JP2001022779A - Interactive information retrieval device, method for interactive information retrieval using computer, and computer-readable medium where program performing interactive information retrieval is recorded - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform information retrieval in a natural interactive style. SOLUTION: An indication sentence analyzing means 2 analyzes an indication sentence inputted in natural language form to extract a retrieval condition and a retrieval executing means 4 retrieves corresponding information on the basis of the extracted retrieval condition. Then an interaction control means 3 holds and manages as context structure data a retrieval process wherein the retrieval condition and the retrieval result based upon it are arranged in time series and controls a flow of interaction between a user and a device according to the contents of the syntax structure data, and an answer sentence generating means 6 generates an answer sentence in natural language form according to the retrieval result. Then the interaction control means 3 when clearly judging that the top of the interaction is traced back to the past automatically puts the syntax back to the state even if the user makes no clear indication and performs information retrieval while advancing natural interaction like interaction between persons between the device and user.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an interactive information search apparatus, an interactive information search method using a computer, and an interactive information search process for searching for user-desired information that satisfies the conditions by giving search conditions. The present invention relates to a computer-readable medium storing a program to be executed.

[0002]

2. Description of the Related Art Conventionally, as an interactive information search apparatus for performing an interactive search operation, for example, Japanese Patent Application Laid-Open No. 7-78171 is known.
Japanese Unexamined Patent Publication (Kokai) No. H10-264, pp. 157-334 is known. In this conventional information retrieval apparatus, the retrieval process is recorded and managed in the form of the history of the dialogue sentence input / output in natural language and the history of the retrieval target set created by the retrieval, and is stored in the user utterance before. When there is a reference to the intermediate result of the search, the search is resumed in a state of going back to the search target set at that time, thereby realizing an efficient search operation. For example, in the dialogue example as shown in FIG. 12, the search target set is changed to ｛X as shown in the network structure of FIG.
X, YY, ZZ} is narrowed down to {AA}. Here, when the user utters Q6 “What is YY town?”, The system outputs “YY” to the previous output sentence A1. Will be found, and the search will be resumed from there. Therefore, there is no need to repeat the search process until “YY” is narrowed down, and the information can be extracted simply by referring to “YY”.

[0003]

However, the conventional interactive information retrieval apparatus has the following problems. In other words, since the user performs backtracking back to the previous state by matching words and phrases that appear in both the new utterance input and the previous input / output sentence, the user clearly indicates the words that appear in the input / output sentence. If it cannot be mentioned, it is difficult to backtrack to such a state by simple dialogue.

[0004] For example, in the dialogue example shown in FIG.
In the search up to A3, the search target was narrowed down to 33,
In the next question, "China no Ha", the number of family restaurants in China is narrowed down from the previous 33 to two. This is shown in FIG. Here, it is assumed that the user understands two Chinese restaurants out of the 33 for the time being, but wants to check an Italian restaurant.
At this time, if it is a natural dialogue between humans, after hearing Q4, if you hear "Italian hana" in Q5, Q4 and Q
Since the content of 5 is an independent exclusion as a search condition, the partner should reject Q4 and the search result, return to the state of A3 immediately before, and answer the result of narrowing down the Italian restaurant from 33 cases. .

[0005] However, in the conventional device, in order to establish such a dialog, first, in response to A3, the system sets 33 family restaurants which are search results to "XX store, ..., XX store, There are 33 cases. "In addition, if the user does not say" It is an Italian restaurant among XX stores, ..., XX stores "in Q5, the system accepts the request. Can not find the position of the desired backtrack by matching the words. For this reason, there is a problem that it is difficult to backtrack with a simple dialogue.

SUMMARY OF THE INVENTION The present invention has been made in view of such a conventional problem, and the apparatus automatically controls the flow of a dialog according to a predetermined criterion, thereby enabling backtracking with a simple dialog. It is another object of the present invention to provide an interactive information search technique capable of searching information in a more natural interactive manner.

[0007]

According to the present invention, there is provided an interactive information retrieval apparatus which analyzes a directive input in a natural language format and extracts a retrieval condition; Search execution means for searching for the corresponding information based on the search conditions extracted by the means, and a search process in which the search conditions and the search results executed by the search execution means based on the search conditions are arranged in time series as context structure data. Retain, manage, and manage users according to the content of the context structure data.
A dialogue control unit that controls a flow of a dialogue between the devices; and a response sentence generation unit that generates a response sentence in a natural language format according to the search result. By repeatedly executing a series of processing,
In a device that searches for information necessary for control, operation or management of onboard equipment in an interactive manner between a user and a system,
The dialogue control means logically disagrees with one of the search conditions already recorded in the context structure data with respect to the newly input descriptive sentence by a search condition extracted by the descriptive sentence analysis means. In some cases, search results and search conditions that are later in time series than the search conditions in the context structure data that are the disjunction are deleted from the context structure data, and the search conditions in the context structure data that are the disjunction are deleted. Is rewritten to the newly extracted search condition, whereby the dialogue is retroactively advanced to the corresponding point in the past.

According to the first aspect of the present invention, a directive sentence input in a natural language format is analyzed by a directive sentence analyzing unit to extract a search condition, and a search is executed based on the extracted search condition. Means for retrieving the corresponding information, and the dialogue control means retains and manages, as context structure data, a search process in which the search conditions and the search results based on the search conditions are arranged in chronological order. The flow of dialogue between the devices is controlled, and a response sentence in a natural language format is generated by the response sentence generation means according to the search result.

Then, the dialogue control means analyzes the newly input instruction sentence and extracts the search condition, when the search condition is logically exclusive of one of the search conditions already recorded in the context structure data. Deletes search results and search conditions that are later in time series than the search conditions in the context structure data that are disjoint from the context structure data, and newly extracts search conditions in the context structure data that are disjoint. By rewriting the search conditions, the dialogue is controlled to proceed to a new dialogue retroactively up to a corresponding point in the past.

[0010] With this, when it is determined that the topic goes back to the previous state in the course of the dialogue, the context is automatically returned to the state even if the user does not explicitly give an instruction. Information retrieval can be performed while a natural conversation like a human being proceeds between the device and the user.

According to a second aspect of the present invention, there is provided an interactive information retrieving apparatus.
A command sentence analyzing unit that analyzes a command sentence input in a natural language format to extract a search condition; a search execution unit that searches for corresponding information based on the search condition extracted by the command sentence analysis unit; A search process in which search conditions and search results executed by the search execution unit based on the search conditions are arranged in chronological order is stored and managed as context structure data, and the flow of interaction between the user and the device is controlled according to the contents of the context structure data. And a response sentence generating means for generating a response sentence in a natural language format according to the search result, and repeatedly executes a series of processes from analysis of the instruction sentence to generation of the response sentence. Thus, in an apparatus for searching for information necessary for control, operation, or management of on-vehicle equipment in an interactive manner between a user and a system, the interactive control means may be newly input. The search condition extracted by the instruction sentence analyzing means for the given sentence does not satisfy the index of the condition limitability predetermined in advance or according to the situation, and the search execution means performs the search based on the search condition. If it is determined that the search target cannot be narrowed down to a predetermined range even after execution, a response sentence is generated to prompt the user to further limit the search conditions without moving the process to the search execution means. Thus, the response sentence generation means is instructed.

In the interactive information retrieval apparatus according to a second aspect of the present invention, a directive sentence input in a natural language format is analyzed by a directive sentence analyzing means to extract a search condition, and a search is executed based on the extracted search condition. Means for retrieving the corresponding information, and the dialogue control means retains and manages, as context structure data, a search process in which the search conditions and the search results based on the search conditions are arranged in chronological order. The flow of dialogue between the devices is controlled, and a response sentence in a natural language format is generated by the response sentence generation means in accordance with the search result.

[0013] The dialogue control means determines that the search condition extracted by the instructional sentence analyzing means for the newly input instructional statement is:
When it is determined that the index does not satisfy the condition limitation index predetermined in advance or according to the situation, and the search target cannot be narrowed down to a predetermined range even if the search execution unit executes the search based on the search condition. Instructs the response sentence generation means to generate a response sentence urging the user to further limit the search conditions without transferring the processing to the search execution means.

