JP2017107276A - Information processing device, information processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform language processing in association of an object existing in an environment with the content of an operation instruction, for even a sentence in which a subject or an object, etc., are omitted.SOLUTION: An information processing device 1 comprises: a dictionary database 10 in which the spatial relative relation of a word and an object co-occurring in a verb representing an operation content is stored; a detection unit 2 for acquiring the data of spatial relative relation of an external object; an input unit 3 for accepting an operation command; a spatial semantic structure analysis unit 4 for referring to the dictionary database 10 and finding the hierarchical structure of the operation command including an omitted word; and a plurality of partial models connected by a hierarchical structure. Each of the partial models is provided with a probability model generation unit 5 for generating a probability model having a first variable relating to a word, a second variable group relating to the spatial relative relation of an object, and a third variable relating to the extent of correspondence of the object to the word, and a collation unit 6 for determining the spatial relative relation of an object having a largest probability value from the data of spatial relative relation detected by the detection unit 2.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an information processing apparatus that performs natural language processing, and more particularly, to an information processing apparatus that analyzes operation instruction contents related to an object (object) in a space.

  Conventionally, it has been known that an operation instruction related to an object in a space is given to a robot by voice (Patent Document 1). However, since the above-described conventional technique does not extract the spatial meaning of the object, it cannot handle the relationship between the relative positions of the objects and the relationship between the operation contents and the objects.

  On the other hand, there is a previous research on a technique that expresses spatial semantic information in a hierarchical tree structure and extracts the spatial semantic structure in it from a natural language sentence by a probabilistic method (non-patented). Document 1, Non-Patent Document 2).

JP 2011-170789 A

T. Koller et al. `` Towards Understanding Natural Language Directions '' Proceedings of the 5th ACM / IEEE International Conference on Human-robot Interaction S. Tellex et al. `` Understanding Natural Language Commands for Robotic Navigation and Mobile Manipulation '' In Proceedings of the National Conference on Artificial Intelligence (AAAI 2011).

  However, in usual conversations, the subject, the object, and the like are often omitted, but Non-Patent Documents 1 and 2 do not provide means for solving such omission. In particular, in the case of Japanese, omission of subjects and objects called zero pronouns is allowed, so there is a need to solve these problems.

  In Non-Patent Document 1, it is assumed that the environment is static like a building. In the technique described in Non-Patent Document 1, since it is necessary to teach the operation contents to the robot in a static environment in advance, there is a situation where it changes dynamically like the drive environment. It was not applicable.

  For example, when considering an environment in which a driver gives driving instructions in words in automatic driving or the like, the environment always changes dynamically, so the technique described in Non-Patent Document 1 cannot be applied and can be applied. However, there is a problem that it can be applied only in a very limited environment that is static and known.

  Accordingly, an object of the present invention is to provide an information processing apparatus that performs linguistic processing by appropriately associating an object existing in an environment with the content of an operation instruction even if the subject, object, etc. are omitted. And

  An information processing apparatus according to the present invention includes a dictionary database storing words and co-occurrence of objects coexisting in a verb representing operation content, and a detection unit that acquires data of spatial relations related to objects existing in external space And an input unit that accepts an input of an operation command sentence from a user, and a hierarchical structure of words included in the operation command sentence input in the input unit, and is omitted in the operation command sentence with reference to the dictionary database A certainty factor that has a spatial semantic structure analysis unit that obtains a hierarchical structure including the wording and a plurality of partial models connected by the hierarchical structure, and the certainty factor is obtained as a function of the certainty factor of each partial model Each of the partial models includes a first variable relating to the wording, a second variable group relating to a spatial relative relation of the objects, A confidence model generation unit that generates a confidence model having a third variable related to the degree to which the object corresponds to the wording, and data on the spatial relative relationship between the objects detected by the detection unit A collation unit that is applied to a group to calculate a certainty factor that is obtained as a function of the certainty factor of the plurality of partial models, and determines a spatial relative relationship of an object that maximizes the certainty factor; and an analysis of the operation command statement And an output unit for outputting the result. Here, the certainty model may be a probability model or an identification model such as a neural network or SVM (Support Vector Machine).

  In this way, by referring to the dictionary database and obtaining the hierarchical structure including the words omitted in the operation command sentence, the words that would be omitted using the information on the spatial relative relationship of the external object. Candidates can be sought. Here, the “word” is a verb, a noun phrase, or the like, and may be a noun plus infix. Then, the subject or object is omitted by sequentially substituting the detected external object into the second variable group and determining the spatial relative relationship of the object having the maximum certainty factor of the certainty factor model. Even in the case of a sentence, language processing can be performed by appropriately associating an object existing in the environment with an operation instruction. In addition, since the values to be assigned to the second variable group are narrowed down by objects existing outside, the calculation process can be reduced.

  In the information processing apparatus of the present invention, the detection unit may perform coordinate conversion of the position of the detected object into a local coordinate system based on the host vehicle based on the map information.

  By converting the coordinates to the local coordinate system in this way, it becomes easier to handle the position of the object, and the calculation process can be reduced.

