KR100420474B1 - Apparatus and method of long sentence translation using partial sentence frame - Google Patents

Abstract

1. TECHNICAL FIELD OF THE INVENTION

The present invention relates to a long sentence translation device and a method using a partial sentence frame.

2. The technical problem to be solved by the invention

The present invention provides a long text translation device for producing a high-quality translation result with high coverage by using a section sentence frame for a long text, and a method and a computer-readable recording medium recording a program for realizing the method. For that purpose.

3. Summary of Solution to Invention

According to another aspect of the present invention, there is provided a long sentence translation method applied to a long sentence translation apparatus using a partial sentence frame, the long sentence division step of dividing a sentence into one or more short sentences according to a preprocessing result; A short sentence translation step of recognizing the divided short sentences and performing a short sentence translation through door frame matching; A determination step of determining whether the entire door frame translation is successful after searching the entire door frame to perform the full door frame translation; And as a result of the determination in the determination step, outputting a translation result if successful, and repeating and outputting a translation result by repeating a process of combining and translating a partial sentence frame and searching and translating the entire sentence frame.

4. Important uses of the invention

The present invention is used in machine translation apparatus and the like.

Description

Apparatus and method of long sentence translation using partial sentence frame}

The present invention provides a long text translation device for producing a high-quality translation result with high coverage by using a section-by-section partial frame for a long text in a door frame-based automatic translation, and a computer program having recorded thereon a program for implementing the method. The present invention relates to a recording medium, and more particularly, to recognize a verse-level partial sentence from a long sentence in order to solve the problem of a significant decrease in the door frame coverage in the case of a long sentence and a decrease in the translation quality of the long sentence in a long sentence. The long sentence translation device and its method, and the computer program for realizing the above method, determine the joining order between sub-frames through section unit structure analysis, and perform repeated door frame matching and door frame combination. To a record carrier It is about.

First, terms used in the present invention are defined as follows.

In the doorframe-based approach, the term protector refers to a word that reflects the characteristics of a sentence, such as verbs and conjunctions, that the ambiguity of parsing begins to increase exponentially. Therefore, by recognizing the protector, it is possible to grasp the form of the sentence and obtain a clue for translation.

In addition, the door frame refers to a simplified form by processing a sentence, the component of the door frame is called a slot, the slot is a protector or a syntax element between the protectors.

The syntax element includes a syntactic character obtained by parsing a part of sentences between protectors, and is expressed as "NP" for a noun phrase and "PP" for a prepositional phrase.

And, the partial sentence frame refers to a door frame that does not include the entire input sentence but includes a sentence at the clause level.

Next, the prior art and its problems will be described.

A doorframe based translation method has been proposed to solve the problem of ambiguity explosion and unlimited generation of band phrases, which has been a problem in the conventional machine translation method.

In the conventional door frame-based automatic translation method, the use of the concept of a door frame composed of a protector reflecting the structure of a sentence and syntax elements therebetween prevents an increase in ambiguity by limiting the scope of parsing and a frame for a predetermined sentence. By discovering it prevents unlimited generation of band phrases and greatly improves the quality of translation.

However, in the conventional doorframe-based automatic translation method, as the sentence length becomes longer, the number of doorframes to be built up rapidly increases, and the doorframe matching success rate drops, thereby causing serious coverage problems.

Therefore, in order to solve the above problem, it is necessary to divide the long text and process it in smaller units. However, the conventional long text segmentation method assumes a structured sentence and uses a limited pattern. In addition, since the sentence unit structure analysis is performed with a few rules in generating the entire translation result by connecting the translation results of each short sentence after the sentence division is performed, there is a limit in generating a natural translation result that connects the whole sentences.

Accordingly, the present invention has been proposed to solve the above-mentioned problem, and a long sentence translation device and method for realizing a high quality translation result having a high coverage using a section sentence structure of a long sentence and a program for realizing the method Its purpose is to provide a computer readable recording medium having recorded thereon.