[0014] Thereby, the search process can be executed after estimating the prospect that the search target can be narrowed down to some extent.
It is possible to improve the efficiency of information search execution by eliminating unnecessary processing.

According to a third aspect of the present invention, there is provided an interactive information retrieval apparatus.
A command sentence analyzing unit that analyzes a command sentence input in a natural language format to extract a search condition; a search execution unit that searches for corresponding information based on the search condition extracted by the command sentence analysis unit; A search process in which search conditions and search results executed by the search execution unit based on the search conditions are arranged in chronological order is stored and managed as context structure data, and the flow of interaction between the user and the device is controlled according to the contents of the context structure data. And a response sentence generating means for generating a response sentence in a natural language format according to the search result, and repeatedly executes a series of processes from analysis of the instruction sentence to generation of the response sentence. Accordingly, in an apparatus for searching for information necessary for control, operation, or management of on-vehicle equipment in a dialogue form between a user and a system, the dialogue control unit may execute the search. A response sentence that prompts the user to relax the search condition if no search object corresponding to the search result is obtained at the time of transferring the processing to the response sentence generating means according to the search result of the column Instruct generation to the response sentence generation means, and, by deleting the search result and the latest search condition that led to the definition from the context structure data, go back to the time immediately before the latest input of the descriptive sentence. This is to control the dialog to proceed anew.

In the interactive information retrieval apparatus according to the third aspect of the present invention, a directive sentence input in a natural language format is analyzed by a directive sentence analyzing means to extract a search condition, and a search is executed based on the extracted search condition. Means for retrieving the corresponding information, and the dialogue control means retains and manages, as context structure data, a search process in which the search conditions and the search results based on the search conditions are arranged in chronological order. The flow of dialogue between the devices is controlled, and a response sentence in a natural language format is generated by the response sentence generation means in accordance with the search result.

When the dialogue control means shifts the processing to the response sentence generation means in accordance with the search result of the search execution means, if no corresponding search target is obtained as the search result, the dialogue control means relaxes the search condition. By instructing the response sentence generation means to prompt the user to perform the search, and deleting the relevant search result and the latest search condition that led to the definition from the context structure data to thereby provide the latest instruction. Control is performed so that a new dialogue proceeds to the time immediately before the sentence input.

Thus, the user can be guided so that the subsequent search can be efficiently performed according to the search result, and the context can be held in an appropriate state in preparation for the next user's utterance. It is possible to search for information while a natural dialog such as a dialog between humans proceeds between the device and the user.

According to a fourth aspect of the present invention, there is provided an interactive information retrieval apparatus.
Claims 1 to 3 perform point search processing in the navigation system, and the point search processing in the navigation system can be executed by a natural conversation between the user and the device.

According to a fifth aspect of the present invention, there is provided an interactive information retrieval method using a computer, which analyzes a directive sentence input in a natural language format to extract a retrieval condition, and, based on the retrieved retrieval condition, retrieves corresponding information. , And retains and manages a search process in which the search conditions and search results based on the search conditions are arranged in time series as context structure data, and controls the flow of dialogue between the user and the computer according to the contents of the context structure data. By repeatedly executing a series of processes for generating a response sentence in a natural language according to the search result, information necessary for control, operation or management of the on-vehicle equipment is searched in an interactive manner between a user and a computer. In the method, the search condition extracted by analyzing the newly input directive is
If it is logically disjointed with one of the search conditions already recorded in the context structure data, the search result and the search condition that are later in time series than the search condition in the context structure data that is the disjunction From the context structure data, and rewriting the search condition in the context structure data, which is the disjunction, to the newly extracted search condition, thereby proceeding a new dialogue to a corresponding point in the past. It is.

According to a sixth aspect of the present invention, there is provided an interactive information search method using a computer, which analyzes a directive sentence input in a natural language format, extracts search conditions, and, based on the extracted search conditions, outputs corresponding information. , A search process in which the search conditions and search results based on the search conditions are arranged in time series is retained and managed as context structure data, and the flow of dialogue between the user and the computer is controlled in accordance with the contents of the context structure data. By repeatedly executing a series of processes for generating a response sentence in a natural language format according to the search result, information necessary for control, operation or management of the on-vehicle equipment is searched in an interactive manner between a user and a computer. In the method, the search condition extracted by analyzing the newly input directive is
When the index of the condition limitability determined in advance or according to the situation is not satisfied and the search target cannot be narrowed down to a predetermined range even if the search is performed based on the search condition, the search processing is performed. Instead, a response sentence that prompts the user to further limit the search conditions is generated.

According to a seventh aspect of the present invention, there is provided an interactive information search method using a computer, which analyzes a directive sentence input in a natural language format, extracts search conditions, and, based on the extracted search conditions, obtains corresponding information. , A search process in which the search conditions and search results based on the search conditions are arranged in time series is retained and managed as context structure data, and the flow of dialogue between the user and the computer is controlled in accordance with the contents of the context structure data. By repeatedly executing a series of processes for generating a response sentence in a natural language format according to the search result, information necessary for control, operation or management of the on-vehicle equipment is searched in an interactive manner between a user and a computer. In the method, when moving to the process of generating the response sentence according to the search result, if no search object corresponding to the search result is obtained, An instruction to generate a response sentence prompting the user to relax the search condition, and the latest instruction by deleting the search result and the latest search condition that led to the definition from the context structure data The dialogue is newly advanced to a point immediately before the input of the sentence.

A computer-readable medium storing a program for performing an interactive information search process according to the invention of claim 8 extracts a search condition by analyzing a descriptive sentence input in a natural language format. Based on the conditions, the corresponding information is searched, and a search process in which the search conditions and the search results based on the search conditions are arranged in time series is stored as context structure data,
Manages, controls the flow of dialogue between the user and the computer in accordance with the content of the context structure data, and repeatedly executes a series of processes for generating a response sentence in the form of a natural language in accordance with the search result. An interactive information retrieval process for retrieving information necessary for control, operation or management of a device in an interactive manner between a user and a computer, wherein a search condition extracted by analyzing a newly input instruction is included in the context. If it is logically disjointed with one of the search conditions already recorded in the structure data, the search result and the search condition in the time series later than the search condition in the context structure data which is the disjunction are described above. By deleting from the context structure data and rewriting the search condition in the context structure data, which is the disjunction, to the newly extracted search condition, The Te advancing a new dialogue is obtained by recording a program for performing an interactive information retrieval process according to claim.

A computer-readable medium having recorded thereon a program for performing an interactive information retrieval process according to the ninth aspect of the present invention analyzes a directive input in a natural language format to extract a retrieval condition, and extracts the retrieved retrieval condition. Based on the conditions, the corresponding information is searched, and a search process in which the search conditions and the search results based on the search conditions are arranged in time series is stored as context structure data,
Manages, controls the flow of dialogue between the user and the computer in accordance with the content of the context structure data, and repeatedly executes a series of processes for generating a response sentence in the form of a natural language in accordance with the search result. An interactive information search process for searching for information necessary for controlling, operating, or managing a device in an interactive manner between a user and a computer, wherein a search condition extracted by analyzing a newly input instruction sentence is determined in advance or If the index does not satisfy the condition limitation index determined according to the situation and the search target cannot be narrowed down to a predetermined range even when the search is performed based on the search condition, the search process is not performed. And a program for performing an interactive information search process characterized by generating a response sentence urging the user to further limit the search conditions.

A computer-readable medium having recorded thereon a program for performing an interactive information search process according to the tenth aspect of the present invention analyzes a sentence input in a natural language format to extract search conditions, and Based on the conditions, the corresponding information is searched, and the search process in which the search conditions and the search results based on the search conditions are arranged in a time series is retained and managed as context structure data, and the user computer is controlled according to the contents of the context structure data. Control the flow of dialogue, and repeatedly execute a series of processes for generating a response sentence in a natural language format according to the search result, thereby providing information necessary for control, operation or management of the on-vehicle equipment to the user and the computer. Is an interactive information search process for searching in an interactive manner with the search result. When the process proceeds to the process of generating the response sentence according to the search result, the search result When no corresponding search target is obtained at all, the user is instructed to generate a response sentence that prompts the user to relax the search condition, and the search result and the latest search that led to the definition are specified. A program for performing an interactive information search process characterized in that a condition and a condition are deleted from the context structure data and a new dialogue is advanced to a point immediately before the input of the most recent instruction sentence.

