JPH083795B2 - Question estimator - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial field of application) The present invention analyzes the contradiction or omission of question contents input to a system, and analyzes the original question contents without contradiction or omission. The present invention relates to a question estimator for seeking

(Prior Art) Recently, many information processing devices such as personal computers and word processors have been introduced into various offices. Along with this, not only professional operators but also general people have more opportunities to use these information processing devices.

By the way, with the development of information processing technology, the processing functions of these information processing devices are diversified, and the processing operations thereof tend to be complicated. Therefore, not only ordinary people but also people who are somewhat skilled in the operation,
The processing operation for a processing function that is rarely used may not be understood. In addition, when an error occurs during the use of the information processing device, the coping method may not be known. As a result, the information processed up to that time, for example, the document information input in the word processor has been lost, and the processing has to be interrupted.

In such a case, generally, a person skilled in the processing operation is made ready for the coping method (processing operation method), or the handling manual of the information processing device is checked. However, it was extremely difficult to find a desired processing operation method when there were no skilled person and the amount of manuals was huge.

Therefore, recently, it is considered that the information such as an operation method to be known is input to the system in the form of a question and the system side responds to this. However, in this case, it is necessary to enter a question in a format that the system can understand, and there is generally a problem in terms of operability. For example, it became necessary to understand the input format of the question and to separately remember the processing operation for entering the question.

Considering man-machine interface, it is considered to input a question in Japanese, for example.
However, for example, if you enter a question using abbreviated expressions peculiar to Japanese language, or if the entered question contains inconsistencies, the system cannot interpret the question and you have no choice but to say "Unable to answer". . For example, when an error occurs, the situation is unknown, and even if you enter a question using only the information within the range that you know, the question cannot be interpreted because the composition requirements of the question are insufficient. There was a problem such as.

(Problems to be Solved by the Invention) In the present invention, when the question input to the system has omissions or inconsistencies as described above, the system cannot deal with this and eventually derives an answer to the question. It was made in view of the problem that it is impossible to do, and the purpose is to estimate the intent of the question and answer the question even if there are omissions or inconsistencies in the entered question contents. It is to provide a question estimating device that can derive the question.

[Structure of the Invention] (Means for Solving Problems) The present invention relates to an item candidate table that stores candidates for the content of each item constituting a question, and a contradiction that stores a contradiction condition between the configuration items. Using a table and a candidate item correspondence table that stores the configuration items necessary to form a question, for example, a question that is selectively entered from among a plurality of item content candidates presented for each configuration item that forms a question The content of each configuration item is obtained for each configuration item unit, and the contradiction between the extracted contents for each configuration item unit is checked with reference to the above contradiction table, and the omission of the configuration item of the above question is described above. Check with reference to the candidate item correspondence table, and if there is a contradiction or omission in the content of the constituent items of the question, refer to the item candidate table, and appropriately refer to the situation of the applied processing function part of the system, etc. Create item candidate list And, in which so as to estimate the original question about the inconsistencies or omissions of questions according to the item candidate list.

(Operation) Thus, according to the present invention, even if there is an omission in the entered question content, or even if the question content contains a contradiction, from the relationship of the content of the constituent items that make up the question,
For example, the contents of the omitted configuration items are estimated and the contents of the inconsistent configuration items are corrected. Then, the lack of the entered question content is compensated, and an omission-free and consistent question is generated which takes intent of the question.

As a result, the system properly interprets the question,
It becomes possible to effectively derive an answer to the question. Therefore, even if a question is input within the range of the understood situation, it is possible to obtain an answer and effectively use the information processing device.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic configuration diagram of a Japanese word processor to which the embodiment system is applied.

Information input through the input device 1 including a keyboard for inputting character codes and a microphone for inputting voice is subjected to predetermined information processing by the processing control device main body 2 and is displayed from a display, a printer or the like. Is output to the output device 3. The external storage device 4 connected to the processing control device 2 is composed of a magnetic disk or the like, and stores information such as a contradiction table between candidates and a candidate item correspondence table described later.

Here, the functional unit of the process control device 2 that executes the process for the question input from the input device 1 is roughly configured as follows.

The input unit 11 forms an interface with the input device 1, and asks a question given from the input device 1 to a question analysis unit.
Giving to twelve. The question analysis unit 12 divides the input question into its constituent item units, and analyzes whether or not there is omission or contradiction between the contents of each constituent item unit. Then, the analysis result is sent to the question item data buffer 20 of the data buffer unit 19.
Stored in.

