CN109918646B - Method, system and device for judging causality of text - Google Patents

Abstract

The invention belongs to the technical field of natural language processing, and in particular relates to a method, system and device for judging causality of texts, and aims to solve the problem of judging causality of texts in robot interaction. The method of the invention includes: based on the language activation model, obtaining the registration event with the highest matching degree for each target text in the input target text pair; based on the registration event corresponding to each target text, according to the stored registration event sequence in each scene , calculate the correlation of the two registered events; based on the target text pair and the correlation of the two registered events, calculate the causal relationship of the target text pair. The present invention can accurately judge the causal relationship of the input target text pair.

Description

Method, system and device for judging causality of text

technical field

The invention belongs to the technical field of natural language processing, and in particular relates to a method, system and device for judging causal relationship of a text.

Background technique

Chapters are almost inseparable from people's daily communication. People often use discourse relations to express and convey the semantic relations between contexts (for example: causal relations, progressive relations, transition relations, etc.). At the same time, as robots gradually play an increasingly important role in people's daily life, how to make robots understand the textual relationship in people's daily expressions has become an unavoidable problem. However, the difficulty of this problem is that there is no good theory and model framework, and the existing methods are limited to the text level and it is difficult to obtain the intrinsic semantics of the text, which leads to many problems in the existing methods.

Recently, some researchers in the field of artificial intelligence have proposed an idea to become contextualized language learning. This idea believes that "the semantics of language mainly comes from the environment". According to this idea, if you want a machine to understand language, the machine should It is perceptible and interactive, and the judgment of textual causality plays an important role in the interaction of robots.

SUMMARY OF THE INVENTION

In order to solve the above problems in the prior art, that is, in order to solve the problem of judging the causal relationship of texts in robot interaction, the first aspect of the present invention proposes a method for judging the causal relationship of texts, and the method includes:

Step S10, based on the language activation model, obtain the registration event with the highest matching degree for each target text in the input target text pair; the target text pair is two input target texts;

Step S20, based on the registration event corresponding to each target text, according to the stored registration event sequence in each scene, calculate the correlation of the two registered events; the registered event sequence in the scene is constructed based on the empirical information of the scene structure. A sequence of registered events with Boolean timing characteristics;

Step S30, based on the target text pair and the correlation of the two registration events obtained in step S20, calculate the causal relationship of the target text pair;

Wherein, the language activation model is obtained by training the machine translation model through experience-language activation training corpus; the experience-language activation training corpus is constructed based on registration events in robot experience.

In some preferred embodiments, in step S30, "calculate the causal relationship of the target text pair", and the method is:

f _r =softmax(tanh(W _c *[s ₁ ; s ₂ ]+W _t *fea _t +b))

Among them, f _r is the causal relationship probability value, including causal relationship probability and non-causal relationship probability; s ₁ and s ₂ are the sentence vectors obtained by the text encoding model of the two target texts in the target text pair; W _c is the preset text The parameter matrix of the vector; W _t is the parameter matrix of the preset Boolean timing characteristics; b is the preset offset.

In some preferred embodiments, step S20 "calculates the correlation of two registration events", and the method is as follows:

Among them, fea _t is the calculated correlation; e ₁ , e ₂ are the registration events matched by the first target text and the second target text in the target text pair, respectively; P(e ₁ ) is the stored structure of all scenes The probability of e ₁ appearing in empirical information; P(e ₂ |e ₁ ) is the appearance probability of e ₂ in the structured empirical information of e ₁ .

In some preferred embodiments, the registration event,

e _i ={obj _i }, i∈R _i

Among them, e _i is the ith registration event; obj _i is the object in the active state in the registration event; R _i is the number of all objects in the language activation model of the object symbol in the registration event.

In some preferred embodiments, the experience-linguistic activation training corpus is expressed as

E _j = {obj _i }: {LS _i }, i∈R _i

Among them, E _j is the jth experience-language activation training corpus; LS _i is the language character corresponding to the object obj _i .

