JPH0232467A - Machine translation method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a machine translation method, and particularly to a machine translation method suitable for efficiently translating identical sentences and sentences of the same type.

[Conventional technology]

In conventional machine translation processing, data such as pre-registered dictionary information and grammatical rules are used to perform syntax analysis, translation generation, etc. on each input sentence to generate a translated sentence (
JP-A-58-4068.4, JP-A No. 59-12157
(Refer to Publication No. 4) 6 In general, the more limited the field of the text is, the more the same or isomorphic sentences are included in the input text. There is no function to check for duplicate text sentences. Therefore, even if a same-sentence or a same-type sentence appears, a wasteful process of translating it again using data such as grammatical rules is performed.

[The invention tries to solve i! KM) In the above conventional technology, even when a sentence that is the same as a sentence that has already been translated or a sentence that is isomorphic appears again in the input text, translation processing using data such as dictionary information and grammatical rules is not performed. This process has to be repeated, which results in wasted processing time.
There was a problem that the processing efficiency of the entire translation process decreased.The purpose of the present invention is to solve this conventional problem and prevent duplicate execution of translation processes for the same sentence or isomorphic sentence in an input text sentence. An object of the present invention is to provide a machine translation method that reduces the CPU time required for the entire translation process for the same text and improves the processing efficiency of translation.

[Means to solve the problem]

In order to solve the above problems, the machine translation method of the present invention is a machine translation method that performs translation processing using data such as word information and grammatical rules registered in advance in a dictionary. A means for patterning an input sentence by regarding the "words consisting of numerical values" as variable words and replacing them with special symbols, a means for temporarily storing the variable words in memory, and using an already registered typical pattern file. means for comparing and matching the pattern of the input sentence and the typical pattern; means for extracting a translation pattern corresponding to the matched typical pattern from the typical pattern file; and a memory for storing special symbols in the extracted translation pattern. The present invention is characterized in that it includes means for generating a translated sentence corresponding to an input sentence by replacing the variable word with the variable word in the input sentence.

Further, means for displaying the pattern of the input sentence on a display device whenever the patterned input sentence does not match the already registered typical pattern, and a translated sentence pattern corresponding to the pattern of the input sentence. The present invention is characterized by the addition of means for inputting from an input device, and means for registering a new pattern in the typical pattern file in association with the pattern of the input sentence and the translated sentence pattern.

Further, when the patterned input sentence does not match the already registered typical pattern, the input sentence pattern is temporarily stored in memory and later displayed all at once on the display device. , means for inputting a translation pattern corresponding to the input sentence pattern from an input device;
Another feature is that a means for newly registering the above-mentioned input sentence pattern group and translation sentence pattern group in the above-mentioned typical pattern file is added. A method that focuses on the notation of uppercase/lowercase letters of words, treats words starting with a capital letter other than the first word in the input sentence as variable words and replaces them with special symbols, and stores the variable words in memory, or By setting attribute conditions for the word extracted from , and checking the attributes of the word by looking it up in a dictionary, the words that match the conditions are regarded as variable words and replaced with special symbols, and the variable words are stored in memory. Furthermore, by focusing on the combinations of words in the input sentence, the words located either before or after a specified word are regarded as variable words and replaced with special symbols, Another feature is that it has means for storing the variable words in a memory.

As a means of comparison and matching between a typical pattern and an input sentence pattern, we focus on the first letter of the first word of the patterned input sentence, and use a typical pattern file that has already been sorted into alphabets and registered. A means of comparing the first character of each pattern in the file with the first character of the above input sentence, and focusing on the specified position word in the patterned input sentence, and using the already specified position word as a keyword. Another feature is that it has means for comparing the word with a keyword in the file using a registered typical pattern file.

As a translation generation means, when replacing a special symbol in a translation pattern extracted corresponding to the input sentence pattern with a variable word in memory, the variable word is looked up in a dictionary and the translation of the variable word is selected. , a means for generating a translated sentence corresponding to the input sentence by replacing the above special symbol with the corresponding translation word, and a means for generating a translated sentence corresponding to the input sentence by replacing the special symbol in the translated sentence pattern extracted corresponding to the above input sentence pattern with the corresponding variable word in memory. Another feature is that it has means for generating a translated word corresponding to an input sentence by replacing it with a translated word.

[Effect]

According to the present invention, in machine translation processing, redundant execution of translation processing using data such as dictionary information and grammatical rules is prevented for the same sentences or isomorphic sentences appearing in input text,
It is possible to reduce the CPU time required for the entire translation process of the same text, and improve the translation processing efficiency.

〔Example〕

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

In this embodiment, a machine translation processing method from English to Japanese will be described.

