JP2697775B2 - Conversation system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a conversation system capable of causing a computer system to perform a desired operation by giving a natural language instruction to the computer system.

[0002]

2. Description of the Related Art Recently, a computer system has been developed as an information processing system for processing various types of information, and is widely used in various industrial fields.

In such an information processing system, the computer system controls the operation for information processing, and when an operator as a user inputs a command or the like to the computer system, a desired operation is performed on the information processing system. Is performed.

In this case, a command input by the operator is usually input by operating a pointing device such as a tablet, a mouse, and a light pen, or an operation of a keyboard.

For example, when a pointing device is used for command input, a plurality of commands to be selected are
It is displayed on a display screen such as an RT, and the operator sequentially proceeds by selecting a desired command from the commands displayed on the display screen.

By the way, the information to be processed has recently become more and more complicated, and accordingly, the types of input commands to be used tend to increase. For this reason, various measures have been considered to make it easier to handle many commands by adopting various methods such as increasing the number of commands displayed on the display screen at one time or displaying them hierarchically. I have.

[0007] However, each of these countermeasures seeks a desired command from many commands, resulting in imposing a burden on the operator on the operation.

On the other hand, when a keyboard is used for command input, a specific format is determined in advance for each command, and it is required to input a command according to this format. Since this format is usually different for each system, an operator using a different system stores a different format for each system in spite of a similar command. The operation becomes extremely complicated, such as the necessity of putting it, and it is easy to cause erroneous operation.

For these reasons, a given information processing system has recently been entered by inputting a natural language in place of command input using a pointing device, a keyboard, or the like, or command input following a fixed predetermined format. Has been put to practical use.

A so-called conversation system using a natural language as an input means is characterized in that it can handle various expressions and ambiguous expressions in a natural language. In other words, a sentence entered by an operator in a natural language is automatically converted into a given command format by a computer and executed, so that the operator can remember command formats that are not originally involved in information processing activities. This has the effect of eliminating the inconvenience of having to do so. However, on the other hand, the meaning of a sentence input in a natural language may change depending on the situation of a conversation that has been performed so far, and it is necessary to ask for an interpretation according to the situation. In other words, this means that a sentence entered in a natural language and a given command are pre-set one-to-one.
Means that they cannot be associated.

For example, in a CAD system in which the layout design of a physical object is designed by a computer, the following cases can be considered. Now, assuming that a sentence input in natural language is (a) "move the desk to the left." (B) "A chair is also input." This means that the "desk" existing in the display screen is moved to the left. Since the sentence of (b) was input immediately after the sentence of (a), the "chair" existing in the display screen was also input. "desk"
Means to move to the left with However, if there are two or more objects that can be designated by the noun “chair” in the display screen, it is necessary to determine which of these “chairs”.

Therefore, when there are a plurality of objects which can be indicated by a noun, an operator is inquired as to which "chair" it is, and an object to which the operator indicates in response to the inquiry is given. It is considered to be specified. But,
In this method, an inquiry is made every time there are a plurality of objects that can be instructed, so that the operator must input many sentences or operate the pointing device to complete the desired operation. However, there is a disadvantage that the operation is still complicated.

On the other hand, in the prior art, a document (“The Repre
Sentation and Use of Fuku
s in a System for Underst
and Dialogs ", BJ Gros
z, ProcofIJCAI "77), a method of determining one object from a plurality of objects using knowledge of the objects incorporated in a knowledge base in advance has been considered. In other words, in this method, in the knowledge base, objects that can be specified simultaneously or immediately before and after are stored as “units of objects”, and, for example, an object a is specified by a noun A in a certain sentence, and If noun B is used in the sentence
Noun B from the object in the "group of objects" to which
Is searched for an object designated by the user.

However, in such a method, a "group of objects" must be defined in the knowledge base in advance, which means that when a knowledge base is constructed by a human, At the same time, or what can be instructed immediately before and after, it is necessary to consider in advance, and there is a problem that construction of a knowledge base becomes very difficult.

[0015]

As described above, in a so-called conversation system in which a conventional natural language sentence is input, when there are a plurality of targets that can be designated, it takes a great deal of time for the operator to specify the desired target. In the case of determining one object from a plurality of objects by using the knowledge of the objects incorporated in the knowledge base, In creating a knowledge base, all objects that can be instructed at the same time or immediately before and after must be taken into consideration, and there is a problem that construction of the knowledge base becomes very difficult.

The present invention has been made in view of the above circumstances, and it is an object of the present invention to provide a conversation system capable of determining a desired object by a simple operation even when there are a plurality of objects that can be designated. .

