JPS58192173A - System for selecting word used in translation in machine translation - Google Patents

Abstract

PURPOSE:To decrease the frequency in correction, by translating a word the translation equivalent of which is not narrowed down to one translated word only with the sentence structure and the analysis of meaning, with a dictionary storing the up-to-date translated words, in a system translating a sentence in the 1st language into the sentence in the 2nd language. CONSTITUTION:Lots of records cmprising translated words, grammatical information and information representing the up-to-date translated words are stored in a dictionary memory 3 of the translation system, as to entries of words and composite words including the entires of the 1st language, and a sentence to be translated is read out from a text file 2 at first, the sentence is sectioned into word and phrases, and records taking the words in the sentence as the entries are read out from the dictionary memory 3, the structure of sentence is analyzed by referencing the grammatical rules and the read-out grammatical information of the 1st language, the setnence is converted into a possible structure by referencing the rules of conversion of structure of the 2nd language, and a suitable translated word is selected. But, if >=2 translated words remain as candidates, an up-to-date translated word flag 13-1 is referenced and the translated word being set on is selected. If not set on, the first translated word is selected.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a system for translating sentences in a first language into sentences in 28 languages, and in particular for words and phrases that cannot be narrowed down to one word by syntax and meaning analysis alone. #
This relates to a method for selecting I.

In general, machine translation systems analyze the syntax of an input sentence, recognize sentence components such as phrases and phrases, and recognize sentence structures, convert sentence structures, and add translations to words. After that, an output statement is generated. At this time, it has multiple meanings,
For so-called polysemous words, it is often difficult to determine the translation based only on information about the word itself and its syntax.

Therefore, in addition to syntactic analysis, it is possible to consider a system that analyzes the entire semantic content of a sentence, but with the current technology, it is impossible to recognize the full meaning with 100% accuracy. Furthermore, even if the meaning is correctly recognized, there are often multiple words that express the meaning in the second word candy. In such cases, the choice of translation must ultimately be left to the user, but in a method where multiple translations are displayed side by side and the user is asked to select, the selection operation is required each time, which places a burden on the user. is large. One method is to preferentially use translated words that are statistically more frequently used, but there is a problem in that the frequency statistics in the text being translated do not necessarily match the general statistics.

The present invention provides a method for determining a translated word when it is impossible to determine a single translated word only from syntactic and semantic analysis, and reduces the proportion of corrections required by the user. purpose.

The present invention is directed to the case where a series of mutually connected sentences are to be translated in succession. -In a text consisting of 逼 sentences, the same word or phrase usually appears many times,
It is often used with one specific meaning in the context of the text.

Furthermore, in order to maintain the consistency of the translation result as a text, it is necessary to unify the translations for the same words and phrases that are used with the same meaning. For this reason, for words that cannot be narrowed down to one translation using fe or F'F based on syntactic and semantic analysis, if the word has also appeared in a previous sentence, the previously used translation is used. It is desirable to allocate it. In this way, it is possible to reduce the number of times the user changes the translation.

An embodiment of the present invention will be described below. FIG. 1 shows the configuration of a translation system according to the present invention, which includes a processing device 1, a text memory 2, a dictionary memory '3, a display device 4 such as a CRT display, and an input device 5 such as a keyboard. The processing device 1 controls the entire system and executes all translation processing. The text memory 2 consists of an input text area for storing the text in the first language to be translated, and an output text area for storing the text in the second language which is the translation result. The dictionary memory is a collection of records with the single @ in the first word as a heading, and the records show translation word 1 grammatical information and the latest used translations, which will be described later, for compound words including headwords and headwords. Information is stored.

The display device 4 is for displaying the translation results, and usually can display the input text and output text side by side.

The input device 5Fi is used to input commands for correcting the translation result Is. Of course, it can also be used as an input device when storing input text in the text memory 2.

FIG. 2 is a diagram showing the structure of records in the dictionary memory 3. As shown in FIG. In this example, the first language is Japanese.

The second language is English. Records are accessed using the heading as a key. The record has 11 headings.
In addition to grammatical information on headwords (quality test).

(inflected forms, digit structure patterns, etc.) 12 and translation information 13 are included. The translated word information field 13F'i is a repeated field equal to the number of translated words, and one field is a latest used translated word flag 13-1 indicating that the translated word is the most recently used translated word for the index. ．．
Translation 13-2. Grammatical information of the translated word (products, declensions, etc.) 1
It consists of 3-3. The latest used translated word flag 13-1 is turned on (1) only when it corresponds to the most recently used translated word, and is turned off (0) when it corresponds to other translated words. In the example in Figure 2, the translation for “handle” is [t
reat J, “deal with J.

``There are three translations of ``handle J'' in the dictionary, and among these, ``rdeal with J'' is the most recently used translation.

FIG. 3 is a flow diagram showing the translation process of the processing device 1. Translation processing is performed one sentence at a time.

