JPH0786857B2 - Conversation system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention converts an operation instruction given using a natural language into a command of an information processing system by referring to a knowledge base, and outputs the information. The present invention relates to a conversation system that functions as an interface for inputting into a processing system.

(Prior Art) In recent years, various information processing systems using computer systems have been developed and widely used in various industrial fields. The operation of such an information processing system is controlled by the computer system, and an operator who is a user of the information processing system inputs a command for instructing the operation to the computer system to obtain the desired information processing system. The action to do is to be performed.

Usually, this kind of operation command is
Input is performed using a pointing device such as a light pen or using a keyboard. When using a pointing device, generally, a plurality of operation commands to be selected are displayed on the screen of a display device such as a CRT,
An operation command is input by selectively instructing a desired command from among them.

By the way, in recent years, the information to be processed tends to be more and more complicated, and the types of commands used in this type of information processing system are also increasing. Therefore, it has been attempted to handle many commands by increasing the number of commands displayed on the screen of the display device at one time or by displaying a plurality of commands hierarchically. However, this has a problem that it becomes necessary to search for a desired command from a large number of commands, which increases the processing load on the operator.

On the other hand, in the case of inputting a command using the keyboard, it is necessary to input the command according to the format because a specific format is determined for each command in advance. The format of this command is generally different for each system, so for operators using multiple different systems, despite similar commands,
It is necessary to input a command in a command format according to the system in use, and there is an adverse effect such as having to remember each of these command formats.

Therefore, recently, it has been attempted to give an operation instruction to the information processing system by using a natural language instead of the command input using the pointing device, the keyboard or the like described above. That is, an interface for converting an operation instruction given using natural language into a command used in an information processing system by referring to a knowledge base in which knowledge about the operation target is stored in advance, and giving this command to the information processing system. It has been attempted to give an operation instruction to an information processing system by using a natural language by using a conversation system that performs a function.

In such a conversation system using natural language input, it is possible to handle various expressions in natural language and ambiguous expressions and to give an operation instruction to the information processing system. In other words, the operation instruction (natural language sentence) input using natural language is automatically converted into the command format of the information processing system and given to the above information processing system, so that the operator needs to remember the command format. It becomes possible to significantly reduce the burden on the operator.

However, the meaning and content of the natural language sentence often changes depending on the situation of the conversation, and it is necessary to ask the interpretation of the natural language sentence according to the situation at that time. For example, in a plant system, operation instructions composed of very similar natural language sentences such as "turn valve A slightly right" and "turn valve B slightly right" may be given. Both of these natural language sentences use a degree expression called "a little", but the degree varies depending on the operation target.

For example, if the valve A is rotated through 360 °, the degree of rotation indicated by [a little] described above is, for example,
If it is about 45 ° and the valve B rotates only 45 °, the degree of rotation indicated by [a little] is, for example,
It will be about 10 °. In this way, the degree of operation indicated by the natural language sentence varies depending on the topic (here, the operation target) at that time. Therefore, when converting a natural language sentence into an operation command, it is necessary to fully consider the degree of expression included in the natural language sentence.

(Problems to be Solved by the Invention) When an operation instruction given in natural language is converted into an operation command of an information processing system and input, the degree of expression included in the natural language sentence is sufficiently considered. However, if the interpretation of the degree expression is incorrect, a wrong operation command will be input.

The present invention has been made in consideration of such circumstances, and an object thereof is, for example, in a difference in the degree of operation indicated by a degree expression in operation instructions input using natural language, It is possible to accurately judge the meaning content of operation instructions shown in natural language and convert commands,
It is to provide a conversation system that enables a natural conversation with an information processing system based on natural language.

[Means for Solving the Problems] (Means for Solving the Problems) The present invention includes a knowledge base in which knowledge about an operation target field of an information processing system is stored in advance, and is provided to the information processing system by using a natural language. A natural language sentence indicating an operation instruction is analyzed by converting a given operation instruction into a command of the information processing system by referring to the knowledge base, and giving the converted command to the information processing system. Then, a means for estimating a topic at each time point of the conversation is provided, and the knowledge of the knowledge base to be referred to when converting the natural language sentence into a command according to the estimated topic is localized, and according to the localized knowledge, By limiting the range of expression included in the natural language sentence, the command is converted into a final command. It is.

