KR20160131730A - System, Apparatus and Method For Processing Natural Language, and Computer Readable Recording Medium - Google Patents

Abstract

The present invention relates to a system, an apparatus, and a method for processing a natural language and a computer readable recording medium, capable of dividing and creating short sentences based on the boundaries between sentences. According to an embodiment of the present invention, the natural language processing apparatus includes a storage unit for storing corpus information including an entity name category to which an entity name word in a sentence belongs and an identifier for distinguishing a boundary between sentences, a communication interface unit for receiving a compound sentence or a plurality of short sentences input into a user device, and a natural language processing unit for creating a plurality of short sentences corresponding to the compound sentence or a complex sentence using the identifier of the corpus information.

Description

TECHNICAL FIELD [0001] The present invention relates to a natural language processing system, a natural language processing apparatus, a natural language processing method, and a computer readable recording medium,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language processing system, a natural language processing apparatus, a natural language processing method and a computer readable recording medium, and more particularly to a natural language processing system such as a voice conversation system, A natural language processing apparatus, a natural language processing method, and a computer readable recording method for dividing a short sentence based on a sentence boundary determined by applying a category of an object name word to a statistical translation technique, Media.

Typically, machine translation means that the computer system automatically translates the natural language sentence F of the input language into the natural language sentence E of the target language. Among the machine translation techniques, statistical machine translation learns machine translation model based on training data and performs machine translation based on learned model. More specifically, when F is given, it is a process to find E that maximizes the probability of E (E | F). That is, E is the best translation result for F. This can be expressed as Equation (1).

Equation (2) can be obtained by decomposing Pr (E | F) by applying Bayes' rule to Equation (1) above.

Pr (F | E) is the translation model's probability of being translated into F when E is given, and indicates how appropriate it is to translate E into F. Translation models are learned based on training data for bilinguals.

Here, Pr (E) is the language model and represents the probability that E will appear in the language, and indicates how natural E is. Language models are learned based on training data for a single language (monolingual).

The conventional natural language processing system analyzes the morpheme information, the syntax structure, and the meaning in the input sentence. At this time, one input sentence is a basic sentence of minimum size or a sentence composed of plural basic sentences, that is, a compound sentence.

Basic sentences composing a compound sentence are connected to each other in various forms.

For example, consider a natural language processing system that recognizes and executes voice commands related to TV programs.

TV users can speak the compound sentence "Record OCN news" and "Show me Family Guy" as a natural language processing system. The sentence "Record OCN news" and "Show me Family Guy" It is connected by a connective.

Also, in some languages such as Korean, sentences may be transformed when a sentence is connected by a conjunction. TV users can spell out the compound sentence "Record Infinite Challenge and play 1 night and 2 days" in the natural language processing system. This compound sentence is composed of basic sentence "Let me record infinite challenge" and " Is connected by a conjunction called "high ".

On the other hand, a TV user can utter a compound sentence called "Record OCN news show me Family Guy" as a natural language processing system. This compound sentence occurs when a TV user fires two sentences without a connection.

However, since such a complex sentence is difficult to be processed by a conventional natural language processing system, the performance of the natural language processing system is deteriorated.

The embodiment of the present invention applies, for example, the category of the object name word to the statistical translation technique in the input Chinese sentence when the Chinese language is changed into the short sentence in a natural language processing system such as a voice conversation system, a query response system and a chat system A natural language processing apparatus, a natural language processing method, and a computer readable recording medium in which a short sentence is divided and generated based on a determined sentence boundary.

A natural language processing system according to an embodiment of the present invention is a natural language processing system for inputting a sentence of a Chinese sentence or a complex sentence, and, when the Chinese sentence or the sentence is inputted, And a natural language processing device for generating a short sentence and providing the generated plurality of short sentences to the user device, wherein the corpus information includes an identifier for distinguishing an object name category and a sentence boundary to which the object name word belongs in the sentence .

The natural language processing apparatus according to an embodiment of the present invention may further include a storage unit for storing corpus information including an object name category to which the object name word in the sentence belongs and an identifier for distinguishing the boundary between the sentences, And generating a plurality of short sentences corresponding to the Chinese or Japanese sentences using the stored identifiers of the corpus information when the Chinese or Chinese sentences are input, And a natural language processing unit for providing the natural language processing unit to the user apparatus.

Wherein the natural language processing unit judges the category of the entity name word in the received Chinese sentence or the complex sentence, extracts corpus information of the storage unit including the determined entity name category, and extracts the plurality of short phrases based on the extracted corpus information Can be generated.

Wherein the natural language processing unit comprises an entity name recognition execution unit for determining a category of the entity name word in the middle or mixed sentence and a control unit for changing the entity name word of the Chinese character to the determined entity name category, And a statistical translation executing unit for obtaining the corpus information and generating the plurality of short phrases based on the acquired corpus information.

The statistical translation executing unit may generate the plurality of short sentences by dividing the Chinese sentence or the plural sentences based on the identifier and restoring the divided Chinese sentences or the plural sentences into the plurality of short sentences.

The natural language processing unit may generate the plurality of short sentences in the same language as the middle or complex sentences and provide the same to the user apparatus.

The natural language processing unit may translate the plurality of short sentences in a language other than the Chinese or Chinese sentences, and may generate a translated sentence including the conjunction and provide the translated sentence to the user device.

The natural language processing unit may determine the middle language when a plurality of short sentences connected without the conjunction at a predetermined time interval are continuously provided.

Wherein the natural language processing unit generates a short sentence based on the corpus information related to the object name category to which the object name word in the inputted short or similar short sentence belongs, can do.

The natural language processing unit can acquire the same corpora information if the object name categories of the first and second Chinese manuscripts including different object name words coincide with each other.

The identifier may include symbol or bit information.

Further, a natural language processing method according to an embodiment of the present invention includes storing corpus information including an entity name category to which an object name word in a sentence belongs and an identifier for distinguishing a boundary between sentences, a Chinese word or a sentence And generating a plurality of short sentences corresponding to the middle or mixed sentences by using the stored identifiers of the corpus or corpus information when the middle or mixed sentences are input, .

Wherein the step of providing the corpus information to the user device comprises the steps of: determining a category of the entity name word in the received Chinese or sentence; extracting corpus information of the storage unit including the determined entity name category; And generating the plurality of short sentences based on the plurality of short sentences.

