JPH01280868A - Preediting support system - Google Patents

Abstract

PURPOSE:To detect no excessive vague part by detecting a vagueness by using the meaning and concept information between clauses, grammatical characteristic information of parallel candidates or the balance information of a sentence structure. CONSTITUTION:A mechanical translation system is constituted of the storage device 201 of a Japanese sentence, a meaning and concept and morphological analyzing dictionary 202, a Japanese analyzer 203, a vague part detector 204, a controller 205, a data storage device 206 and the output device 207 of the alarm of the vague part. The analysis of a linkage to the inputted Japanese sentence is carried out to form a tree structure and detect the vagueness of the parallel Japanese from the balance of the sentence structure, the meaning and the concept of the respective clauses or the grammatical characteristic peculiar to a word understood from this tree structure. Thereby, the detection of the excessive vague parts is suppressed to improve the efficiency in a pre- editing operation.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a system that supports the pre-editing work of Japanese sentences in a machine translation system, and particularly to a system that supports the pre-editing work of Japanese sentences in a machine translation system. The present invention relates to a pre-translation editing support method suitable for supporting pre-editing.

[Conventional technology]

Conventional pre-editing support methods involve morphological analysis of Japanese sentences,
Particles indicating parallel structure (and, ya, and).

etc.), the phrases that exist before and after it.

There are some elements that are taken up as elements constituting a parallel structure and checked with the pre-editor. An example of this type of method is "About the Japanese pre-editor for Japanese-English machine translation" by Shizuka Shindo, Hide Arita, Masatoshi Fukushima, Information Processing Society of Japan, Vol. 3.
1st (late 1985) National Conference Lecture Collection (1349)
~1350 pages), 1985.

[Problem to be solved by the invention]

In the above-mentioned conventional technology, no consideration is given to the semantic concept of phrases, word-specific grammatical characteristics, and JIIi sentence balance, so it is possible to eliminate ambiguity by recognizing the semantic relationships between parallel elements and the role relationships of phrases. Redundant detections may be made for sentences that are understood. Furthermore, when presenting other parallel candidates, there is a possibility that redundant candidates will be raised as well.

An object of the present invention is to provide a pre-editing support method for machine translation that solves the above problems, improves the efficiency of pre-editing work, and reduces the thinking and operational burden on the operator. .

[Means to solve the problem]

The above purpose is to analyze the steps of Japanese morphological analysis and dependency structure analysis used in machine translation in the processing method of a system that supports FJ main body text pre-editing work for machine translation. A step of detecting ambiguous parts related to parallel phrases by considering the semantic concept information, grammatical characteristic information, and syntactic balance of the clauses from the tree structure of the Japanese sentences created using the same analysis method as the above method; This is achieved by providing a step of displaying translated portions regarding parallelization of Japanese sentences and suggesting a pre-editing method based on the detected information.

[Effect]

In this invention, a tree structure is created by analyzing the dependencies of an input Japanese sentence, and the Japanese language is Detect parallel ambiguities in . For this reason, unlike the conventional method of detecting ambiguous parts, which uses all clauses before and after the particle representing a parallel phrase as candidates, it is possible to accurately detect translated parts by suppressing unnecessary ambiguous parts detection. Become. Next, we will display this detection result along with the Japanese sentence, and if we translate the Japanese sentence, how will the translation system understand the meaning, and how can we edit it to obtain the translation with the desired meaning? suggests. This eliminates the need for pre-editors to find out which passages make the syntax ambiguous, and how to correct the sentences to eliminate those ambiguities and enable correct translation. It becomes possible to recognize concretely.

〔Example〕

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

FIG. 1 is a flowchart of processing showing an embodiment of the present invention. In Figure 1, 101 is a Japanese writer Kastep, 102
103 is a morphological analysis step for performing morphological analysis of Japanese sentences, and 103 is a Japanese language analysis step for analyzing the main structure of Japanese sentences.

