Document detection system with improved document detection efficiency

Abstract

A document detection system capable of detecting a desired document from a large number of documents easily and accurately in which the user can make a judgement concerning the appropriateness of the detection result quickly. In the system, those documents which contain a semantic structure of a detection command containing natural language expressions entered by a user are detected. Also, the keywords of each document can be extracted from the summary of each document and those documents whose keywords match with detection keywords specified by a user can be detected. Also, the summary of each detected document can be automatically generated according to text structures of each detected document and displayed along with the detected document itself. Also, the detection processing can be carried out with respect to the summaries of the documents instead of the documents themselves.

Claims

What is claimed is:

1. A document detection system, comprising:

input means for entering a detection command containing detection keywords specified by a user;

document storage means for storing a plurality of detection target documents;

summary generation means for generating a summary of each detection target document stored in the document storage means, and extracting keywords of each detection target document from the summary of each detection target document; and

detection processing means for identifying those detection target documents stored in the document storage means whose keywords extracted by the summary generation means match with the detection keywords in the detection command entered at the input means as detected documents and for retrieving the detected documents.

2. The system of claim 1, further comprising:

input analysis means for determining a semantic structure of the detection command containing natural language expressions entered by a user; and

wherein the summary generation means also extracts a semantic structure of each sentence in the summary of each detection target document, and the detection processing means identifies those detection target documents whose keywords match with the detection keywords and whose summary contain the sentence structure extracted by the summary generation means which matches with the sentence structure of the detection command determined by the input analysis means.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to a document detection system for detecting desired documents from a large number of documents stored in a document database. It is to be noted that the term "retrieval" is often used in the literature of the field instead of the term "detection" used in the following description. The present specification adheres to the use of the term "detection" throughout.

2. Description of the Background Art

In recent years, due to the significant progress and spread of computers, the electronic manipulations of documents are becoming increasingly popular as in the electronic news and electronic mail systems and the CDROM publications of data sources such as dictionaries and encyclopedia that had only been available on papers, and it is expected that this trend of the electronic manipulations of documents will continue at an increasing pace in future.

In conjunction with such electronic manipulations of documents, much attentions have been attracted to a document detection system for detecting desired documents from a large number of documents efficiently, so as to enable the effective utilization of the documents stored in a database system in advance.

As a conventionally available document detection system, there has been a system which uses keywords in combination with logic operators such as AND, OR, NOT or proximity operators for specifying numbers of characters, sentences, and paragraphs that can exist between keywords, and detects a document by using a specified combination of keywords and operators as a detection key.

However, in such a conventional document detection system, it is not necessarily possible to detect the document that is truly desired by a user. Namely, in a case of employing the detection key using the logic operators, when the specified detection key is "computer AND designing", a document having a content of "designing by a computer" as well as a document having a content of "designing of a computer" can be detected, so that at least one of them will have a content irrelevant to a desired content. On the other hand, when the detection key using the proximity operators is employed, the detection is based solely on a physical distance between keywords, so that there is no guarantee that the detected document has a desired content.

Thus, in such a conventional document detection system, the detection result could contain many documents with contents actually irrelevant to the desired content, so that it has been necessary to use the detection key formed by as many keywords that are expected to be related to the desired content as possible. However, in practice, the detection result obtained by using such a detection key formed by a large number of disjunctive keywords would end up containing a considerable number of detection noises and Junks.

For this reason, the conventional document detection system requires an enormous amount of time for an user to single out the desired document by checking each of the detected documents one by one. On the other hand, if the detection key is formed by narrow keywords in order to reduce the number of the detection noises, the probability for the detection error would be increased.

As a result, in the conventional document detection system, it has been difficult to reduce the number of the detection noises without causing the detection error unless the user has a detailed knowledge concerning what kinds of keywords are contained in what kinds of documents, and consequently it has been formidably difficult for an ordinary user without such a detailed knowledge to handle the conventional document detection system effectively.

In addition, in the conventional document detection system, the detection result has been informed by displaying either a number of detected documents or titles of the detected documents alone, so that in order for the user to check each of the detected documents to see if it is the desired document or not, it has been necessary for the user to read the entire content of each of the detected documents one by one, and this operation has been enormously time consuming.

Moreover, in the conventional document detection system, in displaying the titles of the detected documents, the titles are simply arranged in a prescribed order according to the user's query such as an order of descending similarities to the keywords used in the detection key. For this reason, it has been impossible for the user to comprehend the relative relationships among the detected documents and the level of similarity with respect to the detection command for each of the detected documents from the displayed detection result, and consequently it has been difficult for the user to have an immediate impression for the appropriateness of the displayed detection result.

Furthermore, in the conventional document detection system, the detection scheme is limited to that in which each document as a whole is treated as a single entity, so that the document containing the desired content in the background section and the document containing the desired content in the conclusion section will be detected together in mixture. In other words, the detection result contains a variety of documents mixed regardless of viewpoints in which the desired contents appear in the documents. For example, if there is no interest in what had been done in the past, the detected document which matches with the given keywords in the background section will be of no use. Yet, in the conventional document detection system, the documents having different perspectives such as the document containing the desired content in the background section and the document containing the desired content in the conclusion section will not be distinguished, and the mixed presence of these documents in different perspectives makes it extremely difficult for the user to judge the appropriateness of the detection result.

In view of these problems, there has been a proposition for a scheme to reduce the burden on the user to read the entire content of each detected document by displaying only a portion of each detected document. However, in such a scheme, it is often impossible to make a proper judgement as to whether it is the desired document or not unless the relationship of the displayed portion and the remaining portion becomes apparent. For example, when the background section containing the desired content is displayed for one document while the conclusion section containing the desired content for the other document, as these documents cannot be comprehended in a unified viewpoint, it is difficult for the user to make a proper judgement as to which one of these document is the necessary one. As a result, in order to fully comprehend the perspectives of the displayed portions in these documents, the user would be forced to read the entire contents of these documents after all, so that it cannot contribute to the reduction of the burden on the user at all.

