JP7168411B2 - Information processing system and information processing method - Google Patents

Description

The present invention relates to an information processing system and an information processing method for processing information.

Patent Literature 1 discloses a dictionary creation device that learns semantic categories to extend a semantic category dictionary, and can correct the learning results even if an error occurs in the semantic category learning results. This dictionary creation device updates the correspondence between words and semantic categories held in the semantic category dictionary and also updates the semantic category extraction rules in consideration of the result of semantic category assignment by the semantic category assigning unit. In addition to the category learning section, a semantic category editing section is provided that presents the correspondence between words and semantic categories updated by the semantic category learning section and accepts correction of the correspondence between words and semantic categories.

JP 2007-213336 A

However, the conventional technology described above has a problem that it is difficult to increase the number of words in the word dictionary DB. The dictionary creation device of Patent Literature 1 retrieves example sentences including a relationship to be extracted only by keyword retrieval. Therefore, when adding a new example sentence, it is necessary to edit both the extraction rule and the semantic category dictionary.

An object of the present invention is to efficiently collect unregistered words.

An information processing system and an information processing method disclosed in the present application are an information processing system and an information processing method having a processor that executes a program and a storage device that stores the program. a word dictionary database that stores word groups that are word groups that are grouped together; a rule database that stores a tree structure pattern abstracted using the word groups for tree structure data indicating relationships between words in a sentence regarding elements of sentences; wherein the processor acquires a target sentence including a combination of a first element word to which the word group does not apply and a second element word to which the word group does not apply among sentence elements; determining whether or not the target sentence acquired by the acquisition process corresponds to a specific tree structure pattern obtained by excluding a third element corresponding to the word group among the elements of the sentence; an extraction process of extracting words of the third element from the target sentence determined to correspond to the specific tree structure pattern and outputting an extraction result.

According to the representative embodiment of the present invention, unregistered words can be efficiently collected. Problems, configurations, and effects other than those described above will be clarified by the following description of the embodiments.

FIG. 1 is an explanatory diagram showing an example 1 of database maintenance. FIG. 2 is an explanatory diagram showing an example 2 of database maintenance. FIG. 3 is an explanatory diagram of a database maintenance example 3. As shown in FIG. FIG. 4 is a block diagram showing a hardware configuration example of a computer. FIG. 5 is an explanatory diagram showing an example of contents stored in a word dictionary DB. FIG. 6 is an explanatory diagram of an example of contents stored in a rule DB. FIG. 7 is an explanatory diagram of an example of contents stored in a data store. FIG. 8 is an explanatory diagram showing an example of the text. FIG. 9 is an explanatory diagram showing an example of tree structure data and a tree structure pattern. FIG. 10 is an explanatory diagram showing an example of pattern representation. FIG. 11 is an explanatory diagram showing a conversion example using the pattern representation shown in FIG. FIG. 12 is a flowchart showing an example of an information processing procedure by the information processing system. FIG. 13 is an explanatory diagram showing a usage example of the information processing system. FIG. 14 is an explanatory diagram showing a display screen example 1 of the information processing system. FIG. 15 is an explanatory diagram showing a display screen example 2 of the information processing system. FIG. 16 is an explanatory diagram showing a display screen example 3 of the information processing system. FIG. 17 is an explanatory diagram showing a display screen example 4 of the information processing system. FIG. 18 is an explanatory diagram showing example 5 of the display screen of the information processing system. FIG. 19 is an explanatory diagram showing a display screen example 6 of the information processing system. FIG. 20 is a flowchart illustrating an example of a processing procedure in an example of use of the information processing system. FIG. 21 is an explanatory diagram showing an example of word registration. FIG. 22 is a flow chart showing an example of a word registration processing procedure. FIG. 23 is an explanatory diagram showing a display screen example 7 of the information processing system. FIG. 24 is an explanatory diagram showing a display screen example 8 of the information processing system.

In this specification, an example of maintenance for a word dictionary DB (Database) and an example of additional registration of a word to the word dictionary DB will be described separately. Examples of maintenance for the word dictionary DB will be described with reference to FIGS. 1 to 20, and examples of additional registration of words to the word dictionary DB will be described with reference to FIGS. 21 to 24. FIG.

[1. Example of maintenance for word dictionary DB]
FIG. 1 is an explanatory diagram showing an example 1 of database maintenance. FIG. 1 illustrates maintenance of the word dictionary DB 101. FIG. The word dictionary DB 101 stores one or more word groups. A word group is a word group grouped by a predetermined attribute. A predetermined attribute is a characteristic that the word group exhibits. The predetermined attribute specifically includes, for example, verbs in which the subject particle is a verb in the case of "ga" in Japanese sentences, and verbs that co-occur with specific adverbs. In addition, it may be a synonym, a synonym, or a word used in a specific field (investment, medical care, etc.). In FIG. 1, as an example, the word group Ga is a synonym group including "suppress" and "decrease".

The rule DB 102 is a database that stores tree structure patterns that indicate rules. A tree-structured pattern is data obtained by abstracting tree-structured data, which indicates relationships between words in a sentence with respect to elements of the sentence, using word groups. Sentence elements are, for example, subjects, predicates, and objects. The tree structure data is, for example, a syntax tree generated according to phrase structure rules by morphological analysis and phrase structure analysis (hereinafter referred to as syntactic analysis). The rule Ra in FIG. 1 is a tree structure pattern in which a subject (wild card), a predicate, and an object (wild card) are ordered, and the verbs forming the predicate are the word group Ga.

The data store 103 stores text data of various sentences (for example, sentences in academic papers and books, sentences in newspapers and magazines, sentences described on web pages, etc.).

(A) When the data store 103 is searched with the tree structure pattern of the rule Ra (S11), a search result 111 is obtained. All sentences in the search result 111 are text data that satisfy the rule Ra. Here, in (A), it is assumed that "reduce" is added to the word group Ga due to maintenance of the word dictionary DB 101, resulting in the state of (B). (B) When the data store 103 is searched with the tree structure pattern of the rule Ra (S12), a search result 112 is obtained. Even when "reduce" is added to the word group Ga, retrieval is possible without modifying the rule Ra.