[0017]

SUMMARY OF THE INVENTION The present invention provides a conversation system in which a device can be operated by giving an instruction in a natural language, and an input means for inputting a natural language sentence, and the input sentence of the input means is converted into a case frame expression. Case frame expression conversion processing means for performing conversion processing, conversation history storage means for sequentially storing the case frame expressions converted by the case frame expression conversion processing means as a conversation history, and a plurality of objects on the display screen. A distance information recording unit for storing relative distance information; and a case in which an operation word is present in the case frame expression of the input sentence. In the case, the most recently input case frame expression including an object belonging to a set of objects having a sibling relationship with the object specified in the input sentence is referred to as the conversation history case. While specifying with reference to the case frame expression stored in the means, obtains the object specified in the specified case frame expression, based on the relative distance information stored in the distance information storage means. An expression interpolating means for determining an instance closest in distance to the object obtained by the above processing as an object specified in the case frame expression of the input sentence; and an output interpolated by the expression interpolating means. And command output means for controlling the display screen based on a command from the command conversion means.

[0018]

As a result, according to the present invention, when there are a plurality of target objects that can be pointed, identification of the target object is performed using the relative distance relationship between the target objects.
The desired object can be determined easily and automatically at any time, and since the operator can give an operation instruction only in a natural language, the command format must be stored for each different system. Can be excluded.

[0019]

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

FIG. 1 shows a schematic configuration of a conversation system according to the embodiment. In the figure, reference numeral 1 denotes an input unit for enabling input of a natural language sentence. The input unit 1 is connected to a keyboard 2, a mouse 3, and the like.

The input unit 1 is connected to a morphological analysis unit 4, a syntax analysis unit 5, and a semantic analysis unit 6 in this order. The morphological analysis unit 4, syntax analysis unit 5, and semantic analysis unit 6 convert a natural language sentence input from the input unit 1 into a predicate corresponding to a verb and a case slot corresponding to a case particle associated with the verb. This is for converting to a case frame expression. Here, a case slot is set as a filler of a case slot in which a noun preceding the corresponding case particle is applied. Also,
For the syntactic and semantic analysis of natural language sentences, a conventionally used method is diverted. For example, a morphological analysis as a syntactic analysis in the morphological analysis unit 4 (processing of dividing an input sentence for each word) The parsing (processing for analyzing the dependency structure between clauses) by the parsing unit 5 is performed as described in the literature (“Basic Technology of Natural Language Processing”, Hirosato Nomura, edited by the Institute of Electronics, Information and Communication Engineers). The semantic analysis in the semantic analysis unit 6 uses processing as described in the literature (“Japanese Information Processing”, Makoto Nagao, edited by the Institute of Electronics, Communication and Communication Engineers).

The output converted to the case frame expression by the morphological analysis unit 4, syntax analysis unit 5, and semantic analysis unit 6 is sent to the abbreviation expression interpolation unit 7.

The abbreviation expression interpolation unit 7 performs an interpolation process of information not specified in the input sentence with reference to the conversation history 9 and the contents of the relative distance storage 10. In this case, the abbreviation expression interpolating unit 7 executes the interpolation process by using the knowledge base 12.
They refer to the knowledge inside.

The output of the case frame expression interpolated by the abbreviation expression interpolation unit 7 is sent to the command conversion unit 8. The command conversion unit 8 performs a process of converting the output subjected to the interpolation processing into a command format of a target information processing system. In this case, the command conversion unit 8 refers to the knowledge in the knowledge base 12 when executing the conversion processing.

The output converted by the command conversion unit 8 is sent to the task execution unit 13. The task execution unit 13 actually moves an object on the display screen while referring to the image model 14, and here, a given information processing system is employed.

The image model 14 stores the absolute position, size, outline of the shape, and the like of the target object on the screen display at the time of executing the task, and changes the model according to the meaning of the input sentence. I have to. Further, the relative distance between the objects changed by the task execution is extracted by the relative distance extracting unit 11 one by one, and stored in the relative distance storage 10.
Is stored.

The output from the task execution unit 13 is
It is sent to the output unit 15 and displayed on the display screen of the display 16.