Also, there has been a proposition for a scheme to reduce the burden on the user to read the entire content of each detected document by providing a man-made document summary for each stored document in advance in correspondence to each stored document itself and displaying the document summary at a time of displaying the detection result. However, in such a scheme, an enormous amount of human effort is required for preparing the document summary for each document at a time of producing the database itself, which is not practically justifiable unless the database system has a remarkably high utilization rate. Moreover, there are many already existing database systems in which the document summary for each document is not provided, and an enormous amount of human efforts is similarly required for preparing the document summary for each document in such an already existing database system. In addition, the man-made document summary is produced in the very general viewpoint alone, so that there is no guarantee that each document is summarized from a viewpoint suitable for the required detection. As a result, the document summary displayed as the detection result can be quite out of point from the viewpoint of the user with the specific document detection objective, and in such a case, it is possible for the user to overlook the actually necessary document at a time of judging whether each detected document is the desired document or not.

SUMMARY OF THE INVENTION

It is therefore an object of the present invention to provide a document detection system capable of detecting a desired document from a large number of documents easily and accurately.

It is another object of the present invention to provide a document detection system capable of automatically preparing and displaying a document summary for each document in a viewpoint which is efficiently comprehensible for the user, considering the limited visual data processing ability of the human user, such that the user can make a judgement concerning the appropriateness of the detection result quickly.

According to one aspect of the present invention there is provided a document detection system, comprising: input analysis means for determining a semantic structure of a detection command containing natural language expressions entered by a user; document storage means for storing a plurality of detection target documents; and detection processing means for detecting those detection target documents stored in the document storage means which contain the semantic structure of the detection command determined by the input analysis means as detected documents.

According to another aspect of the present invention there is provided a document detection system, comprising: input means for entering a detection command containing detection keywords specified by a user; document storage means for storing a plurality of detection target documents; summary generation means for generating a summary of each detection target document stored in the document storage means, and extracting keywords of each detection target document from the summary of each detection target document; and detection processing means for detecting those detection target documents stored in the document storage means whose keywords extracted by the summary generation means match with the detection keywords in the detection command entered at the input means as detected documents.

According to another aspect of the present invention there is provided a document detection system, comprising: input means for entering a detection command specified by a user; document storage means for storing a plurality of detection target documents; detection processing means for detecting those detection target documents stored in the document storage means which match with the detection command entered at the input means as detected documents; summary generation means for automatically generating a summary of each detected document obtained by the detection processing means according to text structures of each detected document; and detection result output means for displaying each summary generated by the summary generation means.

According to another aspect of the present invention there is provided a document detection system, comprising: input means for entering a detection command specified by a user; document storage means for storing a plurality of detection target documents; summary generation means for generating summaries of the detected documents stored in the document storage means; summary storage means for storing the summaries generated by the summary generation means; and detection processing means for detecting those summaries stored in the summary storage means which match with the detection command entered at the input means.

Other features and advantages of the present invention will become apparent from the following description taken in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic block diagram of a first embodiment of the document detection system according to the present invention.

FIG. 2 is a block diagram for a functional configuration of a main part of the document detection system of FIG. 1.

FIG. 3 is a flow chart for an overall operation of the main part shown in FIG. 2.

FIG. 4 is a flow chart for an operation of a detection control unit in the main part shown in FIG. 2.

FIG. 5 is a detailed block diagram for a functional configuration of an input analysis unit in the main part shown in FIG. 2.

FIG. 6 is a flow chart for an operation of the input analysis unit of FIG. 5.

FIG. 7 is a table of exemplary results obtained at various stages in the flow chart of FIG. 6.

FIG. 8 is a diagrammatic illustration of an exemplary semantic analysis rule used in the flow chart of FIG. 6.

FIG. 9 is a diagrammatic illustration of an exemplary unnecessary expression rule used in the flow chart of FIG. 6.

FIG. 10 is a detailed block diagram for a functional configuration of a detection processing unit in the main part shown in FIG. 2.

FIG. 11 is a flow chart for an operation of the detection processing unit of FIG. 10.

FIG. 12 is a diagrammatic illustration of an exemplary keyword index memory used in the flow chart of FIG. 11.

FIG. 13 is a list of selected Japanese-English correspondences useful in understanding the keyword index memory shown in FIG. 12.

FIGS. 14A and 14B are a flow chart for an operation of a keyword index matching unit and a document file set calculation unit in the detection processing unit of FIG. 10.

FIG. 15 is a diagrammatic illustration of an exemplary semantic structure index memory used in the flow chart of FIG. 11.

FIG. 16 is a flow chart for an operation of a semantic structure index matching unit in the detection processing unit of FIG. 10.

FIG. 17 is a detailed block diagram for a functional configuration of a summary generation unit in the main part shown in FIG. 2.

FIG. 18 is a diagrammatic illustration of exemplary text structure and sentence handled in the summary generation unit of FIG. 17.

FIG. 19 is a diagrammatic illustration of a data structure for the text structure used in the summary generation unit of FIG. 17.

FIG. 20 is a schematic flow chart for a key sentence Judgement unit in the summary generation unit of FIG. 17.

FIG. 21 is a detailed flow chart for a penalty calculation to be carried out by a key sentence Judgement unit in the summary generation unit of FIG. 17.

FIG. 22 is a flow chart for a text re-construction unit in the summary generation unit of FIG. 17.