In this case, the sentences in the search results 112 are all text data that satisfy the rule Ra, and the search results 111 further include "reduce" such as "Z reduces D." and "X is going to reduce E." is added. In this way, only by maintaining the word dictionary DB 101, it is possible to perform a search that satisfies the maintenance result of the word dictionary DB 101 without maintaining the rule DB 102. FIG.

Also, in (B), it is assumed that due to maintenance of the word dictionary DB 101, "reduce" is deleted from the word group Ga, resulting in the state of (A). (A) When the data store 103 is searched with the tree structure pattern of the rule Ra, a search result 111 is obtained. Even if "reduce" is deleted from the word group Ga, retrieval is possible without modifying the rule Ra.

In this case, the sentences in the search result 111 are all text data that satisfy the rule Ra, and in the search result 112, "Z reduces D." and "X is going to reduce E." Not searched. It should be noted that the above-described deletion and addition of words may be executed to change the words. For example, in (B), to change "reduce" to "drop", "reduce" should be deleted from the word group Ga and "drop" should be added. In this way, only by maintaining the word dictionary DB 101, it is possible to perform a search that satisfies the maintenance result of the word dictionary DB 101 without maintaining the rule DB 102. FIG.

FIG. 2 is an explanatory diagram showing an example 2 of database maintenance. FIG. 2 explains maintenance for the rule DB 102 . (A) is the same as (A) of FIG. (B) shows a newly added rule Rb. The rule Rb is a tree structure pattern in which the subject (wildcard), predicate (auxiliary verb (wildcard) and verb), and object (wildcard) word order, and the verb is the word group Ga. In other words, rule Rb is a tree structure pattern in which auxiliary verbs are added to rule Ra.

(B) When the data store 103 is searched with the tree structure pattern of the rule Rb (S13), a search result 210 is obtained. All sentences in the search result 210 are text data that satisfy the rule Rb. Also, when deleting a rule, it is sufficient to delete the rule Rb from the rule DB 102, and there is no need to maintain the word dictionary DB 101. FIG. To change the rules, the deletion and addition of the rules described above may be performed. For example, when rule Ra is changed to rule Rb, rule Ra is called and an auxiliary verb (wild card) is added before the verb (word group Ga). In this way, only by maintaining the rule DB 102, it is possible to perform a search satisfying the maintenance result of the rule DB 102 without maintaining the word dictionary DB 101. FIG.

FIG. 3 is an explanatory diagram of a database maintenance example 3. As shown in FIG. FIG. 3 explains maintenance for the rule DB 102 . When word groups are used in rules, there is no need to register rules in the rule DB 102 for each word in the word group. For example, since rule Ra uses word group Ga, there is no need to register rules Ra1 and Ra2 for each verb in rule DB 102 . As a result, duplication of rules can be suppressed, and memory saving of the rule DB 102 can be achieved.

<Hardware Configuration Example of Information Processing System>
Next, a hardware configuration example of one or more computers 400 configuring the information processing system will be described.

FIG. 4 is a block diagram showing a hardware configuration example of a computer. The computer 400 has a processor 401, a storage device 402, an input device 403, an output device 404, and a communication interface (communication IF 405). Processor 401 , storage device 402 , input device 403 , output device 404 and communication IF 405 are connected by bus 406 . Processor 401 controls computer 400 . A storage device 402 serves as a work area for the processor 401 . Also, the storage device 402 is a non-temporary or temporary recording medium that stores various programs and data. Examples of the storage device 402 include ROM (Read Only Memory), RAM (Random Access Memory), HDD (Hard Disk Drive), and flash memory. The input device 403 inputs data. The input device 403 includes, for example, a keyboard, mouse, touch panel, numeric keypad, and scanner. The output device 404 outputs data. Output device 404 includes, for example, a display and a printer. Communication IF 405 connects to a network and transmits and receives data.

<Example of data stored in database>
Next, examples of contents stored in the word dictionary DB 101, the rule DB 102, and the data store 103 will be described. The word dictionary DB 101, the rule DB 102, and the data store 103 may be implemented by the storage device 402 within the computer 400 shown in FIG. 4, or may be implemented by another computer accessible via the communication IF 405. In the description of the database or table below, the value of the AA field bbb (AA is the field name and bbb is the code) may be written as AAbbb. For example, the value of the group ID field 501 is written as group ID 501 .

FIG. 5 is an explanatory diagram showing an example of contents stored in the word dictionary DB 101. As shown in FIG. The word dictionary DB 101 has a group ID field 501 , an attribute field 502 , a word field 503 and a part of speech field 504 . A combination of values of fields 501 to 504 on the same line defines an entry indicating one word group. A group ID field 501 is a storage area for storing group IDs. A group ID 501 is identification information that uniquely identifies a word group.

The attribute field 502 is a storage area for storing attributes. Attributes 502 are characteristics that the word group exhibits. For example, in Japanese sentences, there are verbs in which the subject particle is a ``ga'' case, and verbs that co-occur with specific adverbs. In addition, it may be a synonym, a synonym, or a word used in a specific field (investment, medical care, etc.).

A word field 503 is a storage area for storing words. A word 503 is a word belonging to the word group. An operator (user or administrator) can add, change, or delete words 503 in the word field 503 .

The part-of-speech field 504 is a storage area for storing parts of speech. The part of speech 504 is a type in which words belonging to a word group are classified according to form and role. Note that the part of speech 504 may specify the form of the word. In the case of English words, verbs are designated, for example, from the base form (present tense), past tense, past participle form, and present progressive form, and nouns are designated from, for example, uncountable nouns, addition nouns, singular forms, and plural forms. and adjectives and adverbs are designated, for example, from the base form, comparative, and superlative. If only the part of speech is specified (no form is specified), all forms of the part of speech 504 may be included.

