KR20060064533A - System and method for processing natural language request - Google Patents

Abstract

The present invention relates to a natural language sentence processing method and a natural language sentence processing system. According to the present invention, when processing a natural language sentence to interpret and respond to the meaning by expressing the input natural language sentence in a formalized structure, the frame for each sentence type is defined, according to the type based on the type of natural language sentence to be input Natural sentences are formalized through frames. When the natural language command is expressed formally, the meaning of the command can be clearly understood through a program. That is, the natural language command can be converted into a formal expression language and interpreted, so that the user can accurately grasp what the user requests through the original natural language command. Therefore, the service required by the user can be automatically provided. According to the present invention, it is possible to operate a computer, a device robot, etc. conveniently and naturally using natural language in an interface between a person and a computer and an interface between a person and a robot.

  Natural Language, Formatting, Frames, Command, Robot

Description

Natural sentence processing system and natural sentence processing method {SYSTEM AND METHOD FOR PROCESSING NATURAL LANGUAGE REQUEST}

1 is a frame structure of a formalized expression language according to an embodiment of the present invention.

2 is a flowchart of a natural-language sentence processing method according to an embodiment of the present invention.

3 is a block diagram of a natural language sentence processing system according to an embodiment of the present invention.

4 is a block diagram of a query response system according to an embodiment of the present invention.

The present invention relates to a method for processing a natural language sentence, and more particularly, to a natural language sentence processing method and a natural language sentence processing system for interpreting and responding to a meaning by expressing an input natural language sentence in a formalized structure.

There are two methods of expressing natural language sentences in a formal structure for controlling a robot or a computer program. First, in-depth natural language processing technique is used to express one natural sentence into one or more logic sentences, and second, more fragmentary and efficient natural language processing technique is used to frame one natural sentence. There was a way to express it as a structure.

A representative example of the technique of expressing logic statements is Discourse Representation Structure (DRS), which is a technique that analyzes a sentence as a parsing tree and scans the tree and transforms it into logical statements with the minimum conversion unit. A sentence formatting language used in DRT (Discourse Representation Theory). The following is a simple example of a DRT.

DRS has the advantage that it can best preserve the meaning of the original natural language sentence and can be easily converted to light processing by a logic programming language such as Prolog. However, parsing the sentence itself is expensive and error-prone. In particular, Korean syntax parsing is known to be more difficult and error-prone, making it difficult to put to practical use.

A representative example of formatting natural language sentences using a frame structure is PAR (Par ameterized Action Representation). PAR is a method of expressing a user's command by using a complex structure including many fields that can comprehensively express various commands of the user. The following is an example of sentence formatting using PAR.

-Original Sentence: A cobbler sold a pair of boots to a hiker.

PAR expression: (SELLING (ACTOR = A COBBLER) (PATIENT = A BOOT) (RECIPIENT = A HIKER))

As shown in the example, PAR classifies behavior through predefined frames such as SELLING, and expresses sentences by pre-defining specific behavior related slots such as ACTOR, PATIENT, and RECIPIENT. Creating a frame from a sentence is simpler than making a logic statement, but you can also create a frame with relatively simple parsing.

However, PAR has a disadvantage in that expressiveness is deteriorated by predefining a frame for each action and a slot for each frame. This is because it is impossible to express behaviors that do not define a frame structure beforehand. In addition, PAR has a power of expression that is limited to the description of behavior, which makes it difficult to express a query. That is, it is not appropriate to express various user commands.

Therefore, the technical problem to be achieved by the present invention is to be able to express and process various languages of the user in the form of a simple structure in the control of a robot or a computer program.

Natural language sentence processing method according to an aspect of the present invention for achieving the above technical problem, as a natural language sentence processing method for interpreting and responding to the meaning by expressing the input natural language sentence in a formal structure, a) Determining the type; And b) formatting the natural language sentence input through a predefined frame for each sentence type.

In addition, the natural language sentence processing system according to the characteristics of the present invention is a natural language sentence processing system for interpreting and responding to the meaning by expressing the input natural language sentences in a formalized structure, by dividing the types of sentences of the input natural language by sentence type A natural language processing module for formatting natural language sentences through a predefined frame is included.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

In the following description, a system for expressing and processing natural language commands in a simple format structure for convenience of explanation is described as an example of a robot system in which real-time responsiveness to a user command is important. Of course, this is not limited.