Wherein the step of providing to the user device comprises the steps of: determining the category of the entity name word in the middle or complex sentence; changing the entity name word of the middle or complex sentence to the determined entity name category; Acquiring the related corpus information, and generating the plurality of short phrases based on the acquired corpus information.

The step of generating the plurality of short sentences may include dividing the middle or complex sentences based on the identifiers, and restoring the divided sentences into the plurality of short sentences.

The step of providing to the user device may generate the plurality of short sentences in the same language as the Chinese or Chinese sentences and provide the same to the user device.

The step of providing to the user device may include the step of determining, when the plurality of short sentences connected without the conjunction are continuously provided at the set time intervals, as the middle sentence.

The step of providing to the user device may include acquiring the same corpus information if the categories of entity names of the first and second Chinese characters including the different entity name words coincide with each other.

The identifier may include symbol or bit information.

Meanwhile, a computer-readable recording medium according to an embodiment of the present invention is a computer-readable recording medium including a program for executing a natural language processing method, the computer program product comprising: Storing a corpus information including a discriminating identifier, receiving a sentence of a Chinese or a Chinese sentence input to the user apparatus, and, when the Chinese sentence or the sentence is inputted, using the identifier of the stored corpus information, Generating a plurality of short sentences corresponding to the complex sentences, and providing the generated plurality of short sentences to the user apparatus.

1 is a diagram showing a natural language processing system according to an embodiment of the present invention;
Fig. 2 is a view showing the detailed structure of the natural language processing apparatus of Fig. 1,
Fig. 3 is a view showing another detailed structure of the natural language processing apparatus of Fig. 1,
FIG. 4 is a diagram showing an example of basic and compound sentence parallel corpus information in FIG. 3;
FIG. 5 is a diagram showing a modified example of an input sentence processed by the natural language processing unit of FIG. 2 or the natural language processing module of FIG. 3;
6 is a diagram illustrating a natural language process according to an embodiment of the present invention;
7 is a flowchart showing a natural language processing method according to the first embodiment of the present invention, and
8 is a flowchart showing a natural language processing method according to the second embodiment of the present invention.

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

1 is a diagram showing a natural language processing system according to an embodiment of the present invention.

1, the natural language processing system 90 according to the embodiment of the present invention may include some or all of the user apparatus 100, the communication network 110, and the natural language processing apparatus 120. [

Including some or all of them here means that at least one component of the network 110 and the natural language processing device 120 is omitted so that the user device 100 performs the natural language processing operation stand-alone, P2P means that the user device 100 and the natural language processing device 120 perform a natural language processing operation in cooperation with a network device in the network 110. In addition, The present invention is not limited to the above-described embodiments.

User device 100 may include display devices such as DTVs, smart phones, desktop computers, laptop computers, tablet PCs and wearable devices capable of searching, voice chatting, querying and chatting capabilities. Of course, any device may be used as long as the above functions are possible even if it is not a display device. For example, assuming a query response, the user device 100 receives a text or voice query from a user requesting an answer through a search window or a microphone, and the received query is transmitted to the natural language processing device (Not shown). At this time, the user device 100 may provide the text-based recognition result to the natural language processing device 120. [ For example, when receiving a voice as a query, the user device 100 receives a voice query through a voice receiver such as a microphone, recognizes a voice query received using a speech engine such as * -Voice The recognition result can be output in a text-based manner.

However, since the natural language processing apparatus 120 can have an engine or program having superior performance as compared with the user apparatus 100, it is more preferable that the natural language processing apparatus 120 generates the text-based recognition result. In other words, the user device 100 transmits only voice signals received through the microphone, and the natural language processing device 120 generates voice recognition and text-based recognition results based on the received voice signals. Therefore, in the embodiment of the present invention, how the recognition result is processed is not particularly limited.

In accordance with an embodiment of the present invention, the user device 100 may receive various types of queries from a user. Here, receiving various types of queries means words and sentences in brief. However, more precisely, even when a word is received, when a plurality of words are received, or when a word is received in the form of a sentence . ≪ / RTI > At this time, in the embodiment of the present invention, the case of the sentence is more preferable. Here, the sentence includes a short sentence, a middle sentence and a complicated form, and may further include a form by a combination of these. The form may include not only a Chinese sentence, but also other cases similar to Chinese sentences. In addition, if the two sentences are related to each other by the relative pronoun, the compound sentence according to the embodiment of the present invention means a sentence in the form of Chinese or Chinese. For example, 'Record OCN News' and 'Show Family Guy' are short paragraphs, respectively. When each of these short sentences is connected to a conjunction such as a consonant conjunction, it forms a Chinese sentence. In other words, "Record OCN News, Show Family Guy" is a Chinese sentence.

However, such conjunctions can be varied in various forms. A good example of this is when you have recorded OCN news, then show Family Guy, and after you have recorded OCN news, show Family Guy. Thus, the conjunction can be varied in various forms. Furthermore, two short sentences can be provided in succession at regular time intervals without conjunction. For example, there are two short sentences: 'Record OCN News and Show Family Guy'. According to an embodiment of the present invention, the user device 100 may receive various forms of Chinese text as described above in the form of text or voice.

In this way, whether the sentence, that is, a short sentence or a middle sentence, provided by the user, the user apparatus 100 can receive a user command by receiving a plurality of sentences having a textual form, for example, a short sentence, For example, suppose you record the OCN News and Show Family Guys above. You will receive two short sentences: 'Record OCN News' and 'Show Family Guy'. At this time, although each short sentence is preferably the same language, it may be provided in a different language form. This could be useful for chat situations where different languages are used. In this way, the user device 100 can easily perform an operation on a query uttered by the user by receiving the Chinese text sentence provided by the user in a short form recognizable by the user device 100 through the natural language processing device 120 You can do it.

The communication network 110 includes both wired and wireless communication networks. Here, the wired network includes an Internet network such as a cable network or a public switched telephone network (PSTN), and the wireless communication network includes means such as CDMA, WCDMA, GSM, Evolved Packet Core (EPC), Long Term Evolution (LTE) to be. Of course, the communication network 110 according to the embodiment of the present invention is not limited to this, and it may be used as an access network of a next generation mobile communication system to be implemented in future, for example, in a cloud computing network under a cloud computing environment. For example, when the communication network 110 is a wired communication network, the access point in the communication network 110 can access the exchange of a telephone office, and in the case of a wireless communication network, the access point can access the SGSN or GGSN (Gateway GPRS Support Node) Or may be connected to various repeaters such as Base Station Transmission (BTS), NodeB, and e-NodeB to process data.

