KR101409413B1 - Method for natural language processing using unification grammar - Google Patents

Abstract

A natural language processing method using a unified grammar is disclosed. According to an embodiment, a method for processing a natural language includes receiving a natural language text, removing a negative word, a refraction affixed word, and an unrecorded word by pre-processing the input natural language text, Value tree to generate a structured feature tree. The feature tree sets the preprocessed natural language text as a root node, generates a syntax feature as a child node of the root node, assigns a corresponding syntax feature value, The morpheme qualities are structured into phrases and the syntactic qualities are structured into sentences by generating morpheme qualities which are the minimum units of language having meaning as nodes and assigning corresponding morpheme qualities.

Description

[0001] The present invention relates to a natural language processing method using a unified grammar,

The present invention relates to a natural language processing technique, and more particularly, to a method for processing a natural language text using a unified grammar in an interactive system capable of recognizing a natural language and responding appropriately to the natural language, and a recording medium storing the method.

Natural language processing refers to artificial intelligence technology that uses a computer to understand a human language or generate and analyze a human language. Natural language comprehension is a task that transforms computerized processing through morphological analysis, semantic analysis, and dialogue analysis through everyday language. Natural language generation is a process of converting texts, Graphics, and so on. Such natural language processing can be utilized in many computer-based applications such as document processing, indexing, language translation, and question answering. In recent years, a dialogue system has come to be developed that recognizes natural language commands of users and provides appropriate responses according to the instructions. Non-patent literature cited below introduces an outline of a dialogue management system for such speech natural language processing.

However, since the language used by humans in everyday life is not completely in conformity with the standard grammar, it may be used in a variety of ways depending on the situation, Especially, it causes problems that the meaning and context can not be accurately grasped at the level of practical use. In order for the conversation system to provide an appropriate response to the user, it must be able to accurately analyze and recognize the user's initial command or utterance.

 The purpose of this paper is to develop a dialogue management system for speech natural language processing. This paper proposes a dialogue management system for speech natural language processing.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide a natural language dialogue system capable of coping with the problem of ambiguity because the information obtained from the natural language text remains at the level of semantic information, And thus the linguistic ambiguity, the boundary ambiguity, the morphological ambiguity and the structural ambiguity are overcome, and the improper response to the ambiguity problem leads to a decrease in the accuracy of conversation of the entire conversation system .

According to an aspect of the present invention, there is provided a method of processing a natural language using at least one processor, the method comprising: inputting a natural language text; Pre-processing the input natural language text to remove a stop word, an inflection affix and an unregistered word; And generating a structured feature tree by associating a feature layered by a predefined unification rule with a corresponding feature value, wherein the feature tree includes a root node, ), Allocates a corresponding syntax property value as a child node of the root node, generates a syntax property as a child node of the root node, generates a morpheme property that is a minimum unit of a language having a meaning as a child node of the syntax property, By assigning a value, the morpheme qualities are structured into phrases, and the phrase qualities are structured into sentences.

The method of processing the natural language according to an exemplary embodiment may further include restoring the prefixed prefix into a circular form when the prefixed natural language text includes a plurality of morphemes combined to form a combined word, have.

The method of processing the natural language according to an exemplary embodiment further includes dividing the preprocessed natural language text into morpheme units using morpheme classification stored in the morpheme database and attaching parts of speech to the divided morpheme .

The method of processing the natural language according to an exemplary embodiment may further include dividing a sentence of the preprocessed natural language text into syntax units using a syntax tag and a syntax rule stored in a syntax property database.

In a method of processing natural language according to an embodiment, the morpheme qualities are selected from the group consisting of LEX (morpheme), POS (part of speech), N (N) , Adjective (ADJ) qualities and adverbs with verbal (V) qualities, LEX (morphemes), POS (parts of speech), SEM (semantic category), QUALIFIER (formula category) with SEM (semantic category), SUBCAT (END) qualities with an ADV qualification, LEX (morpheme), POS (part of speech), CASE (category qualification), LEX (morpheme), POS (part of speech), TYPE , PEND character qualities with LEX (morpheme), POS (part of speech), MOD (linguistic category), and the syntax qualities are SUBJ (subject), OBJ (object word), PRED (SUBJ) and object (OBJ) qualities, HEAD (keyword), and SEM (semantic category) with sentence (S) qualities, HEAD (keyword), CASE (NN) Qualities and ADJP Qualities, HEAD (Keyword), COMP (PRED) qualities with CONJ (mother category).

