KR101117427B1 - Morphological Composition Device And Method Thereof - Google Patents

Abstract

The present invention relates to a morpheme synthesis apparatus and a method thereof, and more particularly, by synthesizing morphemes and generating words in consideration of the intersecting characteristics of Korean, which have a large change in morphemes in forming a word. A morpheme synthesis apparatus for generating Korean expressions and a method thereof.

In addition, the present invention relates to a morpheme synthesis apparatus and a method for generating Korean expressions of various expression forms that may be needed in a natural language system such as various types of expressions, dialects, communication bodies, and age group expressions.

The present invention relates to a morpheme synthesis technique according to natural language generation in the field of natural language processing of artificial intelligence.

Natural language, morphemes, synthesis, preprocessing

Description

Morphological Composition Device And Method Thereof

The present invention relates to morpheme synthesis and its method among natural language generation techniques. More particularly, the present invention relates to a Korean user by generating a word in consideration of a morphological change of a morpheme occurring in the process of forming a word by combining morphemes in a crossword. The present invention relates to a morpheme synthesis apparatus and method for generating natural Korean expressions and generating words of various types of expression forms required by modern natural language generation systems.

The present invention relates to a morpheme synthesis technique according to natural language generation in the field of natural language processing of artificial intelligence.

Natural language generation is a technology that enables a system to deliver data to users using natural language, which is the most natural means of information delivery to humans. Here, natural language is a concept in contrast to machine language used to write a program for a machine and refers to a language used for information exchange while a person writes or speaks to a person (listener, reader, etc.).

In Korean, which is a deadlock with Finnish, Turkish, and Japanese, morphemes, which are the minimum units of semantic expression, must be combined to generate words, which are the minimum units of sentence construction.

In Korean, in particular, serious morphological changes occur when morphemes are combined to form words. Morphological changes in Korean are caused by the interaction between morphemes, unlike refractive words such as English.

Here, the change in the morpheme of the morpheme means that the phoneme constituting the morpheme is changed in the process of forming the word. For example, in the process of forming the word "beautiful", the verb "beautiful answer" combines with the "b" and changes the phoneme to "beautiful".

In the case of English, it is easy to generate natural expressions simply by modifying the word form regularly according to the tenses or first person, or by using a dictionary containing a list of irregular verbs. It is necessary to consider various linguistic characteristics such as verbs (verbs / adjectives), abbreviation phenomena, research, and phonological dimorphism of mothers, which can make expressions that seem natural to users.

Due to this interstitial nature of Korean language, various methods have been proposed to handle morphological changes in morphemes in Korean natural language processing, specifically sub-fields of Natural language understanding and Natural language generation. .

Conventional natural language processing systems have been actively studied for understanding natural language, which requires processing of random input by a user, compared to natural language generation, and various methods for morphological analysis for natural language understanding have been proposed.

The research on morphological unit processing is also focused on the morphological analysis method for natural language understanding, and the morphological synthesis system for natural language generation has been implemented by using the input and output of the conventional morphological analysis system in reverse.

However, neutrality in morphological synthesis and neutrality in morphological analysis are different, and solutions for neutrality resolution are different.

In the case of morphological analysis, the neutrality caused by the lack of information on the meaning and type of morphemes predicted to constitute a word appears. The following is a representative example of the importance of morphological analysis.

I-> I (pronoun) + is: I am

     -> I (verb) + is: be generated or born

     -> Day (verb) +: flying

In the case of morpheme analysis, the parts of the morpheme composing the word (for example, an analysis based on whether the morpheme is a pronoun or a verb) and what the morphemes mean (for example, when the morpheme is a verb) Different morphological analyzes are possible depending on the meaning of "fly" or "be born".

In order to solve the neutrality shown in the morphological analysis, the neutrality can be solved by considering the morphemes around the word that is the morphological analysis target. In the case of the above example, if the morpheme or word outside the word can be considered as follows, the neutrality generated during the analysis can be solved.

I'm in school ....-> I (pronoun) + is: I am

... crops in the field ....-> I (verb) + is: be generated

... while flying ....-> Me (verb) +: flying

On the other hand, in the morpheme synthesis process for natural language generation, what is the morpheme of a given morpheme and what is the meaning of morpheme is generated in morphological analysis because the data given as input of morpheme synthesis includes each morpheme or part of speech There is no problem of neutrality.

However, in the process of morpheme synthesis, there is a problem of necessity to decide whether to apply to a given morpheme, such as abbreviation / phonological dimorphism / irregular use of verbs. The following is an example showing the neutrality that may occur during morphological synthesis.

<Abbreviation>

This + this-> this

           this

Phonological Dimorphism

School + (Purpose)-> School

Love + (purpose)-> love

<Irregular irregularities>

Beautiful +-> fine

Bend + is-> Multiply

In the case of morphological analysis, the word form of the word is known so that when the neutrality problem about the morpheme combination that constitutes the word is solved, the morphological change of the morpheme can be recognized.

However, in the morphological synthesis process, it is necessary to be able to select an appropriate morphological change among various morphological changes of morphemes that may occur during the process of forming a word.

If the morphemes that are to be considered for the morphological changes of the morphemes that may occur in the morpheme combining process are synthesized as they are, the following awkward expression may be generated.

Is destination + / Jeonju +? -> The destination is Jeonju? (Unnatural expression)

                             Is your destination Jeonju? (Natural expression)

What is this +? What is this? (Unnatural expression)

                       What is this? (Natural expression)

In addition, when the morpheme synthesis system is not used, a mechanical representation such as the following example is generated.

Minjung + Siege-> Minjung +

Static + Mainline-> Static +

The following message was sent to Minjung .

In addition, the conventional natural language generation system generates only Korean standard language expressions. For example, in the conventional navigation system, all the expressions are used in the abbreviation, and in the case of the machine translation system, only the standard language expressions such as using the plain text are used.

However, Korean has a variety of expression styles that vary according to the region, age, speaker's social relations, and information exchange (for example, the Internet). There is a need for morphological synthesis.

As a result, in order to generate Korean expressions that are not only suitable for Korean language but also satisfy the demands of the latest natural language generation system, the morphemes can be synthesized according to various expression forms as well as the morphological changes of morphemes according to standard Korean grammar. Skill is required.

In order to solve the above problems, an object of the present invention is to follow the original morphological form regardless of the morphological changes that appear sequentially and the morphological changes of the morphemes previously shown when the morphemes are synthesized. The present invention provides a morpheme synthesis apparatus and a method for synthesizing a word having a natural form in consideration of morphological changes appearing in parallel.