Reference numeral 104 denotes an ambiguous part detection step for detecting ambiguous parts existing in the main structure of the Japanese text, 105 an ambiguous part determination step for determining whether a translated part is detected in the Japanese text, and 106 an ambiguous part detection unit. This is an ambiguous portion warning step in which a message is created and printed for the pre-editing operator regarding the translated portion detected in step.

FIG. 2 is a schematic configuration diagram of a system showing an embodiment of the present invention. This shows a system that suggests pre-edited parts of the Japanese text by grouping them with respect to the Japanese text before translation processing. In FIG. 2, 201 is a storage device for Japanese sentences, 202 is a morphological analysis dictionary for performing morphological S-element analysis of Japanese sentences and obtaining semantic and conceptual information of chanting words, and 203 is a
204 is a Japanese language analysis device for analyzing Japanese; 204 is an ambiguity detection device for detecting ambiguous portions; 206 is an ambiguity detection device for creating display data for ambiguous portions; Storage device for word tree structure data, 207
is an output device for outputting ambiguous portions and pre-editing suggestion information to the pre-editing worker, and 205 is a control device for controlling them. We will then explain the process flow for pre-editing Japanese sentences.

First, a Japanese sentence is input (101), and a tree structure is created.The tree structure of the Japanese sentence is created by performing morphological element analysis (102), dependency analysis (103), etc. on the Japanese sentence. Created by Here, regarding the creation of the tree structure, for example, see Kaji, Okajima, and Yoshimura, "Machine Translation System Based on Semantic Processing," 2 Hitachi Review, Vofl, 6.
9. Na3. pp.

37-46, 1. It can be created using the method described in March 1987.

Next, ambiguous points are detected in the tree structure, and marks indicating ambiguous points are added to the tree structure data (10
4). Note that the ambiguous part detection process (104) may be included in the 11th main word analysis process (103), that is, a method may be adopted in which a mark indicating an ambiguous part is added during the creation of the tree structure data.

Here, if a mark indicating ambiguity is not added to the tree structure data, the processing for the Japanese sentence ends (105); if a mark indicating ambiguity is added to the tree structure data, the process for the Japanese sentence ends (105). From the wooden structure and the marks attached to it,
11 Display data indicating ambiguous parts in the main text is created. Also, using the landmark as a key, data indicating the pre-editing method is called up, and along with the previously created display data,
Note that the ambiguous portion warning output process 106, which is output to the pre-editing operator (106), is repeated as many times as the number of translated portions detected in the 11!1 flavor portion detection process.

FIG. 3 is a detailed flowchart of translation location detection.

The ambiguous part detection section first checks whether there is a parallel structure in the syntax tree created by the Japanese language analysis section, and if no parallel structure is found, ends the ambiguous part detection process. When a row structure is found in the syntax tree, the following conditions are checked for the relationship between the forked clauses that are parallel elements, and a parallel structure that satisfies the conditions is determined to be an unambiguous structure. On the other hand, for a parallel structure that does not satisfy the checking conditions, it is determined that the parallel structure has ambiguity.

The conditions to check are as follows. (However, if two clauses are analyzed as parallel structures by the translation system,
and B. ) (1) By searching the semantic concept dictionary for the semantic concepts A and B, we can find out, for example, ``translation'' (action noun) and ``preediting'' (action noun) in ``device for translation and pre-editing''. ), it is determined that A "translation" and B "pre-editing" have a parallel structure with no ambiguity (301). However, if there is a nominal C that has a similar semantic concept to A after the parallel structure, it is possible that A and C do not constitute a parallel structure, but A and B do. Therefore, judgment is suspended (302).

(The term “clauses with similar semantic concepts” here refers to phrases with matching semantic concept attributes of the phrases obtained in the morphological analysis step.) (2) A and B, respectively. If there are clauses C and D to modify, compare C and D and if they are the same character, for example "rotation" in "rotation speed and direction", A "speed" and B "direction" are ambiguous. It is determined that the parallel structure has no gender (303).

(3) Regarding the clause C modified by the parallel structure clauses A and B, C is a word that is likely to create a parallel structure between procedural clauses, such as "relationship" in "the relationship between a factory and a research institute." In this case, a mark indicating this property is assigned to a phrase with such a property in advance in the form m dictionary, and the property can be determined by referring to the grammatical property information of the clause when checking the condition. Recognizes A'
It is determined that "Factory" and B "Laboratory" have a parallel structure with no ambiguity (304).