That is, the present invention is processed through sentence segmentation, short sentence translation, and substitution in order to solve the coverage problem of the translation that occurs as the sentence length increases in the door frame-based automatic translation and the unnatural sentence generation of the general long sentence translation. By reducing the length of the door frame to be done and performing translation by repeating the door frame matching and the sub-frame frame combination that reflect the whole clause unit structure, it creates a semantically natural band statement and solves the coverage problem of the door frame base. It is an object of the present invention to provide a translation apparatus, a method thereof, and a computer-readable recording medium having recorded thereon a program for realizing the method.

1 is a block diagram of an embodiment of a long sentence translation device using a partial sentence frame according to the present invention.

2 is a flowchart illustrating an embodiment of a long sentence translation method using a partial door frame according to the present invention.

Figure 3 is an example of a long sentence translation (English-Korean translation) using the partial sentence frame according to the present invention.

4 is a flowchart illustrating an embodiment of a sentence division process according to the present invention.

5 is a detailed flowchart illustrating an embodiment of a sentence division process according to the present invention.

* Explanation of symbols for the main parts of the drawings

101: input unit 102: morphological analysis unit

103: part-of-speech determination unit 104: fixed expression recognition unit

105: protector detector 106: partial parser

107: original frame generation unit 108: partial frame processing unit

109: band sentence selector 110: band word generator

In accordance with an aspect of the present invention, there is provided a device for translating a long sentence, comprising: preprocessing means for preprocessing an input long sentence to obtain syntax information; Text frame generation means for generating a text frame for the input text using the syntax information obtained through the preprocessing means; And a partial sentence processing means for translating a long sentence by dividing the original text frame generated by the original text frame generating means into partial sentence frames, recognizing and translating a short sentence, and repeatedly performing the entire door frame search, translation, and partial sentence frame combining and translation process. Characterized in that it comprises a.

On the other hand, the method of the present invention, in the long sentence translation method applied to the long sentence translation apparatus using a partial sentence frame, the long sentence segmentation step of dividing the sentence into one or more short sentences according to the pre-processing result; A short sentence translation step of recognizing the divided short sentences and performing a short sentence translation through door frame matching; A determination step of determining whether the entire door frame translation is successful after searching the entire door frame to perform the full door frame translation; And if the determination result of the determination step, if the success outputs a translation result, and if it fails, repeating and repeating the process of repeating the process of translating and translating the process of searching the entire door frame and the translation result output step It features.

On the other hand, the present invention, in order to translate a long sentence using a partial sentence, a long sentence segmentation function for splitting the long sentence into a sentence or more by performing a sentence division according to the result of pre-processing the long sentence; A short sentence translation function that recognizes the divided short sentences and performs a short sentence translation through door frame matching; A determination function of determining whether the entire door frame translation is successful after searching the entire door frame and performing a full door frame translation; And a result of the determination in the determination function, if successful, outputs a translation result, and if it fails, iterates and repeats the process of combining and translating the partial sentence frame and searching and translating the entire sentence frame. A computer readable recording medium having recorded thereon a program is provided.

As described above, the present invention is characterized by consisting of a process of recognizing a partial sentence frame and analyzing a paragraph unit structure, and translating and replacing a partial sentence frame, searching a whole sentence frame, and combining a partial sentence frame.

That is, in the present invention, the partial sentence frame recognition and the structure analysis between the sub-frame frame is based on the extraction of the split point, applying the split point syntax pattern, the recognition of the short sentence start point, the restoration of the short sentence and the search of the sentence, through the recognition of the short sentence, the starting point-verb verb matching and the door frame search It consists of selecting the correct starting point among several starting point candidates, and the translation process using the structure analysis result between the sub-frames, when the translation is successful in sub-frame translation and substitution, reduced door frame search and sub-frame combination Since the process is repeated step by step, the translation process is performed in a tradeoff between rule-based and pattern-based.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings. Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of an embodiment of a long sentence translation device using a partial sentence frame according to the present invention.

First, the original text entered through the input unit 101 is analyzed for each word through the morpheme analysis unit 102, and the part-of-speech of the words is determined by the part-of-speech determination unit 103.