FIG. 1 is an overall configuration diagram of a machine translation system showing an embodiment of the present invention.

In FIG. 1, 1 is a processor, 2 is a typical pattern file storing typical patterns, 3 is a memory, 4 is an English-Japanese dictionary file, 5 is an English text file, 6 is a display device, and 7 is an input device such as a keyboard. 8 is an English-Japanese grammar rule file; 11 is an input sentence patterning processing unit that regards "words consisting of numerical values" in an input text sentence as variable words and replaces them with special symbols to pattern the input sentence; 1
2 is a comparison processing unit that performs comparison and matching between the input sentence pattern and the typical pattern using the typical pattern file 2;
Reference numeral 13 denotes a translation generation processing unit that replaces special symbols in the translation pattern with variable words in memory to generate a translation corresponding to the input sentence, and 14 a typical pattern registration processing unit that registers typical patterns.

The above-mentioned files are stored on a medium such as a magnetic disk, a floppy disk, or an optical disk.

FIG. 2 is a flowchart of machine translation processing in this embodiment. This is the flow of translation processing in the configuration of FIG. The operation of this embodiment will be explained below according to the flow shown in FIG.

First, the English text 1- to be translated is imported into the memory 3 from the English text file 5 (step 101).
.

In the input sentence patterning processing unit 11 of the processor 1,
Extract the input text sentence by sentence from the English text, extract the words written only with numerical values in the input text sentence, and replace them with special symbols in sequence (Step 1).
02). - Each time multiple words appear in a text sentence.

Change the number of the special symbol and replace it with the word. - If no words are matched, leave the input text as is.

The replaced original variable word is temporarily stored in the memory 3 in association with the input text. In this embodiment, the extracted word is replaced with a special symbol "$$". FIG. 3 shows an example of an input text sentence before the word is replaced with a special symbol, and an example of an input text sentence after the word is replaced.

Next, the comparison processing unit 12 with the typical pattern of the processor 1
In step 103, the replaced English text is compared and matched with a typical pattern registered in advance in the typical pattern file 2 for each one-person catechist sentence. The typical pattern file 2 can be accessed by focusing on the position of the special symbol in the input text sentence and checking in which word in the input text sentence the special symbol is located. After narrowing down the typical patterns, a method is used to access the typical patterns. As a result, the target typical pattern can be accessed more quickly than when all typical patterns are accessed.

FIG. 4 shows typical pattern file 2. and an example of the input text sentence. Here, focusing on the fact that the special symbol is the fourth word in the input text sentence, we narrow down the range to be accessed in the typical pattern file 2, and
A typical pattern that matches the input text sentence is detected.

If a typical pattern that matches the pattern of the input text sentence is detected in the typical pattern file 2, the translation generation processing unit 13 of the processor 1 generates the typical pattern in the typical pattern file 2 as shown in FIG. Extract the translated word pattern that corresponds to the pattern, and use the special symbols in the translated word pattern as
The special symbol in the input text sentence stored in the memory 3 is replaced with the variable word before being replaced (step 104
).

FIG. 6 is a transition diagram of the -human text sentence according to this embodiment. Here, (a) is the input text sentence before replacing with special symbols, (b) is the input text sentence after replacement, (C)
is a typical pattern, (d
) is a translation pattern corresponding to the typical pattern, and (,) is a translation created by replacing special symbols in the translation pattern with the original variable words. The underlined part indicates the replaced part.

In addition, if the typical pattern file 2 is accessed and there is no match with the input text sentence, the dictionary information registered in the English-Japanese dictionary file 4 is added to the input text sentence before replacing it with the special word. , and the data of grammar rules registered in the English-Japanese grammar rule file 8 are used to perform conventional translation processing such as syntactic analysis, intermediate language expression, and translation generation to generate a translation (step 105).

In this embodiment, conventional translation processing is performed on the input text sentences that are not matched with the typical pattern in the typical pattern file 2, but as shown in the flowchart of FIG. When the typical pattern of the sentence and the typical pattern file 2 are not matched, the typical pattern is displayed on the display device 6 (step 201), the corresponding translated word pattern is inputted from the input device 7 such as a keyboard (step 202), and the typical pattern file 2 is displayed. Another method can be considered in which a process of updating the typical pattern file 2 at any time is added to the present embodiment by registering it together with the pattern of the input text sentence after replacement in the pattern file 2 (step 203) (step 106).

In addition, as an expansion method of the step 106 method, patterns of input text sentences after replacement that are not matched can be temporarily stored in the memory 3, and the corresponding translation patterns can be input all at once later. A method of adding Wta to pattern file 2 is also considered.

Although the above is one example, various other expansion methods are possible.