The application processing function unit 13 executes application processing such as a Japanese word processor, and the function execution state is analyzed by the state analysis unit 14.

As will be described later in detail, when the input question has omissions or inconsistencies, the item candidate generation unit 15 causes the external storage device 4 to operate.
An item candidate table is read out from the item candidate table data buffer 21 and an item candidate is generated based on the function status of the application processing function unit 13 obtained by the state analysis unit 14, and the information is generated in the item candidate list. Is stored in the item candidate list / data buffer 22 as

The question forming unit 16 creates a question without omission or contradiction according to this item candidate list, and the answer searching unit 17 searches for an answer to the created question. The answer obtained by this search is output from the output unit 18 to the output device 3 and presented to the operator.

FIG. 2 shows the flow of the processing procedure from the analysis processing of a question to the acquisition of the answer in the system configured as described above. The details of this method will be described below along this flow.

When a question is input to the input unit 11 (step a), the question analysis unit 12 first divides the question into constituent item units, and obtains the contents of each constituent item (step b). Incidentally, as will be described later, it is possible to display the candidates of each of the constituent items that make up the question, detect the selected and designated candidate from the candidates, and input the question.

Then, when the contents of the respective constituent items of the entered question are respectively obtained, it is checked whether or not the question is omitted or inconsistent (step c). The omission of the question and the analysis of the contradiction are performed by using, for example, the inter-candidate contradiction table and the candidate item correspondence table stored in the external storage device 4 in the data formats shown in FIGS. 3 (a) and 3 (b), respectively. Done.

That is, the configuration items that make up the question are [target] [action]
Assuming that there are three [contents of description], as shown in FIG. 3A, the item contradiction table stores the contradiction conditions between the contents of the configuration items. For example, when the target of the question is {ruled line}, there is no {insert} as the operation of the word processor, so {insert} is described as the content of the operation that contradicts the target content {ruled line}. Similarly, as the content of the operation inconsistent with the target content {underline}, {insert,
Copy, move} is described.

By referring to such an item contradiction table, the contradiction between the item contents divided for each constituent item of the input question can be examined.

Further, in the candidate item correspondence table shown in FIG. 3B, the information of the constituent items necessary to compose the question is stored in correspondence with, for example, the type of [description content]. That is, when the type of [description content] is {method}, each information of [target] and [action] is required to compose the question, and the type of [description content] is {concept}. If yes, you need Target information to configure that question, and Behavior
It is described that the information of is unnecessary.

With reference to such a candidate item correspondence table, it is checked whether or not there are omissions in the configuration items of the input question, that is, whether or not all the information of the configuration items necessary to compose the question are given. To be

And if there are no omissions or inconsistencies in the entered questions,
According to the contents of each constituent item of the question, for example, the answer search data shown in FIG. 4A is searched (step d), and the question is identified by the pointer attached to the question (set of contents of constituent item). For example, an answer as shown in FIG. 7B is presented (step e).

In addition, the answer to the question is not only in text format,
It may be given in the form of animation, voice, or the like.

Here, the analysis processing for omission or contradiction of the questions will be described in more detail.

Now, for example, it is assumed that the item candidates are presented for each component item of the question as shown in FIG. 5, the item candidates are selected and designated, and the question is input.

For example, in the example shown in FIG. 5, {document} {ruled line} {underline} ... is shown as a candidate for the configuration item [target], and {create} {delete} {move} {copy as a candidate for the configuration item [action]. }, And {method} {attention} {situation} ... as candidates for the configuration item [description content]. It should be noted that each of these candidates is obtained, for example, by referring to an item candidate table that stores candidates that the item content can take for each configuration item.

In the example shown in FIG. 5A, candidates a1, b2, and c1 are selected, and "I want to know how to delete a sentence." Is displayed in the example shown in FIG. 5 (b).
The candidates 2, b1 are selected and "I want to know how to create ruled lines. ]
The question has been input.

Such as "I want to know how to delete sentences. The question analysis unit 12 divides the question into constituent item units like “Sentence / Delete / Method (I would like to know)” [Target] [Operation] [Explanation content] and omits the content. Whether or not there is a contradiction is analyzed with reference to the inter-candidate contradiction table and the candidate item correspondence table.