In addition, the fixed expression recognizer 104 bundles words that are treated like a word or phrase, such as idioms, compound nouns, salmon, and the like, and attaches new parts of speech corresponding thereto. Recognizing fixed expressions like this can simplify parsing and setting up doorframes.

The protector finder 105 detects parts of speech or words that play an important role in a sentence, for example, a verb, a conjunction, a related company, a symbol, and the like, and attaches a display as a protector. No analysis is performed on these.

The partial parser 106, which performs partial parsing on the words between the protectors, attaches an appropriate syntax tag to them. This is a partial parsing with the exception of protectors, so ambiguity rarely occurs and the analysis time is significantly reduced.

In addition, the text frame generation unit 107 generates a text frame for the input sentence using the protector and the syntax tag determined above. The original text frame is used to find the same text frame in the text frame database. If there is no text frame that exactly matches the text, the input text frame is divided into section-level subframes to cover the text. Do it. This is handled by the partial frame processing unit 108 corresponding to the core of the present invention.

That is, the partial sentence processing unit 108 recognizes the partial sentence frame in the original frame and performs translation on the divided partial sentence frame. When the translation of each partial frame is performed, the entire translation is performed by the door frame connecting the whole door frame, and if there is no door frame connecting the whole door, after translating the partial door frame through the structural analysis, Repeat the process of searching for and translating the entire reduced frame by replacing with one node.

At this time, the process of translating a partial sentence frame consists of searching the original frame database, selecting a band sentence frame, and generating a band word. The text frame found by searching the database has several band text frames, and one of them is selected according to the context. The band-frame contains only realistically structured translations that already contain all the structural features of the band-language.

The band sentence selector 109 selects a band sentence frame suitable for the current context by using air information and syntax / meaning information. The band word generator 110 is responsible for structurally converting each slot of the band sentence frame from the selected band sentence frame into a complete sentence. The final band sentence made to correspond to the input sentence is output to the printing apparatus 112 or the display apparatus 114 through the printing unit 111 or the display control unit 113.

2 is a flowchart illustrating an embodiment of a long sentence translation method using a partial door frame according to the present invention.

First, the operation flow is briefly described. A preprocessing process such as morpheme analysis, part-of-speech determination, fixed expression recognition, protector discovery, and partial classification analysis is performed on an English sentence (201).

Next, sentence division is performed on a row of slots, which are the result of the preprocessing, to divide the English long sentence into one or more short sentences (202).

Thereafter, the divided short sentences are recognized and short sentence translation is performed through door frame matching (203). Subsequently, if it is determined that the short translation is successful (204) and the door frame matching fails in the short translation, the short sentence translation process is performed again with another sentence division candidate in consideration of the possibility of an error in the sentence division. Proceed to next step.

Subsequently, when the short sentence is completed, the part corresponding to the partial sentence frame is replaced with the syntax node corresponding to the sentence symbol, and the entire door frame translation is performed by searching for the reduced entire door frame (205). Subsequently, it is determined whether the entire door frame translation is successful (206), and if the entire door frame search is successful and the translation is made, the translation result is output and ends.

If the entire door frame translation fails (206), the partial sentence frame is combined and attempted to be translated according to the intersect structural analysis rule (207). Subsequently, it is checked whether the partial sentence translation is successful (208) and if successful, performs the door frame search and translation (205) for the entire reduced door frame, and if the partial sentence translation fails, the partial sentence combining and translation process (207). Repeat this.

Now, the above-described operation flow will be described in detail.

In the sentence division process 202, the starting point of all the subclauses is recognized from the partial syntax analysis result. In other words, the splitting point of a sentence corresponds to the starting point of every subclause. Recognition of the starting point is achieved through a predetermined syntax pattern and pre-built database information, and it is assumed that the starting point of all clauses is recognized. The syntax pattern for starting point recognition consists of a combination of punctuation, conjunctions, affiliates, and interrogators. In case of ambiguity of the starting point by comma or parallel noun phrases in recognition of the starting point, or when it is excluded due to the omission of the conjunction, the pre-established starting point pattern for the starting point is applied. Therefore, the starting point candidate is recognized by the predetermined starting point pattern, and the starting point is removed and added by the starting point pattern of the database. If there is a starting point ambiguity that is not resolved later, all candidates are recognized and selected in later steps.