Another object of the present invention is a morpheme synthesis capable of expressing Korean in various forms of expression that vary depending on the dialect according to the age and region desired by the user, the age or social relationship of the speaker, the Internet expression, the buzzword, and the information exchange. An apparatus and a method thereof are provided.

In order to achieve the above object, the morpheme synthesis apparatus according to the present invention is a morpheme synthesis apparatus for synthesizing a morpheme sequence input from a natural language generator, and receives and analyzes morpheme sequence information including the morpheme sequence or expression form information. Selecting an appropriate morpheme synthesis rule from the stored morpheme synthesis rules, characterized in that to generate a word by synthesizing the morphemes according to the morpheme synthesis rules.

Here, the morpheme synthesis apparatus performs a preprocessing for morpheme synthesis on the communication unit receiving the morpheme sequence information including the morpheme sequence or the expression form information transmitted from the natural language generation device and the received morpheme sequence, and the synthesis to be applied to the morpheme synthesis. A preprocessing unit for selecting a rule, a memory unit for storing a morphological synthesis rule to be applied to the morphological synthesis for the morphological column including the received morphological column or the expression form information, and a condition included in the synthesis rule selected in the preprocessing unit for the morphological column A morpheme synthesis unit for extracting a synthesis rule that satisfies all the conditions of the morpheme column, modifying and calculating the morpheme according to the extracted synthesis rule, and a word synthesis unit configured to generate a word and a natural language sentence for the synthesized morpheme It is characterized by.

The morpheme synthesis apparatus may further include an expression style designation unit capable of directly setting a desired expression style.

In addition, the morpheme synthesis rule is characterized in that each formed of a data structure as follows.

Here, the phoneme condition is a condition for specifying the phoneme and the parts of speech of the morpheme, the grammar condition is a condition for the grammar applied according to the grammatical properties of the morpheme, and the system condition is a morpheme change related to three or more morphemes. Is a condition considered, and the correction operation is a morpheme correction formula to be applied when the condition is satisfied.

And it is characterized in that it is formed of a data structure as follows that further includes the presentation form information received or specified from the user.

Here, the expression style condition is a condition for defining various expression styles such as dialect, acronym, written word, buzzword, and internet dialogue (communication).

In addition, the morpheme synthesis unit satisfies the expression style condition after performing a morpheme correction operation by a synthesis rule that satisfies other conditions except for the expression style condition when the expression style information is included in the morpheme column data or a specific expression style is designated. The morpheme correction operation is further performed by the composition rule.

The preprocessing unit filters elements that are not subject to morpheme synthesis among the received morpheme sequences, filters expressions inserted between words using parentheses, and converts morpheme forms that are not directly morpheme synthesized among the received morpheme sequences. The morphological synthesis rules to be applied to the morphological synthesis are modified and selected from the memory unit.

The preprocessor recognizes both ends of the parenthesis symbol, recursively inputs a phrase or sentence inserted between both ends of the parenthesis symbol, inputs a morpheme string combining the morphemes before and after the parenthesis symbol to a morpheme synthesis unit. The morpheme synthesis unit may sequentially synthesize a phrase or sentence inserted between the recursive input parenthesis symbols and a morpheme sequence combining morphemes before and after the parenthesis symbol.

The morpheme synthesis module may further include a morpheme synthesis condition module for extracting a condition included in the synthesis rule selected for the morpheme column and a synthesis rule that satisfies all the conditions of the morpheme column, and a modification operation included in the extracted synthesis rule. It includes a morphological synthesis operation module for performing the operation.

Here, the morpheme synthesis condition module is characterized by extracting a synthesis rule that satisfies all the conditions by repeatedly checking whether all the conditions are satisfied by sequentially comparing the selected synthesis rules with the morpheme sequence.

The word synthesizing unit rearranges the synthesized morphemes according to the morpheme order information received at the time point received by the communication unit to generate a word.

Meanwhile, the morpheme synthesis method according to the present invention is a morpheme synthesis method for synthesizing morphemes by receiving morpheme sequence information including morpheme sequence or expression form information from a natural language generator, and a preprocessing unit converts the morpheme sequence or expression form information. Performing pre-treatment by analyzing the morphological heat including and morphological synthesis by selecting a suitable morphological synthesis rule among the synthesis rules selected for the pre-treated morphological heat and synthesizing the morpheme with the word synthesis unit using the word for the synthesized morpheme Characterized in that it comprises the step of generating.

The performing of the preprocessing may include filtering out elements that are not subject to morpheme synthesis among the input morpheme strings and transmitting them to the word generator, filtering expressions inserted between words using parentheses, and receiving morpheme strings. And modifying a form of the morpheme which is not directly morphologically synthesized into a form suitable for morphological synthesis and extracting and selecting morphological synthesis rules to be applied to morphological synthesis from the memory unit.

And filtering expressions inserted between words using the parentheses, extracting and recursively inputting expressions inserted between the words using the parentheses, and combining the morphemes before and after the parentheses and inputting them into the morpheme synthesis unit. It is characterized by.

In the synthesizing of the morphemes, the selected synthesis rules may be sequentially compared with the input morpheme column to extract a synthesis rule that satisfies all conditions, and performs a morpheme synthesis operation on the modified operation defined in the extracted synthesis rule. If there are residual synthesis rules, the process of extracting a synthesis rule satisfying all conditions is compared with the inputted morpheme column, and repeating the process of performing a morpheme synthesis operation on the modified operation defined in the extracted synthesis rule. .

And extracting a synthesis rule that satisfies all the conditions comprises comparing a condition of an input morpheme column with a condition included in the selected synthesis rule, and a phoneme condition for specifying phonemes and parts of speech included in the selected synthesis rule. It is characterized by extracting a synthesis rule that satisfies both the grammatical conditions applied according to the grammatical properties of the morphemes and the system conditions considering the morphological morphological changes related to three or more morphemes.

In this case, when the input morpheme column includes expression style information or a specific expression style is specified, a synthesis rule satisfying the remaining conditions except for the expression style condition is extracted to perform a morphological modification operation, and then additionally satisfies the expression style condition. Extract the rule and perform a morphological modification operation.

The generating of the word may include generating a word by rearranging the synthesized morpheme according to the morpheme order information at the time when the word synthesis unit receives the communication unit.