In addition, hereinafter, a formalized expression for a natural language sentence according to a feature of the present invention is called a FURRL (Formalized User Request Representation Language). At this time, the term is a term given for convenience of description, it is also clear that the scope of the present invention is not limited by the term.

FURRL can be defined in a frame structure, and can express simple commands, conditional commands, and query statements. Examples of each sentence class are as follows.

(1) Simple command: turn on the TV, come on, open the window after 1 minute, etc.

(2) Condition Command: Wake if it rains, turn on the boiler if it's cold, wake up at 11 o'clock if it rains tomorrow, etc.

(3) Question: Do you like the weather in Seoul tomorrow? Is Gyeongbu Express Station blocked? Who's at home? Etc

For each sentence class, FURRL can provide three frames:

(1) Action: Frame for expressing simple command.

(2) ConditionAction: Frame to express condition instruction.

(3) Query: Frame for expressing query statement.

Simple commands and query statements are the same because they are expressed in one frame in the structure of a formal statement, but condition commands can be expressed in two types of frames, including two elements in one sentence: condition and action. . Two frames of conditional commands can represent conditions and commands, respectively.

The internal structure of three frames of Action, CondtionAction, and Query is as shown in FIG.

ConditionAction includes subframes Condition and Action, and Action, Condition, and Query each include the same slot. Details of each slot are as follows.

<Slot specification>

(1) Theme

Topic concepts include concepts that are the core subject of a sentence. Here are some examples:

(2) Assistive Theme

Auxiliary concepts are concepts that limit the scope of the concept by subsidizing the subject concept in the sentence. Here are some examples:

(3) Theme Time

Topic time includes the concept of time that modifies the concept of a topic in a sentence. An example is shown below.

(4) Theme Space

The subject space includes a spatial concept that modifies the subject concept. An example is as follows.

(5) Object

The object concept includes an object concept that is closely related to the subject, and various sentence construct concepts such as the object of the action and the subject of the action may be the object. Here is an example:

(6) Object Time

Object time includes the concept of time that modifies the concept of objects.

(7) Object Space

Object space includes the concept of space that modifies the concept of objects.

(8) Object Identity

The object discrimination concept is a concept that limits the conceptual scope of the object concept, and functions similar to the auxiliary theme. An example is as follows.

(9) Modality

Aspects express aspects of user commands such as allow, prohibit, and possibility. The vocabulary expressing aspects is as follows.

<Space-time expression>

In FURRL, space-time knowledge is a value of a space-time related slot and is expressed by combining a space-time expression predicate and a space-time concept. The concept of space-time refers to a specific time or space, and includes 'now', 'yesterday', 'birthday', 'house', and 'seoul'. The space-time expression predicate expresses the concept of 'formula' on the concept of space-time. Concepts such as before, after, and after are expressed in space-time predicates.

(1) time expression predicate

Predicates that can be used to express time concepts are listed below.

The concept of time can be divided into several different categories, each of which must have different meanings of time and different forms of expression. The concept of time can be divided into three categories:

(1) Quantitative point in time: 1 pm, January 20, 2004, etc.

(2) Quantitative interval time: 2 minutes, 2 hours, 3 days, etc.

(3) Qualitative time: morning, evening, yesterday, yesterday, etc.

Among these, (1) and (2) express time quantitatively, and (3) express time somewhat qualitatively using commercial vocabulary. The distinction between time point and segment time depends on whether the concept of time refers to a specific point in time or a time interval of arbitrary length.

The start time and the interval time can be expressed as DateTime and Duration frames as shown below.

(2) Division of time point and section time

The time point and the segment time are sometimes difficult to determine within a sentence. First, assuming that FURRL is a syntactic structure for formally expressing simple commands to robots or interactive agents, the distinction between viewpoint time and space time is based on simple distinction rules without in-depth interpretation of time expression vocabulary. It can be distinguishable.

For easy and clear section time discrimination, the pattern to be interpreted as the section time is determined, and other time representations may be configured to represent the view point time. The following is an example of a pattern for distinguishing the segment time from the viewpoint time.

As shown in the -FURRL definition, the edge time always contains only quantitative time, not qualitative time.

-The segment time is attached to the ending of the before, before, and every time.