As described above, when the starting point of all the sections is recognized, the starting point of the short sentence is recognized and translated (203), and the end point recognition of the corresponding short sentence is attempted. Recognition of the starting point of a short sentence is achieved by recognizing all the starting points as the starting point of the short sentence, not the "that" clause, the relation clause, and the interrogation clause, and the end point of the short sentence is achieved through the door frame search and the starting point to the verb matching. When the short sentence recognition is performed, the sentence sentence restoration is executed when the subject or object of the sentence is omitted by inversion or omission according to the type of the clause to which the short sentence belongs. The sentence restoration adds a noun slot before the verb when the door frame except the conjunction and the related verb starts with a verb and adds a noun slot when the object is omitted in the clause led by the relative clause and the interrogator. By restoring such sentences, the door frame coverage of the partial sentence frame can be increased and smooth translation can be generated. In case of short sentence matching in short sentence translation, the short sentence translation process is re-executed to another sentence division candidate in consideration of the possibility of error in sentence division.

When the short sentence is completed as described above, the part corresponding to the partial sentence frame is replaced with the syntax node corresponding to the sentence symbol, and the entire door frame translation is performed by searching for the reduced entire door frame (205). If the whole door frame is searched successfully and the translation is made, the translation result is printed and then terminated. If it fails, the partial frame combining and translation 207 is attempted according to the intersect structure analysis rule, and if successful, the door frame searching and translation 205 is performed on the entire reduced frame. If the partial sentence combining and translation fails, the partial sentence combining and translation process 207 and the entire sentence search and translation process 205 are repeated.

At this time, the inter-parallel parallel analysis rule used for the partial statement combining is as follows.

1. The "that" clause, the relation clause and the question clause clause: depend on the immediately preceding clause

2. T "that" clause, (T) and "that" clause / relational clause: parallel to the nearest "that" clause / relational clause

3. T (n) V: Connect to a connection without the nearest front (n) V or (n) V

4. (T) and nV: parallel with the highest similarity with all the nV in front

5. (T) and V: parallel to the nearest front V (consideration of numbers and tense)

'T' represents a punctuation such as a comma, 'n' represents a noun phrase, 'V' represents a verb, and the equation for calculating the similarity between each starting point is as shown in Equation 1 below.

Similarity S = w1 * Sim (C-C) + w2 * Sim (n-n) + w3 * Sim (V-V)

In order to calculate the similarity of each item, the objects to be considered are Sim (CC): Vocabulary, Sim (nn): Type, Semantic Code, Vocabulary, Sim (VV): Type, Vocabulary, Tense, Number Same as First Person (TV).

In the process of performing the unit-by-section combining according to the analysis rule, first, a dependency list corresponding to each slot of the door frame to be processed currently is constructed. Dependency analysis is performed as follows by the analysis rule at each starting point and main verb.

1. In rule 1 of the analysis rule, assign the starting point of dependency [i] .depend <-1, dependency [i] .link <-immediately preceding.

2. In rule 2-4 of the analysis rule, dependency [i] .depend <-0, dependency [i] .link <-assign the starting point of the clause in which the clause is parallel.

3. For any other dependent connection clause, assign the starting point of dependency [i] .depend <-0, dependency [i] .link <-immediately preceding.

4. When relation clauses are included in a sentence, connect one main verb to one clause for a main verb that is not infinitive or participle in a slot to analyze the end point of the relation clause. At this time, these connecting links should not be crossed. When such a link is constructed for the whole sentence, assign a starting point for each main verb with dependency [i] .depend <-0, dependency [i] .link <-.

From the dependency list obtained in the above process, the depth list of each slot is obtained as follows.

depth [i] = depth [dependency [i] .link] + dependency [i] .depend;

absence dependency [i] .link depth [i] = depth [i-1];

Based on the depth obtained through the above process, the joining of the clauses is performed in the order of the deepest depth.