The communication network 110 may include an access point. The access point includes a small base station such as a femto or pico base station, which is installed in a large number of buildings. Here, the femto or pico base station is classified according to the maximum number of user devices 100 that can be connected on the classification of the small base stations. Of course, the access point includes a user equipment 100 and a short-range communication module for performing short-range communication such as ZigBee and Wi-Fi. The access point may use TCP / IP or RTSP (Real-Time Streaming Protocol) for wireless communication. In this case, the short-range communication is performed by various standards such as RF (Radio Frequency) and UWB (Ultra Wide Band) communication such as Bluetooth, Zigbee, IrDA, UHF and VHF . Accordingly, the access point can extract the location of the data packet, specify the best communication path to the extracted location, and forward the data packet to the next device, e.g., the user device 100, along the designated communication path. The access point may share a plurality of lines in a general network environment, and may include, for example, a router, a repeater, and a repeater.

The natural language processing device 120 quickly predicts the boundary between two sentences for an input sentence received from the user device 100 in the form of a short or middle sentence, or more precisely a sentence in the form of a Chinese sentence, and a simplified machine language use. Here, the simplified machine language can be regarded as a language used internally to change the input Chinese sentence into the form of a short divided sentence. For example, the natural language processing device 120 may be provided with a sentence such as "Letting KBS know who the main character of Harry Potter is, KBS play" and "Letting KBS go with the wind" If you have learned that 'Harry Potter' and 'Destroy with the Wind' are learned by learning to belong to the object name category (or object name attribute, object type type) (Or substitution, translation) in the form of a sentence, and apply the statistical translation technique on the basis thereof to quickly generate a plurality of short sentences. Here, the object name category means the category to which the object name word belongs in the sentence. For example, the object name or word name may be 'Harry Potter', 'disappear with the wind', 'OCN' as a channel name, '* Bama' as a character name, Can be classified into a movie, a channel name, a character, and the like. However, in other sentences the category of the object name 'Harry Potter' could be a person.

Here, the statistical (mechanical) translation technique outputs a Chinese sentence B that has been trained (or learned) in A, when various types of Chinese sentences, or more precisely, a Chinese sentence tagged with an object name category are A, In other words, as a result of the processing of the A sentence, a sentence B in which a plurality of short sentences are connected to a specific identifier is generated. For example, the Chinese character B can be an identifier for two short sentences, such as '@ movie record # @ movie show' inserted with a sentence separator. Based on the sentence separator, the natural language processing device 120 predicts the boundaries of the two short sentences and generates two short sentences B such as "Record Harry Potter" and " To the user device 100, or to change it to a different language and provide it to the user device 100. Or may translate the two short sentences into another language and then reconnect them to the user and provide them to the user device 100. Here, although the identifier is described as a symbol, it may be in the form of bit information, and thus is not particularly limited. Also, considering that the translation result may be changed in an ungrammatical or word order, the above 'Record @ movie' and 'Show @ movie' may be in the form of 'Record @ Movie # @ Movie'. Therefore, in the embodiment of the present invention, the output form of the above result is not particularly limited.

For example, when a user provides a text in the form of a text through a search window of the user device 100, the natural language processing device 120 can use the text as it is. However, in the case where a speech signal is provided by speech, a text-based recognition result may be obtained by using the speech engine of the present invention. Of course, if the user device 100 recognizes such a voice sentence and provides a text based result, the process may be omitted. Therefore, in the embodiment of the present invention, the manner in which the natural language processing apparatus 120 receives the sentences provided by the user and provides the divided two short sentences to the user apparatus 100 is not particularly limited I will not.

More specifically, the natural language processing apparatus 120 according to an embodiment of the present invention may store the entity name corpus information for determining the category of the entity name, that is, the attribute of the inputted sentence. For example, "Harry Potter" in the input sentence could be a movie, but it could also refer to a person. Accordingly, the natural language processing apparatus 120 performs training and learning based on a lot of entity name corpus information stored therein, thereby quickly distinguishing categories of entity names in the corresponding input sentences. In other words, the object name recognition model can be learned based on the entity name corpus information, and the object name category can be quickly and automatically found for various short and middle sentences inputted based on the learned object name recognition model. This process becomes the process of recognizing the object name.

Also, in order to apply the Chinese name tagged with the object name, or more precisely, the object name category to the statistical translation technique, the natural language processing device 120 may store various parallel corpus corpus information mixed with short and middle Chinese words. Even in the case of the parallel corpus information, it is also possible to train the parallel corpus information so that a fast segmented short phrase conversion can be performed on the input sentence, that is, the sentence having the form replaced with the internal machine language according to the embodiment of the present invention . For example, a statistical translation model is trained on the basis of parallel corpus information, and an input sentence tagged with an object name category is divided and restored into a plurality of basic sentences based on a trained statistical translation model.

Here, restoration is a process of finding an original sentence for an input sentence shifted to the above-mentioned various forms. For example, a statistical translation model trained by input corpus information may output a plurality of basic sentences directly through the execution of a mathematical algorithm or program. Such restoration may also allow for various operations. For example, when replacing the input Chinese name with the object name category, the object name is changed to the object name before the change in the replaced object name category in the divided sentence because the object name is known through a separate storage. In the case of Korean, For example, if the information is matched to be changed to "after" or "after" as a result of searching the DB, the information may be changed That is what you should do. However, in the embodiment of the present invention, such an operation is preferably processed by a translation model. As described above, various restoration methods may be possible, so that the above description is not particularly limited.

In summary, in the system training step, a basic sentence, that is, a sentence composed of a plurality of basic sentences automatically or manually from a short sentence, that is, a composite sentence or a middle sentence, is constructed The statistical translation model is learned based on training data consisting of pairs of sentences. In the system implementation phase, a trained statistical translation model is applied to the input sentence so that multiple basic sentences can be divided and restored from the compound sentence.

Meanwhile, up to now, the natural language processing apparatus 120 has applied the statistical translation model to output the trained corpus information in the input sentence, and divides and restores the input sentence based thereon. For this purpose, the natural language processing device 120 determines the category of the object name in order to improve the performance, and executes a statistical translation model based on the category. However, in the embodiment of the present invention, even if it is not a statistical translation model, separate corpus information trained thereby may be stored in the DB for use. In other words, the object name category is determined in the input sentence, that is, the short sentence or the middle sentence, and the DB is searched based on the determined object name category to extract the matching corpus information. Then, it can be divided and restored into a plurality of short sentences based on the extracted identifier of the corpus information. Therefore, in the embodiment of the present invention, the above translation model is not particularly limited.