In some preferred embodiments, the sequence of events registered in each scenario described in step S20, the acquisition method thereof includes:

Step A10, in the working environment of the robot, the object information obtained by the interaction between the robot and the object obtained by the robot's own sensing device; the object information includes the object attribute and object state information obtained by the interaction between the robot and the object;

In step A20, the object information obtained in step A10 is subjected to structured processing, and is organized and stored in chronological order, as structured experience information obtained from the operating working environment of the robot.

Step A30, remove the objects in the inactive state in the structured experience information, and obtain the registration event sequence in the corresponding scene.

In some preferred embodiments, the structured empirical information is expressed as

E _0-t ={f ₀ ,f ₁ ,f ₂ ,...,f _t }

f _t ={obj ₁ :a ₁ ;...obj _i :a _i ;...obj _n :a _n ;}

Among them, E _0-t is all experience from time 0 to time t; f _t is all object information input at time t; obj _i is the ith object, a _i is the activation state of object obj _i , a _i = When it is 1, it means that the ith object is activated, and when a _i =0, it means that the ith object is not activated, i∈[1,n].

In a second aspect of the present invention, a system for judging a causal relationship of a text is proposed, which includes a registration event matching module, a correlation computing module, and a causal relationship computing module;

The registration event matching module is configured to obtain the registration event with the highest matching degree for each target text in the input target text pair based on the language activation model; the target text pair is two input target texts;

The correlation calculation module is configured to calculate the correlation of two registered events based on the registered event corresponding to each target text and according to the stored registered event sequence in each scene; the registered event sequence in the scene is based on the scene structure A registration event sequence with Boolean timing characteristics constructed from the transformed empirical information;

The causal relationship calculation module is configured to calculate the causal relationship of the target text pair based on the target text pair and the correlation of the two registered events obtained in step S20;

Wherein, the language activation model is obtained by training the machine translation model through experience-language activation training corpus; the experience-language activation training corpus is constructed based on registration events in robot experience.

In a third aspect of the present invention, a storage device is provided, in which a plurality of programs are stored, and the programs are adapted to be loaded and executed by a processor to realize the above-mentioned method for judging the causality of a chapter.

In a fourth aspect of the present invention, a processing device is provided, including a processor and a storage device; the processor is adapted to execute various programs; the storage device is adapted to store multiple programs; the programs are adapted to be loaded by the processor And execute it to realize the above article causality judgment method.

Beneficial effects of the present invention:

An accurate causal relationship judgment result can be obtained by performing correlation calculation through the registration event sequence of the scene in the present invention, and further combining the registration events of the target text pair to perform the causal relationship calculation of the target text pair.

Description of drawings

Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

1 is a schematic flowchart of a method for judging a causal relationship of a text according to an embodiment of the present invention;

FIG. 2 is an example of judging the causal relationship of a text according to an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not All examples. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the related invention, but not to limit the invention. In addition, it should be noted that, for the convenience of description, only the parts related to the related invention are shown in the drawings.

It should be noted that the embodiments in the present application and the features of the embodiments may be combined with each other in the case of no conflict.

The basic idea of the invention is to use the signal input of the robot working platform to build a solidified network for the robot, and then use the deep neural network technology to achieve better modeling of text semantics to achieve the purpose of final causal relationship judgment.

A method for judging the causal relationship of a text of the present invention, as shown in Figure 1, the method includes:

Step S10, based on the language activation model, obtain the registration event with the highest matching degree for each target text in the input target text pair; the target text pair is two input target texts;

Step S20, based on the registration event corresponding to each target text, according to the stored registration event sequence in each scene, calculate the correlation of the two registered events; the registered event sequence in the scene is constructed based on the empirical information of the scene structure. A sequence of registered events with Boolean timing characteristics;

Step S30, based on the target text pair and the correlation of the two registration events obtained in step S20, calculate the causal relationship of the target text pair;

Wherein, the language activation model is obtained by training the machine translation model through experience-language activation training corpus; the experience-language activation training corpus is constructed based on registration events in robot experience.