Similarly, "I want to know how to create ruled lines. The question analysis unit 12 divides the question into units of configuration items, such as ruled line / creation /-(I want to know) [Target] [Operation] [Description], and omits the content. Whether or not there is a contradiction is analyzed with reference to the inter-candidate contradiction table and the candidate item correspondence table.

Specifically, first, it is checked whether or not the information of [description content] is input. If the input question does not include the [Explanation content] configuration item, it is determined that this is omitted. When the information of [Description content] is input, the candidate item correspondence table shown in FIG. 3B is checked according to the content. Then, it is checked whether or not the information of the configuration item defined in the candidate item correspondence table is input, and if the information of the required configuration item is not input, it is determined that this is omitted.

For example, if [Description] is {method}, [Target]
You can see that and the Action information is needed to construct the question. Therefore, when each information of [Target] and [Operation] is not input, it is determined that this is omitted.

Then, next, with reference to the candidate conflict table shown in FIG. 3 (a), the conflict between the contents entered for each component is checked. If the relationship between the contents is described in the above contradiction table, it is determined that there is a contradiction.

The question "I would like to know how to delete sentences" mentioned above by the judgment processing for such omissions and inconsistencies. Is determined to have no omissions or contradictions, and the question "I want to know how to create ruled lines." ”Is omitted in the constituent items of the [description content], and it is determined that there is no contradiction.

Then, the entered question is analyzed for omission or inconsistency, and if judged, the state analysis unit 14 is activated and the state of the applied processing function unit 13 at that time is checked (step f). That is, it is possible to check what kind of applied processing the processing control device 2 was executing. As a result, for example, the status of error occurrence is internally requested.

Then, the item candidate generation unit 15 refers to the status information of the application processing function unit 13 and the item candidate table as a display example in FIG. 5 to refer to the item candidate list / data buffer.
For example, an item candidate list as shown in FIG. 6A is created in 22 (step g). The item candidate list is realized by providing score entry columns corresponding to the respective contents of the item candidate table. And "I want to know how to make ruled lines. ] Question is input, and at that time, the state analysis unit
When it is indicated from 14 that the application processing function unit 13 is handling ruled lines, the item candidate list is created as follows.

First, since the state analysis unit 14 indicates that the applied processing function unit 13 is handling ruled lines, a score higher than {ruled lines} stored in the item candidate table data buffer 21 is high, for example, 10 points. give. Then, the [operation] that does not conflict with the content of {ruled line} is checked from the conflict table, and a low score is given to the item candidates that do not conflict. Specifically, since the only content of the [operation] that contradicts the {ruled line} is {insert}, 5 points are given to {create} {move} {delete} etc., respectively.

After that, the question item data buffer 20 is referred to and the contents of the configuration items stored therein are obtained. in this case,
Since the information of {ruled line} and {created} is stored respectively, a high score, for example, 20 points is given to the scored items of {ruled line} and {created} in the item candidate list. As a result, the score item of {ruled line} stores 30 points including the previous score.

At the same time, since not only the above {ruled line} but also {document} etc. exist in the contents of the constituent items that can be {created}, connect to the above {ruled line} in the item of [Target] according to the above conflict table. Give a low score for each possible candidate.

Subsequently, the [explanatory content], which requires the respective contents of [object] and [operation], is checked from the configuration item correspondence table, and points are added to the respective configuration item candidates for which the combination is established.

By performing such point addition processing, for example, the sixth
As shown in FIG. 7A, a score will be given to each candidate. At this time, candidates that are given a score of a certain value or less may be omitted from the item candidate list because there is almost no possibility of forming a question.

After that, the question forming unit 16 creates a question without omission or contradiction as follows according to the item candidate list.

First, from the item candidate list above, select item 1 of the [Description content] item.
Extract one candidate X (step h). The extraction of the candidate X is performed, for example, in descending order of score. Then, with respect to the candidate X of the [description content], the candidate item correspondence table is referred to, and it is determined whether or not the [target] is required (step i). When [target] is required, one candidate Y of the item of [target] is extracted from the item candidate list (step j). The extraction of the candidate Y is performed in the same manner, for example, from the highest score.