  In the information processing apparatus of the present invention, the detection unit acquires time-series data of the spatial positional relationship of the object, and when the state of the object is changed, the state of the object different from the current acquired in the past And the spatial semantic structure analysis unit obtains a hierarchical structure by distinguishing a word related to the current object and a word related to the past object when the word related to the past object is included. Also good.

  In the information processing apparatus according to the present invention, the detection unit obtains time-series data of the spatial positional relationship of the object, and when the state of the object is changed, the state of the future object to be predicted is obtained. The spatial semantic structure analysis unit may detect a hierarchical structure by distinguishing a word related to the current object and a word related to the future object when the word related to the future object is included. Good.

  By treating an object whose state has changed in the external environment as a past object or future object in this way, even if the operation command statement includes a command related to a past or future object, the language can be appropriately Processing can be performed.

  An information processing apparatus according to the present invention acquires a dictionary database storing words and co-occurrence of objects that co-occur on verbs representing operation contents, and time-series data of spatial relative relationships related to objects existing in an external space. A detection unit that detects, as a past object, a state of an object different from the current acquired in the past when the state of the object has changed, an input unit that receives an input of an operation command sentence from a user, and the dictionary database Referring to, the operation command sentence input in the input unit is analyzed, and when a word related to a past object is included, a word related to a current object is distinguished from a word related to a past object. A spatial semantic structure analysis unit for obtaining a hierarchical structure, and a plurality of partial models connected by the hierarchical structure. A certainty factor model in which the certainty factor is obtained as a function of the certainty factor of each partial model, wherein each partial model includes a first variable relating to the wording and a second variable group relating to a spatial relative relation of the objects. , A confidence model generation unit that generates a confidence model having a third variable related to the degree to which the object corresponds to the wording, and data of spatial relative relations of the objects detected by the detection unit. A collation unit that is applied to a group of variables to calculate a certainty factor obtained as a function of the certainty factor of the plurality of partial models, and determines a spatial relative relation of an object that has the maximum certainty factor, and the operation command statement And an output unit for outputting the analysis result.

  In addition, the information processing apparatus of the present invention includes a dictionary database that stores words and co-occurrence of objects that co-occur on verbs representing operation contents, and time-series data of spatial relationships of objects that exist in external space. And when the state of the object is changed, a detection unit that detects the predicted state of the future object as a future object, an input unit that receives an input of an operation command from the user, and the dictionary database Referring to the analysis of the operation command sentence input in the input unit, if the word about the future object is included, the word about the current object and the word about the future object are distinguished It has a spatial semantic structure analysis unit for determining the structure and a plurality of partial models connected by the hierarchical structure. A certainty factor model in which the certainty factor is obtained as a function of the certainty factor of each partial model, wherein each partial model includes a first variable relating to the wording, a second variable group relating to a spatial relative relation of objects, and the object A confidence model generation unit that generates a confidence model having a third variable relating to the degree corresponding to the wording, and data of the spatial relative relationship between the objects detected by the detection unit. A collation unit that calculates the certainty factor obtained as a function of the certainty factor of the plurality of partial models and determines the spatial relative relationship of the object having the maximum certainty factor, and the analysis result of the operation command statement An output unit.

  As described above, the detection unit detects past objects and future objects when the state of the object is changed based on the time series data of the spatial relative relationship, while the spatial semantic structure analysis unit detects the operation. Analyzing the statement to distinguish the current object from the past object or future object to obtain the hierarchical structure, so in the calculation of confidence, the past object is used as a partial model of the past object, and the future object is used as a part of the future object. Appropriate language processing analysis can be applied to the model.

  The information processing method of the present invention is a method executed by the information processing apparatus, and the program of the present invention is a program for realizing the method executed by the information processing apparatus.

  The present invention can perform language processing by appropriately associating an object existing in the environment with the contents of an operation instruction even if the subject, object, etc. are omitted.

It is a figure which shows the structure of the information processing apparatus of embodiment. It is a figure which shows the example of the data memorize | stored in the dictionary database. It is a figure which shows the example of a driving scene. It is a figure which shows the example of the structure analysis result of an operation command sentence. It is a figure which shows the example of a probability model. It is a figure which shows the operation | movement in which information processing apparatus acquires the data of the feature-value of an external object. It is a figure which shows the operation | movement which analyzes the content, when the operation command is made from the driver. (A) is a figure showing an example of a parking lot. (B) It is a figure which shows the coordinate system set to the parking lot. (C) is a figure which shows the example of the primary environment object which linked | related the basic attribute of the object and the option attribute to the coordinate system. It is a figure which shows the example of the secondary environmental object which prescribed | regulated the route from the own vehicle position to a parking space.

  Hereinafter, an information processing apparatus according to an embodiment of the present invention will be described with reference to the drawings. In the present embodiment, it is assumed that the information processing apparatus is an apparatus mounted on an automobile, and an operation command instructed by the driver is analyzed in the driving scene of the automobile.