FIG. 23 is a diagrammatic illustration of an exemplary summary sentence memory in an individual data storage unit in the main part of FIG. 2.

FIG. 24 is a table representing an exemplary display priority rule dictionary in the detection control unit shown in FIG. 2.

FIG. 25 is a diagrammatic illustration of an exemplary data content of the individual data storage unit shown in FIG. 2.

FIG. 26 is an illustration of an exemplary detection result display used in the system of FIG. 1.

FIG. 27 is a diagrammatic illustration of an exemplary data content for the detection result in the individual data storage unit shown in FIG. 2.

FIG. 28 is an illustration of an exemplary detection result screen display in the system of FIG. 1, for the exemplary data content of FIG. 27.

FIG. 29 is a table representing an exemplary analysis result learning dictionary in the detection control unit shown in FIG. 2.

FIG. 30 is a flow chart for an operation of the system according to a first variation of the first embodiment.

FIG. 31 is a table of exemplary analysis results obtained by the input analysis unit shown in FIG. 2 in the first variation of the first embodiment.

FIGS. 32A and 32B are diagrammatic illustrations of an exemplary semantic structure index memory before and after the operation of FIG. 30, respectively.

FIG. 33 is a flow chart for an operation of the detection control unit shown in FIG. 2 in the first variation of the first embodiment.

FIG. 34 is a table of exemplary analysis results obtained by the input analysis unit shown in FIG. 2 in the second variation of the first embodiment.

FIG. 35 is a table of another exemplary analysis results obtained by the input analysis unit shown in FIG. 2 in the second variation of the first embodiment.

FIG. 36 is a table of exemplary analysis results obtained by the input analysis unit shown in FIG. 2 in the third variation of the first embodiment.

FIG. 37 is a diagrammatic illustration of an exemplary bibliographical matter analysis rules used in the fourth variation of the first embodiment.

FIG. 38 is a diagrammatic illustration of an exemplary unnecessary expression rule used in the fourth variation of the first embodiment.

FIG. 39 is a diagrammatic illustration of a display priority score conditions used in the fifth variation of the first embodiment.

FIG. 40 is a diagrammatic illustration of one exemplary semantic structure index memory in the fifth variation of the first embodiment.

FIG. 41 is a diagrammatic illustration of another exemplary semantic structure index memory in the fifth variation of the first embodiment.

FIG. 42 is a diagrammatic illustration of an exemplary semantic structure index memory in the sixth variation of the first embodiment.

FIG. 43 is a diagrammatic illustration of an exemplary keyword extraction rule dictionary used in the seventh variation of the first embodiment.

FIG. 44 is a flow chart for an operation of the input analysis unit and the detection processing unit shown in FIG. 2 in the seventh variation of the first embodiment.

FIG. 45 is a table of exemplary semantic structures handled in the eighth variation of the first embodiment.

FIG. 46 is an illustration of an exemplary detection result screen display used in the tenth variation of the first embodiment.

FIG. 47 is a diagrammatic illustration of an exemplary summary information memory in the individual data storage shown in FIG. 2 in the fourteenth variation of the first embodiment.

FIG. 48 is an illustration of an exemplary detection result display for a list of document identified in the fourteenth variation of the first embodiment.

FIG. 49 is an illustration of an exemplary detection result display for a summary in the fourteenth variation of the first embodiment.

FIG. 50 is an illustration of an exemplary detection result display for an original document in the fourteenth variation of the first embodiment.

FIG. 51 is an illustration of an exemplary screen display for display change command input used in the fifteenth variation of the first embodiment.

FIG. 52 is an illustration of an exemplary detection result display for a summary in the fifteenth variation of the first embodiment.

FIG. 53 is an illustration of an exemplary detection result display for an original document in the fifteenth variation of the first embodiment.

FIGS. 54A and 54B are a flow chart for an operation in the fifteenth variation of the first embodiment.

FIG. 55 is a flow chart for a pointer shifting operation in response to a previous page button in the fifteenth variation of the first embodiment.

FIG. 56 is an illustration of an exemplary detection result display for a summary in the fifteenth variation of the first embodiment in response to a next page button.

FIG. 57 is an illustration of an exemplary detection result display for an original document in the fifteenth variation of the first embodiment in response to a next page button.

FIG. 58 is an illustration of an exemplary detection result display for a list of document identified in the fifteenth variation of the first embodiment.

FIG. 59 is an illustration of an exemplary detection result display for a summary in the fifteenth variation of the first embodiment corresponding to the list of FIG. 58.

FIG. 60 is an illustration of an exemplary detection result display for an original document in the fifteenth variation of the first embodiment corresponding to the list of FIG. 58.

FIG. 61 is an illustration of an exemplary screen display for summary display item change command input used in the sixteenth variation of the first embodiment.

FIG. 62 is an illustration of an exemplary detection result display for a summary in the sixteenth variation of the first embodiment.

FIG. 63 is a flow chart for an operation in the sixteenth variation of the first embodiment.

FIG. 64 is a diagrammatic illustration of an exemplary summary sentence table used in the sixteenth variation of the first embodiment.

FIG. 65 is an illustration of an exemplary detection result display for an original document in the sixteenth variation of the first embodiment corresponding to the summary of FIG. 62.

FIG. 66 is a block diagram for a functional configuration of a main part of the document detection system according to the second embodiment of the present invention.

FIG. 67 is a flow chart for an operation of a detection control unit in the main part shown in FIG. 66.

FIG. 68 is a detailed block diagram for a functional configuration of an input analysis unit in the main part shown in FIG. 66.

FIG. 69 is a flow chart for an operation of the input analysis unit of FIG. 68.

