JPH08180057A - Method and device for retrieving document - Google Patents

Abstract

PURPOSE: To provide a document retrieval method and a device therefor capable of providing more useful information to a user who thinks of translating a retrieval input sentence. CONSTITUTION: The combination of most similar clauses between the retrieval input sentence of a first language inputted from an input part 1 and a retrieval object sentence inside a translation storage part 12 is obtained in an optimum combination calculation part 7. From the combination, in a word correspondence detection part 9, the work of the retrieval object sentence corresponding to the word of the retrieval input sentence is detected and further, the word of the sentence of a second language corresponding to the word of the retrieval object sentence is detected from the translation storage part 12. Also, the word of the second language which is the equivalent of the word provided in the retrieval input sentence is retrieved from a equivalent storage part 13 in an equivalent retrieval part 8 and a word replacing part 11 replaces the word of the sentence of the second language corresponding to the word of the retrieval object sentence detected in the word correspondence detection part 9 with the word or word string of the second language which is the equivalent of the word of the retrieval input sentence retrieved in the equivalent retrieval part 8.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text search method and apparatus suitable for searching a text similar to a search input text input by a user from a large amount of text.

[0002]

2. Description of the Related Art In recent years, with the spread of text input devices such as Japanese word processors and optical character readers (OCRs), the texts previously stored in the form of paper or microfilm have been digitized into magnetic disks. It has come to be stored in an external storage device such as a magnetic disk or a magneto-optical disk. Further, a text search technique for searching a document required by a user from a large amount of digitized document data has been developed and is being developed.

In the conventional search technique, it is general to search for a sentence in which a character string matches an input search sentence, or a word in which it is used, and the sentence as a whole is similar. I couldn't search for the sentence.

On the other hand, in a method in which a sentence is divided into a certain sentence constituent unit, for example, a bunsetsu unit, the degree of similarity is calculated for each bunsetsu, and the most similar bunsetsu combination is obtained, the whole sentence is similar. You can search the sentence you are doing. As an application of this search technique, it is possible to search for a dictionary example. An example of a dictionary is composed of a set of sentences of a Japanese sentence (which is the first language) and a translated sentence of the second language (for example, English). Using the above search technique, if a Japanese sentence similar to the input Japanese sentence is searched from the example, the parallel sentence, that is, the sentence in the second language, will be displayed when the input Japanese sentence is translated into the second language. Give useful information to.

However, this method cannot use the presented sentence of the second language as it is, and the user
A new translation had to be created with reference to the sentence in the second language.

Also, in this method, Japanese (first language) is used.
However, the information obtained from the second language could not be utilized even though the second language was also a language.

[0007]

As described above, in the conventional document retrieval method, the parallel translation of the retrieved search target sentence is merely a reference when the user translates, and if any correction is made, the translation of the input sentence is translated. There was a problem that the user had to think about whether or not to create a new translated sentence. In addition, since the parallel translation of the retrieved search target sentence is only for the purpose of being referred to by the user, there is a problem that the linguistic information cannot be used.

The present invention has been made in view of the above-mentioned drawbacks, and an object thereof is to process and present a bilingual translation of a retrieved retrieval target sentence so as to match the translation of the retrieval input sentence input by the user. A document search method and apparatus are provided.

Another object of the present invention is to provide a sentence in the second language, which is a parallel translation of the retrieved sentence in the first language, to the second language as an input sentence for the retrieval. It is an object of the present invention to provide a document search method and apparatus that can present useful information such as searching for example sentences in a second language that could not be searched for in the first language.

[0010]

Means and Actions for Solving the Problems First of the Invention
In the document search method and apparatus for searching the search target sentence similar to the search input sentence in the first language, the configuration according to the viewpoint is a search target sentence in the first language and a sentence in the second language which is a parallel translation thereof. And a word dictionary in which a word or a word string in the second language that is a translation for the word in the first language is paired and stored.
Using the words included in the search input sentence of the language of as a key, a word in the second language that is a translation is searched from the word dictionary, and in the search target sentence obtained as described above, in an expression similar to the search input sentence. If there is a word corresponding to the word of the search input sentence, the corresponding word of the sentence of the second language that is paired with the obtained search target sentence is a word that is a translation of the word of the search input sentence. Alternatively, it is characterized in that the word string is replaced as it is.