  FIG. 1 is a diagram illustrating a configuration of an information processing apparatus 1 according to the embodiment. The information processing apparatus 1 includes a detection unit 2 that detects data related to an external environment, an input unit 3 that inputs an operation command from a driver, and a spatial semantic structure analysis unit that analyzes a hierarchical structure of the input operation command sentence 4, a probability model generation unit 5 that generates a probability model of an operation command sentence based on the hierarchical structure, and obtains a probability value by substituting the data of the spatial relative relationship of the detected object with respect to the probability model, A collation unit 6 that collates objects related to the command, and an output unit 7 that outputs an analysis result of the operation command based on the collation result in the collation unit 6 are provided. The information processing apparatus 1 includes an electronic control unit (ECU) as hardware. Such a program for controlling the ECU is also included in the scope of the present invention.

  A camera 11 and a positioning device 12 are connected to the information processing apparatus 1. The camera 11 images the periphery of the vehicle and transmits image data obtained by the imaging to the detection unit 2. The positioning device 12 is a device that measures the position of the host vehicle. For example, the positioning device 12 may receive a signal from a GPS satellite with a GPS receiver and may measure the position, or may be based on data obtained from a gyro / acceleration sensor or the like. The current position may be calculated.

  The detection unit 2 analyzes the camera image captured by the camera 11, extracts objects such as other vehicles, pedestrians, roadside objects, parking spaces, and road paints from the image, and the spatial relative relationship of the extracted objects. Get the data. Here, the spatial relative relationship data is a positional relationship between objects existing in the space. For example, the object A and the object B are adjacent to each other, or the distance between the object A and the object B is OOm. The object A has a relationship such as existing between the object B and the object C.

  The detection unit 2 may acquire not only the spatial relative relationship between the objects but also the attribute data of the objects. The object attributes are, for example, the type, size, color, etc. of the object. The detection unit stores the spatial relative relationship and attributes of the objects in an environment database (hereinafter referred to as “environment DB”) 9. In this specification, the spatial relative relationship and attributes of an object are combined and referred to as an object feature amount.

  The detection unit 2 is connected to a map database (hereinafter referred to as “map DB”) 8 and can read map data around the current position of the vehicle from the map DB 8. The detection unit 2 converts the position of the object obtained from the camera image into a coordinate system based on the host vehicle.

  Further, the detection unit 2 detects the data of the object over time, and stores the data in the environment DB 9 as a past object when the state of the object has changed, for example, the object is moving. If an object stays in the same place for more than a certain time, the time series may be segmented and divided into three different past objects according to the time of arrival at that point and the time of movement from that place, In the case of staying from the initial time, it may be divided into two past objects of staying time and moving time.

  Further, the detection unit 2 predicts the future movement of the object from the movement of the object from the past to the present, and stores it as a future object in the environment DB 9. Future objects are updated at a predetermined cycle.

  For example, when a vehicle (object) parked in the parking space leaves, the vehicle currently moving is the current object, and the vehicle parked in the parking space acquired in the past is the past object. Further, when it can be predicted from the current movement of the vehicle that the vehicle is heading toward the exit of the parking lot, the vehicle that will be at the exit of the parking lot in the future is the future object. Thus, the past object and the future object are virtual objects that do not exist at present. Details of the feature amount data of the object detected by the detection unit 2 and stored in the environment DB 9 will be described later.

  The input unit 3 includes a microphone that inputs voice and a voice recognition unit. The input unit 3 converts the voice uttered by the driver into text data and passes it to the spatial semantic structure analysis unit 4.

  The spatial semantic structure analysis unit 4 divides the operation command sentence input from the input unit 3 into morphemes having substantial meanings alone, and analyzes the hierarchical structure of the divided morphemes. If the input operation command sentence is a compound sentence, it is divided into simple sentences and the hierarchical structure of each simple sentence is analyzed.

  The spatial semantic structure analysis unit 4 identifies an instruction target corresponding to an anaphor such as “there” or “that”. In the case of Japanese, words that can be easily guessed by the listener are frequently omitted, but in such a so-called “zero pronoun” sentence, the spatial semantic structure analysis unit 4 is omitted. Analyze that the word is hidden.

  Further, the spatial semantic structure analysis unit 4 performs a tense (current, past, future) modality analysis, and the object included in the operation command statement is any of a past object, a (current) object, or a future object. Decide what. For example, in the case of an operation command sentence “stop where there was a red car”, it is determined that “red car” is a past object. In the case of an operation command sentence “when the previous car stops, park next to it”, it is determined that “the previous car” is a future object.

  FIG. 2 is a diagram showing an example of data stored in a dictionary database (hereinafter referred to as “dictionary DB”) 10. The dictionary DB 10 stores data of noun phrases that co-occur with the verb in the driving scene in association with the verb used in the driving scene, and features (spatial relative relationship / attribute) co-occurring with the verb. Is stored. For example, “in the parking space”, “on the roadside”, and “near the entrance” are stored as noun phrases co-occurring with the verb “stop”.

