention;

FIGS. 29A an 29B are schematic diagrams illustrating an example of a plain text that is displayed on a display and that is to be subjected to the authoring process according to the embodiment of the invention;

FIG. 30 is a schematic diagram illustrating an example of a text displayed on the display after being subjected to morphological analysis in the authoring process according to the embodiment of the invention;

FIG. 31 is a schematic diagram illustrating an example of a manner of displaying candidates in terms of morphological elements during the authoring process according to the embodiment of the invention;

FIG. 32 is a schematic diagram illustrating an example of a text displayed on the display after being determined in terms of morphological elements during the authoring process according to the embodiment of the invention;

FIG. 33 is a schematic diagram illustrating an example of a manner of displaying an undefined word during the authoring process according to the embodiment of the invention;

FIG. 34 is a schematic diagram illustrating an example of a manner of presenting a subwindow for processing an undefined word during the authoring process according to the embodiment of the invention;

FIG. 35 is a schematic diagram illustrating an example of a manner of processing an undefined word in the subwindow during the authoring process according to the embodiment of the invention;

FIG. 36 is a schematic diagram illustrating an example of a text displayed after defining an undefined word during the authoring process according to the embodiment of the invention;

FIG. 37 is a schematic diagram illustrating an example of a text displayed after completion of morphological analysis during the authoring process according to the embodiment of the invention;

FIG. 38 is a schematic diagram illustrating an example of a text including tags representing document structures added during the authoring process according to the embodiment of the invention;

FIG. 39 is a schematic diagram illustrating an example of a manner of displaying candidates for words modified by a modifier, during the authoring process according to the embodiment of the invention;

FIG. 40 is a schematic diagram illustrating an example of a manner of adding a tag using a subwindow during the authoring process according to the embodiment of the invention;

FIG. 41 is a schematic diagram illustrating an example of a manner of displaying a heading and tags associated with sentences during the authoring process according to the embodiment of the invention;

FIG. 42 is a schematic diagram illustrating an example of a manner of displaying a text after being tagged during the authoring process according to the embodiment of the invention; and

FIG. 43 is a schematic diagram illustrating an example of a manner of displaying words cataphorically referred to by another word, during the authoring process according to the embodiment of the invention.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

The present invention is described in further detail below with reference to preferred embodiments, as to the following items.

First, the configuration of a document data providing system is described. After that, a document processing apparatus, that is included in the document data providing system and that receives document data, is described. The data structure of document data is then described. Thereafter, described is an embodiment of an authoring apparatus for generating document data to be supplied to the document processing apparatus according to the present invention. More specifically, the following items will be described:

1. Configuration of Document Data Providing System

2. Configuration of Document Processing Apparatus

3. Document Data Structure

4. Manual Categorization of Document Data

    • 4.1 Procedure
    • 4.2 Indexing
    • 4.3 Browsing, Generation of Categories, and Categorization
    • 4.4 Creation/Registration of the Categorization Model


    • 5. Automatic Categorization of Document Data
    • 5.1 Procedure
    • 5.2 Automatic Categorization


    • 6. Summarization

    7. Reading Aloud Document

    8. Configuration of Authoring Apparatus

    9. Authoring Process

    1. Configuration of Document Data Providing System

    FIG. 1 illustrates an example of the configuration of a document data providing system.

    In the present embodiment, the document data providing system includes mainly a document processing apparatus 1, a server 3, an authoring apparatus 2, and a document provider 4.

    The document provider 4 provides original text data (original document) including no tags that will be described later (hereinafter, such a text including no tags will be referred to as a plain text).

    In the present embodiment, the document provider 4 has a document generator 4a for generating a plain text.

    Note that the document provider 4 is not necessarily required to have the document generator 4a, and what is essential to the document provider 4 is to provide a plain text. Instead of generating a document, the document provider 4 may provide a plain text received from an external document producer.

    The document provider 4 may store a plain text, to be provided, on a storage medium 81 such as a floppy disk or an optical disk whereby the plain text may be provided to the authoring apparatus 2 or other apparatus.

    Alternatively, the document provider 4 may transmit a plain text to the authoring apparatus 2 via a cable communication medium (such as a public telephone line, a private communication line, or the Internet) or a wireless communication medium (such as a satellite communication line or a wireless telephone line).

    The authoring apparatus 2 includes an authoring processor 2a, a document generator 2b, and an authoring program 2c. The hardware configuration of the authoring apparatus 2 and the operation thereof will be described later.

    When the authoring apparatus 2 receives a plain text via the storage medium 81 or the communication line 6, the authoring apparatus 2 may perform an authoring process upon the received plain text thereby generating document data to be supplied to the document processing apparatus 1 serving as a user terminal that will be described later. Herein, the "document data" is generated by adding various tags as will be described later. Thus, such document data will also be referred to as a tag file.

    The authoring processor 2a performs the authoring process in accordance with the authoring program 2c.

    When the authoring apparatus 2 has the document generator 2b, the authoring apparatus 2 is capable of generating a plain text without necessarily having to receive a plain text from the document provider 4, and the authoring apparatus 2 is capable of performing an authoring process upon the generated plain text thereby generating document data (tag file).

    The authoring program 2c may be stored in advance in an internal storage device (RAM, ROM, hard disk) in the authoring apparatus 2. Alternatively, an external authoring program 5 may be downloaded into the authoring apparatus 2 via the storage medium 81 or the communication line 6 so that the authoring processor 2a may operate in accordance with the downloaded authoring program 5.

    The document data generated by the authoring apparatus 2 is supplied to the server 3 and stored in a database 3a. Although not shown in FIG. 1, the supplying of document data from the authoring apparatus 2 to the server 3 may also be achieved using a storage medium or a communication line.

    The server 3 supplies document data stored in the database 3a to an end user's apparatus such as the document processing apparatus 1 via a storage medium 32 such as a floppy disk or an optical disk or a communication line 6.

    An end user may perform various processes upon received document data using the document processing apparatus 1 as will be described later thereby obtaining various high-level documents.

    Note that FIG. 1 shows only one example of the system configuration, and a practical system may be configured in various manners.

    For example, the system may include a large number of document providers 4, authoring apparatus 2, and servers 3. An authoring apparatus 2 may be disposed on the side of a document provider 4. Furthermore, a document provider 4, an authoring apparatus 2, and a server 3 may be combined into a single apparatus.

    2. Configuration of Document Processing Apparatus

    The document processing apparatus 1, that is included in the document data providing system and that processes received document data, is described in further detail below.

    As shown in FIG. 2, a document processing apparatus includes a main unit 10 including a controller 11 and an interface 12, an input unit 20 used by a user to input data or a command to the main unit 10, a communication device 21 for transmitting and receiving a signal to or from an external device, a display unit 30 for displaying an output from the main unit 10, a write/read unit 31 for writing and reading information onto and from a recording medium 32, an audio output unit 33, and an HDD (hard disk drive) 34.

    The main unit 10 including the controller 11 and the interface 12 serves as the core of the document processing apparatus 1.

    The controller 11 includes a CPU 13 for processing a document, a RAM 14 serving as a volatile memory, and a ROM 15 serving as a nonvolatile memory.

    The CPU 13 executes a program in accordance with a procedure stored in the ROM 15, wherein the CPU 13 temporarily stores data in the RAM 14 if necessary.

    Operations performed by the controller 11 include, as will be described in detail later, categorization of given document data, summarization, generation of a file used to output data by voice, and document analysis required in the above operations. Programs and application software required for the above operations are stored in the ROM 15, the HDD 34, or the storage medium 32.

    As described above, the document processing program used by the controller 11 may be stored in advance in the ROM 15 or may be loaded from the storage medium 32 or the HDD 34. Alternatively, the document processing program may be downloaded from an external server via the communication device 21 (communication line 6) and a network such as the Internet.

    The interface 12 is connected to the controller 11, the input unit 20, the communication device 21, the display 30, the write/read unit 31, the audio output unit 33, and the HDD 34.