Also, the natural language processing unit 120 may process the input sentences in different ways after determining whether the inputted sentences are short sentences or middle sentences. For example, in the case of a short text, a conventional natural language processing method can be used, and a technique according to an embodiment of the present invention can be applied only to a Chinese text. This will save the cost of building the system. However, for the sake of efficiency, it is preferable that the embodiment of the present invention is designed such that when both short and middle sentences are provided as input sentences, the system is designed to be able to process all of them.

As a result of the above configuration, the embodiment of the present invention can divide and restore a long and complicated input sentence into a plurality of short sentences, thereby improving the performance of the system. In other words, it is possible to improve the system performance because the system which can only process the existing short text can process both short and long text forms.

FIG. 2 is a view showing the detailed structure of the natural language processing apparatus of FIG. 1, and it is exemplified that the natural language processing apparatus is constituted by hardware.

2, a natural language processing apparatus 120 according to an embodiment of the present invention includes a communication interface unit 200, a natural language processing unit 210, and a storage unit 220, .

Here, including some or all of them may mean that some components such as the communication interface unit 200 are omitted, or that some components such as the storage unit 220 are integrated with other components such as the natural language processing unit 210 And the like, and the description will be made in order to facilitate a sufficient understanding of the invention.

The communication interface unit 200 receives the natural language of the short or middle Chinese characters provided by the user device 100. [ At this time, the natural language may be a sentence provided by an operation such as a search, a query response, or a chat of the user device 100, or may be performed through speech recognition in the case of a search or a query response. In this case, it is preferable that a text-based recognition result is provided as an input sentence, but when a voice signal is simply input, the natural language processing unit 210 may be provided with a text-based recognition result.

The communication interface unit 200 may provide the input natural language to the natural language processing unit 210 and may transmit the result processed by the natural language processing unit 210 to the user device 100 under the control of the natural language processing unit 210. Here, the processed result can be provided in the same language as a plurality of short sentences divided into input compound sentences, i.e., Chinese sentences. Or a plurality of short sentences changed to another language. Furthermore, a plurality of short sentences changed to another language may be provided as a conjunction, that is, a conjunction of an initially input compound sentence. Of course, the conjunction will also be the same as the translation language. For example, the Korean word '~' or 'and' may be changed to ',' or 'and' in English. Based on this, the user device 100 may perform an operation according to the voice utterance of the user, a query response may be performed, and a chat operation may be possible.

The natural language processing unit 210 performs the entity name recognition operation on the inputted various types of Chinese characters. For example, as described above, the categories of the object names of the inputted sentences are determined based on the entity name corpus information stored in the storage unit 220 and the recognition model. At this time, in the embodiment of the present invention, it is found that in the inputted sentence, for example, "Harry Potter" or "disappear with the wind" belongs to the category of the movie.

In addition, the natural language processing unit 210 can substitute the input Chinese characters using the category information of each entity name with a machine language for applying the statistical translation model to the input Chinese character. In other words, if you enter the Chinese phrase "Record Harry Potter and show up with the wind", you know that it belongs to the category of "Harry Potter" and "Gone with the Wind" Change it in the form of 'show movie'. Such data may be processed substantially in the form of bit information using '0' and '1'. These matters are not particularly limited because they can be changed by the system designer as much as possible. In other words, in the case of processing with bit information, it may be more useful when using a DB instead of a mathematical translation model.

Of course, the natural language processing unit 210 may be replaced with the same data information as in the above case, for example, even if a Chinese word " Record Harry Potter and show love and soul " is input. In other words, the embodiment of the present invention is not intended to accurately recognize the entity name through the entity name recognition process, but rather to divide the Chinese character input through the generalized category information of the entity name and the statistical translation technique into a plurality of short sentences will be.

In this way, the natural language processing unit 210 extracts the complex sentence corpus information about the input sentence tagged with the category information by applying the trained statistical translation model to the basic and compound sentence parallel corpus corpus information, and generates a plurality of short sentences . For example, the natural language processing unit 210 applies a translation model and displays '@ movie record' and 'show @ movie' for 'record @ movie' and 'show @ movie'. Then, the natural language processing unit 210 determines the boundary of the sentence based on this, and generates two independent short sentences according to the determination result. For example, you would create two shorts: "Record Harry Potter" and "Show me how to disappear with the wind."

In this way, the natural language processing unit 210 can recognize the category information of the object name of the input sentence, replace it with a simple machine language, apply the substituted information, that is, the machine language to the statistical translation model, do.

Until now, the natural language processing unit 210 has output a result through a mathematical algorithm such as an entity recognition model or a statistical translation model. That is, since these models are trained models for various information, it is possible to output the results quickly for various types of input sentences. However, as described above, according to the embodiment of the present invention, the object name is searched in the DB to extract information on the object name category, and information on the extracted plurality of object name categories is retrieved from the DB storing the foot ball information The related corpus information can be extracted. It is also possible to divide and restore the sentence based on the identifier information of the extracted corpus information, so that the embodiment of the present invention is not limited to the above contents.

Meanwhile, the natural language processing unit 210 includes a control unit and a natural language processing execution unit, which are not shown in a separate drawing, and the control unit may include a physically separated CPU and memory. The natural language processing execution unit may include a program for executing natural language processing according to an embodiment of the present invention. Accordingly, at the time of initial operation of the system, the CPU can bring the program of the natural language processing execution unit into the memory and perform the natural language processing execution operation. Of course, even if this is not the case, it is possible to receive the processing result only after the CPU executes the natural language processing execution unit, so that the embodiment of the present invention is not limited to the above contents.

The storage unit 220 includes a hardware such as a memory or a DB, and a software storage unit such as a registry. The storage unit 220 may store the entity name corpus information, basic and compound sentence parallel corpus information as described above. Actually, such information is output under the control of the natural language processing unit 210. However, the storage unit 220 may provide all the corresponding information at the time of initial operation of the system in response to a request of the natural language processing unit 210, And thus the information is not particularly limited in how the information is processed in the embodiment of the present invention. In this regard, the storage unit 220 may be omitted from the configuration of the apparatus and integrated into the natural language processing unit 210. [

Meanwhile, the communication interface unit 200, the natural language processing unit 210, and the storage unit 220 are configured as hardware modules physically separated from each other. However, each module stores software for performing the above operations and executes the software It will be possible. However, the corresponding software is a set of software modules, and each module can be formed of hardware, so it will not be limited to the configuration of software or hardware. For example, the storage unit 220 may be a hardware storage or a memory. However, since it is possible to repository information in a software way, it is not limited to the above contents.