In other words, according to the topic of conversation, for example, the degree of operation indicated by the degree expression is specified by localizing knowledge, so that the operation instruction given in the natural language sentence can be appropriately converted into a command. It is a feature.

(Operation) According to the present invention, when an operation instruction (natural language sentence) is given to an information processing system using natural language, a topic is estimated by an analysis process for the natural language sentence, and the topic is estimated according to the estimated topic. A knowledge base in which knowledge about an operation target field of the information processing system is stored in advance is localized. Then, according to this localized knowledge, the natural language sentence is converted into a command.

As a result, even if the given natural language sentence includes an ambiguous operation expression such as a degree expression, the operation indicated in the above degree expression by a topic expression that identifies the operation target. Since it is possible to convert the natural language sentence into a command used in the information processing system after specifying the degree of, it is possible to accurately convert the content indicated by the operation instruction shown in the natural language into the command. It becomes possible to give it to the information processing system, and it becomes possible to realize a conversation with high naturalness.

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

FIG. 1 is a block diagram showing a schematic processing function of the embodiment system.

This system is a morphological analyzer that analyzes the operation instructions (natural language sentences) given to the information processing system in natural language.
1, a syntactic analysis unit 2 and a semantic analysis unit 3 are provided. The topic estimation unit 4 estimates the topic at each time point of the conversation indicated by the natural language sentence based on the analysis results by the analysis units 1, 2, and 3. Further, the command conversion unit 5 preliminarily stores in the knowledge base 6 according to the analysis result of the natural language sentence by each of the analysis units 1, 2 and 3 and the information of the topic at each time point of the conversation obtained by the topic estimation unit 4. While localizing the stored knowledge about the operation target, the localized knowledge is referred to, the natural language sentence is converted into a command used in the information processing system, and the command is output to the information processing system.

Next, the outline of the process of deriving the meaning content of the operation instruction from the conversation input as a natural language sentence in the embodiment system configured as described above will be described.

The gist of the present invention is to determine, from an operation instruction (natural language sentence) given using a natural language, a scale of a degree expression according to the topic and a semantic content of the operation instruction to determine an operation intended by the natural language sentence. It is to convert the instruction into a command used in the information processing system properly, and not to the processing of natural language itself. Therefore, for the syntactic / semantic analysis processing of the natural language sentence indicating the operation instruction, various conventionally used language analysis methods may be appropriately used.

For example, a morphological analysis as a syntactic analysis for a given natural language sentence, that is, a process of dividing a natural language sentence into words, a syntactic analysis for the natural language sentence, that is, a dependency between clauses forming the natural language sentence. For the processing of analyzing the structure, it is sufficient to use the processing method shown in, for example, “Basic technology of natural language processing; Hirogo Nomura, edited by Institute of Electronics, Information and Communication Engineers”. Further, as the semantic analysis processing of the natural language sentence, for example, a processing method as shown in “Japanese Information Processing; Makoto Nagao, edited by Institute of Electronics and Communication Engineers” may be used.

Further, here, the semantic expression in the natural language sentence is represented by, for example, a case frame expression. Therefore, the case frame is composed of a predicate corresponding to the verb of the input sentence and a case slot corresponding to the case particle associated with the verb, and a noun preceding the corresponding case particle is set as a filler in the case slot. Therefore, in this embodiment, a given natural language sentence is subjected to morphological analysis, syntactic analysis, and semantic analysis to be converted into a case frame, and then a topic is obtained based on the case frame.

Incidentally, what kind of operation control is performed according to this command after changing the meaning expression represented by the case frame expression to a specific command format is also not the gist of the present invention, and therefore its explanation is omitted here. .

Now, as an operation instruction to the information processing system, for example, a natural language sentence such as "turn the valve A slightly to the right" as shown in FIG. 2 is given, and this natural language sentence is subjected to morphological analysis / syntactic analysis / semantic analysis. Then, it is assumed that the case frame as shown in FIG. 3 is obtained. Then, in the topic estimation process, the topic of the natural language sentence is estimated as follows according to the case frame.

