KR19980031976A - English Long Segmentation Method for English-Korean Machine Translation System - Google Patents

Abstract

The English long sentence segmentation method for improving the success rate and translation speed of the English-Korean machine translation system. The English long sentence segmentation method for an English-Korean machine translation system according to the present invention comprises: a split point extraction process of extracting possible split point candidates such as equivalence conjunctions, subordinate conjunctions, relative pronouns, punctuation marks, etc. from input sentences; Dividing the split point candidate and the candidate into sentence segments; Determining a sentence type of the divided sentence segment; Loads a partition pattern database, which is a set of patterns that describe sentence types, parts of speech, and words that each split point candidate should have on the left and right to become actual sentence split points, from a large amount of sentences collected in advance. And attempting pattern matching by extracting a split pattern of each split point candidate of the input makeup from a split pattern database, and determining whether to determine a sentence split point according to whether the match is successful. In this case, the method includes determining a split point candidate as a split point, and inserting replacement words specified in a pattern that successfully matches the sentence segments before and after the split point to restore the divided incomplete sentence to a complete sentence. The English long sentence segmentation method according to the present invention has the effect of improving the translation success rate and the translation rate by dividing the English long sentence into short sentences that can be successfully translated by the English-Korean machine translation system.

Description

English Long Segmentation Method for English-Korean Machine Translation System

The present invention relates to an English-Korean machine translation system, and more particularly, to an English long sentence segmentation method for improving translation system success rate and speeding up translation speed.

The English-Korean machine translation system of the general transfer method consists of three stages of interpretation, transformation and generation. The interpretation step includes English morpheme analysis, syntax analysis, and semantic analysis process, and the generation step includes Korean phrase generation and Korean morpheme generation process.

The performance of the machine translation system, that is, the translation success rate and the translation speed, becomes very low when the English sentence becomes longer. English long texts increase the complexity of the translation process, causing translations to fail or produce incorrect translations. Specifically, parsing failures and errors for English texts in the interpretation phase cause translation failures and incorrect translations.

In order to solve this problem, it is necessary to divide the English text into short sentences.

An object of the present invention is to provide a method for dividing an English long sentence into a short sentence that can be successfully translated to improve the performance of the English-Korean machine translation system.

1 is a flowchart illustrating an operation process of an English-Korean machine translation system to which the English long sentence segmentation method according to the present invention is applied.

FIG. 2 is a flowchart illustrating a sentence division method in the sentence division process illustrated in FIG. 1.

3 shows an example of sentence division.

4 shows an example of a finite state automata for sentence type classification.

The English long sentence segmentation method of the English-Korean machine translation system according to the present invention achieves the above object. A split point extraction process for extracting; Dividing the split point candidate and the candidate into sentence segments; Determining a sentence type of the divided sentence segment; Loads a partition pattern database, which is a set of patterns that describe sentence types, parts of speech, and words that each split point candidate should have on the left and right to become actual sentence split points, from a large amount of sentences collected in advance. Extracting a split pattern of each split point candidate of the input makeup from a split pattern database, attempting pattern matching, and determining whether to determine a sentence split point according to whether the matching is successful; And when the pattern matching process is successful, determining a split point candidate as a split point, and inserting replacement words specified in a pattern that successfully matches the sentence segment before and after the split point, and restoring the partitioned incomplete sentence into a complete sentence. It is characterized by. Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating an operation process of a system to which an English palmprint splitting method according to the present invention is applied.

First, in step 102, a proper part-of-speech sequence is determined by performing a morpheme analysis on the input English sentence S.

In step 104, the sentence division is performed on the part-of-speech sequence that is the result of the morpheme analysis process (step 102) by the long sentence segmentation method according to the present invention. For example, the input English sentence S is divided into split statements S1, S2, S3, ....

Step 106 checks whether the statements S1, S2, S3,, and the like are valid, and creates a parsing tree showing the analysis structure if appropriate.

Step 108 analyzes the meaning of the parse tree, which is the result of English parsing. For this purpose, the semantic analysis step searches each node of the parsing tree and assigns the corresponding semantic meanings according to the branched form and contents.

In step 110, the conversion to the intermediate expression is performed by applying a conversion rule for resolving lexical and structural differences between English and Korean.

In step 112, a Korean phrase is generated from an intermediate expression that is a result of the generation process (step 110).

In step 114, a morpheme is generated on the result of the syntax generation process (step 112), and the final translations K1, K2, K3 ,, for the sentence S1, S2, S3 ,, are output.

2 is a flowchart illustrating a sentence division method in step 104.

In step 202, all possible split point candidates are extracted while reading the input sentence S from the beginning to the end. Here, the split point candidates used are as follows.