Other details will be described with reference to FIGS. 4 and 5. FIG.

Fig. 3 is a diagram showing another detailed structure of the natural language processing apparatus of Fig. 1, in which the natural language processing apparatus is constituted by software, and Fig. 4 is an example of basic and compound sentence parallel corpus information .

Referring to FIG. 3, a natural language processing apparatus 120 'according to another embodiment of the present invention may include a natural language processing module 300 and a storage module 310. Referring to FIG. This may be the natural language processing unit 210 and the storage unit 220 of FIG.

2, if the natural language processing module 300 of FIG. 3 corresponds to the communication interface unit 200 and the natural language processing unit 210 of FIG. 2, the storage module 310 may be a storage unit 220, respectively.

According to the embodiment of the present invention, the natural language processing module 300 may include only the entity name recognition execution unit 300-1 and the statistical translation execution unit (or translation execution unit) 300-2, The recognition model 300-3, the object name recognition training unit 300-5, the statistical translation model 300-7, and the statistical translation training unit 300-9. Here, the inclusion of a part or the whole is the same as the preceding meaning.

Based on the entity name recognition model 300-3, the entity name recognition execution unit 300-1 automatically finds the entity name, or more precisely, the category of the entity name in the input sentence. For example, in the sentence "Do you know who starred in Harry Potter?", It automatically finds that the word "Harry Potter" is a movie.

To do this, the entity name recognition training unit 300-5 generates an entity name recognition model 300 - 1 based on the entity name corpus or the corpus information 310-1 included in the storage module 310, 3) to learn (or train).

For such object name recognition training, the entity name corpus, that is, corpus information (310-1) generated by the system designer or the user is required. The entity name corpus 310-1 may be composed of lists of sentences in which the part corresponding to the entity name is indicated by the entity name tag. For example, the sentence "Who starred in <movie> Harry Potter </ movie>?" In the object name corpus means that the word "Harry Potter" belongs to the object name category movie.

In addition, the statistical (machine) translation executing unit 300-2 divides and restores a tagged object name category into a plurality of basic sentences based on the statistical translation model 300-7. For example, you can split and restore the composite sentence "record @movie and show me @movie" into the basic sentences "record @movie" and "show me @movie". In the process, it is possible to convert it to the original form and output it. In other words, you can print the sentence in the form of "record Harry Potter" and "show me Gone with the Wind".

In addition, the statistical translation execution unit 300-2 may be configured so that even when a plurality of basic sentences form a compound sentence, i.e., a Chinese sentence, such as "@movie record and play @movie" It can be split and restored to the basic sentence "@movie record" and "@movie play" and output.

Furthermore, the statistical translation executing unit 300-2 can maintain a basic sentence when a basic sentence, that is, a short sentence is input. This case may also be provided based on the statistical translation model 300-7 trained in the basic and compound sentence parallel corpus information 311. [

For the above execution, the statistical translation executing unit 300-2 substitutes the object name value of the input sentence corresponding to the portion corresponding to the object name in the sentence of the basic and compound sentence parallel corpora information 311 into the object name category do. For example, if "OCN news" in the "Record OCN news" category is the movie object name category, replace it with "Record @movie".

In this case, the object name substitution is intended to improve the performance of statistical translation for word segmentation and restoration by grouping the words constituting the object name. This has been fully explained above.

For such an implementation, the natural language processing module 300 trains the statistical translation model 300-7 through the statistical translation training section 300-9. For this, basic and complex sentence parallel corpus information 311 is required.

The basic and complex sentence parallel corpus information 311 forms a list by combining a plurality of basic sentences and a corresponding compound sentence. This is as shown in FIG. In Figures 4 (a) to 4 (c), the input corresponds to one compound sentence in the list. This is substituted information. The output also corresponds to a plurality of basic sentences matching the substituted information. Item 1 (item 1) represents a case where a compound sentence is substituted without a conjunction, and item 2 (an item 2) is an example of a compound sentence including a conjunction (and). Item 3 (item 3) indicates a case where a short text is inputted. As described above, the corpus information 311 may be stored in the DB in the form of 2-bit information.

 4, a plurality of basic sentences of the basic and compound sentence parallel corpus information 311 are connected to a specific sentence boundary delimiter such as an identifier, e.g., "# ". For example, "Record OCN news # show me Family Guy" concatenates the two sentences "Record OCN news" and "Show me Family Guy".

The basic and complex sentence parallel corpora information 311 is generated automatically or manually from basic sentence corpus information 317 (313, 315). The basic sentence corpus information 317 is composed of a list of basic sentences. For example, "Record OCN news" is a basic sentence. The basic corpus information 317 may be constructed separately from the entity name corpus information 310-1, and the same storage module may be used.

The automatic compound sentence generation module 310 automatically generates a compound sentence by connecting two or more basic sentences of the basic sentence corpus information 317. For example, a basic sentence of "Record OCN news" and a basic sentence of "Show me Family Guy" are automatically created and a compound sentence called "Record OCN news and show me Family Guy" is created.

However, in the case of a compound sentence which is difficult to be generated by the compound sentence automatic generation module 313, the system designer or the user may manually generate the compound sentence using the compound sentence manual generation module 315. [ The system designer or the user manually generates a compound sentence based on two or more basic sentences of the basic sentence corpus information 317 in the step of manually generating a compound sentence. Furthermore, in the process of manually generating a compound sentence, for example, a system designer may manually generate a compound sentence based on the sentence. Since the basic and compound sentence parallel corpus corpora information 311 can be constructed in various forms as described above, the embodiment of the present invention is not particularly limited to how information is constructed.

The statistical translation and training department 300-9 trains the statistical translation model 300-7 based on the basic and complex sentence parallel corpus information 317.

 Then, the statistical translation executing unit 300-2 divides and restores the inputted short text or divides the Chinese text based on the trained statistical translation model, and outputs it.

