JPH01273171A - Document rewriting system and automatic translating system - Google Patents

Abstract

PURPOSE:To execute a translation processing after having converted a sentence to a sentence which can be translated easily and to shorten the translation time by providing a pre-processing before the translation processing. CONSTITUTION:A pre-processing part 1 executes processings such as a standardization of a notation, a long sentence division, a connecting word supplement, an indispensable case supplement with respect to an original language sentence, obviates or reduces ambiguity and incompleteness of the original language sentence, and converts it to a sentence which can be translated easily. Subsequently, a translation processing part 6 inputs the sentence which has been converted to the sentence which can be translated easily by obviating or reducing ambiguity and incompleteness in the pre-processing part 1, and translates it to an object language. In such a way, pre-editing by a person's help is scarcely required, and the processing time can be shortened as the whole translation.

Description

DETAILED DESCRIPTION OF THE INVENTION (1) Technical field to which the invention pertains The present invention relates to a natural language processing method, more specifically, a document rewriting method for rewriting input sentence data into plain sentence data.

The present invention relates to an automatic translation method that automatically translates the language of input sentence data into sentence data of a different language.

(2) Conventional technology It is generally difficult to automatically translate sentences in their original form. Therefore, in conventional translation methods, before starting the translation process, the sentences are made easier to translate by applying Brieddate. Users were required to convert.

However, in the case of the conventional translation method.

■ Dispersion of Briedet results.

(Preedit results vary depending on the user) ■　Incompleteness of preedit.

(Because the preedit is incomplete, some sentences remain that cannot be translated.) ■ Man-hours for the preedit.

(Pre-editing requires manpower and time) and other problems, making it an unsuitable translation method.

■, for ■.

(1) Human power for translation processing It is not possible to narrow down the linguistic phenomena that appear in sentences, and it becomes necessary to deal with all linguistic phenomena.

(2) Translation processing becomes complex and large-scale, making the process less predictable.

(3) As a result, the quality of the translated text deteriorates.

■For.

(4) When looking at the translation process as a whole, the processing time becomes longer.

There was a drawback of fil~(4).

(3) Purpose of the Invention An object of the present invention is to provide a document rewriting method and an automatic translation method that can output high-quality translations that solve the above-mentioned problems.

(4) Structure of the invention (4-1) Differences between features of the invention and conventional technology Regarding the language (original language) of input sentence data.

At least one of a notation standardized knowledge dictionary, a long sentence division knowledge dictionary, a connective word complementary knowledge dictionary, and a case element complementary knowledge dictionary is provided, and by referring to the notation standardized knowledge dictionary,
By obtaining the words that require notation standardization and the standard form of the word that requires notation standardization, and rewriting the word that requires notation standardization into the standard form, and by referring to the long sentence division knowledge dictionary, Obtain the number of characters in the sentence that defines the sentence, the cutting point at which the division is performed, and the connecting word that should be added after the division,
A process of dividing a long sentence, a process of complementing a connective word according to the logical relationship between two sentences by referring to the relevant connective word complementary knowledge dictionary, and an abbreviation process by referring to the relevant case element complementary knowledge dictionary. A first means (preprocessing) for performing at least one process of supplementing the case element in the preceding sentence or the following sentence, and a separate system using the output sentence data from the first means as input. The main feature is that there is a second means for translating and outputting sentence data in the language (target language).

Conventional technology is because most of the processing that conventionally required one-person pre-editing can be accomplished as pre-processing.

■ By processing the source language rewriting function, which was distributed among the translation processing functions, all at once during preprocessing, it is now possible to clearly separate the translation function into preprocessing function and translation function. become.

■ The quality and nature of input sentences to translation processing become uniform, and the language processing functions required for translation processing can be significantly reduced compared to conventional functions.

■ Because of this, the translation process can be realized compactly and ■ As a result, the quality of the translated text can be improved.

■ Since manual translation is almost unnecessary, the overall translation processing time is shortened.

Each point is different.

(4-2) Embodiment FIG. 1 is a basic configuration diagram of the present invention. ■ performs processing such as standardization of binary notation, long sentence division, connective word completion, and mandatory case completion on the input source language sentence, and eliminates or reduces ambiguity and incompleteness of the input source language sentence. , a preprocessing unit that converts the text into a text that is easy to translate; 2 is a knowledge dictionary search unit that searches a knowledge dictionary to obtain the necessary knowledge in the preprocessing unit l; and 3 is knowledge that systematizes the knowledge used in the preprocessing unit 1. Dictionary group 4 is a user dictionary registration unit that registers words determined to be unknown words or difficult-to-translate words in the preprocessing unit 1 into the user dictionary; 5 is a user dictionary registration unit that registers unknown words or difficult-to-translate words; A user dictionary, 6 is a translation processing unit that takes as input a sentence that has been converted into an easy-to-translate text with ambiguity and incompleteness resolved or reduced in the preprocessing unit 1, and translates it into the target language; 7 is a translation processing unit Section 6 is a translation dictionary search section that searches for a group of translation dictionaries and user dictionaries that have necessary information, and 8 is a source language analysis dictionary, a source language-target language conversion dictionary, a target language generation dictionary, etc. necessary for the translation processing section 6. 9 is an output unit that outputs the translation result of the translation processing unit 6 to a recording medium or a terminal, and 10 is an automatic translation device with natural language preprocessing. FIG. 2 is a schematic flowchart of the operation of the automatic translation device IO with natural language preprocessing.