[0026]

According to the interactive information retrieval apparatus of the first aspect of the present invention, when it is determined that the topic goes back to an earlier state in the course of the dialogue, the user explicitly gives an instruction. The context can be automatically returned to that state without doing so, and information retrieval can be performed while a natural conversation like a human being proceeds between the device and the user.

According to the interactive information retrieval apparatus of the present invention, it is possible to execute the retrieval processing after estimating the prospect of retrieving the retrieval target to some extent. Can be improved.

According to the interactive information retrieval apparatus of the present invention, the user is guided so that the subsequent retrieval can be efficiently advanced according to the retrieval result, and the context is prepared for the next user's utterance. Can be kept in an appropriate state, and information can be searched while a natural conversation such as a conversation between humans proceeds between the device and the user.

According to the interactive information search device of the fourth aspect, the point search process in the navigation system can be executed by a natural dialog between the user and the device.

According to the interactive information retrieving method using a computer according to the fifth aspect of the present invention, when it is determined that the topic is clearly going back to a previous state in the progress of the dialog, the user explicitly specifies the topic. , The context can be automatically returned to that state without instructing the user, and information retrieval can be performed while a natural conversation like a human being proceeds between the device and the user.

According to the interactive information search method using a computer according to the sixth aspect of the present invention, the search processing can be executed after estimating the prospect of narrowing down the search target to some extent, and the information search can be performed without wasteful processing. Execution efficiency can be improved.

According to the computer-based interactive information search method of the present invention, the user is guided so that the subsequent search can be efficiently advanced according to the search result, and the next user's utterance is uttered. , The context can be kept in an appropriate state, and information retrieval can be performed while a natural conversation such as a conversation between humans proceeds between the device and the user.

The computer-readable medium having recorded thereon the program for performing the interactive information search processing according to the invention of claim 8 is a program for causing a computer to read and execute the program for the interactive information search processing recorded thereon. If it is determined that the topic is obviously back to the previous state in the course of the dialogue, the context can be automatically returned to that state without the user's explicit instruction, and the human Information retrieval can be performed while such a natural conversation proceeds between the device and the user.

According to the ninth aspect of the present invention, there is provided a computer-readable medium having recorded thereon a program for performing an interactive information search process by causing a computer to read and execute the program for an interactive information search process recorded thereon. The search process can be executed after estimating the prospect that the search target can be narrowed down to some extent, and the execution efficiency of the information search can be improved by eliminating unnecessary processes.

The computer-readable medium having recorded thereon the program for performing the interactive information search process according to the tenth aspect of the present invention is a program for causing a computer to read and execute the program for the interactive information search process recorded thereon. In addition to guiding the user so that subsequent searches can be performed efficiently according to the search results, the context can be maintained in an appropriate state in preparation for the next user's utterance, and the interaction between humans can be performed. Information search can be performed while such a natural conversation proceeds between the device and the user.

[0036]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to the drawings. The interactive information search device of the present invention is a computer that incorporates a program for performing an interactive information search process and executes the program to act as the interactive information search device on the entire computer. When each is divided, the configuration as shown in FIG. 1 is obtained. FIG. 1 shows a configuration of an interactive information retrieval apparatus according to one embodiment of the present invention. The apparatus includes an input unit 1 for inputting a directive in a natural language format, a directive analysis processing unit 2 for performing a natural language analysis process on the input directive and extracting a search condition, and managing a search process. A dialogue control unit 3 for controlling a flow of a dialogue between a user and a device; a search execution unit 4 for searching for information based on search conditions; a storage device 5 for holding a database to be searched; It comprises a response sentence generation processing unit 6 for generating a response sentence in a natural language format, and an output unit 7 for outputting the generated response sentence to the user.

The input unit 1 is a voice input device such as a microphone for inputting a directive by voice, and a character input device such as a keyboard for inputting a text by text. However, in particular, when the present device is mounted on a vehicle and used, by using the former voice input device,
The burden on the user for inputting the instruction can be reduced.

When the instruction sentence is given by voice, the instruction sentence analysis processing unit 2 first performs a phoneme analysis process, and then performs a series of natural language analysis processes such as morphological analysis, syntax analysis, and semantic analysis. Then, search conditions are extracted in light of the meaning of the analyzed directive. The apparatus of the present embodiment performs interactive information search in the point search processing of the navigation system. Examples of search conditions in this case are shown in the table of FIG. The search conditions here are roughly classified into seven types: point attributes, point-to-point distances, point-to-point directions, adjacent roads, point departure and arrival times, point departure and arrival frequencies, and point-to-point departure and arrival relations. According to the difference, they are further subdivided into 20 types. In the table of FIG. 2, the rightmost column shows an example of a specific expression in a directive corresponding to each condition, but the directive analysis processing unit 2 determines which of the twenty types of the input directive is It is determined whether or not the condition is satisfied, and each condition component such as a corresponding reference point and a constraint value is extracted.

The dialogue control unit 3 is further divided into functional components.
2. It is composed of a target specificity determining unit 303 and a search result determining unit 304.

The context information management unit 301 holds the search conditions extracted by the instruction sentence analysis processing unit 2 and the search results searched by the search execution unit 4 based on the search conditions as context structure data representing a search process. Add, delete, or update the content accordingly. It is assumed that this context structure data is held, for example, in the form shown on the left side of FIG. Note that the right side of the figure shows an example of a dialogue between the user and the device that led to the construction of the data as shown on the left side.

The rejection condition information management unit 302 logically rejects the search conditions extracted by the directive sentence analysis processing unit 2 with each condition in the context structure data already stored in the context information management unit 301. Determine if there is. The target-specificity determination unit 303 determines whether or not the search conditions extracted by the directive sentence analysis processing unit 2 can sufficiently narrow down the search target when the search is performed, before performing the search.
The search result determination unit 304 determines whether the search target has not yet been narrowed down, has been sufficiently narrowed down, or has not been obtained at all as a result of the search performed by the search execution unit 4. Sort the process.

The search execution unit 4 searches the database stored in the storage device 5 for a point or the like that satisfies the search conditions, according to the search conditions extracted and written in the context structure data. The storage device 5 includes a CD, a DVD,
An HDD or other large-capacity storage medium and a reading device therefor. In the case of the apparatus of the present embodiment, for example, a list of spot data in which spot names, positions, attributes, etc. are recorded as shown in FIG. It is assumed that the list of the traveling history data accumulated according to the above is made into a database and stored. It should be noted that the points may include points registered by the user himself, in addition to those provided by the apparatus in advance. User registration points and travel history data are automatically added as the vehicle travels, and such information is stored in a storage medium such as a flash memory that can be written separately from the points provided by the device. You may make it the structure which memorize | stores. In this case, the search execution unit 4 can also recognize the user registration point without distinguishing it from the point provided by the apparatus.

The response sentence creation processing unit 6 generates a response sentence to be transmitted to the user from the apparatus according to the instruction of the dialog control unit 3 according to the search result. The response sentence is text-synthesized according to a predetermined template and specific information such as a search result provided by the dialog control unit 3. Further, when outputting the response sentence by voice, the response sentence generation processing unit 6 also executes speech synthesis based on the synthesized response sentence text.

The output unit 7 is a voice output device such as a speaker when the response sentence is output by voice, and is a display device such as a liquid crystal monitor or a CRT when the response text is output. Also, both of these may be provided simultaneously.

Here, the above-mentioned instruction sentence analysis processing unit 2, dialog control unit 3, search execution unit 4, and response sentence creation processing unit 6
U, an internal memory, and if necessary, an integrated computer including a sound chip and the like.
Further, each processor may be configured by an individual processor.

Next, the operation of the interactive information retrieval apparatus having the above configuration will be described. FIG. 5 shows a flowchart of a series of processes from the input of the instruction sentence to the generation of the response sentence. This figure 5
By repeatedly executing the flowchart shown in
The user can proceed with the search in an interactive manner with the present apparatus and find a desired point.