In order to more clearly describe the method for judging the causality of the text of the present invention, the following is a detailed description from three aspects: 1. The construction of the registered event sequence in each scene; 2. The construction of the language activation model; 3. The textual causality judge.

1. Construction of registered event sequences in each scene

In step A10, in the working environment of the robot, the object information obtained by the interaction between the robot and the object obtained by the robot's own perception device; the object information includes object attributes and object state information obtained by the interaction between the robot and the object.

The robot is placed in the working environment, and the sensing device of the robot is used to receive the signals in the environment (that is, the object information obtained by the interaction between the robot and the object). The information obtained is continuously updated through the continuous interaction of the robot in the working environment. For example, in a scene, there are a large number of objects interacting with the robot. These objects themselves have properties, and these properties have corresponding states. In particular, after the robot interacts with the object, the state of the object changes. Robots need to perceive these properties and states to obtain information about these objects. For example, the attributes and states of the object "cow" on the robot face are displayed, and the object is denoted as obj2, with name=cow).

In step A20, the object information obtained in step A10 is subjected to structured processing, and is organized and stored in chronological order, as structured experience information obtained from the operating working environment of the robot.

The object information obtained from the interaction between the robot and the object from step A10 is stored in the long-term memory of the robot. The specific implementation method is as follows: the robot stores the collected object attribute state information in the long-term memory. In long-term memory, we keep corresponding boolean nodes for the properties and states of different objects.

In the structuring process, these signals are processed and stored in a structured manner; the specific implementation is as follows: structure points are reserved for these information Boolean nodes inside the robot, so as to save the information. And, the information will be organized and saved in chronological order. That is to say, let the input of a time t be f _t , then the expression of structured empirical information is:

E _0-t ={f ₀ ,f ₁ ,f ₂ ,...,f _t }

Among them, E _0-t is all experience from time 0 to time t; f _t is all the object information (attribute, state) input at time t.

At the same time, the experience at each moment contains the activation information of all objects (represented by Boolean values), and its formal expression is:

f _t ={obj ₁ :a ₁ ;...obj _i :a _i ;...obj _n :a _n }

Among them, obj _i is the ith object, a _i is the activation state of the object obj _i , when a _i = 1, it means that the ith object is activated, when a _i = 0, it means that the ith object is not activated, i∈[ 1,n].

For example, at the fifth moment f ₅ ={obj ₁ :0;obj ₂ :1;obj ₃ :1;obj ₄ :1}, it means that obj ₂ , obj ₃ , and obj ₄ in all objects at this moment are in the active state, obj ₁ is inactive.

Step A30, remove the objects in the inactive state in the structured experience information, and obtain the registration event sequence in the corresponding scene.

For example, for the above f ₅ , the corresponding registration sequence in the scene is T ₅ ={obj ₂ ; obj ₃ ; obj ₄ ;}.

2. The construction of language activation model.

(1) Construct the registration event in the robot experience.

According to the experience of the robot, these objects represent a state of a boolean value, and if the robot receives the corresponding signal in the working environment, the corresponding object will be activated. An event is an activated state when all objects belonging to it are activated. Among them, the registration event expression is:

e _i ={obj _i }, i∈R _i

Among them, e _i is the ith registration event; obj _i is the object in the active state in the registration event; R _i is the number of all objects in the language activation model of the object symbol in the registration event.

For example, the registration event obtained based on the above f ₅ is e ₅ ={obj ₂ ; obj ₃ ; obj ₄ }, at this time R _i =[2,3,4]. When _e5 is activated, it can be expressed as e5:1={obj2: ₁ ;obj3: ₁ ; _obj4 : ₁ ;}.