  Further, for example, in association with the words “stop” and “in the parking space”, the feature quantity “parking space: True forward: True distance: [1, 5] is on: {}” is stored. This is because, if there is a phrase “parking space” that co-occurs in “stop” in the operation statement, “parking space” exists “true” (True), and the distance is 1 to 5 m. This indicates that a spatial feature amount (True) co-occurs.

  A noun phrase that co-occurs in a verb is obtained by learning a large amount of operation command sentences used in a driving scene as teacher data. Also, with respect to the feature amount data, a large amount of video data obtained by photographing a driving scene is analyzed, and the movement of the car co-occurring with the verb included in the operation command sentence issued by the driver is detected, and the dictionary DB 10 is generated. .

  The dictionary DB 10 may include a plurality of dictionary DBs depending on the driving scene. For example, by having a plurality of dictionary DBs such as a dictionary DB used in a parking lot, a dictionary DB used in an urban area, and a dictionary DB used on a highway, words used in each driving scene can be analyzed appropriately.

  The probability model generation unit 5 generates a probability model of the operation command sentence based on the hierarchical structure of the operation command sentence analyzed by the spatial semantic structure analysis unit 4. The probabilistic model has a configuration in which a plurality of nodes are connected in a hierarchical structure, and each node uses a first random variable λ as a word for an operation command sentence, and second data as a feature quantity of an object. It has a random variable γ and a third random variable φ relating to the degree to which the wording corresponds to the feature quantity of the object.

  In FIG. 5, which will be described later, for convenience of illustration, each node has one second random variable γ. However, in practice, there are usually a plurality of feature quantities for one wording. , One node has a plurality of second random variables λ. The third random variable φ is “1” (True) when the value of the first random variable λ corresponds to the value of the second random variable γ, and “0” ( False). Nodes having the first random variable λ, the second random variable γ, and the third random variable φ constitute a “partial model”, and the partial models are assembled to form a probability model. The probability value of the probability model is obtained as a function of the probability value of each partial model. Here, the product of the probability values of the partial models is obtained as the probability value of the entire probability model.

  In the present embodiment, since the object is to obtain feature quantities corresponding to the first random variable λ and the second random variable γ, “1” (True) is substituted for the random variable φ. Note that the probability variable φ is used as a variable when learning the dictionary DB 10.

  FIG. 3 is a diagram illustrating an example of a driving scene described below. In this scene, let us consider a case where an operation instruction is issued from the driver: “If a red car next to the vending machine goes out, stop there”. Although FIG. 3 is shown in black and white for the sake of convenience, it is assumed that of the three vehicles parked at the parking lot, the two vehicles with the vehicle body shown in black are red vehicles. In addition, although the vehicle C1 shown with the dotted line has parked in the parking space in the past, it has shown that it came out of the parking space now. In this example, the operation instruction of the driver is intended to stop at the place P1 where the red car C1 has stopped after the red car C1 leaves, but after the red car C1 leaves Since the red car C1 does not exist, there is a possibility that it is erroneously determined that the instruction is to stop next to the red car C2.

  FIG. 4 is a diagram illustrating a spatial semantic structure analysis result of an operation command sentence “If a red car beside the vending machine goes out, stop there”. The sentence “When the red car next to the vending machine goes out, stop there” is broken down into the landmark L “When the red car next to the vending machine goes out” and the verb V “Stop” Is done. Here, the landmark L "When the red car next to the vending machine goes out" is a Spatial Description Clause, and the landmark L "next to the vending machine" and "Go out" It is broken down into a verb V and an object F called “red car”. Further, the landmark L “next to the vending machine” is a spatial descriptive phrase, and is decomposed into a landmark L “vending machine”, a spatial relationship SR “next”, and an object F “p red car”. . In the scene shown in FIG. 3, “p red car” does not exist, but is a virtual object (past object) indicating that a red car has existed in the parking space P1 next to the vending machine in the past.

  As described above, the spatial semantic structure analysis unit 4 analyzes the operation command sentence to obtain a hierarchical structure as shown in FIG. In addition, the spatial semantic structure analysis unit 4 analyzes the modality of the operation command sentence “If the red car next to the vending machine goes out, stop there”, and the expression of the assumption method is included. The past object corresponding to the anaphora is generated. In the case of the assumption method expression, the object is not necessarily a past object, but may be a future object. The modality analysis of the operation command statement can be performed using a known technique.

  FIG. 5 is a diagram showing an example in which a probability model is generated based on the spatial semantic structure analysis result shown in FIG. The probability model generation unit 5 generates a probability model including a node corresponding to the past object ξ as shown in FIG. Each node represents a probability value, and the probability value of the entire probability model can be obtained by multiplying the probability value of each node.