In the case of Japanese, generally, the topic is presented by being followed by a specific topic presentation expression such as a sub particle [ha] or [about]. In addition, when a topic is not presented by such an auxiliary particle or topic presentation expression, a word that occupies a major case such as [objective] or [nominative] that constitutes the sentence becomes a topic. Further, in the case where the conversation continues, the topic is not always specified every time the natural language sentence is input. If the topic is not clearly indicated in the input sentence, the topic immediately before that is generally continued.

Therefore, in the case of the natural language sentence shown in FIG. 2, if the case frame is obtained as shown in FIG.
Since [Valve A] is explicitly indicated in the sentence, this [Valve A] is taken out as a topic here. Then, the analysis processing determines that the operation instruction for the topic [valve A] is [turn to the right] and the degree of the operation instruction is [a little].

On the other hand, the knowledge about the operation target field indicating the operation instruction and its degree is registered in the knowledge base 6 in advance. The command converting unit 5 uses the knowledge registered in advance in the knowledge base 6 as described above, and performs the processing corresponding to the topic obtained as described above.

That is, the knowledge registered in the knowledge base 6 is composed of a plurality of slots for each knowledge, for example, a part of which is shown in FIG. For example, knowledge about the operation instruction of [turn] has three slots of [target], [direction], and [degree].

Further, each knowledge in the knowledge base 6 is hierarchically structured. Specifically, in this example, the knowledge [little] is positioned in the lower hierarchy of the knowledge [degree], and the knowledge [valve A] and the knowledge [valve B] are positioned in the lower hierarchy of the knowledge [valve].

Further, in this knowledge base 6, the values assigned to the slots are classified and set for each object that can be a topic of conversation. For example, as [immediate] immediate slots, [valve A] and [valve B] are classified and set according to the topic, and each of these immediate slots [valve A] and [valve B] is [45 °]. And the value of [10 °] are respectively substituted.

By referring to the knowledge base 6 as described above, for example, when the topic is [valve A], an immediate slot of [45 °] is obtained as the degree corresponding to the degree expression [little] there. , If the topic is [bulb B], the degree corresponding to the degree expression [little] is [10
°] is obtained. Therefore, in the example of the input sentence shown in FIG. 2 described above, [45 °] is required as the degree of operation for turning the valve A.

After that, the case frame expression described above is converted into a command. The format of the command to be converted generally differs depending on the information processing system to be operated, but for example, the fifth
As shown in the figure, when the command format is set as "action (target / value, direction / value, degree / value)", the target is [Valve A], and the direction is [Right]. , And the degree is set to the immediate value [45 °] corresponding to the above-mentioned topic [valve A], and the command “operation (target / valve A, direction / right, degree / 45 °)” is obtained.

As described above, according to the present invention in which the case frame expression of the input sentence is converted into the command by referring to the knowledge localized according to the topic as described above, even when the topic is changed, the topic is correctly handled. It can be converted into a command. Moreover, the command can be correctly converted into a command that matches the meaning and content desired by the operator. Therefore, it is possible to easily give an operation instruction to the information processing system by using a natural language, and it is possible to effectively eliminate the conventional inconvenience that the command format must be stored for each different system. .

The topic is clearly presented by a particle [ha] or a specific expression [about] in the natural language sentence, and if the topic is not presented in the natural language sentence at each time point, it has been used until then. The topic of the conversation that has been continued is used as it is. Therefore, since it is not necessary to input and clearly indicate the subject matter to be discussed each time, it is possible to reduce the trouble of inputting the natural language sentence.

The above processing procedure will now be described in some detail.

FIG. 6 shows a knowledge storage structure in the knowledge base 6. The knowledge registered in this knowledge base 6 is composed of a plurality of slots. For example, each of these slots is composed of [slot name], [slot value], and [viewpoint].

The [slot name] is a name for distinguishing each slot, and the [slot value] is a value set for the slot. [Slot value] for one slot is described from a plurality of viewpoints.

