KR100286650B1 - Method for generating predicated feature according to table driven system - Google Patents

Abstract

PURPOSE: A method for generating a predicated feature is provided to generate a natural Korean predicate by generating a feature ordering of the Korean predicate using a predict feature ordering and lexical table, and by easily and effectively processing a discordance of word order, a causative and a passive expression, and an irregular conjugation between the Japanese and the Korean. CONSTITUTION: A Korean-Japanese translation device receives meaning paths of morpheme meaning-determined(401). A present morpheme is set as the first morpheme of a meaning path(402). It is judged whether a meaning path of the present morpheme is "null" capable of indicating an end of a sentence to be translated(403). In case that the meaning path of the present morpheme is "null", a Korean equivalent term for the meaning path of the morpheme is connected(404). In case that the meaning path of the present morpheme isn't "null", it is judged whether the present morpheme is a declinable word(405). In case that the present morpheme isn't the declinable word, the present morpheme is set as next morpheme(406). In case that the present morpheme is the declinable word, a PFOLT(Predict Feature Ordering and Lexical Table) is initialized(407). The next morpheme is loaded to the present morpheme(408). It is judged whether the present morpheme is a predict feature ordering(409). The PFOLT is activated. The meaning path of the present morpheme is searched. The meaning path is allocated to the present morpheme(410).

Description

Predicate Aspect Flow Generation Method by Table Drive Method

1 is a block diagram showing the configuration of an automatic machine translation system.

2 is a configuration diagram of the boolean upstream vocabulary table (PFOLT).

3 is a block diagram showing a process of generating predicate phase currents.

4A and 4B are flowcharts showing a method of generating Korean according to the present invention.

<Explanation of symbols for main parts of drawing>

11: Japanese morpheme interpreter 12: one-to-one converter

13: Korean Generator

[Purpose of invention]

[Technical field to which the invention belongs and the prior art in that field]

The present invention relates to a method of generating predicate modalities by a table-driven method. In particular, the Japanese and Korean word order mismatch is adjusted, negation, ministry, passive expression processing, and endings using a predicate modality lexical table in Japanese-Korean machine translation. The present invention relates to a method for generating predicate modalities using a table-driven method that generates more natural Korean sentences by processing heteromorphic forms, irregularities, and phonetic abbreviations.

Although the word order is consistent between the words of Korean and Japanese bilingualism, there are many differences in word order between grammatical morphologies in the predicate. In addition, the negative expression of grammatical morphemes in Japanese is often substituted as a term of negative meaning in Korean. In addition, there may be a case where one-to-one correspondence between bilingual morphemes does not exist or the order of listing differs. Therefore, in the generation of Korean language, auxiliary words should be rearranged according to Korean grammar rules. For example, the Japanese sentence "ぃ き ま せ ん" is translated "Korean" when it is translated into Korean. Looking at the Japanese sentence, the order of (honoring-negative) is the order of (negative-honoring) in Korean.

Because of this, it is difficult to generate natural Korean predicates in the Japanese-Korean translation system that is mainly used because of the similarity between Japanese and Korean grammar systems.

Accordingly, the present invention has been invented to solve the above problems, and an object of the present invention is to drive a table using a predicate side-current vocabulary table expressing partial order relations of grammatical form rumors appearing in a predicate in Korean and Korean connection information. The present invention provides a method for generating predicate side currents.

In order to achieve the above object, the apparatus according to the present invention is a method for generating a predicate modulus in the process of translating Japanese into Korean, using a predicate modality lexicalization table to add a Korean morpheme to each ending following a verb stem. A step of activating a semantic feature to give a symbol, a negative expression processing step of judging whether a semantic meaning feature comes after the verb stem, and if the verb has a negative expression in the dictionary, converting the negative qualifier into a corresponding negative expression, and the verb Determines whether ministry-type semantics come after stem, and if there is a ministry-type expression in the dictionary, ministry expression processing step of converting ministry-type semantics into ministry-type expression, and whether passive semantic features come after the verb stem. If there is a passive expression in the dictionary by judging, the passive semantic A passive expression processing step of changing a string, a heteromorphic ending processing step of determining whether a heteromorphic semantic feature comes after the verb stem, and selecting an accessible form according to a connection test between adjacent morphemes, and irregularities if the verb is an irregular verb A phonological abbreviation / irregular processing step of performing the abbreviation process if the abbreviation is to be performed.