    Under the control of the controller 11, the interface 12 inputs data via the input unit 20, inputs and outputs data from and to the communication device 21, outputs data to the display 30, inputs and outputs data from and to the write/read unit 31, outputs data to the audio output unit 33, and inputs and outputs data from and to the HDD 34. In the above operation, the interface 12 adjusts timing of inputting or outputting data between various parts described above and also converts data format as required.

    The input unit 20 is used by a user to input data or a command to the document processing apparatus 1. The input unit 20 may include a keyboard and a mouse. Using the input unit 20, the user may enter a keyword via the keyboard. The user may also select, using the mouse, an element of an electronic document displayed on the display 30.

    Hereinafter, an electronic document handled by the document processing apparatus 1 is also referred to simply as a document. Furthermore, the term "element" is used to describe various elements of a document. Elements include a document itself, a sentence, and a word.

    The communication device 21 serves to receive a signal that is transmitted by an external apparatus to the document processing apparatus 1 via a communication line 6. The communication device 21 also serves to transmit a signal over the communication line 6.

    If the communication device 21 receives one or more document data from an external apparatus such as a server 3, the communication device 21 transfers the received document data to the main unit 10. The communication device 21 also transmits data to an external apparatus via the communication line 6.

    The display 30 serves as an output device of the document processing apparatus 1, for displaying characters and/or image information. The display 30 may include a cathode ray tube (CRT) or a liquid crystal display (LCD). The display 30 may display one or more windows in which characters and/or graphic images are displayed.

    The write/read unit 31 serves to write and read data to and from a storage medium 32 such as a floppy disk or an optical disk.

    Although in the present embodiment, a floppy disk (magnetic disk) or an optical disk is employed as the storage medium 32, other types of removable storage media such as a magnetooptical disk, a memory card, and a magnetic tape may also be employed. As for the write/read unit 31, a device (such as a disk drive or a card drive) adapted to writing/reading data to and from an employed medium may be used.

    In the case where a document processing program to be used to process a document is stored on the storage medium 32, the write/read unit 31 may read the document processing program from the storage medium 32 and transfer it to the controller 11.

    When document data is stored on the storage medium 32, the write/read unit 31 may read such a data from the storage medium 32 and transfer it to the controller 11. This provides another way for the document processing apparatus 1 to acquire document data.

    Furthermore, after processing document data by the document processing apparatus 1, the controller 11 may store the resultant document data on the storage medium 32 using the write/read unit 31.

    The audio output unit 33 serves as an output device of the document processing apparatus 1, for providing a voice output corresponding to a document.

    More specifically, the audio output unit 33 outputs a voice signal synthesized by the controller in accordance with document information (read-out file) that will be described later.

    The HDD 34 serves as a mass storage device used by the document processing apparatus 1 to store a large amount of data. The HDD 34 writes and reads information under the control of the controller 11.

    The HDD 34 is used to store various application programs such as a voice synthesis program executed by the controller 11. The HDD 34 may also be used to store document data input to the document processing apparatus 1.

    3. Document Data Structure

    The data structure of document data is described below. In the present embodiment, a document is processed in accordance with attribute information described by a tag attached to a document. Tags used in the present embodiment include a syntactic tag and a semantic/pragmatic tag wherein the syntactic tag indicates the structure of a document and the semantic/pragmatic tag makes it possible for a machine to understand the contents of documents written in various languages.

    A syntactic tag may be used to described the internal structure of a document.

    The internal structure, to be represented by tags, includes elements such as a document, sentences, and words which are linked to one another by normal links or reference links, as shown in FIG. 3.

    In FIG. 3, open circles represent elements. Open circles at the bottom represent elements in the lowest level in a document. Solid lines represent normal links indicating direct connections between elements such as sentences or words. Broken lines represent reference links indicating dependence between elements.

    The internal structure of a document is composed of, in order from the highest level to the lowest level, a document, a subdivision, a paragraph, a sentence, a subsentential segment, . . . , and a word element, wherein the subdivision and the paragraph are optional.

    Tags may also be used for the semantic purpose. For example, when a word has a plurality of senses (meanings), a tag may be used to specify a particular sense.

    In the present embodiment, tagging is performed according to XML (Extensible Markup Language) similar to HTML (Hyper Text Markup Language).

    Some examples of tagging are described below. In these examples, tags are described within brackets < >. Note that tagging is not limited to those examples but tagging may be performed in many different ways. Two specific examples of documents including tags are shown below, where one of the example is written in English and the other is written in Japanese. Note that internal structures of documents written in other languages can also be described using tags.

    When a sentence "Time flies like an arrow." is given, tagging may be performed as follows.

    <sentence><noun phrase: word sense="time0">time</noun phrase><verb phrase><verb: word sense="fly1">flies</verb><adverb phrase><adverb: word sense=like0>like</adverb><noun phrase>an<noun: word sense="arrow0">arrow</noun></noun phrase></adverb phrase></verb phrase>.</sentence>

    In the above example, <sentence>, <noun>, <noun phrase>, <verb>, <verb phrase>, <adverb>, and <adverb phrase> are used to indicate a sentence, a noun, a noun phrase, a verb, a verb phrase, an adjective/adverb (including preposition and postposition phrases), and an adjective/adverb phrase, respectively. That is, the syntactic structure of the sentence is described by those tags.

    A start tag is placed immediately before an element and a corresponding end tag is placed immediately after that element. Herein, end tags placed immediately after the respective elements include a symbol "/" to indicate that the tags are end tags. The term "element" is used herein to describe a syntactic element such as a phrase, a paragraph, or a sentence.

    The expression, word sense="time0", indicates that word "time" is used herein to describe the 0th sense of a plurality of senses of word "time". More specifically, although "time" has senses as a noun, an adjective, and a verb, "time" is used herein as a noun (first sense). Similarly, word "orange" has three senses, namely, the name of a plant, one of colors, and one of fruits, which can be distinguished from each other by specifying a word sense.

    In the present embodiment, the syntactic structure of document data may be displayed in a window 101 on the display 30, as shown in FIG. 4. In the window 101, word elements are displayed in a subwindow 103 on the right side, and the internal structure of a sentence is displayed in a subwindow on the left side.

    In this specific example in FIG. 3, a part of a sentence "Acustom characterBcustom characterCcustom charactercustom charactercustom charactercustom charactercustom charactercustom charactercustom charactercustom charactercustom charactercustom character" (Convention B was held in C City under the leadership of Mr. A. Some of newspaper companies, including usual and popular newspaper companies, have announced, on their papers that they will restrict themselves in terms of insertion of photographs of Mr. A.) is shown in the window 101. This document may be tagged, for example, as follows.

    <document><sentence><|adverb phrase: relation="place"><noun phrase><adverb phrase: place="Ccustom character"><document><sentence><adverb phrase: relation="place"><noun phrase><adverb phrase: place="Ccustom character"><adverb phrase: relation="subject"> <noun phrase: identifier="Bcustom character"><adverb phrase: relation="position"><person name: identifier="Acustom character">Acustom character</person name></adverb phrase><organization name: identifier="Bcustom character">Bcustom character</organization name></noun phrase>custom character</adverb phrase>custom character</adverb phrase><place name: identifier="Ccustom character">Ccustom character</place name></noun phrase>custom character</adverb phrase><adverb phrase: relation="subject"><noun phrase: identifier="press"; syntax="parallel"><noun phrase><adverb phrase>custom character</adverb phrase>custom character</noun phrase>custom character<noun>custom character</noun></noun phrase>custom character</adverb phrase><adverb phrase: relation="object"><adverb phrase: relation="content"; subject="press"><adverb phrase: relation="object"><noun phrase><adverb phrase><noun: coreference="Bcustom character">custom character</noun>custom character</adverb phrase>custom character</noun phrase>custom character</adverb phrase>custom charactercustom character</adverb phrase>custom character</adverb phrase><adverb phrase: relation="location">custom character</adverb phrase>custom charactercustom character</sentence></document>

    As can be seen, the structure of the document is described by pairs of tags <***> and </***>.