Meanwhile, a computer readable recording medium on which a program for executing a method of processing a natural language as described above on a computer is recorded.

The embodiments of the present invention can collectively acquire the morpheme, syntax, and qualitative information structured hierarchically from the natural language text by generating a structured qualification tree by linking the qualities layered by the uniformization rule and the corresponding qualification value, Therefore, by considering semantic information, part-of-speech information, and syntax information at the word level, it is possible to solve the problem of ambiguity in the conversation system based on the conventional natural language processing. As a result, Conversational accuracy improves.

1 is a block diagram illustrating an interactive system using natural language processing in which embodiments of the present invention may be utilized.
2 is a flowchart illustrating a natural language processing method using a unified grammar according to an embodiment of the present invention.
3 is a block diagram illustrating an interactive system including a natural language processing apparatus using a unified grammar according to an embodiment of the present invention.
4 is a diagram illustrating morpheme qualities adopted by embodiments of the present invention.
5 is a diagram illustrating a process of analyzing a sentence in a morphological unit and attaching a part of speech in a natural language processing method according to an embodiment of the present invention.
FIG. 6 is a diagram illustrating syntax qualities adopted by embodiments of the present invention.
7 is a diagram illustrating a process of analyzing a sentence in syntax units in the natural language processing method according to an embodiment of the present invention.
8 is a diagram illustrating a feature tree structured by a unification rule in a natural language processing method according to an embodiment of the present invention.

Prior to describing the embodiments of the present invention, the characteristics of the natural language processing and the dialog processing using the same, and the problems therefrom are briefly introduced, and then the technical means employed by the embodiments of the present invention are sequentially .

Artificial intelligence, including natural language processing, has long been an interesting subject for scholars in the field of human-computer interaction. Recently, a dialogue system has been put into practical use in which a human being understands and intends to understand human conversation and performs an action such as an appropriate response in conversation with a machine such as a computer or a robot through a language. Can be utilized for a voice conversation agent of a smart phone, a voice conversation agent of a car, a voice conversation robot, a consumer electronic product capable of voice conversation or a text conversation computer application program. The general structure of this dialog system is as follows.

1 is a block diagram showing an interactive system 100 using natural language processing in which embodiments of the present invention can be utilized and includes a natural language processing unit 10, a dialogue management unit 30, and a natural language generation unit 50 can do.

The natural language processing unit 10 receives the dialogue from the user and understands it. To this end, the natural language processing unit 10 analyzes the inputted natural language by referring to a dictionary, grammar rules and a constraint database 20 for storing various words constituting a human language, and finds meaning therefrom.

The dialogue management unit 30 determines through the natural language processing unit 10 what response is to be made in response to the language in which the meaning is grasped. For this, a response template (templete) suggesting an appropriate response method according to a human conversation situation and a domain knowledge database 40 that can be utilized in a specific field can be referred to, Will find the basic elements that make up the appropriate response corresponding to the user's input natural language.

The natural language generation unit 50 generates a natural language response based on the basic elements of the response generated through the dialogue management unit 30 by referring to the language information 60 constituting the human language and provides the generated natural language response to the user.

As described above, it can be seen that the natural language processing unit 10, which inputs the natural language first from the user's utterance point and grasps its meaning, is most important. Particularly, in the conventional natural language conversation system, various processing processes (e.g., morphological analysis process, part-of-speech tagging process, parsing process, semantic analysis process, and the like) for grasping the meaning based on unprocessed input text ) Is selectively performed to extract the language information from the input text and use it as the basic information of the dialogue management.

This conventional natural language dialogue system has a problem that the information that can be obtained from the natural language text remains at the level of semantic information, part-of-speech information or syntactic information at the word level, so that linguistic ambiguity, boundary ambiguity, It is pointed out that the structural ambiguity can not be solved. That is, since the ambiguity problem of the input text can not be solved through the natural language processing unit 10 for the first time, it is needless to say that the result of the processing through the dialogue management unit 30 to the natural language generation unit 50 is improper. As a result, it can be seen that the processing failure of the natural language processing unit 10 causes a decrease in conversation accuracy of the entire conversation system.

Therefore, embodiments of the present invention described below deal with the basic characteristics of a morphological feature, a syntactic feature, and a sentence feature of a natural language inputted in an interactive system using a natural language text as an input medium, We will structure the language by uniting them according to the combining rules of sentence qualities, and propose a method to utilize such structured quality-based language for dialogue.

