JPH01312666A - Pre-edition supporting system - Google Patents

Abstract

PURPOSE:To raise the efficiency and the quality of pre-editing related to the supplement of an essential case by analyzing the essential case, and detecting all declinable words lacking in the essential case, and outputting a warning. CONSTITUTION:When a sentence is inputted, at first, by analyzing the essential case, it is analized what clause fulfills the role of the essential case for what declinable word. Next, by detecting all the declinable words lacking in the essential case according to the result and by outputting the warning, the declinable word not having the clause to fulfill the role of the essential clause at all is detected, and in the case that such a declinable word exists in the sentence, the warning is issued. The worker of the pre-editing sees the warning, and corrects the inputted sentence. Thus, the efficiency and the quality of the work of the pre-editing related to the supplement of the essential case is improved.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a pre-editing support method for a pre-editing support device for machine translation, and in particular, to improve the success rate of machine translation. The present invention relates to a pre-editing support method suitable for supporting the efficient rewriting of target Japanese sentences.

[Conventional technology]

Conventional pre-editing support devices for machine translation include, for example, Arita, Fukushima, and Shindo, "About pre-editing in Japanese-English machine translation systems," Information Processing Society of Japan Natural Language Processing Study Group Materials, 48-7, 1985. As discussed on March 27th.

It detects ambiguity in the meaning of attached words, ambiguity in dependencies, and unregistered words, presents candidates for the correct answer, and lets the user select the candidate.

[Problem to be solved by the invention]

Conventional pre-editing support devices such as those described above do not have a function to detect i° (verbs, adjectives, adjective verbs) that lack an essential case, so finding words that lack an essential case is difficult. This must be done purely by humans, and there is a problem in that both the efficiency of the pre-editing work related to the addition of essential cases and the quality of each work are poor.

SUMMARY OF THE INVENTION An object of the present invention is to provide a pre-editing support method that solves these conventional problems and improves the efficiency and quality of pre-editing regarding replenishment of essential cases.

[Means to solve the problem]

To achieve the above objects, the present invention includes a step of analyzing essential cases to support pre-editing of input sentences to increase the success rate of machine translation, and detecting sentences in which all essential cases are missing. However, it is characterized by providing a step for outputting a warning.

[Effect]

In the present invention, when a sentence is input, first.

In the step of analyzing the essential case, it is analyzed which clause among the clauses in the input sentence plays the role of the essential case for which usage. Next, based on the result, the step of detecting a predicate that lacks all mandatory cases and outputting a warning detects a predicate that does not have any clause that plays the role of a mandatory case, If such a use i is present in a sentence, a warning will be issued. The pre-editing operator sees the warning and corrects the input sentence.

As described above, according to the present invention, a predicate that is missing all essential cases is automatically detected, so that the pre-editing support device for pre-editing requires less time to complete the essential cases. The efficiency and quality of editing work improves. Also,
A warning is issued only when all essential cases are missing, so compared to a system that issues a warning when even one essential case is missing, there are fewer redundant detections and fewer pre-edits due to over-detection. There is almost no burden on the worker.

〔Example〕

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

This embodiment is a device that supports the work of correcting an input sentence for a system that translates from Japanese to English. 1st
The figure is a diagram showing the flow of processing in an embodiment of the present invention. The processing of this device is 101 literary skills, 1 morphological analysis/phrase recognition,
02, modification consists of five processes: analysis 103, mandatory case analysis 104, and detection/warning output 105 of all mandatory cases missing.

FIG. 2 is a diagram showing the configuration of this embodiment. As shown in FIG. 2, this device includes a central processing unit 201, a storage medium 202, a storage medium 203, a printer 204, and a CRT (
It consists of a cathode ray tube) 205 and a human-powered device 206. The central processing unit 20 analyzes the input sentence while controlling each component according to the processing procedure stored in the storage medium 203, detects a predicate that lacks all mandatory cases, and issues a warning. It performs the function of outputting.

The storage medium 202 is for storing work data essential for processing. The storage medium 203 is for storing the entire processing procedure and the dictionary 1 grammar rules necessary for analyzing the input sentence.

