JP3358100B2 - Japanese question message analysis method and device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for analyzing a Japanese question message, and more particularly, to analyzing a Japanese message input in a text format used in an information retrieval system, and efficiently reducing an information request in the message. The present invention relates to a method and an apparatus for analyzing a Japanese question message in a Japanese message structuring apparatus that structurally structures.

[0002]

2. Description of the Related Art In an information retrieval system, a major point is how to express an information request desired by a user. Generally, there is a format in which a user selects a plurality of keywords by himself and logically combines the selected keywords to express an information request. For example, if you do not know your name (store name) and want to search for a phone number,
The occupation (or occupation type) and the address of the telephone number to be searched are entered in the form of "occupation AND address" and a keyword of "restaurant business AND Minato-ku" is input. In this case, a keyword "sushi restaurant AND Minato-ku Azabu" is input when narrowing down because a considerable number of search target data has appeared. In order to further narrow down the search data, the keyword "sushi restaurant AND 3 Azabu, Minato-ku" is input. This example shows a case where keywords are combined by a logical expression (AND). Conversely, there is a method of changing a keyword using a logical expression (OR) and relaxing a search condition to search a database. .

[0003] Other methods include a method of arriving at necessary information by repeating browsing based on a topic in a search target field, and a method of expressing an information request by a pseudo natural language sentence. There is also a method in which a question message described by a user is subjected to language analysis, and information requests are extracted and structured.

[0004]

However, the above-mentioned conventional method of logically combining selected keywords is as follows.
It is difficult to express diversified and complicated information requirements of users in the form of logical combination of keywords.

[0005] In addition, the method of repeating browsing based on the topic of the target field and the method of expressing by a pseudo-natural language sentence have a problem that when users are diversified and complicated requests cannot be coped with. In addition, a method of linguistically analyzing a question message described by a user to derive an information request is based on the fact that a question message usually includes many sentences other than a sentence expressing an information request. In order to express the information request, it is difficult to clarify the strategy that defines which sentence should be focused on. Therefore, perform a uniform linguistic analysis process over the entire question message. There is a need.

[0006] The present invention has been made in view of the above points, and when a user uses an information retrieval system,
Japanese Query Message Analysis Method and Apparatus that Allows Users to Give Information Requests to Information Retrieval Systems Using Natural Question Messages Against Difficulty of Expressing Information Requests by Logical Combination of Keywords The purpose is to provide.

[0007]

FIG. 1 is a diagram for explaining the principle of the present invention. The present invention relates to a method for analyzing a Japanese question message which analyzes a Japanese question message input in a text format, wherein a Japanese question message is input (step 1) and the input processing target Japanese question message is analyzed. , Paying attention to a specific expression string,
An information request sentence describing the core part of the information request is extracted (step 2), and the reference expression included in the extracted information request sentence is resolved based on the characteristics of the information request sentence (step 3). If the plurality of information request statements described above state information requests for the same object, the structures of the plurality of information request statements are integrated to structure the main information request in the Japanese question message (step 4). Then, a structured information request statement is output (step 5).

Further, according to the present invention, when extracting the information request sentence (step 2), the information request which is a functional role played by the Japanese sentence constituting the Japanese question message and the sentence of the goal statement which is the purpose of the user are provided. Referring to a sentence type determination dictionary in which a combination of the type and the information request and the expression character string pattern for determining the goal statement is stored in advance, it is determined whether the Japanese question message corresponds to the specified sentence type. Judgment is made by comparing character strings, and a sentence having an information request function is extracted from the Japanese question message.

Further, according to the present invention, when an information request sentence is extracted, a morpheme / syntax analysis is performed on the extracted information request sentence, and a unique semantic structure pattern is previously specified for each entry in the Japanese question message. By referring to the stored semantic structure pattern dictionary, the entry is compared with the analysis result of the information request sentence that has been subjected to morpheme / syntax analysis, and the semantic structure of the information request sentence is obtained.

Further, according to the present invention, when the reference expression is eliminated (step 3), the pattern of the reference expression that can be included in the information request sentence in the Japanese question message, the search direction for searching for the reference destination of the reference expression, and Referencing the sentence type of the Japanese sentence to be searched and the linguistic expression unit of the reference expression in advance by referring to the reference expression resolution control table, and checking whether the sentence type of the searched Japanese sentence corresponds to the specified sentence type Is determined, and if it corresponds, the reference expression resolution process is controlled.

FIG. 2 is a diagram showing the principle of the present invention. The present invention relates to a Japanese-language question message analyzer for analyzing a Japanese-language question message input in a text format, comprising: an input unit 1 for inputting a Japanese-language question message; An information request sentence extracting unit 4 that extracts an information request sentence that describes a core part of an information request by focusing on a specific expression character string, and an information request sentence extracted by the information request sentence extracting unit 4. When a reference expression is included, a reference expression canceling unit 10 that resolves the reference expression based on the characteristics of the information request sentence, and a case where the plurality of extracted information request sentences state an information request for the same object. , Unifying the structure of multiple information request statements,
Information request structuring means 11 for structuring a main information request in a Japanese question message, and information request structuring means 1
Output means 12 for outputting the information request sentence integrated by 1
And

Further, the information request sentence extracting means 4 includes a function for requesting information, which is a functional role played by a Japanese sentence constituting a Japanese question message, and a sentence type for a goal statement, which is a purpose of a user. The sentence type determination dictionary 2 in which combinations of expression character string patterns for determining the goal statement are stored in advance, and the Japanese question message corresponds to the specified sentence type with reference to the sentence type determination dictionary 2. A sentence type determination unit 3 that determines whether a sentence has a character string, and a sentence type determination activation unit 41 that activates the sentence type determination unit 3 and extracts a sentence having an information request function from a Japanese question message.

The Japanese-language question message analyzing apparatus according to the present invention comprises: a basic Japanese-language analysis unit 5 for performing morpheme and syntax analysis on the information request sentence extracted by the information request sentence extraction unit 4; , A semantic structure pattern dictionary 6 in which a unique semantic structure pattern is stored in advance for each entry, and an analysis result of the entry and the information request sentence analyzed by the Japanese basic analysis means 5 with reference to the semantic structure pattern dictionary 6. And a semantic structure analyzing means 7 for determining the semantic structure of the information request sentence by comparing

[0014] The above-mentioned reference expression eliminating means 10 includes a reference expression pattern that can be included in an information request sentence in a Japanese question message, a search direction for searching for a reference destination of the reference expression, and a Japanese sentence to be searched. Using the reference expression elimination control table 8, the reference expression elimination control table 8, and the sentence type determination unit 3 in which the language type of the reference expression and the linguistic expression unit of the reference expression are stored in advance. And a cancellation control means 9.