FIG. 2 shows an expression format of knowledge in the knowledge base 12. In this case, the knowledge in the knowledge base 12 is composed of a plurality of slots, and each slot is composed of a slot name and a slot value. That is, a combination of a plurality of slots is one piece of knowledge. Here, the slot name 1 is a name for distinguishing each slot, and the slot value is a value set for the slot.
FIG. 3A shows a frame representing a class “desk”, in which “desk” has an object as an upper class and has “desk 1” as an instance. Here, “superC” is a slot name for setting an upper class as a slot value, and “ins” is a slot for setting an instance as a slot value. FIG. 3B shows a frame representing the instance “desk 1”. In this case, “Desk 1” has “Desk” as a class, and indicates that “Table cloth 1” is subordinate. “Cls” is a slot for setting a class name to which the instance belongs, and “with” is a slot name for indicating a belonging relationship.

FIG. 4 shows an example of the expression “movement” as knowledge in the knowledge base 12. In this case, the target slot,
It has a direction slot, a degree slot, and a command format slot, and each slot value is a variable {x, {y,}
It is associated with z. That is, when the command is taken out, the slot value of the command format slot is taken out, which means that the variables in the command are respectively associated with the slot values of other slots.

Thus, the outline of the knowledge expression format in the knowledge base 12 can be represented as shown in FIGS. 13 (a) and 13 (b).

Next, the operation of the embodiment configured as described above will be described.

Now, when a natural language is input by the operator from the input unit 1, a sentence based on this natural language is
The morphological analysis unit 4, the syntax analysis unit 5, and the semantic analysis unit 6 perform morphological analysis, syntactic analysis, and semantic analysis, respectively, and are converted into case frame expressions.

From this state, the abbreviation expression interpolation unit 7 executes the interpolation processing of the input sentence.

FIG. 5 is a flowchart showing the processing in the abbreviation expression interpolation unit 7. In this case, the abbreviation interpolation unit 7
Will interpolate the omission information based on the case frame expression that is the result of the semantic analysis. First, it is determined whether or not a word about “operation” exists in the input case frame expression (step A1).

If it exists, the class of the object specified in the case frame expression to be processed is set to a (step A2), and the case frame expression to be processed is set to F (step A3). .

The object in the case frame expression most recently input with reference to the conversation history 9 is set to b (step A4). In this case, the conversation history 9 is shown in FIG.
As shown in (1), the analysis result of the input sentence of the operator input so far is stored. In this example, FIG.
1A shows a state immediately after the processing of the input sentence shown in FIG. 1A is completed, and stores a case frame expression corresponding to FIG. The pointer P points to the case frame expression of the immediately preceding sentence. Each time the conversation progresses, the case frame expression processed at each point in time is sequentially stored.
Move accordingly. Therefore, in this case, the object b in the case frame expression indicated by the pointer P is extracted.

Next, step A8 is executed. In step A8, an instance belonging to a is extracted from the knowledge base 12, and an instance closest to the object b is obtained by referring to the relative distance storage 10, and is set to c. In this case, the relative distance storage 10 stores the task execution unit 1
Relative distance extraction unit 1 for extracting the relative distance between the objects from 3
The relative distance between the centers of different objects is stored as shown in FIG. In the illustrated example, “desk 1” and “chair 1” are 100, and “desk 1” and “chair 2” are 6
00, “desk 1” and “chair 3” are 700, “chair 1” and “chair 2” are 650, and “chair 1” and “chair 3” are 75.
0, indicating that "chair 2" and "chair 3" are at a distance of 50. Here, one unit of the distance is a horizontal or vertical pixel on the display. The reason is that it is only necessary to obtain a relative magnitude relation. The display screen is updated each time a certain input sentence is processed and the conversation proceeds, but the relative distance at each time point is stored. The pointers B1 and E1 indicate the relative distances of the display screen immediately before the immediately preceding input sentence is processed. On the other hand, the pointers B2 and E2 indicate the display screen immediately after the immediately preceding input sentence is processed, respectively. It shows the relative distance of. Therefore, here
If there are a plurality of instances belonging to a, which one is the correct object is determined based on the relative distances indicated by the pointers B1 and E1, and this is set as the instance c closest to the object b.

Then, the object a of F is rewritten with c (step A9), and F is output to the command conversion section 8 as a result (step A10).

On the other hand, if it is determined in step A1 that there is no word for "operation" in the input case frame expression, step A5 is executed to interpolate the omission information. In step A5, the class of the object specified in the input is set to a, and by referring to the knowledge base 12, the classes of all the objects having a sibling relationship with this class a are found. Is A.

Next, the most recently input frame expression is checked with reference to the conversation history 9 to find a frame expression in which an object belonging to A is specified. This frame representation is F, and the object referred to there is b (step A6), and b in the frame representation F is a
(Step A7).