The dictionary contains data such as parts of speech, conjugation information, semantic features, case-dominant patterns, and codes representing grammatical characteristics of each word, and the key to this is the visit of each word to Korea.

The printer 204 is for printing output information of the device, such as IF notification. The CRT 205 is for displaying device output information such as warnings. Input device 2
06 is for inputting sentences to be processed, control commands, etc., and usually a Japanese input device having a kana-kanji conversion function is used. Of course, as the sentence to be processed, Japanese sentence data stored on a magnetic tape or magnetic disk may be input via a magnetic tape drive or a magnetic disk drive.

Details of each process in FIG. 2 are shown below.

(1) Literary Power 101 When a Japanese sentence is input from the human power device 206, the data of the Japanese sentence is stored in the storage medium 202.

(2) Morphological analysis/clause recognition 102 While performing dictionary lookup based on the principle of longest match, use the word adjacency conditions (part of speech, conjugation type, and conjugation form of the front word, and constraints regarding the part of speech of the back word), Perform morphological analysis of the input sentence and divide it into words 6. Next, apply a rule to eliminate the multi-item side (for example, word X, which can be both a verb and a noun, is immediately followed by a noun. Then, apply a rule such that the part of speech of word X is a noun, and narrow down the part of speech of each word to one. Furthermore, after that, each word is grouped into a phrase based on the standard that the phrase from an independent word to before the next independent word is a phrase.

(3) Dependency analysis 103 Based on information such as parts of speech, function words, conjugated forms, etc., the possibility of a dependency relationship between each clause is determined, and under the non-intersection condition,
Dependency relationships between phrases are set so that each phrase is related to a phrase as close as possible. However, as the type of reception, MYOU in the case of conjunctive modification, M in the case of conjunctive modification.
In the case of TAI and noun chain, TEMN is set. Furthermore, dependency modifies relationships, and dependency modifies relationships (for example, a verb that acts like a function word, such as "through", should only have one case element). FIG. 3 is a diagram showing an example of data generated as a result of modification analysis. Figure 3 (A).

Items (B) and (C) said, ``Reliability has been increased by providing a backup system.'' ``My brother was walking on the fence.'' ``I read the book I bought.'' The dependency in the sentence ``'' indicates a relationship. In this way, by analyzing the dependencies, a network structure with each clause as a node is generated.

(4) Required case analysis 104 Determines deep cases (required cases) based on case dominant patterns registered as dictionary data of predicates.

Change the dependency arc label (modification type) to MYO
Change from U or MTAI to a deep case code. A case-dominant pattern is a pattern of essential cases that is dominated by the word, and is a particle code.

It is described using a combination of deep case codes and semantic features. Figure 4 (A) and (B) are similar to Figure 3 (A) and (B).
FIG. 3 is a diagram showing the results of performing essential case analysis on the data of . The circle in the figure is a case code that means the object case. In addition,
If the use i modifies a case element, the deep case before the synopsis of the modifier light is also used.

Determine based on case dominance pattern. In this case, change the label of the arc that extends from the predicate to the outline of the modifier light to a code with an E at the beginning of the case code of the deep case, and then change the dependency from the case element of the modifier light to the predicate. Create an arc for the embedded sentence that is different from the arc in , and set the deep case code as the label for that arc. For example, if we apply essential case analysis to the data in Figure 3(c), we will see Figure 4(
c) data is available. Here, 1 from "bought"
The label E○ on the arc attached to ``Book'' is a case code representing an object case with an E added to the beginning, indicating that it is an object case as an embedded sentence. Also,"
The dotted line arc extending from ``book'' to ``bought'' is the arc for the embedded sentence. This arc shows that the object case of ``I bought J'' is ``1 book ノ''.

(5) Detection and warning output of all mandatory qualifiers missing 105 Figure 5 shows:
FIG. 7 is a diagram showing a detailed flow of processing for detecting the absence of all mandatory requirements and outputting a warning. The flow will be explained according to FIG. The data generated by the essential case analysis 104 is to be processed,
First, focus on the first bunsetsu node (process 301).Next, check whether the central word of the bunsetsu node you are looking at is idiom d (process 302).If it is not a term, process 30
Proceed to step 6, and if it is a predicate, proceed to the next process 303, and the clause node of interest is Myou.