  The matching unit 6 calculates a probability value of the entire probability model by substituting a value for the random variable included in the probability model, and obtains a random variable that maximizes the probability value of the entire probability model. The matching unit 6 sequentially substitutes the detected feature data into the probability model, and obtains the probability value of the partial model at that time. The value to be substituted for the second random variable γ is feature value data detected by the detection unit 2. Specifically, when the spatial semantic structure analysis unit 4 performs the spatial semantic structure analysis based on the dictionary DB 10, among the feature quantities that co-occur in the words of the operation command sentence, the feature quantities having a strong co-occurrence relationship. Are stored as candidates, and the stored feature value data is used when collation is performed.

  In addition, in the operation command sentence, when a word is omitted by a directive, zero pronoun, or the like, the partial model is substituted by substituting the word into the first random variable λ of the target node of the probability model. Find the probability value of. At this time, the value to be substituted into the first random variable λ is obtained when the spatial semantic structure analysis unit 4 performs spatial semantic structure analysis based on the dictionary DB 10 and analyzes that the wording is omitted. Among the words that co-occur with the words of the operation command sentence, words having a strong co-occurrence relationship are stored as candidates, and the stored word data is used when collation is performed.

  The matching unit 6 calculates the probability value of the entire probability model by taking the product of the probability values of the partial models. Thereby, the object instruct | indicated in the operation command sentence and the abbreviation text can be determined appropriately.

  The output unit 7 outputs the analysis result of the operation command sentence based on the object data obtained by the collation unit 6. In other words, by confirming the corresponding object in the collation unit 6, an analysis result indicating what operation the instruction statement instructs the object in the external space can be obtained, and the result is output. To do.

  6 and 7 are flowcharts showing the operation of the information processing apparatus 1 according to the present embodiment. FIG. 6 shows an operation in which the information processing apparatus 1 acquires feature amount data of an external object, and FIG. 7 shows an operation in which the contents are analyzed when an operation command is issued from the driver. The operation shown in FIG. 6 and the operation shown in FIG. 7 are performed in parallel. Specifically, the feature amount data shown in FIG. 6 is always acquired. When an operation instruction is issued from the driver, the feature amount data shown in FIG. Operation is performed.

  First, acquisition of feature amount data of an external object will be described. As illustrated in FIG. 6, the information processing apparatus 1 processes an image captured by the camera 11 to detect an external object (S10), and uses the position of the detected external object as a reference to the local vehicle coordinate system. (S11). The local coordinate system is, for example, a coordinate system in which the own vehicle is a reference, the traveling direction of the own vehicle is the vertical axis, the direction orthogonal to the traveling direction is the horizontal axis, and the size of one vehicle or a half thereof is a scale. It is. Then, the information processing apparatus 1 stores the feature value data converted into the local coordinate system in the environment DB 9.

Here, an example is given and demonstrated about the feature-value of an object. However, the method of defining the feature amount is not limited to the method described below.
The objects include a primary environment object that defines the environment itself and a secondary environment object that defines a positional relationship in the environment. The primary environment object has basic attributes (ID, coordinates) that are attributes included in all objects, and optional attributes (type, color, size, etc.) included in some objects.

  Examples of primary environmental objects include road segments, parking spaces, landmarks, own vehicles, road paints, other vehicles, vending machines, fences, road lamps, poles, traffic signs, etc. Defined by the data shown.

  The primary environment object includes event data as data having no coordinates. For example, parking, moving forward, moving backward, moving toward the target, turning left, turning right, changing lanes, and the like.

  The secondary environment object also has basic attributes (ID, a set of primary environment objects) and optional attributes (type, etc.). Examples of secondary environmental objects include empty parking spaces, positional relationships (upper, next, middle, front, back, cross, right, left), order, and parking route.

  8A is a diagram showing an example of a parking lot, FIG. 8B is a diagram showing a coordinate system set in the parking lot in FIG. 8A, and FIG. 8C is a basic attribute of an object in the coordinate system. It is a figure which shows the example of the primary environment object which linked | related and the option attribute. As shown in FIG. 8C, data of feature quantities of objects such as parking spaces, lanes, trucks, small trucks, and gray bangs are associated with the local coordinate system.

  FIG. 9 is a diagram illustrating an example of a secondary environment object that defines a route from the vehicle position to the parking space. For example, “2” defines a route from coordinates (2,0) through (1,0) (1,1) (1,2) to the parking space at coordinates (0,2). Yes.

  As shown in FIGS. 8 and 9, feature amount data can be defined by giving the primary environment object and the secondary environment object defined above to the local coordinate system. The feature amount data has been described above. The information processing apparatus 1 always senses the external environment to detect an external object, converts the position of the detected external object to the local coordinate system, and extracts and stores the above-described feature amount data.

  With reference to FIG. 7, an operation when an operation command statement is input from the driver will be described. When the driver issues an operation command by voice, the information processing apparatus 1 accepts an input of an operation command sentence at the input unit 3 (S20). The information processing apparatus 1 divides the input operation command sentence into morphemes, performs anaphora analysis, and supplements the directives and zero pronouns omitted in the operation command sentence (S21).