-Nominated conjunctions: and, but, for ,,,

Dependent conjunctions that lead to adverb clauses: before, after, after, because, since, as, if, unless, though ,,,

-Relative pronouns: who, which, that, what, whose, whom ,,,

-comma, quotation marks ,,,

Step 204 divides the split point candidate and the candidate. Each divided section is called a sentence segment. 3 shows an example of sentence division. For example, it can be seen that the sentence segment SEG3 is defined between the split point candidates BP3 and BP4.

In step 206, a sentence type is determined using a finite state automata for each sentence segment. Figure 4 shows an example of a finite state automata for different types of length. In FIG. 4, NP_cat, VP_cat, and PP_cat are sets of parts of speech belonging to noun phrases, verb phrases, and prepositional phrases, respectively.

In step 208, the partition pattern database is searched for each split point candidate of the input sentence, and pattern matching is performed to determine whether to split the sentence. Specifically, a sentence similar to a sentence type appearing before and after a split point is searched by referring to pre-sampled sentences and a translation example corresponding thereto, and when there is a similar sentence, the sentence is searched before and after the split point by referring to a translation example of the searched sentence. This is the process of obtaining a string to be inserted.

The following describes the notation used to describe the partition pattern of the partition candidate points constituting the database used in step 208. First, the target split point candidate is specified, and on the left, the word, sentence type, and part of speech that may appear on the left of the split point are described, and on the right, the word, sentence type, and part of speech that may appear on the right of the split point are described. The left substitution describes the substitution to be inserted at the end of the sentence segment before the split point when the pattern matching succeeds, and the right substitution describes the substitution to be inserted before the sentence segment after the split point.

Split Pattern Notation for Split-Point Candidates

{Split point candidate leftrightleft replace right replace}

-leftright notation

no mark word: words

% SentenceType: (w _f , w _s ) [cat_w _f , cat_w _s ]

Part of article: (feature value ,,, / feature value)

w _f : First word of sentence segment

w _s : last word of sentence segment

cat_w _f : part of w _f

cat_w _s : part of w _s

Sentence type: S, NP, VP, VP_ing, VP_ed, PP

Part of speech: ADJ (adjective), ADV (adverb), DET (definitive), NOUN (noun), PREP (preposition), PRON (pronoun), AUXI (verb), VERB (verb), CONJ (adverb), PUNCT ( Separator)

left-right replacement

NO_MARK

IS (provisional), IS + verb,

IO (target term)

Patterns of Split Point Candidates in a Split Pattern Database and Their Applications

(1) an example of the conjunction and; {and% S @VERB: (FORM pres ed) IS}

I typed a sentence; I was a student and studied hard.

SEG1 = S ^ SEG2 = VP

Split sentence; I was a student. IS studied hard.

Explanation; The split point candidate and has the left sentence segment as S, and is a split point when the part of the word immediately after the split point candidate is a verb (the FORM feature must have a value of pres or ed). At this time, there is no substitute for the left sentence segment, and the word IS is inserted before the right sentence segment.

(2) an example of the split pattern of commaa; {comma including IS is}

Input sentences; Republicans would limit the right, including food stamps and Medicaid.

SEG1 = S ^ SEG2 = VP_ing

Segmented sentences; Republicans would limit the right IS is food stamps and Medicaid.

Explanation; The split point candidate comma becomes the split point when the word immediately after the split point candidate is including, regardless of what comes to the left sentence segment. Insert IS and the verb is in front of the right sentence segment.

In step 210, if the pattern matching performed in step 208 succeeds, the split point candidate is determined as the split point, and the left and right substitute words specified in the successful pattern are inserted into the sentence segment before and after the split point, and then restored to a complete split statement.

As described above, the long text segmentation method according to the present invention has the effect of improving the translation success rate and the translation rate by dividing the English long text into short sentences that the machine translation system can successfully translate.

In addition, by translating the fragments of the original text, there is an advantage that it is possible to perform a robust parsing and translation even for a sentence that uses a partly wrong grammar.

Claims (1)

In English long sentence segmentation method for English-Korean machine translation system, The English long sentence segmentation method of the English-Korean machine translation system The English long sentence segmentation method of the English-Korean machine translation system includes a segmentation point extraction process of extracting possible segmentation point candidates such as equalized conjunctions, dependent conjunctions, relative pronouns, punctuation marks, etc. from input sentences; Dividing the split point candidate and the candidate into sentence segments; Determining a sentence type of the divided sentence segment; Loads a partition pattern database, which is a set of patterns that describe sentence types, parts of speech, and words that each split point candidate should have on the left and right to become actual sentence split points, from a large amount of sentences collected in advance. And attempting pattern matching by extracting a split pattern of each split point candidate of the input makeup from a split pattern database and determining whether to determine a sentence split point according to whether the matching is successful; and If the pattern matching process is successful, the method includes determining a split point candidate as a split point, and inserting replacement words specified in a pattern that successfully matches the sentence segment before and after the split point to restore the incomplete fragmented sentence into a complete sentence. Long way of dividing.