Check whether there are child nodes connected by arcs with arc labels other than MTAI (
Process 303). If it has, proceed to process 306; if it does not, proceed to the next process 304.0 Next, it is checked whether an arc for an embedded sentence is extended to the phrase node of interest (process 304). . If the arc for the embedded sentence is stretched, the process advances to step 306; if not, the process advances to the next step 305. At this point, the clause node of interest has all essential cases. Since it is obvious that the clause is missing, a mark indicating this is added to this clause node (processing 305), and it is checked whether processing has been completed for all clause nodes (processing 306). ).

If the processing has not been completed, the process returns to step 302, focusing on the first clause node (step 307). If the processing has been completed, a warning is output regarding the phrase of the usage ° that lacks all essential cases, using the mark given in step 307 as a clue.

FIG. 6 shows an example of the warning.

6(A) and (B) are respectively FIG. 5(A).

As a result of performing processing 105 on the data in (B),
This is the warning that was output. By seeing this warning,
Regarding Figure 6 (A), the pre-editor said, ``We have increased reliability by providing a backup system.'' and Figure 6 (B).
): ``As I was walking, I was looking over the fence.

” Correct the sentence as follows. Regarding the data in Figure 5(C), the case i° is missing all essential cases.
does not exist, so no warning is output.

As described above, through the processes (1) to (5) described above, if a predicate that lacks all essential cases exists in the input sentence, a warning regarding the predicate is output.

〔Effect of the invention〕

According to the present invention, the following four effects can be obtained.

(1) According to the present invention, since the pre-editing support device detects the use i° that is missing all essential cases, omissions in pre-editing regarding the addition of essential cases are reduced.
The quality of pre-editing will improve, as will the efficiency of pre-editing.

(2) ■ Machine translation does not necessarily require all essential cases to be satisfied (for example, even if tsutsume is omitted, if the object is present, it can be translated in the passive voice), ■
There are two reasons why Japanese is often omitted. If we were to detect and report when there is a predicate in a sentence in which even one essential case is omitted, it would result in over-detection and This actually increases the burden on the editing worker. However, according to the method of the present invention, a warning is issued only when all essential cases are missing, so there is no possibility of over-detection.

(3) Due to incorrect analysis of machine-processed dependencies, a warning may be issued that all essential cases are missing. Therefore, conversely, this warning may suggest to the pre-editor that the modification caused by machine processing is incorrectly analyzed. Therefore, according to the present invention, it is possible to improve the quality and efficiency of pre-editing regarding dependency.

(4) If there is an ``i'' in a sentence that lacks all essential cases, it will be difficult not only for machine translation systems but also for humans to understand the sentence.Therefore, there is a warning about the lack of all essential cases. This invention can also be used to support elaboration of a single sentence.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the processing flow of an embodiment of the present invention, and FIG.
The figure is a block diagram of an embodiment of the present invention, Figure 3 is a diagram showing an example of the result of dependency analysis, Figure 4 is a diagram showing an example of the result of mandatory case analysis, and Figure 5 is a diagram showing an example of the result of mandatory case analysis. FIG. 6 is a diagram showing details of the process of detecting a lack and outputting a warning, and is a diagram showing an example of a warning regarding the lack of all mandatory requirements. 101...Literature ability, 102...Morphological analysis/Phrase recognition. 103... Dependency analysis, 104... Essential case analysis. 105... Detection/warning output of all mandatory qualifications missing, 201.
...Central processing unit, 202...Storage medium, 203...
・Storage medium, 204...Printer, 205...CR
T. $ 10 Kaya 2 causes 30 + A) (B) Fig. 4 (^) (C,) Kaya 5 wards) l zero deposit 1 (regarding IJ, 7 hTζ. Total I zo inclination 7? - nυ four banks Noyoto No. 1 (fi＼ ム fig. (A) Kokushi fly t1 Esu ζ hi 1 and J Soyama)

Claims (1)

[Claims] 1. In a method that supports pre-editing of input sentences to increase the success rate of machine translation, a step of analyzing essential cases;
A pre-editing support method comprising the steps of: detecting a predicate that lacks all essential cases and outputting a warning.