[0015]

According to the present invention, a sentence describing a core part of an information request is extracted from an input Japanese question message based on a specific expression character string. At this time, if the extracted sentence does not match the predetermined sentence type, by controlling not to be an information request sentence, a sentence that is not directly related to the information request is not extracted. In the subsequent processing, useless processing is avoided.

In the present invention, analysis and cancellation of reference information are performed using characteristics as information request sentences. This allows
If the extracted information request sentence includes a reference requirement for indicating the main subject, a reference destination of the reference expression is obtained, referenced, identified, and the reference expression is resolved.

Further, according to the present invention, the identity of a plurality of information request objects is determined, and when the identity can be confirmed, the semantic structures of a plurality of information request sentences are integrated into a single semantic structure. It is possible to structure and output an information request in a Japanese message to be processed.

Thus, the user of the information retrieval system can perform complicated and diverse information request instructions by searching the database by inputting a natural Japanese question message.

[0019]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 3 shows the configuration of the Japanese-language question message analysis device according to one embodiment of the present invention. The Japanese question message analyzer shown in FIG. 1 includes an input unit 1, a sentence type determination dictionary 2, a sentence type determination unit 3, an information request sentence extraction unit 4, a Japanese basic analysis unit 5, a semantic structure pattern dictionary 6, and a semantic structure. It comprises an analysis unit 7, a reference expression elimination control table 8, a reference expression elimination control unit 9, a reference expression elimination unit 10, an information request structuring unit 11, an output unit 12, and a processing table 13.

When a text-based Japanese question message is input, the input unit 1 divides the message into sentences and stores the message in a processing table 13 described later with reference to FIG.
Note that the input unit 1 may be connected to a file system or a terminal of a specific computer system, or may be connected to a computer network that has been rapidly developing and spreading in recent years.

The sentence type determination dictionary 2 stores a combination of a sentence type (information request type / goal assertion type) and an expression character string pattern for determining the sentence type in advance. is there. The sentence type determination unit 3 refers to the sentence type determination dictionary 2 to determine whether a sentence extracted from the input Japanese question message to be processed corresponds to the specified sentence type by comparing character strings. .

The information request sentence extracting unit 4 includes a processing table 13
Reads the Japanese question message in the sentence type determination unit 3
To extract a sentence having an information request function from the Japanese question message. For details, the information request sentence extracting unit 4
Specifies the sentence type to be extracted as "information request", starts the sentence type determination unit 3, and extracts a sentence directly describing the information request included in the input Japanese question message.
At this time, the sentence type determination unit 3 refers to the sentence type determination dictionary 2 and performs character string collation between the entry and the Japanese sentence to be processed.

The Japanese basic analysis unit 5 performs a morpheme / syntax analysis on a Japanese sentence. The analysis method uses an existing system, and the details are not specified in the present invention. The semantic structure pattern dictionary 6 stores a semantic structure pattern unique to the Japanese question message in advance, and is referred to by the semantic structure analysis unit 7.

The semantic structure analysis unit 7 analyzes the contents of the semantic structure pattern dictionary 6 and the results of analysis (morpheme, syntax analysis result) of the entry and the information request sentence extracted by the information request sentence extraction unit 4. ) To determine the semantic structure of the information request sentence. Reference expression resolution control table 8
The search direction to search the reference destination of the reference expression in advance, the sentence type of the Japanese sentence to be searched, the reference destination, using the reference expression that can be included in the information request sentence in the Japanese question message and the semantic attribute in the semantic structure as keys Is a table in which the linguistic expression unit of the expression is stored, and is referred to by the reference expression resolution control unit 9.

The reference expression elimination control unit 9 controls the reference expression elimination process using the reference expression elimination control table 8 and the sentence type determination unit 3. More specifically, based on the semantic structure of the information request sentence input from the semantic structure analysis unit 7, a candidate sentence that may include a reference destination is selected based on the information on the search direction obtained from the reference expression resolution control table 8. I do. At this time, the reference expression elimination control unit 9 activates the sentence type determination unit 3 using the sentence type information obtained from the reference expression elimination control table 8, and determines the eligibility of the candidate sentence.

The reference expression elimination unit 10 obtains a reference destination of the reference expression. More specifically, when the candidate expression is obtained by the reference expression elimination control unit 9, the reference expression elimination unit 10 that actually obtains the reference destination of the reference expression is activated, and the necessary reference expression included in the input information request sentence is started. Perform the cancellation. Here, the resolution of the reference expression means, for example, if the object of the information request is indicated by the description of "the cable necessary for that", if it is not possible to determine what "the" indicates , "Cable" cannot be defined, so that the reference expression "that" is searched, the pointed content is acquired, and the reference expression is eliminated.

When a plurality of information request sentences are extracted from the input Japanese message, the information request structuring unit 11
By determining the identity of the object of the information request, the information request in the Japanese message to be processed is structured. In detail, by the above processes, the identity of the object of the information request is determined, and if the identity can be confirmed, the semantic structures of the plurality of information request sentences are integrated and processed as a single semantic structure. Structure the information request in the target Japanese message.

The output unit 12 extracts a target Japanese question message and outputs a structured information request to a device connected at a later stage. The hardware for realizing the Japanese-language question message analysis device may be any device including a CPU and a memory, for example, and can be realized by a device such as a computer system.

FIG. 4 is a flowchart showing the outline of the operation of the Japanese-language question message analyzing apparatus according to one embodiment of the present invention. Step 101) When the user inputs a Japanese question message in which the information request is described in text format by the input unit 1 as shown in FIG. 5, the message is transferred to the processing table 13 as shown in FIG. You.

Step 102) The information request extraction unit 4
For the Japanese question message acquired from the processing table 13, the sentence type to be extracted is set to "information request" and the sentence type determination unit 3 is started. Step 103) The sentence type determination unit 3 performs character matching between an entry in the sentence type determination dictionary 2 storing a combination of the sentence type and an expression character string pattern for determining the sentence type, and a Japanese sentence to be processed. A sentence (information request sentence) directly describing the information request included in the Japanese question message is extracted and transferred to the information request sentence extracting unit 4. For example, FIG.
In the example above, the third sentence "I'm looking for the necessary connection cable," and the fourth sentence "Do you know where you are selling? ] Is applicable.