First, in step S501, the instruction sentence analysis processing unit 2 analyzes the instruction sentence newly input from the input unit 1 and extracts a search condition. This process is executed by determining which of the 20 types of search conditions shown in FIG. 2 the given sentence applies as described above, and extracting the components of the corresponding conditions.

For example, when a directive "restaurant within 2 km" is input, the directive analysis processing unit 2 performs a natural language process to determine that this is "the reference point and the upper limit of the distance from the reference point." Retrieval of target point satisfying the constraint "is recognized. Accordingly, this corresponds to the search condition c in FIG. 2, and “there” is first extracted as information corresponding to the reference point. In this case, since "there" is a pronoun and refers to a place that was a topic just before this directive was input in a series of dialogues, the information is already stored in the context information management unit 30.
1 should be described in the context structure data held. Therefore, the information in the context structure data to which "there" corresponds is further referred to as a reference point.

The target point is a target of the search in this instruction, and after executing the search, the result is embedded as an answer. This is shown in the figure <Answer>
Is shown. However, in this instruction, “restaurant” is specified as an attribute that defines the target point. Strictly speaking, this corresponds to the search condition that stipulates that the attribute belongs to the specified point attribute, as shown in Fig. 2A. In this way, it is used as a noun that directly indicates the target point in combination with other search conditions. The point attribute that has been set will be treated as an attribute that uniquely defines the search target, in distinction from the case of a. Hereafter, this is referred to as the "primary attribute"
I will call it. In the expression examples shown in the rightmost column of FIG. 2, the nouns "Family Restaurant", "Dealer", and "Stand" at the end of an example sentence in the form of an adjective phrase + a nominative form
Are all primary attributes. Further, even when the search condition is not combined with another search condition, for example, when only "gas station" is input, only the first primary attribute may be specified. However, there may be cases where primary attributes have already been defined in previous conversations,
In the subsequent processing, when a search condition is added to the context structure data, it may be processed again as a point attribute belonging condition of a.

Now, returning to the above-mentioned example of “restaurants within 2 km therefrom”, regarding the constraint values among the components of the search condition, in this case, the upper limit value of the distance indicated by a specific numerical value Is extracted as “2 km”. The constraint form is a flag for further finely identifying the above seven types of condition, and is uniquely determined by which of the subdivided search condition items. In this case, it is the “upper limit” as described above.

As for natural language processing such as syntactic analysis, semantic analysis, and identification of pronoun anaphores required in the above processing, generally known natural language processing techniques are described in the processing described here. And limited application.

Some of the search conditions shown in FIG. 2 include <answer> in the constraint value at the target point.
This is not an instruction to directly proceed with the search for a point, which is the object of the present embodiment, but is an instruction to obtain reference information for that purpose. For example, it is assumed that the target point is narrowed down to some extent while the dialogue is progressing, and an instruction sentence “How far is here?” Is input. In this case, it is determined that the search condition corresponds to the search condition e in FIG. 2, and the word "here" is interpreted as indicating the current location as the reference point, and the current location information is extracted. The information in the context structure data is referred to as the target point, as in the previous example, as in the previous example, as the supposed destination of “there”.

The constraint form is defined as "inquiry" as an instruction for inquiring a specific value of the distance. Then, a search is executed with the constraint value as the search target instead of the point. By receiving such a constraint value inquiry instruction, the user can obtain information that is a clue for searching for a place through a natural conversation. As shown in FIG. 2, in the constraint value inquiry instruction, the constraint form is all "inquiry", and the target point must be explicitly or implicitly specified in the instruction.

Further, it is also possible to receive instructions in the form of a combination of the above-mentioned items (a) to (d) in FIG. For example, the directive "Stands other than Nisseki along the Atsugi Kaido" is an adjoining road condition with a constraint value of "Atsugi Kaido" and a location attribute exclusion condition of "Nisseki" as a constraint value. Combination. The instruction sentence "Convenience store performed 5 or more times in the last month" is a combination of an arrival time condition that uses "last month" as a constraint value and a frequency lower limit condition that uses "5" as a constraint value. However, in this case, "Convenience store that went last month" and "Convenience store that went more than 5 times"
Are different from the simple combination of
Condition types that are accepted as one condition may be provided in addition to the 20 types shown in FIG.

Further, although not shown in FIG. 2, an instruction sentence for directly specifying a point name may be considered as a kind of point search instruction. For example, when an instruction sentence “Totsuka Station” is input, it can be considered that this is an instruction to search for a location whose name is “Totsuka Station”. If the name of each point is uniquely given, the target point can be directly obtained without actually executing the search processing. For convenience, a point search condition specifying the name is extracted. The following processing may be performed as a matter of course.

Returning to FIG. 5, after analyzing the directive and extracting the search condition, in step S502, the directive analysis processor 2 checks whether the search condition has been extracted from the input directive. The input sentence may actually be other than a point search instruction and a constraint value inquiry instruction. In such a case, the process branches to NO in step S502.

For example, if the search progresses and the target point is narrowed down to only one point, an instruction such as "set as the destination" is input to set this as the destination for route guidance. May be. In this case, the apparatus sets the corresponding point as the destination as instructed, and further shifts the processing to route search and the like. Further, when the user cannot find a desired point, an instruction such as “redo” may be input in order to start the search again from the beginning. Then, in this case, all of the context structure data up to that point are erased and the initial state is restored. Further, although the user intended to instruct a point search or a constraint value inquiry, the input sentence had a grammatical error or a lack of necessary information. It may not be possible to extract. In such a case, a failure of the condition extraction is responded to the user, and the state before the instruction is input is returned. As described above, when an instruction other than the point search and the constraint value inquiry is input, or when the search condition of the point search and the constraint value inquiry cannot be extracted, the process is shifted to a process according to the situation. Since such separate processing is not the object of the present invention, hereinafter, only the processing when the search condition can be extracted will be described (branch to YES in step S502).

In step S 503, the discharge condition determination unit 3
02 is already extracted from the context information management unit 30
In light of the context information data held in No. 1, it is determined whether or not both contain a condition that is logically exclusive. The logical exclusion here means that
This means that when two point sets satisfying each of the two search conditions are taken, it is clear that the intersection of both sets is an empty set. FIG. 6 shows a list of search conditions that are logically exclusive based on the 20 types of search conditions shown in FIG. 2 that are considered in the present embodiment.
The case (1) will be described. For example, “Chinese”
And "Italian" and "Japanese food" are treated as having no common terms with each other. " In the present embodiment, the point attributes are defined more explicitly in a hierarchical manner, and two child attributes having a common parent attribute are considered to be exclusive. For example, a parent attribute of "restaurant" is defined and its child attributes are "Chinese", "Italian", "French",
"Japanese food" is defined. Under the parent attribute "automobile dealer", a series name such as "Nissan", "Toyota", "Honda" is defined as a child attribute. As a result, the exclusivity of the point attribute and the information necessary for the attribute inheritance process at the time of the point attribute search can be defined at a time.

In the case of (5) in FIG. 6, even if there is a difference in the number of times specified in the two search conditions, if the difference is small, the two conditions are not considered to be exclusive in some situations. For example, when the instruction sentence “Golf course performed 5 times” is input, it is difficult to think that the number of times that the user has performed for each point is considered to be accurately stored, and the number of times 5 is always strictly defined. I can't say that. Therefore, in the present embodiment, the number of times specified in the frequency condition is treated as an approximate number, and the specified number of times n is allowed to have a tolerance of, for example, up to 0.2n. That is, in response to a search instruction with a condition of “performed 5 times”, actually 4 to
Processing is performed such as searching for points that have been performed six times. Therefore, in order for the two search conditions in (5) to be disjoint, the tolerance is defined as described above, where n and m (n <m) are designated, and n + [0 .2n] <m- [0.2m] is a necessary condition. The same applies to the case (6) in FIG. It should be noted that the definition of the tolerance is not limited to the above, and the conditions for determining rejection are appropriately determined accordingly.