FIG. 70 is a table of exemplary results obtained at various stages in the flow chart of FIG. 69.

FIG. 71 is a detailed block diagram for a functional configuration of a detection processing unit in the main part shown in FIG. 66.

FIG. 72 is a flow chart for an operation of a detection result processing control unit in the main part shown in FIG. 66.

FIG. 73 is an illustration of an exemplary document to be handled in the second embodiment.

FIG. 74 is a flow chart for an operation of an internal document relation data analysis unit in the main part shown in FIG. 66.

FIG. 75 is a flow chart for an operation of an external document relation data analysis unit in the main part shown in FIG. 66.

FIG. 76 is a diagrammatic illustration of an exemplary summary information memory in the individual data storage shown in FIG. 66.

FIG. 77 is a diagrammatic illustration of an exemplary internal document relation data in the individual data storage shown in FIG. 66.

FIG. 78 is a diagrammatic illustration of an exemplary external document relation data in the individual data storage shown in FIG. 66.

FIG. 79 is a detailed block diagram for a functional configuration of a detection result output unit In the main part shown in FIG. 66.

FIGS. 80A, 80B, 80C, 80D, 80E, 80F, and 80G are flow charts for various elements in the detection result output unit of FIG. 71.

FIGS. 81A, 81B, 81C, 81D, 81E, and 81F are illustration of exemplary detection result screen displays at various stages in the operation of the detection result output unit of FIG. 71.

FIG. 82 is a detailed block diagram for a functional configuration of the detection processing unit shown in FIG. 66 for the first variation of the second embodiment.

FIG. 83 is a diagrammatic illustration of a relation order table used in the first variation of the second embodiment.

FIG. 84 is a diagrammatic illustration of a data structure for the text structure analysis result used in the first variation of the second embodiment.

FIG. 85 is a flow chart for an operation of a rearrangement unit in the detection result output unit of FIG. 71.

FIG. 86 is a flow chart for a matching relation extraction processing in the operation of FIG. 85.

FIG. 87 is a detailed block diagram for a functional configuration of a summary generation unit shown in FIG. 2 in the second variation of the second embodiment.

FIG. 88 is a detailed block diagram for a functional configuration of a summary generation unit shown in FIG. 2 in the third variation of the second embodiment.

FIGS. 89A and 89B are illustrations of exemplary window displays used in the third variation of the second embodiment.

FIG. 90 is a detailed block diagram for a functional configuration of the detection result output unit shown in FIG. 66 in the fourth variation of the second embodiment.

FIGS. 91A and 91B and flow charts for an operation of a summary modification processing unit in the detection result output unit of FIG. 90.

FIG. 92 is an illustration of an exemplary summary display before the operation of FIG. 91.

FIG. 93 is an illustration of the exemplary summary display during the operation of FIG. 91.

FIG. 94 is an illustration of the exemplary summary display after the operation of FIG. 91.

FIG. 95 is a flow chart for an operation of a document selection processing unit in the detection result output unit shown in FIG. 90 in the fifth variation of the second embodiment.

FIG. 96 is a flow chart for an operation of a summary modification processing unit in the detection result output unit of FIG. 90 in the fifth variation of the second embodiment.

FIG. 97 is an illustration of an exemplary summary display before the operation of FIG. 96.

FIG. 98 is an illustration of the exemplary summary display during the operation of FIG. 96.

FIG. 99 is an illustration of the exemplary summary display after the operation of FIG. 96.

FIG. 100 is a detailed block diagram for a functional configuration of the detection result output unit shown in FIG. 66 in the sixth variation of the second embodiment.

FIG. 101 is a flow chart for an operation of a document selection processing unit in the detection result output unit of FIG. 100.

FIGS. 102A and 102B are flow charts for an operation of a summary detailing processing unit in the detection result output unit of FIG. 100.

FIG. 103 is a flow chart for an operation of a summary simplification processing unit in the detection result output unit of FIG. 100.

FIG. 104 is a flow chart for an operation of an original document display processing unit in the detection result output unit of FIG. 100.

FIG. 105 is an illustration of an exemplary summary display before the operation of FIG. 102.

FIG. 106 is an illustration of the exemplary summary display after the operation of FIG. 102.

FIG. 107 is a flow chart for an operation of a document selection processing unit in the detection result output unit shown in FIG. 100 in the seventh variation of the second embodiment.

FIG. 108A and 108B are flow charts for an operation of a summary detailing processing unit in the detection result output unit shown in FIG. 100 in the seventh variation of the second embodiment.

FIG. 109 is a flow chart for an operation of a summary simplification processing unit in the detection result output unit shown in FIG. 100 in the seventh variation of the second embodiment.

FIGS. 110A and 110B are illustrations of the exemplary summary and original document displays obtained in the seventh variation of the second embodiment.

FIG. 111 is a flow chart for an operation of a document structure analysis unit in the summary generation unit in the eighth variation of the second embodiment.

FIG. 112 is an illustration of an exemplary original document handled in the eighth variation of the second embodiment.

FIG. 113 is an illustration of the exemplary paragraph data obtained from the original document of FIG. 112 by the operation of FIG. 112.

FIG. 114 is a flow chart for an operation of a text re-construction unit in the summary generation unit in the eighth variation of the second embodiment.

FIG. 115 is an illustration of the exemplary summary obtained from the original document of FIG. 112 by the operation of FIG. 114.

FIG. 116 is a block diagram for a functional configuration of a main part of the document detection system in the ninth variation of the second embodiment.

FIG. 117 is a block diagram for a functional configuration of a main part of the document detection system in the tenth variation of the second embodiment.