If you need to describe your knowledge from multiple perspectives,
A viewpoint is set for each slot value. This means that a certain slot value can be set differently depending on the viewpoint, and in the present invention, a thing that can become a topic is set for this viewpoint.

Therefore, the above knowledge [a little] is specifically expressed as shown in FIG. From the knowledge structure shown in FIG. 7,
It is shown that there is a degree as a higher hierarchy (classification) of knowledge [a little], and its immediate values are [45 °] and [10 °]. However, the immediate value [45 °] represents a value from the viewpoint of knowledge [valve A], and the immediate value [10 °] represents a value from the viewpoint of knowledge [valve B]. Furthermore, this 7th
In the figure, the view [nv] in [super C], which represents the higher-ranked slots, represents the value in all views.

Further, FIG. 8 shows an example of expression of knowledge [turn] in the knowledge base 6. This knowledge [turn] is [target],
It has slots of [direction], [degree], and [command format], and each slot value is variable [$ x], [$ y],
It is associated with [$ z].

Then, when the command is fetched by referring to the knowledge base 6, the slot value of the above [command format] slot is fetched. At this time, the variables in the command format slot are associated with the slot values of other slots.

Now, the topic estimation unit 4 extracts the topic in the natural language sentence according to the processing procedure shown in FIG. That is, the topic estimation unit 4 estimates a topic based on the input sentence (natural language sentence) and its morphological analysis result / syntactic analysis result / semantic analysis result.

Specifically, first, it is determined whether a topic presentation expression is present in the natural language sentence (step a1). Then, if the topic presentation expression exists, the word preceding the topic presentation expression is extracted as the topic (step a2).

On the other hand, if the topic presentation expression does not exist, it is then determined whether or not the sub-particle [ha] or [mo] exists in the natural language sentence (step a3). If the sub particle [ha] or [mo] is present, the word preceding the sub particle [ha] or [mo] is extracted as the topic at that time (step a4).

If the auxiliary particle [ha] or [mo] does not exist, it is then determined whether or not the target case exists in the natural language sentence (step a5). When the existence of the target case is detected, the target case is extracted as a topic (step a6).

If none of them exist, the previous (up to now) topic is inherited as it is (step a7).

After the topic is extracted as described above, the command conversion unit 5 next executes the command conversion processing according to the processing procedure shown in FIG. 10, for example.

The command conversion unit 5 receives the case frame expression of the natural language sentence described above as an input, and first extracts the knowledge corresponding to the verb that is the core of the case frame expression from the knowledge base 6, and defines this knowledge as knowledge A. Yes (Step b
1). Then, the topic estimated by the topic estimation unit 4 is substituted into the variable [TOPIC] (step b2).

Next, the following processing is repeated for all case slots in the case frame expression.

That is, first, the knowledge corresponding to the slot value of each case slot is taken out from the knowledge base 6, and this knowledge is set as the knowledge B (step b3). Next, the value of the immediate slot in the knowledge B is taken out (step b4). At this time, the value of the viewpoint equal to the variable [TOPIC] is taken out. The variable [TOP
If there is no view equivalent to [IC], the value of the immediate slot is used as the slot value of [nv].

Next, this value is substituted as the value of the slot of the knowledge A corresponding to the case slot of the case frame expression that is the current processing target (step b5). Also at this time, the variable [TOPIC]
If there is a view equal to, substitute it for the corresponding slot value. Further, as the knowledge slot value, for example, when the same variable name is used over a plurality of slots, the same value is set to the location of the same variable name.

The above process is repeated until values are obtained for all case slots in the case frame representation (step b
6). As a result, the content specified by the input natural language sentence is set to the slot value, and at the same time, the value is set to the variable in the command format slot.

Finally, the slot value in the command format slot is taken out and output as a command. At this time, it is assumed that the slot whose viewpoint is the one indicated by the variable [TOPIC] is taken out, and the slot value of the viewpoint [nv] is taken out as a command only when the viewpoint that is the variable [TOPIC] does not exist (step b7). ).

Specifically, for example, when a natural language sentence as shown in FIG. 2 is given, a case frame representation as shown in FIG. 3 is obtained as a result of the semantic analysis. Therefore, in the case of this example, in the topic presenting unit 4 described above, the topic presenting expression and the auxiliary particles [wa] and [mo] do not exist, but since the target case [valve A] exists, this target case [ Estimate valve A] as a topic.