In particular, embodiments of the present invention define morpheme qualities, syntax qualities, and sentence qualities suitable for an application field in which an interactive system is to be used, construct a qualitative dictionary corresponding thereto, and then structure the syntax qualities by unitizing morpheme qualities In addition, it suggests a technical means to unify the syntactic qualities and structure them into sentence qualities. Therefore, the conversation system can acquire any predefined qualities information using the structured sentence through the above process, and thus can be utilized as key information for performing intelligent conversation.

Hereinafter, embodiments of the present invention for solving the above-mentioned technical problems will be described in detail with reference to the drawings. In the following description and the accompanying drawings, detailed description of well-known functions or constructions that may obscure the subject matter of the present invention will be omitted. It is to be noted that the same components are denoted by the same names and reference numerals as possible throughout the drawings.

FIG. 2 is a flowchart illustrating a natural language processing method using a unified syntax according to an embodiment of the present invention. The following steps are implemented in an interactive system or a natural language processing apparatus that processes natural language using at least one processor. . Also, steps 210, 220, and 260 that are necessarily performed are indicated by solid lines, and steps 230, 240, and 250 that may be performed selectively are indicated by dashed lines. The optional steps 230, 240, and 250 may be utilized as basic information for generating the feature tree of step 260, but may optionally be utilized as needed. It is to be understood that while these processes are all described below, optional steps 230, 240 and 250 are not necessarily all performed.

In step 210, the conversation system receives natural language text from the user. Of course, the user's utterance is initially analog data, but is preferably converted into digital data that is easy for the conversation system to process and input.

In step 220, the conversation system preprocesses the natural language text input through step 210 to remove stop words, inflection affixes, and unregistered words. Negative words mean simple words, adverbs, prepositions, and infinitives that are not considered necessary for language processing. By filtering these negatives, refined natural language text can be obtained. For example, "aya", "huck", "the", "a", "hmm" Refraction affix is divided according to what role the morpheme plays in forming the word, and refraction of one of the morphemes forming the periphery of the word, not the morpheme (called the base) responsible for the center of the word inflection). For example, "-", "-", "-" and "-" included in "Laugh", "Laugh", "Laugh", "Laugh", "Laugh" , "-", "-" and "-" correspond to the inflection, especially the inflectional suffix (often called the "mother"). It is preferable that the unlabeled word is deleted in order to prevent the natural language processing means of the conversation system or a term or word which is not registered in the dictionary and is not recognized in advance so as to prevent improper operation for understanding the meaning of the conversation. It may also be stored and processed. This may be the case of a new language of the Internet or a language that is hard to find easily. For example, if the expression "good" is inputted, which is an Internet profanity word of "completely good", an unregistered word " There will be. Furthermore, through the preprocessing step of step 220, it is possible to use a method of correcting misplaced punctuation or removing unnecessary cases.

In step 230, if the prefixed natural language text is combined with a plurality of morphemes combined into one, the dialog system may restore the prefixed word to a round shape before joining. It is also called the affixing language. It has the intermediate character of the isolated word and the inflection. It is a combination of the affix to the root and the function of each word in the sentence. In the step 230, when such a ploys are detected in the natural language text, the ploys can be restored to the original form before the ployment for easier semantic recognition and morphological classification. For example, you can restore the words "go" and "please" to "go + go" and "go + go + to +" respectively.

In step 240, the conversation system divides the preprocessed natural language text into morpheme units using the morpheme classification stored in the morpheme database, and attaches the part-of-speech to the divided morpheme. The morpheme is the minimum unit of the language having meaning. In step 240, the inputted natural language text is divided into these morpheme units, and the meaning of each word is grasped. For example, for the expression "I", the division of the morpheme unit such as "I (pronoun) + (investigation)" and the part of speech can be attached. Or the same expression "I" can be analyzed as "fly (verb) + b (tubular ending)", securing both, and then using the tagger, You can choose.

Typically, the problematic parts in the morpheme analysis step include unrecognized words, misspellings, and spacing errors. To prevent these errors, the preprocessing has been performed through step 220. [ Another problem is compound noun decomposition, which can lead to confusion as the composition can be decomposed in various ways, and it is necessary to select the most suitable decomposition result among these options. To this end, embodiments of the present invention may use probability based table parsing to select an appropriate outcome from various decomposed analysis results.