Regarding the variable word replacement method, in addition to the numerical variable replacement method explained in the above embodiment, a method that replaces variable words such as proper nouns by paying attention to the uppercase/lowercase letters of the words in the input text sentence can be considered. . FIG. 8 is a flowchart of the replacement method.

That is, the input text sentence is divided into words, and words starting with a capital letter excluding the first word of the sentence are extracted from the word group (step 301), and the words are regarded as variable words and replaced with special symbols ( Step 302) (Step 102').

In addition, in the above step 102' method, words starting with a capital letter are simply extracted, but when further extracting words with detailed conditions such as "place name word", it is necessary to check the attribute information of the word. . Therefore, step 1 above
A method that further incorporates dictionary lookup into the method of 02' can be considered. FIG. 9 is a flowchart of the replacement method.

That is, the processor 1 performs a dictionary lookup in the English-Japanese dictionary file 4 for the word extracted in step 102' (step 401), checks that the word has attribute information of a place name, and then performs a dictionary lookup in the English-Japanese dictionary file 4 (step 402). ), the word is regarded as a variable word and replaced with a special symbol (step 403) (step 102'').

Furthermore, another method may be considered in which variable words are extracted and replaced based on a combination of words in the input text sentence. 1st
In the case of an English text sentence related to weather reporting as shown in Figure 0, the word between rth eJ and rAreaJ is inevitably determined to be a place name, so the input text sentence and "
If "the" and rAreaJ match, the words between them can be regarded as variable words and replaced with special symbols.

FIG. 11 is a flowchart of the replacement method.

That is, the input text sentence is divided into words, and words that have a fixed positional relationship with the specified word are extracted (step 5).
01), the word is regarded as a variable word and replaced with a special symbol (step 502) (step 102'').

In step 104, regarding the method of replacing special symbols in the translation pattern with variable words in the original input text sentence, in this embodiment, the corresponding words are replaced as they are, but the dictionary lookup in the English-Japanese dictionary file 4 is used. One possible method is to replace the translated words of the variable words. FIG. 12 is a flowchart of the replacement method.

That is, as shown in FIG. 12(a), in addition to the above step 102# method, after replacing variable words in the text sentence with special symbols, a step is performed to select the translation of the corresponding word when looking up the dictionary. 601).

There is a method (step 602) in which the word is stored in the memory 3 in correspondence with the variable word (step 104'), and a method (step 104') in FIG.
As shown in b), there are two possible methods: one is to look up the dictionary again when replacing the special symbol in the translated word pattern (step 701), and the other is to select a translated word for the variable word (step 702) (step 104'').

In addition, in this embodiment, the access method for the typical pattern file 2 is to narrow down the range to be accessed by focusing on the position of the special symbol and search for a typical pattern that matches more quickly. The following method is also possible.

In addition to the access method in step 103, focus on the first character of the first word of the input text sentence, use this character as a key to match with the typical pattern file 2 sorted alphabetically, narrow down the typical pattern file 2, and then , a method of accessing typical patterns can be considered.

Further, by focusing on a word (or words) at a pre-specified position in the input text sentence and performing matching with the typical pattern file 2 using the word as a keyword, the typical pattern file 2 is narrowed down, and then One possible method is to access typical patterns.

In FIG. 13, focusing on the word at the second position and the word at the fifth position in the input text sentence, a typical pattern file 2 is created that has the second word and the fifth word as keywords in advance. The typical pattern file 2 is narrowed down by matching. However, the greater the number of words used as keywords at this time, the greater the effect of narrowing down the typical pattern file 2, but on the other hand, it takes time to match the keywords, so set an appropriate number depending on the length of the input text sentence. That is important.

Note that the methods shown in each of the above embodiments do not exclude the currently used machine translation methods. In this example, the English-Japanese machine translation method was described, but by fully considering the application field etc., Japanese-English machine translation and
It is also applicable to machine translation methods for other languages. In addition, in this embodiment, a machine translation method has been explained, but step 10
It is also possible to introduce the processes from 1 to 1o4 into a conventional machine translation method as a tool. Further, in this embodiment, although the application field is text related to weather reporting, this method is also effective for translation processing of other standard texts such as manual texts and economic texts.

〔Effect of the invention〕

As explained above, according to the present invention, in order to avoid performing conventional translation processing using data such as dictionary information and grammatical rules for the same sentence or isomorphic sentence that repeatedly appears in the input text sentence, Translation processing for the entire text can be significantly sped up. Furthermore, by sequentially updating the typical pattern files, the translation time can be reduced as more texts are translated. Furthermore, the frequency of appearance of identical sentences and isomorphic sentences increases as the field of the target sentence is limited, so it can be especially effective in speeding up the translation process of input sentences in a limited field. .