Further, instead of directly replacing the corresponding word of the sentence of the second language that is paired with the obtained sentence to be searched with the word or word string that is the translation of the word of the search input sentence, The word or word string that is the translation of the word in the sentence may be transformed and replaced. The modification here can be performed according to the word attribute (utilization, tense, etc.) of the corresponding word of the sentence of the second language that forms a pair with the obtained search target sentence. Further, it is also possible to transform the sentence according to the syntactic attribute of the corresponding word of the sentence of the second language that forms a pair with the obtained sentence to be searched.

In the configuration according to the first aspect, after searching for a search target sentence similar to the search input sentence, the bilingual sentence of the search target sentence is directly replaced with the translation word or word string of the word of the search input sentence. Or you can transform it and replace it,
Instead of simply presenting a reference sentence for translating the search input sentence, it is possible to present a sentence corrected to a form close to the translated sentence of the search input sentence.

According to a second aspect of the present invention, in a document retrieval method and apparatus for retrieving a retrieval target sentence similar to a retrieval input sentence, a sentence in the first language and its parallel translation are provided.
Sentences in the same language are stored in association with each other, search results are obtained from the set by the search in the first language, and the bilingual sentence in the second language is further input in the second language. It is characterized by performing a search.

Further, according to the present invention, when the sentence of the second language in the configuration according to the second aspect is searched, the result that the parts judged to be similar at the time of the search of the first language are similar is obtained. Another feature is that they are presented with priority.

Of the search results of sentences in the second language,
It is also characterized in that the same result as the search result in the first language is preferentially presented. In the configuration according to the second aspect, after searching the search target sentence similar to the search input sentence,
The bilingual translation of the search target sentence is further searched as a search input sentence, and the useful information for the search input sentence is preferentially presented from the search result, so that the linguistic characteristic of the second language is utilized for the user. It can provide useful information.

[0016]

Embodiments of the present invention will be described below with reference to the drawings. [First Embodiment] FIG. 1 is a block diagram of a document retrieval apparatus according to the first embodiment of the present invention.

The document retrieval apparatus of FIG. 1 has an input unit 1, a control unit 2, a display unit 3, a word division unit 4, a phrase synthesis unit 5, a phrase similarity calculation unit 6, an optimum combination calculation unit 7, and a translated word search unit 8. ,
A word correspondence detection unit 9, a word transformation unit 10, a word replacement unit 11,
It is composed of a parallel translation storage unit 12 and a translated word storage unit 13.

The input unit 1 is for inputting a character string and various control codes (commands, etc.), and is an input device composed of, for example, a keyboard or an OCR. The control unit 2 controls the entire apparatus, and is, for example, a central processing unit (CPU). Here, the control unit 2 includes each unit in the apparatus (word dividing unit 4, phrase matching unit 5, phrase similarity calculating unit 6, optimal combination calculating unit 7, translation word searching unit 8, word correspondence) for executing document search processing. The detection unit 9 and the word transformation unit 10) are controlled.

The display unit 3 is for displaying a character string of an input sentence or a character string of a retrieval target sentence, and is, for example, a CRT display device or a liquid crystal display device. The parallel translation storage unit 12 forms a search target sentence that is a first language (for example, Japanese) and a sentence of a second language (for example, English) that is a parallel translation thereof into a group, and expresses words and expressions of each other's words. It is for storing information in a form that can be dealt with.

The translated word storage unit 13 stores words in the first language,
It is for storing a word or a word string in the second language, which is the translated word, as a set. The parallel translation storage unit 12
The translated word storage unit 13 is realized by a large-capacity external storage device such as a hard disk device or an optical disk device.

The word dividing unit 4, the phrase synthesizing unit 5, the phrase similarity calculating unit 6, the optimum combination calculating unit 7, the translated word searching unit 8, the word correspondence detecting unit 9 and the word transforming unit 10 are functions necessary for the document searching process. The elements are realized by their own processing routines.