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

In Japanese, a verb is combined with a verb to express the meaning of a predicate, whereas in Korean, a verb is added to a verb by a supplementary verb, a premature ending, and a ending. These supplementary verbs can be generally classified as follows, and when several are used simultaneously, they have the following order.

* Passive:-Come. -Oh. * Intention:-To try.

* Behavior:-let him.

* Pretend: -b. -B.

* Highlight:-Oh. * Progress:-It is. -Go. -Come on.

* Service:-Awesome. -I will. * Of course:-

* Complete:-Ah. -Oh yeah. * Possible:-can be.

* Try:-Ah. * Negation:-not. -I can't. -Don't do it.

* STOP:-Everything. * Herb:-You may.

* Hope:-I want to. -I want to. -To be.

* Reason:-because -B

Estimation: It's a -d shape. Will.

Estimation: It's a -d shape. Will.

* Illustration: -d. * Quotes:

Of these, only the auxiliary words required for the Korean-Japanese translation are shown in Table 1.

The mother is divided into a fresh ending mother and a ending mother. The word ending belongs to the classification of the mother, and the order of appearance on the surface gate is fixed. The word ending plays the role of adding the meaning of the word together with the supplementary word and the word ending. The general word endings and their relations are as follows. Table 2 lists the word endings required in the Japanese-Korean translation. The ending ending was represented.

* Respect:-city-,-practice-

* Tense: (Future)-I'm-

(Current) -b-,-is-

(Past)

Humility: -O-,-,-,-ㅂ-,-wet

* Recall: -The-

* Guess:-

The word endings belong to the classification of mothers, but since the order of appearance on the surface door is fixed as above, these processes are treated like the processing of the modal verbs.

When a Japanese sentence is input to a Japanese automatic machine translation system (COBALT-JK: Collocation-Based Language Translator from Japanese to Korean), it is translated through a Japanese morpheme interpretation step, a Japanese-Korean translation step, and a Korean generation step.

A block diagram showing the configuration of the automatic machine translation system shown in FIG. 1 is shown.

In the Japanese morpheme analyzer 11, the Japanese morpheme analyzer 11 analyzes in morpheme units using the Japanese language connection table (a) and the Japanese-Korean dictionary (b). A Japanese semantic path sentence (path) whose morpheme has been interpreted is input to a Japanese translator 12. In this case, the most appropriate meaning is determined by calculating the matching score using the salmon pattern described by the semantic code of the noun semantic dictionary (c) and the verbal semantic dictionary (d). do.

The semantic path sentence (path) whose meaning is determined is input to the Korean generator 13, where the predicate modality vocabulary table (PFOLT) (e) is used to activate each aspect in the predicate and readjust it to match the Korean word order. By creating a natural predicate. In addition, by checking the connection by the Korean connection table f, processing of heteromorphism of the modal verb, etc., a natural Korean semantic path is output.

The morphological analysis of "彼 は 家 に ぃ き ま せ ん." Is performed as follows.

[(1) 彼 (1)] [(63) は (170)] [(1) 家 (1)] [(34) に (151)] [(200) ぃ (221)] [(155) (13)] [(21) ま (221)] [(154) せ (19)] [(61) ん (15)] [(132). (217)]

Each Japanese morpheme has left and right access information in order to classify parts according to parts of speech and utilization as described above. The connection information is determined according to the characteristics of the parts of speech and the utilization, and the nouns have the left connection 1 and the right connection 1, and the dichotomy "は" has the left connection 63 and the right connection 170.

A morpheme can have multiple parts of speech. Each part-of-speech checks whether it is possible to connect with the parts of speech before and after the part-of-speech, and the connection test uses a connection table composed of a matrix of whether the parts of speech are connected to each other. For example, all parts of speech that can be accessed for "ぃ" above are first tested for connection with the previous investigation "に". If the search and verb stem are accessible, and then all the parts of speech that can be connected to "ぃ" and the connection test of "KI" in the back, the "KI" can be followed. Only a stem.

Other morphemes also determine their parts-of-speech by checking the connection based on the left and right connection information.

The result of the morphological interpretation is sent to a translator. In one and one converter 12, the semantic determination of morphemes is made based on the salmon pattern. In other words, the most appropriate meaning is determined by calculating the similarity score between the previously described salmon pattern and the input sentence in order to solve the homonym problem that occurs mainly in surveys and verbs.