    For example, a pair of tags <document> and </document> indicates the range of a document, and a pair of tags <sentence> and </sentence> indicates the range of a sentence. A pair of tag <noun phrase: identifier="Bcustom character"> and </noun phrase> is used to describe a noun phrase "Acustom characterBcustom character" with an identifier "Bcustom character".

    Thus, the internal structure of the sentence is described by tags as shown in the subwindow on the left side of FIG. 4.

    In the above sentence, syntax="parallel" indicates that "custom character" and "custom character" are parallel in relation. Herein, "parallel" elements are such elements having the same dependency. When no particular dependency is specified, "<noun phrase: relation=x><noun>A</noun><noun>B</noun></noun phrase>" indicates that A depends on B. Expression, relation=x, indicates a relational attribute.

    A relational attribute describes a relation between elements in terms of syntax, meaning, and rhetoric. More specifically, a relation attribute describes a grammatical function such as a subject, an object, and an indirect object, a theme/role such as an acting person, a person receiving an action, and a beneficiary, and a rhetorical relation such as a reason and a result. In the present example, relatively simple syntactic functions such as a subject, object, and indirect object are described by relational attributes.

    Furthermore, in the present example, the attributes of proper nouns such as "Acustom character", "Bcustom character", and "Ccustom character" ("Mr. A", "Convention B", "City C") are described by tags <place name>, <person name>, and <organization name>. By attaching a tag <place name>, <person name>, or <organization name>, it is possible to indicate that a tagged word is a proper noun.

    4. Manual Categorization of Document Data

    4.1 Procedure

    In the document processing apparatus 1 of the present embodiment, when document data is input from the outside via the communication device 21 (or via the write/read unit 31), the document data is categorized in accordance with the content thereof. Although in the following description, document data is assumed to be input from the outside via the communication device 21, the categorization may also be performed in a similar manner when document data is input from a removable medium such as a floppy disk via the write/read unit 31.

    In general, categorization is performed either in a manual fashion by a user in accordance with the content of given document data or in an automatic fashion by the document processing apparatus 1.

    Categorization is performed on the basis of a categorization model that will be described later. In the initial state, the document processing apparatus 1 has no categorization model. Therefore, when the document processing apparatus 1 is in the initial state, it is required to manually generate a categorization model and perform categorization. If a categorization model has been generated, it becomes possible to automatically categorize given document data.

    First, the manual categorization process to be performed initially is described. That is, when the document processing apparatus in the initial state receives document data from the outside, the manual categorization process is performed by the controller 11 in accordance with an operation performed by a user so as to generate a categorization model and categorize the document data.

    The outline of the manual categorization process is first described with reference to FIG. 5, and then each step of the process is described in further detail.

    In step F11 in FIG. 5, the receiver 21 of the document processing apparatus 1 receives a document. In this step F11, the receiver 21 receives one or more documents via, for example, a communication line. The received one or more documents are transferred to the main unit 10 of the document processing apparatus 1. The controller 11 stores the one or more documents into RAM 14 or the HDD 34.

    In step F12, the controller 11 of the document processing apparatus 1 extracts words characterizing the plurality of documents received via the receiver 21 and generates an index for each document. The controller 11 stores the generated index 11 in the RAM 14 or the HDD 34.

    As will be described later, the index of each document includes a proper noun and/or other words that characterize the document. Therefore, categorization or retrieval can be performed using an index.

    In step F13, a user reads a document as required. In this step, the document processing apparatus 1 performs an operation in response to a command issued by the user. The next step F14 is also performed in response to an operation of the user.

    The document data input to the document processing apparatus 1 is displayed on the screen of the display 30 in response to a command issued by the user so that the user can read it.

    When the user reads a document, the user may issue various commands by clicking an icon or the like on the screen so as to perform various processes such as summarization that will be described later. When the user reads a document in the manual categorization process, step F14 is performed to generate categories and categorize the document.

    In step F14, the controller 11 generates and displays categories in accordance with an operation performed by the user. The user then specifies a category for each document data. In response, the controller 11 categorizes and displays document data.

    In step F15, the controller 11 generates a categorization model on the basis of categories generated by the user in step F14 and also on the basis of categorization performed by the user for each document data.

    The categorization model includes data that represents correspondence between categories and elements of indexes (generated in step F12) of respective documents. That is, the categorization model represents how documents are categorized.

    In step F16, the resultant categorization model is registered. The registration is performed by the controller 11 by storing the categorization model in the RAM 14.

    By performing the process shown in FIG. 5 in the above-described manner, one or more document data input to the document processing apparatus 1 in the initial state are manually categorized, and a categorization model is generated.

    The respective steps in the process shown in FIG. 5 are described in further detail below.

    4.2 Indexing

    In step F14, the controller 11 generates an index for each document data input.

    A specific example of an index generated for certain document data is shown below.

    <index: date="AAAA/BB/CC"; time="DD:EE:FF"; document address="1234">

    <user's operation history: maximum summary size="100">

    <selection: number of elements="10">PictureTel<custom character></selection>

    </user's operation history>

    <summary>Primary Minister X did not tell a specific amount of tax reduction, in a press conference.</summary>

    <word: word sense="0003"; central activation value="140.6">not tell</word>

    <word: word sense="0105"; identifier "X"; central activation value="67.2">Prime Minister</word>

    <person name: identifier "X"; word: word sense="6103"; central activation value="150.2">Prime Minister X</word></word /person name>

    <word: word sense="5301"; central activation value="120.6">ask</word>

    <word: word sense="2350"; identifier "X"; central activation value="31.4">Prime Minister</word>

    <word: word sense="9582"; central activation value="182.3">emphasize</word>

    <word: word sense="2595"; central activation value="93.6">tell</word>

    <word: word sense="9472"; central activation value="12.0">noticed</word>

    <word: word sense="4934"; central activation value="46.7">did not tell</word>

    <word: word sense="0178"; central activation value="175.7">excuse</word>

    <word: word sense="7248"; identifier="X"; central activation value="130.6">I</word>

    <word: word sense="3684"; identifier="X"; central activation value="121.9">Prime Minister</word>

    <word: word sense="1824"; central activation value="144.4">appeal</word>

    <word: word sense="7289"; central activation value="176.8">show</word>

    </index>

    In the above example, <index> and </index> indicate the start and end positions, respectively, of the index. <date> and <time> indicate the date and the time, respectively, at which the index was generated. <summary> and </summary> indicate the start and the end, respectively, of the summary.

    <word> and </word> indicate the start and end of a word.

    word sense="0003" indicates the third word sense of a word. The other tags are used in a similar manner. As described earlier, in order to distinguish a plurality of word senses of a word, numbers are assigned in advance to the respective word senses, and a particular word sense is specified by the number assigned to that word sense.

    <user's operation history> and </user's operation history> indicate the start and end of a user's operation history. <selection> and </selection> indicate the start and end of a selected element. maximum summary size="100" indicates that the maximum summary size is set to 100 characters. number of elements="10" indicates that the number of selected elements is 10.

    As can be seen from the above example, the index of a document includes one or more proper nouns and/or word senses that characterize the document.

    The indexing process in step F12 is described in further detail below with reference to FIGS. 6 to 9. Note that FIG. 6 illustrates the indexing process for one document data. When indexing is performed for a plurality of document data, it is required to perform the process shown in FIG. 6 for each document data.

    FIG. 8 illustrates the details of step F31 shown in FIG. 6, and the details of step F43 are shown in FIG. 9.

    In the indexing process shown in FIG. 5, spreading of activation values is first performed in step F31 in FIG. 6.

    The spreading of activation values is a process in which the central activation values associated with elements in document data are spread depending on the internal structure of a document such that high central activation values are given to elements having significant relations with elements having high central activation values.

    More specifically, initial central activation values are first given to the respective elements of a document, and the central activation values are then spread depending upon the internal structure, that is, the link structure, of the document.

    The central activation values are determined depending upon the internal structure represented by tags, and they can be used to extract distinctive words characterizing the document.

    The controller 11 performs the spreading of activation values in step F31 and stores the resultant central activation values associated with the respective elements into the RAM 14.

    The spreading of activation values in step F31 is described in further detail below with reference to FIGS. 7 to 9.