FIG. 118 is an illustration of a possible manner of detection result display in the document detection system according to the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

Referring now to FIG. 1, the first embodiment of a document detection system according to the present invention will be described in detail.

In this first embodiment, the document detection system has an overall configuration as shown in FIG. 1 in which a central processing means 1 is connected with a memory means 2, a display means 4 through a display controller 3, and an input means 6 through an input controller 5.

The central processing means 1 is formed by a processor for carrying out various processing operations such as the input analysis, the detection processing, and the summary generation. The memory means 2 is formed by a memory medium such as a semiconductor memory, a magnetic disk memory, an optical disk memory, etc. for storing documents as detection targets. The display means 4 is formed by a display device such as a liquid crystal display and a plasma display for displaying the text content of the document obtained as the detection result while also outputting other graphic, audio, and speech data contained in the document, under the control of the display controller 3. The input means 6 is formed from input devices such as a keyboard and a mouse for entering the detection commands and other inputs from the user, under the control of the input controller 6.

Here, it is to be noted that the document detection system may contain more than one of the central processing means 1 or more than one of the memory means 2, all of which are connected together by paths or a network, if desired.

In further detail, a main portion of the document detection system of this first embodiment has a functional configuration as shown in FIG. 2, which comprises: an input unit 11 for entering an input sentence in a natural language from the user; an input analysis unit 12 for carrying out various analyses of the input sentence including the morphological analysis, the syntactic analysis, and the semantic analysis; a detection processing unit 13 for carrying out the detection processing for detecting documents according to a syntactic analysis result obtained from the input sentence and a detection key constructed by using keywords extracted from the input sentence; a summary generation unit 14 for producing a summary of each detected document; a document storage unit 15, connected with the detection processing unit 13, for storing the document database; a detection result output unit 17 for outputting the results obtained by the input analysis unit 12, the detection processing unit 13, and the summary generation unit 14; an individual data storage unit 16, connected with the input analysis unit 12, the detection processing unit 13, the summary generation unit 14, and the detection result output unit 17, for storing individual data including the detected documents; and a detection control unit 18 for controlling the operations of the other processing modules including the input unit 11, the input analysis unit 12, the detection processing unit 13, the summary generation unit 14, and the detection result output unit 17, while managing user interactions.

In this FIG. 2, the document storage unit 15 and the individual data storage unit 16 belong to the memory means 2 in the overall configuration of FIG. 1, while the input unit 11, the input analysis unit 12, the detection processing unit 13, the summary generation unit 14, the detection result output unit 17, and the detection control unit 18 belong to the central processing means 1 in the overall configuration of FIG. 1. Also, in this FIG. 2, the thick arrows indicate data lines while thin arrows indicate control lines.

In this functional configuration of FIG. 2, the detection control unit 18 controls each processing module to realize the processing sequence as indicated in the flow chart of FIG. 3, as follows.

First, at the step 301, the input sentence in a form of input character string entered from the input unit 11 is transferred to the input analysis unit 12, at which the morphological analysis, the syntactic analysis, and the semantic analysis of the input character string are carried out. The results obtained by the input analysis unit 12 including the keywords in the input sentence are then stored in the individual data storage unit 16.

Next, at the step 302, the input analysis unit 12 produces a detection key by using the keywords in the input sentences stored in the individual data storage unit 16 in conjunction with logic operators as described in detail below.

Then, at the step 303, the detection processing unit 13 is activated to carry out the detection processing using the produced detection key on the document database stored in the document storage unit 15. A set of documents detected at this step 303 are temporarily stored in the individual data storage unit 16.

Next, at the step 304, the detection processing unit 13 judges whether more than one documents have been detected at the step 303 and the syntactic and semantic analyses results for the input character string are stored in the individual data storage unit 16, or not.

In a case of the affirmative, next at the step 305, the syntactic and semantic analyses results for the detected documents prepared in advance and stored in the document storage unit 15 are matched with the syntactic and semantic analyses results for the input character string stored in the individual data storage unit 16. A set of documents detected and matched at this step 305 are temporarily stored in the individual data storage unit 16. Otherwise, the step 305 is skipped.

Next, at the step 306, the text contents of the detected and matched documents stored in the individual data storage unit 16 are taken out from the document storage unit 15, and the summary generation unit 14 is activated to produce a summary for each detected and matched document from the text content. The summary for each detected and matched document obtained by the summary generation unit 14 is then stored in the individual data storage unit 16.

Then, at the step 307, the summary generation unit 14 checks the matching of the keywords or the syntactic and semantic analyses results for the input sentence in the summary for each detected and matched document. The check result obtained at this step 307 is then stored in the individual data storage unit 16.

Finally, at the step 308, the detection result output unit 17 is activated to display the detection result containing the document names or the summaries of those documents remaining in the individual data storage unit 16 after the step 307, in an order according to the check result obtained at the step 307. Here, the detection result output unit 17 can display and change the data stored in the individual data storage unit 16 in accordance with the commands entered from the user through the input unit 11, such that the user can select the desired document from the displayed detection result.

Now, the detailed operation of each processing module in this first embodiment will be described in detail.

First, the detection control unit 18 operates according to the flow chart of FIG. 4, as follows.