The word dividing unit 4 divides the character string of the input sentence of the first language (hereinafter referred to as a search input sentence) input from the input unit 1 into word units. Phrase synthesizer 5
Is to synthesize a clause consisting of words divided by the word division part.

The bunsetsu similarity calculation unit 6 detects the bunsetsu of the search input sentence obtained by the bunsetsu synthesizing unit 5 and the sentence of the first language to be searched stored in the parallel translation storage unit 12 (hereinafter referred to as a search target). (Referred to as a sentence) to calculate the degree of similarity with the clause.

The optimum combination calculation unit 7 obtains a combination of clauses that are most similar to each other on the basis of the similarity calculation result. The translated word search unit 8 searches the translated word storage unit 13 for the second language, which is a translated word, by using the word divided by the word dividing unit 4 from the search input sentence as a key.

The word correspondence detection unit 9 searches the word of the search target sentence corresponding to the word of the search input sentence from the combination of clauses obtained by the optimum combination calculation unit 7, and also corresponds to the word of the search target sentence. The word of the sentence in the second language is detected from the parallel translation storage unit 12.

The word transforming unit 10 detects a word or word string in the second language, which is a translation of the word of the search input sentence, from the sentence in the second language of the search target sentence detected by the word correspondence detecting unit 9. It is transformed according to the word attribute and the syntax attribute of the word.

The word replacing unit 11 is the second language or word string which is a translation of the word of the search input sentence transformed by the word transforming unit 10, and is the second sentence of the search target sentence detected by the word correspondence detecting unit 9.
It replaces words in sentences in the language.

Next, an outline of the operation of the configuration shown in FIG. 1 will be described. First, a character string in the first language, which is a search input sentence, is input from the input unit 1. This search input sentence is sent to the word dividing unit 4 via the control unit 2 and divided into word units. FIG. 2A is a diagram showing a situation in which the character string of the search input sentence is analyzed and divided into word units. At this time, the word division unit 4 divides the divided unit into parts of speech and inflections,
Information on word attributes (called morpheme information) such as tenses is added.

The result of word division is sent to the phrase synthesizing unit 5. The phrase synthesizing unit 5 synthesizes the words obtained by the word dividing unit 4 to generate a phrase as shown in FIG.
The phrase similarity calculation unit 6 extracts an arbitrary sentence from the parallel translation storage unit 12 as a search target sentence. In the parallel translation storage unit 12, in addition to the character string of the search target sentence,
Words and phrases are stored. An example of the contents stored in the parallel translation storage unit 12 is shown in FIG.

The phrase similarity calculation unit 6 compares the one phrase of the search target sentence retrieved from the parallel translation storage unit 12 with the one phrase of the search input sentence obtained by the phrase synthesis unit 5 to calculate the similarity. The operation is performed by a brute force of (number of clauses of search input sentence) × (number of clauses of search target sentence). Based on this calculation result, the optimum combination calculation unit 27 finds the most similar combination for the entire sentence.

The control unit 2 judges from the optimum combination obtained by the optimum combination calculation unit 7 whether or not the search input sentence and the search target sentence are similar, and if they are similar, the search target sentence. Is the search result for the search input sentence. In this case, the translated word search unit 8 searches for a translated word from the translated word storage unit 13 using each word of the search input sentence as a key.

The word correspondence detection unit 9 detects the word of the search target sentence corresponding to the word of the search input sentence from the combination of clauses obtained by the optimum combination calculation unit 7, and further detects the search target sentence from the parallel translation storage unit 12. The word of the translated word corresponding to the word of is detected.

The word transforming unit 10 determines the translation of the word of the search input sentence obtained by the translation searching unit 8 according to the word attribute or the syntax attribute of the word of the translation of the search target sentence detected by the word correspondence detection unit 9. Deform.

The word substituting unit 11 replaces the word of the translation of the search target sentence detected by the word correspondence detecting unit 9 with the translation of the word of the search input sentence transformed by the word transforming unit 10. The result is displayed on the display unit 3.

At this time, if the unsearched sentence to be searched remains in the parallel translation storage unit 28, one unsearched sentence is taken out from it and the above operation is repeated. The above is the outline of the operation of the configuration of FIG.