    FIG. 7 illustrates an example of a link structure associated with some elements.

    Note that FIG. 7 does not illustrate all elements of a document and the entire link structure associated therewith but illustrates a part of the link structure in the vicinity of elements E1 and E2. Of elements E1-E8 shown in FIG. 7, E1 and E2 are taken as examples in the following description.

    Herein, we assume that the element E1 has a central activation value equal to e1 and the element E2 has a central activation value equal to e2.

    These two elements E1 and E2 are connected to each other by a link L12 (normal link or reference link).

    The link L12 has an end point T12 connected with the element E1 and also has an end point T21 connected with the element E2.

    The element E1 is also connected with elements E3, E4, and E5, via links L13, L14, and L15, respectively. The links L13, L14, and L15 have end points T13, T14, and T15, respectively, connected with the element E1.

    Similarly, the element E2 is also connected with elements E6, E7, and E8, via links L26, L27, and L28, respectively. The links L26, L27, and L28 have end points T26, T27, and T28, respectively, connected with the element E2.

    The spreading of activation values over such a link structure is described below with reference to FIGS. 8 and 9.

    In step F41 in FIG. 8, before starting the spreading of activation values associated with the document data, an index of which is to be produced, initial central activation values are defined for all elements included in the document.

    The initial central activation values are determined such that, for example, a proper noun and other elements selected by a user have high values.

    The controller 11 sets to zero the end-point activation values of end points T(xx) of reference links and those of normal links via which elements are connected to one another. The controller 11 stores the resultant initial end-point activation values in the RAM 14.

    In step F42, the controller 11 initializes a counter for counting the number of elements Ei of the document. More specifically, the controller 11 sets the counter value i of the element counter to 1. When i=1, the counter points to a first element (for example, element E1 in FIG. 8).

    In step F43, the controller 11 recalculates the central activation value for an element pointed to by the counter.

    By way of example, the recalculation of the central activation value for the element E1 is described in detail with reference to FIG. 9.

    In the recalculation of the central activation value, end-point activation values of the element are first recalculated, and a new central activation value is determined using the current central activation value and the recalculated end-point activation values.

    In step F51 in FIG. 9, the controller 11 initializes the counter for counting the number of links connected at one end thereof with an element Ei (E1 in this specific example) of a document. More specifically, the controller 11 sets the counter value j of the link counter to 1. When j=1, the link counter points to a first link L(yy) connected with an element Ei. In the specific example shown in FIG. 7, a link L12 is pointed to as a first link associated with the element E1.

    In step F52, the controller 11 determines, by referring to a relational attribute tag, whether or not the link pointed to by the link counter, that is, the link L12 between elements E1 and E2, is a normal link. If the link L12 is a normal link, the controller 11 advances the process to step F53. However, the controller 11 advances the process to step F54 if the link L12 is a reference link.

    In the case where the link L12 is a normal link and thus the process goes to step F53, the controller 11 calculates a new end-point activation value for the end point T12 at which the element E1 is connected to the normal link L12.

    The end-point activation value t12 of the end point T12 is obtained by adding the central activation value e2 of the element E2 and the end-point activation values (t26, t27, t28) of all end points (T26, T27, T28) of the element E2 linked to the element E1 except for the end point connected to the link L12 and then dividing the resultant sum by the total number of elements included in the document.

    The controller 11 determines the new end-point activation value of the end point connected the normal link by performing the above-described calculation using end-point activation values and the central activation value read from the RAM 14. The determined end-point activation value is stored in the RAM 14. Thus, the end-point activation value t12 for the end point T12 is updated.

    On the other hand, in the case where it is determined in step F52 that the link L12 is a reference link and thus the process goes to step F54, the controller 11 calculates a new end-point activation value of the end point T12 at which the element E1 is connected to the link L12. In this case, the calculation is performed as follows.

    The end-point activation value t12 of the end point T12 is obtained by adding the central activation value e2 of the element E2 and the end-point activation values (t26, t27, t28) of all end points (T26, T27, T28) of the element E2 linked to the element E1 except for the end point connected to the link L12. (In this case, unlike the calculation for normal links, the resultant sum is not divided.)

    The controller 11 determines the new end-point activation value of the end point connected the reference link by performing the above-described calculation using end-point activation values and the central activation value read from the RAM 14. The determined end-point activation value is stored in the RAM 14. Thus, the end-point activation value t12 for the end point T12 is updated.

    After performing step F53 or F54, the controller 11 determines, in step F55, whether to go to step F57. That is, the process goes to step F57 if it is determined in step F55 that the calculation is not completed for all links. In step F55, the counter value j is incremented, and the process returns to step F52.

    Thus, the counter value becomes j=2, and the counter points to the second link (for example, L13) connected to the element E1. The end-point activation value t13 of the end point T13 at which the element E1 is connected to the link L13 is calculated, in a similar manner as described above, by performing step F52 and the following steps.

    In step F55, the controller 11 determines whether the new end-point activation value has been calculated for all links connected to an element Ei (E1 in this specific example) pointed to by the current counter value i, and the controller 11 performs the calculation until the new end-point activation value has been determined for all end points of the current element Ei.

    That is, the above-process is performed repeatedly while incrementing the counter value j in step F57 thereby determining new end-point activation values t12, t13, t14, and t15 of end points T12, T13, T14, and T15 of the element E1. When all end-point activation values have been determined, the process goes from step F55 to F56.

    In step F56, the new central activation value ei for the element Ei is determined using the new end-point activation values determined in the above process.

    The new central activation value ei is determined by adding the sum of new end-point activation values of the element Ei to the current central activation value ei of the element Ei. For example, in the case of the element E1 shown in FIG. 7, the new central activation value e1(new) is given by


    After determining the central activation value ei of the element Ei pointed to by the current counter value i, the controller 11 stores the resultant central activation value ei in the RAM 14. Thus, the central activation value ei of the element Ei is updated. (The old central activation value is further held for use in step F45 that will be described later.)

    After updating the central activation values in step F43 shown in FIG. 8 in the manner described above with reference to FIG. 9, the controller 11 advances the process to step F44 shown in FIG. 8. In step F44, the controller 11 determines whether the central activation values have been updated for all elements of the document. More specifically, the controller 11 determines whether the counter value i has become equal to the total number of elements included in the document.

    If the updating of the central activation value is not completed for all elements, the controller 11 advances the process to step F47. In step F47, the controller 11 increments the counter value i and returns the process to step F43.

    For example, at the time when the process for the element E1 is completed, the counter value i is incremented to i=2 so as to point to the element E2.

    Thus, step F43 (that is, the process shown in FIG. 9) is repeated to calculate the central activation value for the element E2.

    Although a further detailed description is not given herein because step F43 is performed in a similar manner, the end-point activation values t21, t26, t27, and t28 of the end points T21, T26, T27, and T28 of the element E2 are updated, and then the new central activation value e2(new) is determined in accordance with the following equation:


    In the process shown in FIG. 8, step F43 is performed repeatedly to calculate the central activation value while incrementing the counter value i in step F47 so as to change the element pointed to by the counter value, until the central activation value has been updated for all elements included in the document.

    When the updating of the central activation value is completed for all elements included in the document, the process goes from step F44 to F45.

    In step F45, the controller 11 calculates the mean value of variations in the central activation value of all elements contained in the document. That is, the mean value of differences between the new and old central activation values of all elements is calculated.

    More specifically, the controller 11 reads from the RAM 14 the old central activation values and the updated new central activation values for all elements. The controller 11 then calculates the differences between the new and old central activation values and divides the sum of differences by the total number of elements thereby determining the mean value of variations in central activation values of all elements.

    The controller 11 then stores into the RAM 14 the mean value of the variations in the central activation values of all elements.

    In the following step F46, the controller 11 determines whether the mean value calculated in step F45 is less than a predetermined threshold value.

    If the mean value is less than the threshold value, the controller 11 terminates the process of spreading activation values. However, when the mean value is not less than the threshold value, the process returns to step F42 to repeat the above-described process.