Namely, at first, the detection control unit 18 awaits for the input from the input unit 11 at the step 401. Then, when the input of the input character string at the input unit 11 is detected, the input analysis unit 12 is activated at the step 402. Next, the detection control unit 18 awaits for the end of the processing at the input analysis unit 12 at the step 403. Then, when the end of the processing at the input analysis unit 12 is detected, the detection processing unit 13 is activated at the step 404. Next, the detection control unit 18 awaits for the end of the processing at the detection processing unit 13 at the step 405. Then, when the end of the processing at the detection processing unit 13 is detected, the summary generation unit 14 is activated at the step 406. Next, the detection control unit 18 awaits for the end of the processing at the summary generation unit 14 at the step 407. Then, when the end of the processing at the summary generation unit 14 is detected, the detection result output unit 17 is activated at the step 408. Next, the detection control unit 18 awaits for the end of the processing at the detection result output unit 17 at the step 409. Then, when the end of the processing at the detection result output unit 17 is detected, the operation returns to the step 401 to repeat the process of the steps 401 to 409 for the next input sentence.

Next, the input analysis unit 12 has a detailed functional configuration as shown in FIG. 5, which comprises: a morphological analysis unit 120, a syntactic analysis unit 121, a semantic analysis unit 122, an unnecessary expression extraction rule application unit 123, a content word extraction unit 124, a detection key production unit 125, an analysis dictionary 126 and an analysis grammar 127 utilized in the morphological, syntactic, and semantic analyses, an unnecessary expression extraction rule dictionary utilized by the unnecessary expression extraction rule application unit 123, and a related word dictionary 129 utilized by the detection key production unit 125.

With this functional configuration of FIG. 5, the input analysis unit 12 operates according to the flow chart of FIG. 6, as follows.

First, at the step 601, the morphological analysis is carried out on the input sentence at the morphological analysis unit 120 by using the analysis dictionary 126, to divide the input sentences into words. Then, at the steps 602 and 603, the syntactic analysis and the semantic analysis are carried out at the syntactic analysis unit 121 and the semantic analysis unit 122, respectively, by using the analysis dictionary 126 and the analysis grammar 127. Here, the details of the morphological, syntactic, and semantic analyses to be carried out at these steps 601 to 603 are not essential to the present invention, and any known schemes can be adopted. For example, the schemes disclosed in "Japanese Word Dictionary", Japan Electronic Dictionary Research Institute, Ltd., April 1990, chapter 3, pp. 27-33, can be utilized here.

Next, at the step 604, the unnecessary expression extraction rule specified by the unnecessary expression extraction rule dictionary 128 is applied at the unnecessary expression extraction rule application unit 123, to delete a partial structure which coincides with each unnecessary expression specified by the unnecessary expression extraction rule, and then at the step 605, the structure obtained up to the step 604 is stored into the individual data storage unit 16.

Next, at the step 606, the content words in the structure obtained up to the step 604 are extracted at the content word extraction unit 124, and then at the step 607, the detection key is produced at the detection key production unit 125 by using the content words extracted at the step 606 along with appropriate logic operators.

Next, at the step 608, additional detection keys are also produced at the detection key production unit 125 by looking up related words of each extracted content word and replacing each extracted content word in the detection key produced at the step 607 by each looked up related word. Then, all the content words in the detection key as well as their related words are set as detection target keywords in the detection keys.

Finally, at the step 609, all the detection keys produced at the steps 607 and 608 are stored into the individual data storage unit 16.

As a concrete example of the result of the operation by this input analysis unit 12, FIG. 7 shows various results obtained at various stages in the flow chart of FIG. 6 described above, for a particular input sentence in Japanese (with English translation provided in parentheses). More specifically, for the input sentence indicated in (a) of FIG. 7, the morphological analysis result appears as indicated in (b) of FIG. 7 in which the input sentence is divided into words. Then, the syntactic analysis result appears as indicated in (c) of FIG. 7 in which the tree structure representing the syntactic structure of the input sentence is generated. Then, the semantic analysis result appears as indicated in (d) of FIG. 7 in which the tree structure representing the semantic structure of the input sentence is generated.

In the semantic analysis, the semantic analysis rule specified in the analysis dictionary 126 such as that indicated in FIG. 8 for example is applied. Namely, this semantic analysis rule indicated in FIG. 8 specifies the semantic structure corresponding to the syntactic structure in which the Japanese verb "mochiiru" (meaning "use") appears between a noun and a so called sahen-noun (i.e., a noun that can be turned into a verb by affixing a Japanese verb "suru" (meaning "do") at the end) as a so called instrument case relationship between the noun and the sahen-noun. The similar semantic analysis rules are also specified in the analysis dictionary 126 for the other frequently used verbs such as "employ", "invoke", etc. It is noted here that in FIGS. 7 and 8, "wo" indicates a Japanese postpositional word, "rentai" indicates a symbol for a participial adjective, "nitsuite" indicates a Japanese expression meaning "about", and "obj" indicates a symbol for an objective case relationship.

Then, the unnecessary expression extraction result appears as indicated in (e) of FIG. 7 in which a partial structure which coincides with the unnecessary expression specified by the unnecessary expression extraction rule in the unnecessary expression extraction rule dictionary 128 is deleted from the tree structure obtained by the semantic analysis unit 122. Here, as shown in FIG. 9 for example, the unnecessary expression extraction rule indicates the partial structure to be deleted as "-obj.fwdarw.KNOW". It is to be noted, however, that the unnecessary expression extraction rule may indicate a word to be deleted instead of the partial structure to be deleted.

Then, the content word extraction result appears as indicated in (f) of FIG. 7 in which the content words "example" and "machine translation" are extracted from the tree structure obtained by the unnecessary expression extraction rule application unit 123.

Finally, the detection key production result appears as indicated in (g) of FIG. 7 in which the detection key is formed by connecting the content words extracted by the content word extraction unit 124 with an appropriate logic operator "+".

Next, the detection processing unit 13 has a detailed functional configuration as shown in FIG. 10, which comprises: a keyword index matching unit 131 and a semantic structure matching unit 132 which are connected with the document storage unit 15 and the individual data storage unit 16, and a document file set calculation unit 133 connected with the keyword index matching unit 131 and the semantic structure matching unit 132.