As a result of the above configuration, the natural language processing module 300 performs entity name recognition on an input sentence such as Chinese text, divides a compound sentence by performing statistical translation based on the result, and generates a plurality of restored basic sentences The performance of the system 110 'may be improved.

The concept of training and execution in FIG. 3 will be described in more detail as follows.

The execution referred to in FIG. 3 means that a machine or device receives input and finds out what the person is curious about and outputs. For example, in the case of emotional recognition that recognizes emotions by looking at a human face, the input is a human face and the output is a person's emotions (eg, joy, sorrow, ...). When taking a statistical-based approach to this implementation, execution can consist of an algorithm and a model. The algorithm is included in the execution part, and the model describes the relationship between input and output. For example, when a person is happy, he or she makes a facial expression, while when it is sad, what kind of expression is a model and mathematically expressed. Algorithms describe the process of finding emotions by looking at human faces based on these models.

Execution above explained that the machine receives input and finds out what it is curious about. Training is to build a model of the relationship between input and output, where the machine receives both input and output. For this training, one must make training data consisting of a pair of direct input and output. For example, when 1000 face images are recorded, if the person's face of each image records the emotions of the person expressing, the training data is inputting the face image and outputting the emotion. In the field of natural language processing, this data is called a corpus.

From this training and implementation perspective, we will look at entity recognition and (machine) translation.

The recognition of the object name means to find out that "disappear with the wind" is an object name of the movie category (or type) in a sentence such as "Let the wind disappear with the start time". That is, the input is a natural language sentence, and the output is the object name category. The object name category handled by Samsung Electronics TV may be movie title, genre, actor, time, etc.

Accordingly, for example, the entity name corpus information 310-1 of FIG. 3 may be composed of sentences having a type of entity name attached thereto. For example, it can be expressed as follows.

/ Dead Poet Society / Movie / Let me know the start time.

/ Gladiator / movie / let me know the start time.

/ Today / time / / u / se / actor / pro (gram) that came out?

Then, the entity name recognition training unit 300-5 constructs an entity name recognition model 300-3 for the relationship between sentences and entity names based on the entity name corpus. In the training data above, you can see the phrase "Let me know the start time" on the right side of the object name that represents the movie name. Such knowledge can be expressed, for example, mathematically.

The entity name recognition execution unit 300-1 finds the entity name, that is, the entity name category, based on the entity name recognition model 300-3 when an input sentence is given.

Then, looking at the (machine) translation for sentence segmentation and restoration, for example, the translation means translating the sentence of the specific language "Train is fast" into a sentence of another language, such as "train is fast". However, in the embodiment of the present invention, it is aimed to divide the compound sentence "Record OCN news and show me Family Guy" into several basic sentences such as "Record OCN news" and "Show me Family Guy". Therefore, in order to systematically express such a process, several basic sentences are expressed in sentence boundaries separated by '#' symbol like "Record OCN news # show me Family Guy".

That is, the input is a compound sentence, and the output can be regarded as a plurality of basic sentences. At this time, the above-mentioned object name recognition technique is applied to a sentence to be translated or changed. For example, in the sentence "Record OCN news and show me Family Guy", it is judged that "OCN news" and "Family Guy" are program names and substitute "Record @movie and show me @movie" . The reason for applying the entity name recognition technique is to prevent the boundary of the sentence boundary from being separated first in the object name (eg, to prevent it from being split between "Family" and "Guy"), and second, It is intended to enhance the accuracy of the machine translation view by expressing the same thing as a word but semantically the same.

The basic and complex sentence parallel corpus information 311 consists of input and output of translation and is generated automatically and manually. The basic and compound sentence parallel corpus can be summarized as in <Table 1>. Table 1 contains the contents of FIG.

input

Print
record @movie show me @movie

record @movie # show me @movie
record @movie and show me @movie

record @movie # show me @movie
record @movie

record @movie
who starred in @movie?

who starred in @movie?

The Statistical Translation and Training Department (300-9) builds a statistical translation model (300-7) on the relationship between the input sentence and the output sentence based on the above corpus. The above training data show that between "record @movie and show me @movie", "and" is replaced (or translated) by "#" and other words are replaced by oneself, Express. Of course, it can be represented by 2-bit information, and it can be used anyway after storing this form in DB.

The statistical translation executing unit 300-2 divides and restores a plurality of basic sentences from a compound sentence by performing a translation process based on the statistical translation model 300-7 when an input sentence is given. At this time, the input sentence is a sentence in which the entity name, that is, the entity name category is found by the entity name recognition execution unit 300-1 and the entity name is substituted.

FIG. 5 is a diagram showing a modified example of an input sentence processed by the natural language processing unit of FIG. 2 or the natural language processing module of FIG. 3;

Referring to FIG. 5 together with FIG. 3 for convenience of explanation, the entity name recognition execution unit 300-1 of FIG. 4 inserts a sentence "Tells who is the main character of Harry Potter and KBS plays" Can be input.

Then, the object name recognition execution unit 300-1 automatically inputs the object name category for the object name of the input sentence, as shown in FIG. 5 (b), through the object name recognition model 300-3 for the input sentence Find out. Through this, "Harry Potter" is a movie and "KBS" is a channel name. In the embodiment of the present invention, the entity name category information is provided as an input sentence, and the entity name is tagged or the entity name category is tagged input sentence.

The statistical translation executing unit 300-2 of Fig. 3 replaces the entity name with the entity name type in the input sentence. Accordingly, as shown in FIG. 5C, the input sentence is changed to a sentence such as "Tell me who @movie is the main character, @channel name is played".

Then, the statistical translation executing unit 300-2 generates (or extracts) a sentence " Let me know who the main character is in the movie # @ channel name "through the statistical translation model 300-7 as shown in (d) )can do.

On the basis of this, the statistical translation execution unit 300-2 judges the boundary of the basic sentence based on the sentence delimiter (#) as an identifier for distinguishing the boundaries of the sentence, and calculates the original sentence for each basic sentence, that is, "Harry Potter Create a sentence that says, "Tell me who is the main character" and "Let me play KBS."

This process can be regarded as restoring the original sentence, that is, the basic sentence, while the statistical translation executing unit 300-2 performs the division on the input sentence.

6 is a diagram illustrating a natural language processing process according to an embodiment of the present invention.