(2) Based on the constructed registration events, construct an experience-language activation training corpus.

This step can be regarded as a symbol translation process, therefore, the training corpus is in the form of parallel corpus. Specifically, the operation is as follows: another internal object of experience is solidified into a sequence of symbols, for example, obj ₁₅ =Obj_15. When the text refers to the symbol, the symbol and related concepts are activated. For example, in our actual experiment, obj ₁₅ (ie Obj_15) means "cow" in Chinese and "cow" in English. This step constructs a parallel corpus of such linguistic symbols corresponding to empirical internal symbols, as follows:

E _j = {obj _i }: {LS _i }, i∈R _i

Among them, E _j is the jth experience-language activation training corpus; LS _i is the language character corresponding to the object obj _j .

For example, if the language characters corresponding to the objects obj ₂ , obj ₅ , and obj ₁₅ are Adam, attack, and cow, respectively, the training language chat can be expressed as

obj ₂ , obj ₅ , obj ₁₅ : Adam attack cow

(3) The language activation model is obtained by training the machine translation model through the experience-language activation training corpus.

According to the experience-language activation training corpus, the training is based on the machine translation model. In this embodiment, a sequence-to-sequence neural network machine translation model is used, of course, other machine translation models may also be used.

3. Discourse causality judgment

Step S10, based on the language activation model, obtain the registration event with the highest matching degree for each target text in the input target text pair; the target text pair is two input target texts.

The construction process of the language activation model has been described in detail above. Here, the corresponding registration event can be obtained directly by changing the model. The key process of text processing by the model can be shown below.

(1) Translate the input text into the symbol sequence of the registration event through the language activation model, that is, the object symbol contained in the registration event inside the robot. Its formula is as follows:

y ₁ , y ₂ , y ₃ ,...,y _j =P(y _i |x ₁ ,x ₂ ,x ₃ ,...,x _n )

Among them, y ₁ , y ₂ , y ₃ ,…,y _j is the object symbol sequence; x ₁ ,x ₂ ,x ₃ ,…,x _n is the language symbol sequence; P(y _i |x ₁ ,x ₂ ,x ₃ ,...,x _n ) is to decode (ie translate) the object symbols according to the language symbol sequence.

In some embodiments, two word vector sets _Embl and _Embe may be constructed for the linguistic side and the empirical symbol side, respectively.

(2) The language activation model activates the corresponding experience information of the robot according to the input target text. According to the translated object symbol sequence, all registered events stored in the robot are matched, and the event with the highest matching degree is selected as the activated event. Among them, a long and short-term memory model LSTM can be used as a tool for generating symbol sequence vectors to generate sequence vectors of object symbols, as shown in the following formula

LSTM(Emb _e [obj ₁ ],Emb _e [obj ₂ ],...,Emb _e [obj 1 ]) _{= ve}

Then, based on the cosine similarity of the vectors, the event with the highest matching degree is obtained.

Among them, v _e is the vector obtained by the LSTM network and Emb _e of the object symbol sequence after translation of the target text; v _i is the vector obtained by the _ith registration event in the language activation model through the LSTM network and Emb _e ; sim(vi , v _e ) is the cosine similarity between these two vectors.

The registration event with the highest cosine similarity is selected as the activation event, that is, the registration event corresponding to the input target text.

Step S20, based on the registration event corresponding to each target text, according to the stored registration event sequence in each scene, calculate the correlation of the two registered events; the registered event sequence in the scene is constructed based on the empirical information of the scene structure. A sequence of registered events with Boolean timing characteristics.

Remove the objects in the inactive state in the structured experience information, and obtain the registered event sequence in the corresponding scene

When judging causality, there are two sentences entered each time. Therefore, we need to find two registration events at one time. At the same time, according to the obtained events e ₁ , e ₂ , it is necessary to obtain the corresponding Boolean timing characteristics of the two events in the sequence of registered events in each scene stored by the robot, so as to calculate the correlation of the two registered events.