Next, the operation will be explained according to FIGS. 1 and 2.

The preprocessing unit 1 executes a plurality of preprocessing items on the source language text that is input to the automatic translation device with natural language preprocessing 10 . n in FIG. 2 counts the number of preprocessing items. The preprocessing unit 1 requests the knowledge dictionary search unit 2 to transfer the knowledge necessary for each preprocessing item. The knowledge dictionary search unit 2 selects and searches for a knowledge dictionary storing knowledge necessary for the preprocessing item currently being processed from the knowledge dictionary group 3 (step 2), and transfers the information to the preprocessing unit 1. Using the transferred information, the preprocessing unit 1 completes the process for the preprocessing item currently being processed in step 3.
) Proceed to the first pre-processing item (that is, enter the process of step 6 shown in FIG. 2).

In addition, when a word that needs to be registered in the user dictionary occurs in the processing of each preprocessing item (step 4), a request is sent each time to the user dictionary registration unit 4, and the cough word is stored in the user dictionary 5. (Step 5).

When all the preprocessing items are completed in the preprocessing unit l, the preprocessed source language text is sent to the translation processing unit 6.

The translation processing unit 6 performs a process of translating the preprocessed source language text into a target language text (step 7). At this time, the translation processing unit 6 performs this process by using information obtained from the translation dictionary group 8 and the user dictionary 5 by the translation dictionary search unit 7, as necessary. In addition,
Methods for this translation processing include the Pipoft method and the transfer method, but these are not particularly limited here.

The target language sentence obtained by the translation processing section 6 is sent to the output section 9. The output unit 9 outputs the target language sentence.

Output light specified in advance by the user (recording media such as hard disks, displays, printers, etc.)
(Step 8).

Next, we will specifically discuss the case where the source language is Japanese using example sentences.

FIG. 3 shows an input Japanese sentence used as an example sentence. The input Japanese sentence is preprocessed by a preprocessing unit l, and
The preprocessed input Japanese text shown in FIG. 4 is sent to the translation processing section 6.

The input Japanese text in Figure 3 is processed in the preprocessing unit 1 using the four preprocessing items shown in Figure 4: ■ Standardization of notation, ■ Long sentence division, ■ Conjunction word completion, and ■ Case element completion. The input Japanese text is then converted into the input Japanese text that has been subjected to the preprocessing shown in FIG. 4, and has not been processed by other preprocessing items.

Next, the processing operations for each of the preprocessing items ① to ② will be described in detail.

■ Standardization of notation (replacing synonymous words and the same word with 9 different notations (e.g. color and color) with a single word notation that is determined as standard) Preprocessing unit 1
In , when the preprocessing item comes to "standardization of notation",
The preprocessing unit 1 requests the knowledge dictionary search unit 2 to transfer information regarding "standardization of notation." In the knowledge dictionary search section 2, “
Information regarding "standardization of notation" is obtained by searching the knowledge dictionary used for "standardization of notation" in the knowledge dictionary group 3, and is transferred to the preprocessing unit 1.

Figure 5 shows an example of the contents of a knowledge dictionary used for standardization of notation in a table format. 5-3 is the standard notation for the notation, and 5-4 is the part of speech of the word in the standard notation.

The preprocessing unit 1, which has received the information from the knowledge dictionary search unit 2, compares the word in the input sentence with the notation in the column 5-1 and the part of speech in the column 5-2, and identifies all words that match both. Rewrite it in standard notation 5-3. In the example shown in FIG. 3, the record indicated by the arrow in FIG. 5 corresponds to the record, and is rewritten as "Haneda Airport-Tokyo International Airport" (■ in FIG. 4).