From this state, the process proceeds to step A8 in the same manner as described above, the instance belonging to a is extracted from the knowledge base 12, and the instance closest to the object b is obtained by referring to the relative distance storage 10, and this is obtained. It is assumed that c is set, and the target a of F is rewritten with c (step A9), and F is output to the command conversion unit 8 as a result (step A10).

FIG. 6 is a flowchart showing the processing in the command conversion section 8. The command conversion unit 8 converts the case frame expression in which the omitted information output from the omitted expression interpolation unit 7 has been interpolated into a command in the task execution unit 13 (information processing system). In this case, the knowledge corresponding to the verb that is the case in the case frame expression is extracted from the knowledge base 12, and this knowledge is set to D (step B1). Next, since the name of the knowledge in the knowledge base is stored in the slot value of the case slot, the knowledge E corresponding to the value of the case slot is extracted using this name as a clue (step B21), and the correspondence of the knowledge D is obtained. Set E to the slot value to be used. Such an operation is repeated for all case slots in the case frame expression (step B2). This allows
The values in the variables in the command format slot are set as appropriate. Finally, the slot value of the command format slot is taken out and output as a command (step B3).

Next, the above processing will be described based on an actual example.

For example, it is assumed that a state as shown in FIG. 10A is displayed on the display screen of the display 16, and from this state, a natural language sentence shown in FIG. Then, this sentence is subjected to morphological analysis, syntactic analysis, and semantic analysis by the morphological analysis unit 4, syntactic analysis unit 5, and semantic analysis unit 6, and as shown in FIG. It is output as a case frame expression composed of the corresponding predicate and the case slot corresponding to the case particle associated with the verb.

The operations of the morphological analysis unit 4, the syntax analysis unit 5, and the semantic analysis unit 6 are not the gist of the present invention, but
In addition, since it is a well-known technique as described above, its description is omitted.

Then, the semantic expression represented by the case frame expression is converted into a command format corresponding to a specific system, and this is subjected to actual execution processing in a given information processing system such as a CAD system.

First, the case frame expression is sent to the abbreviation expression interpolation unit 7. Then, in the above-described flowchart shown in FIG. 5, step A2 is executed first, and “desk” is set to a. Next, step A3 and step A
By executing the step 4, the case frame expression shown in FIG. On the other hand, since the input sentence at this time is the first one, nothing is set in the conversation history 9 and nothing is set in b.

Then, the process proceeds to step A8. in this case,
In the knowledge base 12, only "desk 1" has an instance belonging to the class a (= "desk"), so that "desk 1" is set in c. Step A9
Then, this “desk 1” is replaced with the target “desk” of the case frame expression F shown in FIG. 12A, and the result is output to the command conversion unit 8.

In the command conversion unit 8, in the flowchart shown in FIG. 5 described above, the case verb in the case frame expression at this time is “move”. ) Is taken out (step B1). Then, in step B2, slots of knowledge "move" are sequentially set for all case slots of the case frame expression. That is, the target “desk 1” is set to the target slot, and the direction “left” is set to the direction slot. Thereafter, when the command format slot of knowledge "move" is taken out in step B3, a command shown in FIG. 9A is obtained. Then, by passing this command to the task execution unit 13, the display screen of FIG. 10A is updated to the state shown in FIG. 10B.

Next, when the natural language sentence shown in FIG. 11B is input, the result of the semantic analysis is expressed as shown in FIG. 12B, which is input to the abbreviated expression interpolation unit 7. Is done. Since the verb is omitted in this expression, the case frame expression stored in the conversation history 9 is checked to interpolate the information. In this case, since the specified noun is "chair", the class a becomes "chair" in step A5. Further, when a set of classes having a sibling relationship with the object “chair” in the knowledge base 12 is A, such an object is only “desk”.
= {“Desk”}. Since the input sentence in which the object belonging to A is specified is the case frame expression indicated by the pointer P in the conversation history 9, this is set to F.

In the case frame expression indicated by the pointer P in the conversation history 9 (FIG. 7), b = “desk 1”, a =
In step A7, “b” is replaced with “a” as “chair”, whereby the expression shown in FIG. 14B is obtained.

Next, in step A8, there are three instances belonging to "a", "chair 1", "chair 2", and "chair 3", and it is necessary to identify which of them is correct. In the distance indicated by the relative distance memories B1 and E1, b (= “desk 1”) is closest to “chair 1” in this case.
Therefore, this is set to c.

Finally, in step A 9, “desk 1” in the case frame expression F is replaced with “chair 1” and output to the command conversion section 8.