The meaning path thus determined is sent to the Korean language generator 13. The Korean language generator 13 aims to generate natural predicates by activating the predicate phase currents and performing delayed word order adjustment, negative words, ministry words, passive processing, heteromorphism processing, phonological abbreviation and irregular processing.

The data structure called Predict Feature Ordering and Laxicalizing Table (PFOLT) expresses the ordering relationship of the semantic qualities as above. PFOLT expresses semantic features such as tense and aspect according to the grammatical order of Korean language, and each feature has the corresponding Korean morpheme. 2 shows the structure of the PFOLT. Here, 'sCAU', 'sNEG', etc. represent semantic features, and each feature is arranged from left to right according to the order relation in the Korean predicate. Each feature has a Korean form. For example, in the case of the 'sNEG' feature, it has a form called "not".

Using the predicate modality vocabulary table as described above, the Korean predicate modality is generated in the following five steps. 3 is a block diagram illustrating a process of generating predicate phase currents.

1) Activation of meaning qualities-First, find the maxims and activate the qualities of the mother and the verbs that add the meaning of the maxims. If a morpheme has more than one semantic feature, all of them are activated in PEOLT.

2) Negative word processing-Check if the negative feature is activated in PFOLT and perform negative word processing. If the negative qualities are activated, the negative words in the verbs are replaced with the previously registered negative expressions to form a new verb stem and delete the negative qualities. This translates into negative expressions, if any, to ensure a natural translation.

3) Processing of Ministry Expressions and Passive Expressions-The Korean-Korean dictionary and the ministry form and the passive form of the verbs are registered in the Korean-Korean dictionary. When activating the ministry or passive qualities of PFOLT, if there are ministry-type and passive expressions registered in advance for ministry-type and passive-type qualities, the meaning-qualities are changed to ministry-driven expression.

4) Processing of auxiliary forms, premature endings, and mother endings-Form predicates by concatenating Korean morphemes with activated features in the order of PFOLT. At this time, according to the connection test between each morpheme, this form of a mother is determined by selecting a connectable form.

5) Phonological Abbreviation, Irregular Processing-Irregular processing is performed according to the irregular information expressed in the connection information of the verb and auxiliary verb, and the abbreviation is processed by examining the phonological conditions.

4 (a) and (b) show flowcharts showing the Korean language generating method according to the present invention.

In step 401, the translator 12 receives the semantic paths of the morphemes determined by the translator 12, and sets the current morpheme as the first morpheme of the semantic path. In step 403 it is determined whether the current morpheme is "NULL" which marks the end of the sentence to be translated.

If the current morpheme is " NULL " indicating the end of the sentence to be translated, the process proceeds to step 404 where the Korean translation of the processed morpheme semantic path is concatenated and the Korean language generation ends.

If the semantic path of the current morpheme is not "NULL", the flow proceeds to step 405 to determine whether the current morpheme is verbal. If the current morpheme is not a verb, the current morpheme is changed to the next morpheme in step 406, and the process of step 403 or less is repeated.

If the current morpheme is a verb, then in step 407 the predicate phase flow lexicalization table (PFOLT) is initialized. In step 408, the next morpheme is placed in the current morpheme, and in step 409 it is determined whether the current morpheme is a predicate morphology. If the current morpheme is a predicate aspect, in step 410, the predicate morphology vocabulary table is activated to retrieve the semantic path of the current morpheme, assign the semantic path to the current morpheme, and repeat the process of step 408.

If the current morpheme is not a predicate modality, the flow proceeds to step 411 where "negative" is activated, and it is determined whether information exists in the negative field of the dictionary. If there is no information in the negative field of the dictionary, the flow proceeds to step 413. If the information exists in the negative field of the dictionary, the content of the negative field is combined between verb verbs in step 412. Step 413 "Ministry" is activated, and it is determined whether information exists in the ministry field of the dictionary.

If there is no information in the ministry field of the dictionary, the flow proceeds to step 415. If the information exists in the ministry field of the dictionary, in step 414 the contents of the ministry field are combined between verb verbs. Proceeding to step 415, "passive" is activated, and it is determined whether or not information exists in the prior driven field. If there is no information in the driven field of the dictionary, the flow advances to step 417. If the information exists in the driven field of the dictionary, in step 412, the contents of the driven field are combined between the verbs.