■ Long sentence division (dividing a long sentence that is difficult to translate into multiple short sentences that are easy to translate, and assigning connections between them) When the preprocessing item comes to "Long sentence division" in the preprocessing section 1, The preprocessing unit 1 requests the knowledge dictionary search unit 2 to transfer information regarding “long sentence division”. In the knowledge dictionary search section 2,
Information regarding "long sentence division" is obtained by searching the knowledge dictionary used for "long sentence division" in the knowledge dictionary group 3, and is transferred to the preprocessing unit 1. FIG. 6 shows an example of the contents of a knowledge dictionary used for long sentence segmentation in a table format. In FIG. 6, 6-1 indicates the length of a sentence indicating the minimum number of characters in a sentence to be divided into long sentences.

Reference numeral 6-2 indicates a cutting point indicating at which position the long sentence is to be cut, based on the connection relationship between two simple sentences, and 6-3 indicates a conjunction after cutting, indicating a conjunction to be inserted after cutting.

The preprocessing unit 1 that received the transfer from the knowledge dictionary search unit 2 determines that the input sentence is longer than the number of characters indicated in the “sentence length” 6-1,
If the sentence is made up of two or more simple sentences, it is determined where to cut it and what conjunction should be used after cutting from the transferred information (see FIG. 6) and executes it. The example shown in FIG. 3 has a length of 49 characters and is made up of two or more simple sentences, so it is subject to long sentence division.

The record indicated by the arrow in Figure 6 corresponds to the record, and is cut off at the part where continuous use is discontinued (``... is out of order'' in Figure 3) (■ in Figure 4).
It should be noted that the fact that the column 6-3 of "connection after disconnection" in the record is blank indicates that the connective after disconnection is not uniquely determined at this point.

■ Connecting word completion (complementing the most appropriate connecting word between two short sentences based on the logical relationship between them) In the preprocessing unit 1,
When the preprocessing item is "connective word completion," the preprocessing section 1 requests the knowledge dictionary search section 2 to transfer information regarding "connective word completion." The knowledge dictionary search section 2 obtains information regarding "connective word completion" by searching the knowledge dictionary used for "connective word completion" in the knowledge dictionary group 3, and transfers this information to the preprocessing section 1. FIG. 7 shows an example of the contents of a knowledge dictionary used for connective word completion in a table format. In Figure 7, 7-1
7-2 shows the connection word that should be inserted when there is a relationship between the preceding sentence and the following sentence. is a conjunction to be inserted.

The preprocessing unit 1 that has received the transfer from the knowledge dictionary search unit 2 grasps the logical relationship between the sentence immediately before and the sentence immediately following the insertion position of the connective word (this method is not particularly limited), 7-I
Compare with IfA and insert the connective word of the matching record. In the example shown in Figure 3, the logical relationship between the immediately preceding sentence and the immediately following sentence is "cause-effect," so the record indicated by the arrow in Figure 7 corresponds, and "for this reason" is inserted as a conjunction. (
Figure 4 ■).

■ Case element completion (when there is no essential case in the sentence to be translated, complementing the word corresponding to the essential case from the preamble or post-sentence of the cough sentence) In the preprocessing section 1, if the preprocessing item is ``case element When the preprocessing section 1 reaches the point of "completion," the preprocessing section 1 requests the knowledge dictionary search section 2 to transfer information regarding "case element completion." The knowledge dictionary search section 2 obtains information regarding "case element completion" by searching the knowledge dictionary used for "case element completion" in the knowledge dictionary group 3, and transfers this information to the preprocessing section 1. Figure 8 shows an example of the contents of a knowledge dictionary used for case element completion in a table format. Completion from the previous sentence indicates which case in the previous sentence is to be completed by the word filling the required case when completing the required case from the previous sentence. 8-3 is the required case that is being completed. Completion from the subsequent sentence, which indicates which case in the subsequent sentence is to be completed by the word filling the subsequent sentence, 8-
4. When completing, the meaning category of the word to be completed is
This is a category match with a predicate pattern that shows whether it is necessary to be included in the semantic category of the case filled in by completion in the case structure of the predicate of the simple sentence being completed.

The preprocessing unit 1, which has received the transfer from the knowledge dictionary search unit 2, selects a record in which the essential case to be completed matches the case element to be completed in column 8-1, and completes it from the preamble, or
After determining whether to complete from the post-sentence (the method is not particularly limited), case element completion is performed according to the description of the record. In the example shown in Figure 3, it is the obligatory case and the object case that we are trying to complete.

The record indicated by the arrow in FIG. 8 is selected. The preprocessing unit 1 decides to complete from the preamble and attempts to complete the word "Tokyo International Airport" which is a locative case in the preamble according to the description of the record.The column 8-4 of the record is "necessary" Therefore, we check the semantic category of ``Tokyo International Airport'' and the semantic category of the object case of ``to use'', which is a simple phrase to be complemented. As a result.

Since "Tokyo International Airport" can be the object case of the term "use,""Tokyo International Airport" is complemented as the object case (
Figure 4 ■).