Step 104) Information request sentence extraction unit 4
Acquires the information request sentence from the sentence type determination unit 3 and transfers it to the Japanese basic analysis unit 5. The information request sentence extracting unit 4
Will be described in detail with reference to FIG. Step 105) The Japanese basic analysis unit 5 analyzes the morpheme / syntax of the information request sentence transferred from the information request sentence extraction unit 4, and transfers the analysis result to the semantic structure analysis unit 7.

Step 106) The semantic structure analysis unit 7
The semantic structure of the information request sentence is obtained by comparing the entry in the semantic structure pattern dictionary 6 storing the unique semantic structure pattern in the Japanese question message with the analysis result of the Japanese basic analysis unit 5. The obtained meaning structure is transferred to the reference expression elimination control unit 9.

Step 107) Reference expression cancellation control section 9
Stores the reference expression pattern that can be included in the information request sentence in the Japanese question message, the search direction for searching for the reference destination of the reference expression, the sentence type of the Japanese sentence to be searched, and the linguistic expression unit of the reference expression. The reference expression elimination control table 8 is searched based on the semantic structure of the input information request sentence, and a candidate sentence that may include a reference destination is selected.

Step 108) Reference expression cancellation control section 9
Activates the sentence type determination unit 3 to determine the eligibility of the selected candidate sentence. Step 109) When the candidate expression is obtained by the reference expression elimination control unit 9, the reference expression elimination unit 10 for actually obtaining the reference destination of the reference expression is activated, and the necessary reference expression included in the input information request sentence is started. Perform the cancellation.

Step 110) Information request structuring unit 11
If a plurality of information request sentences are extracted from the Japanese message input by the above process, the same is determined for the target sentence of the information request. Step 111) If the identity can be confirmed, the information request structuring unit 11 integrates the semantic structures of the plurality of information request units and structures the information request in the Japanese message to be processed as a single semantic structure. .

Step 112) The output unit 12 extracts from the target Japanese question message by the above procedure,
The structured information request is output to a device connected at a later stage. Next, a Japanese question message analysis system using a Japanese question message analysis device will be described in detail using the Japanese question message shown in FIG. 5 as an example.

I) Input Processing First, a Japanese question message shown in FIG. The input unit 1 decomposes and stores the input Japanese question message as a set of sentences into a table structure as shown in FIG. The entry of the processing table 13 shown in FIG. 6 includes five fields. The first field has a sentence number, the second field has a sentence character string, and the third field.
The fourth field stores a sentence type, the fourth field stores a sentence subtype, and the fifth field stores a pointer to a memory area that holds information on the structure of a sentence.
In the third to fifth fields of each entry in the processing table 13, “NIL”, which is a symbol indicating that no information exists, is stored as an initial value. Input unit 1
Transfers the contents of the processing table 13 shown in FIG.

Ii) Information Request Sentence Extraction Processing Next, the outline of the operation logic of the information request sentence extraction unit 4 will be described. FIG. 7 is a flowchart showing a detailed operation of the information request sentence extracting unit according to one embodiment of the present invention. Step 201) First, as an initial value, the count i of the sentence of the input message is set to 1, and the total number of the sentences of the input message is set to N.

Step 202) From the input unit 1, a Japanese question message is input to the processing table 13 in the form shown in FIG. Step 203) The information request sentence extracting unit 4 activates the sentence type determining unit 3 by designating the sentence type to be extracted as "information request".

Step 204) Information request sentence extraction unit 4
Moves to step 205 if the sentence type directly describing the information request included in the input Japanese question message can be obtained, and moves to step 206 otherwise. Step 205) If the sentence type of the information request can be obtained, the extracted sentence type (that is, the information request) and the sentence subtype obtained as a side effect are entered in the corresponding entry of the processing table 13. Here, the sentence subtype indicates the gist of the sentence acquired for the sentence type "information request", and the third sentence in FIG. 6 "I am looking for a necessary connection cable," Is sentence subtype "get goods". The information request is entered in the third field of the entry on the processing table 13 (FIG. 6), and the sentence subtype is entered in the fourth field.

Step 206) The count i of the message is incremented. Step 207) If the message count i is larger than N, the process is terminated. If there is a message to be processed, the process proceeds to Step 202.

Next, the sentence type determination dictionary 2 will be described. FIG. 8 shows an example of a sentence type determination dictionary according to an embodiment of the present invention. As shown in the figure, the sentence type determination dictionary 2 is composed of two parts: an expression character string table as shown in the figure (A), and a sentence type determination table as shown in the figure (B). .

The expression character string table of FIG. 9A has a table structure in which each entry has three fields. The first field has an entry ID, and the second field has the function of the expression character string. Function I that expresses
D, an expression character string is stored in the third field.

The sentence type determination table shown in FIG. 4B has a table structure having three fields.
Field stores a sentence type, a second field stores a sentence subtype, and a third field stores a function type that represents a combination of a sentence type and a character string pattern for determining the sentence subtype. A list of pointers to the first element of a structure called a list is stored.

In the sentence type of the first field, "information request" and "goal statement" are set as shown in FIG. The “goal statement” expresses a user's own goal (goal). Function ID of the third field
Each cell element in the list has two fields. The first field in the function ID list field stores a function ID, and the second field stores a list of pointers to cells that can follow.

The sentence type determination unit 3 compares the entry of the sentence type determination table of the sentence type determination dictionary 2 shown in FIG. 8B with a Japanese sentence to be processed, thereby obtaining a sentence to be processed. Determines whether has the specified statement type. The outline of the operation logic of the sentence type determination unit 3 will be described below.