The above process is described with reference to FIG. 3 as follows.

3 is an example of a long sentence translation (English-Korean translation) using the partial sentence frame according to the present invention.

First, after performing the starting point recognition indicated by '/' in the partial parsing result of the original text, a dependency list is obtained for each starting point and the verb. Arrows and numbers in the figure indicate dependency [i] .link and dependency [i] .depend for each starting point and main verbs. Through this dependency list, a depth list is obtained as shown in FIG. 3 using the first depth of a sentence as '0'.

Then, the part corresponding to the short sentence among the first recognized partial sentence frames is first translated and replaced with 's' as shown in the figure, and the door frame search for the entire door frame 301 is performed.

If the door frame search and translation is successful, the translation result is output and finished. If the door frame fails, 'pTs' is combined into one door frame according to the depth list value. If the door frame is successful, the translation of the entire door frame 302 is performed. Is attempted. Likewise, if the doorframe search fails, the doorframe combination of 'nCs' is attempted according to the depth list, and the same operation is repeated.

Therefore, by describing the pattern in a proper process, high coverage can be obtained through combining the high translation quality by the pattern method and the stepwise sub-framework by the structural analysis rule.

4 is a flowchart illustrating an example of a sentence division process according to the present invention.

As shown in the figure, for long palm segmentation, first, starting points of all short sentences are extracted from a column of slots which are partial parsing results (401), and starting point candidates are extracted using a starting point pattern (402).

Next, after extracting the end point candidates of the non-relative short sentence corresponding to the start point candidate (403), the end point candidate of the short sentence connected to the related company is extracted using the end point pattern (404).

Thereafter, after extracting and restoring the short candidate, door frame matching is performed to extract the short sentence for the entire sentence (405).

5 is a detailed flowchart of an embodiment of a sentence division process according to the present invention.

First, the starting point candidates of all short sentences are extracted from the column of the slot, which is the result of partial parsing (501, 502).

In this case, step 501 is a process of extracting starting point candidates of short sentences using only syntax information such as a conjunction, a related company, and a punctuation mark, and may be intuitively extracted from a column of input slots. Syntax information used for starting point extraction is as follows.

-Conjunctions: and, but, or

Dependent conjunctions: if, when, before, until, as, because

-Noun conjunctions: that

-Related companies: who, which, that, whose

Comma, quotation marks ("")

-The beginning of a sentence

Next, step 502 is a process of finding a starting point candidate by matching patterns stored in a pre-built short starting point pattern database with an English sentence to be divided. Matching of the English sentence and the short sentence starting point pattern is performed by attempting matching with the short sentence starting point pattern for all or part of the slot sequence between the starting points recognized by step 501, and extracting the starting point candidate when the matching is successful.

The short-term starting point pattern database is a set of patterns for extracting starting point candidates that are not extracted in step 501 due to the omission of a conjunction, and the pattern for the starting point that is not extracted in step 501 from a large number of pre-collected English sentences. Collect and build. The short sentence starting point pattern is a pattern constructed to recognize the starting point by describing syntax information and context information around the starting point using the slot type and feature information. The short sentence starting point pattern is an English sentence. Describe the syntax information and context information of the starting point by describing in the order shown in. Slot names described in the short start point pattern include n, V, T, C, and p, which represent symbols of noun phrases, verb phrases, commas and the like, conjunctions including related words, and prepositional phrases. The feature of a slot is indicated by describing the type of feature, the operator for checking the feature value, and the feature value in the [] after the slot name. / Indicates the starting point position. An example of a short start point pattern is shown below.

V [etype == [D5, T5]] n p / n V [etype == [T1]] p

The pattern is a pattern for estimating the starting point for a sentence in which the noun clause “that” is omitted, as in the sentence “… had assured the U.S. government throughout the day / Russian troops would not cross into Kosovo ...”. T5, D5, T1, etc. represent the verb form.