As described above, the morpheme synthesis apparatus and method according to the present invention are parallel according to the morphemes of the morphemes regardless of the morphological changes that appear sequentially according to the morphological changes of the morphemes shown previously and the morphological changes of the morphemes shown previously. Natural words can be synthesized by taking into account the changes in the shape of the words, resulting in excellent reliability and system performance.

In addition, there is an excellent effect of synthesizing morphemes in consideration of dialects according to the times and regions, the age and social relations of the speaker, internet expressions, buzzwords, and information styles that vary depending on the situation in which information is being exchanged.

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

1 is a system configuration diagram of a natural language generation system according to a preferred embodiment of the present invention, and FIG. 2 is a detailed block diagram of the morpheme synthesis apparatus of FIG. 1.

1 and 2, the natural language generation system according to the present invention analyzes the natural language generator 10 for generating natural language and the morpheme information and expression style information transmitted from the natural language generator and applies morpheme synthesis rules And a morpheme synthesis apparatus 20 for generating a natural language sentence, and a client 30 receiving the natural language sentence generated from the morpheme synthesis apparatus.

Here, the expression style information will be defined collectively information about the style to be applied in the relationship with the user of the client terminal, such as standard language, dialect, written language, colloquial language, high word, low word, term used in the Internet, and trendy words.

The expression style information is information input from the outside of the natural language generating apparatus 10, and when there is no input from the outside, the natural language generating apparatus 10 may transmit only the morpheme thermal information to the morpheme synthesis apparatus 20.

In general, natural language generation is to generate a corresponding morpheme string after grasping meaning through a morphological analysis of a natural language sentence input through a client. As well as self-explanatory to those having a part from the core of the present invention will not be described in detail will be described only the word generation and natural language sentence generation according to the morphological synthesis of the generated natural language.

The morpheme synthesis apparatus 20 performs preprocessing for morpheme synthesis on the communication unit 210 and the received morpheme sequence, which receive morpheme column information including morpheme column information or expression form information transmitted from the natural language generator. The preprocessor 220 selects a synthesis rule to be applied to the morpheme synthesis, the memory unit 230 which stores the received morpheme column information, and the morpheme synthesis rule to be applied to the morpheme, and whether the conditions of the selected synthesis rule are satisfied by the preprocessor. It may be configured to include a morpheme synthesis unit 240 for synthesizing the morphemes by determining whether or not, and a word synthesis unit 250 for synthesizing the words with respect to the synthesized morphemes.

The morpheme synthesis apparatus 20 may further include an expression style designation unit capable of designating a specific expression style desired by the client.

Synthesis rules stored in the memory unit 230 are all rules that can handle changes in morphological synthesis when appearing according to grammar of Korean, such as verbal rules and irregular use, phonological dimorphism and preprocessing, affirmative 'yi', and dialect. Say the rules.

The synthesis rule may include a condition for applying to synthesis and a correction operation defining a morphological correction formula when the condition is satisfied.

The phoneme condition may be classified into a phoneme condition for specifying the phoneme and parts of speech of the morpheme to be combined and a condition condition for solving the neutrality that may occur in the morpheme synthesis process.

In addition, the state conditions include three expression forms that define various expression forms such as dialect, acronym, written word, buzzword, and internet dialogue (communication), and grammar conditions applied according to the grammatical properties of each morpheme. These morphemes can be subdivided into system conditions that take into account relevant morphological changes.

The synthesis rule including the condition and the modification operation as described above may be generated as the following data structure and stored in the memory unit.

The preprocessing unit 220 filters out elements that are not morphological synthesis targets (such as punctuation marks and question marks) among the received morphemes and transmits them to the word synthesis unit 250 and uses parentheses between words. It is responsible for filtering out inserted expressions and for modifying the shape of received morphemes such as symbols, units, numbers, and special characters into a form suitable for morphological synthesis (eg m-> meter).

Pre-processing is required when additional expressions are inserted using symbols in the middle of a word, such as the use of parentheses, or when units (m, ℓ) or numbers forming words without sentence marks are inserted.

For example, such as "Taiwan free time signal ", the noun "free time" and the survey "a" are separated by '(自由 時 自)' in parentheses, although they are not in the same word. Therefore, it is impossible to deal with phonological variants. As described above, the problem that another word / phrase / sentence is inserted in a word using a symbol such as parentheses is not limited to Korean. In the present invention, the pretreatment unit processes the above phenomenon through a pretreatment process.

Filtering of symbols filters out symbols that are commonly used when writing sentences such as question marks, commas, and periods. The symbol can be given a part-of-speech, 'S', to prevent morphological compositing rules from being applied. However, it is desirable to perform the symbol filtering process through the preprocessor to improve the performance of the system by reducing the process of applying the morphological synthesis rules.

The expression inserted in the word is not only a notation of the original foreign language expression as in the above example, but also a new sentence may be inserted.

Therefore, the preprocessor 220 may find the inserted expressions using conditions such as parentheses so as not to affect the process of generating the word, and the found insertion expressions may be used to form an appropriate sentence again using a morpheme input unit. have.

For example, .... in a beautiful shop (which has been in operation since 2000) , the phrase "being in 2000" that is enclosed in parentheses in a sentence like .. do.

Here, the case of processing the inserted expression will be described in detail in the following embodiments.

The preprocessor 220 may be used to process symbols or numbers that form words without being sentence symbols.

For example, in the expressions Σ is and λ is, Σ (sigma) and λ (epsilon) are symbols, respectively, and it is difficult to determine whether to apply the morpheme synthesis rule through phoneme comparison. Therefore, the preprocessor stores the pronunciation of each symbol in the memory unit, performs appropriate conversion when the symbols are input, and transmits them to the morpheme synthesis unit.

In addition, in the process of generating the actual word in the word combining unit, the symbols must be processed in the form of the original symbol, so that the preprocessor transmits the above information to the word combining unit so that the intended expression can be generated.

For example, in the case of "Σ + / is", the preprocessing unit 220 converts the sigma + is into a form that can be morphologically synthesized, and the morphological synthesis unit synthesizes "sigma + is". The word synthesizing unit 250 converts the "sigma" into the original symbol "∑" to generate the word "∑".

In addition, numbers are also represented by Arabic numeral symbols, but because they actually have Korean pronunciations, preprocessing is necessary. In particular, numbers may be processed according to the units attached to them.

For example, '1 ml' and '1 horse' are pronounced differently as 'one milliliter' and 'one horse', respectively, so that the preprocessing unit performs processing to match the appropriate pronunciation according to the situation.