The correlation of two registered events is calculated as:

Among them, fea _t is the calculated correlation; e ₁ , e ₂ are the registration events matched by the first target text and the second target text in the target text pair, respectively; P(e ₁ ) is the structure of all the stored scenes The probability of e ₁ appearing in empirical information; P(e ₂ |e ₁ ) is the appearance probability of e ₂ in the structured empirical information of e ₁ .

Step S30: Calculate the causal relationship of the target text pair based on the target text pair and the correlation between the two registered events obtained in step S20.

The method for calculating the causal relationship of the target text pair is:

f _r =softmax(tanh(W _c *[s ₁ ; s ₂ ]+W _t *fea _t +b))

Among them, f _r is the causal relationship probability value, including the causal relationship probability Cause and the non-causal relationship probability Non-Cause; s ₁ and s ₂ are the sentence vectors obtained by the two target texts in the target text pair through the text encoding model; W _c is The parameter matrix of the preset text vector; W _t is the parameter matrix of the preset Boolean time series feature; b is the preset offset.

In this embodiment, the classification model of softmax is used, and the softmax function will output the probability values of the two classifications in the two-class classification, and take the largest one as the output. For example: softmax will output a probability value Cause: 0.7, Non-Cause: 0.3, then we choose Cause as the output.

The text encoding model can use LSTM (Long Short Term Memory Model), BOW (Bag of Words Model Based on Word Vectors), CNN (Convolutional Neural Network Model), and other models.

As shown in Figure 2, the target text pair input in this example is S1: Adam attacks the cow with sword, S2: and Adam gains some beef. In Part1, the registration events e1: Adam attacks cow sword, e2: Adam gains some beef are obtained through the language activation model. In Part2, based on the perception signal sequence received by the robot in each scene (S plus a number represents the ith scene, each scene consists of many consecutive moments, such as T9-T10-T11, etc., each moment contains A large number of events), find that e1 appears in scenes S87, S72, and e2 appears in scenes S54, S87, S72, and locate the moment of appearance by arrows in the figure (that is, the event locates the perception in the robot's memory In the signal sequence), e1 is located at time T0 of scene S87, time T1 of S72, e2 is located to time T11 of scene S54, time T2 of S87, time T3 of S72, and calculate the correlation of the two registered events; in Part3 Perform text encoding on the target texts S1 and S2 respectively to obtain sentence vectors; in Part4, perform causal judgment based on the correlation obtained in Part2 and the two sentence vectors obtained in Part3, and output the judgment result.

The code realization of the present invention can be realized by using the python programming language and the TensorFlow deep learning framework, and the development platform is Ubuntu Linux 16.04. At the same time, the present invention can also rely on the general artificial intelligence platform Malmo released by Microsoft as the working environment of the robot. Since the written program does not use any platform-dependent code, the described system implementation can also run on the Windows operating system.

A text causality judgment system according to an embodiment of the present invention includes a registration event matching module, a correlation calculation module, and a causal relationship calculation module;

The registration event matching module is configured to obtain the registration event with the highest matching degree for each target text in the input target text pair based on the language activation model; the target text pair is two input target texts;

The correlation calculation module is configured to calculate the correlation of two registered events based on the registered event corresponding to each target text and according to the stored registered event sequence in each scene; the registered event sequence in the scene is based on the scene structure A registration event sequence with Boolean timing characteristics constructed from the transformed empirical information;

The causal relationship calculation module is configured to calculate the causal relationship of the target text pair based on the target text pair and the correlation of the two registered events obtained in step S20;

Wherein, the language activation model is obtained by training the machine translation model through experience-language activation training corpus; the experience-language activation training corpus is constructed based on registration events in robot experience.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, for the specific working process and related description of the system described above, reference may be made to the corresponding process in the foregoing method embodiments, which will not be repeated here.