FIG. 9 is a flowchart showing the outline of the operation of the sentence type determination unit according to one embodiment of the present invention. The processing shown in FIG. 4 includes four blocks (steps 300, 400, 5).
00, 600). See Figure 1 for details.
0, FIG. 11, FIG. 12, and FIG. In the following flowcharts, the following symbols and functions are shown.・ I, j, k, L, M, N ... integer ・ R1, R2, R3, R4 ... set ・ ｛｝ ... empty set ・ {A, B, C} ... has A, B, C Set · + ··· Operation to create a union to the set · S: Character string of target Japanese sentence · L (i): Entry of expression string list with entry ID = i · L (i). fid, L (i). string: Function ID and expression character string in the entry ・ T (i): i-th entry in the sentence type determination table ・ T (i). type: sentence type in the entry • T (i). pointer (k): function ID in the entry
K-th pointer in the list search (X, Y): a function that returns the starting position of the character string X in the character string Y when the character string Y is a partial character string of the character string X. Returns 0 if not a substring. Comb (i, X): a function for generating a set having a set that can be generated by selecting an element of an i child from the set X as an element.
If the number of elements of X is N, comb (N, X) = {X}. Check-overlap (X): X is an element of the set R2,
It has the structure ｛…, {pos-i, length-i, fid-i},…｝. check-overlap (X) checks all the elements of X, and from the information on the head position (pos) and length (len) of the substring,
This function checks whether an overlap occurs in each partial character string stored in X and returns yes / no. Check-list (pointer, X): pointer is a pointer stored in the function ID list of a certain entry in the sentence type determination table. X is an element of the set R3, and has a structure of {..., {pos-i, length-i, fid-i ...}}. check-list (pointer, X) checks whether the function ID (fid-i) in each element of X and the function ID stored in each element cell in the function ID list pointed to by the pointer all match, This function returns yes / no. • compute-length (X): X is an element of the set R4,
{..., {pos, length-i, fid-i}, ...}, i, k}. compute-length (X) scans each element of the set {..., {pos, length-i, fid-i}, ...} which is the first element of X, and sums the character string length (len-i). Ask for. • compute-items (X): X is an element of the set R4,
{..., {pos, length-i, fid-i}, ...}, i, k}. compute-items (X) calculates the number of elements of a set {..., {pos, length-i, fid-i},.

Step 300) First, the sentence type determination unit 3 refers to the expression character string table of FIG. 8A to obtain an expression character string that is a partial character string of the target Japanese sentence. FIG. 10 shows the details of the processing. Step 400) A possible combination pattern in the target sentence of the expression character string extracted in step 300 is obtained. FIG. 11 shows the details of the processing.

Step 500) The combination pattern and the sentence type determination table (FIG. 8B) are collated to determine the sentence type. FIG. 12 shows the details of the processing. Step 600) If matching with a plurality of entries in the sentence type determination table is successful, one entry is selected. FIG. 13 shows details of the processing.

FIG. 10 is a flow chart for explaining the first operation of the sentence type judging section according to one embodiment of the present invention. FIG. 10 shows the detailed operation of the above-described step 300. First
Operation finds an expression character string that is a partial character string of the target Japanese sentence. Step 301) i is set to 1, N is set to the number of entries in the expression character string list, and R1 is an empty set ｛｝.

Step 302) The start position (pos) of the partial character string is pos = search (S, L (i) .string)}
If 0, the process proceeds to step 303, where pos = serach (
S, L (i). If (string) = 0, the process proceeds to step 304. Step 303) R1 = R1 + ｛pos, length (L
(i) .string), L (i). fid｝;

Step 304) i is incremented (i = i + 1). Step 305) If i> N, proceed to step 306; if i ≦ N, proceed to step 302. Step 306) If R is an empty set (｛｝), the process ends because the target Japanese sentence does not have the specified sentence type. If R is not an empty set, the process proceeds to step 302.

FIG. 11 is a flow chart for explaining the second operation of the sentence type judging section according to one embodiment of the present invention. The second operation is to determine a possible combination pattern of the extracted expression string in the target sentence. Step 401) N = the number of elements of R1, and R2 = comb
(1, R1) +... + Comb (N-1, R1) + R1.

Step 402) Set i-1 and M = R2
And R3 = ｛｝. Step 403) It is checked whether or not overlap occurs in each of the partial character strings stored in R2 (i). If no overlap occurs (no), R3 = R3 + R2
(I).

Step 404) i is incremented (i = i + 1). Step 405) If i> L, proceed to step 406; if i ≦ L, proceed to step 402. Step 406) If R3 = ｛｝, the target Japanese sentence does not have the specified sentence type, and the process ends. If R3 is not an empty set, the process proceeds to step 403.

FIG. 12 is a flow chart for explaining the third operation of the sentence type judging section according to one embodiment of the present invention. The third operation is to compare the combination pattern with the sentence type determination table to determine the sentence type. Step 501) Set i = 1, set the number of entries in the sentence type determination table to L, and set R4 to an empty set (｛｝).

Step 502) If the sentence type T (i) .type in the entry does not match the specified sentence type, the process ends. Step 503) Set j = 1, and set M to the number of entries in the function ID list of the i-th entry T (i) in the sentence type determination table (FIG. 8B).

Step 504) Set k = 1 and set N to R3
Set the number of elements of Step 505) The pointer check-list (T (i) .pointer (j), stored in the function ID list of the function ID list in a certain entry of the statement type determination table.
R3 (k)), the function ID stored in the element cell
Are all the same, R4 = R4 + ｛R3 (k),
i, j}.

Step 506) Increment k (k = k + 1). Step 507) If k> N, proceed to step 508; if k ≦ N, proceed to step 505. Step 508) Increment j (j = j +
1).

Step 509) If j> M, proceed to step 510; if j ≦ M, proceed to step 505. Step 510) Increment i (i = i +
1). Step 511) If i> L, proceed to step 512; if i ≦ L, proceed to step 505.

Step 512) R4 is an empty set (｛｝)
If so, the processing is terminated assuming that the target Japanese sentence does not have the specified sentence type. Step 513) If the number of elements of R4 is 1, a set of the sentence type and the sentence subtype of the element is returned, and the process ends. If the number of elements of R4 is greater than 1, step 5
Move to 04.

FIG. 13 is a flow chart for explaining the fourth operation of the sentence type judgment unit according to one embodiment of the present invention. The fourth operation is to select one entry when matching with a plurality of entries in the sentence type determination table is successful. Step 601) Select = 1, length = compute-
length = R4 (1), items = compute-items
(R4 (1)).

Step 602) When i = 2, N is R4
Set the number of entries. Step 603) length <compute-items (R4
If (i)), the process proceeds to step 604, where length ≧
If it is compute-items (R4 (i)), step 60
Move to 6.

Step 604) items> compute-item
If s (R4 (i)), the process proceeds to step 605,
If items ≦ compute-items (R4 (i)), the process proceeds to step 606. Step 605) select = i, length = compute-leng
th (R4 (i)), items = compute-items (R4
(I)). Step 606) Increment i (i = i +
1).

Step 607) If i> N, proceed to step 608. If i ≦ N, step 603
Move to Step 608) Return the set of the sentence type and the sentence subtype of the element of N and end the process.