With this functional configuration of FIG. 10, the detection processing unit 13 operates according to the flow chart of FIG. 11, as follows.

First, at the step 1101, the detection key stored in the individual data storage unit 16 is taken out to the keyword index matching unit 131, and then at the step 1102, the detection using the taken out detection key is carried out on a keyword index memory in the document storage unit which stores all words of all the documents in the document database, so as to obtain those documents which contain the same words as those used in the detection key.

Next, at the step 1103, a set calculation for the detected documents obtained at the step 1102 is carried out by the document file set calculation unit 133 according to the detection key, and the set calculation result is stored in the individual data storage unit 16.

Then, at the step 1104, the semantic structure matching unit 132 Judges whether more than one documents are stored in the individual data storage unit 16 as the detected documents obtained at the step 1103 and the semantic structure of the input character string is stored in the individual data storage unit 16, or not. In a case of the affirmative, the semantic structure of the input character string is taken out from the individual data storage unit 16 at the step 1105, and the detection using the taken out semantic structure is carried out on a semantic structure index memory in the document storage unit 15 which stores all the semantic structures of all the documents in the document database, so as to obtain those documents which have the same semantic structure as that of the input character string, and then the result obtained at the step 1106 is stored in the individual data storage unit 16 at the step 1107. Otherwise, the steps 1105 to 1107 are skipped.

In order to illustrate the keyword detection part of the operation by this detection processing unit 13 more concretely, FIG. 12 shows an exemplary content of the keyword index memory.

Namely, in this keyword index memory of FIG. 12, each kanji character involved in the keywords is assigned to a unique address, and each keyword formed by a plurality of kanji characters is specified by the link data registered after each kanji character, so as to reduce the required memory capacity and simplify the necessary detection procedure.

For example, for the Japanese keyword "kikai" (meaning "machine") formed by two kanji characters as indicated on the first line in FIG. 13, the first character registered at the address 00053 in the head character storage region has a link data "00935" specifying the second character of the keyword as that registered at the address 00935 in the subsequent character storage region. In addition, this second character at the address 00935 also has the link data "01201" specifying the third character, and the third character at the address 01201 has the link data "01309" specifying the fourth character, for another keyword "kikai-honyaku" (meaning "machine translation") formed by four kanji characters as indicated on the last line in FIG. 13 which contains the above described keyword "kikai" as a first part. Furthermore, the second character at the address 00935 also has a file data "file 4 (34, 35, 72, 86)" indicating that the keyword "kikai" is contained in the document data having the file name "file 4" at sentence No. 34, 35, 72, and 86. Similarly, the fourth character at the address 01309 has two file data "file 25 (18, 42)" and "file 21 (23)", indicating that the keyword "kikai-honyaku" is contained in the document data "file 25" at sentence No. 18 and 42 and the document data "file 21" at sentence No. 23. On the other hand, the first character at the address 00091 in the head character storage region is common to two keywords "jikken" (meaning "experiment") and "Jitsurei" (meaning "example") as indicated on the second and third lines in FIG. 13, so that it has two link data "01003" and "01004" specifying the respective second characters for these two keywords. In FIG. 12, an isolated "0" functions as a separator for separating the character, link data, and the file data. Also, the first characters of the keywords are registered in the continuous head character storage region in a sorted order such as that of JIS (Japanese Industrial Standard) codes.

Thus, the keyword index matching unit 131 looks up this keyword index memory to make a matching of each detection target keyword in the detection key, and obtains the documents containing the same detection target keyword as the detection key according to the document data registered for the matched keyword in the keyword index memory.

More specifically, the keyword index matching unit 131 and the document file set calculation unit 133 operate according to the flow chart of FIGS. 14A and 14B as follows.

First, at the step 1301, a variable "i" indicating an index of each keyword to be detected is set to 1 and a variable "N" indicating a total number of keywords to be detected is set to a number of keywords involved in the detection key as an initialization.

Then, at the step 1302, whether "i" is greater than "N" or not is judged. Unless "i" is greater than "N" at the step 1302, next at the step 1303, the first character of the keyword "i" is detected in the head character storage region of the keyword index memory, and a block registering that first character is designated as a block "A". Here, as the head character storage region stores the first characters in a sorted order, so that the block registering the first character of the keyword "i" can be obtained easily by carrying out the binary search.

Next, at the step 1304, a variable "k" indicating a character position in the keyword "i" is set to 2, and then at the step 1305, whether "k" is greater than a character string length of the keyword "i" or not is judged. Unless "k" is greater than the length of the keyword "i", next at the step 1306, the link data registered in the block "A" is looked up to obtain a block registering the "k"-th character of the keyword "i", and an obtained block is newly designated as a block "A". Then, at the step 1307, whether the block "A" obtained at the step 1306 actually exists in the subsequent character storage region of the keyword index memory. If the block "A" actually exists, the variable "k" is incremented by one at the step 1308 and the operation returns to the step 1305 described above for the incremented value of the variable "k".

In a case "k" is greater than the length of the keyword "i" at the step 1305, next at the step 1309, the document file name registered as the document data in the block "A" is set to the document file set "i" for the keyword "i", and then the variable "i" is incremented by one at the step 1310 and the operation returns to the step 1302 described above for the incremented value of the variable "i".

In a case the block "A" does not exist at the step 1307, next at the step 1311, the document file set "i" for is the keyword "i" is set as an empty set, and then the variable "i" is incremented by one at the step 1310 and the operation returns to the step 1303 described above for the incremented value of the variable "i".