FIG. 6 is an explanatory diagram showing an example of contents stored in the rule DB 102. As shown in FIG. The rule DB 102 has a rule ID field 601 and a tree structure pattern field 602 . A combination of values of fields 601 and 602 on the same line defines an entry indicating one rule. A rule ID field 601 is a storage area for storing rule IDs. A rule ID 601 is identification information that uniquely identifies a rule. The tree structure pattern field 602 is a storage area for storing tree structure patterns. The operator can add, change, or delete the tree structure pattern 602 in the tree structure pattern field 602 . 1 to 3 show rules in which verbs in the tree structure pattern 602 are word groups and subjects and objects are wildcards. A word group may be applied and other words may be wildcarded. Also, the rule may apply a plurality of word groups to one tree structure pattern 602 .

FIG. 7 is an explanatory diagram showing an example of the contents stored in the data store 103. As shown in FIG. The data store 103 has a headword field 701 , a text field 702 and a tree structure data field 703 . A combination of values of fields 701-703 on the same line defines an entry for one sentence.

The headword field 701 is a storage area for storing headwords and is used for index searches. The headword field 701 has a plurality of annotation fields (three annotations a0 field 710 to annotation a2 field 712 in FIG. 7). The annotation a0 field 710 is a storage area for storing entry word 701 preset as annotation a0. An annotation a1 field 711 and an annotation a2 field 712 are storage areas for storing headwords 701 as annotations a1 and annotations a2. Annotation a1 field 711 and annotation a2 field 712 are blank in the initial state, and annotation a1 and annotation a2 are added by index update described later.

A text field 702 is a storage area for storing text. The text 702 is text data from which the tree structure data 703 is analyzed. The tree-structured data field 703 is a storage area for storing tree-structured data obtained by parsing the text according to the phrase structure rule.

<Examples of various data>
FIG. 8 is an explanatory diagram showing an example of the text 702. As shown in FIG. FIG. 8 shows an example of the text st1 in English, but it is not limited to English and may be in other languages such as Japanese.

FIG. 9 is an explanatory diagram showing an example of tree structure data and a tree structure pattern. The tree structure data tr1 is a syntax tree obtained by parsing the text st1 of FIG. 8 according to the phrase structure rules. In the tree structure data tr1, "POS" indicates the part of speech, and "ROOT" indicates the root of the syntax tree. A string of 1 to 3 capital letters indicates the type of part of speech (noun, verb, etc.). The tree structure pattern tp1 is a pattern edited by deleting unnecessary information from the tree structure data tr1 by the operator. This tree structure pattern tp1 indicates a word order rule in which the subject is a wildcard, the predicate is the verb "spin off", and the object is a wildcard.

FIG. 10 is an explanatory diagram showing an example of pattern representation. The pattern representation 1000 is used when an information processing system executes information processing. Further, by recognizing the pattern expression 1000, the operator can edit the tree structure data 703 to generate the tree structure pattern 602. FIG. In the pattern expression 1000, “_” is leaf node (leaf of the syntax tree) judgment, “|” is option, “#” is subtree (subtree in syntax tree) extraction, “!” is negation, and “*” is 0 times. Occurrences of the above subtrees, "+" indicates one or more occurrences. Note that the pattern representation 1000 shown in FIG. 10 is an example.

FIG. 11 is an explanatory diagram showing a conversion example using the pattern representation shown in FIG. In the tree structure data tr11, the selection of increase or cause whose part of speech (POS) is a verb (VP) is converted to a call (\dic.) of a word group with a group ID 501 of affect. As a result, a tree structure pattern tp11 containing word groups is generated. Note that such conversion is executed by the operator's editing operation.

<Example of information processing procedure>
FIG. 12 is a flowchart showing an example of an information processing procedure by the information processing system. The information processing system waits for a maintenance request (step S1201: No). A maintenance request is given from an instruction from the processor 401 , from a terminal via the communication IF 405 , or from the input device 403 . If there is a maintenance request (step S1201: Yes), the information processing system determines whether the maintenance request is for a word or for a rule (tree structure pattern) based on information included in the maintenance request. (step S1202).

If the maintenance request is for a word (step S1202: word), the information processing system determines whether the maintenance request for a word is for addition or deletion of a word based on the information included in the maintenance request for a word (step S1202: word). step S1203). If it is a word addition (step S1203: add), the information processing system identifies the word group to which the word is to be added from the word dictionary DB 101 (step S1204). Specifically, for example, when the word maintenance request includes the group ID of the addition destination, the information processing system selects the group ID 501 as the addition destination of the word to be added included in the word maintenance request. Identify the word group specified by .

Further, if the maintenance request for the word does not include the group ID of the addition destination, the information processing system may automatically identify the word group of the addition destination. For example, if the word to be added is a word extracted from the text 702 included in the maintenance request related to the word, the information processing system identifies from the word dictionary DB 101 a word group having attributes corresponding to the feature of the text. . Then, the information processing system adds the word to be added to the specified word group of the addition destination (step S1205), and returns to step S1201.

Also, in step S1203, if the word is to be deleted (step S1203: delete), the information processing system deletes the word to be deleted included in the maintenance request related to the word from the word group to be deleted in the word dictionary DB 101. (step S1206) and returns to step S1201. The word group to be deleted is, for example, all entries in the word dictionary DB 101 if the group ID 501 is not specified in the maintenance request for the word, and if the group ID 501 is specified, the entry specified by the group ID 501. is.

In step S1202, if the maintenance request is for a rule (step S1202: rule), the information processing system determines whether the rule-related maintenance request is for addition or deletion of a rule. This determination is made based on the information available (step S1207). In the case of addition of a rule (step S1207: add), the information processing system adds to the rule DB 102 the rule to be added included in the maintenance request regarding the rule (step S1205), and returns to step S1201.

Also, in step S1207, if the rule is to be deleted (step S1207: delete), the information processing system deletes the entry of the rule ID 601 included in the maintenance request related to the rule from the rule DB 102 (step S1209), Return to step S1201.

<Use example of information processing system>
FIG. 13 is an explanatory diagram showing a usage example of the information processing system. (1) The information processing system acquires the text stc1 from the data store 103 . In (1), the information processing system may acquire the sentence stc1 by directly specifying it, or may acquire the sentence stc1 by an index search using the headword 701 . (2) The information processing system converts the obtained sentence stc1 into tree-structured data trc by syntactic analysis. In (2), the information processing system may execute the syntax analysis, the information processing system transmits the sentence stc1 to another computer, and the other computer executes the syntax analysis to obtain the tree structure data trc as information. May be returned to the processing system. If the tree-structured data trc has already been generated, the information processing system calls the tree-structured data trc associated with the text stc1 from the data store 103 .