The outline of the operation of the sentence type determination unit 3 for the Japanese question message shown in FIG. 5 will be described with reference to the flowcharts of FIGS. (1) First sentence, "I will ask you a question for the first time. ] Does not match any of the character strings stored in the expression character string table of FIG. Therefore, it is not determined as an information request statement.

(2) Next, the second sentence “I want to connect a 17-inch monitor to VA066”. ] Matches and matches only the expression of entry ID = 13 stored in the expression character string table of FIG. The entry function ID is 800, and according to step 500, this is not determined to be a combination of expression character string patterns that define the sentence type of "information request".
Therefore, it is not determined as an information request statement.

(3) Next, the third sentence “I am looking for a connection cable necessary for that,
With 0, it is checked against only the expression character string of entry ID = 3 stored in the expression character string table of FIG. This entry has a function ID of 100 and a step 500
Thus, this is determined to be a combination of expression character string patterns that define the sentence type of “information request”. At the same time, the fact that the sentence subtype is "target article presentation" is obtained as a side effect.

(4) Fourth sentence "Do you know which store you are selling? Is Step 3
00, the three entries stored in the expression string table of FIG. 8B are compared with the expression strings of entry ID = 5, entry ID = 7, and entry ID = 11. , Respectively, 200, 300,
400. Set R1 resulting from step 300
R1 = {1,11,200}, {12,7,30
0}, {19, 10, 400}. Also, the set R which is the processing result of step 400
3, R3 = {1, 11, 200}, {12, 7, 3}
00, 19, 10, 400, 1, 1
1,200｝, {12,7,300}, ｛｛1,1
1,200｝, ｛19,10,400 ｛｛, ｛｛12,
7,300, {19,10,400}, {1,1
1,200｝, {12,7,300}, {19,10,
400｝｝｝. Further, when this R3 is compared with the sentence type determination table in step 500, as a result, the set R4 becomes R4 = {12, 7, 300}, {19, 10, 4}
00, 1, 2, {1, 11, 200}, {1
2,7,300}, {19,10,400}, 1,
1｝｝ is obtained. This indicates that there are two combinations of partial character strings giving the sentence type of "information request". Therefore, the process proceeds to step 600. In step 600, the length of the character string covered by the matched expression string is counted and compared. Here, the second of R4
Character string length covered by the combination given by the element (= 2
8) is longer than the character string length (= 17) covered by the combination provided by the first element, so the combination provided by the second element is selected. Therefore, sentence type = "information request"
Is confirmed, and the sentence subtype = “obtain article” is obtained as a side effect.

(5) Finally, the fifth sentence, "Thank you. Does not match any of the character strings stored in the expression character string list in FIG. 8B, and therefore is not determined to be an information request sentence (step 300) as in the case of the first sentence. As a result of the above processing, the sentence type (“information request”) and the sentence subtype (“acquisition of goods”) are requested for the third sentence “I am looking for the necessary connection cable.” The information is stored in the third and fourth fields of the corresponding entry of the processing table 13. This situation is shown in FIG.

The sentence of the Japanese question message whose sentence type is determined to be “information request” by the sentence type determining unit 3 is a Japanese basic analyzing unit that performs morphological and syntactic analysis on the input Japanese sentence. 5 is transferred. iii) Basic Japanese Analysis Process FIG. 15 shows an outline of the result of the basic Japanese analysis of the third sentence in the Japanese question message shown in FIG. As shown in the figure, the result of the basic analysis is expressed as a syntax tree structure in which a clause is a node, a dependency relation between the clauses is a branch, a label to a branch of an attribute of the dependency relation, and a final clause is a root, The address of the root node (pointer to the root node) to the fifth field of the processing table 13 is stored. Here, it is assumed that each node holds the result of the morphological analysis for each phrase. Note that the processing of morphological analysis and syntax analysis is provided in a normal Japanese processing system.
The present invention is not limited by the present invention.

The sentence type determination unit 3 determines that the sentence type is “information request”, and the sentence of the Japanese question message analyzed by the Japanese basic analysis unit 5 is:
It is transferred to the semantic structure analysis unit 7. The semantic structure analysis unit 7
A prepared semantic structure pattern dictionary 6 that stores unique semantic structure patterns in Japanese question messages.
Is compared with the analysis result of the Japanese basic analysis unit 5 to determine the semantic structure of the information request sentence.

FIG. 16 shows an example of entries in the semantic structure pattern dictionary 6 according to one embodiment of the present invention. The left side of the arrow indicates a syntax tree structure to be integrated, and the right side of the arrow indicates a syntax tree structure to be converted when the collation is successful. As shown in the figure, the dictionary entry is expressed as a partial syntax tree structure. In the same figure, "@HEA
"D", "$ X", etc. are variables for holding the matching subtree structure, and "TYPE ==
Expressions such as “information request” indicate restrictions on each node itself.
For example, the expression has a corresponding sentence type of “information request”.
Expresses the constraint that it must be Similarly, “SubType ==“ presentation of target article ”expresses a constraint on a sentence subtype.

The semantic structure analysis unit 7 performs the matching process so that as much of the input syntax tree structure as possible is covered by the partial syntax tree structure entered in the semantic structure pattern dictionary 6, and the matching is successful. In this case, the data is converted into a designated semantic structure expressed by a set of a syntax tree structure and pairs of attributes and attribute values. The processing result of the Japanese basic analysis unit 5 shown in FIG.
17A and 17B, the entry is converted into a semantic structure as shown in FIG. In FIG. 17, ": n
Indicators indicating semantic attribute names such as "eded-for" and ": item-for" are given by corresponding entries in the semantic structure pattern dictionary 6.

The semantic structure which is the result of the processing by the semantic structure analyzing unit 7 is held by storing a pointer to the storage area in the fifth field of the processing table 13.
This result is further transferred to the reference expression elimination control unit 9 that controls the reference expression elimination process using the reference expression elimination control table 8 and the sentence type determination unit 3.

Iv) Reference expression elimination processing The reference expression elimination control unit 9 determines the reference destination based on the search direction information obtained from the reference expression elimination control table 8 based on the semantic structure of the input information request sentence. Is selected. At this time, the sentence type determining unit 3 is activated using the sentence type information obtained from the reference expression elimination control table 8 to determine the eligibility of the candidate sentence.

