KR20060064443A - Apparatus for automatic translation through 2-step syntactic analysis and method thereof - Google Patents

Abstract

Automatic translation apparatus and method through two-step syntax analysis according to the present invention includes a morpheme analysis unit for analyzing the morphemes for the input sentence; A tagging unit for determining a part-of-speech candidate for each of the morphemes; A syntax analysis unit for parsing the tagged input sentence based on a verb phrase and then parsing the entire sentence of the input sentence; And a conversion generator for generating a translation of the input statement based on the syntax tree generated as a result of the parsing. The tagging ambiguity and the noun phrase chunking ambiguity and the parallel and attachment ambiguity are separated from the parsing ambiguity. In the first step, the tagging ambiguity and noun phrase chunking ambiguity are resolved, and the two-step parallel and attachment ambiguity are resolved to increase the analysis efficiency without significant performance degradation.

Automatic translation, syntax analysis, sentence division, basic noun phrases, verb phrase analysis

Description

Apparatus for automatic translation through 2-step syntactic analysis and method

1 is a block diagram showing the configuration of an automatic translation apparatus through two-step syntax analysis according to the present invention.

FIG. 2 is a detailed functional block diagram of the parser of FIG. 1.

3 is a flowchart illustrating a process of an automatic translation method through two-step syntax analysis according to the present invention.

4 is a flowchart illustrating the syntax analysis step of FIG. 3 in more detail.

<Description of the symbols for the main parts of the drawings>

110: morphological analysis unit 120: tagging unit

130: parser 140: transform generator

The present invention relates to an automatic translation apparatus and method, and a recording medium thereof, and more particularly, to an automatic translation apparatus for generating a final translation through syntax analysis consisting of two steps of a verb phrase analysis step and an entire sentence analysis step on an input original text, and A method and a computer readable recording medium having recorded thereon a program for realizing the method.

In the conventional rule-based automatic translation method, there is a problem in that the speed and translation performance are deteriorated due to the ambiguity of the syntax analysis and the unlimited generation of band phrases. The ambiguity problem grows explosively, especially as the sentence gets longer, and there is a sentence segmentation method in which the sentence is divided into smaller units and processed before parsing.

However, since sentence splitting is performed with limited information before structural analysis, there is a high possibility of incorrect results and limitations in performance when a wide range of contexts are required. And if there is a failure in the sentence splitting, there is a problem that the entire parsing fails, the whole translation fails, causing a great side effect.

The technical problem to be achieved by the present invention was devised to solve the above problems, by separating the syntax analysis process into a verb phrase analysis step and the whole sentence analysis step to minimize side effects due to sentence division, and also in parallel and modifier attachment The present invention provides an automatic translation apparatus and method having high analysis accuracy and efficiency by reducing ambiguity and a computer-readable recording medium recording a program for realizing the method.

In order to achieve the above technical problem, an automatic translation apparatus through two-step syntactic analysis according to the present invention includes a morpheme analysis unit for analyzing a morpheme for an input sentence; A tagging unit for determining a part-of-speech candidate for each of the morphemes; A syntax analysis unit for parsing the tagged input sentence based on a verb phrase and then parsing the entire sentence of the input sentence; And a transformation generator for generating a translation of the input sentence based on the syntax tree generated as a result of the parsing.

In order to achieve the above technical problem, an automatic translation method using two-step syntax analysis according to the present invention comprises: determining a part-of-speech candidate after analyzing a morpheme for an input sentence; Generating a partial tree list by parsing the tagged input statement and parsing the split phrase around the verb phrase; Recognizing a part-of-speech list and a basic noun phrase in the partial tree list and parsing the entire sentence; And generating a translation sentence of the input sentence based on a syntax tree generated as a result of parsing the entire sentence.

In order to achieve the above technical problem, a computer-readable recording medium that records a program capable of executing an automatic translation method through two-step syntax analysis according to the present invention on a computer may be analyzed by morphemes for input sentences. Determining a candidate; Generating a partial tree list by parsing the tagged input statement and parsing the split phrase around the verb phrase; Recognizing a part-of-speech list and a basic noun phrase in the partial tree list and parsing the entire sentence; And generating a translation sentence of the input sentence based on a syntax tree generated as a result of parsing the entire sentence.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. For convenience of description, the apparatus and method will be described simultaneously.

1 is a block diagram showing a configuration of an automatic translation apparatus through two-step syntax analysis according to the present invention, and FIG. 2 is a detailed functional block diagram of the syntax analysis unit 130 of FIG. 3 is a flowchart illustrating a process of an automatic translation method through two-step syntax analysis according to the present invention, and 4 is a flowchart illustrating the syntax analysis step of FIG. 3 in more detail.