Returning to the flowchart of FIG. 5, all the search conditions in the context structure data already held are:
If the search condition p does not conflict with the newly extracted search condition p, the process proceeds to step S504, and the context information management unit 301 adds p to the context structure data. The same applies to the case where context structure data has not yet been acquired in the initial state of the dialogue. As shown in the left side of FIG. 3, for example, the context structure data is stored in a form in which two different types of structural type data are linked to each other by a reference pointer.

The structural type data A and B shown by double frames in FIG. 3 describe a set of points obtained as a result of a search according to given search conditions. Or later,
This is called a “point set frame”. Further, the structured data a and b indicated by single-line frames describe search conditions themselves, and are referred to as “search condition frames”. The search condition frames a and b hold the respective slots for storing the search condition components shown in FIG.
Among these, the “reference point” slot and the “target point” slot have pointers (for example, “A” or “B”) that refer to the corresponding point set frame, or invalid values that indicate that nothing is referred to. − ”Is written. Also,
The point set frames A and B include “point set”, “attribute”,
Each slot of “name” and “search condition” is held. The “point set” slot is, for example, $ 12149,1
1595｝, I
The corresponding point set is stored in the form of a list of D numbers. In the “attribute” slot, when the above-mentioned first attribute is specified in the input instruction, for example, “stand”, this is written. If the instruction directly specifies a location name, the location name is written into the “name” slot (for example, “Totsuka Station”). Further, in the "search condition" slot, reference pointers to search condition frames describing search conditions that have led to the definition of the point set are stored as a list arranged in chronological order in the order in which the conditions were given. When a plurality of search conditions are given at the same time by one instruction, the search conditions are recorded in a form clearly specified. For example, in the first directive, the search condition p
1 is given, then p2 and p3 are given by one directive, and finally p4 is given, the search condition slot contains {(p1), (p2, p3), (p4)}. Record as That is, the value of the search condition slot is a list having a double structure in which each list element is further a list of search condition pointers. Furthermore, among these connected frames, for a point set frame that is currently focused on as a search target in a dialogue,
A mark MK indicating the point of interest is given. In FIG. 3, the point set frame B is the point of interest.

The process of establishing the context structure data shown in FIG. 3 will be described with reference to the dialogue example shown on the right side of FIG. First,
When the instruction sentence "Totsuka Station" is input, it can be understood that this is a place name directly specified by the processing in the preceding stage, and no search condition is given. Therefore, no search condition frame is generated, and only the point set A is generated. Also, without performing a search, ID number 1 of Totsuka Station
Since 3042 is uniquely determined, a list in which the point set slot is the only element is stored. Further, at this stage, since there is only one point set frame A, this is set as a point of interest.

Next, when the "stand within 1 km from the stand" is input, the search condition relating to the distance between points and the primary attribute "stand" are extracted. In general, even in a directive sentence analysis process, even if a point attribute is determined to be a primary attribute, it is associated with a point set frame (usually a point of interest) already existing in the context, and If a value has already been described in the attribute slot of the point set frame, the preceding point attribute is not added to the primary attribute but is added to the context structure data as a point attribute belonging condition (a in FIG. 2). . Even in this case, formally, “stand” may be a point attribute belonging condition of “Totsuka Station” which is the point of interest. However, here, the search condition includes the demonstrative pronoun "There", and the priority corresponding to the point of interest is higher in the demonstrative pronoun than in the primary attribute.
Eventually, it is determined that nothing corresponding to "stand" is not present in the current context. Therefore, a point set frame B in which “stand” is described as the primary attribute in the attribute slot is newly generated.

On the other hand, the upper limit condition of the distance between points (C in FIG. 2)
, A search condition frame a is also generated. The reference point is a point set frame A to which the demonstrative pronoun “there” corresponds, and the target point is a newly generated B. Point of interest M
Since K indicates the target point of the latest search condition,
Move from A to B. Further, a is added as a list element to the search condition slot of the point set frame B, and the search condition frame a and the point set frame B hold a bidirectional reference pointer. Since A is not the search target of the condition a, a is not recorded in the search condition slot of A. Thereafter, when the subsequent search processing is executed, five points are searched as points that satisfy the relevant condition, as shown by the system response in the dialog example of FIG. At this point, the point set slot of the point set frame B includes:
The list of the ID numbers of these five points is written.

Further, when an instruction of "has been along National Road No. 1" is input, adjacent road conditions are extracted by pre-processing. If there is no primary attribute, all of the extracted search conditions are considered to define the current point of interest. Therefore, a new search condition frame b is generated, and the target point becomes the current point of interest B. On the other hand, point set frame B
Is added to the search condition slot of. Since B is still the target point of the latest search condition, the point of interest MK does not change.

Then, as a result of executing a search for B,
Assuming that two points are searched as shown in the dialogue example, the point set slot of the point set frame B is rewritten, and the context structure data has the form shown in FIG. in this way,
As for the search conditions, all the given conditions are retained in chronological order unless the processing of the exclusion condition described later is performed. Only the latest search results are kept.

As an additional process of adding the extracted search condition to the context structure data, another example in which the point attribute extracted as the primary attribute is transferred to the point attribute belonging condition will be described. In the state illustrated in FIG. 3, it is assumed that an instruction “Nisseki west of Totsuka Station” is further input. In this case, according to the preceding process, the point-to-point azimuth condition and the primary attribute "sunstone" are extracted from this instruction. The reference point of the point-to-point direction condition is given by the point name "Totsuka Station", and there is no demonstrative pronoun corresponding to the current point of interest.
This time, “Nisseki” corresponds to the point of interest B. Since “Stand” is already assigned to the attribute slot of the point set frame B as the primary attribute, “Nisseki” is transferred to the point attribute belonging condition. Therefore, as shown in FIG. 7, two new search condition frames c and d are generated,
In both cases, the point set frame B is referred to as the target point.

Since c and d are simultaneously defined in the search condition slot of the point set frame B, (c,
d). Note that “Totsuka Station” specified by the point name again can be handled as a uniquely determined constant, and therefore may be duplicated in the context structure data. Therefore, in the example of FIG. 7, a new point set frame C is generated. Of course, the corresponding point set frame A is searched in the context structure data, and A is referred to in the reference point slot of the search condition frame c. Is also good.

As described above, in step S504, the newly extracted search condition is added to the context structure data. On the other hand, if the exclusion condition is found in step S503, the processing is performed after step S505. To correct the context structure data.

The exclusivity of the search condition is determined between the newly extracted search condition and each condition described in the search condition slot of the point set frame which is the point of interest in the context structure data. And Now, the search condition described in the double list structure in the search condition slot at the point of interest is represented by ｛(p _{1
1, p 1 2, ...)} , (p 2 1, p 2 2, ...), ...,
_{(..., p i j, p} i j + 1, ...), ... and}. here,
Search p _{i j,} out of at least j or more simultaneously given search condition i th point with respect to the target point,
For convenience, it indicates the j-th element. In addition, since there are generally a plurality of newly extracted search conditions,
1, p2, ..., pk.

In step S505, the search conditions p1, p
2, ..., those given in the earliest time among the p _{i j} is the conditions and disjoint of pk, i.e., find what i is minimum. Hereinafter, the point at which the exclusion condition is given will be referred to as the “exclusion point”.

[0072] If it is only one search condition given new, i.e. if k = 1, the p _{i j} is a disjoint by process described below is guaranteed to be at most one I have. However, if multiple search conditions are given at the same time, a disjoint for their respective p _i j
May exist, so it is necessary to identify the minimum value at the time of exclusion. Now, among the new search conditions, pk1 is p
_{If il} j 'and pk2 are exclusions from p _i2 j "and i1≤i2, the earliest time when the exclusion condition is given is i1.

[0073] Therefore, in step S506, from the search conditions slot of the target point, the earliest mutually exclusive point i1 later given search conditions, ie i> to delete all of the i1 becomes p _{i j.} Also, the search condition frames referred to by the deleted pointers are all deleted unless they are referred to from other frames.