FIG. 18 is a flowchart showing the operation of the reference expression cancellation control unit according to one embodiment of the present invention. Step 701) The reference expression elimination control unit 9 scans the semantic structure of the information 9 to be processed and extracts the reference expression included in the linguistic expression analysis part represented as the attribute value.
For example, a reference expression based on a context instruction such as “that” or “that” is extracted.

Step 702) Initially, count i = 1, N is the number of reference expressions extracted, and L is the position of the statement to be processed from the beginning of the message. Step 703) Next, as the semantic attribute key in the reference expression and the semantic structure, the search direction for searching the reference destination of the reference expression, the sentence type of the Japanese sentence to be searched, and the linguistic expression unit of the reference target idiom. The stored reference expression elimination table 8 is retrieved and the control information is obtained.

Step 704) Hereinafter, step 703
Based on the result of the above, the following processing is performed in order from the sentence closest to the sentence including the reference expression to be processed along the designated search direction one by one until the reference destination is obtained. First, if the search direction D is the same sentence, the process proceeds to step 705; otherwise, the process proceeds to step 708.

Step 705) Reference expression cancellation control section 9
Starts the sentence type determination unit 3 and, when the search direction D is the same sentence, starts the reference expression resolution unit 10 and actually obtains the reference destination of the reference expression. Step 706) If the processing is successful, go to Step 7
07, and if unsuccessful, the flow proceeds to step 719.

Step 707) Replace the part of the reference expression to be processed in the semantic structure with the semantic structure for the reference destination, and proceed to step 719. Step 708) The count j is set to 1. Step 709) If the search direction D is forward, the reference expression elimination control unit 9 proceeds to Step 710; otherwise, proceeds to Step 711.

Step 710) The (Lj) -th sentence is set as a candidate sentence, and the flow advances to step 712. Step 711) When the search direction D is other than “forward”, the reference expression elimination control unit 9 sets the (L + j) th sentence as a candidate sentence.

Step 712) The sentence type judgment section 3 is started. Step 713) The sentence type determination unit 3 determines whether the candidate sentence has the specified sentence type. If so, the process proceeds to step 714; otherwise, the process proceeds to step 717. Move to

Step 714) If the above judgment is successful, the reference expression elimination control unit 9 activates the reference expression elimination unit 10 and actually obtains the reference destination of the reference expression. It is assumed that this processing itself is realized by an existing technology such as checking the identity of the expression notation and the semantic identity using a thesaurus, and details thereof are directly related to the gist of the present invention. Since there is no description, the description is omitted.

Step 715) If the reference destination of the reference expression has been obtained, the flow shifts to step 716; otherwise, the flow shifts to step 717. Step 716) Replace the part of the reference expression with the semantic structure for the reference destination, and proceed to Step 720.

Step 717) The count j is incremented (j = j + 1). Step 718) When j <1 when the search direction D is “forward” or j <the number of sentences in the question message when the search direction D is “backward”, the process proceeds to step 719, otherwise. , To step 708.

Step 719) The count i is incremented (i = i + 1). Step 720) If the count i is larger than the extracted number N of reference expressions, the process ends. If the count i is less than the number of reference expressions, the process proceeds to Step 703.

In the operation of the above-described flowchart, a basic analysis of a candidate sentence including a reference destination is generally required to resolve a reference expression. In the present invention, it is assumed that the reference expression elimination unit 10 activates the Japanese basic analysis unit as needed. Hereinafter, the operation of the reference expression cancellation control unit 9 with respect to the example of FIG. 17 based on the operation logic of FIGS. 18 and 19 will be described.

Here, an example of the configuration of the reference expression elimination control table 8 sufficient to explain the example of FIG. 17 will be described. FIG.
9 shows an example of a reference expression elimination control table according to an embodiment of the present invention. The reference expression elimination control table 8 has a table structure in which each entry has five fields. First
The field is a reference expression, the second field is the corresponding semantic attribute, the third field is the search direction of the candidate sentence, the fourth field is the sentence type of the candidate sentence, and the fifth field is the reference destination. The linguistic expression unit of the expression is stored.
First, in step 701, the <value> portion, which is the attribute value for the semantic attribute ": needed-for", is extracted as a unique reference expression. This determination is made by identifying the part of speech as a demonstrative from the result of the morphological analysis for "that".

Next, the reference expression elimination control table 8 is searched in step 703 to obtain a search direction D = “forward”, a sentence type T = “goal statement”, and a goal statement S = “proposition part of goal statement”. Can be Next, in step 709, the second (= 3-1) sentence becomes a candidate sentence, and the sentence type determining unit 3 determines whether the sentence has a sentence type of "statement of goal". Here, the description of the operation of the sentence type determination unit 3 is omitted, but when the operation logic shown in FIGS. 9 to 13 is performed using the sentence type determination dictionary 2 shown in FIG. 8, the sentence type becomes “goal statement”. Is determined. Accordingly, the process proceeds to step 705, and the reference destination is the second sentence of the message, "I want to connect a 17-inch monitor to VA066." Connect to VA066>. As described above, the reference expression elimination control unit 9 for the present example ends the processing. As described above, the semantic structure of the third sentence in the Japanese question message in FIG. 5 is as shown in FIG.
The head address of the storage area of the structure is stored in the fifth field of the processing table 13.

In the fourth sentence of the Japanese question message shown in FIG. 5, the target sentence "connection cable" where the user inquires "where the store is sold" is omitted. . Such linguistic phenomena are called zero pronouns, and are a kind of reference expression frequently appearing in Japanese discourse structure. The resolution of the reference destination is also performed by the reference expression control unit 9.
It is performed by Since it is assumed that the identification process for the zero pronoun is also realized by the existing technology, the details of the operation of the reference expression resolving unit 10 regarding the resolution of the reference destination of the zero pronoun are not specified here. It is assumed that <cable> is identified as a reference. The semantic structure of the fourth sentence by the processing so far is as shown in FIG.

When only one information request sentence is extracted from the input Japanese question message by the above processing and its semantic structure is obtained, it is output from the output unit 12 as an information request in the message. v) Information request integration processing On the other hand, when a plurality of information request sentences are extracted and their semantic structures are obtained, the information request structuring unit 11 determines the identity of the information request object in each information request sentence. However, if the identity can be confirmed, the information request in the Japanese message to be processed is structured as a single semantic structure by integrating the semantic structures of a plurality of information request sentences. This integration processing is performed by natural language processing and unification processing of feature configurations widely known in the field of research and development of artificial intelligence. According to the above processing procedure, the information request extracted and structured from the target Japanese question message is output via the output unit 12 to a device connected to the subsequent stage.