After extracting all the starting point candidates, the short end candidates corresponding to the starting point candidates are extracted (504 and 505). Here, the process 504 extracts an end point candidate of a short sentence corresponding to a start point of a short sentence that is not connected to an affiliated company, and extracts a starting point of the next short sentence as an end point of the short sentence for a single non-connected short sentence.

In operation 505, an endpoint candidate of a short sentence connected to a related company is recognized. The endpoint candidate is extracted using a short endpoint pattern. The short end pattern is a pattern constructed to recognize the end point by describing the syntax information and context information around the end point using the slot type and the quality information for the end point of the short sentence that does not become the end point of the short sentence. The syntax information and context information of the endpoint are described by describing the types of slots appearing before and after the endpoint and the columns of the corresponding features in the order in which they appear in the English sentence.

The short end pattern is described in the same form as the short start pattern. The pattern matching for end point estimation is a short sentence starting with a related company and end point pattern that is matched by attempting pattern matching on all or part of slot columns between two start point candidate intervals when one or more main verbs exist between two start point candidates. If present, it is extracted as an endpoint candidate.

In this case, when ambiguity occurs in recognition of candidates of the starting point and the end point, all candidates are recognized, and the correct starting point is selected by eliminating the ambiguity through subsequent short sentence matching. Short sentence matching is used to resolve the ambiguity between starting point candidate and end point candidate extraction and to determine the correct starting point and end point. The short sentence matching the short sentence means that the short candidate is a correct short form.

In order to attempt the short sentence matching, a short candidate is first extracted between the starting point candidate and the corresponding end point candidate from the beginning of the English sentence (506).

When the short candidate is extracted, when the subject or object of the sentence is omitted by inversion or omission according to the type of the clause to which the short sentence belongs, the sentence restoration of the short candidate is performed (507). Sentence restoration adds a noun slot in front of a verb when the frame of a sentence except a conjunction and an associated verb starts with a verb, and adds a noun slot when the object is omitted in a clause led by a relative clause and a question. Perform.

Subsequently, the sentence restored short sentence candidate is attempted to be matched with the short sentence frame obtained from the short sentence frame database (508). If the match is successful (509), and if the match is successful, the short sentence candidate is extracted. Proceed to extraction process 506.

Subsequently, the short sentence candidates are extracted in the same form with respect to the remaining sentences in which the short sentences are extracted, the sentence restoration is performed, and the door frame matching is repeated (506 to 510). That is, when the English sentence to be divided into short sentences remains, the short sentence candidate between the other starting point candidate and the end point candidate is repeatedly performed from the short candidate extraction process 506 to divide the entire English sentence into short sentences.

As described above, the method of the present invention may be implemented as a program and stored in a recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.) in a computer-readable form.

The present invention described above is not limited to the above-described embodiments and the accompanying drawings, and various substitutions, modifications, and changes are possible in the art without departing from the technical spirit of the present invention. It will be clear to those of ordinary knowledge.

The present invention as described above, by reducing the length of the door frame to be processed through sentence division and substitution, and performing translation by repeating the step-by-step matching and sub-frame frame combination to reflect the entire clause unit structure, the semantically natural band statement While creating, it has the effect of solving the coverage problem of the door frame base.

That is, the present invention has an effect of obtaining a high coverage and high quality long sentence translation through a highly accurate partial sentence recognition, paragraph unit structure analysis, stepwise door frame application, and partial sentence combination in a door frame-based automatic translation.

Claims (10)