  Subsequently, the information processing apparatus 1 performs a spatial semantic structure analysis, analyzes the hierarchical structure of the morphemes obtained by the division (S22), extracts the spatial meaning based on the analysis result (S23), For example, a hierarchical structure as shown in FIG. 4 is obtained. Next, the information processing apparatus 1 generates a probability model as shown in FIG. 5, for example, based on the hierarchical structure of the operation command sentence (S24).

  Next, the information processing apparatus 1 sequentially substitutes the object read from the environment DB 9 into the probability model, obtains the probability value of the entire probability model, and designates the object substituted when the probability value is maximized as the operation command statement. The matching process for obtaining the object corresponding to is performed (S25). Then, the information processing apparatus 1 outputs the analysis result of the operation command sentence (S26). The configuration and operation of the information processing apparatus 1 according to the present embodiment have been described above.

  In the present embodiment, the spatial semantic structure analysis unit 4 analyzes that there are omitted words and generates a probability model based on this. As a result, it is possible to appropriately perform processing that compensates for words omitted in the operation command sentence, and it is possible to intuitively give instructions for spatial operations without handling operations such as a monitor. In particular, in the case of Japanese, unlike English, omission of the subject and object called zero pronouns is allowed, so the matching process could not be performed without specifying them first, but according to this embodiment, Appropriate association processing can be performed.

  Further, in the present embodiment, the spatial semantic structure analysis unit 4 performs a tense (current, past, future) modality analysis, and the objects included in the operation command statements are past objects, (current) objects, Decide which of the future objects. On the other hand, since the detection unit 2 stores data of virtual past objects and future objects that do not currently exist based on the result of detecting the objects in the environment, the operation command statement relates to past or future objects. Even when the operation is included, the association process can be appropriately performed. Thereby, even when the environment changes dynamically, the language processing can be performed by appropriately associating the objects in the environment with the operation command statements. Moreover, since the assignment to the random variable is narrowed down to the objects existing outside, the search process can be reduced.

The information processing apparatus of the present invention has been described in detail with reference to the embodiment. However, the information processing apparatus of the present invention is not limited to the above-described embodiment.
In the above-described embodiment, an example in which data of noun phrases co-occurring with a verb is stored in the dictionary DB 10 in association with a verb has been described, but a noun classified by an infix may be associated. By adding infix information to nouns, verbs can be used according to the surface case of the noun (ga, ha, wo, ni, kara, het, to, yori, de, etc.) A database with strong co-occurrence of nouns can be generated. In addition, the dictionary DB may define the co-occurrence relationship based on deep case information such as an action case, a target case, and a place case, not necessarily information on the surface case.

  In the above-described embodiment, an example using a probability model has been described as a model for calculating the certainty of correspondence between a word and an object. However, the certainty model is not limited to the probability model, and the neural network is identified. A model or SVM may be used. That is, each partial model is configured by a neural network or SVM, and the reliability of the entire certainty model is determined as a product of the certainty of each partial model.

  In the embodiment described above, an example including both processing for obtaining a sentence omitted in an operation command sentence by spatial semantic structure analysis and processing for generating a probability model using a past object and a future object is given. The above two processes do not necessarily have to be set, and an apparatus in which only one of them is mounted is also included in the scope of the present invention.

  INDUSTRIAL APPLICABILITY The present invention can perform language processing by appropriately associating an object existing in the environment with the contents of an operation instruction even if the subject, object, etc. are omitted, and is useful for language analysis processing. .

DESCRIPTION OF SYMBOLS 1 Information processing apparatus 2 Detection part 3 Input part 4 Spatial semantic structure analysis part 5 Probabilistic model generation part 6 Collation part 7 Output part 8 Map DB
9 Environment DB
10 Dictionary DB
11 Camera 12 Positioning device

Claims (13)