(3) The information processing system generates a tree-structured pattern from the tree-structured data trc by the editing operation of the operator, and uses it as a rule Rc. Here, it is assumed that the verb word group Gb is applied to the predicate of the rule Rc.

(4) The information processing system extracts the subject "X" of the sentence stc1 as the annotation a1 from the tree structure pattern of the rule Rc, and extracts the object "A" of the sentence stc1 as the annotation a2. , to be displayed on the display screen.

(5) The information processing system registers rule Rc in rule DB 102 . Note that if a rule with the same content has already been registered, the information processing system does not register the rule Rc in the rule DB 102 .

(6) The information processing system registers the tree structure data trc of (2) and the annotations a1 and a2 of (4) in the entry of the sentence stc1 in the data store 103 . As a result, it is possible to automatically generate the headword 701 of the acquired text stc1, and to improve the efficiency of subsequent index searches.

(7) The information processing system searches texts other than the text stc1 in the data store 103 to identify the text stc2 corresponding to the rule Rc. The word "K" is registered as an annotation a2 (index update). As a result, the headword 701 can be automatically generated by spreading to other texts stc2, and subsequent index searches can be made more efficient.

Next, examples of display screens in the example of use shown in FIG. 13 will be described with reference to FIGS. 14 to 19. FIG. The display screens of FIGS. 14 to 19 are display screens displayed by a computer 400 in the information processing system.

FIG. 14 is an explanatory diagram showing a display screen example 1 of the information processing system. Display screen 1400 has sample tab 1401 , validate tab 1402 , and index tab 1403 . In FIG. 14, a sample tab 1401 is displayed. Sample tab 1401 has search keyword input field 1411 , search button 1412 , and save button 1415 . A search keyword input field 1411 is an input field for the operator to input a search keyword. The search button 1412 is a button for performing an index search for the headword 701 in the data store 103 and extracting the corresponding text 702 by the operation of the operator. In this example, an index search will be described, but a full-text search of the text 702 may also be used.

In FIG. 14, it is assumed that "spin off" is entered in the search keyword input field 1411 and the search button 1412 is pressed. As a result, as shown in (1) of FIG. 13, the headword 701 in the data store 103 is searched by index, and the corresponding text 702 is displayed as the search result 1413. FIG. Each text in the search result 1413 has a check box 1414, and the information processing system selects the text with the check box 1414 checked by the operator. In FIG. 14, it is assumed that text st1 is selected. A save button 1415 is a button for saving the text selected by the check box 1414 from the search result 1413 . By pressing the save button 1415 , the text st1 whose checkbox 1414 is checked is saved in the data store 103 .

FIG. 15 is an explanatory diagram showing a display screen example 2 of the information processing system. Display screen example 2 is an example of a display screen when the validate tab 1402 is selected with the check box 1414 checked in the display screen example 1 of FIG. The validate tab 1402 has a confirmation area 1501 , a copy area 1502 , an analysis button 1503 , an annotation button 1504 , an add button 1505 and an edit area 1506 . The confirmation area 1501 has a selected sentence display area 1510 , an annotation a1 display area 1511 and an annotation a2 display area 1512 . Selected sentence display area 1510 displays the text selected by checking check box 1414 in display screen example 1 of FIG. An annotation a1 display area 1511 is an area for displaying an annotation a1 (subject). An annotation a2 display area 1512 is an area for displaying an annotation a2 (object).

In display screen example 2, the annotation a1 display area 1511 has a text input field 1513 for annotation a1. The operator refers to text st1 in selected sentence display area 1510 and enters a phrase (for example, “Nichiritsu”) corresponding to annotation a1 (subject) in text input field 1513 for annotation a1. The annotation a2 display area 1512 has a text input field 1514 for annotation a2. The operator refers to text st1 in selected sentence display area 1510 and enters a phrase (for example, “home appliance”) corresponding to annotation a2 (object) in text input field 1514 for annotation a2.

The combination of the text st1 displayed in the confirmation area 1501, the word "Nichiritsu" input in the text input field 1513 for annotation a1, and the word "home appliance" input in the text input field 1514 for annotation a2 is confirmed. data set 1500 for

A copy button 1515 is a button for copying the text of the selected sentence display area 1510 to the copy area 1502 by the operator's operation. A copy area 1502 is an area where the text st1 of the selected sentence display area 1510 is copied and displayed by pressing a copy button 1515 . An analysis button 1503 is a button for parsing the text st1 copied to the copy area 1502 (corresponding to (2) in FIG. 13). An annotation button 1504 is a button for extracting an annotation of the text st1 from the tree structure pattern edited in the editing area 1506 (corresponding to (4) in FIG. 13). An add button 1505 is a button for adding the tree structure pattern edited in the editing area 1506 to the rule DB 102 as a rule (corresponding to (5) in FIG. 13).

FIG. 16 is an explanatory diagram showing a display screen example 3 of the information processing system. A display screen example 3 is an example of a display screen when the copy button 1515 and the analysis button 1503 are pressed in the display screen example 2 of FIG. When the copy button 1515 is pressed by the operator's operation, the selected text st1 is copied to the copy area 1502 . Next, when the analysis button 1503 is pressed by the operator's operation, the tree structure data tr1 obtained by parsing the selected text st1 is displayed in the editing area 1506 (corresponding to (2) in FIG. 13).

FIG. 17 is an explanatory diagram showing a display screen example 4 of the information processing system. A display screen example 4 is an example of a display screen when the tree structure data tr1 in the editing area 1506 is edited in the display screen example 3 of FIG. For example, by the operator's operation, the information processing system assigns "a0", "a1", and "a2" indicating annotations to words to be extracted as annotations. "a0", "a1", and "a2" define rules. The annotation a0 is an annotation not to be extracted that serves as an extraction reference for the other annotations a1 and a2. That is, if the annotation a0 is a word, it is a non-extraction-target word that matches the other text, and if the annotation a0 is a word group, it is a non-extraction-target word group that includes the words of the other text. Since annotation a1 is defined as a subject (noun phrase (NP)) for annotation a0 in tree structure pattern tp1, and annotation a2 is defined as an object (noun phrase (NP)) for annotation a0 in tree structure pattern tp1, Noun phrases corresponding to the rule are extracted from other texts.