FIG. 22 shows an outline of the operation logic of the information request structuring unit 11. The operation of the information request structuring unit 11 in a situation where the meaning structures of the two information request sentences shown in FIGS. 21A and 21B are obtained from the Japanese question message shown in FIG. 5 will be described. Step 801) The information request structuring unit 11 checks the identity of an item having an attribute value for the semantic attribute “: head”. If the identity is not established, proceed to step 802,
If the condition holds, the process proceeds to step 803.

Step 802) If the identity is not established, it is determined that the integration is impossible and the process is terminated. Step 803) Compare the attribute values of the semantic attribute sentence subtype in each semantic structure, and replace the attribute value of the semantic attribute sentence subtype in each semantic structure with the attribute value having the most detailed information.

Step 804) Perform unification processing between the semantic structures for each information request statement (if they have the same semantic attribute, check the identity of each attribute value. All the identity of these is confirmed. If possible, take the sum of each semantic structure).

Step 805) If the unification succeeds, the process proceeds to step 806. If the unification fails, the process proceeds to step 807. Step 806) If the unification succeeds, the result is used as an information request structure, and the process ends.

Step 807) If the unification fails, each semantic structure is set to an independent information request structure, and the processing is terminated. Next, a detailed description will be given according to the flowchart of FIG. First, in step 801, the identity of the partial structure having the indicator “: head” is checked. In the input examples of FIG. 16 and FIG. 17, both of those having the sign “: head” match with the noun “cable”, so that this is successful. Even if these words are not the same, "floppy" and "FD"
When words with the same meaning but different notations are used, such as "secondary storage" and "hard disk", they may be regarded as the same word in context context. In the present invention, it is assumed that such a check of semantic identity is achieved by an existing notation fluctuation absorption technique or a semantic check using a thesaurus, and details thereof are not specified.

Next, in step 803, an attribute value to be given to the sentence subtype attribute of the semantic structure obtained by integration is determined. This attribute value is assumed to represent the most detailed information in the “sentence subtype” attribute in each semantic structure to be integrated. This attribute is shown in FIG.
FIG. 2A shows “SEEK-OBJECT” and FIG.
In 1 (B), it is “HOW-TO-GET”. Therefore, the latter is selected as expressing more detailed information. In the actual processing, the attribute value of the “sentence subtype” attribute in the structure of FIG. 21A is rewritten from “SEEK-OBJECT” to “HOW-TO-GET”.

Next, the control of the process proceeds to step 804. In step 804, a structure integration process called unification of a feature structure, which is widely known in the field of artificial intelligence and natural language processing, is performed. This process
This is a process of checking the consistency of an attribute-attribute value pair included in each target structure, and finding an integrated structure having the minimum information that encompasses the information of each structure. For the example being described, the structure shown in FIG. 23 is generated as the final information request structure and transferred to the output unit 12. In this example, FIG.
Since there is no information other than the “: head” indicator in the structure shown in FIG. 1 (B), the structure in which only the “sentence subtype” attribute in the structure of FIG. 21 (A) is rewritten is finally integrated as it is. It becomes an information request structure. This result is transferred to a device connected to the subsequent stage via the output unit 12.

In contrast to such an example, FIG. 24A shows an example in which, in a structure to be integrated, a sum of information in both structures of an integration source is obtained, and a partial structure having the same marker is obtained. FIG. 24B shows an example in which integration is not possible due to lack of identity. In this example, both structures have a "needed-for" sign, but they cannot be combined because their terms describe different situations. In such a case, two information request structures are output via the output unit 12.

The Japanese-language question message analyzer of the present invention can be applied as follows. For example, a question message posted on an electronic bulletin board in a computer network is analyzed at a certain timing using the Japanese question message analyzer of the present invention, and information requests in those are extracted, thereby conforming to the extracted information request. It is possible to search for information to be performed from a database of question / answer messages stored on a computer network.

Further, when the Japanese question message analysis device of the present invention is applied to service provision by telephone,
Since the data input to the input unit 1 is text data, by providing a function of converting the result of voice input voice data recognized by voice recognition processing into text data at a stage prior to the input unit 1, voice input can be performed. Can also retrieve information. At this time, the operator on the center side can visually recognize the search result on the display unit and reply to the user by telephone voice. Further, it is also possible to convert the result of searching the database in response to the information request output from the output unit 12 into voice and return it to the user.

The present invention is not limited to the above embodiments, but can be variously modified and applied within the scope of the claims.

[0104]

As described above, according to the present invention, it is possible to structure an information request described in a natural Japanese message in which instructions of complicated and various information requests to the information retrieval system are described. Therefore, the Japanese message structuring function according to the present invention can be a component for realizing an information search system that is more flexible than the conventional one that expresses an information request in a logical combination format of keywords by being incorporated into the information search system. .

In the apparatus for structuring a Japanese question message according to the present invention, a basic language analysis process requiring grammatical rules such as morphological analysis and syntax analysis, and a reference expression resolution process for identifying a reference destination of a reference expression Is applied only to the necessary part selected based on the identification of the question message from the whole message to be processed, and (1) the processing is fast; (2) a normal message to process the whole message There is an advantage that higher analysis accuracy is achieved as compared with the analysis method. Further, since the reference expression elimination process is controlled based on the semantic structure analyzed based on the characteristics of the question message, it is possible to expect high accuracy in reference expression elimination.

Further, the present invention can be variously applied to various message processing systems and the like.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a configuration diagram of a Japanese question message analysis device according to an embodiment of the present invention.

FIG. 4 is a flowchart showing an outline of the operation of the Japanese-language question message analysis device according to one embodiment of the present invention.

FIG. 5 is a diagram illustrating an example of a Japanese question message input according to an embodiment of the present invention;

FIG. 6 is a diagram showing an example of disassembling the message of FIG. 5 according to one embodiment of the present invention.

FIG. 7 is a flowchart showing a detailed operation of an information request sentence extracting unit according to one embodiment of the present invention.

FIG. 8 is a diagram illustrating an example of a sentence type determination dictionary according to an embodiment of the present invention.

FIG. 9 is a flowchart illustrating an outline of an operation of a sentence type determination unit according to an embodiment of the present invention.

FIG. 10 is a flowchart illustrating a first operation of a sentence type determination unit according to one embodiment of the present invention.

FIG. 11 is a flowchart illustrating a second operation of the sentence type determination unit according to one embodiment of the present invention.