Above, four of the preprocessing items processed by the preprocessing unit 1 have been described in detail using examples.

The human-powered Japanese text that has been subjected to the preprocessing shown in FIG. 4 is sent to the translation processing unit 6, where it is converted into the target language.

Sent to the output section 9, 1. It is output from the output section 9.

In this example, we have described four preprocessing processes: standardization processing for binary notation, long sentence division processing, connective word completion processing, and case element completion processing. It is also possible to recognize the relationship between the bases that make up a compound word and perform paraphrasing (eg, document rewriting method → document rewriting method).

Due to this structure and operation, compared to conventional methods.

■ Conventionally, the source language rewriting function, which was distributed among each function of translation processing, is processed all at once during preprocessing, so that the translation function can be combined with the preprocessing function and Translation II! It becomes possible to clearly separate and configure the i-function and the i-function.

■ The quality and nature of input sentences to translation processing become uniform, and the language processing functions required for translation processing can be significantly reduced compared to conventional functions.

■ Because of ■, translation processing can be realized compactly.

■ As a result, the quality of the translated text can be improved.

■ Since manual pre-editing is almost unnecessary, the overall translation processing time is shortened.

There were improvements in each point.

(5) As described in detail, according to the present invention, by providing pre-processing before translation processing, the number of brieddates performed by two people can be minimized, and further.

Translation processing can be performed after eliminating or reducing the ambiguity and incompleteness of human source language sentences and converting them into sentences that are easy to translate, which has the advantage of shortening translation time and outputting high-quality translated sentences.

Also, what is the document rewriting method consisting only of the first means?
j! Since it is possible to rewrite unwritten sentence data into sentence data that eliminates or reduces incompleteness and ambiguity, this method can be used as preprocessing for systems that require analysis of sentence data (such as database search systems). , it is possible to narrow down the linguistic phenomena to be analyzed by the system,
The design of the analysis method can be facilitated.

[Brief explanation of the drawing]

Figure 1 is a basic configuration diagram showing the configuration of an embodiment of the present invention.9 Figure 2 is a schematic flow of the operation of the present invention.Figure 3 is an input Japanese sentence used as an example sentence.Figure 4 is the figure 3. Figure 5 shows the preprocessed input Japanese text after the input Japanese text passes through the preprocessing section. Figure 5 shows the knowledge used to standardize the notation used in "standardization of notation," which is one of the preprocessing items. Dictionary content example, Part 6
The figure shows an example of the contents of a knowledge dictionary used for long sentence segmentation, which is one of the preprocessing items, and Figure 7 shows one of the preprocessing items.
Figure 8 is an example of the contents of a knowledge dictionary used for case element completion, which is used for ``case element completion'', which is one of the preprocessing items. It is. 1... Preprocessing section, 2... Knowledge dictionary search section, 3...
Knowledge dictionary group, 4... User dictionary registration section, 5... User dictionary. 6... Translation processing unit, 7... Translation dictionary search unit, 8...
- Translation dictionary group, 9...output section. Patent applicant Nippon Telegraph and Telephone Corporation

Claims (2)

[Claims] (1) For the source language of input sentence data, at least one of a notation standardized knowledge dictionary, a long sentence division knowledge dictionary, a connective word complementary knowledge dictionary, and a case element complementary knowledge dictionary is provided, and the notation standardized knowledge dictionary is provided. A process of obtaining a word that requires notation standardization and a standard form of the word that requires notation standardization, and rewriting the word that requires notation standardization into the standard form, and the long sentence division knowledge dictionary. By referring to , obtain the number of characters in the sentence that defines the long sentence, the cutting point for dividing, and the connective word to be added after dividing, and refer to the process of dividing the long sentence and the connective word supplementary knowledge dictionary. The process of complementing connective words according to the logical relationship between two sentences by , a document rewriting method characterized by executing at least one process. (2) For the source language of the input sentence data, at least one of a notation standardized knowledge dictionary, a long sentence division knowledge dictionary, a connective word complementary knowledge dictionary, and a case element complementary knowledge dictionary is provided, and the notation standardized knowledge dictionary is provided. A process of obtaining a word that requires notation standardization and a standard form of the word that requires notation standardization, and rewriting the word that requires notation standardization into the standard form, and the long sentence division knowledge dictionary. By referring to , obtain the number of characters in the sentence that defines the long sentence, the cutting point for dividing, and the connective word to be added after dividing, and refer to the process of dividing the long sentence and the connective word supplementary knowledge dictionary. The process of complementing connective words according to the logical relationship between two sentences by , a first means comprising a document rewriting method characterized by executing at least one process; and inputting output sentence data from the first means, translating it into sentence data of a target language of a different system, An automatic translation method comprising: a second means for outputting.