[Brief explanation of the drawing]

Fig. 1 is an overall configuration diagram of a machine translation method showing an embodiment of the present invention, Fig. 2 is a flow chart showing the processing flow of an embodiment of the present invention, and Fig. 3 is a variable of an input text sentence in the present invention. FIG. 4 is a diagram showing a part of a typical pattern file in the present invention and an example of an input text sentence; FIG. 5 is a diagram showing an example of the internal structure of a typical pattern file in the present invention. 6 is a transition diagram of the process from an input text sentence to obtaining a translated sentence according to an embodiment of the present invention, FIG. 7, FIG. 8, FIG. 9,
11 and 12 are flowcharts showing processing operations according to another embodiment of the present invention, and FIG. 10 is a diagram showing an example of a method for replacing variable words in an input text sentence in another embodiment of the present invention. , FIG. 13 is a diagram showing an example of a typical pattern file access method according to the present invention. 1: Processor, 2: Typical pattern file, 3: Memory, 4: English-Japanese dictionary file, 5: English text file, 6: Display device, 7: Input device such as keyboard, 8: English-Japanese grammar rule file, 11: Input sentence patterning processing unit, 12: Typical pattern comparison processing unit, 13: Translation generation processing unit. 14: Typical pattern registration processing section. Fig. Cow Fig. Fig. ○

Claims (1)

[Claims] 1. In a machine translation method that performs translation processing using data such as word information and grammatical rules registered in advance in a dictionary, "words consisting of numerical values" in an input text sentence to be translated means for patterning the input sentence by regarding it as a variable word and replacing it with a special symbol; means for temporarily storing the variable word in memory; and means for patterning the input sentence using an already registered typical pattern file. means for comparing and matching with a typical pattern; means for extracting a translation pattern corresponding to the matched typical pattern from the typical pattern file; and replacing special symbols in the extracted translation pattern with the variable word in memory. A machine translation method comprising means for generating a translated sentence corresponding to an input sentence. 2. means for displaying the pattern of the input sentence on a display device whenever the patterned input sentence does not match the already registered typical pattern; and a translated sentence pattern corresponding to the pattern of the input sentence. Claim 1 further includes means for inputting from an input device, and means for newly registering the pattern of the input sentence and the translated sentence pattern in the typical pattern file in correspondence with each other. machine translation method. 3. means for temporarily storing the pattern of the input sentence in memory and displaying it all at once on a display device when the patterned input sentence does not match the already registered typical pattern; , a means for inputting a translation pattern corresponding to the input sentence pattern from the input device, and a means for newly registering the input sentence pattern group and the translation pattern group in correspondence with each other in the typical pattern file. A machine translation method according to claim 1, characterized in that: 4. Paying attention to the uppercase/lowercase letters of each word in the input sentence, consider words in the input sentence that begin with a capital letter, excluding the first word, to be variable words and replace them with special symbols, and store the variable words in memory. 2. The machine translation method according to claim 1, further comprising means for patterning an input sentence by . 5. For the word extracted from the above input sentence, set the attribute condition of the word, look it up in a dictionary, check the attribute of the word, and then consider the word that matches the condition as a variable word and replace it with a special symbol. 5. The machine translation method according to claim 4, further comprising means for patterning an input sentence by storing variable words in a memory. 6. Focusing on the combinations of words in the input sentence above, consider the words located either before or after a specified word as variable words and replace them with special symbols, and store the variable words in memory. 2. The machine translation system according to claim 1, further comprising means for patterning input sentences by storing them. 7. Focusing on the first letter of the first word of the patterned input sentence, using a typical pattern file that has already been sorted in alphabetical order and registered, the first letter of each pattern in the file and the above input sentence are Claim 1, characterized in that it has means for comparing the first character of
Machine translation method described in section. 8. Focusing on the specified position word in the patterned input sentence, compare the word with the keyword in the file using a registered typical pattern file that already has the specified position word as a keyword. A machine translation system according to claim 1, characterized in that the machine translation system comprises means. 9. When replacing the special symbol in the translation pattern extracted corresponding to the input sentence pattern with the variable word in memory, look up the variable word in the dictionary, select the translation of the variable word, and replace the special symbol with the variable word in the memory. 2. The machine translation method according to claim 1, further comprising means for generating a translated sentence corresponding to an input sentence by replacing symbols with the translated word. 10. When performing a dictionary lookup for the variable word extracted from the input sentence, a means for selecting a translation of the variable word and storing it in memory; 6. The machine translation method according to claim 5, further comprising means for generating a translated word corresponding to the input sentence by replacing a special symbol with a translated word of the variable word in the memory.