And transmitting only the morpheme thermal information that needs to be applied to the morpheme synthesis rule to the morpheme synthesis unit 240, and selecting a synthesis rule list including the phoneme rule and the state condition applicable to the morpheme synthesis from the memory unit 230. In charge of. If the presentation style information is received together, the presentation style condition to be applied to the morpheme synthesis is also selected.

The morpheme synthesis unit 240 performs a morpheme synthesis operation by defining a condition to which the synthesis rule selected by the preprocessor is applied and determining whether the condition is satisfied.

3 is a detailed block diagram of the morpheme synthesis unit of FIG. 2.

Referring to FIG. 3, the morpheme synthesis unit 240 includes a morpheme synthesis condition module for extracting a synthesis rule that satisfies all the conditions by examining a phoneme condition and a state condition among a morpheme column and a selected synthesis rule list. According to the present invention, the morpheme synthesis operation module may be configured to perform a morpheme synthesis operation.

The morpheme synthesis condition module sequentially inputs a list of all the synthesis rules in which the morpheme column is selected to extract a rule that satisfies the conditions (self condition and state condition) of the morpheme column, and the morpheme synthesis operation module satisfies the condition. According to the rule, the operation to modify the morpheme is performed, and the modified morpheme is combined to form a morpheme.

More specifically, the morpheme synthesis rules can be classified into categories such as verbal rules and irregular use, phonological dimorphism and preprocessing, affirmative 'yi', and expression style, and will be described in detail.

1. Rules and Irregular Usage of Proverbs

Proverbs means morphemes that correspond to adjectives and verbs in morphemes, and in Korean, various morphological changes, such as utilization, occur during the process of combining a verb with a mother. In particular, the use of verbs is divided not only into the use of rules that necessarily appear according to the spelling of the term being combined and the ending of the ending, but also into irregular use, which may or may not appear depending on the verb.

Iii) In the case of the use of the rule, if the end of the verb is 'ㄹ', 'ㄹ' is dropped before the ending of the beginning of 'b / ㅂ / ㅅ / ㅗ'.

For example, changes that occur during morphological synthesis, such as "flesh + everything-> live", "flesh + ney-> live", "flesh + live-> live", can be attributed to the spelling, especially the bonds, that make up the morpheme. It can be conditioned by spelling the boundary. Therefore, in the morpheme synthesis rule, the following conditions are specified using the morpheme phoneme to be combined in the morpheme synthesis part.

In the synthesis rule, morpheme 1 denotes the last phoneme condition of the front morpheme, and morpheme 2 specifies the front phoneme condition of the back morpheme.

Also, because most rules are applicable to combinations of morphemes with specific parts of speech, such as the use of verbs, it is possible to specify parts of speech as conditions.

If the morpheme column satisfies the phoneme condition for the application of the synthesis rule, the morpheme synthesis operation module should modify the form of the two morphemes into a representation for generating a correct expression.

For example, if morpheme 1 is "flesh" and morpheme 2 is "floating", delete the last letter "d" of the former morpheme (modification 1) and delete the previous letter "ㅂ" of the latter morpheme. The morpheme is synthesized by "living" by performing the operation of inserting (correcting 2) the last element of the front stem. The notion of substitution used in the general syntax can also be implemented with a combination of delete and insert operations.

The following is an embodiment of a rule for using the drop 'd'.

(Rule 1) Rules for dropping 'ㄹ'

ㄹ (V) / ㄴ (E) D / #

V (V) / ㅂ (E) D / #

ㄹ (V) / ㅅ (E) D / #

V (V) / ㅗ (E) D / #

In the rule, "D" means the operation to delete the last one from the front morpheme, and the back # means no change in the back morpheme. If you need to delete multiple phonemes in the delete operation, add D as many times as you want to delete.

Ii) Unlike regular use of rules, irregular use is a phenomenon in which usage occurs or does not occur depending on usage, even if a given word and ending are spelled with the same conditions.

 In the case of 'ㅂ' irregularity, when the stem word is 'ㅂ', it is changed to 'ㅗ / TT' in front of a mother that starts with '-er' or a vowel that requires a mediated vowel. It turns into 'oh' and the rest turns to 'right'.

For example, "Multiply + Multi"-> "Multiply" / "Multiply + Ni"-> "Guni" /

           "Product + Arthur"-> "come and go" (exception, back is multiplied)

As can be seen in the above example, a verb ending with 'ㅂ' is changed in the process of being combined with a specific ending. However, as can be seen in the exception, this change does not necessarily occur. In the case of the product of the word 'beautiful' which means beautiful, the word is used. In this case, the natural language generator has information on which of the two 'product-' verbs is intended to be used, and the morpheme synthesis apparatus according to the present invention can specify additional features of each morpheme in the morpheme synthesis rule. If you make it so that the word can be generated by reflecting the change of irregular morphemes such as irregular utilization.

"CD" is a grammar condition for determining whether to apply a morpheme synthesis rule according to grammatical properties of a morpheme, "C" is a grammar condition of a front morpheme, and "D" is a grammar condition of a rear morpheme. Condition values in actual morpheme units are expressed in bit units rather than numbers.

irregular
Group 1 irregular
Group 2 irregular
Group 3 irregular
Group 4 Phonological
Morphology contraction
Step 1 contraction
Step 2 contraction
Step 3 Binary Binary Binary Binary Binary Binary Binary Binary

The data format shown in the above table is an optimized method for Bit Operation, and as long as it is possible to specify several conditions for one morpheme, how to actually specify the conditions may vary depending on the development method.

In the grammar condition, one morpheme may have various grammatical characteristics, such as a morpheme capable of irregular use and abbreviation of a verb. Thus, the condition can be expressed in bit unit data for specifying a plurality of grammatical attributes for each morpheme.

The following is a rule and an embodiment thereof for utilizing a dropout.

10 / ㅂ (V) / ㅓ (E) D ㅝ / D Patch: Gib + ㅓ => tilt

10 / ㅂ (E) D ㅘ / D Multiply: Multiply + Ah => Gowa

10 / ㅂ (V) / ㄴ (E) D TT / Multiply: Multiply + b => fine

10 / ㅂ (V) / ㄹ (E) D TT / spicy: map + ㄹ => spicy

10 / ㅂ (V) / hour (E) D TT / Multiply: Multiply + Hour + Multi =>

10 / ㅂ (V) / ㅁ (E) D TT / Spicy: Map + ㅁ => Spicy

10 / ㅂ (V) / ㅡ (E) D TT / D Multiply: Multiply + That => Goni

    ㅂ (V) / ㅡ (E) # / # Multiply: Multiply + That => Multiply

In the above rule, "10" means that the expression form information indicating the use of the irregular morphology should be input together in the front morpheme, and that no condition is required in the rear morpheme.