-Pattern 1: A single time pattern. All expressions with the following time vocabulary are considered interval time.

a. ~ Minutes: 3 minutes later, 5 minutes before

b. ~ Sec: after 30 seconds, 20 seconds before

c ~ time: 1 hour later, 2 hours before

d. ~ Day: 1 day later, 2 months ago

e. ~ Month: After 1 month, 2 months before

f. ~ Month: One month before, three months later

g. ~ Year: 1 year ago, 2000 years ago, 10000 years ago (when expressing year in time point, order after 2000, before 1970)

-Pattern 2: In the case of a compound time expression, if a pattern other than the interval time expression pattern is included, it should be interpreted as a viewpoint time. Here are some examples of compound time representations.

a. 3 hours 2 minutes ago (expression of segment time)

b. Before 3:02 (expression of time)

c. After 3 minutes and 20 seconds (expression of segment time)

d. After 3:03:20 (expression of time)

e. After 4 months and 5 days (expression of segment time)

f. After April 5 (time expression)

g. 3 months 20 days before (expression of segment time)

h. Before March 20 (time expression)

i. After 5 months in 1920 (expressing segment time)

j. After May 1920 (time expression)

k. 4 hours after 3 days (expression of segment time)

l. After 3 days and 4 o'clock

(3) Example of time point representation

The following table shows some of the commands, including the time point, and a formal representation of the time information contained in the command.

(4) Example of segment time expression

The following table shows some of the commands that represent the edge time and the formal representation of the time information contained in the command.

(5) spatial expression predicates

Also, as in the case of time expression, a predicate that expresses space is also required. The table below is a list of predicates used to express knowledge related to space.

Based on the foregoing, an example of how a natural language instruction is formatted using FURRL is as follows.

At this time, in the embodiment table, the slot name is abbreviated for convenience, and the abbreviation for each slot is as follows.

The following are examples of sentence formatting. The various slot values appearing in the formal expression sentences are presented by combining English words suitable for expressing the meaning of the corresponding sentence on the left side, and do not directly indicate the formal language according to the features of the present invention. The following example shows how various natural language commands can be effectively represented by a FURRL frame structure.

Hereinafter, a natural language sentence processing method according to an exemplary embodiment of the present invention will be described with reference to FIG. 2.

First, a frame is defined for each type of natural language sentence. For example, as described above, the natural language sentence is largely classified into a simple command, a condition command, and a query statement. As a frame for expressing a command, a Query frame may be defined as a frame for expressing a query statement (S100).

In this case, the Action frame and the Query frame may be represented as one frame, and the ConditionAction frame may be expressed as two frames, Condition and Action.

Each frame includes one or more slots, for example, the Theme, Assistive Theme, Thematic Time, Thematic Space, Object, Object Identity, Object Cardinality, Object Time, Object Space, Modality slots.

In addition, the value of the space-time-related slot is expressed by combining the space-time expression predicate and the space-time concept. For example, the time expression predicates include Before, After, At, From, To, Every, and the like. The time concept used together may be classified into quantitative viewpoint time, quantitative interval time, and qualitative time. The time point and the section time may be represented by DateTime frames and Duration frames, respectively. Space expression predicates include space: From, space: To, space: At, and so on.

Details of each frame and slot have already been mentioned, so they are omitted here.

Next, when the natural language sentence is input from the user (S101), the natural language sentence is converted into a formalized expression sentence through a frame according to the type of the input sentence.

For example, when the input sentence is a condition command, the natural sentence is formatted through a ConditionAction frame. At this time, since the ConditionAction frame further includes a Condition frame and an Action frame, in the case of a condition command, the ConditionAction frame is formatted through the Condition frame and the Action frame (S103 and S107).

More specifically, when a natural language of "wake up when it rains tomorrow morning" is input, the sentence is formatted through a ConditionAction frame as a condition command. At this time, if the "morning tomorrow morning" corresponding to the condition through the Condition frame, "wake up" corresponding to the action action frame is formatted.

"When it rains tomorrow morning", "When it comes to" is "topic concept", "rain" is "object concept", "tomorrow" and "morning" are "topic time", and the time expression as "At" Since is used, it can be expressed as follows.

Condition (theme = fall, object = rain, theme time = At (tommorrow), theme time = At (morning))

In addition, since "wake up" in "wake up" is a "topic concept", it can be expressed as follows.

Action (theme = wake)

Thus, "Wake up when it rains tomorrow morning" can be expressed as:

ConditionAction (Condition (theme = fall, object = rain, theme time = At (tommorrow, theme time = At (morning))

                Action (theme = wake))

In addition, when the input sentence is a simple command or query word, it is formatted through an Action frame or a Query frame, respectively (S103 and S105).