In addition, subtrees and "lemmma" (basic forms of words) that are subjectively unimportant to the operator are deleted by the operator's operation. Also, as shown in FIG. 11, a word defined in the tree structure data tr1 may be changed to a description of calling a word group containing the word.

FIG. 18 is an explanatory diagram showing example 5 of the display screen of the information processing system. Display screen example 5 is an example of a display screen when the annotation button 1504 is pressed in display screen example 4 of FIG. When the annotation button 1504 is pressed by the operator, the information processing system displays the character strings corresponding to the annotations a1 and a2 corresponding to the tree structure pattern tp1 (rule) edited in the editing area 1506 on the copy screen. Extract from the selected text st1 and display the extraction result 1800 (corresponding to (4) in FIG. 13). In this case, "Japanese electronics maker Nichiritsu" is extracted as the noun phrase of the annotation a1, and "its home appliance and industrial equipment divisions" is extracted as the noun phrase of the annotation a2. The extracted noun phrases of annotations a1 and a2 are displayed in annotation a1 display area 1511 and annotation a2 display area 1512, respectively.

As a result, the operator compares the phrase "Nichiritsu" input in the text input field 1513 for annotation a1 with the noun phrase "Japanese electronics maker Nichiritsu" of annotation a1 extracted according to the rule, and determines the likelihood of the rule. can be confirmed. Similarly, the operator compares the phrase "home appliance" entered in the text input field 1514 for annotation a2 with the noun phrase "its home appliance and industrial equipment divisions" of annotation a2 extracted according to the rule, can confirm the certainty of

Further, when the add button 1505 is pressed, the character string (edited tree structure data tr1) in the editing area 1506 becomes the tree structure pattern tp1 and is registered as a rule in the rule DB 102 (( 5)).

FIG. 19 is an explanatory diagram showing a display screen example 6 of the information processing system. A display screen example 6 is an example of a display screen when the index tab 1403 is selected in the display screen example 5 of FIG. Index tab 1403 has update button 1900 . When the operator presses the update button 1900, the information processing system updates the selected text st1 with the tree structure data tr1, the noun phrase "Japanese electronics maker Nichiritsu" of the annotation a1, and the noun phrase "Japanese electronics maker Nichiritsu" of the annotation a2. By associating "its home appliance and industrial equipment divisions" and registering it in the data store 103, the index of the entry of the selected text st1 is updated (corresponding to (6) in FIG. 13).

Similarly, the information processing system registers the noun phrase of annotation a1 and the noun phrase of annotation a2 that correspond to the rule of tree structure pattern tp1 in the data store 103 in association with the other text. , the index of the entry of the other text is updated (corresponding to (7) in FIG. 13).

<Example of processing procedure in example of use of information processing system>
FIG. 20 is a flowchart illustrating an example of a processing procedure in an example of use of the information processing system. As shown in FIG. 14, the information processing system receives input of a search keyword in the search keyword input field 1411 (step S2001), and executes an index search using the input search keyword by pressing a search button 1412 ( step S2002). The information processing system saves the text selected by the operator's operation, as shown in FIG. 14 (step S2003).

Next, as shown in FIG. 15, the information processing system sets the confirmation data set 1500 by the operator's operation (step S2004). Then, as shown in FIG. 16, the information processing system obtains tree structure data tr1 by parsing the selected text st1 (step S2005). Further, the information processing system registers the tree structure pattern tp1 edited from the tree structure data tr1 in the rule DB 102 when the operator presses the add button 1505 (step S2006). Pressing the add button 1505 corresponds to step S1207: add in FIG. 12, and registering the tree structure pattern tp1 corresponds to step S1208 in FIG.

Then, as shown in FIG. 18, when the annotation button 1504 is pressed by the operator, the information processing system converts the words and phrases from the selected text st1 to the annotation a1 according to the rules of the tree structure pattern tp1. The word/phrase of the annotation a2 is extracted and displayed as an extraction result 1800 (step S2007).

Note that the operator can repeatedly edit the tree structure pattern tp1, and the information processing system may register the tree structure pattern tp1 as a rule each time. In this case, in step S2007, the information processing system extracts annotations from the selected text for each tree structure pattern tp1. Thereafter, the information processing system updates the index of the extracted annotations in the data store 103 as shown in FIG. 19 (step S2008).

In this way, the information processing system described above has a word dictionary DB 101 and a rule DB 102, and the processor 401 performs a maintenance request reception process, and the maintenance request received by the reception process is a word-related maintenance request. a maintenance process of performing maintenance on the word group to which the word belongs to the word dictionary DB 101, and performing maintenance of the tree structure pattern on the rule DB 102 if the maintenance request is a maintenance request for a tree structure pattern. Run.

Thereby, the word dictionary DB 101 and the rule DB 102 can be maintained independently. In other words, the information processing system maintains only one of the word dictionary DB 101 and the rule DB 102 . Therefore, even if a certain word group in the word dictionary DB 101 is maintained, there is no need to maintain rules using the word group in the rule DB 102 . Conversely, maintenance of a rule in the rule DB 102 does not require maintenance of word groups used in the rule. Therefore, maintenance of the database can be facilitated.

In addition, when the maintenance request for a word is a word addition request, the processor 401 executes a specification process of specifying the attribute of the word group to which the word should belong based on the word. In the maintenance process, the processor 401 specifies Add the word to the attribute word group identified by the process.

As a result, when there is a request to add a word, the word is additionally registered in the corresponding word group in the word dictionary DB 101, but there is no need to maintain rules using the word group in the rule DB 102. FIG. Therefore, it is possible to facilitate maintenance when registering words.

In the maintenance process, processor 401 deletes the word from the word group to which the word belongs when the maintenance request for the word is a word deletion request.