It should be noted that, the system for judging the causality of the text provided by the above-mentioned embodiment is only illustrated by the division of the above-mentioned functional modules. The modules or steps in the embodiments of the present invention are further decomposed or combined. For example, the modules in the above embodiments may be combined into one module, or may be further split into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing each module or step, and should not be regarded as an improper limitation of the present invention.

In the storage device according to an embodiment of the present invention, a plurality of programs are stored therein, and the programs are adapted to be loaded and executed by a processor to implement the above-mentioned method for judging the causal relationship of a chapter.

A processing device according to an embodiment of the present invention includes a processor and a storage device; the processor is adapted to execute various programs; the storage device is adapted to store multiple programs; the programs are adapted to be loaded and executed by the processor to realize The above chapter causality judgment method.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and relevant description of the storage device and processing device described above can refer to the corresponding process in the foregoing method embodiments, which is not repeated here. Repeat.

Experimental verification:

In the training corpus constructed in the experiments, each example consists of two sentences, each describing a separate event. By training on this corpus, the method of the present invention is compared with traditional plain text-based methods. Table 1 shows the basic statistics of the corpus of textual causality constructed in this experiment.

Schedule 1

causation non-causal relationship total Acceptance quantity 1619 3233 4852 98.2%

Attached table 2 presents the results of the present invention and several existing methods for discriminating the causal relationship of the chapters. Among them, the upper part of the table expresses the three current mainstream neural network methods (BOW, CNN, LSTM) for processing sentence information. The experiment takes two sentences as input and examines whether the target model has the ability to discriminate whether the sentence pair is a causal relationship. It can be seen that the performance of the present invention (Our model) on the data set is significantly better than that of the traditional text method, wherein Model+BOW, Model+CNN, Model+LSTM in Table 2 respectively indicate that BOW, CNN, LSTM text encoding model.

A distinct difference from traditional textual methods is that the present invention uses empirical information gathered from a robotic platform. This process is very similar to that of humans, so the key to the present invention is to propose a new method that is different from traditional thinking, which emphasizes the importance of text-related experience.

Schedule 2

Those skilled in the art should be aware that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps Can be placed in random access memory (RAM), internal memory, read only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or as known in the art in any other form of storage medium. In order to clearly illustrate the interchangeability of electronic hardware and software, the components and steps of each example have been described generally in terms of functionality in the foregoing description. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods of implementing the described functionality for each particular application, but such implementations should not be considered beyond the scope of the present invention.

The terms "first," "second," etc. are used to distinguish between similar objects, and are not used to describe or indicate a particular order or sequence.

The term "comprising" or any other similar term is intended to encompass a non-exclusive inclusion such that a process, method, article or device/means comprising a list of elements includes not only those elements but also other elements not expressly listed, or Also included are elements inherent to these processes, methods, articles or devices/devices.

So far, the technical solutions of the present invention have been described with reference to the preferred embodiments shown in the accompanying drawings, however, those skilled in the art can easily understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims (10)

1. a method for judging causal relationship of a chapter, is characterized in that, the method comprises: Step S10, based on the language activation model, obtain the registration event with the highest matching degree for each target text in the input target text pair; the target text pair is two input target texts; Step S20, based on the registration event corresponding to each target text, according to the stored registration event sequence in each scene, calculate the correlation of the two registered events; the registered event sequence in the scene is constructed based on the empirical information of the scene structure. A sequence of registered events with Boolean timing characteristics; Step S30, based on the target text pair and the correlation of the two registration events obtained in step S20, calculate the causal relationship of the target text pair; Wherein, the language activation model is obtained by training the machine translation model through experience-language activation training corpus; the experience-language activation training corpus is constructed based on registration events in robot experience. 2. The method for judging a causal relationship of a text according to claim 1, wherein in step S30, "calculate the causal relationship of the target text pair", and its method is: f _r =softmax(tanh(W _c *[s ₁ ; s ₂ ]+W _t *fea _t +b)) Among them, f _r is the causal relationship probability value, including causal relationship probability and non-causal relationship probability; s ₁ and s ₂ are the sentence vectors obtained by the text encoding model of the two target texts in the target text pair; W _c is the preset text The parameter matrix of the vector; W _t is the parameter matrix of the preset Boolean timing characteristics; b is the preset offset. 3. The method for judging the causal relationship of a text according to claim 1, wherein the step S20 "calculates the correlation of two registered events", and its method is:

Among them, fea _t is the calculated correlation; e ₁ , e ₂ are the registration events matched by the first target text and the second target text in the target text pair, respectively; P(e ₁ ) is the structure of all the stored scenes The probability of e ₁ appearing in empirical information; P(e ₂ |e ₁ ) is the appearance probability of e ₂ in the structured empirical information of e ₁ . 4. The method for judging a causal relationship of a text according to claim 1, wherein the registration event, e _i ={obj _i }, i∈R _i Among them, e _i is the ith registration event; obj _i is the object in the active state in the registration event; R _i is the number of all objects in the language activation model of the object symbol in the registration event. 5. The method for judging a causal relationship of a text according to claim 4, wherein the experience-language activation training corpus is expressed as E _j = {obj _i }: {LS _i }, i∈R _i Among them, E _j is the jth experience-language activation training corpus; LS _i is the language character corresponding to the object obj _i . 6. The method for judging a causal relationship of a text according to claim 1, wherein the sequence of events registered in each scenario described in step S20, the acquisition method thereof comprises: Step A10, in the working environment of the robot, the object information obtained by the interaction between the robot and the object obtained by the robot's own sensing device; the object information includes the object attribute and object state information obtained by the interaction between the robot and the object; In step A20, the object information obtained in step A10 is subjected to structured processing, and is organized and stored in chronological order, as structured experience information obtained by the operating working environment of the robot; Step A30, remove the objects in the inactive state in the structured experience information, and obtain the registration event sequence in the corresponding scene. 7. The method for judging a causal relationship of a discourse according to claim 6, wherein the structured empirical information is expressed as E _0-t ={f ₀ , f ₁ , f ₂ , . . . , f _t } f _t = {obj ₁ : a ₁ ; ...obj _i : a _i ; ... obj _n : a _n ;} Among them, E _0-t is all experience from time 0 to time t; f _t is all object information input at time t; obj _i is the ith object, a _i is the activation state of object obj _i , a _i = When it is 1, it means that the ith object is activated, and when a _i =0, it means that the ith object is not activated, i∈[1,n]. 8. A textual causality judgment system, characterized in that the system comprises a registration event matching module, a correlation calculation module, and a causal relationship calculation module; The registration event matching module is configured to obtain the registration event with the highest matching degree for each target text in the input target text pair based on the language activation model; the target text pair is two input target texts; The correlation calculation module is configured to calculate the correlation of two registered events based on the registered event corresponding to each target text and according to the stored registered event sequence in each scene; the registered event sequence in the scene is based on the scene structure A registration event sequence with Boolean timing characteristics constructed from the transformed empirical information; The causal relationship calculation module is configured to calculate the causal relationship of the target text pair based on the target text pair and the correlation between the two registered events obtained by the correlation calculation module; Wherein, the language activation model is obtained by training the machine translation model through experience-language activation training corpus; the experience-language activation training corpus is constructed based on registration events in robot experience. 9 . A storage device, wherein a plurality of programs are stored, wherein the programs are adapted to be loaded and executed by a processor to implement the method for judging the causality of a chapter according to claims 1-7 . 10. A processing device, comprising a processor and a storage device; the processor is adapted to execute various programs; the storage device is adapted to store a plurality of programs; characterized in that the programs are adapted to be loaded and executed by the processor to The method for judging the causal relationship of the chapters described in claims 1-7 is realized.

Images