In the long sentence translation apparatus using a partial sentence frame, Preprocessing means for preprocessing the received palmprint to obtain syntax information; Text frame generation means for generating a text frame for the input text using the syntax information obtained through the preprocessing means; And Partial frame processing means for translating long texts by dividing the original text frame generated by the original text frame generating means into partial text frames, recognizing and translating short texts, and then repeatedly performing the entire door frame search, translation, partial text frame combining and translation process Long sentence translation device using a partial sentence frame comprising a. The method of claim 1, The partial frame processing means, After the original text frame generated by the original text frame generating means is divided, the partial text frame is recognized and the translation is performed for each divided partial text frame. If this does not exist, the translation of the input sentence by repeating the process of combining and translating the partial sentence frames through paragraph unit structure analysis and searching and translating the reduced entire sentence frame by substituting the partial sentence into one node is repeated. Long sentence translation device using a partial sentence frame characterized in. The method of claim 2, The partial frame processing means, A long sentence translation device using a partial sentence frame, wherein the partial sentence frame is translated by performing a text frame database search, a band sentence frame selection, and a band word generation process. The method according to any one of claims 1 to 3, The partial frame processing means, After extracting the starting point of short sentence using syntax information, starting point candidate is extracted using starting point pattern. Extracting the end point candidates of the short sentence that are not the relational ambassador corresponding to the starting point candidate and extracting the end point candidates of the short sentences connected to the related company using the end point pattern, A long sentence translation device using a partial sentence frame, characterized in that after extracting and restoring a short sentence candidate, door frame matching is performed to extract a short sentence for all sentences. In a long sentence translation method applied to a long sentence translation device using a partial sentence frame, A palmprint segmentation step of segmenting the palmprint into one or more fragments according to a preprocessing result; A short sentence translation step of recognizing the divided short sentences and performing a short sentence translation through door frame matching; A determination step of determining whether the entire door frame translation is successful after searching the entire door frame to perform the full door frame translation; And As a result of the determination of the determination step, if it is successful, output the translation result, and if it is unsuccessful, iteratively repeats the process of translating and combining the partial sentence frame and the process of searching and translating the entire sentence frame. Long sentence translation method using a partial sentence frame comprising a. The method of claim 5, wherein The palmprint division process of the palmprint division step, A first starting point candidate extracting step of extracting starting point candidates of short sentences using syntax information and extracting starting point candidates using starting point patterns; Extracting end point candidates using the extracted start point candidates and extracting end point candidates using end point patterns; And After extracting and restoring short sentence candidates, short frame extraction process is performed to extract the short sentence for the whole sentence by performing door frame matching Long sentence translation method using a partial sentence frame comprising a. The method of claim 6, The short extraction process, A short code extraction process of extracting a short candidate between the starting point candidate and the corresponding end point candidate; A sentence restoration process of restoring the extracted short sentence candidates into a grammatically complete sentence; Determining a starting point candidate and an end point candidate as a starting point and an end point according to a matching success and attempting to match the restored short sentence candidate with a short sentence; And An iterative process that repeats the short candidate extraction process until the entire long sentence is divided into a set of short sentences. Long sentence translation method using a partial sentence frame comprising a. The method according to claim 6 or 7, The first starting point candidate extraction process, A second starting point candidate extracting step of extracting short sentence starting point candidates such as conjunctions, affiliated companies, punctuation marks, etc. from a slot column that is a result of partial parsing of the input long sentence; And A third starting point candidate extracting process for extracting starting point candidates not extracted in the second starting point candidate extracting process through matching of a short text starting point pattern due to omission of a conjunction Long sentence translation method using a partial sentence frame comprising a. The method according to claim 6 or 7, The first endpoint candidate extraction process, A second end point candidate extracting step of extracting a starting point of the next short sentence as an end point of the short sentence for the short sentence not connected to the related company; And Third endpoint candidate extraction process for extracting the end points of short sentences connected to related companies using short end pattern Long sentence translation method using a partial sentence frame comprising a. In order to translate a long sentence using a partial sentence, in a long sentence translation device having a processor, A palmprint segmentation function for partitioning a sentence into one or more fragments according to a preprocessing result; A short sentence translation function that recognizes the divided short sentences and performs a short sentence translation through door frame matching; A determination function of determining whether the entire door frame translation is successful after searching the entire door frame and performing a full door frame translation; And As a result of the determination in the determination function, if the result is successful, the translation result is output, and if the failure is repeated, the repetition and the translation result output function are performed to repeat the process of combining the partial sentence frame and the process of searching and translating the entire sentence frame. A computer-readable recording medium having recorded thereon a program for realizing this.