JPH0673134B2 - Machine translation system - Google Patents

Description

TECHNICAL FIELD The present invention relates to a machine translation system for translating a source language sentence into a sentence of another language, and in particular, ambiguity occurs as a source language (hereinafter, referred to as “Japanese”). Or, the present invention relates to a machine translation system provided with a means for extracting sentences that are totally strange as Japanese (hereinafter, these are collectively referred to as "non-sentences").

2. Description of the Related Art Conventionally, in a machine translation system, an input Japanese sentence is first subjected to morphological analysis, and then subjected to syntactic analysis and the like, and the syntactic analyzed sentence is converted into another language (for example, English,
It will be translated into French). However, all the input Japanese sentences are not necessarily correct Japanese, so it was checked whether they were non-sentences during the above-mentioned parsing process. Alternatively, a method is adopted in which non-sentence checking is not performed and the input Japanese sentence is directly parsed and translated.

A way to check for non-sentences during parsing like this is
There was a problem that unnecessary processing was performed until a non-sentence was found. In addition, when parsing is performed using a grammar that finds a non-sentence, there is a problem that the parsing becomes complicated and the processing speed is reduced.

On the other hand, if the syntactic analysis is performed without checking the non-sentence and the translation result is output, the translation result may be completely incorrect.

In this way, if a translation system is constructed using a grammar that checks non-sentences during parsing, the cost of the system will increase, and even if parsing takes a long time, the result may end up not being obtained. .

Aim The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a non-sentence determination extraction unit capable of efficiently extracting the above-mentioned "non-sentence" to improve the speed of syntactic analysis. It is to provide a machine translation system capable of reducing the capacity of a grammar dictionary for parsing.

Configuration The above object of the present invention is to provide a source language sentence input receiving unit of a machine translation system for translating a source language sentence into a sentence of another language,
A plurality of conditions for detecting an inappropriate sentence (non-sentence) as an input source language sentence to the machine translation system are defined in advance, and a plurality of statistics for the number of cases corresponding to the plurality of conditions of the input source language sentence are collected. And a non-sentence determination and extraction unit that extracts the non-sentence by comparing the results calculated by the plurality of statistical units with the condition to determine whether or not the sentence is a non-sentence. Achieved by a featured machine translation system.

Embodiment Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. FIG. 2 is a schematic configuration diagram of a machine translation system according to an embodiment of the present invention.

In FIG. 2, reference numeral 1 is an input device for inputting a natural language sentence, for example, a character input device such as a keyboard or OCR. Reference numeral 2 is a morphological analyzer that divides an input sentence into words using a Japanese dictionary and connection table and adds information such as part of speech. Reference numeral 3 denotes a non-sentence extraction device, which is a characteristic component of the present embodiment, and is grammatically and usage-wise using morphological statistics (a large number of characters, a large number of phrases, etc.), a parameter list 10 and a parameter list 11. It is judged whether or not it has a feature of non-sentence, and if it is not a sentence, it is displayed to the user, and if it is not a sentence, it is input to the translation device 4 and translated. Reference numeral 4 denotes a translation device, which uses a translation grammar and translation dictionary to parse, convert, and generate an input sentence, and translate it from Japanese into another language (for example, English or French).

An output device 5 outputs a non-sentence extracted by the non-sentence extraction device 3 or a foreign language translated by the translation device 4, such as a CRT display or a printer.

FIG. 1 is a block diagram of the non-sentence extracting device 3 in the embodiment shown in FIG.

In FIG. 1, reference numeral 6 denotes a non-sentence determination device that determines a non-sentence using a statistical result of the determination table 12 and a threshold value. If one of the statistical devices determines that the sentence is non-sentence, the process is interrupted and displayed to the user. At this time, the reason why the sentence is determined to be non-sentence is also displayed. 7 is a statistical device that counts the number of characters and phrases in the input sentence and fills in the judgment table 12, 8 is a parameter list
A statistical device that uses 10 to obtain the statistics in the sentence of each parameter, 9 is a statistical device that obtains the statistics from the sentence of each parameter using the parameter list 11, and 10 is a key word or attribute for determining a non-sentence ， Parameter list in which the attribute value is entered, 11 is the parameter list in which the relationship of the key words for determining the non-sentence is entered, 12 is the processing name, parameter name, statistical result, threshold item In the statistical result, the count number of the number of parameters in each process is entered.

FIG. 3 is a diagram showing a configuration example of a parameter list 10 in FIGS. 3 and 2. The parameter list 10 is composed of three sets of word attribute names, attribute values, and notations. For example, the hiragana “no” constitutes a parameter list from the attribute name “part of speech”, the attribute value “case particle”, and the notation “no”.

FIG. 4 is a diagram showing a configuration example of the parameter list 11 of FIG. This parameter list 11 is used to indicate the context of words and phrases. First, the list is divided into two parts, the front part and the rear part. The front part consists of attribute names, attribute values, and notations, and the latter part consists of attributes. It consists of name, attribute value, notation, and evaluation value.