In step S507, the newly obtained
.., Pk obtained at the time point i1
Search condition slot of the point set frame of the point of interest as a condition
Add to At this time, the search terms already written
P_i1For j and exclusion, the exclusion condition
Shall be replaced with For example, given at time i1
(P_i11, p_i12, p_i13, p
_i14) is written and newly obtained search condition
Of p1, p2 and p3, p1 is p_i13 and p3 is p
_i1If it is 2, the search condition given at time i1
The case is (p_i11, p3, p1, p_i14, p2)
Can be replaced. Search condition pointer replaced at this time
p_i12, p_i1Also delete search condition frame referenced by 3
Is done. Also, the search condition ｛(p₁
1, p₁2, ...), (p₂1, p ₂2, ...), ..., (p
_i1-11, p_i1-12, ...)｝
Unaffected. This causes the dialogue context to stay at time i1
This means that it has been traced back.

Further, in step S508, the point set slot of the point set frame of the point of interest whose search condition has been rewritten is cleared to an empty list. This is to clearly indicate that the point set held so far has lost the association with the rewritten search condition, and not to lose the consistency in the point set frame.

FIG.
This will be described with reference to the dialog example shown in FIG. In the dialogue up to A4 in FIG. 14, the context structure data is in the state shown in FIG. That is, with respect to the point set frame A of the point of interest designating “family restaurant” as the primary attribute, “east from Yamato station”, “west from Tsurugamine”,
Three search conditions a, b, and c of "Chinese" are given.

Here, when a new instruction sentence "Italian no ha" is input, a search condition as shown as a search condition frame d in FIG. 9 is extracted. Exclusivity between d and the search condition frames a, b, and c given up to that time is examined. It turns out that it is. Since the search condition frame c is the condition most recently given to the point set frame A of the point of interest,
The search conditions prior to the rejection time corresponding to step S506 are not deleted, and the rejection conditions c are rewritten with new search conditions d. Then, the point set slot of the point set frame A is cleared, and eventually, the context structure data becomes as shown in FIG. That is, the point set frame A of the point of interest
Means "Italian family restaurant east of Yamato station and west of Tsurugamine station."
After the context structure data correctly traces back to A3 in FIG. 14, Q5 is given again as a condition.

When step S504 or step S508 is completed, the process proceeds to step S509. Here, the target specificity determination unit 303 determines whether or not the search target can be sufficiently narrowed by the given search condition. For that purpose, it is necessary to determine an index for determination in advance. Hereinafter, this index will be referred to as an index of conditional limitation. In the present embodiment,
The following two indices are defined.

One is a case where only one point attribute condition (A or B in FIG. 2) is given in the dialogue start state. In the case of the point attribute belonging condition (A in FIG. 2), this means that the primary attribute has been given. In general, it is expected that there are a considerable number of points belonging or not belonging to a certain point attribute. For example, in the example of the dialog shown in FIG. 14, when the first instruction Q1 "family restaurant" is input, since the "family restaurant" is extremely large in the database, the search is executed after the conditions are further limited. Is efficient. Therefore, in such a case, it is determined in step S509 that the condition limitability is not sufficient.

Another index of condition limitation is that when the number of elements of the point described in the point set slot of the point set frame of the point of interest exceeds a certain value, the newly input search condition is This is a case where it is determined that there is only a slight limitation in light of the search conditions given up to that time. For example, in the dialogue example in FIG. 14, it is known that there are 259 “family restaurants east of Yamato Station” when the dialogue proceeds to A2. Here, assuming that a descriptive sentence of “other than Denny's” is input as Q3, “Denny's” is a child attribute of “family restaurant”, and a point attribute having the same family restaurant as a parent attribute is “Sky Lark”; There are many such as "royal hosts". Therefore, if only "family restaurant" is specified for the point attribute, it is expected that searching for "other than Denny's" will not be able to narrow down the points sufficiently, so the condition limitability is not sufficient. I will do it. However, if the point set is not 259 at the stage of A2 but is narrowed down to, for example, about 10 points, the number of point sets does not increase any more even if the search is performed for "other than Denny's". In such a case, it is determined that the condition limitation is sufficient.

The index of the condition limiting property is not limited to the example given here, and another index may be appropriately determined. For example, in the above example of the first index, if only the primary attribute is specified, the condition limitability is not sufficient. A small number. Therefore, for example, the location attribute given only to the user registration location shown in FIG. 4 may have sufficient condition limitation even if it is designated as the primary attribute.

If it is determined in step S 509 that the condition limitability is sufficient, the process proceeds to step S 510, where the search execution unit 4 performs the processing according to the search conditions given to the point set frame of the point of interest. Search for a point. At this time, the previous search result given to the point set slot in the point set frame of the point of interest may be appropriately used. For example, after a new search condition is extracted by Q3 in FIG. 14, the search execution unit 4 may search for “family restaurants located east of Yamato station and west of Tsurugamine station” for all point data. Instead, the search can be speeded up by examining 259 “family restaurants east of Yamato Station” that are “west of Tsurugamine Station”. However, if the search condition has been rewritten due to the existence of the exclusion condition, as in Q5 in FIG. 5, the point set slot is also cleared in step S508. The Italian family restaurant east of Tsurugamine Station and west of the east ”will be searched again from the data of all points. In any case, it is assumed that the search result is newly written in the point set slot of the point set frame of the point of interest.

Here, a supplementary description will be given of a method of inheriting attributes, particularly for a search relating to point attributes. As described above, in the present embodiment, each point attribute is defined as having a hierarchical configuration. Therefore, in the search for the location attribute, one of all attributes below the specified attribute
If at least one point is assigned to a point, the point is regarded as having the specified attribute. For example, ID = 11595 and ID = 121 shown in FIG.
For the 49 spots, only "Showa Shell" and "Nisseki" are given as spot attributes. However, since these are lower-level attributes (child attributes) of the location attribute "gas station", these locations are searched when "gas station" is specified as a search condition. vice versa,
If the point attribute exclusion condition is given as “other than gas station”, these points are not searched.

On the other hand, if it is determined in step S509 that the condition limitability is not sufficient, step S511
In, a flag indicating that the search has not been executed is added to the newly input search condition, and the process proceeds to a subsequent step. This is to maintain consistency between the point set slot in the point set frame, which is the point of interest, and the search condition slot, as in step S508. Thereafter, when a new search condition is given and it is determined that the condition limitability is sufficient, and the process proceeds to step S510, the search execution unit 4 gives the latest given one of the points described in the point set slot. In addition to the retrieved search conditions, a search is made for a search that satisfies the search condition given a flag indicating that the search has not been executed. For example, in the dialogue example of FIG.
In the case where the condition “other than Denny's” is given, if the search is not performed according to the above-described determination of the condition limitability, the condition “other than Denny's” is included in the search condition slot of the point set frame of the point of interest. Despite being added, the point set slot is 2 obtained in A2
59 cases are held. Therefore, if a search non-execution flag is given to the condition "other than Denny's", then in the next Q4, for example, a condition "within 5 km from Tsurugamine Station" is given, and the condition limitation is confirmed, and the search is executed. When performing the search, the search execution unit 4 can search for the items other than “Denny's within 5 km from Tsurugamine Station” from the previous 259 cases by combining the unexecuted condition and the newest condition. become.

When the search is executed in step S510 and the search result is written in the point set frame of the point of interest, in step S512, the search result determination unit 304 checks the number of searched points and performs the subsequent processing according to the number of points. Sort out. Here, the processing is described as (i) the number of points is large,
If the target is not sufficiently narrowed down, (ii) the number of points is 0
If not, but it ’s somewhat targeted,
(Iii) Divide into three cases where the number of points is 0.

[0086] Regarding the distinction between (i) and (ii), specific criteria need to be clearly defined. To do so, the threshold value of the number of points is determined in advance by a constant. For example, when the number of points is 10 or more, (i), and when the number of points is 9 or less, (ii)
It may be. Alternatively, before writing the result newly searched in the preceding step S510 to the point set slot of the point set frame of the point of interest, the number of points n obtained before the search is held, and A value multiplied by a small constant ratio may be used as a threshold value for discriminating between (i) and (ii). For example, if the fixed ratio is set to 0.7 and the number of points before the search is supposed to be 200, (i) if the number of newly searched points is 140 or more, and (ii) if the number of newly searched points is 139 or less. In addition, such as combining these two,
How to determine the threshold value is not limited to the illustrated one.