Referring to Figure 1 and 3, looking at the overall flow of the automatic translation device through a two-step syntax analysis according to the present invention, the morpheme analysis unit 110 performs a morphological analysis on the input sentence, tagging unit 120 In step S310, tagging determines a part-of-speech candidate for each morpheme. The parser 130 analyzes the tagged input sentence based on the verb phrase and then parses the entire input sentence (steps S320 to S350). In more detail, sentence segmentation is performed on the input sentence resulting from the tagging result (step S320). Then, verb phrase-centered syntax analysis is performed on each split statement, and the base noun phrase (Base NP) is recognized in the optimal syntax tree (S340). Here, a basic noun phrase means a noun phrase that does not include a noun or a verb phrase below it. If the basic noun phrase is recognized, a full sentence parsing is performed on the second recognized result (step S350). The conversion generation unit 140 generates a translation sentence of the input sentence based on the syntax tree generated as a result of the final operation of the parser 130 and finally outputs the translation sentence (S360).

2 and 4, the two-step syntax analysis process will be described in more detail. First, the sentence dividing unit 210 receives the tagged sentence and generates a sentence dividing point (S410). Then, the primary sentence analysis unit 220 performs syntax analysis based on the sentence splitting point, and receives an optimal N (N is a natural number) part-of-speech candidate for each morpheme. That is, parsing is performed with tagging ambiguity, and the tagging performance is increased by determining the part-of-speech tag by reflecting the result of parsing. In the process of parsing, constraints are attached to the closest attachable nodes for parallel and modifiers. This is to increase the analysis efficiency by removing parallel and attachment ambiguity, which causes many ambiguities in syntax analysis, because parallel and modifier attachments have little effect on parts of speech and basic noun phrase chunking. This verb phrase parsing is mainly aimed at performing correct verb phrase parsing. Therefore, even if sentence division is made, the information necessary for the analysis of one verb phrase is localized compared to the entire sentence, so that a relatively accurate analysis can be performed (step S420).

The basic noun phrase recognition unit 230 outputs a basic noun phrase that appears as a result of performing the verb phrase unit parsing as described above, and the sentence division unit 210 parses the sentence in the state where the sentence is split. The analysis result is shown in the form of a partial tree list rather than a syntax tree for the entire sentence, and an optimal partial analysis tree list is determined through a Viterbi algorithm (step S340).

When the optimal partial tree list is generated, the secondary syntax analysis unit 240 recognizes the parts of speech list and the basic noun phrase, and performs the entire sentence analysis again. In this case, the sentence splitting point, parallelism, and constraints are removed, and the entire sentence is parsed. In this case, since the part-of-speech tag is determined as one and the basic noun phrase chunking is performed, the number of input charts is reduced and the ambiguity is reduced, thereby enabling parsing of the entire sentence. In addition, even if there is a sentence segmentation error, the sentence segmentation error does not affect the error unless the segmentation error divides the inside of the verb phrase and causes the verb phrase analysis error (step S440).

Now, a practical example of analyzing the following example by applying a two-step syntax analysis method according to the present invention is presented.

1. Input Statement: Former Gov. Stratton commuted her term to 75 years, making her eligible for parole, as one of his last acts in office.

2. Tagging Result: If N = 2 above

Former / JJ / NNP Gov./NN Stratton / NNP commuted / VBN / VBD her / DT / PRP term / NN / VBP to / TO 75 / CD years / NNS / NN, making / VBG / JJ her / DT / PRP eligible / JJ / NN for / IN parole / NN / JJ, as / IN one / PRP / CD of / IN his / DT / PRP last / JJ / VB acts / NNS / VBZ in / IN office / NN / NNP.

3. Segmentation Result: <Former / JJ / NNP Gov./NN Stratton / NNP commuted / VBN / VBD her / DT / PRP term / NN / VBP to / TO 75 / CD years / NNS / NN>, <making / VBG / JJ her / DT / PRP eligible / JJ / NN for / IN parole / NN / JJ, as / IN one / PRP / CD of / IN his / DT / PRP last / JJ / VB acts / NNS / VBZ in / IN office / NN / NNP>

4. Perform primary parsing:

(S (NP Former / JJ Gov./NN Stratton / NNP)

(VP commuted / VBD

(NP her / DT term / NN)

(PP to / TO (NP 75 / CD years / NNS))))

,

(VP making / VBG

(NP her / PRP)

(ADJP eligible / JJ)

(PP for / IN parole / NN))

,

(PP as / IN

(NP (NP one / PRP)

(PP of / IN

(NP (NP his / DT last / JJ acts / NNS)

(PP in / IN (NP office / NN))))))

5. Recognize parts of speech and basic noun phrases

(NP Former Gov Stratton) commuted / VBD (NP her term) to / TO (NP 75 years), making / VBG (NP her) eligible / JJ for / IN (NP parole / NN), as / IN (NP one) of / IN (NP his last acts) in / IN (NP office)

As seen from above, the input of secondary parsing is a state of speech ambiguity removed and the overall sentence length reduced.