A dictionary database that stores the words that co-occur on verbs representing the operation content and the spatial relative relationships of the objects;
A detection unit that acquires data of a spatial relative relationship with respect to an object existing in the external space;
An input unit for receiving an input of an operation command from a user;
Spatial semantic structure for obtaining a hierarchical structure of words included in an operation command sentence input in the input unit and for obtaining a hierarchical structure including words omitted in the operation command sentence with reference to the dictionary database An analysis unit;
A certainty factor model having a plurality of partial models connected in the hierarchical structure, the certainty factor being obtained as a function of the certainty factor of each partial model, wherein each partial model is a first variable related to the wording, A certainty factor model generating unit that generates a certainty factor model having a second variable group related to a spatial relative relationship between objects, a third variable related to the degree to which the object corresponds to the wording,
Applying the spatial relative relationship data of the objects detected by the detection unit to the second variable group to calculate a certainty factor obtained as a function of the certainty factor of the plurality of partial models, the certainty factor is A collation that determines the spatial relative relationship of the largest object,
An output unit for outputting an analysis result of the operation command statement;
An information processing apparatus comprising:   The information processing apparatus according to claim 1, wherein the certainty model is a probability model.   The information processing apparatus according to claim 1, wherein the detection unit performs coordinate conversion of the position of the detected object into a local coordinate system based on the host vehicle based on map information. The detection unit acquires time-series data of the spatial positional relationship of the object, and when the state of the object has changed, detects the state of the object different from the current acquired in the past as a past object,
4. The spatial semantic structure analysis unit according to any one of claims 1 to 3, wherein, when a word relating to a past object is included, the hierarchical structure is determined by distinguishing a word relating to a current object and a word relating to a past object. The information processing apparatus described in 1. The detection unit obtains time-series data of the spatial positional relationship of the object, and when the state of the object is changed, detects the state of the predicted future object as a future object,
5. The spatial semantic structure analysis unit according to any one of claims 1 to 4, wherein, when a word regarding a future object is included, the hierarchical structure is determined by distinguishing a word regarding a current object and a word regarding a future object. The information processing apparatus described in 1. A dictionary database that stores the words that co-occur on verbs representing the operation content and the spatial relative relationships of the objects;
A detection unit that acquires time-series data of spatial relations related to objects existing in the external space, and detects the state of an object different from the current acquired in the past as a past object when the state of the object has changed When,
An input unit for receiving an input of an operation command from a user;
Referring to the dictionary database, the operation command sentence input in the input unit is analyzed, and when a word related to a past object is included, a word related to a current object and a word related to a past object; Spatial semantic structure analysis unit that determines the hierarchical structure by distinguishing
A certainty factor model having a plurality of partial models connected in the hierarchical structure, the certainty factor being obtained as a function of the certainty factor of each partial model, wherein each partial model is a first variable related to the wording, A certainty factor model generating unit that generates a certainty factor model having a second variable group related to a spatial relative relationship between objects, a third variable related to the degree to which the object corresponds to the wording,
Applying the spatial relative relationship data of the objects detected by the detection unit to the second variable group to calculate a certainty factor obtained as a function of the certainty factor of the plurality of partial models, the certainty factor is A collation that determines the spatial relative relationship of the largest object,
An output unit for outputting an analysis result of the operation command statement;
An information processing apparatus comprising: A dictionary database that stores the words that co-occur on verbs representing the operation content and the spatial relative relationships of the objects;
A time series data of a spatial relative relationship with respect to an object existing in the external space, and when the state of the object is changed, a detection unit that detects a predicted state of the future object as a future object;
An input unit for receiving an input of an operation command from a user;
By referring to the dictionary database, analyzing the operation command sentence input in the input unit, and when a word regarding a future object is included, a word regarding a current object and a word regarding a future object Spatial semantic structure analysis unit that determines the hierarchical structure by distinguishing
A certainty factor model having a plurality of partial models connected in the hierarchical structure, the certainty factor being obtained as a function of the certainty factor of each partial model, wherein each partial model is a first variable related to the wording, A certainty factor model generating unit that generates a certainty factor model having a second variable group related to a spatial relative relationship between objects, a third variable related to the degree to which the object corresponds to the wording,
Applying the spatial relative relationship data of the objects detected by the detection unit to the second variable group to calculate a certainty factor obtained as a function of the certainty factor of the plurality of partial models, the certainty factor is A collation that determines the spatial relative relationship of the largest object,
An output unit for outputting an analysis result of the operation command statement;
An information processing apparatus comprising: An information processing method for performing language processing of an operation command statement for an object existing in an external space,
An information processing device acquiring spatial relative data regarding an object existing in external space;
The information processing apparatus receives an input of an operation command sentence from a user;
The information processing apparatus obtains a hierarchical structure of words included in an input operation command sentence, and refers to a dictionary database that stores words that co-occur in a verb representing operation contents and a spatial relative relationship between objects. A step of obtaining a hierarchical structure including words omitted in the operation command sentence;
The information processing apparatus includes a plurality of partial models connected in the hierarchical structure, and a certainty factor model in which a certainty factor is obtained as a function of the certainty factor of each partial model, wherein each partial model includes the wording Generating a confidence model having a first variable relating to, a second variable group relating to a spatial relative relationship of objects, and a third variable relating to a degree to which the object corresponds to the wording;
The information processing apparatus calculates the certainty factor obtained as a function of the certainty factor of the plurality of partial models by applying the spatial relative relationship data of the detected object to the second variable group, and the certainty factor Determining the spatial relative relationship of the object for which
The information processing apparatus outputting an analysis result of the operation command statement;
An information processing method comprising: An information processing method for performing language processing of an operation command statement for an object existing in an external space,