In step S512, the target points are narrowed down to some extent, and if it is determined that this corresponds to the above case (ii), the process proceeds to step S513, where the response sentence generation processing unit 6 informs the user of the obtained points. Is generated and notified to the user through the output unit 7. After this, the system enters a state of waiting for input of a directive.

If the target point is not sufficiently narrowed down in step S512, that is, if it is determined to correspond to (i), or if the condition limitation is determined to be insufficient in step S509, and the processing in step S511 is completed. Moves to the process of step S514. Here, since the response sentence generation processing unit 6 has not narrowed down the target points, the response sentence generation processing unit 6 generates a response sentence urging the user to further limit the search conditions. At the same time, a specific example of a more limited search condition may be created, and the user may be suggested what kind of instruction should be input next. For example, "The target cannot be narrowed down.
How about searching within 0m? Is created and output. As a specific example of the condition limitation, based on the most recently input search condition, that is, based on the search condition described at the end of the list in the search condition slot of the point set frame of the point of interest, its sufficient condition (not enough Conditions that are not necessary) may be created. If there are a plurality of latest search conditions, one of them is appropriately selected. FIG. 10 illustrates a method of creating a sufficient condition for each condition type. After creating the response sentence for urging the limitation of the condition in this way, the response sentence is output in the same manner as in step S513, and the process enters a state of waiting for the next instruction sentence.

If no target point is obtained in step S512 (iii), the process proceeds to step S515, where the context information management unit 301 determines whether or not the search condition slot of the point set frame of the point of interest is Delete the last search condition described at the end of the list. This is a process for invalidating the current search process and returning the context structure data to a state before the current instruction is input. As a result, the next instruction sentence is smoothly connected to the previous conversation context, and the flow of the conversation can be made natural. For example, in the dialogue example of FIG.
It is assumed that the target point was not obtained as a result of the search according to Q4. The context structure data in this case is the one shown in FIG. 8, and the point set slot of the point set frame A of the point of interest is an empty list. This is because, as a result of the search in step S510, no target point is obtained, and an empty list is written in the slot. Therefore, in the processing of this stage,
The search condition frame c which is the last element of the search condition slot list of the point set frame A is deleted, and the search condition slot list becomes {(a), (b)}. In this state, when "Italian no ha" of Q5 is input next, this is added to the slot, and eventually, the same context structure data as that shown in FIG. 9 is obtained. It should be noted that the point set slot in the point set frame of the point of interest is not limited to the empty list, and the main stage may be processed so as to return to the previous state. To do so, when writing the search result in step S510, the search result before that may be temporarily stored. In the present example, FIG.
In step S510, if no target point is obtained as a result of the search in step S510, if the point set slot of the point set frame of the point of interest is cleared to the empty list, the point held there is cleared. ID
, That is, the 33 point sets indicated by A3 in the dialogue example are temporarily stored. After that, the latest search condition c is deleted in the processing of this stage, and the 33 point sets saved in the temporary storage are returned to the point set slot in the point set frame of the point of interest. By such processing, in the subsequent search corresponding to Q5, the search execution unit 4 reads “East from Yamato Station,
You can find the "Italian" point from 33 family restaurants west of Tsurugamine Station, and you can improve the execution speed.

Further, in step S516, the response sentence generation processing unit 6 generates a response sentence urging the user to further relax the search condition since the target point cannot be obtained.
At this time, a more specific example of the search condition that is more relaxed may be created, and the user may be suggested what kind of instruction should be input next. For example, a response sentence such as "Cannot search for an object. Why not search within 5km from Totsuka Station?" As a specific example of the condition relaxation, based on the latest search condition deleted in the previous stage,
The necessary conditions (necessary but not sufficient conditions) may be created. If there are multiple recent search conditions,
One of them is appropriately selected. The necessary condition for each condition type is created by an operation reverse to the operation shown in FIG. 10 (operation in which the arrow is turned in the opposite direction). After creating a response sentence for prompting the relaxation of the condition in this way, this is output in the same manner as in step S513, and the process enters a state of waiting for the input of the next instruction sentence.

As described above, after the series of procedures in FIG. 5 is completed and the input of the directive is awaited, if a further directive is input, the procedure from step S501 is repeated again, and the dialogue is started. The search process will proceed in a format. Note that, even when the content of the input instruction is not a point search instruction, but is a restriction value inquiry instruction such as "distance to here", the search is performed in step S504, similarly to the point search instruction. A condition frame is generated,
Appended to the context structure data. In step S510, the search is performed in such a manner as to obtain the constraint value slot of the generated search condition frame. No judgment is made as to whether the condition is exclusive or restrictive. In addition, although the response may be appropriately changed in accordance with the search result, no restriction or relaxation of the condition is proposed.

As an example, FIG. 11 shows the state of the context structure data in the case where "distance from here" is further input in the dialogue example shown on the right side of FIG. The constraint form slot of the newly created search condition frame c is shown in FIG.
As described above, it is always "inquiry". Also,
Although the point set frame B, which is the point of interest, is referred to by the target point slot, conversely, c is not added to the search condition slot of the point set frame B. This is because the search condition frame c is not a condition that defines the point set of the point set frame B. Furthermore, since the point of interest does not move from the point set frame B, when a point search is instructed by the next instruction sentence input, the subsequent search processing should be continued in exactly the same manner as when the search condition frame c does not exist. Can be.

As described above, the interactive information search apparatus according to the present embodiment can efficiently and quickly perform a search operation in a natural dialogue between humans, and therefore, does not impose a heavy burden on the user. Thus, an information search function suitable for use in a vehicle can be realized.

In the above embodiment, the point search processing in the car navigation system has been described.
The application system of the apparatus of the present invention is not limited to this. For example, for a program search process in a car stereo, a corresponding database as shown in FIG. 4 and a search condition of a search condition as shown in FIG. If a specification is given and the rejection of the search condition as shown in FIG. 6 and the inclusion relation as shown in FIG. 10 can be clearly defined, the present invention can be applied to any information search processing.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a functional configuration of an embodiment of an interactive information search device according to the present invention.

FIG. 2 is a table showing specifications of search conditions used by the apparatus according to the embodiment.

FIG. 3 is an explanatory diagram showing an example of a dialog corresponding to the configuration of context structure data in the apparatus according to the embodiment.

FIG. 4 is an explanatory diagram showing a storage form of point data and travel history data in the device according to the embodiment.

FIG. 5 is a flowchart of an interactive information search process executed by the apparatus according to the embodiment.

FIG. 6 is a table showing a list of search conditions, which are exclusive, in the apparatus according to the embodiment.

FIG. 7 is an explanatory diagram showing another example of context structure data in the device according to the embodiment.

FIG. 8 is an explanatory diagram showing still another example of context structure data in the device according to the embodiment.

FIG. 9 is an explanatory diagram showing still another example of context structure data in the device according to the embodiment.

FIG. 10 is a table showing a list of methods for creating sufficient search conditions.

FIG. 11 is an explanatory diagram showing an example of context structure data including a constraint value inquiry instruction.

FIG. 12 is an explanatory view of a dialogue example correctly understood in a conventional example.

FIG. 13 is an explanatory diagram showing a network structure of a conversation context understood in a conventional example.

FIG. 14 is an explanatory diagram of a dialogue example that is not correctly understood in the conventional example.