Next, regarding the details of the operation of the configuration of FIG.
This will be described with reference to the flowcharts of FIGS. 4 and 5. First, when a search input sentence is input from the input unit 1 (step S1), the control unit 2 displays the search input sentence on the display screen of the display unit 3 and outputs it to the word dividing unit 4. Here, it is assumed that, for example, a Japanese character string (character string in the first language) "It takes at least 5 minutes from here to the station" is input as the search input sentence.

The word dividing unit 4 analyzes the character string of the input search input sentence and divides it into word units (step S).
2). That is, as shown in FIG. 2A, the word segmentation unit 4 indicates that “it takes 5 minutes from here to the station” in the search input sentence “here”, “kara”, “station”, “to”, “ It is divided into words such as "ha", "at least", "5", "minutes", "ha", and "take", and information on each part of speech and word attribute information such as conjugations and tenses is added.

When the word dividing process by the word dividing unit 4 is completed, the control unit 2 activates the phrase synthesizing unit 5. The phrase combining unit 5 combines the words obtained by the word dividing unit 4 to generate a phrase as shown in FIG. 2B (step S3). Here, five clauses of “from here”, “to the station”, “at least”, “for 5 minutes”, and “take” are generated.

When the bunsetsu synthesizing unit 5 completes the bunsetsu synthesizing process, the bunsetsu similarity calculating unit 6 is activated by the control unit 2. The phrase similarity calculation unit 6 extracts an arbitrary sentence as a search target sentence from the parallel translation storage unit 12 (step S
4).

In the parallel translation storage unit 12, in addition to the character string of the sentence to be searched, the character string is divided into words by the processing of the word dividing unit 4 (or a function equivalent to the word dividing unit 4) in advance. The clauses generated by the processing of the clause synthesizer 5 (or the function equivalent to the clause synthesizer 5) are stored. As a search target sentence, for example, when a character string “It takes 5 hours from here to the port” is stored in the parallel translation storage unit 12, the parallel translation storage unit 12 is further processed by the processing of the word dividing unit 4 in FIG. "Here", like (a),
The words "kara", "minato", "to", "ha", "5", "time", "ha", and "take" are stored, and as shown in FIG. The phrases “from here”, “to the port”, “5 hours”, and “take” are stored by the processing of the part 5.

The phrase similarity calculator 6 compares one phrase of the search input sentence obtained by the phrase synthesizer 5 with one sentence of the search target sentence retrieved from the parallel translation storage unit 12 to calculate the similarity. (Step S5). This similarity calculation is performed by a brute force of (number of clauses of search input sentence) × (number of clauses of search target sentence), and is repeated for all combinations of clauses (step S6).

The phrase similarity calculation unit 6 obtains a phrase having a predetermined similarity or higher by this similarity calculation. Here, the combination of “from here” in the clause of the search input sentence and “from here” in the search target sentence has a high degree of similarity (match), and “to the station” in the search input sentence and “ports in the search target sentence” "Ha" is somewhat similar (similar).

When the phrase similarity calculation unit 6 completes the similarity calculation, the control unit 2 activates the optimum combination calculation unit 7. The optimum combination calculation unit 7 obtains a combination of phrases that are the most similar as a whole sentence between the search input sentence and the search target sentence, based on the similarity calculation result of the phrase similarity calculation unit 6 (step S7). . Here, the optimum combination of the phrase of the search input sentence and the phrase of the search target sentence is, as shown in FIG. 6, “from here” of the search target sentence to “from here” of the search input sentence, as shown in FIG. “To the station” is the search target sentence, “To the port”, “5 minutes” is the search target sentence, “5 hours is”, and “Search” is the search target sentence. It is calculated that the combination that associates “takes” is optimal.

When the optimum combination calculating unit 7 finds the optimum combination of the phrase of the search input sentence and the phrase of the search target sentence, the control unit 2 determines that the search input sentence and the search target sentence are similar to each other from the optimum combination. Whether or not the search target sentence can be used as the search result (step S8). In order to simplify the description, the criterion here is that the search input sentence and the search target sentence are similar to each other if the combination includes at least one clause that is similar (for example, the degree of similarity is not zero). It is what
Of course, this criterion depends on the required search accuracy, such as whether all the phrases have a certain degree of similarity or not, or whether a certain proportion of the number of phrases has a certain degree of similarity or more. Can be appropriately changed and set.