Also, attaching part of speech can be performed through a tagger, and one analysis result suitable for a context can be selected from various analysis results output through the behavior analysis unit. In this case, it is possible to use a large-scale parts-of-speech corpus to select a suitable analysis result by using the information as a hint for solving the ambiguity located on the left and right of the context. For implementation, a hidden Markov model (HMM) can be utilized.

In step 250, the conversation system divides the sentence of the preprocessed natural language text into syntax units by using the syntax tag and the syntax rule stored in the syntax property database. In order to prevent the problem of ambiguity in parsing, if necessary, we may precede parsing and clause-based parsing first, and then perform parsing using these small-scale analysis results. In step 250, the dialog system decomposes the constituent elements of the sentence and analyzes the hierarchical relationship between the decomposed elements to determine the structure of the sentence.

In step 260, the dialog system creates a structured feature tree by linking the qualities layered by the predefined unification rule with the corresponding feature values. Here, the feature tree sets the preprocessed natural language text as a root node in step 220, generates a syntax feature as a child node of the root node, assigns a corresponding syntax feature value, The morpheme qualities are structured into phrases and the phrase qualities are structured into sentences by generating morpheme qualities which are the minimum unit of language having meaning as nodes and allocating corresponding morpheme qualities. In other words, through the structured feature tree of step 260, the dialogue system structures the syntactic features by unifying the morpheme qualities, and structures them into sentence qualities by unifying the syntactic qualities, thereby acquiring predefined qualitative information comprehensively Respectively.

FIG. 3 is a block diagram illustrating an interactive system including a natural language processing apparatus using a unified grammar according to an embodiment of the present invention. In the interactive system of FIG. 1, . In FIG. 3, the configuration is shown focusing on the natural language processing process, and the subordinate configuration is omitted. In addition, the essential components (11, 17) in natural language processing are indicated by solid lines, and the optional components (12, 13, 14, 15, 16) are indicated by dotted lines. Since each configuration corresponds to each of the steps of FIG. 2 already described above, the configuration is focused on the device characteristic and the connection relationship.

The preprocessing unit 11 receives the natural language text, preprocesses it, and removes the negatives, the refraction affixes, and the unrecognized words.

The circular restoration unit 12 receives the output value of the preprocessing unit 11 to check whether or not a prefixed word is included. If the prefixed word is included, the circular restoration unit 12 restores the prefixed word to its original form.

The morphological analysis unit 13 receives the output values of the preprocessing unit 11 or the circular restoration unit 12 and divides the natural language text into morpheme units using the morpheme classification stored in the morpheme quality database 14, Attach part of speech.

The parsing unit 15 receives the output values of the preprocessing unit 11, the circular restoration unit 12, or the morpheme analysis unit 13 and generates a natural language word using syntax tags and syntax rules stored in the syntax / Split text sentences into syntactic units.

The feature tree generating unit 17 receives the output values of the preprocessing unit 11, the circular restoration unit 12, the morpheme analysis unit 13, or the syntax analysis unit 15, And associates the attribute values to generate a structured feature tree.

Finally, the dialogue manager 30 outputs the natural language dialogue or response using the feature structure of the feature tree generated through the feature tree generating unit 17. [ In a strict sense, such a conversation manager 30 is included in the conversation system 300, but is not included in the concentrated natural language processing means of the present invention.

Hereinafter, three embodiments that can be derived through the dialog system 300 of FIG. 3 will be presented. Each of the embodiments can be varied by the addition or subtraction of the optional configuration 12, 13, 14, 15, 16 shown in FIG.

Example  One)

The dialog system 300 according to the first embodiment includes a preprocessing unit 11, a morpheme analysis unit 13, a morpheme quality database 14, a syntax analysis unit 15, a syntax quality database 16, 17, and a dialogue management unit 30. The dialog system 300 according to the first embodiment performs basic preprocessing for natural language text input, performs morpheme analysis using the morpheme qualities database 14, and outputs morpheme analysis results and syntax quality database 16 To perform parsing. The result is generated into a structured feature tree by using the unifying syntax through the feature tree generating unit 17, and the dialog management unit 30 utilizes the generated morpheme feature information and syntax feature information.

Example  2)

The dialog system 300 according to the second embodiment is an example implemented in the simplest form and includes a preprocessing unit 11, a morpheme analyzing unit 13, a morpheme database 14, a quality tree generating unit 17, Only the dialogue management unit 30 can be implemented. The dialog system 300 according to the second embodiment performs morpheme analysis using the morpheme quality database 14 after basic natural language preprocessing. At this time, since there is no syntax information, the qualification tree generation unit 17 can be structured into a list structure of merely morpheme qualities. Finally, the dialogue management unit 30 outputs the response using the structured list of qualities.