    As a result of spreading activation values, the central activation values of elements related to elements having high central activation values become high.

    However, if the spreading of activation values is performed only once, there is a possibility that the central activation value of an element, which should be increased to achieve the purpose of the indexing process, is not increased to a sufficiently high level. More specifically, although the central activation values of elements directly linked to an element having a high initial central activation value are increased to sufficiently high levels by one execution of the activation spreading process, the central activation values of elements that are not directly linked to an element having a high initial value are not increased to sufficiently high levels even when those elements are important to create the index.

    To avoid the above problem, the spreading of activation values is performed as many times as required to satisfy the condition in step F46. That is, the spreading of activation values is performed repeatedly until the central activation values for all elements have substantially converged, thereby ensuring that the central activation values of all important elements are increased.

    The central activation values of all elements can converge via the iterations of spreading activation values, because the central activation values of the respective elements are updated using central activation values calculated in the previous iteration. However, if the number of iterations is too great, the calculations are continued uselessly after the central activation values for all elements have converged.

    To avoid the above problem, the mean value of variations in the central activation values between two successive iterations is calculated in step F45, and it is determined in step F46 whether the mean value have fallen within a predetermined small range. Thus, the calculation is terminated when the central activation values have substantially converged.

    After completion of the spreading of activation values in FIGS. 8 and 9 (step F31 in FIG. 6), the controller 11 advances the process to step F32 shown in FIG. 6.

    In step F32, the controller 11 evaluates the central activation values determined in step F31 for the respective elements and extracts elements having central activation values greater than a predetermined threshold value. The controller 11 stores the extracted elements in the RAM 14.

    In the next step F33, the controller 11 reads the extracted elements from the RAM 14. The controller 11 then extracts all proper nouns included in the extracted elements and adds the extracted proper nouns to the index. Proper nouns have no word sense and they are not described in a dictionary. Thus, proper nouns are handled separately from the other words. Herein, as described earlier, a "word sense" refers to a particular meaning of a word having a plurality of meanings.

    It is possible to determine whether each element is a proper noun, by checking an associated tag described in a document. For example, in the internal structure represented by tags as shown in FIG. 3, relational attributes represented by tags indicate that "Acustom character", "Bcustom character", and "Ccustom character" (("Mr. A", "Convention B", "City C") are "person name", "organization name", and "place name", respectively, and thus they are proper nouns. The controller 11 adds the extracted proper nouns to the index and stores the result in RAM 14.

    In the next step F34, the controller 11 extracts, from the elements extracted in step F32, word senses other than the proper nouns and adds the extracted word senses to the index. The result is stored in the RAM 14.

    By performing the above process, an index such as that described above with reference to the specific example is obtained. That is, words characterizing a document including tags are detected, and an index is generated by listing the detected words. The significance of words included in a document is evaluated on the basis of the central activation values determined by means of spreading activation values depending upon the internal structure of the document.

    Because indexes generated in the above-described manner include word senses and proper nouns that characterize documents, indexes can be used to retrieve a desired document.

    In addition to the word senses and the proper nouns that characterize the document, the index also includes the document address representing the storage location of the RAM 14 (or the HDD 34) where the document is stored.

    4.3 Browsing, Generation of Categories, and Categorization

    The process of generating the index described above with reference to FIGS. 6 to 9 is performed in step F12 shown in FIG. 5. When the manual categorization process shown in FIG. 5 is performed, after the completion of generating the index, a user reads a document and manually categorizing the document, in steps F13 and F14.

    In step F13 in FIG. 5, as described earlier, the user can read a document displayed on the display 30.

    In step F14, the user generates categories and categorizes document data into categories generated.

    The operations in steps F13 and F14 and other related operations performed by the controller 11 are described below with reference to specific examples.

    FIGS. 10 and 11 illustrate specific examples of documents displayed on the display 30.

    FIG. 10 shows a document categorization window 201 used to categorize documents in accordance with a categorization model that will be described in detail later. In this specific example, the document categorization window 201 serves as a graphic user interface (GUI) for categorization of documents.

    The document categorization window 201 includes operation control buttons 202 such as a position reset button 202a used to reset the window into an initial state, a browser button 202b used to browse documents, and an exit button 202c used to exit from the window 201.

    The document categorization window 201 includes subwindows serving as document category displaying areas 203, 204, 205, etc., corresponding to categories based on the categorization model.

    The document category displaying area 203 is used to display miscellaneous topics. That is, documents that have not been categorized yet are indicated in the document category displaying area 203. For example, documents that are received in step F11 in FIG. 5 (and that are to be categorized) are indicated in the document category displaying area 203 entitled "miscellaneous topics".

    On the other hand, the document category displaying area 204 is used to indicate documents categorized in, for example, "business news".

    The document category displaying area 205 is used to indicate documents categorized in, for example, "political news".

    The other document category displaying areas having no reference numerals in FIG. 10 may also be used to indicate documents categorized in particular categories.

    When documents are categorized in particular categories, document icons and document titles of documents are displayed in corresponding document category displaying areas 203, 204, etc. When a document has no title, a sentence representing the summary of the document is displayed.

    The size of each document category displaying area 203, 204, etc., is not fixed. That is, the size of each document category display area can be changed to a desired size by moving the subwindow frames 211, 212, 213, etc., by means of dragging or the like. The number of document category displaying areas can be changed by a user to an arbitrary value.

    The title (such as "Political News") of each document category displaying area 203, 204, etc., may be arbitrarily set and changed by a user.

    The number of document category displaying areas and the titles thereof correspond to the number of categories and categories, respectively, defined in the categorization model that will be descried later. That is, the number of categories and the titles of the categories of the categorization model are set when a user sets the document category displaying areas or the title thereof in the categorization window 201 by using the mouse or the keyboard of the input unit 20.

    FIG. 11 illustrates an example of a browser window 301 used by a user to browse documents.

    For example, if a user clicks the browser button 202b in the categorization window 201 after selecting a document by clicking the corresponding icon or the like in the categorization window 201 shown in FIG. 10, then the controller 11 opens the browser window 301 as shown in FIG. 11 and displays the selected document therein.

    The browser window 301 includes a file name displaying area 302 for displaying the file name of a selected document data file, a document displaying area 303 for displaying document data corresponding to the file name displayed in the file name displaying area 302, a summary displaying area 304 for displaying a summary of the document displayed in the document displaying area 303, and a keyword displaying area 305 used to input and display a keyword. Furthermore, the browser window 301 includes operation control buttons 306 such as a Summarize button 306a used to start summarization, an undo button 306b used to cancel an operation, and a read-out button 306c used to execute a read-aloud operation.

    In the browser window 301, a user can read a document displayed in the document displaying area 303. When the entire document is not displayed at a time in the document displaying area 303, a part of the document is displayed. In this case, the use can read the entire document by scrolling the document.

    If the user clicks the Summarize button 306a, a summary of the document displayed in the document displaying area 303 is generated and displayed in the summary displaying area 304.

    The operation performed by the controller 11 to generate a summary text will be described later.

    On the other hand, if the user clicks the read-out button 306c, the document displayed in the document displaying area 303 or the summary thereof is read aloud.

    The process of reading-aloud a document will be described later.

    The categorization window 201 and the browser window 301 are displayed on the display 30 not only during the manual categorization process shown in FIG. 5 but also during other processes in response to a request issued by the user. For example, in the manual categorization process shown in FIG. 5, information about the types and the contents of received documents are displayed in the categorization window 201 or the browser window 301, and thus the user can acquire such information via the categorization window 201 or the browser window 301.

    More specifically, if one or more documents are received in step F11 shown in FIG. 5, an index is generated in step F12 for the received documents. After that, the titles of the received documents are displayed in the document category displaying area 203 entitled "Miscellaneous Topics" in the categorization window 201 shown in FIG. 10.

    Using the categorization window 201, the user manually categorizes the documents displayed in the document category displaying area 203. If the user cannot guess the content of a document from the title thereof, the user may display the document in the browser window 301 shown in FIG. 11 and read the content thereof. That is, in step F13 shown in FIG. 5, the user reads a document if reading is required for the above purpose.