When the information processing device acquires time-series data of spatial relations regarding objects existing in the external space and the state of the object has changed, the state of the object different from the current acquired in the past is displayed as the past object. Detecting as a
The information processing apparatus receives an input of an operation command sentence from a user;
The information processing apparatus refers to a dictionary database that stores words and words that co-occur on verbs representing operation contents and a spatial relative relationship between objects, analyzes an operation command sentence input by a user, and relates to past objects If wording is included, distinguishing the wording for the current object and the wording for the past object to determine the hierarchical structure;
The information processing apparatus includes a plurality of partial models connected in the hierarchical structure, and a certainty factor model in which a certainty factor is obtained as a function of the certainty factor of each partial model, wherein each partial model includes the wording Generating a confidence model having a first variable relating to, a second variable group relating to a spatial relative relationship of objects, and a third variable relating to a degree to which the object corresponds to the wording;
The information processing apparatus calculates the certainty factor obtained as a function of the certainty factor of the plurality of partial models by applying the spatial relative relationship data of the detected object to the second variable group, and the certainty factor Determining the spatial relative relationship of the object for which
The information processing apparatus outputting an analysis result of the operation command statement;
An information processing method comprising: An information processing method for performing language processing of an operation command statement for an object existing in an external space,
The information processing apparatus acquires time-series data of spatial relations regarding objects existing in the external space, and detects the predicted future object state as a future object when the state of the object has changed. Steps,
The information processing apparatus receives an input of an operation command sentence from a user;
The information processing apparatus refers to a dictionary database that stores words that co-occur on verbs representing operation contents and spatial relations of objects, analyzes an operation command sentence input by a user, and relates to a future object. If wording is included, distinguishing the wording for the current object and the wording for the future object to determine the hierarchical structure;
The information processing apparatus includes a plurality of partial models connected in the hierarchical structure, and a certainty factor model in which a certainty factor is obtained as a function of the certainty factor of each partial model, wherein each partial model includes the wording Generating a confidence model having a first variable relating to, a second variable group relating to a spatial relative relationship of objects, and a third variable relating to a degree to which the object corresponds to the wording;
The information processing apparatus calculates the certainty factor obtained as a function of the certainty factor of the plurality of partial models by applying the spatial relative relationship data of the object detected by the detection unit to the second variable group. And determining a spatial relative relationship of the object having the maximum certainty factor,
The information processing apparatus outputting an analysis result of the operation command statement;
An information processing method comprising: A program for performing language processing of operation statements for objects existing in an external space,
Obtaining spatial relative data for objects present in external space;
A step of accepting an input of an operation command from a user;
The hierarchical structure of the words included in the input operation instruction sentence is obtained, and the operation instruction sentence is omitted in the operation instruction sentence with reference to the dictionary database storing the words that co-occur in the verb representing the operation contents and the spatial relative relationship of the objects. A step to obtain a hierarchical structure including
A certainty factor model having a plurality of partial models connected in the hierarchical structure, the certainty factor being obtained as a function of the certainty factor of each partial model, wherein each partial model is a first variable related to the wording, Generating a confidence model having a second variable group relating to a spatial relative relationship of objects, a third variable relating to a degree to which the object corresponds to the wording;
Applying the spatial relative relationship data of the detected object to the second variable group to calculate a certainty factor obtained as a function of the certainty factor of the plurality of partial models, and for the object having the maximum certainty factor Determining a spatial relative relationship;
Outputting an analysis result of the operation command statement;
A program that executes A program for performing language processing of operation statements for objects existing in an external space,
Acquiring time-series data of spatial relations regarding objects existing in the external space, and detecting a state of an object different from the current acquired in the past as a past object if the state of the object has changed; ,
A step of accepting an input of an operation command from a user;
Referring to the dictionary database that stores the words that co-occur in the verb representing the operation contents and the spatial relative relationship of the objects, the operation command sentence input by the user is analyzed, and the words related to the past objects are included. In some cases, distinguishing the wording about the current object from the wording about the past object to obtain a hierarchical structure,
A certainty factor model having a plurality of partial models connected in the hierarchical structure, the certainty factor being obtained as a function of the certainty factor of each partial model, wherein each partial model is a first variable related to the wording, Generating a confidence model having a second variable group relating to a spatial relative relationship of objects, a third variable relating to a degree to which the object corresponds to the wording;
Applying the spatial relative relationship data of the detected object to the second variable group to calculate a certainty factor obtained as a function of the certainty factor of the plurality of partial models, and for the object having the maximum certainty factor Determining a spatial relative relationship;
Outputting an analysis result of the operation command statement;
A program that executes A program for performing language processing of operation statements for objects existing in an external space,
Obtaining time-series data of a spatial relative relationship with respect to an object existing in the external space, and detecting a predicted future object state as a future object when the state of the object is changed;
A step of accepting an input of an operation command from a user;
Referring to a dictionary database that stores words that co-occur on verbs that represent operations and spatial relations of objects, it analyzes the operation commands entered by the user and includes words about future objects In some cases, determining the hierarchical structure by distinguishing the language of the current object from the language of the future object,
A certainty factor model having a plurality of partial models connected in the hierarchical structure, the certainty factor being obtained as a function of the certainty factor of each partial model, wherein each partial model is a first variable related to the wording, Generating a confidence model having a second variable group relating to a spatial relative relationship of objects, a third variable relating to a degree to which the object corresponds to the wording;
Applying the spatial relative relationship data of the objects detected by the detection unit to the second variable group to calculate a certainty factor obtained as a function of the certainty factor of the plurality of partial models, the certainty factor is Determining the largest spatial relationship of the objects;
Outputting an analysis result of the operation command statement;
A program that executes

Images