Example  3)

The dialog system 300 according to the third embodiment exemplifies the most complex form of the dialog system 300 and includes a preprocessing unit 11, a circular restoration unit 12, a morpheme analysis unit 13, a morpheme database 14, An analysis unit 15, a syntax property database 16, a property tree creation unit 17, and a dialogue management unit 30. The dialog system 300 according to the third embodiment performs the circular restoration after the basic natural language preprocessing and then performs the morphological analysis using the morpheme qualities database 14 and the morphological analysis result and the syntax quality database 16 To perform parsing. The result is generated as a structured feature tree by using the unifying syntax through the feature tree generating unit 17, and the dialog management unit 30 utilizes the generated morpheme feature information and syntax feature information.

In summary, embodiments of the present invention perform basic natural language preprocessing until the natural language text is input and processed by the dialogue management unit 30, and then use the morpheme quality database 14 and the syntax / sentence quality database 16 Perform morphological analysis and parsing, respectively. Each of the units divided into the morpheme unit and the syntactic unit unit sentence qualities unit is unified and structured by the qualification tree generation unit 17 as a combination of morphemes as a syntax and a combination of syntax as a sentence. The structured qualities are preferably expressed in a tree structure that can be easily processed through a computer system. The completed structured qualities tree provides various kinds of qualitative information necessary for conversation to the dialogue management unit 30. Thus, a more natural and intelligent conversation management becomes possible because the qualification tree further acquires information related to various conversations from the qualities.

FIG. 4 is a diagram illustrating morpheme qualities adopted by embodiments of the present invention and includes noun qualities, verb qualities, adjective qualities and adverb qualities, END qualities, and PEND qualities. Each morpheme has the following details.

The noun qualities are LEX (morpheme), POS (part of speech), SEM (semantic category), verb V qualities are LEX (morpheme), POS (part of speech), SEM (semantic category), SUBCAT ), ADJ qualities and adverb qualities have LEX (morpheme), POS (part of speech), SEM (semantic category) and QUALIFIER (qualification category), and the PARTICLE and PP qualities are LEX POS has a POS (morpheme), POS (part of speech) and PEND character has a LEX (morpheme), POS (part of speech), CASE ), POS (Part of Speech), and MOD (Calligraphy Category).

FIG. 5 is a diagram illustrating a process of analyzing a sentence in morphological units and attaching parts of speech in the natural language processing method according to an embodiment of the present invention, and assumes a situation in which a sentence "I go to school" .

The input sentence is divided into tokens, each token is again divided into specific morphemes, and the parts are attached to the divided morphemes. As can be seen from FIG. 5, it can be seen that various morphological analyzes are possible even in the case of one divided token, and the most suitable analysis result among these options will be selected in consideration of the context.

FIG. 6 is a diagram illustrating syntax qualities adopted by embodiments of the present invention, including sentence (S), subject (SUBJ) and object (OBJ), complex noun (NN), and adjective , And predicate (PRED) qualities. Each syntactic feature has the following details:

(SUBJ), OBJ (object), and PRED (predicate). The SUBJ and OBJ qualities are HEAD, CASE, COMP, (ADJP) qualities have HEAD, SEM, and PRED qualities are HEAD, COMP, CONJ, and so on. Lt; / RTI >

7 is a diagram illustrating a process of analyzing a sentence in syntax units in a natural language processing method according to an embodiment of the present invention. [A] "I worked very hard." And [B] "He could not very well quit his job. "

FIG. 7 shows a result of top-down analysis of the sentence texts input to the root node, from which the respective sentences are analyzed. That is, each component is extracted from the original text of the sentence, a tree is constructed according to the structure, and finally, a feature value corresponding to the syntax feature is assigned.

FIG. 8 is a diagram illustrating a feature tree structured by a unification rule in a natural language processing method according to an embodiment of the present invention, assuming a situation where a sentence "I am going to school" is input.

As described above in detail with reference to FIG. 2, embodiments of the present invention propose a structured feature tree for hierarchically structuring syntax and morpheme attributes according to a unification rule, and assigning feature values to each layer. Accordingly, it can be seen that the original text of the sentence is located in the root node in the feature tree shown in FIG. 8, and syntax qualities are generated as its child nodes and the syntax qualities are assigned. In addition, morpheme qualities are generated as child nodes of syntax qualities, and morpheme qualities are assigned. In other words, the natural language inputted first through the structured property tree can be analyzed comprehensively, and it is possible to provide direct information about each component.