Referring to FIG. 6, the natural language processing apparatus 120 according to the embodiment of the present invention stores corpus information including an entity name category for a word in a sentence and an identifier for separating a boundary of the sentence (S600). Here, the corpus information may be understood as basic and compound sentence parallel corpus information shown in FIG. Of course, the step S600 may include a state in which the statistical translation model is trained based on the stored corpus information.

Then, the natural language processing device 120 receives the Chinese text input by the user as a natural language from the user device 100 (S610). Here, a Chinese sentence means a sentence formed by a plurality of short sentences including a conjunction or connected without a conjunction. Also, connecting without a connection shows a case where a short sentence is continuously provided at regular time intervals.

The natural language processing unit 120 determines words included in the two basic sentences in the received Chinese sentence, that is, the object name category of the object name (S620). This can be determined by the entity name recognition model trained in the input sentence. For example, even if Harry Potter is included in different Chinese texts, it could be a movie title or a character. Therefore, even though the same entity name is used, the entity name category can be accurately determined by the entity name recognition model trained in various sentences. In addition, if there is "Harry Potter" and "disappearing with the wind" in the interlocutors, they can be judged to be the same object category, or movie category.

When the entity name category is determined as described above, the natural language processing device 120 obtains foot group information related to the determined entity name category and generates a plurality of short phrases based on the obtained foot name information (S630). In this process, the natural language processing unit 120 replaces the word corresponding to the object name in the Chinese character with the object name category, for example, when the attribute of the object name in the input Chinese character is determined. Then, Chinese sentences including the replaced object name category are applied to the statistical translation model to obtain corpus information. Here, the corpus information to be output differs in that the corpus information has the same form as the form in which the entity name category is substituted in the input Chinese sentence, and includes an identifier such as a sentence separator between the two sentences.

Accordingly, the natural language processing unit 120 identifies the boundary between two sentences in the middle sentence based on the identifiers for distinguishing sentences, and divides the middle sentence into a plurality of short sentences based on the identifications, thereby restoring the sentence. Since this has been described fully with reference to Figs. 4 and 5, further explanation is omitted.

Thereafter, the natural language processing device 120 may provide the generated plurality of short sentences to the user device 100 (S640). The plurality of short sentences provided at this time may have various forms. For example, when the user device 100 processes a voice command such as a DTV, a plurality of short sentences, for example, 'Record Harry Potter'. Can be provided in two short sentences such as 'Let's play KBS'. Also, in the case of a video display device such as a PC or a chatting PC, the natural language processing device 120 may connect the sentence to the original connection and then provide it again or translate it into another language. In this case, the operation can be performed in such a manner that the two short sentences are respectively translated and then the connection is connected. The original sentence restored after the division may be changed into various forms and may be provided to the user device 100.

7 is a flowchart showing a natural language processing method according to the first embodiment of the present invention.

7, the natural language processing apparatus 120 according to the present embodiment includes object name categories for a word in a sentence, more precisely, an object name word, and an identifier for distinguishing the boundaries of the sentence Is stored (S700). Here, the storage includes a state in which the statistical translation model is trained based on the stored corpus information.

Next, the natural language processing device 120 receives the Chinese text input by the user as a natural language from the user device 100 (S710). Here, a Chinese sentence means a sentence formed by a plurality of short sentences including a conjunction or connected without a conjunction. Also, connecting without a connection shows a case where a short sentence is continuously provided at regular time intervals.

Then, the natural language processing unit 120 generates a plurality of short sentences based on the identifiers of the corpus information related to the object name categories determined in the received Chinese sentences (S720), and then transmits the generated short sentences to the user device 100 .

In addition to the above, the details relating to FIG. 7 have been fully described with reference to FIG. 6, and a further explanation will be omitted.

8 is a flowchart showing a natural language processing method according to the second embodiment of the present invention.

Referring to FIG. 8 of the accompanying drawings, the natural language processing apparatus 120 according to the second embodiment of the present invention includes object name categories for object name words in a sentence and an identifier for distinguishing boundaries of sentences The trained translation model is constructed in the corpus information (S800).

Then, the natural language processing device 120 receives the Chinese text input by the user as a natural language from the user device 100 (S810). Here, a Chinese sentence is a sentence composed of a plurality of short sentences, for example, a first short sentence and a second short sentence including a conjunction or connected without a conjunction. Also, connecting without a connection shows a case where a short sentence is continuously provided at regular time intervals.

In addition, the natural language processing apparatus 120 changes the object name word in the received Chinese sentence into the object name category (S820). For example, even if the object name words in the input Chinese sentences are different from each other, they can be regarded as being replaced with the same category value if they have the same category. In other words, even though the object name is different from 'Harry Potter' and 'Harry Potter', since the movie belongs to the same category, the object name word is replaced with the category value, for example, '@movie'. In the embodiment of the present invention, since the boundary of the sentence is quickly predicted and the sentence boundary is divided into two sentences with respect to the input Chinese sentence, the performance can be considered to be much faster.

Then, the natural language processing apparatus 120 applies (or generates) the corpus information related to the object name category by applying the Chinese text substituted for the object name category to the translation model (S830). Here, the translation model trains the various corpus information, thereby outputting the corpus information related to the category of the object name determined in the input Chinese sentence.

Then, the natural language processing unit 120 divides the input Chinese sentence based on the identifier of the generated corpus information, that is, the sentence syntax symbol, and restores the original Chinese sentence. For example, based on the corpus information, "Harry Potter recording and KBS play" is divided into two sentences: "Record Harry Potter" and "Play KBS." Then, Can be restored in two sentences: "Let me record Harry Potter" and "Let me play KBS."

As described above, in the embodiment of the present invention, not only the object name recognition model such as a mathematical model and the statistical translation model are applied, but the corpus information is searched in the DB based on only the object name category information, A method of generating two short sentences is described. However, in the embodiment of the present invention, in order to divide Chinese sentences into short sentences, it is necessary to identify the category of the object name in the inputted sentence, to grasp the boundaries of the two sentences based on the obtained sentence, Method.

Furthermore, although a description has been given of processing a plurality of sentences of sentences of received short sentences, middle sentences and sentences, and combinations thereof (eg, short sentences + middle sentences, short sentences + complex sentences, and middle sentences + complex sentences) The restored plurality of short sentences can be provided to the user device 100 of FIG. 1 in the form of a Chinese sentence or a double sentence combined again. Therefore, the present invention is not limited to the above short sentences.