FIG. 5 is a diagram showing a configuration example of the judgment table 12 according to the present invention. This judgment table 12 is composed of a processing name, a parameter name, a statistical result, and a threshold value.
Shows the processing by, and in the parameter name, for example,
Enter the number of characters in the sentence, the number of characters / the number of phrases, and decide the threshold value in advance. Here, the threshold value of the number of characters is 40 and the number of phrases / the number of characters is 0.6. However, it is not limited to this value. Further, as the statistical result, since the result obtained by taking statistics by the statistical device 7 is entered with respect to the sentence actually input, nothing is usually entered. Statistical processing 2 indicates statistical processing by the statistical device 8,
In the parameter name, for example, the number of "no", the number of continuous cancellations, and the number of parallel particles are entered, and the threshold values are "3", "3", and "2", respectively. This numerical value is not limited as in the statistical processing 1 described above. The statistical processing 3 shows the statistical processing by the statistical device 3, in which the number of "..., ~" is entered in the parameter name, and the threshold value is "1". Therefore, when the non-sentence determination device 6 looks at the determination table 12 as described above and the statistical result is equal to or more than the threshold value, the input sentence is regarded as a non-sentence.

FIG. 6 is a flowchart showing the non-sentence extraction processing according to the present invention. This shows the operation of the non-sentence extraction device of FIG. Hereinafter, description will be given according to the flowchart of FIG.

First, the number of characters in the sentence morphologically analyzed by the morphological analyzer 2 is counted and entered in the judgment table 12 (step 60
1), the number of clauses is counted and entered in the judgment table 12 (step 602), the statistical result of the judgment table 12 and the threshold value are compared, and the non-sentence judgment device 6 judges the non-sentence (step 60).
3). When the non-sentence determining device 6 determines that the sentence is not a sentence (step 604), the output device 5 shown in FIG. When the non-sentence determination device 6 determines that the sentence is not a non-sentence (step 604), the statistical device 8 counts each parameter in the input sentence for the parameter list 10 and writes it in the determination table 12 (step 60).
5) The non-sentence determination device 6 compares the statistical result of the determination table 12 with the threshold value to determine the non-sentence (step 606). If it is determined as a non-sentence (step 607), the output device 5
The non-sentence is displayed by (step 611). When the sentence is not a non-sentence (step 607), each parameter of the parameter list 11 is counted in the input sentence by the statistical device 9 and entered in the decision table 12 (step 608), and the statistical result of the decision table is obtained by the non-sentence determination device 6. And the threshold value are compared to determine non-sentence (step 609). If it is determined that the sentence is not a sentence (step 610), the output device 5 displays it by the user (step 611). If it is determined that the sentence is not a non-sentence (step 610), the process ends.

FIG. 7 is a diagram showing an example of an input sentence. It can be readily seen that the "book on his desk bookshelf" as shown in this example has four clauses, but parsing this sentence causes ambiguity. Ie his desk, desk bookshelf, bookshelf books,
His bookshelf, desk bookshelf, bookshelf books, his books, desk bookshelves, bookshelves books, his desk, desk books, bookshelves books can be considered. Therefore, it is possible to prevent processing deterioration by extracting such an input sentence as a non-sentence and not performing syntax analysis or the like. Hereinafter, the non-sentence extraction processing until the "book on the bookshelf of his desk" is regarded as a non-sentence by the book non-sentence extraction device 3 will be described. First, using the input device 1, "book on the bookshelf of his desk"
When is input, the morphological analysis device 2 next performs morphological analysis. The morphologically analyzed sentence “book on the bookshelf at his desk” is input to the non-sentence extraction device 3, and whether or not it is a non-sentence is checked. Hereinafter, the non-sentence extraction device 3 shown in FIG.
The non-sentence extraction process will be described with reference to FIG.

First, the morphologically analyzed sentence "book on the bookshelf of his desk" is statistically numerically 7 and the statistics of the number of characters, the number of phrases, and the number of phrases / number of characters are taken. Then, the result is entered in the judgment table 12. The statistical results are: number of characters = 8, number of phrases = 4, number of phrases / number of characters = 0.5
Becomes An example of this judgment table 12 is shown in FIG. The non-sentence determination device 6 looks at the determination table 12 as shown in FIG. 8 and determines whether or not the statistical result is larger than the threshold value. However, in this example, since both are smaller than the threshold value, the result of statistic 1 is not a non-sentence. The process proceeds to the statistical device 8. The statistical device 8 uses the parameter list 10 to search for "books on the bookshelf of his desk" using the parameters, counts them, and fills in the judgment table 12. The statistical result by this statistical device 8
Shown in the figure. Now, pay attention to a parameter whose part of speech is a case particle and whose notation is "no" in the parameters of the parameter list 10 (for the sake of explanation, other parameters are ignored). Then,
In the input sentence "book on his desk bookshelf", the case particle "no" is 3
One is included. Enter this in Judgment Table 12. Therefore,
In the non-sentence determination device 6, when the statistical result of the case particle “no” and its threshold value are compared, the statistical result is equal to the threshold value, so “the book on the bookshelf of his desk” can be a non-sentence. There is a
The processing is interrupted here, and the output device 5 shown in FIG. 2 displays to the user that "books on the bookshelf of his desk" is non-sentence, and at the same time displays the factors determined to be non-sentence. .