According to the above-described embodiments of the present invention, by creating a structured quality tree by linking the qualities layered by the uniformization rule and the corresponding qualities, the morpheme, syntax, and qualitative information hierarchized from the natural language text are synthesized , Thereby simultaneously solving the problem of ambiguity in the dialogue system based on the conventional natural language processing by considering the semantic information of the word level, part of speech information, and syntax information at the same time. As a result, Interpretation improves conversation accuracy throughout the conversation system.

In addition, various information necessary for conversation that can not be obtained from pure natural language text can be acquired from the morphological and syntactic qualities of the qualitative tree adopting the embodiments of the present invention and utilized for conversation, Thereby enabling more intelligent conversation. Furthermore, the embodiments of the present invention can be applied to the management of a conversation using natural language text after voice recognition in a voice conversation agent of a smart phone, a voice conversation agent of a car, a voice conversation robot, and a home appliance capable of voice conversation.

Meanwhile, the embodiments of the present invention can be embodied as computer readable codes on a computer readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer system is stored.

Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device and the like, and also a carrier wave (for example, transmission via the Internet) . In addition, the computer-readable recording medium may be distributed over network-connected computer systems so that computer readable codes can be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present invention can be easily deduced by programmers skilled in the art to which the present invention belongs.

The present invention has been described above with reference to various embodiments. 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 invention as defined by the appended claims. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is defined by the appended claims rather than by the foregoing description, and all differences within the scope of equivalents thereof should be construed as being included in the present invention.

100: Conversation system
10: Natural Language Processing Department
20: dictionary / grammar rules / constraint database
30:
40: Response Template / Domain Knowledge Database
50: Natural language generating unit
60: Language information database
300: Interactive system using structured feature tree
11: preprocessing unit 12: circular restoration unit
13: Morphological analysis part 14: Morphological property database
15: parsing unit 16: syntax / sentence quality database
17: qualification tree generation unit

Claims (5)

A method for processing natural language using at least one processor,
Receiving natural language texts;
A stop word which is recognized to be unnecessary for language processing by preprocessing the input natural language text and a reflex affix which is a morpheme that only deals with inflection of a word in a morpheme forming a peripheral part of a word other than a base, removing inflection affix and unregistered word;
Restoring the prefixed word into a circular form before combining when a plurality of morphemes are combined and represented in the preprocessed natural language text and an affix to the root is combined to include a word that indicates the function of each word in the sentence; And
Generating a structured feature tree by associating a feature layered by a predefined unification rule with a corresponding feature value,
Wherein the feature tree sets the preprocessed natural language text as a root node, generates a syntax feature as a child node of the root node, assigns a corresponding syntax feature value, and assigns the meaning as a child node of the syntax feature Wherein the morpheme qualities are structured into phrases and the phrase qualities are structured into sentences by generating morpheme qualities that are the smallest units of the language and allocating corresponding morpheme qualities. delete The method according to claim 1,
Dividing the preprocessed natural language text into morpheme units using a morpheme classification stored in a morpheme quality database, and attaching parts of speech to the divided morpheme. The method according to claim 1,
Further comprising the steps of: dividing the sentence of the preprocessed natural language text into syntax units using syntax tags and syntax rules stored in the syntax property database. The method according to claim 1,
The morpheme qualities include,
(N) qualities with LEX (morpheme), POS (part of speech), SEM (semantic category)
(V) qualities with LEX (morpheme), POS (part of speech), SEM (semantic category), SUBCAT (subcategory)
Adjective (ADJ) and adverb (ADV) qualities with LEX (morpheme), POS (part of speech), SEM (semantic category), QUALIFIER
LEX (morpheme), POS (part of speech), CASE (category category)
End qualities with LEX (morpheme), POS (part of speech), TYPE (end category)
(PEND) qualities with LEX (morpheme), POS (part of speech), MOD (linguistic category)
The phrase qualities include,
Sentence (S) qualities with SUBJ (subject), OBJ (object word), PRED (predicate)
Subject (SUBJ) and object (OBJ) qualities with HEAD (keyword), CASE (category), COMP
(NN) and ADJP qualities with HEAD, SEM (semantic category)
A predicate (PRED) qualifier having a HEAD (keyword), a COMP (complement), and a CONJ (mother category).

Images