As a result, when there is a request to delete a word, the word is deleted from the corresponding word group in the word dictionary DB 101, but there is no need to maintain rules using the word group in the rule DB 102. FIG. Therefore, it is possible to facilitate maintenance when deleting words.

Also, in the maintenance process, if the tree structure pattern maintenance request is a tree structure pattern addition request, the processor 401 registers the tree structure pattern in the rule DB 102 if the tree structure pattern does not exist in the rule DB 102 .

As a result, when there is a request to add a tree structure pattern, the tree structure pattern is additionally registered in the rule DB 102 as a new rule, but there is no need to maintain word groups used in the new rule in the word dictionary DB 101. . Therefore, it is possible to facilitate maintenance when registering a tree structure pattern.

Also, in maintenance processing, the processor 401 deletes the tree structure pattern from the rule DB 102 when the maintenance request for the tree structure pattern is a deletion request for the tree structure pattern.

As a result, when there is a request to delete a tree structure pattern, the tree structure pattern is deleted from the rule DB 102, but there is no need to maintain word groups used in the tree structure pattern in the word dictionary DB 101. FIG. Therefore, it is possible to facilitate maintenance when registering a tree structure pattern.

Also, the processor 401 can access the data store 103 storing a plurality of sentences, and extracts specific tree structure data that is the analysis result of a specific sentence in the data store 103 that contains a specific word among the plurality of sentences. , an acquisition process for acquiring a specific tree structure pattern abstracted using a specific word group including a specific word, and a specific word in the specific tree structure pattern obtained by the acquisition process from the specific tree structure data Extraction processing that extracts words included in words that co-occur with groups (for example, subjects and objects when a specific word group is a predicate verb), and the words extracted by extraction processing can be displayed on the display screen In the maintenance process, processor 401 outputs a specific is registered in the rule DB 102.

As a result, words corresponding to a specific tree structure pattern can be displayed as annotations for a specific sentence. Therefore, for example, if the operator has previously selected a word or phrase that co-occurs with a specific word group for a specific sentence, by comparing the selected word with the annotation, a specific tree structure pattern can be obtained. The probability can be confirmed and registered in the rule DB 102 .

Also, the processor 401 can access the data store 103 storing a plurality of sentences, and extracts specific tree structure data that is the analysis result of a specific sentence in the data store 103 that contains a specific word among the plurality of sentences. , an acquisition process for acquiring a specific tree structure pattern abstracted using a specific word group including a specific word, and a specific word in the specific tree structure pattern obtained by the acquisition process from the specific tree structure data An extraction process for extracting words included in words co-occurring with a group, and an update process for updating the data store 103 by associating the words extracted by the extraction process with a specific sentence are executed. , the processor 401 registers the specific tree structure pattern in the rule DB 102 when the maintenance request for the specific tree structure pattern is a request to add the specific tree structure pattern.

As a result, words corresponding to a specific tree structure pattern can be associated and registered as annotations of a specific sentence, and the specific tree structure pattern used for the association can be registered in the rule DB 102 as a rule. This also allows, for example, when searching the data store 103, a particular sentence can be extracted from the data store 103 by performing an index search using the associated annotation as a headword.

In addition, the processor 401 extracts from other tree structure data, which is the analysis result of sentences other than the specific sentence among the plurality of sentences, a word that is included in a word that co-occurs with a specific word group in a specific tree structure pattern. The data store 103 is updated by extracting other words and associating other words extracted by the extraction process with other sentences.

As a result, for other sentences in the data store 103 as well, other words corresponding to specific tree structure patterns can be associated and registered as annotations of other sentences, and the rules defined by the specific tree structure patterns can be registered. It can spill over into other sentences.

[2. Example of additional registration of words to word dictionary DB 101]
As described above, the information processing system has a rule DB 102 that defines tree structure patterns representing grammatical structures as rules, and a word dictionary DB 101 that collects words representing relationships. With these, the information processing system defines a grammatical structure for relationships such as sentence elements, and extracts annotations a0 to a2 as relational annotations from the text data. The information processing system compares the rule DB 102 and the word dictionary DB 101 and extracts words that match both the rule DB 102 and the word dictionary DB 101 . Thus, in order to be able to extract a large amount of relational information, it is necessary to add words to the word dictionary DB 101 .

However, there is a problem that it is difficult to increase the number of words in the word dictionary DB101. In Japanese Patent Laid-Open No. 2002-200011, since example sentences including relationships to be extracted are searched only by keyword search, example sentences that do not match the tree structure rule are also searched. Therefore, when adding a new example sentence, it is necessary to edit both the rule DB 102 and the word dictionary DB 101 .

For this reason, the information processing system extracts relational annotations (for example, a combination of words corresponding to subjects and objects) and rules indicated by tree structure patterns in the rule DB 102 (for example, sentence elements (for example, subjects, By using a combination of predicate and object), candidates for registration (for example, predicate) are obtained from a large amount of text data. Since it is guaranteed that the example sentences containing the registration candidates thus obtained match the rules indicated by the tree structure pattern, simply by adding the acquired registration candidates to the word dictionary DB 101, new relationships can be created from the example sentences. Extraction becomes possible.

Therefore, the performance of the information processing system can be improved by updating only the word dictionary DB 101 without editing the rule DB 102 . Editing the rule DB 102 requires a certain understanding of the grammatical structure, so it is difficult to edit unless it is trained personnel. In actual operation, for example, personnel familiar with the development of the rule DB 102 create a certain amount of the rule DB 102, and personnel who only update the word dictionary DB 101 update the word dictionary DB 101 using the information processing system. Work sharing is possible.

Note that, in this example, the above [1. Maintenance Example for Word Dictionary DB], it is assumed that the rule DB 102 is constructed and headwords 701 (annotations a0 to a2) are added to the text 702 in the data store 103 . It is also assumed that "enter" has been registered as one of the word groups in the word dictionary DB. "enter" is a word group including "enter" and "participate".