FIG. 12 is a flowchart illustrating a third operation of the sentence type determination unit according to one embodiment of the present invention.

FIG. 13 is a flowchart illustrating a fourth operation of the sentence type determination unit according to one embodiment of the present invention.

FIG. 14 is a diagram illustrating an example of a processing table after the message determination in FIG. 5 according to an embodiment of the present invention;

FIG. 15 is a diagram illustrating an outline of a result of a basic Japanese analysis of a Japanese question message according to an embodiment of the present invention.

FIG. 16 is a diagram illustrating an example of an entry of a semantic structure pattern dictionary according to an embodiment of the present invention.

FIG. 17 is a diagram illustrating an example of conversion to a semantic structure according to an embodiment of the present invention.

FIG. 18 is a flowchart illustrating an operation of a reference expression cancellation control unit according to an embodiment of the present invention.

FIG. 19 is a flowchart illustrating an operation of a reference expression cancellation control unit according to an embodiment of the present invention.

FIG. 20 is a diagram illustrating a reference expression elimination control table according to an embodiment of the present invention.

FIG. 21 is a diagram showing a semantic structure of an embodiment of the present invention.

FIG. 22 is a flowchart showing the operation of the information request structuring unit according to one embodiment of the present invention.

FIG. 23 is a diagram illustrating an example of an integration process according to an embodiment of the present invention.

FIG. 24 is a diagram showing an example where integration is possible / an example where integration is not possible according to an embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input part, input means 2 Sentence type judgment dictionary 3 Sentence type judgment part, sentence type judgment means 4 Information request sentence extraction part, Information request sentence extraction means 5 Japanese basic analysis part, Japanese basic analysis means 6 Semantic structure pattern dictionary 7 semantic structure analysis unit, semantic structure analysis unit 8 reference expression elimination control table 9 reference expression elimination control unit, reference expression elimination control unit 10 reference expression elimination unit, reference expression elimination unit 11 information request structuring unit, information request structuring unit 12 output unit and output means 13 processing table 41 statement type activation means

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-63-106042 (JP, A) JP-A-2-297157 (JP, A) JP-A-5-233730 (JP, A) Tatsuo Ota, Question and answer System construction, FUJITSU, Japan, Fujitsu Limited, September 10, 1994, Vol. 45, No. 5, 436 to 442 (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 17/30 JICST file (JOIS)

Claims (8)

(57) [Claims] 1. A method of analyzing a Japanese question message input in a text format, the method comprising: analyzing a Japanese question message input in a text format; Paying attention to a specific expression character string, an information request sentence that describes a core part of the information request is extracted. When the extracted information request sentence includes a reference expression, the information request sentence The reference expression is resolved based on the characteristics of the information request statement, and when the plurality of extracted information request statements state an information request for the same object, the structures of the plurality of information request statements are integrated. A method of analyzing a Japanese question message, comprising structuring a main information request in the Japanese question message and outputting a structured information request sentence. 2. When extracting the information request sentence, an information request which is a functional role played by a Japanese sentence constituting the Japanese question message, a sentence type of a goal statement which is a user's purpose, and the information Referencing a sentence type determination dictionary in which a combination of a request and an expression string pattern for determining the goal statement is stored in advance, and collating the character string to determine whether the Japanese question message corresponds to a specified sentence type. The method according to claim 1, wherein a sentence having an information request function is extracted from the Japanese question message. 3. When the information request sentence is extracted, morpheme / syntax analysis is performed on the extracted information request sentence, and a unique semantic structure pattern is previously stored for each entry of the Japanese question message. 2. The Japanese question message analysis according to claim 1, wherein the entry is compared with an analysis result of the information request sentence that has been subjected to morphological / syntactic analysis with reference to the semantic structure pattern dictionary. Method. 4. When resolving the reference expression, a pattern of a reference expression that can be included in the information request sentence in the Japanese question message, a search direction for searching for a reference destination of the reference expression, and a search target. The reference sentence type of the Japanese sentence and the linguistic expression unit of the reference expression are referenced in advance to determine whether or not the sentence type of the Japanese sentence to be searched corresponds to the specified sentence type with reference to the reference expression resolution control table. 2. The Japanese question message analysis method according to claim 1, wherein, in a corresponding case, the reference expression resolution processing is controlled. 5. A Japanese-language question message analyzing apparatus for analyzing a Japanese-language question message input in a text format, comprising: input means for inputting the Japanese-language question message; An information request sentence extracting unit that extracts an information request sentence that describes a core part of an information request by focusing on a specific expression character string of the message; and the information request sentence extracted by the information request sentence extraction unit. When a sentence includes a reference expression, a reference expression resolving means for resolving the reference expression on the basis of the characteristics of the information request sentence, and a plurality of extracted information request sentences describing an information request for the same object. Information request structuring means for unifying the structure of the plurality of information request sentences to structure a main information request in the Japanese question message; Japanese inquiry message analyzing apparatus characterized by an output means for outputting the information request statements that have been integrated by Zoka means. 6. The information request sentence extracting means includes: an information request which is a functional role played by a Japanese sentence constituting the Japanese question message and a sentence type of a goal statement which is a user's purpose; A sentence type determination dictionary in which a combination of expression character string patterns for determining a goal statement is stored in advance, and referring to the sentence type determination dictionary, whether the Japanese question message corresponds to a specified sentence type And a sentence type determination activating unit that activates the sentence type determination unit and extracts only an information request sentence having an information request function from the Japanese question message. The Japanese question message analyzer according to claim 5. 7. A Japanese basic analysis means for performing morpheme / syntax analysis on the information request sentence extracted by the information request sentence extraction means, and a unique semantic structure pattern for each of the Japanese question messages for each entry. And referring to the semantic structure pattern dictionary stored in advance, and comparing the entry with the analysis result of the information request sentence analyzed by the Japanese basic analysis means, 6. The Japanese question message analysis device according to claim 5, further comprising a semantic structure analyzing means for obtaining the semantic structure of the Japanese question message. 8. The reference expression resolving means includes: a reference expression pattern that can be included in the information request sentence in the Japanese question message; a search direction for searching for a reference destination of the reference expression; Using the reference expression elimination control table pre-stored the sentence type of the word sentence and the linguistic expression unit of the reference expression, the reference expression elimination control table, and the sentence type determination means, identify the characteristics of the object of the information request. 5. The Japanese-language question message analysis device according to claim 4, further comprising: reference-expression-resolving control means for controlling a reference-expression resolving process.