FIG. 15 is an explanatory diagram showing a network structure of a conversation context that is not understood in the conventional example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input part 2 Instruction sentence analysis processing part 3 Dialog control part 4 Search execution part 5 Storage device 6 Response sentence generation processing part 7 Output part 301 Context information management part 302 Exclusivity condition judgment part 303 Target specificity judgment part 304 Search result Judgment unit

Claims (10)

[Claims] An instruction analysis means for analyzing an instruction input in a natural language format to extract a search condition, and a search for searching for corresponding information based on the search condition extracted by the instruction analysis. Executing means, holding and managing, as context structure data, a search process in which the search conditions and the search results executed by the search execution means based on the search conditions are arranged in a time series, between the user and the device according to the contents of the context structure data. A dialogue control means for controlling a flow of a dialogue; and a response sentence generation means for generating a response sentence in a natural language format according to the search result. A series of steps from analysis of the instruction sentence to generation of the response sentence By repeatedly executing the processing, an interactive information search device that searches for information necessary for control, operation, or management of the on-vehicle equipment in an interactive manner between the user and the system is provided. The dialogue control means is such that the search condition extracted by the instruction sentence analysis means for a newly input instruction sentence is logically inconsistent with one of the search conditions already recorded in the context structure data. In this case, a search result and a search condition that are later in time series than the search condition in the context structure data that is the disjunction are deleted from the context structure data, and the search condition in the context structure data that is the disjunction is deleted. An interactive information search apparatus, characterized in that control is performed so that a new dialogue is advanced to a corresponding point in the past by rewriting to the newly extracted search condition. 2. A directive analyzing means for analyzing a directive input in a natural language format to extract a search condition, and a search for searching corresponding information based on the search condition extracted by the directive analyzing means. Executing means, holding and managing, as context structure data, a search process in which the search conditions and the search results executed by the search execution means based on the search conditions are arranged in a time series, between the user and the device according to the contents of the context structure data. A dialogue control means for controlling a flow of a dialogue; and a response sentence generation means for generating a response sentence in a natural language format according to the search result. A series of steps from analysis of the instruction sentence to generation of the response sentence By repeatedly executing the processing, an interactive information search device that searches for information necessary for control, operation, or management of the on-vehicle equipment in an interactive manner between the user and the system is provided. The dialogue control means determines that the search condition extracted by the descriptive sentence analysis means with respect to the newly input descriptive sentence does not satisfy an index of the condition limitability determined in advance or according to the situation, and When it is determined that the search target cannot be narrowed down to a predetermined range even when the search execution unit executes a search based on the search condition, the processing is not transferred to the search execution unit, and the user is prompted to further limit the search conditions. An interactive information search device for instructing the response sentence generating means to generate a response sentence prompting the user. 3. A directive analyzing means for analyzing a directive input in a natural language format to extract a search condition, and a search for searching corresponding information based on the search condition extracted by the directive analyzing means. Executing means, holding and managing, as context structure data, a search process in which the search conditions and the search results executed by the search execution means based on the search conditions are arranged in a time series. A dialogue control means for controlling a flow of a dialogue; and a response sentence generation means for generating a response sentence in a natural language format according to the search result. By repeatedly executing the processing, an interactive information search device that searches for information necessary for control, operation, or management of the on-vehicle equipment in an interactive manner between the user and the system is provided. The dialogue control means, when transferring processing to the response sentence generation means according to the search result of the search execution means, relaxes the search condition when no search object corresponding to the search result is obtained. Instructs the response sentence generation means to prompt the user, and deletes the search result and the latest search condition that led to the search result from the context structure data to thereby provide the latest instruction. An interactive information retrieval apparatus characterized in that the interactive information retrieval apparatus is controlled so as to advance a new dialog to a time immediately before a sentence is input. 4. The interactive information search device according to claim 1, wherein a point search process is performed in a navigation system. 5. Analyzing a sentence input in a natural language format to extract a search condition, and based on the extracted search condition,
A search process in which the corresponding information is searched, and the search condition and the search result based on the search condition are arranged in a time series is retained and managed as context structure data, and the flow of the dialogue between the user and the computer according to the content of the context structure data By repeatedly executing a series of processes for generating a response sentence in a natural language format according to the search result, thereby providing information necessary for control, operation or management of the on-vehicle equipment in an interactive format between the user and the computer. In the interactive information search method using a computer, the search condition extracted by analyzing a newly input instruction is logically different from one of the search conditions already recorded in the context structure data. In the case of exclusion, the search result and the search condition in the time series later than the search condition in the context structure data which is the exclusion are obtained from the context structure data. Delete
And a dialogue using a computer characterized by renewing a search condition in the context structure data, which is the disjunction, with the newly extracted search condition, thereby advancing a new dialogue to a corresponding point in the past. Type information search method. 6. A search condition is extracted by analyzing an instruction sentence input in a natural language format, and based on the extracted search condition,
A search process in which the corresponding information is searched, and the search condition and the search result based on the search condition are arranged in a time series is retained and managed as context structure data, and the flow of the dialogue between the user and the computer according to the content of the context structure data By repeatedly executing a series of processes for generating a response sentence in a natural language format according to the search result, thereby providing information necessary for control, operation or management of the on-vehicle equipment in an interactive format between the user and the computer. In the interactive information search method using a computer to search in, the search condition extracted by analyzing the newly input instruction sentence does not satisfy the index of the condition limitability predetermined in advance or according to the situation, If the search target cannot be narrowed down to a predetermined range by executing the search based on the search condition, the search process is not performed and the search process is not performed. And generating a response sentence urging the user to limit the search conditions. 7. A search condition is extracted by analyzing a directive sentence input in a natural language format, and based on the extracted search condition,
A search process in which the corresponding information is searched, and the search condition and the search result based on the search condition are arranged in time series is retained and managed as context structure data, and a flow of dialogue between the user and the computer is performed according to the content of the context structure data. By repeatedly executing a series of processes for generating a response sentence in the form of a natural language according to the search result, the information necessary for control, operation or management of the on-vehicle equipment in the interactive form between the user and the computer In the interactive information search method using a computer to search for, in the process of generating the response sentence according to the search result, if no search target corresponding to the search result is obtained at all, the search Instructs the user to generate a response sentence that prompts the user to relax the conditions, and also compares the search result with the latest search condition that led to the definition. An interactive information search method using a computer, characterized in that a new dialogue is advanced to a point immediately before input of the latest instruction sentence by deleting from the context structure data. 8. Analyzing a directive input in a natural language format to extract search conditions, and based on the extracted search conditions,
A search process in which the corresponding information is searched, and the search condition and the search result based on the search condition are arranged in a time series is retained and managed as context structure data, and the flow of the dialogue between the user and the computer according to the content of the context structure data By repeatedly executing a series of processes for generating a response sentence in a natural language format according to the search result, thereby providing information necessary for control, operation or management of the on-vehicle equipment in an interactive format between the user and the computer. In the interactive information search process, the search condition extracted by analyzing the newly input instruction is logically inconsistent with one of the search conditions already recorded in the context structure data. In the case of, the search result and the search condition that are later in time series than the search condition in the context structure data which is the contention are deleted from the context structure data, and A program for performing an interactive information search process, characterized in that a search condition in a certain context structure data is rewritten to the newly extracted search condition, whereby a new dialogue is advanced to a corresponding point in the past. Computer readable media. 9. Analyzing a directive input in a natural language format to extract search conditions, and based on the extracted search conditions,
A search process in which the corresponding information is searched, and the search condition and the search result based on the search condition are arranged in a time series is retained and managed as context structure data, and the flow of the dialogue between the user and the computer according to the content of the context structure data By repeatedly executing a series of processes for generating a response sentence in a natural language format according to the search result, thereby providing information necessary for control, operation or management of the on-vehicle equipment in an interactive format between the user and the computer. In the interactive information search processing, the search condition extracted by analyzing the newly input instruction does not satisfy the index of the condition limitability predetermined in advance or according to the situation, and If the search target cannot be narrowed down to a predetermined range even if the search is executed based on the conditions, the search processing is not performed and the search conditions are further limited. Interactive information retrieval process recording a computer-readable medium storing a program for performing, characterized in that to generate a response sentence prompting the user to. 10. Analyzing a directive input in a natural language format to extract a search condition, searching for corresponding information based on the extracted search condition, and retrieving the search condition and a search result based thereon. A search process arranged in a series is retained and managed as context structure data, a flow of dialogue between a user and a computer is controlled in accordance with the content of the context structure data, and a response sentence in a natural language format is generated according to the search result. It is an interactive information search process of searching for information necessary for control, operation or management of the on-vehicle equipment in an interactive manner between a user and a computer by repeatedly executing a series of processes that perform At the time of proceeding to the process of generating the response sentence, if no corresponding search target is obtained as the search result, the user is prompted to relax the search condition. By instructing the generation of an answer, and by deleting the search result and the latest search condition that led to the definition from the context structure data, a new dialogue can be performed retroactively to the time immediately before the input of the latest instruction. A computer-readable medium having recorded thereon a program for performing an interactive information search process.