In this example, a word group is defined as a predicate word group, and a new predicate word is additionally registered in the word dictionary DB 101 based on the relationship between the subject, predicate, and object in the example sentence. However, additional registration to the word dictionary DB 101 is not limited to predicates as long as the word corresponds to a word group. Although the example sentences are written separately in English and Japanese, the distinction between them does not matter.

<Additional word registration example>
FIG. 21 is an explanatory diagram showing an example of word addition registration, and FIG. 22 is a flow chart showing an example of a word registration processing procedure.

(1) The information processing system searches for an example sentence 2001 corresponding to one of rules R1, R2, . S2201). In this example, in the retrieved example sentence 2001, "various companies" corresponds to the annotation a1, and "in the retail market" corresponds to the annotation a2.

Note that the example sentence 2001 is text data that exists on a network outside the information processing system (for example, a web page on the Internet or document data in a database server). Alternatively, the example sentence may be the text 702 to which the headword 701 is not added in the data store 102 .

(2) The information processing system syntactically analyzes the example sentence 2001 retrieved in (1) to generate tree structure data 2002 (step S2202). The tree structure data 2002 is, for example, a syntax tree generated according to phrase structure rules by syntactic analysis (morphological analysis and phrase structure analysis) (see FIG. 9). In order to improve the accuracy of processing after (3), the user may edit the tree-structured data 2002 into a tree-structured pattern as shown in FIG.

(3) The information processing system compares the tree structure data 2002 obtained in (2) with the rule R in the rule DB 102 (step S2203). At this time, the information processing system does not use the condition C of the annotation a0 for the comparison because the word dictionary DB 101 is not applied in (3). Specifically, the information processing system removes condition C from rule R, for example.

For example, for rule R1, condition C defines the word group "enter." Therefore, "\dic.enter" corresponding to "enter" (that is, "enter" and "participate") is excluded from rule R1.

(4) The information processing system obtains an extraction result 2003 from the example sentence 2001 by the comparison in (3) (step S2204). For example, by comparing with rule R1, the information processing system extracts the subject "various companies" that meet the condition of annotation a1 of rule R from example sentence 2001 as extraction result 2003. FIG.

In addition, the information processing system extracts the object "to the retail company" that satisfies the condition of the comment a2 of the rule R from the example sentence 2001 as the extraction result 2003. FIG. In addition, since the condition C of the annotation a0 is excluded in (2) above, the information processing system extracts the predicate "appearing" that does not correspond to the condition C ("enter" and "participate") as the extraction result 2003. can do. The original form of the verb "appearing" extracted by excluding this condition C becomes a candidate for registration in the word dictionary DB 101. FIG. Thereby, unregistered words can be efficiently collected. Note that the extraction result 2003 should include at least registration candidates.

(5) The information processing system confirms whether or not the registration candidate "appear" in the extraction result 2003 of (4) can be newly added to the word dictionary DB 101 (step S2205). Specifically, for example, the information processing system determines whether the text 702 corresponding to the extraction result 2003 exists in the data store.

If the text 702 corresponding to the extraction result 2003 exists, the registration candidate "appearing" of the extraction result 2003 has already been registered in the word dictionary DB 101 . Therefore, there is no need to register the registration candidate "Appear" (step S2206: No), and the information processing system does not execute the next process (6) (step S2207).

On the other hand, when the text 702 corresponding to the extraction result 2003 does not exist, the registration candidate "appear" in the extraction result 2003 becomes an unregistered predicate in the word dictionary DB 101 . There is a need to register the registration candidate "appear" (step S2006: Yes). Therefore, the information processing system executes the following process (6) (step S2207).

(6) The information processing system additionally registers the registration candidate "appear" of the extraction result 2003 in the word dictionary DB 101 (step S2207). In this additional registration, as described above, the information processing system may additionally register the registration candidate without executing (5). If it matches the registered predicate, it is simply overwritten, so there is no problem even if the process (5) (step S2205) is not executed.

However, if (5) is not executed, even if the data store 103 does not have text corresponding to the annotations a1 and a2 of the example sentence 2001, the registration candidate "appear" may be additionally registered in the word dictionary DB 101. be. Thus, by executing the process (5) (step S2205) and then executing the process (6) (step S2207), the accuracy of new registration to the word dictionary DB 101 can be improved (suppression of erroneous registration). can be planned.

In addition, in the process (6) (step S2207), regardless of whether the process (5) (step S2205) is executed or not, the information processing system displays the registration candidate "appearing" as the computer 400 operated by the user. may be displayed on the display screen, and the user may be prompted to approve or disapprove of the registration by the user's operation. This allows the user to confirm whether or not to perform additional registration before additional registration.

Thus, the information processing system can collect only the example sentences 2001 that match the rule R, and can improve the efficiency of expansion of the word dictionary DB 101 . Also, the word dictionary DB 101 can be expanded without updating the rule DB 102 .

<Screen transition example>
Next, screen transition of the display screen in the example of word addition registration described above will be described.

FIG. 23 is an explanatory diagram showing a display screen example 7 of the information processing system. FIG. 24 is an explanatory diagram showing a display screen example 8 of the information processing system. The INVESTIGATE tab 2300 and the DICT tab 2400 are used for additional word registration on the display screen 1400 .

FIG. 23 shows an example of a display screen in the process (1) (step S2401) described above. In FIG. 23 the user selects the INVESTIGATE tab 2300 . As a result, the display screen 1400 becomes the screen shown in FIG. When the user selects the target relationship type (for example, "penalize") from the "Select Annotator" pull-down and clicks the SEARCH button 2302, the information processing system "penalizes" using the existing rule R and the word dictionary DB 101. Example sentences 2302A and 2302B indicating the relationship between are searched, and a search result 2303 is displayed on the display screen 1400. FIG.

If a word for narrowing down is entered in the "Please Input keyword" column 2304, only example sentences 2302A and 2302B containing the word in the sentence are retrieved and displayed. The user checks the check box 2305 at the left end of the search result 2303 for example sentences 2302A and 2302B that correctly extract the relation "penalize", and clicks the STORE button 2306 to temporarily store them. The user then selects the DICT tab 2400 . As a result, the display screen 1400 becomes the display screen shown in FIG.