By these steps, the document file set for each keyword in the detection key is obtained. Then, in a case "i" becomes greater than "N" at the step 1302, next at the step 1312, the variable "i" is reset to 2, and the document file set 1 for the keyword 1 is set as the current document file set. Then, until the variable "i" exceeds "N" at the step 1313, the step 1314 in which the set calculation to update the current document file set to be a logical product ("AND") of the document file set "i" and the current document file set and the subsequent step 1315 in which the variable "i" is incremented by one are repeated, so as to eventually obtain the final current document file set containing those documents which contain all the keywords in the detection key.

On the other hand, in order to illustrate the semantic structure detection part of the operation by this detection processing unit 13 more concretely, FIG. 15 shows an exemplary content of the semantic structure index memory.

Namely, in this semantic structure index memory of FIG. 15, each semantic structure specified by a set of target word, relation symbol, and source word is enlisted along with the document file names and the sentence No. containing each semantic structure. This semantic structure index memory is prepared in advance by carrying out the analysis of all the sentences of each document in the document database stored in the document storage unit 15. Thus, in the example shown in FIG. 15, the semantic structure of "instrument" case relationship between the source word "machine translation" and the target word "example" is contained in the document "file 25" at the sentence No. 18, the document "file 21" at the sentence No. 23, etc.

More specifically, the semantic structure index matching unit 132 operates according to the flow chart of FIG. 16 as follows.

First, at the step 1501, the semantic structure index memory is looked up for the target word in the semantic structure of the input character string. Then, only when the coinciding target word is found out in the semantic structure index memory at the step 1502, the relation symbol registered for the coinciding target word is looked up and matched with the relation symbol in the semantic structure of the input character string at the step 1503. Then, only when the coinciding relation symbol is found out in the semantic structure index memory at the step 1504, the source word registered for the coinciding relation symbol is looked up and matched with the source word in the semantic structure of the input character string at the step 1505. Then, only when the coinciding source word is found out in the semantic structure index memory at the step 1506, the document file names and sentence No. registered for the coinciding source word is stored into the individual data storage unit 16 along with the matched semantic structure itself.

Next, the summary generation unit 14 has a detailed functional configuration as shown in FIG. 17, which comprises: a document structure analysis unit 141, a text structure analysis unit 142, a key sentence judgement unit 143, and a text re-construction unit 144.

The document structure analysis unit 141 analyzes the chapter structure of each document by extracting title data indicating the chapters and sections in each document. Here, the details of the document structure analysis to be carried out by this document structure analysis unit 141 is not essential to the present invention, and any known schemes can be adopted. For example, the scheme disclosed in Doi, M., et al.: "Research on Model Based Document Processing System DARWIN", Human-Computer Interaction-INTERACT '87, H. J. Bullinger, B, Shackel (Ed.), Elsevier Science Publishers B.V. (North-Holand), 1987, pp. 1101-1106, can be utilized here.

The text structure analysis unit 142 analyzes the logical structure of the sentences of each chapter or section by extracting rhetorical expressions used in the sentences of each chapter or section. Here, again, the details of the text structure analysis to be carried out by this text structure analysis unit 142 is not essential to the present invention, and any known schemes can be adopted. For example, the scheme disclosed in Sumita, K., et al.: "A Discourse Structure Analyzer for Japanese Text", Proceedings of the International Conference on Fifth Generation Computer Systems 1992, Institute for New Generation Computer Technilogy (Ed.), pp. 1133-1140, can be utilized here.

The key sentence judgement unit 143 judges the key sentences according to the document structure and the text structure obtained by the document structure analysis unit 141 and the text structure analysis unit 142.

The text re-construction unit 144 generates a summary of each document from the key sentences obtained by the key sentence judgement unit 143.

In this manner, for an exemplary sentence shown in (b) of FIG. 18, a text structure shown in (a) of FIG. 18 can be obtained for example. In this example of FIG. 18, the text structure indicates that the second sentence has a relationship of "extension" with respect to the first sentence, while the third sentence has a relationship of "direction" with respect to the first and second sentences.

In this summary generation unit 14, the text structure is expressed by a data structure as shown in FIG. 19, in which a block for each node includes a "text structure Rel" indicating a relationship of each node, "text structure L" and "text structure R" indicating links to the nodes located on the left and right of each node, respectively, and a "text structure P" indicating a penalty for each node. Here, for a terminal node, the "text structure Rel" indicates a pointer to a corresponding sentence, while each of the "text structure L" and "text structure R" indicates the value "-1".

The key sentence judgement unit 143 operates according to the flow chart of FIG. 20, in which a penalty calculation at the step 2001 is carried out recursively for all the nodes. After the penalty is calculated for all the nodes, the terminal node with less penalty can be judged as more important.

In further detail, the penalty calculation for each node is carried out according to the flow chart of FIG. 21 as follows.

Here, the key sentence judgement unit 143 initially has an input in a form of the text structure obtained by the text structure analysis unit 142 and an initial value of the penalty. Then, first, at the step 2101, whether the text structure is a terminal node or not is judged. Namely, when the value indicated by the "text structure L" is "-1", it can be judged as the terminal node. In a case of the terminal node, the entered penalty value P is set to the "text structure P" indicating the penalty at the step 2102.

On the other hand, in a case of the non-terminal node, the relationship indicated by the "text structure Rel" is judged as one of a right stressed type, a left stressed type, and neither one of these, at the steps 2103 and 2106. This judgement is made according to the relationship table provided in the key sentence judgement unit 143 in advance. For example, the relationship table has the entries similar to the following (for a case of using Japanese):

    ______________________________________
    SERIAL RELATION      right stressed type
    DIRECTION            right stressed type
    EXEMPLIFICATION      left stressed type
    ______________________________________