More specifically, when the natural language "come to the kitchen" is input, the sentence is formatted through the Action frame as a simple command.

In this case, since the "and" and "and" are "topic", "kitchen" is "topic space", and "space: To" is used as a space expression predicate, the simple command may be expressed as follows. have.

Action (theme = come, theme space = space: To (kitchen))

Or, "Will it rain tomorrow?" When a natural language is input, the sentence is formatted through a Query frame as a query word.

In this case, "W" in "tomorrow rain?" "W" is "topic concept", "Rain" is "object concept", "tomorrow" is "topic time", At can be used as a time expression predicate of the time point The query term may be expressed as follows.

Query (theme = fall, object = rain, theme time = At (tommorrow))

In this way, the natural language command can be expressed in a simple form structure, thereby improving the efficiency of sentence analysis and formatting. In particular, when a relatively simple command is mainstream and real-time responsiveness is important, when the natural language command processing method according to the characteristics of the present invention is applied to a robot or an interactive agent, the responsiveness to the natural language command can be improved.

That is, as shown in FIG. 3, the natural language command processing system may include a natural language recognition module 100, a natural language processing module 200, and a response module 300.

The natural language recognition module 100 may transmit the input natural language sentence to the natural language processing module 200 without performing a separate conversion process.

The natural language processing module 200 may include a sentence analysis module 201 and a formal sentence conversion module 203 to express the natural language sentence received from the natural language recognition module 100 in a simple format structure.

More specifically, the sentence analysis module 201 determines the type of the input natural language sentence. According to the above-described example, the sentence analysis module 201 may determine whether a simple command, a condition command, or a query statement is present.

The formal sentence converting module 203 may express the natural language sentence in a simple format structure through the corresponding frame according to the type of natural language sentence determined by the sentence analysis module. For example, if a natural language sentence is a simple command, it can be expressed through a Action frame, if a natural language sentence is a condition command, a ConditionAction frame, and if a query word is a simple form structure, each frame and frame Since the description of the slots has already been described above, they will be omitted.

The response module 300 may receive sentence information expressed in a simple formal structure from the natural language processing module 200 to interpret the sentence information and provide an appropriate response according to a requirement. At this time, according to the present invention, it is possible to increase the speed of converting a natural language sentence into a form sentence, and as a result, the sentence interpretation and the response speed of the response module can also be increased, thereby increasing user convenience for the system.

More specifically, as shown in FIG. 4, the question answering system may be described as an embodiment according to the features of the present invention.

The question and answer system may include a speech recognition module as a specific example of the natural language processing module 200 and the natural language recognition module 100, and a specific example of the response module 300 may include a target setting module, a planning module, and a plan. It may include an execution module and a voice speech module.

The voice recognition module recognizes a voice command transmitted from the user and transmits the natural language sentence to the natural language processing module 200.

As described above, the natural language processing module 200 expresses the natural language sentence in a simple format structure through an Action frame, a ConditionAction frame, or a Query frame according to the type of natural language sentence.

The goal setting module interprets the meaning of a user command through a format sentence sentence received from the natural language processing module 200 and sets a goal to be processed by the system in response to the user command.

The planning module executes the service planning of the system for performing the goal based on the system goal specification received from the goal setting module, and the plan execution module executes the process according to the service planning.

In other words, according to the present system, the user's command can be received as an input, the meaning of the user's requirements can be determined, and the user's requirements can be given to the user. In the case of an essential system, user convenience can be maximized.

What has been described above is only an embodiment of the present invention, and the scope of the present invention is not limited thereto, and various applications can be made to those skilled in the art, for example, used in the present invention. The type of the frame and the configuration of the slots forming the frame are not limited to those described above. The scope of the present invention will include all technical components recognized as the claims described below and equivalents thereof.

As described above, according to the present invention, the structure of a formal statement can be simplified and intuitively expressed so that a formal statement can be expressed without complicated and accurate parsing of a user's command. Therefore, it is possible to improve the real-time for the conversion of user statements and improve the practicality of the technology.

In addition, the expression power can be easily expanded through a universal expression structure that is not dependent on the behavior of the expression, such as PAR. Therefore, more kinds of questions can be formalized and processed.

As a result, it is easy to build services of the system that must understand user commands and provide meaningful responses, and the user interface is improved.