6. Parse whole sentence

(S (NP Former Gov Stratton)

(VP commuted

(NP her term)

(PP to (NP 75 years))

(PRN,

(VP making

(NP her)

(ADJP eligible)

(PP for

(NP parole)))

,)

(PP as

(NP (NP one)

(PP of

(NP his last acts)

(PP in

(NP office)))))))

The original sentence division resulted in an error that the prepositional phrase attached to making, such as (Former Gov. Stratton commuted her term to 75 years), (making her eligible for parole, as one of his last acts in office) In the parsing process, the insertion phrase is recognized to produce the correct parsing result. The result of the analysis is the final translation through the conversion and generation of the target language.

The automatic translation method through two-step parsing according to the present invention can also be embodied as computer readable codes on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, which are also implemented in the form of a carrier wave (for example, transmission over the Internet). It also includes. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

What has been described above is only one embodiment for implementing the automatic translation method and apparatus through two-step syntax analysis according to the present invention, the present invention is not limited to the above-described embodiment, the scope of the claims Without departing from the gist of the present invention claimed in the present invention, anyone of ordinary skill in the art will have the technical spirit of the present invention to the extent that various modifications can be made.

According to the automatic translation apparatus and method through two-step syntax analysis according to the present invention as described above, tagging ambiguity and noun phrase chunking ambiguity and parallel and attachment ambiguity among syntax parsing ambiguities are separated and tagging ambiguity and noun phrase chunking in one step. By solving ambiguities and solving two-level parallelism and attachment ambiguities, there is an effect of increasing the analysis efficiency without significant performance degradation. This is made possible by doing this independently because parallel and attachment ambiguities have little effect on determining parts of speech tags and basic noun phrases.

Claims (9)

A morpheme analysis unit for analyzing a morpheme with respect to an input sentence; A tagging unit for determining a part-of-speech candidate for each of the morphemes; A syntax analysis unit for parsing the tagged input sentence based on a verb phrase and then parsing the entire sentence of the input sentence; And And a translation generator for generating a translation of the input sentence based on the phrase tree generated as a result of the parsing. 2. The method of claim 1, wherein the parser is A sentence division unit that performs sentence division on the tagged input statement; A primary parsing unit which performs verb phrase-centered parsing on each partitioned statement and selects an optimal partial tree list; A basic noun phrase recognition unit recognizing a basic noun phrase in a syntax analysis result of the primary parser; A second parsing unit configured to perform full parsing based on the part-of-speech tag of the optimal partial tree list and the recognized basic noun phrase; And And a translation generator for generating a final translation after performing the conversion according to the phrase tree. The method of claim 2, wherein the primary parser is Automatic translation through two-step parsing, characterized by having the tagging ambiguity of the best N (N is a natural number) candidate, analyzing the syntax of the partitioning statement by limiting parallel and modifier attachment, and selecting the optimal subtree list Device. The method of claim 2, wherein the secondary parser is Automatic translation device through two-step parsing, characterized in that to remove the parallel and attachment constraints and parse the entire sentence by inputting the parts of speech tag and the basic noun phrase of the optimal partial tree list. (a) analyzing the morphemes for the input sentence and determining candidates for parts of speech according to the morphemes; (b) generating a partial tree list by performing sentence division of the tagged input statement and parsing the fragment statement around the verb phrase; (c) recognizing a part-of-speech list and a basic noun phrase in the partial tree list and parsing the entire sentence; And and (d) generating a translation of the input sentence based on the syntax tree generated as a result of the step (c). 2. The method of claim 5, wherein step (b) (b1) generating a sentence splitting point for the input sentence; (b2) receiving an optimal N part-of-speech candidate for each morpheme as an input and performing syntax analysis on the boundary of the sentence content; And (b3) generating an optimal partial tree list based on the parsing result of step (b2). 2. The method of claim 6, wherein step (b3) (b31) having a tagging ambiguity of the best N (N is a natural number) part-of-speech candidate, analyzing the syntax of the partitioning statement by limiting parallel and modifier attachment, and selecting an optimal subtree list; Automatic translation through two-step parsing.  The method of claim 5, wherein step (c) Removing the sentence splitting points, parallelism, and constraints, and performing syntax parsing on the entire sentence. 2.  (a) analyzing the morphemes for the input sentence and determining candidates for parts of speech according to the morphemes; (b) generating a partial tree list by performing sentence division of the tagged input statement and parsing the fragment statement around the verb phrase; (c) recognizing a part-of-speech list and a basic noun phrase in the partial tree list and parsing the entire sentence; And and (d) generating a translation of the input sentence based on the syntax tree generated as a result of the step (c).

Images