In step 417, it is determined whether "I <number of PFOLT" (where I is the room of the presently searching aspect of the predicate aspect class lexicalization table). If I &lt; the number of PFOLTs, it is determined in step 418 whether there is a modality set to &quot; 1 &quot; in the current room of the predicate modality vocabulary table.

If there is no current flow set to "1" in the current room, in step 419 the room I of the phase flow is increased to "1", and the process of step 417 is repeated. If there is a phase current set to "1" in the current room, it is determined in step 420 whether the phase current PFOLT (I) set to "1" in the current room is accessible to the stem of the verb. If not, proceed to step 422. If connectable, in step 421 the phase information of the PFOLT (I) is combined with the stem of the verb to perform an irregular / phonic abbreviation process.

In step 422, if it is determined whether or not this type exists in the PFOLT, if the present type is present, the process of step 420 or less is repeated, and if the type is not present, the method proceeds to step 419, where the phase I of the phase flow is "1". Increment and repeat the process from step 417.

If it is determined in step 417 that the number of I <PFOLT is not equal, the flow advances to step 406 to change the current morpheme to the next morpheme and repeats the steps 403 and below.

As described above, according to the present invention, by using the predicate modality vocabulary table, the modality of the Korean predicate is generated to prevent inconsistency, negation, ministry, passive expression, morphological processing, phonological abbreviation, and irregular processing of Japanese and Korean. You can easily and effectively create natural Korean predicates.

Claims (3)

In the method for generating a predicate modality in the process of translating Japanese into Korean, using a predicate modality vocabulary table, a step of activating a semantic feature for assigning a Korean morpheme to each ending following a verb stem and the verb stem If there is a negative expression in the dictionary after judging whether or not there is a negative meaning feature, a negative expression processing step of converting the negative feature into the negative expression, and a dictionary by judging whether a ministry-type semantic feature comes after the verb stem. The ministry expression processing step of converting the ministry-type semantic features into ministry-type expressions if there is a ministry-type expression in the image, and judging whether the passive-type semantic features come after the verb stem, and if there are passive-type expressions in the dictionary, A passive expression processing step of changing the passive expression, and the verb stem The heteromorphic ending processing step of determining whether a semantic feature of the morphological form comes in and selecting an accessible form according to a connection test between adjacent morphemes, and if the verb is an irregular verb, irregular processing is performed, and if abbreviation should occur, abbreviation processing A method of generating a predicate phase current according to a table driving method, comprising a phonological contraction / irregular processing step. The method of claim 1, wherein the predicate modality lexicalization table expresses semantic features such as tense and modality according to the grammatical order of Korean language, and each feature has a corresponding Korean morpheme as its content. A predicate modality generation method using a table driving method characterized by expressing an ordering relationship of qualities. In the method of generating predicate modalities in the process of translating Japanese into Korean, one or more translators receive input the semantic path sentences of the morphemes that have been determined, and place the current morpheme as the first morpheme of the semantic path sentence to end the sentence to be translated. Determining whether to display "NULL"; if the current morpheme is "NULL" indicating the end of a sentence to be translated, connecting the Korean translation of the processed morpheme semantic path and ending the Korean generation; and If the path of meaning is not "NULL", find a verbal morpheme, initialize the predicate morphology table (PFOLT), place the next morpheme in the current morpheme, and determine whether the current morpheme is the predicate morpheme. If the current morpheme is a predicate modality, the predicate modality lexicalization table is activated to search for the semantic path of the current morpheme and to determine its meaning. Allocating a non-path, performing a negative expression processing, a ministry expression processing, a passive expression processing, a mother's heteromorphic processing, and a phonological abbreviation / irregular processing, if the current morpheme is not a predicate morphology; In the case where it is determined by determining whether the room I of the present aspect being searched in the predicate counter-current vocabulary table is smaller than the number of individuals in the PFOLT, the present condition set to "1" exists in the current room of the predicate phase-current vocabulary table. And if there is no phase current set to "1" in the current room, increase the room I of the phase flow to "1", repeat the above steps, and set the phase current to "1" in the current room. If present, the present invention is provided with a step of performing an irregular / phonic abbreviation process and a heteromorphic process, when the current flow PFOLT (I) set to "1" in the present room is accessible to the stem of the verb. Method of generating predicate countercurrent by the table drive method.