In other words, natural language commands can be expressed and processed in a simple format structure, which increases efficiency and accuracy in natural language sentence analysis and formatting, and is applied to robots, interactive agents, etc. where relatively simple commands are mainstream and real-time responsiveness is important. User convenience can be improved.

Claims (27)

In the natural language sentence processing method for interpreting and responding to the meaning by expressing the input natural language sentence in a formal structure, a) determining the type of natural language sentence to be input; And b) formatting natural language sentences input through a predefined frame for each sentence type. How to handle natural language sentences. The method of claim 1, The frame, It contains an Action frame for a simple command sentence, a ConditionAction frame for a condition command sentence, and a Query frame for a query statement character. How to handle natural language sentences. The method of claim 2, The condition action frame includes a condition frame and an action frame. How to handle natural language sentences. The method according to any one of claims 1 to 3, The frame includes a plurality of slots for expressing components of the natural language sentence. How to handle natural language sentences. The method of claim 4, wherein The slot is, Theme, Assistive Theme, Theme Time, Theme Space, Object, Object Identity, Object Time, Object Space ( Each of one or more of Object Space and Modality. How to handle natural language sentences. The method of claim 5, Step b), Classifying the elements of natural sentences and assigning them to each slot of the frame How to handle natural language sentences. The method of claim 6, Step b), Temporal knowledge among the components of a natural language sentence is assigned to the value of a time-related slot and is expressed in combination with a time expression predicate. How to handle natural language sentences. The method of claim 7, wherein Step b), The time knowledge is divided into a viewpoint time and an interval time, and is represented by a date time frame and a duration frame as values of time-related slots, respectively. How to handle natural language sentences The method of claim 7, wherein The time expression predicate Contains one or more of Before, After, At, From, To, Every How to handle natural language sentences. The method of claim 7, wherein The time slot is At least one of the subject time slot and the object time slot How to handle natural language sentences. The method of claim 6, Step b), Spatial knowledge among the components of natural language sentences is assigned to values of space-related slots and is expressed in combination with spatial expression predicates. How to handle natural language sentences. The method of claim 11, The spatial expression predicate contain one or more of space: From, space: To, or space: At How to handle natural language sentences. The method of claim 11, The space related slots, At least one of the subject space and the object space How to handle natural language sentences. In the natural language sentence processing system for expressing the input natural language sentence in a formal structure to interpret and respond to the meaning, A natural language processing module for classifying natural language sentences by classifying sentence types of natural language input and formatting natural language sentences through predefined frames for each sentence type; Natural language sentence processing system. The method of claim 14, The natural language processing module A sentence analysis module classifying types of natural language sentences; And It includes a format sentence conversion module for formatting natural language sentences through the frame based on the type of sentence Natural language sentence processing system. The method of claim 15, The natural language processing module, Based on whether it is a simple command, a condition command, or a query statement, the natural language sentence is formatted through an action frame, a condition action frame, and a query frame, respectively. Natural sentence processing system The method of claim 16, The condition action frame includes a condition frame and an action frame. Natural language sentence processing system. The method according to any one of claims 14 to 17, The frame includes a plurality of slots for expressing components of the natural language sentence. Natural language sentence processing system. The method of claim 18, The slot is, Theme, Assistive Theme, Theme Time, Theme Space, Object, Object Identity, Object Time, Object Space ( To represent one or more of Object Space and Modality. Natural language sentence processing system. The method of claim 19, The natural language processing module The components of natural language sentences are separated and assigned to each slot of the frame. Natural language sentence processing system. The method of claim 20, The natural language processing module, The temporal knowledge among the components of natural language sentences is assigned to the values of time-related slots, which are formatted by combining them with time expression predicates. Natural language sentence processing system. The method of claim 21, The natural language processing module, The temporal knowledge is divided into a time point and an interval time, and expressed as a value of a time-related slot through a date time frame and a duration frame, respectively. Natural language sentence processing system. The method of claim 21, The time expression predicate, Contains one or more of Before, After, At, From, To, Every Natural language sentence processing system. The method of claim 21, The time slot is One or more of the subject time slot and the object time slot Natural language sentence processing system. The method of claim 20, The natural language processing module, Spatial knowledge among the components of natural language sentences is assigned to the values of space-related slots. Natural language sentence processing system. The method of claim 25, The spatial expression predicate is contain one or more of space: From, space: To, or space: At Natural language sentence processing system. The method of claim 25, The space related slots, At least one of the subject space slot and the object space slot Natural language sentence processing system.

Images