    In step F14, using the categorization window 201, the user may add, update, and delete a category, as required. In response to an operation performed by the user, the controller 11 changes the manner in which the document category displaying areas 203, 204, etc., are displayed (that is, the number, the size, and the title of document category displaying areas are modified).

    If the user creates or modifies a category (the title of a document category displaying area), the creation or modification is reflected in the categorization model that will be described later.

    After creating a category as required, the user categorize the respective documents displayed in the document category displaying area 203 into proper categories corresponding to document category displaying areas. Thus, documents are manually categorized by the user.

    More specifically, the user drags, using the mouse of the input unit 20, the icons of documents displayed in the document category displaying area 203 entitled "Miscellaneous Topics" into document category displaying areas corresponding to desired categories.

    For example, the user may create a document category displaying area entitled "Sports" and may drag the icon of a document on a sport displayed in the document category displaying area entitled "Miscellaneous Topics" into the document category displaying area entitled "Sports".

    After being dragged, the icons and the titles of the respective documents are displayed in document category displaying areas into which the documents have been dragged.

    4.4 Creation/Registration of the Categorization Model

    In step F15 shown in FIG. 5, after completion of the manual categorization, the controller 11 creates a categorization model including a plurality of categories on the basis of the categorization that has been manually performed by the user. More specifically, the controller 11 creates a categorization model by gathering indexes of a plurality of documents categorized into categories. After that, the controller 11 categorizes the plurality of documents into corresponding categories defined in the categorization model.

    The categorization model consists of a plurality of categories in which documents are categorized, and the categorization model represents the correspondence between each category and documents.

    As described above, an index is generated for each document in step F12. The categorization model has a data structure in which the indexes of the respective documents are related to the corresponding categories in which the documents are categorized. An example of such a categorization model is shown in FIG. 12A.

    In the example shown in FIG. 12A, the categorization model includes categories "sport", "company", "computer", etc., which have been created by the user using the categorization window 201. Note that the categorization model may include a category that is not given by a user but that has been predefined. A document category displaying area corresponding to such a predefined category may also be displayed in the categorization window.

    In the categorization model, correspondence between each category and indexes IDX1, IDX2, . . . is described. That is, the indexes of the respective documents are related to the corresponding categories in which the documents are categorized.

    The indexes related to the respective categories are the same as those of documents displayed in the document category displaying areas corresponding to the respective categories in the categorization window 201.

    For example, index IDX1 is related to category "sport" because a user has created a document category displaying area entitled "sport" in the categorization window 201 and dragged the icons of a document having index IDX1 into the document category displaying area entitled "sport".

    As described earlier, each index includes one or more proper nouns and word senses other than the proper nouns, and also includes a document address.

    As shown in FIG. 12A, one or more indexes are related to each category. Because each index includes one or more proper nouns and word senses other than the proper nouns and also includes a document address, the categorization model may also be represented as shown in FIG. 12B.

    In the example shown in FIG. 12B, the categorization model has index fields for describing proper nouns, word senses other than proper nouns, and document addresses.

    In this categorization model, proper nouns "Mr. A", etc., are related to category "sport". Similarly, proper noun "Mr. B", etc., are related to "company", C Company", "G Company", etc., to "computer", "D species", etc., to "plant", "Mr. E", etc., to "art", and "Mr. F", etc., to "event".

    Similarly, word senses such as "base ball (4546)", "grand (2343)", "labor (3112)", "employment (9821)", "mobile (2102)", "cherry-1 (11111)", "orange-1 (9911)", "cherry-2 (11112)", "orange-2 (9912)", and "cherry-3 (11113)" are related to the corresponding categories.

    Furthermore, document addresses such as "SP1", "SP2", "SP3", . . . , "S01", "S02", "S03", . . . , "C01", "C02", "C03", . . . , "PL1", "PL2", "PL3", . . . , "AR1", "AR2", "AR3", . . . , and "EV1", "EV2", "EV3", . . . are also related to the corresponding categories.

    Herein, "cherry-1", "cherry-2", and "cherry-3" represent the first word sense (11111), the second word sense (11112), and the third second word sense (11113), respectively, of "cherry". Similarly, "orange-1", and "orange-2" represent the first word sense (9911) and the second word sense (9912), respectively, of "orange". More specifically, for example, "orange-1", represents an orange that is one of plants, and "orange-2" represents an orange color.

    For general nouns other than proper nouns, not words but word senses are used because a word can have a plurality of meanings.

    In step F15 shown in FIG. 5, a categorization model is generated in the above-described manner on the basis of manual categorization performed by a user. In the next step F16, the generated categorization model is registered, that is, stored in the RAM 14 (or the HDD 34).

    Thus, by generating and registering the categorization model, documents are categorized.

    After generating and registering the categorization model steps F15 and F16 shown in FIG. 5, the categorization model is updated via an automatic categorization process that will be described later, or via a modification of a category or a further manual categorization process performed by a user.

    If the categorization model is updated, the date and time of update is written in the categorization model. In the example shown in FIG. 12, the date and time of update is written as "1998:12:10:19:56:10".

    5. Automatic Categorization of Document Data

    5.1 Procedure

    In the document processing apparatus 1 according to the present embodiment, once a categorization model is generated, it becomes possible to perform an automatic categorization process to automatically categorize document data input from the outside via the communication device 21 or the like.

    That is, when the document processing apparatus 1 receives document data from the outside, the automatic categorization process is performed to categorize the received document data, as is described in detail below.

    In the following description, it is assumed that the automatic categorization process is performed each time one document is received. However, the automatic categorization process may be performed each time a predetermined number of documents have been received. Alternatively, the automatic categorization process may be performed when the window shown in FIG. 9 is opened. In this case, the automatic categorization process may be performed for all documents that have been received at that time.

    The outline of the automatic categorization process is shown in FIG. 13.

    In step F21 in FIG. 13, the receiver 216f the document processing apparatus 1 receives a document. In this step F21, the receiver 21 receives one or more documents via, for example, a communication line. The received one or more documents are transferred to the main unit 10 of the document processing apparatus 1. The controller 11 stores the one or more documents into RAM 14 or the HDD 34.

    In the next step F22, the controller 11 generates an index for each document data received in step F21.

    In step F23, the controller 11 automatically categorizes each document with an index into one of categories of the categorization model. The controller 11 stores the categorization result in the RAM 14. Each step in the automatic categorization process will be described in further detail later.

    In step F24, the controller 11 updates the categorization model on the basis of the result of automatic categorization performed upon the new document in step F23.

    In step F25, the controller 11 registers the resultant categorization model updated in step F24, by storing it in the RAM 14.

    Thus, by performing the process shown in FIG. 13 in the above-described manner, the document data input to the document processing apparatus 1 is automatically categorized in accordance with the categorization model.

    That is, in the automatic categorization process, an index is first generated for a received document, and then the document is automatically categorized. Furthermore, proper nouns, word senses, and the document address described in the index are related to a category on the categorization model as shown in FIG. 12 (thereby updating the categorization model).

    Steps F21 and F22 are performed in a similar manner to steps F11 and F12 in the manual categorization process described above. That is, the indexing process in step F22 is performed in a similar manner as described above with reference to FIGS. 6 to 9, and thus it is not described in further detail herein.

    In step F24, the categorization model is updated on the basis of the result of the automatic categorization performed in step F23.

    The automatic categorization in step F23 is performed in a different manner from the manual categorization process, as will be described below.

    5.2 Automatic Categorization

    FIG. 14 illustrates details of the automatic categorization process in step F23 shown in FIG. 13.

    In step F61 in FIG. 14, the controller 11 determines the number P(Ci) of proper nouns that are included in both the set of proper nouns belonging to the category Ci defined in the categorization model and the set of words extracted from the document received in step F21 and employed as elements of the index of the document. The controller 11 stores the calculated number P(Ci) into the RAM 14.

    In step F62, the controller 11 determines the word sense relevance values between all word senses included in the index of the document and all word senses included in each category Ci by referring to a word sense relevance table in FIG. 16 that will be described later. The controller 11 then calculates the sum R(Ci) of the word sense relevance values.