FIG. 24 shows an example of a display screen in the processes (2) to (7) described above (steps S2402 to S2407). In FIG. 24, the user clicks the SEARCH button 2401 to search for predicate candidates. As a result, predicate candidates 2402A and 2402B are displayed as extraction results 2402 together with example sentences 2302A and 2302B. In example sentences 2302A and 2302B, character strings of annotations a0 to a2 to be extracted are highlighted.

Names of tree structure patterns 2403A and 2403B indicating the rule R used for extraction are also displayed. Here, registration candidates that are not registered in the word dictionary DB 101 are preferentially displayed. Registration candidates already registered in the word dictionary DB 101 are also displayed together with the name 2404 of the corresponding word dictionary DB 101 . The user confirms the displayed registration candidates and adds appropriate registration candidates to the word dictionary DB 101 called by the corresponding tree structure patterns 2403A and 2403B. In FIG. 24, the word "impose" is additionally registered.

As a result, the word dictionary DB 101 is expanded, and more relationships can be extracted. In addition, by clicking the update button 1900 of the index tab 1403 after expanding the word dictionary DB 101, the computer 400 displaying the display screen 1400 in the information processing system is updated to the computer 400 that can access the data store 103. An instruction to write annotations a0 to a2 is sent.

Computer 400 having access to data store 103 writes annotations a0-a2 to example sentences 2302A and 2302B stored in data store 103. FIG. As a result, the headwords 701 of the example sentences 2302A and 2302B corresponding to the extraction result 2402 are updated. By repeating such operations, more extraction results 2402 can be used for additional registration of words, so more word candidates can be collected, and the number of words registered in the word dictionary DB 101 increases.

It should be noted that the present invention is not limited to the embodiments described above, but includes various modifications and equivalent configurations within the scope of the appended claims. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and the present invention is not necessarily limited to those having all the described configurations. Also, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Moreover, the configuration of another embodiment may be added to the configuration of one embodiment. Moreover, other configurations may be added, deleted, or replaced with respect to a part of the configuration of each embodiment.

In addition, each configuration, function, processing unit, processing means, etc. described above may be realized by hardware, for example, by designing a part or all of them with an integrated circuit, and the processor realizes each function. It may be realized by software by interpreting and executing a program to execute.

Information such as programs, tables, files, etc. that realize each function is stored in storage devices such as memory, hard disk, SSD (Solid State Drive), or IC (Integrated Circuit) card, SD card, DVD (Digital Versatile Disc) recording Can be stored on media.

In addition, the control lines and information lines indicate those considered necessary for explanation, and do not necessarily indicate all the control lines and information lines necessary for mounting. In practice, it can be considered that almost all configurations are interconnected.

101 word dictionary DB
102 Rule DB
103 data store 400 computer 401 processor 402 storage device

Claims (8)

An information processing system having a processor that executes a program and a storage device that stores the program,
a word dictionary database that stores word groups, which are word groups grouped by predetermined attributes;
a rule database that stores a tree structure pattern abstracted using the word groups for tree structure data indicating relationships between words in a sentence regarding elements of sentences;
The processor
Acquisition processing for acquiring a target sentence including a combination of a word of a first element to which the word group does not apply and a word of a second element to which the word group does not apply among sentence elements;
a determination process of determining whether or not the target sentence acquired by the acquisition process corresponds to a specific tree structure pattern obtained by excluding a third element corresponding to the word group among the elements of the sentence;
an extraction process of extracting words of the third element from the target sentence determined by the determination process to correspond to the specific tree structure pattern, and outputting an extraction result;
An information processing system characterized by executing The information processing system according to claim 1,
The processor
An information processing system, characterized by executing a registration process of registering a word of the third element in the target sentence in a word group corresponding to the third element excluded from the specific tree structure pattern. The information processing system according to claim 1,
The information processing system, wherein in the extraction process, the processor outputs the extraction result in a displayable manner. The information processing system according to claim 3,
The processor
registration for registering the word of the third element in the target sentence in a word group corresponding to the third element excluded from the specific tree structure pattern when a registration instruction is input after the extraction result is displayed; An information processing system characterized by executing processing. The information processing system according to claim 1,
having a data store that stores a set of sentences;
The processor
Information characterized by executing confirmation processing for confirming whether or not a sentence composed of a combination of a word of the first element and a word of the second element exists in the data store, and outputting a confirmation result. processing system. The information processing system according to claim 2,
having a data store that stores a set of sentences;
The processor
performing a confirmation process for confirming whether a sentence composed of a combination of the words of the first element and the words of the second element exists in the data store;
In the registration process, if the confirmation process confirms that a sentence composed of a combination of the word of the first element and the word of the second element does not exist in the data store, the processor performs registering the word of the third element in the word group corresponding to the third element excluded from the specific tree structure pattern;
An information processing system characterized by: The information processing system according to claim 3,
having a data store that stores a set of sentences;
The processor
performing a confirmation process for confirming whether a sentence composed of a combination of the words of the first element and the words of the second element exists in the data store;
The processor
When it is confirmed by the confirmation process that a sentence composed of a combination of the word of the first element and the word of the second element does not exist in the data store, and a registration instruction is input after the extraction result is displayed 7. An information processing system, characterized by executing a registration process of registering the word of the third element in the target sentence in a word group corresponding to the third element excluded from the specific tree structure pattern. A processor that executes a program, a storage device that stores the program, a word dictionary database that stores word groups that are groups of words grouped according to a predetermined attribute, and a relationship between words in a sentence regarding sentence elements. An information processing method by an information processing system having a rule database for storing a tree structure pattern abstracted from the tree structure data using the word group,
The processor
Acquisition processing for acquiring a target sentence including a combination of a word of a first element to which the word group does not apply and a word of a second element to which the word group does not apply among sentence elements;
a determination process of determining whether or not the target sentence acquired by the acquisition process corresponds to a specific tree structure pattern obtained by excluding a third element corresponding to the word group among the elements of the sentence;
an extraction process of extracting words of the third element from the target sentence determined by the determination process to correspond to the specific tree structure pattern, and outputting an extraction result;
An information processing method characterized by executing

Images