2. Phonological Dimorphism

There are two main categories of subjects that indicate that a word is a subject. If there is a final syllable in the preceding noun, we use 'yi', and in the absence of a final syllable, we use the survey 'a'. Like a / an in English, there are pronunciations in Korean that are naturally felt according to the phoneme of the last syllable of the previous morpheme. This is called phonological dimorphism.

This phonological form is mainly shown in the connection of nouns and investigations, and its type is divided into whether there is a support in the last syllable of a noun. Two problems arise in dealing with this phonological variant.

First, the presence or absence of the support is difficult to express simply. Second, there are many cases in which nouns and surveys are input due to the structure of the input morpheme column.

The first problem can be specified in the manner of directly expressing the phoneme of the Hangul, such as the rules and irregular use of the verb, but it is impossible to check the presence or absence of the support only by this, in the present invention as follows for the morpheme condition item Several special characters can be used, which allow for the expression of phonological variants.

^: The beginning of morphemes

$: End of stem

! Negation

@ : Consonant

# : collection

%: Initiality

& : neutrality

*: Jongsung

The following is a synthesis rule and an embodiment thereof for phonological dimorphism using the above special characters.

00 / & (N) / (J) # /

= 00 / *! (N) / (J) # /

00 / * (N) / | (J) # / | DD Lean + Go => Lean

= 00 / &! (N) / | (J) # / | DD

Let's look at an example of a rule that uses other special characters.

00 / ^ Fu (V) / ㅓ (E) D ㅓ / D Solve: Poe + U => Fur

The above rule is a rule for irregular use of 'TT' which is one of irregular use. In case of 'TT' irregularity, only one word 'fu-' appears. '^ Fu' means that the last syllable of the morpheme is 'fu'. And the condition that it must be at the front of the morpheme with no more syllables before it. As such, special irregularities, such as the use of 'TT' irregularities, can be implemented by using only special characters. In the case of '$', it plays the same role as '^' for the back morpheme where the test direction of the phoneme is reversed.

And for the '!' This means that the phoneme of the next and the phoneme of the input morpheme should not match.

3. Positive designator 'yi'

In Korean, there is a morpheme called '-i-', and it is usually inserted between morphemes in the same form as the ending word so that the word has a positive meaning. The affirmative designator 'I' is not only used very frequently in Korean, but also consists of only the vowel 'ㅣ' which is easy to combine with other vowels.

In particular, as shown in the following example, it is necessary to generate a word in consideration of three or more morphemes in the morphological synthesis process.

Ⅰ) President + Lee / VCP + => President

Ii) woman + tooth / VCP + => woman

Iii) was woman + two / VCP + was => woman

In the above example, i) is used as it is because there is a support in the front morpheme, and in the case of ii) a positive designation is eliminated because there is no support in the front morpheme. Finally, in case of ⅲ), the third morpheme starts with 'ㅓ' which can be combined with the vowel 'ㅣ', so it is shortened to the third morpheme and merged into one syllable. In particular, in the case of i), the affirmative designator is eliminated first in the process of processing the previous two morphemes, so that in the process of combining 'yes' in the next step, it is not considered the existence of the affirmative designator. Can be generated.

Therefore, in order to solve the above problems, the morpheme synthesis apparatus according to the present invention may define a synthesis rule to consider three or more morphemes.

The following is an embodiment of the synthesis rule for the positive designator 'yi' as described above.

Here, the frontmost "B" is the system condition for the global state for the affirmative 'yi', and the following "CD" is the grammar condition for verbal rules and irregular use.

<Positive designator 'yi'>

0 10 / & (N) / | (VCP) # / D / 1 Excitation + Tooth => Excitation (Global 0 → 1)

<Mother related rules>

1 10 / # / ㅓ (E) # / D ㅕ / 0 female + was +> female (global state 1)

As described above, the system condition may be inserted in the case of considering a morphological form change related to three or more morphemes such as "Woman + was +", and the positive designator 'yi' is combined according to the preceding two morpheme combinations. In this case, the global state is changed from 0 to 1, and accordingly, the following two morphemes are changed to 'was'.

In addition to generating a synthesis rule for performing an operation according to the conditions for morphological synthesis as described above, it is also possible to store a record of morphemes disappeared due to dropouts and to generate information necessary for generating various expressions such as dialect. Can provide.

System condition consists of data of bit type optimized for Bit operation as described in grammar condition, and it is used for processing purpose such as positive designation 'yi', auxiliary verb 'ha', pronoun 'this / that / it'. Can be.

4. Expression form

All of the above mentioned synthesis rules are intended for the purpose of generating standard words. However, in Korean, there are several dialects that are divided according to regional dialects, archaic expressions over a relatively long time period, buzzwords (e.g. Hassam / Haoche) in relatively short periods of time, and the speaker's age, gender, and relationship between speakers. There are many different ways of expression, including

Accordingly, the morpheme synthesis apparatus according to the present invention is configured to specify expression style conditions in order to be able to easily select and generate these various ways of expression.

To do this, all the basic morphemes entered by default are transformed into a form suitable for generating basic standard word expressions through the rules described so far. Then, rules for converting to the appropriate form are applied according to the designated presentation style information.

Here, "A" at the foremost part is a part that contains expression style conditions such as dialect. The above condition is also made in the form of bit data for bit operation. Through this, the field of the bit data can be divided into ones for specifying the type of expression, and the combination of bits can be used to simply select various types of expressions.

For example, Gyeongsang-do + youth + brief expression / Jeolla-do + example height + written word / Chungcheong-do + youth + speech expression, etc. can be set the expression style conditions, the synthesis rules according to the conditions can be applied to synthesize the morphemes. .

 The following is a morphological synthesis rule and an example of the result for generating a gradient dialect.

1000 / ㅆ (E) / ㅅ (E) D / ㅆ / # / 0

1000 /-(E) / (E) DD / | // # / 0 went + to go => chew

1000 /-(E) /? (E) DD / | D / DDDD went / 0 ++ == thin

4 is a flowchart schematically showing the processing of the morpheme synthesis unit according to the preferred embodiment of the present invention.