When the control unit 2 determines that the search input sentence and the search target sentence are similar to each other, the control unit 2 activates the translated word search unit 8. The translated word search unit 8 uses each word of the search input sentence, that is, each word (here, independent word) divided by the word dividing unit 4 for the search input sentence as a key, and uses the translated word storage unit 1 as a key.
A second language, which is a translation, is searched from 3 (step S
9). Specifically, as shown in FIG. 2C, the translated word of "station" is searched for as "station" and the translated word of "minute" as "minute".

When the translated word search process by the translated word search unit 8 is completed, the control unit 2 activates the word correspondence detection unit 9.
The word correspondence detection unit 9 detects the word of the search target sentence corresponding to the word of the search input sentence from the phrase combination obtained by the optimum combination calculation unit 7 (step S10). Specifically, from the set of the phrase “to station” in the search input sentence and the phrase “to port” in the search target sentence in FIG. 6, “station” corresponds to “port”, and “5 minutes” From "and 5 hours", it is detected that "minute" and "hour" correspond.

The word correspondence detection unit 9 further includes a parallel translation storage unit 1.
The word of the sentence of the second language corresponding to the word of the sentence to be searched is detected from 2 (step S11). This parallel translation storage unit 12
As shown in Fig. 3 (c), the sentence "It takes five hours to go to the seaport from he" is a translation of the search target sentence "It takes 5 hours from here to the port".
“Re” is stored as a set, and at the same time, information on each word (part of speech and word attribute) is also stored. The parallel translation storage unit 12 also stores the correspondence between the words of the search target sentence and the words of the bilingual sentence, as shown in FIG. Therefore, when the search target sentence is “it takes 5 hours from here to the port”, the word in the second language corresponding to “port” is
The second language corresponding to "seaport" and "time" is "hou
r "is detected. Finally, as shown in FIG. 7, it can be seen that the word “station” in the search input sentence corresponds to the word “seaport” in the parallel translation of the search target sentence, and similarly, “minute” corresponds to “hour”.

When the detection process by the word correspondence detecting unit 9 is completed, the control unit 2 activates the word transforming unit 10 and the word replacing unit 11 in order. The word transforming unit 10 and the word replacing unit 11 perform a process of transforming and replacing a sentence (inside word) in the second language, which is a parallel translation of the search target sentence, based on the detection result of the word correspondence detecting unit 9 ( Step S12). Details of the processing in step S12 will be described.

First, based on the detection result of the word correspondence detection unit 9, the process of transforming and replacing the word in the parallel translation of the search target sentence (step S12) is performed according to the word attribute of the parallel translation word of the search target sentence. A method of performing transformation / replacement, or a method of performing transformation / replacement in consideration of syntactic attributes related to parallel words of a search target sentence can be applied.

Therefore, the details of the processing in step S12 are modified according to the word attribute of the bilingual word of the search target sentence.
An example of replacement will be described with reference to the flowchart of FIG.

First, the word transforming unit 10 searches the word correspondence detecting unit 9 for the second word or word string in the second language, which is a translation of the search input sentence searched by the translation searching unit 8, for the second search target sentence.
It is transformed according to, for example, the word attribute (inflection, tense, etc.) of the word of the sentence of the language (step S20). Specifically, the word "station" in the search input sentence is replaced by the word "seap" in the parallel translation of the search target sentence.
Since it corresponds to "ort", the translated word "station" from the word attribute "singular" of "seaport" is transformed into the singular "station". Similarly, the word "minute" in the search input sentence corresponds to the parallel translation word "hour" in the search target sentence, so the translated word "minute" from the word attribute "plural" in "hour" is plural Is transformed into "minutes".