    That is, the controller calculates the sum R(Ci) of word sense relevance values for words on the categorization model other than proper nouns. The controller 11 stores the calculated sum of word sense relevance values into the RAM 14.

    The word sense relevance value is described below.

    The word sense relevance value is calculated in advance for each word sense contained in an electronic dictionary provided in the document processing apparatus 1, and the calculated word sense relevance values are stored as shown in FIG. 16. That is, if the controller 11 performs the process shown in FIG. 15 once, the obtained relevance values can be used in the automatic categorization process shown in FIG. 14.

    More specifically, the process shown in FIG. 15 is performed by the controller 11 as described below.

    First, in step F71, the controller 11 generates a word sense network in accordance with explanations of word senses described in the electronic dictionary.

    More specifically, the word sense network is generated in accordance with the explanations of the respective word senses described in the dictionary and the referential relations of word senses appearing in the explanations.

    The internal structure of the network is described by tags such as those described above. The controller 11 of the document processing apparatus 1 sequentially reads word senses and explanations thereof described in the electronic dictionary stored in the RAM 14 and generates a network.

    The controller 14 stores the generated word sense network in the RAM 14.

    Instead of generating a network by the controller 11 of the document processing apparatus 1 using the dictionary, a network may also be obtained by receiving from the outside via the receiver 21 or by installing from the storage medium 32 via the write/read unit 31.

    Similarly, the electronic dictionary may also be obtained by receiving from the outside via the receiver 21 or by installing from the storage medium 32 via the write/read unit 31.

    In step F72, spreading of central activation values of elements of the respective word senses is performed over the word sense network generated in step F71. In this activation spreading process, the central activation values associated with the respective word senses are given in accordance with the internal structure described by tags using the dictionary. The process of spreading activation values is performed in the manner described above with reference to FIG. 8.

    In step F73, one word sense Si is selected from elements constituting the word sense network generated in step F71. In the next step F74, the initial central activation value ei of the element Ei corresponding to the word sense Si is changed, and the change Δei in the central activation value from the initial value is calculated.

    In the next step F75, the change Δej in the central activation value ej of an element Ej corresponding to another word sense Sj in response to the change Δei in the central activation value of the element Ei is determined.

    In step F76, the difference Δej obtained in step F75 is divided by Δei obtained in step F74. The resultant ratio Δej/Δei is employed as the word sense relevance value of the word sense Si with respect to the word sense Sj.

    In step F77, it is determined whether the word sense relevance values have been calculated for all possible combinations between one word sense Si and all other word senses Sj.

    If word sense relevance values have not been calculated for all possible combinations, the process returns to step F73 to calculate the word sense relevance value for a remaining combination.

    In the loop from step F73 to F77, the controller 11 sequentially reads values required for the calculation from the RAM 14 and calculates the word sense relevance values in the above-described manner. The controller 11 sequentially stores the calculated word sense relevance values into the RAM 14.

    If it is determined in step F77 that the word sense relevance values have been calculated for all possible combinations of two word senses, the process is terminated.

    In the calculation of word sense relevance values, as can be seen from the above description, when the central activation value of a certain word sense is changed, if the central activation value of some other word sense changes to a great degree, then that word sense is regarded as having a high relevance.

    That is, if the central activation value of a certain word sense is changed in step F74, this change results in changes in the central activation values of word senses related (linked) to that word sense. Therefore, the relevance of word senses with respect to a certain word sense can be determined from the relative changes. (As described earlier, the central activation value of an element Ei is given by the sum of the current central activation value and the end-point activation values associated with that element Ei. Herein, the end-point activation values of the element Ei depend upon the central activation value and end-point activation values of elements linked to the element Ei. Therefore, if an element Ej has a high degree of relevance to the element Ei, a change in the central activation value of the element Ei generates a large change in the central activation value of the element Ej.)

    By performing the above-described process for all possible combinations of two word senses, the relevance values are obtained for all possible combinations of two word senses.

    A word sense relevance value is defined between each word sense and another word sense, as shown in FIG. 16. In the example of the word sense relevance table shown in FIG. 16, word sense relevance values are normalized such that they take a value within the range from 0 to 1. In the example shown in FIG. 16, the word sense relevance values among "computer", "television", and "VTR" are described in the table. Herein, the relevance value between "computer" and "television" is 0.55, and that between "computer" and "VTR" is 0.25. The relevance value between "television" and "VTR" is 0.60.

    Referring again to FIG. 14, after performing step F62 using the word sense relevance values which have been calculated in advance in the above-described manner, the controller 11 performs step F63 to calculate the document category relevance value Rel(Ci) of a document with respect to category Ci according to the following equation:

    where coefficients ml and nl are constants representing the degrees of contributions of the respective values to the document category relevance.

    In the above process, the controller 11 calculates, according to the above equation, the document category relevance value Rel(Ci) using the number P(Ci) of common elements calculated in step F61 and the sum R(Ci) of word sense relevance values calculated in step F62.

    The controller 11 stores the calculated document category relevance value Rel(Ci) into the RAM 14.

    The coefficients ml and nl may be set to, for example, 10 and 1, respectively.

    The values of coefficients ml and nl may also be determined statistically. In this case, the controller 11 calculates the document category relevance value Rel(Ci) using various values of ml and nl, and employs optimum values.

    In step F64, the controller 11 categorizes the document into category Ci if the document category relevance value of the document becomes highest for category Ci and if the document category relevance value Rel(Ci) is greater than a threshold value.

    That is, the controller 11 calculates document category relevance values with respect to a plurality of categories, and selects a category corresponding to the highest document category relevance value. If the document category relevance value corresponding to the selected category is greater than the threshold value, the controller 11 categorizes the document into the selected category. Thus, the document is automatically categorized into a correct category.

    If the highest document category relevance value is not greater than the threshold value, the document is not categorized into any category.

    After performing the automatic categorization in step F23 in FIG. 14, which is described in further detail in FIG. 14, the categorization model is updated and registered in steps F24 and F25, respectively, in accordance with the result of the automatic categorization. Thus, the entire process associated with the automatic categorization is completed.

    In this way, the document data input to the document processing apparatus 1 is automatically categorized, and displayed in a corresponding document category displaying area in the document categorization window 201 shown in FIG. 10, thereby informing the user of the reception of the document.

    6. Summarization

    Now, the process of generating a summary of document data is described.

    As described earlier, a user can select a document and read the selected document displayed in the browser window 301 shown in FIG. 11. The browser window 301 can be opened from the categorization window 201 shown in FIG. 10 when the above-described manual categorization process is performed in step F13 or at any other time.

    For example, if the user clicks the browser button 202b in the categorization window 201 after selecting a document, the browser window 301 is opened and the selected document is displayed in the document displaying area 303 as shown in FIG. 17.

    When the entire document is not displayed at a time in the document displaying area 303, a part of the document is displayed.

    When a summary has not been generated yet, nothing is displayed in the summary displaying area 304 as shown in FIG. 17.

    If the summarize button 306a in the browser window 301 is clicked, a summary of the document displayed in the document displaying area 303 is generated and displayed in the summary displaying area 304 as shown in FIG. 18.

    More specifically, in response to the Summarize button 306a being clicked by the user, the controller 11 performs a summarization process for generating a summary text and the displays the generated summary text as described below.

    The process of generating a summary from a given document is performed on the basis of the internal structure, represented by tags, of the document.

    The summary is generated depending on the size of the summary displaying area 304. The sizes of the document displaying area 303 and the summary displaying area 304 can be changed by moving the boundary 312.

    That is, the summary is generated such that the resultant summary has a size (document length) corresponding to the size of the summary displaying area 304 at the time when a summarization command is issued.

    FIG. 19 illustrates the process performed by the controller 11 to generate a summary text in response to the Summarize button 306a being clicked.

    In step F81 in FIG. 19, the controller 11 spreads activation values. In the present embodiment, a summary is generated by employing elements having high degrees of significance represented by the central activation values obtained by means of spreading activation. When a given document includes tags representing the internal structure, central activation values determined by means of spreading activation in accordance with the. internal structure described by tags can be assigned to the respective elements.