Referring to FIG. 4, the preprocessing unit 220 receives a first word of the input morpheme string, analyzes the first word, selects synthesis rules to be applied to morpheme synthesis, and the morpheme synthesis unit checks conditions according to the synthesis rule selected by the preprocessor. And perform a stemming operation.

More specifically, the morpheme synthesis condition module of the morpheme synthesis unit extracts the first synthesis rule from the list of synthesis rules selected from the preprocessing unit, checks whether both the phoneme condition and the condition of the first word are satisfied, and if not, the remaining synthesis The process of extracting the second composition rule among the rules and checking whether the condition is satisfied is repeated, and the process is repeated until the composition rule satisfying all the conditions of the morpheme column is extracted.

When the synthesis rule is extracted through the above process, the morpheme synthesis operation module performs morpheme synthesis through morpheme modification according to the correction operation defined in the extracted synthesis rule.

As described above, when the morphological synthesis operation on the first morphological column is applied, it is checked whether the residual morphological column exists, and when the residual morphological exists, the process applied to the first morpheme on the remaining morphological stem is completed. Repeat until done.

5 is a flowchart schematically showing a morpheme synthesis method according to a preferred embodiment of the present invention.

Referring to FIG. 5, first, a morpheme string generated from a natural language generator or a morpheme string including expression style information is received through a communication unit.

The preprocessing unit performs preprocessing on the received morpheme strings, and selects synthesis rules that meet the condition from the parts of speech and phoneme relations of the morpheme rows among the synthesis rules stored in the memory unit.

Here, the preprocessing filters out elements that are not subject to morpheme synthesis, such as symbols, to the word generator, filters expressions inserted between words using parentheses, and forms the morphemes of the received morphemes in a form suitable for morpheme synthesis. It means the process of modifying and transmitting to the morpheme synthesis unit.

The morpheme synthesis unit then applies the selected synthesis rule to perform morpheme synthesis.

When the morpheme synthesis is completed through the above process, the word synthesizer rearranges the morphemes transmitted by the preprocessing unit and the morphemes synthesized through the morpheme synthesis unit according to the morpheme order information at the point of time at which the communication unit is received. Will generate

Hereinafter, a process of synthesizing morphemes by applying a synthesis rule will be described in detail with reference to FIGS. 4 and 5 and the following examples.

For illustrative purposes, it is assumed that the list of morpheme synthesis rules is as follows.

01: 0/0/1/0 / ㅎ / V / ㅗ / E D / # / 0 // Irregular dropping

02: 0/0/1/0 / H / V / ㄹ / E D / # / 0

03: 0/0/1/0 / ㅎ / V / ㅁ / E D / # / 0

04: 0/0/1/0 / H / V / B / E D / # / 0

05: 0/0/1/0 / H / V / ㅂ / E D / # / 0

06: 0/0/1/0 / ㅂ / V / ㅓ / E D ㅝ / D / 0 // 'ㅂ' irregular

07: 0/0/1/0 / ㅂ / V / ㅏ / E D ㅘ / D / 0

08: 0/0/1/0 / ㅂ / V / B / E D TT / D / 0

09: 0/0/1/0 / ㅂ / V / ㄹ / E D TT / D / 0

10: 0/0/1/0 / ㅂ / V / ㅂ / E D TT / D / 0

11: 0/0/1/0 / ㅂ / V / ㅗ / E D TT / D / 0

12: 0/0/1/0 / ㅂ / V / H / E D TT / D / 0

13: 0/0/1/0 / ㅂ / V / ㅁ / E D TT / D / 0

14: 0/0/1/0 / ㅂ / V / ㅡ / E D TT / D / 0

15: 0/0/2/0 / thing / N / ㄴ $ / J D / # / 0 // noun '~ thing' / '~ what' abbreviation

16: 0/0/2/0 / one / N / ㅣ $ / J DD / D ㅔ / 0

17: 0/0/2/0 / one / N / ㅡ $ / J D / D / 0

18: 0/0/2/0 / thing / N / $ / J D / DD / 0

19: 0/0/2/0 / $ what / N / # / J DDD \ / DD / 0

19: 0/0/2/0 / # / N / 's / J # / D / 0

20: 0/0/0/0 / # / N / ㅣ $ / VCP # / D / 1 // no finality + positive specifier 'yi'

21: 0/1/0/0 / # / VCP / J / J # / J / 0 // Missed Positive Specifier 'I' + Survey

22: 0/1/0/0 / # / VCP / J / J # / J / 0

23: 0/1/0/0 / # / VCP / preset $ / J # / this / 0 // reinsert missing positive specifier

24: 0/1/0/0 / # / VCP / Lira $ / J # / this / 0

24: 0/1/0/0 / # / VCP / ㅁ / J # / Im / 0

25: 0/0/0/0 / @ / N / | / J # / DDDDDD / 0 // phonological dimorphism processing

26: 0/0/0/0 / # / N / | / # J / DDDDDD / 0

27: 0/0/0/0 / @ / N / this | J / J # / DDDD / 0

28: 0/0/0/0 / # / N / is | J # / DDDD / 0

29: 1/0/0/0 / night / J // DD

30: 2/0/0/0 / crab / J // D ㅣ / # / 0

31: 2/0/0/0 / night / J // DD / # / 0

32: 4/0/0/0 / night / J // DD / 3 / # / 0

Example 1: The blue color is even better.

Input morpheme column: blue (1) + b / color + 2 / + more / product (1) + y ++.

First, the preprocessor receives the first word "blue (1) + ㄴ", the morpheme is the morpheme where the dropping irregularity occurs, and the front morpheme is the verb (V). Send to morphological synthesis section.

The morpheme synthesis unit compares the condition of the synthesis rule 1 and the input morpheme 'blue (1) + b', and the synthesis rule 1 matches the grammatical condition that 'ㅎ' dropping irregularities but In contrast to 'b', since the first synthesis rule is 'ㅗ', the phoneme condition does not match, so the condition of the second synthesis rule among the remaining synthesis rules is compared with the input morpheme.

In the above process, the fourth synthesis rule matches the condition that the dropping irregularity is 'ㅎ', but the morpheme in which the rear morpheme is input also satisfies the consonant condition of 'b', thus performing a correction operation according to the fourth synthesis rule.

The correction operation defined in the synthesis rule 4 is "D / # / 0", which deletes one phoneme from the front morpheme, so the morpheme column after the operation of deleting one phoneme becomes 'para (1) + b'.