When the word transforming unit 10 transforms the sentence in the second language, which is a parallel translation of the sentence to be searched, this time, the word replacing unit 11 uses the result of the word transforming unit 10 to perform the transformed search. A process of replacing the word of the parallel sentence of the search target sentence with the translated word of the word of the input sentence is performed (step S21). Specifically, the parallel translation of the search target sentence `` It takes five hours to go to the seaport from her
"hours" and "seaport" in "e" are "minutes"
, "Station" is replaced by "It t," which is similar to the translated sentence of the search input sentence "It takes 5 minutes from here to the station".
akes five minutes togo to the station from here ''
A sentence in the second language is obtained.

When the transformation / replacement process (step S12) including the steps S20 and S21 by the word transformation unit 10 and the word substitution unit 11 is completed, the control unit 2 completes the sentence in the second language after the transformation / replacement. Is output to the display unit 3 (step S13). Then, the control unit 2 transfers control to the phrase similarity calculation unit 6, and if there is a retrieval target sentence that has not been taken out (unretrieved) in the translated word storage unit 13 (step S14), the translated word storage unit 13 Only one sentence to be retrieved is retrieved (step S4). The processing from step S4 is the same as when the search target sentence "It takes 5 hours from here to the port" is extracted. However, if it is determined in step S8 that the search input sentence and the search target sentence are not similar, the search target sentence is ignored and the process jumps to step S14. If it is determined in step S14 that no unsearched search target sentence remains in the translated word storage unit 13, the series of processes (document search process) shown in the flowchart of FIG.
Ends.

FIG. 8 shows a display example of search and word replacement according to this embodiment. FIG. 8A is a display example of a search input sentence, 8B is a display example of a search target sentence and its bilingual sentence, and FIG. 8C is a transformation and replacement of bilingual words in accordance with the search input sentence. It is a display example of a result. In the figure, the shaded portions indicate that the words of the translated or replaced or transformed target words are highlighted.

Next, the details of step S12 for transforming and replacing a word based on the result of the word correspondence detection unit 9 will be described as an example of the case where transforming / replacement is performed according to the syntactic attribute relating to the bilingual word of the search target sentence. First, referring to the flowchart of FIG.

First, in order to consider this processing, it is assumed that a character string such as "it takes at least 1 minute from here to the station" is input as a search input sentence. When the optimum combination of the search target sentence “it takes 5 hours from here to the port” is obtained through the processes of steps S1 to S7, as shown in FIG. 9, “from here”-“here” "To", "to the station"-"to the port",
"1 minute"-"5 hours", "takes"-"takes".

In this case, the translation word search unit 8 causes the translation word "station" for "station" and the translation word "a" for "1" as shown in FIG.
(one) "and" minute "are searched for translation" minute "(step S9). Then, as shown in FIG. 11, the word correspondence detection unit 9 determines that the word “station” of the search input sentence is the parallel translation word “seaport” of the search target sentence and “1” is the “five”.
It is detected that "minute" corresponds to "hour" (steps S10 and S11).

The word transforming unit 10 searches the word correspondence detecting unit 9 for the second word or word string in the second language, which is a translation of the word of the search input sentence searched by the translation searching unit 8, and searches for the second search target sentence. It is transformed according to the word attribute of the word of the sentence of the language (step S30). Specifically, the word "station" in the search input sentence
Corresponds to the word “seaport” in the bilingual translation of the search target sentence, so the translated word “station” of the word “station” is transformed into the singular “station” from the word attribute “singular” of “seaport”. Similarly, the word "1" corresponds to the word "five",
Since there is no word attribute of "five", the translated word "a" of the word "1"
Or transformed into "one". Also, the word "minute"
Corresponds to the word "hour", the word attribute "plural" of "hour" is transformed into the plural "minutes" of the translated word "minute" of the word "minute".

The word substituting unit 11 uses the result of the word transforming unit 10 to replace the word in the bilingual sentence of the search target sentence with the translation of the word in the transformed search input sentence (step S).
31). Specifically, the parallel translation of the search target sentence "It takes fiv
e hours to go to the seaport from here ''
"e", "hours", and "seaport" are "a (on
e) "," minutes "," station ",
`` It takes a minutes to go to the station from here
The sentence "" is obtained. The sentence in the second language in which the words are replaced by the word replacing unit 11 (the word replacing unit 1
The result 1) is returned to the word transformation unit 10.