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments. That is, within the scope of the present invention, all of the components may be selectively coupled to one or more of them. In addition, although all of the components may be implemented as one independent hardware, some or all of the components may be selectively combined to perform a part or all of the functions in one or a plurality of hardware. As shown in FIG. The codes and code segments constituting the computer program may be easily deduced by those skilled in the art. Such a computer program may be stored in a non-transitory computer readable medium readable by a computer, readable and executed by a computer, thereby implementing an embodiment of the present invention.

Here, the non-transitory readable recording medium is not a medium for storing data for a short time such as a register, a cache, a memory, etc., but means a medium which semi-permanently stores data and can be read by a device . Specifically, the above-described programs may be stored in non-volatile readable recording media such as CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM,

While the invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present invention.

100: user equipment 110:
120: Natural language processing device 200: Communication interface part
210, 300: natural language processing unit (module) 220, 310: storage unit (module)
300-1: entity name recognition execution unit 300-2: statistical translation execution unit
300-3: entity name recognition model 300-5: entity name recognition training section
300-7: Statistical translation model 300-9: Statistical translation training department

Claims (20)

In a natural language processing system,
A user apparatus for inputting a sentence of a Chinese or a Chinese sentence; And
And a natural language processing device for generating a plurality of short sentences corresponding to the middle or complex sentence using the identifier of the previously stored corpus information when the middle or complex sentence is input and providing the generated plurality of short sentences to the user device and,
The corpus information may include:
The name of the object in the sentence, the name of the object to which the word belongs, and an identifier for distinguishing the category boundary and the sentence boundary. In the natural language processing apparatus,
A storage unit for storing corpus information including an identifier for distinguishing an object name category to which an object name word in a sentence belongs and a boundary between sentences;
A communication interface unit for receiving a sentence of a Chinese or a Chinese sentence input to the user device; And
A natural language processing unit for generating a plurality of short sentences corresponding to the middle or complex sentence using the identifier of the stored corpus information when the middle or complex sentence is inputted and providing the generated plurality of short sentences to the user apparatus;
Included natural language processing device. 3. The method of claim 2,
Wherein the natural language processing unit judges the category of the entity name word in the received Chinese sentence or the complex sentence, extracts corpus information of the storage unit including the determined entity name category, and extracts the plurality of short phrases based on the extracted corpus information And outputs the generated natural language processing result. 3. The method of claim 2,
The natural language processing unit,
An entity name recognition execution unit for determining the category of the entity name word in the middle or complex sentence; And
A statistical translation unit for converting the entity name word into the determined entity name category in the middle or complex sentence, acquiring the corpus information related to the changed entity name category, and generating the plurality of short phrases based on the acquired corpus information Execution part;
Wherein the natural language processing apparatus comprises: 5. The method of claim 4,
Wherein the statistical translation executing unit divides the Chinese or Chinese sentences based on the identifier and restores the divided Chinese or Chinese sentences to the plurality of short sentences to generate the plurality of short sentences. 3. The method of claim 2,
Wherein the natural language processing unit generates the plurality of short sentences in the same language as the middle or complex sentence and provides the generated short sentences to the user apparatus. 3. The method of claim 2,
Wherein the natural language processing unit translates the plurality of short sentences in a language other than the Chinese or Chinese sentences, and generates a translated sentence including the conjunction, and provides the translated sentence to the user device. 3. The method of claim 2,
Wherein the natural language processing unit determines that the Chinese character is a Chinese character when a plurality of short sentences connected without the conjunction at predetermined time intervals are continuously provided. 3. The method of claim 2,
Wherein the natural language processing unit generates a short sentence based on the corpus information related to the object name category to which the object name word in the inputted short or similar short sentence belongs, Wherein the natural language processing apparatus comprises: 3. The method of claim 2,
Wherein the natural language processing unit acquires the same corpus information when entity name categories of the first middle and second Chinese sentences including different entity name words coincide with each other. 3. The method of claim 2,
Wherein the identifier includes sign or bit information. In a natural language processing method,
Storing corpus information including an entity name category to which an object name word in a sentence belongs and an identifier for distinguishing a boundary between sentences;
Receiving a sentence of a Chinese or a Chinese sentence input to the user device; And
Generating a plurality of short sentences corresponding to the middle or complex sentence using the identifier of the stored corpus information when the middle or complex sentence is inputted and providing the generated plurality of short sentences to the user apparatus;
Including natural language processing methods. 13. The method of claim 12,
Wherein providing to the user device comprises:
Determining a category of the object name word in the received Chinese or Chinese sentence;
Extracting corpus information of the storage unit including the determined entity name category; And
Generating the plurality of short sentences based on the extracted corpus information;
The natural language processing method comprising the steps of: 13. The method of claim 12,
Wherein providing to the user device comprises:
Determining a category of the object name word in the middle or complex sentence;
Changing the object name word of the Chinese or Japanese sentence into the object name category judged as the object;
Acquiring the corpus information related to the changed entity name category; And
Generating the plurality of short sentences based on the acquired corpus information;
The natural language processing method comprising the steps of: 15. The method of claim 14,
Wherein the step of generating the plurality of short sentences includes:
Dividing the Chinese or Chinese sentence based on the identifier; And
And restoring the divided Chinese sentences or plural sentences into the plurality of short sentences
The natural language processing method comprising the steps of: 13. The method of claim 12,
Wherein providing to the user device comprises:
And generating the plurality of short sentences in the same language as the middle or complex sentences, and providing the short sentences to the user apparatus. 13. The method of claim 12,
Wherein providing to the user device comprises:
Wherein when a plurality of short sentences connected without the conjunction are provided continuously in a set time interval, it is determined as the middle sentence. 13. The method of claim 12,
Wherein providing to the user device comprises:
And if the entity name categories of the first and second Chinese characters including mutually different entity name words coincide with each other, acquires the same corpora corporation information. 13. The method of claim 12,
Wherein the identifier comprises a symbol or bit information. A computer-readable recording medium containing a program for executing a natural language processing method,
In the natural language processing method,
Storing corpus information including an entity name category to which an object name word in a sentence belongs and an identifier for distinguishing a boundary between sentences;
Receiving a sentence of a Chinese or a Chinese sentence input to the user device; And
Generating a plurality of short sentences corresponding to the middle or complex sentence using the identifier of the stored corpus information when the middle or complex sentence is inputted and providing the generated plurality of short sentences to the user apparatus;
A computer readable medium having computer readable program code embodied thereon.

Images