After applying the synthesis rule 4 to the morpheme column as described above, and performing the conditional check by repeating the process of comparing the stem cells with the remaining 5 ~ 14 rules, if the rule to be compared no longer exists, the synthesis rule 4 is applied. The morphological heat is transmitted to the word generator to generate 'blue'.

For the remaining morphological columns, "color + two", "+ more", "product (1) + did + many +.", The same process is repeated to generate a word.

Example 2: .... was a friend.

Input Morphs: Friend + Yes + Yes +.

First, a period (.) Is directly transmitted to the word generator in the preprocessing process.

Then, among the remaining four morphemes, a synthesis rule 15 to 20, which is a synthesis rule for 'friend + i' and a noun, is selected and input to the morpheme synthesis unit.

Synthesis rule 20, which satisfies the condition of the input morpheme column, is applied to the morpheme column through the sequential condition checking process for the synthesis rules 15 ~ 20 as in Example 1, and the morpheme column is converted to 'friend + #' Is converted from 0 to 1.

After 'friend + two' is treated as 'friend + #', the composition rules 21 ~ 24 and 25 ~ 28, which are the synthesis rules for '# +' and positive designator (VCP), are added to the morpheme synthesis section. Is entered.

Here, the synthesis rules Nos. 21 to 24 are applicable rules when the system state (global state) is 1, so that the system state is changed from 0 to 1 through the above process, and thus is applicable to the currently input morpheme column. Therefore, through the sequential condition checking process for the synthesis rules 21 ~ 24, '# +' satisfies all the conditions of the synthesis rule 21, and according to the operation 'ㅕ / 0' defined in the synthesis rule 21 above. Modify "# +" to "# +".

And because we handled the missing positive designator, we set the system's state back to zero.

Finally, after the synthesis process for '+ was', the final morpheme column becomes 'friend + # + was +', and the word generator generates the transformed morpheme row and the period (.) Passed directly from the pretreatment block. Combine to create the final word, 'was a friend'.

Example 3: A beautiful store (run since 2000) ...

Input morpheme: beautiful + is / shop + (+2000 + from + operation + has + ㅁ +) + this |

The preprocessing part receives the first word 'beautiful answer +', the morpheme is the morpheme where the 'ㅂ' dropping irregularity occurs, and the front morpheme is the verb (V). The word "beautiful" is generated according to the correction operation ("D TT / D / 0") defined in the synthesis rule 14 which satisfies the condition by comparing the selected synthesis rule with the morpheme synthesis unit.

Subsequently, the preprocessor recognizes 'store + (from +2000+ years') as a word having a compound structure from the symbol '(' and further reads the morpheme string of the next word.

Recognized that the word or sentence inserted from the symbol ')' in the 'Operation + Begin + ㅁ +) +' + 'end' is over and is inserted in the parentheses from 2000+ The + 'is recursively inputted to the morpheme synthesis section to generate the phrase' beginning from 2000 '.

Then, the morpheme column 'shop + i | ga', which is located on both sides of the '(' and ')', is input to the morpheme synthesis part, and the morpheme string 'shop + ga' is generated according to the synthesis rule 28.

The word generation unit combines 'Beautiful', '(Operated since 2000)' and 'House + Ga' inputted from the morpheme synthesis unit in the order of the original input. Produces the word '

Example 4: What is it?

Input morpheme column: this (2) + what / (2) +? [One]

                 What is this (2) + / (2) +? [2]

                 What is this (2) + / (2) +? [4]

This example shows the change of expression according to the abbreviation and expression style.The grammatical column 'This (2) + I' is attached with the grammar attribute value 2 which means that it is a morpheme to be abbreviated. It is modified to 'I + ㅔ' by the synthesis rule.

Subsequently, what (2) + 'is also modified to' what + what 'by Synthesis No. 19.

In addition, [1], [2], and [4] expression form information were additionally inputted in the input morpheme column. When the expression form is [1], the synthesis rules of Nos. 29 to 30 are applied. Rule number is applied, and in case of [4], rule number 32 is additionally applied.

Therefore, the final generated sentences are made of different expressions according to the expression style information of the input sentence as follows.

[1: Gyeongsang-do dialect] What is it?

[2: Chungcheongdo dialect] What is this?

[4: 'Hasam' body] What is this?

In order to evaluate the superiority of the morpheme synthesis apparatus according to the present invention was carried out experiments using data consisting of 1.5 million words.

The data contains information about which morphemes are given, and which parts of speech each morpheme is. In this experiment, 300,000 were used in the preparation of the synthesis rule, and 1.2 million were used for the performance evaluation. In addition, instead of making rules for all irregular uses, only 100 irregular words were made and applied.

In order to confirm the effect of the objective morpheme synthesis apparatus, the result of combining the morphemes without using the synthesis system according to the present invention to generate a word (A) and the morpheme synthesis rule by the morpheme synthesis system according to the present invention are used. The synthesized results (B) were compared, and the results are shown in Table 1 below.

A B Error count 105,870 29,722 accuracy 91.17% 97.53%

As can be seen from Table 1, when the morpheme synthesis rule according to the present invention is applied (B), the performance of about 6.9% is improved compared to the case of simply combining the input morpheme (A).

In the case of applying the morpheme synthesis rule according to the present invention, the number of errors is 29,722, which is grammatically correct, but is largely divided into errors due to different expressions, errors due to data itself, errors due to lack of irregular information, and errors due to lack of morpheme synthesis rules.

Here, considering that the remaining errors except the errors due to the lack of the morpheme synthesis rule are available expressions regardless of the performance of the morpheme synthesis apparatus, the words that are incorrectly generated due to the lack of the actual rule are 18% of the total errors. It was only.

Therefore, when considering only actual error, the accuracy is 99.83%, which is 8.76% better than the simple combination, and the error rate is only 0.27%, which means that one error occurs in more than 200 sentences. It can be seen that the accuracy of the morpheme synthesis apparatus according to the invention is very high.

Thus, if sufficient morphological compositing rules are applied, grammatical errors can be minimized.

While the present 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 invention.

1 is a system configuration diagram of a natural language generation system according to a preferred embodiment of the present invention, and FIG. 2 is a detailed block diagram of the morpheme synthesis apparatus of FIG. 1.

3 is a detailed block diagram of the morpheme synthesis unit of FIG. 2.