1. In the first step (101), a sentence to be translated is read from the text file 2. Next, the second step (102
), a record is read from the dictionary memory 3 in which the sentence is divided into words and phrases, and all the candy in the n1 sentences are set as the headline candy. In the third step (103), referring to the grammatical rules of the first sentence and the grammatical information attached to the words in the sentence read out in the second step,
The sentence structure is analyzed. In the fourth step (1 "04"), the structure conversion rule from the first '@ candy to the second word candy is referred to, and the structure is converted into a structure that is possible as a sentence in the second language. In the fifth step (105), , gives an appropriate translation!FD for the first enemy word.The details of the 5th step are shown in Figure 4.At the 5th step, the translation is narrowed down to grammatically acceptable translations. However, improbable translations are discarded due to their semantic connectivity with other haku (10
51). As a result, if the translation is narrowed down to one translation, that translation is selected (1052). If two or more translated words remain as candidates, it is checked whether there is a translated word whose latest used translated word flag is O/ among the candidates, and if so, the translated word is selected (1053). For example, suppose you use the dictionary shown in Figure 2, and the candidate translation for "handling" is "dea".
If "l with J" and "handle J" remain, "deal with J" is selected. If there is no translation like the one above, the first registered word in the dictionary among the candidates is selected. (10
54). As a result of each of the above steps, the structure of the sentence and the translated word based on the second wording are determined.

In addition, the 4th. The fifth step is not executed serially in this order, but is executed while exchanging information 'tf with each other. In the sixth step (106), a sentence in the second language is generated by morphological synthesis or the like and displayed on the display device 4. Here, the user checks the displayed translation results. Step 7 (107) is a process based on user input, in which the translated text is corrected according to the command input from the input device 5.

When the correction is completed (if there is no correction, proceed immediately from step 6), the corrected translated text is output to the text file 2 in step 8 (108). Finally step 9 (1
09), the latest used translated word information is updated. That is, among the most recently used translated word flags for the words in the input sentence, only those corresponding to translated words used in the translated text (or the corrected text if there is a correction) are turned on, and the others are turned off. for example,
Referring to the dictionary in Figure 2, find the translation for ``handle'' [dea
If a user modifies a translated sentence using 'l with J' to 'i-handle J', [h
The latest used translation flag for andle J is turned on,
The latest translation flag 1 for "deal with J" will be turned off.
Translation of one sentence is completed in the first to ninth steps, and the process moves on to the translation of the next sentence.

As a modification of this embodiment, when considering the translation for #L new moon, if derived words are treated as a group, a more desirable result can be obtained in terms of consistency of translation. That is,
Suppose that the second term ``Tsume α'' is used as a translation for a phrase A in the first term.

Here, if the first phrase B is waiting for the derivative word β of α as a translation word, the latest used translated word information for B is updated as if β had been used. For example, "tr
If the translated word "treatment J" is used, the latest used monthly salary flags for "treat" in FIG.

In addition, in the example, the latest used translation word flag is set to ON for only one of the translation words for one entry word, but the translations for one entry word are divided into grammatically and semantically different groups. If possible, it is also possible to consider the latest usage for each group.

According to the present invention, even if syntactic/semantic analysis of a single sentence cannot narrow down to one translation phase, one translation word can be selected, and the selected translation word has a probability of being appropriate for the text currently being translated. It has a high value. Therefore, it is possible to reduce the frequency with which the user modifies the translation results, and the burden on the user can be significantly reduced.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the configuration of the translation system of the present invention, and FIG. 2 is a diagram showing the configuration of records in the dictionary memory in the method of the present invention.
FIG. 3 is a flowchart of translation processing, and FIG. 4 is a flowchart of translated word selection processing showing an embodiment of the method of the present invention. 1... Processing device, 2... Text memory, 3...
Dictionary memo 18 4...display device, 5...input device,
11... Heading candy, 12... Comparison candy grammar information, 1
3... Translation information, 13-1... Latest used translation word flag, 13-2... Translation word, 1st ward 2nd figure\3 figure

Claims (1)

[Claims] 1. A translation device comprising a means for translating a sentence in a first word candy into a corresponding sentence in a second word candy, displaying it on a display device, and correcting the displayed translated sentence. For words that have multiple translations, we need a dictionary that is structured so that it can store information that distinguishes the most recently used translation from other translations, and narrows it down to a single translation by syntactic and semantic analysis of the input sentence. To generate a translation using the latest used translation for words that cannot be used, and to update the dictionary so that when the translation of a certain word is changed by the user, the changed translation is used as the latest used translation of the word. A translation word selection method in machine translation featuring the following. 2. In the method described in paragraph 1, the 21st! The derivative words in candy are considered as a group, and when a certain translated word is used, the latest used translation information is also provided for the first word that has a derivative word of the translated word as a translation, as if the derived word had been used. A translation selection method in machine translation characterized by updating.