On the other hand, the command conversion unit 5 receives the case frame expression shown in FIG. 3 as an input, and first, the knowledge [turn] shown in FIG. 8 corresponding to the core verb [turn].
From the knowledge base 6. However, since the topic is sought as [valve A] as described above, the [valve A] is set to the variable [TOPIC].

Since the case slots in the case frame expression currently being processed are [target case], [direction case], and [degree case], the slot value of each of these case slots is set to the corresponding slot. . Specifically, first, the target slot value [valve A] is taken out from the knowledge base 6. At this time, since the knowledge [valve A] does not have an immediate slot whose viewpoint is equal to the variable [TOPIC], the above [valve A] is directly substituted into the target slot. Similarly, for the direction slot, [right] is set to the slot value.

On the other hand, the slot value [a little] of the grade is set to the knowledge base 6
When taken out further, there is an immediate value whose view is [valve A] as shown in FIG. Therefore, in this case, the same topic [45 °] as the current topic [valve A] indicated by the variable [TOPIC] is extracted. Then, this value is set in the knowledge [turn] grade slot. At this time, since each variable in the slot value of the command format slot corresponds to the slot value of another slot, it is set at the same time when setting each slot value. Therefore, in the case of this example, the desired command can be obtained by extracting the value of the command format slot, and the natural language sentence input as described above is suitable for the command according to its semantic content. It is possible to convert accurately.

The present invention is not limited to the above embodiment. In this embodiment, when the slot value corresponding to the topic is taken out, it is explained that the same view as the topic is taken out. However, using the hierarchical relation expressed by the knowledge base 6,
Of course, it is also possible to take out a more semantically similar view. Also, the flow of topic extraction processing and the processing procedure of command conversion can be variously modified. 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 in detail above, according to the present invention, an operation instruction input using a natural language sentence is appropriately extracted from a localized topic of a portion adapted to the topic, and the natural language sentence is extracted. It can be properly converted into a command having the meaning shown in. Therefore, even an input sentence (natural language sentence) in a conversation in which the topic changes can be correctly converted into a command and the burden on the operator can be significantly reduced. A great effect is practically exhibited.

[Brief description of drawings]

FIG. 1 shows a conversation system according to an embodiment of the present invention. FIG. 1 is a diagram showing a schematic configuration of the embodiment system, and FIG.
The figure shows an example of an input sentence given using natural language,
FIG. 3 is a diagram showing a case frame expression for the input sentence of FIG. 1, and FIG. 4 is a diagram showing a knowledge structure in a knowledge base used in the embodiment. Further, FIG. 5 is a diagram showing an example of a virtual command format for explaining the present invention, FIG. 6 is a diagram showing a knowledge representation format in the knowledge base, and FIGS. 7 and 8 are knowledge representations, respectively. Is a diagram showing a concrete example of the above, FIG. 9 is a diagram showing a flow of processing in the topic estimation unit, and FIG. 10 is a diagram showing a flow of processing in the command conversion unit. 1 ... Morphological analysis unit, 2 ... Syntax analysis unit, 3 ... Semantic analysis unit,
4 ... Topic estimation unit, 5 ... Command conversion unit, 6 ... Knowledge base.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-63-106042 (JP, A) JP-A-2-207303 (JP, A) JP-A-2-232728 (JP, A) JP-A-1- 233559 (JP, A)

Claims (1)

[Claims] 1. A processing system comprising a knowledge base in which knowledge about an operation target field operated by an information system is stored in advance, and an operation instruction given to the information processing system using a natural language is referred to the knowledge base. In the conversation system for converting the command into the above information processing system and giving the converted command to the information processing system, a means for estimating a topic at each point of conversation by analyzing a natural language sentence indicating an operation instruction, and Means for localizing the knowledge of the knowledge base referred to when converting the natural language sentence into a command, and limiting the range of the degree expression included in the natural language sentence according to the localized knowledge, Conversation system comprising means for converting into a general command.