(4) For A and B, count the number of clauses included in each parallel component and compare the numbers to check whether the balance of the number of clauses between the parallel components is disrupted (305 ), for example, in the sentence ``A house made of wood and paper,'' if a machine translation system analyzes that ``wood'' and ``a house made of paper'' are parallel,
The number of clauses with parallel elements centered around is 0, while the number of clauses with parallel elements centered around ``house'' is 2, so A ``tree'' and B ``house'' are ambiguous. , it is determined that the structure is parallel.

Figure 4 is an example of the tree structure of a Japanese sentence created by a machine translation system. This Japanese sentence is an example sentence in which machine translation systems are likely to make mistakes. In other words, the example sentence ``Pre-editing for document input and translation'' can be interpreted in two ways: (Document input and translation) pre-#Q collection (document input) and (Translation first part g!).

When performing dependency analysis of parallel phrases in a translation system where analysis progresses uniquely, for example, even if the analysis is performed based on the recognition that phrases with similar word meanings are likely to have a parallel relationship, the correct translation is not necessarily generated, because there are two action nouns, "translation" and "pre-editing", as parallel partner candidates for the action noun "document input". Therefore, if the example sentence is translated by a machine translation system without pre-editing, there is a risk that a translated sentence with a meaning different from that intended by the writer of the Japanese sentence may be generated. The pre-editing support system deals with such cases and displays a warning and instructions on how to edit it.

FIG. 5 is an example of a machine translation system displaying a warning for example sentences that are likely to make mistakes. The pre-editing operator checks this information on the display device and, for example, if he or she determines that the meaning is (document input) and especially pre-editing for translation, the pre-editing operator can edit the original text by inputting the document and [pre-editing for translation]. By making corrections, it becomes possible to correct ambiguous parts of the Japanese sentence and obtain the intended translation.

〔Effect of the invention〕

According to the present invention, in order to detect ambiguity using semantic concept information, grammatical characteristic information, syntactic balance information, etc. between clauses that are parallel candidates, clauses with the same part of speech before and after the particle expressing parallelism are simply detected. This has the effect that detection errors that tend to occur in ambiguity detection using parallel candidates are less likely to occur. Furthermore, it is possible to suggest specific and detailed pre-editing methods for ambiguities detected by the present invention.

[Brief explanation of the drawing]

Figure 1 is a flowchart of a method showing an embodiment of the present invention;
The figure is a schematic configuration diagram of a system according to an embodiment of the present invention, FIG. 3 is a detailed flowchart of the ambiguous point detection unit 104 in FIG.
The figure is an explanatory diagram showing a tree structure of a Japanese sentence with ambiguity, and FIG. 5 is a diagram showing an example of display showing ambiguous parts in the tree structure of the Japanese sentence. 101... Japanese sentence input step, 102... Morphological analysis step, 103... Dependency analysis step, 104... Ambiguous part detection step, 105... Ambiguous part determination step, 106... Ambiguous point warning step, 201... Japanese sentence storage device, 202.
...Semantic concept and morphological analysis dictionary, 203...Japanese language analysis device, 204...Ambiguous point detection device, 205.
...Control device, 206...Data storage device, 207
...Kaya 1 ■ Kaya 2 concave ￥-3Σ

Claims (1)

[Claims] 1. In the processing method of a system that supports pre-editing work in machine translation, there are two steps: analyzing Japanese morphemes using an analysis method similar to that used in machine translation, and a step similar to the analysis method used in machine translation. a step of analyzing a tree structure representing a dependency structure of a Japanese sentence using an analysis method;
A step of detecting ambiguous points in the parallel structure existing in the tree structure by considering the balance between the semantic conceptual information, grammatical characteristic information, and syntax of the bunsetsu, and detecting ambiguous points in the Japanese sentence based on the detected information. A pre-editing support method characterized by providing a step of displaying and suggesting a pre-editing method.