    The process of spreading activation in step F81 is performed in a similar manner to the. process described earlier with reference to FIGS. 7-9. As described earlier, the spreading activation is a process in which the central activation values associated with elements are spread such that if an element has significant relation with an element having a high central activation value, then a high central activation value is given to the former element. The activation spreading process causes both an anaphoric (coreferential) expression and an antecedent thereof to have the same central activation value. On the other hand, the central activation values of the other elements decrease. The central activation values are determined in accordance with the internal structure represented by tags, and they are used to extract keywords characterizing the document.

    In the next step F82, the controller 11 sets a parameter ws such that ws represents the current size of the summary displaying area 304 in the browser window 301 displayed on the display 30. That is, the parameter ws represents the maximum allowable number of characters that can be displayed in the summary displaying area 304 such that The controller 11 initializes the summary string s (internal register for storing a summary). That is, the initial value of the summary string s is set such that s(0)="". The controller 11 stores the maximum allowable number ws of characters and the initial value s(0) of the string s into the RAM 14.

    In step F83, the controller 11 sets the counter value i of a counter for counting the number of iterations.

    Then in step F84, the controller 11 extracts a skeleton of a sentence having an ith greatest mean central activation value from the document.

    Herein, the mean central activation value refers to the mean value of central activation values of elements included in a sentence.

    The controller 11 reads a string s(i-1) from the RAM 14 and adds the string of the extracted sentence skeleton to the string s(i-1) thereby generating a string s(i). The controller 11 stores the resultant string s(i) into the RAM 14.

    In the first iteration, because the string s(i-1) has an initial value s(0), the sentence skeleton extracted in this first operation is employed as the string s(i) and stored into the RAM 14.

    When step F84 is performed in the following iterations, a newly extracted sentence skeleton is added to the current string s(i) (that is, string s(i-1) at that time).

    Furthermore, in step F84, the controller 11 generates a list L(i) of elements that are not included in the sentence skeleton, wherein elements are listed in the order of descending central activation values. The controller 11 stores the resultant list L(i) into the RAM 14.

    The summarization algorithm employed in step F84 is to select sentences in the order of central activation values from the highest value to the lowest value on the basis of the result of spreading of activation values and extract sentence skeletons of selected sentences. The skeleton of a sentence is made up of essential elements extracted from the sentence. Elements that can be essential include a head, a subject, an object, an indirect object, and an element having a relational attribute as to possessor, cause, condition, or comparison. When a coordination structure is essential, elements included directly in the coordination structure are employed as essential elements. The controller 11 generates a sentence skeleton by joining essential elements of a selected sentence and adds it to the summary.

    In step F85, the controller 11 determines whether the length of the string s(i) is greater than the maximum allowable number ws of characters that can be displayed in the summary displaying area 304 of the browser window 301.

    This step F85 is necessary to generate the summary such that the summary has a length corresponding to the size of the summary displaying area 304.

    If the length of the string s(i) is less than the maximum allowable number ws of characters, the controller 11 advances the process to step F86.

    In step F86, the controller 11 compares the central activation values of elements of a sentence having an (i+1)th highest mean central activation value of sentences included in the document with the highest central activation value among those of elements included in the list L(i) generated in step F84.

    That is, a sentence (a candidate having highest priority among the remaining sentences) whose mean central activation value is next in magnitude to that of a sentence that has been employed in step F84 as a part of the summary is compared with the central activation values of elements that have been regarded as being not essential and omitted from the skeletons of sentences employed in step F84 to generate the summary.

    Thus, in step F86, it is determined whether an element omitted from the sentence skeleton employed in the previous step F84 should be now added to the summary or an element of another sentence should be added.

    If the highest central activation value among those of elements in the list L(i) is higher than those of elements of the sentence having the (i+1)th highest mean central activation value, an element is selected from the elements that were not employed in the sentence skeleton in the previous step F84 and the selected element is added to the summary string.

    In this case, the controller 11 advance the process to step F88 and selects an element having the highest central activation value from the list L(i) and adds the selected element to the current string s(i) thereby generating a string ss(i).

    The controller ll then removes the selected element from the list L(i).

    In step F89, the controller 11 determines whether the length of the string ss(i) is greater than the maximum allowable value ws. If not, the process returns to step F86.

    On the other hand, if it is determined in step F86 that the sentence having the (i+1)th highest mean central activation value includes an element having a higher central activation value than the highest central activation value among those of elements in the list L(i), it is determined that an element to be further added to the summary string should be selected from a sentence other than the sentence employed in the previous step F84. In this case, the process goes to step F87, and the counter value i is incremented. Then, the process returns to step F84.

    That is, a skeleton is extracted from the sentence that have been determined, in step F84, to have the (i+1)th highest mean central activation value, and the extracted skeleton is added to the string s(i).

    Thus, elements having high central activation values are selected in step F84 or F88, and the selected elements are added to the summary string. on the other hand, in step F85 or F89, the length of the string s(i) or ss(i) is compared with the maximum allowable number ws of characters, thereby ensuring that the number of characters included in the string becomes closest to but not greater than the maximum allowable number ws.

    If it is determined in step F85 that the length of the string s(i) is greater than the maximum allowable value ws, then the controller 11 advances the process to step F90 and employs the previous string s(i-1) instead of the new string s(i) that includes a skeleton selected and added in the previous step F84.

    That is, when a sentence skeleton is added to the summary string in step F84, if the resultant summary string includes a greater number of characters than the maximum allowable number ws, it is determined that the previous string s(i-1), which does not include the sentence skeleton employed in the immediately previous step F84 to form the current string s(i), includes as many characters as possible below the limit ws. Thus, the previous string s(i-1) is employed as a final summary string.

    When the string s(i) is generated for the first time in step F84 (i=1, in this case), if it is determined in step F85 that the number of characters included in the string s(i) is greater than the maximum allowable number ws, the string s(i-1) becomes identical to the initial string s(0) (null string) given in step F82, and thus no summary string is generated.

    This can occur when the size of the summary displaying area 304 is too small. In this case, the user may expand the size of the summary displaying area 304 on the screen and click the Summarize button 306a to start the process shown in FIG. 19.

    If it is determined in step F85 that the number of characters included in the string s(i) is not greater than the maximum allowable number ws, the controller 11 advances the process to step F86 as described above and selects an element to be further added to the summary string.

    In step F89, it is determined whether the number of characters included in the string ss(i) is greater than the maximum allowable number ws.

    If yes, the controller 11 advances the process to step F91 and employs, as the summary string, the previous string s(i) that does not include an element added in the immediately previous step F88 to form the current string.

    That is, when an element is added to the string in step F88, if the resultant summary string includes a greater number of characters than the maximum allowable number ws, it is determined that the previous string 6(i), which does not include the above-described element, includes as many characters as possible below the limit ws. Thus, the previous string s(i) is employed as a final summary string.

    If wsy=ws, it is determined that a summary has been generated in the summarization process such that the length of the summary matches the size of the summary displaying area 304. The content of the summary is made up of a skeleton of one or more sentences having high mean central activation values and one or more elements that are not included in skeletons but have high central activation values.

    The resultant summary is stored in the RAM 14 and the entire summary is displayed in the summary displaying area 304 in a fixed fashion as shown in FIG. 18.

    When the user reads the summary displayed in the summary displaying area 304, if the user wants a longer or shorter summary, the user may click the Summarize button 306a after increasing or decreasing the size of the summary displaying area 304 in the browser window 301.

    In response, the process shown in FIG. 19 is performed, and a summary having a length matching the specified size of the summary displaying area 304 is generated and displayed.

    7. Reading-Aloud Process

    As described above, when the document processing apparatus 1 receives a document including a tag, the document or a summary thereof is displayed so that a user can read it. Furthermore, the document processing apparatus 1 is capable of outputting a voice that reads aloud the received document.

    In this case, a read-aloud program stored in the ROM 15 or the HDD 34, i