4 is a flowchart schematically showing a process of a morpheme synthesis unit according to a preferred embodiment of the present invention.

5 is a flowchart schematically showing a morpheme synthesis method according to a preferred embodiment of the present invention.

* Description of the symbols for the main parts of the drawings *

10: natural language generator 20: morpheme synthesis device

210: communication unit 220: preprocessing unit

230: memory unit 240: morpheme synthesis unit

250: word synthesizer 30: client terminal

Claims (18)

delete A morpheme synthesis apparatus for synthesizing the morpheme heat input from a natural language generator, Receives and analyzes the morpheme sequence or morpheme sequence information including the expression form information, selects a morpheme synthesis rule corresponding to the morpheme sequence and the expression form information from the stored morpheme synthesis rules, and applies the morpheme according to the morpheme synthesis rule. To synthesize words, The morpheme synthesis device A communication unit configured to receive morphological column information including morphological column or expression form information transmitted from the natural language generating device; A preprocessing unit performing preprocessing for morphological synthesis on the received morphological columns and selecting a synthesis rule to be applied to morphological synthesis; A memory unit for storing a morpheme synthesis rule to be applied to morpheme synthesis for the received morpheme sequence or the morpheme sequence including the expression form information; A morpheme synthesis unit configured to extract a synthesis rule that satisfies all the conditions of the morpheme column among the conditions included in the synthesis rule selected by the preprocessing unit with respect to the morpheme column, and to modify and operate the morpheme according to the extracted synthesis rule; And a word synthesis unit configured to generate a word and a natural word sentence with respect to the synthesized morpheme. 3. The method of claim 2, A stemming apparatus, characterized in that it further comprises a representation style designation unit that can set the desired representation style directly by the client. 3. The method of claim 2, The morpheme synthesis rule Morphological synthesis apparatus, characterized in that each formed in the following data structure.

Here, the phoneme condition is a condition for specifying the phoneme and the parts of speech of the morpheme, the grammar condition is a condition for the grammar applied according to the grammatical properties of the morpheme, and the system condition takes into account the morphological changes related to three or more morphemes. Condition, and the correction operation is a morpheme correction formula to be applied when the condition is satisfied. The method of claim 4, wherein The morpheme synthesis rule A morpheme synthesis device, characterized in that formed in the following data structure further comprising the presentation form information received or specified by the user.

Here, the expression style condition is a condition for defining various expression styles such as dialect, acronym, written word, buzzword, and internet dialogue (communication). The method of claim 5, The morpheme synthesis unit Stemming column data contains representational form information or a specific representational form And a morpheme correction operation is performed according to a synthesis rule that satisfies the expression style condition after the morpheme correction operation is satisfied. 3. The method of claim 2, The preprocessing unit Filter elements that are not the subject of morpheme synthesis among received morpheme columns, filter expressions inserted between words using parentheses, and modify morphemes that cannot be directly stemmed from morpheme sequences And extracting and selecting morpheme synthesis rules to be applied to morpheme synthesis from the memory unit. The method of claim 7, wherein The preprocessing unit Recognizes both ends of the parenthesis symbol, extracts a word or sentence inserted between both ends of the parenthesis symbol, recursively inputs the morpheme string combining the morphemes before and after the parenthesis symbol, And the morpheme synthesis unit sequentially synthesizes a phrase or sentence inserted between the recursive input parenthesis symbols and a morpheme sequence combining morphemes before and after the parenthesis symbol. 3. The method of claim 2, The morpheme synthesis unit A morpheme synthesis condition module for extracting a condition included in the synthesis rule selected for the morpheme column and a synthesis rule that satisfies all conditions of the morpheme column; And a morpheme synthesis operation module configured to perform a morpheme synthesis operation according to the correction operation included in the extracted synthesis rule. The method of claim 9, The morpheme synthesis condition module And comparing the selected synthesis rules sequentially with the morpheme column to repeatedly check whether all conditions are satisfied and extracting a synthesis rule that satisfies all conditions. The method according to claim 2 or 8, The word synthesis unit And a word is generated by rearranging the synthesized morphemes according to the morpheme order information received at the time point received by the communication unit. A morpheme synthesis method for synthesizing morphemes by receiving morpheme sequence information including morpheme sequence or expression form information from a natural language generator, Performing a preprocessing by analyzing a morpheme string including the transmitted morpheme string or expression form information by a preprocessor; Synthesizing the morpheme by selecting a morpheme synthesis rule corresponding to the morpheme column and the expression pattern information among the synthesis rules selected for the preprocessed morpheme column; And a word synthesis unit generating a word for the synthesized morpheme. The method of claim 12, Performing the pretreatment is Filtering elements which are not subject to morpheme synthesis among the input morpheme strings and transmitting the filtered elements to a word generator; Filtering expressions inserted between words using parentheses; Modifying a form of the morpheme which is not directly morphologically synthesized among the received morphological columns into a morphological synthesis form; And extracting and selecting the morpheme synthesis rules to be applied to the morpheme synthesis from the memory unit. The method of claim 13, Filtering expressions inserted between words using the parentheses is The expressions inserted between words using the parentheses are extracted and recursively inputted, and the morpheme synthesis method is characterized by combining the morphemes before and after the parenthesis. The method of claim 12, Synthesizing the morphemes Comparing the selected synthesis rules with sequentially inputted morpheme columns to extract a synthesis rule that satisfies all conditions, and performing a morpheme synthesis operation on the modified operation defined in the extracted synthesis rule; If there are residual composition rules, the composition rule extracts a composition rule that satisfies all the conditions by comparing with the input morpheme column, and repeats the process of performing a morpheme synthesis operation on the modified operation defined in the extracted composition rule. Synthetic method.  The method of claim 15, Extracting a synthesis rule that satisfies all the above conditions Compares the condition of the input morpheme column with the conditions included in the selected synthesis rule, and the grammar condition applied according to the grammatical properties of the morpheme and the phoneme condition for specifying the phoneme and the parts of speech of the morpheme included in the selected synthesis rule. A method of morpheme synthesis, comprising extracting a synthesis rule that satisfies all system conditions in consideration of morphological morphology changes related to three or more morphemes. The method of claim 16, The input stem column contains representation form information or a specific representation form is specified. A morpheme synthesis method comprising extracting a synthesis rule that satisfies a condition other than an expression form condition and performing a morpheme correction operation, and then additionally extracting a synthesis rule that satisfies the expression form condition to perform a morpheme correction operation. delete

Images