FIG. 10 is a block diagram of a non-sentence extraction device 3 showing another embodiment. Here, the configuration is the same as that of FIG. 1 except for the non-sentence determination device 6, so the description is omitted and the non-sentence determination device 6 will be described.

The non-sentence determination device 6 once moves the process to the statistical device 7, and then determines the determination table 12 until the statistical device 9 sends a signal indicating the end of processing.
I can't go see. That is, when the processing is completed in the order of statistical device 7 → statistical device 8 → statistical device 9, the decision table 12 is looked at, and non-sentence is determined.
In step 12, all statistical results that are equal to or greater than the threshold value are extracted, and all the factors of the non-sentence are displayed on the output device 5 to the user.

FIG. 11 is a flowchart of a non-sentence extraction process showing another embodiment. This shows the processing of non-sentence extraction by the non-sentence extraction device 3 in FIG. Hereinafter, description will be given according to the flowchart of FIG.

First, the number of characters in the input sentence is counted by the statistical device 7,
Fill in the judgment table 12 (step 1101), count the number of phrases, and fill in the judgment table 12 (step 1102). Next, each parameter of the parameter list 10 is counted by the statistical device 8 in the input sentence and entered in the judgment table 12 (step 1103). Furthermore, each parameter of the parameter list 11 is counted by the statistical device 9 in the input sentence and entered in the judgment table 12 (step 1104). The statistical results of the judgment table 12 entered by the above processing are compared with the threshold value to judge non-sentence (steps 1105, 1106). If it is determined to be non-sentence, the data is transferred to the output device 5 to display that it is a non-sentence (step 1107). When it is not a non-sentence, the translation device 4 performs syntax analysis, conversion, generation, etc., and outputs another language to the output device 5.

Here, the point that the other embodiment is different from this embodiment is that a plurality of factors determined to be non-sentences are displayed to the user. Since all the items of the statistical processing 1, 2, and 3 are filled in the judgment table 12, the non-sentence judging device 6 looks at this and extracts the factors of the non-sentence (the statistical result is the threshold value). All the above parameters are sent to the output device 5 for display.

As described above, in this embodiment, the input sentence subjected to morphological analysis is compared with a predetermined non-sentence condition to determine whether or not the sentence is a non-sentence, and then the syntax is determined. Since parsing is performed, the capacity of the grammar dictionary and the like required for parsing can be reduced. Further, since this non-sentence extraction can be mechanically processed to some extent, the processing speed can be improved. Further, since unnecessary parsing and the like need not be processed, the parsing processing speed can be improved. Further, in the present embodiment, since the input sentence is translated except for the non-sentence, the quality of the translated sentence can be finally improved.

Effect As described above, according to the present invention, in natural language processing such as a machine translation system, it is possible to extract a non-sentence from an input sentence with a simple hardware configuration before performing a parsing process. In addition, the processing speed of parsing can be improved, and the capacity of the grammar dictionary for parsing can be saved.

[Brief description of drawings]

FIG. 1 is a block diagram of a non-sentence extraction device showing an embodiment of the present invention, FIG. 2 is a block diagram of a machine translation system to which the present invention is applied, and FIGS. 3 and 4 are parameter lists according to the present invention. FIG. 5 is a diagram showing a configuration example, FIG. 5 is a diagram showing a configuration example of a judgment table according to the present invention, FIG. 6 is a flow chart showing non-sentence extraction processing by the non-sentence extraction device of FIG. 1, and FIG. FIG. 8 is a diagram showing an example, FIG. 8 is a diagram showing a configuration example of a judgment table showing the statistical results of the input sentence of FIG. 7, and FIG. 10 is a non-sentence extracting device according to another embodiment of the present invention. 11 is a flow chart showing a non-sentence extraction process by the non-sentence extraction device of FIG. 1: input device, 2: morphological analysis device, 3: non-sentence extraction device, 4: translation device, 5: output device, 6: non-sentence determination device, 7, 8, 9: statistical device, 11, 12: parameter list .

Claims (2)

[Claims] 1. A source language sentence input receiving unit of a machine translation system for translating a source language sentence into a sentence of another language, in advance, an inappropriate sentence (non-sentence) as an input source language sentence to the machine translation system. A plurality of conditions to be detected are set in advance, and a plurality of statistical means for taking statistics corresponding to the plurality of conditions of the input source language sentence are compared with the results calculated by the plurality of statistical means and the conditions. A machine translation system comprising: a non-sentence determination extraction unit that determines whether or not a sentence is a non-sentence and extracts a non-sentence. 2. The statistical means is characterized by the number of characters in the source language sentence,
The machine translation system according to claim 1, wherein the number of phrases or the number of words or phrases having a specific grammatical attribute is counted, and these upper limit values are set as a determination condition.