The candidate Y of this [object] is also referred to the above-mentioned weather item correspondence table, and it is judged whether or not [action] is required (step k). If it is necessary, the candidate Z of [action] is determined.
Are extracted one by one in order from the highest score (step l). In this case, whether or not the extracted [action] candidate Z is inconsistent with the [target] candidate Y is checked by referring to the inconsistency table (step m). Repeat the extraction.

Through such a series of processing, a combination of constituent item contents (candidates X, Y, Z) without any contradiction or omission, that is, a question candidate is obtained from the item candidate list, and is given to each of the candidates X, Y, Z. The total score is calculated and registered (step h).

The above process is repeated for all combinations of the candidates X, Y, Z in the item candidate list (step o).

After that, the combination of the candidates X, Y, Z having the highest score is obtained from the combination of the above candidates X, Y, Z and the total score thereof (step p), and this is the original question intended by the input question. Suppose there is. Then, the answer is retrieved for the question generated in this way (step d), and the obtained answer is output (step e).

According to such a processing procedure, even if the entered question is incomplete, that is, "I want to know the ruled line creation. [Explanation content]
Even if the content of is not input, it is effectively inferred that the content is {method}. As a result, taking intent of the entered question, "I want to know how to create ruled lines. That question will be inferred and generated correctly. Then, it becomes possible to effectively obtain the answer to the question.

The present invention is not limited to the above embodiment. For example, the number and types of configuration items that make up a question are
It may be determined according to the type of question expected to be input. Instead of calculating the score for all combinations of candidates in the item candidate list and adopting the candidate combination that takes the maximum value as the question, ask the combination of item candidates that was initially obtained without contradiction. May be adopted as. Alternatively, a plurality of inferred questions may be presented to the user, and an answer may be asked for a question selected and designated from the questions. In addition, the present invention can be variously modified and implemented without departing from the scope of the invention.

[Effects of the Invention] As described above, according to the present invention, even when an omission or a contradiction is included in the input question content, a question that takes intent of the question is estimated, and the estimated question is estimated. You can ask for the answer. Therefore, it is possible to easily and effectively perform the process operation in the question form, and it is possible to effectively operate the information processing device, which is a great practical effect.

[Brief description of drawings]

FIG. 1 shows an embodiment system of the present invention. FIG. 1 is a schematic configuration diagram of a question processing unit of a Japanese word processor configured by adopting the embodiment system, and FIG. 2 is a processing procedure in the embodiment system. FIG. 3 (a) is a diagram showing a configuration example of a candidate contradiction table, FIG. 3 (b) is a diagram showing a configuration example of a candidate item correspondence table, and FIGS. 4 (a) and 4 (b). ) Is a diagram showing an example of a storage structure of answer data to a question, FIGS. 5 (a) and (b) are diagrams showing a configuration example of an item candidate table and a question input example, and FIGS. 6 (a) and (b) are items. It is a figure which shows the structural example of a candidate list. 1 ... Input device, 2 ... Processing control device, 3 ... Output device, 4 ... External storage device, 11 ... Input unit, 12 ... Question analysis unit, 13 ... Applied processing function unit, 14 ... State analysis part, 15 ...
… Item candidate generation unit, 16 …… Question composition unit, 17 …… Answer search unit, 18 …… Output unit, 19 …… Data buffer unit, 20 …… Question item data buffer, 21 …… Item candidate table data buffer, 22 …… Item candidate list / data buffer.

Claims (1)

[Claims] 1. An extraction means for dividing input question information into constituent item units to extract the contents of each constituent item, and a contradiction table table storing contradiction conditions corresponding to the contents between constituent items. By referring to the contradiction determining means for checking the contradiction between the contents extracted in units of the configuration items and the candidate item correspondence table table storing the configuration items necessary to compose the question about the contents of the specific configuration item. Omission determination means for checking omission of configuration items of question information, state detection means for detecting the processing state of the device, and item content for each configuration item when it is determined that the question information is inconsistent or omitted The item candidate list table storing the candidates is generated to generate the item candidate list, and the candidate contents of each constituent item of the item candidate list are used as the extracted contents of the constituent items of the device status information and question information. A candidate list generating means for performing weighting according to the above, and sequentially extracting candidate contents of each high-priority constituent item according to the weight information given to this item candidate list, and combining candidate contents of these respective constituent items to generate a question candidate. By providing the question candidate generating means and the control means for making the final contradiction determination and omission determination based on the question candidate, it is possible to estimate the original question for the question with contradiction or omission. Characteristic question estimation device.