In the command conversion unit 8, in the flowchart shown in FIG. 6, the case verb at this time is “move”.
(Shown in FIG. 4) (step B)
1). Then, in step B2, slots of knowledge "move" are sequentially set for all case slots of the case frame expression. That is, the target “chair 1” is set to the target slot, and the direction “left” is set to the direction slot. Thereafter, when the command format slot of knowledge "move" is taken out in step B3, a command shown in FIG. 9B is obtained. By transferring this command to the task execution unit 13, the "chair 1" is moved, and the display screen of FIG. 10B can be updated to the state shown in FIG. 10C.

It should be noted that the present invention is not limited to the above-described embodiment, but can be carried out by appropriately modifying it without changing the gist. For example, in the above-described embodiment, the case where the target object is indicated by a direct noun has been described. However, by using the topic presented by the particle “ha”, the target object closest to the topic in terms of distance is given priority. It is also possible to employ a process for identifying the data. Also, in the input processing of “chair”, a spatial focus area is determined from the distance relationship between the immediately preceding pointing object and the current pointing object, and a movable object within the focusing area is determined. It is also possible to add a process of moving at the same time. That is, when the immediately preceding pointing object is “desk 1” and the pointing object at the present time is “chair 1”, the objects near both of them are simultaneously moved.

[0056]

According to the present invention, when there are a plurality of targets that can be pointed, the target to be pointed is identified by using the relative distance relationship between the targets. The determination of an object can be made easily and automatically at any time. In addition, since the operator can give operation instructions only in natural language, it is possible to configure a smooth conversation system such as eliminating the inconvenience of having to store a command format for each different system.

[Brief description of the drawings]

FIG. 1 is a diagram showing a schematic configuration of an embodiment of the present invention.

FIG. 2 is a diagram showing an expression format of knowledge in a knowledge base.

FIG. 3 is a diagram showing a specific example of a knowledge expression in a knowledge base.

FIG. 4 is a diagram showing a specific example of a knowledge expression in a knowledge base.

FIG. 5 is a flowchart for explaining processing of an abbreviation expression interpolation unit.

FIG. 6 is a flowchart for explaining processing of a command conversion unit.

FIG. 7 is a diagram showing a part of relative distance storage.

FIG. 8 is a diagram showing a part of a conversation history.

FIG. 9 is a view showing a command corresponding to the input sentence shown in FIG. 11;

FIG. 10 is a diagram illustrating an example of a display screen according to the embodiment of the present invention.

FIG. 11 is a diagram showing an example of an input sentence.

FIG. 12 is a diagram showing a case frame expression obtained by performing morphological analysis, syntax analysis, and semantic analysis on the input sentence of FIG. 11;

FIG. 13 is a diagram showing an outline in a knowledge base.

FIG. 14 is a diagram illustrating a state in which omission information is interpolated by an omission expression interpolating unit.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... input part, 2 ... keyboard, 3 ... mouse, 4 ... morphological analysis part, 5 ... syntax analysis part, 6 ... semantic analysis part, 7 ... abbreviation interpolation part, 8 ... command conversion part, 9 ... conversation history, 10 …
Relative distance storage, 11 relative distance extraction unit, 12 knowledge base, 13 task execution unit, 14 image model, 15 output unit, 16 display.

 ──────────────────────────────────────────────────続 き Continuation of the front page (56) References Teruhiko Ukita, Kazuo Sumita, Satoshi Kinoshita, Makie Amano, Hiroshi Sano, “Natural Language Input Device Operation Guidance System”, ICOT Research Bulletin, No. 563 (1988)

Claims (1)

(57) [Claims] An input means for inputting a natural language sentence and a case frame expression conversion processing for converting an input sentence of the input means into a case frame expression in a conversation system capable of operating a device by giving an instruction in a natural language. Means,
Conversation history storage means for sequentially storing case frame expressions converted by the case frame expression conversion processing means as a conversation history; and distance information storage for storing relative distance information between a plurality of objects on a display screen. If the word about the action is present in the case frame expression of the input sentence, the object in the most recently input case frame expression is obtained. Otherwise, the object is specified in the input sentence. The most recently input case frame expression including an object belonging to a set of objects having a sibling relationship with the object is specified with reference to the case frame expression stored in the conversation history storage means, and the specified case frame expression is specified. The target object specified in the case frame expression is obtained, and based on the relative distance information stored in the distance information storage means, the target object closest to the target object obtained by the above processing is obtained. An expression interpolating means for determining an instance as an object specified in a case frame expression of an input sentence; a command converting means for converting an output interpolated by the expression interpolating means into a predetermined command format; Output means for controlling a display screen based on a command from the means.