The word transforming unit 10 checks the result of the word replacing unit 11 using the syntax attribute (step S32), detects a word that does not match the syntax attribute, and transforms the word (step S33). .

More specifically, the word transforming unit 10 is, for example, as shown in FIG.
It holds syntactic attributes like 2.
Or, if there is one, it is singular, and if there is a numeral other than one immediately before it (for example, two, three), it is plural. ” Therefore, in step S32, the word transforming unit 10 uses the syntax attribute held by the word transforming unit 10 as a result of the word replacing unit 11, that is, "It takes a minutes t".
By checking "go to the station from here", [a minutes] is
Or, if one exists, it is detected that it does not match the singular syntax attribute. In this case, the word transforming unit 10 transforms "minutes" into a singular "minute" in accordance with the syntax attribute in step S33, and passes the result to the word replacing unit 11 again.

The word substituting section 11 performs the processing of step S31 again, and returns "It takes a minute to go to the sta".
Since the sentence "tion from here" is obtained, the result is returned to the word transformation unit 10 again. Since the word which does not match the syntax attribute is not detected this time, the transformation / replacement process (step S12 in the flowchart of FIG. 4) including the steps S30 to S33 of FIG. 5B by the word transformation unit 10 and the word substitution unit 11 is performed. Ends. Then, the control unit 2 causes the display unit 3 to display the result of the transformation / replacement process for the parallel translation of the search target sentence.

The case where the word transforming unit 10 and the word replacing unit 11 transform and replace a word in the sentence of the second language which is a parallel translation of the search target sentence has been described above. It is also possible to adopt a configuration in which the replacing unit 11 also performs a process of transforming and replacing a word in the sentence of the first language that is the search target sentence. For this reason, the word transformation unit 10 sends the word of the search input sentence to the word correspondence detection unit 9
A processing function of transforming the word of the corresponding sentence to be detected detected in step 1 according to the word attribute or the syntax attribute is added. Further, the word replacing unit 11 is provided with a processing function of replacing the word of the corresponding search target sentence detected by the word correspondence detecting unit 9 with the word of the search input sentence transformed by the word transforming unit 10.

In this case, after the process of step S12 in the flowchart of FIG. 4, a process of transforming / replacement of the word of the search target sentence is added. The details of the process are step S1.
5A, which is a detailed flowchart of the process of FIG.
The flow is the same as in (b). Figure 5 if necessary
In the flowcharts of (a) and (b), "word translation" and "parallel translation word" should be replaced with "word", and "search input sentence translation" should be replaced with "search input sentence word".

That is, the word transforming unit 10 is, for example, as shown in FIG.
Similar to the process of step S20 in (a), the word of the search input sentence is transformed according to the word attribute of the word of the search target sentence checked by the word correspondence detection unit 9. Specifically, the word "station" in the search input sentence corresponds to the word "port" in the search target sentence, and since the word attribute of this "port" is "noun", it is not used, so the word "station" is also used. No deformation is performed. Similarly, since the word attribute of "minute" is also "noun", the corresponding word "time" is not transformed.

The word substituting unit 11 uses the result of the word transforming unit 10, for example, in the same manner as the process of step S21 in FIG. Replace words Specifically, in the example of the search target sentence "It takes 5 hours from here to the port",
By replacing "port" and "time" with "station" and "minute", respectively, the output "it takes 5 minutes from here to the station" is obtained.

FIG. 13 shows a display example of search and word replacement according to the present embodiment in which the process of transforming and replacing the word of the search target sentence is added. FIG. 13A shows a display example of a search input sentence, FIG. 13B shows a display example of a search target sentence and its parallel translation sentence, and FIG. 13C shows a parallel translation sentence of the search target sentence according to a word attribute. FIG. 13 is a diagram showing a situation in which a word is transformed / replaced.
FIG. 13D shows a display example in which the words of the bilingual sentence of the search target sentence are transformed / replaced according to the syntax attribute, and FIG. 13E shows a display example in which the words of the search target sentence are transformed / replaced according to the word attribute. is there.
In the figure, the shaded portion indicates that the word of the translated word that is to be replaced / transformed or has been replaced / transformed is highlighted as in FIG.

The present invention is not limited to the above embodiment. In the present embodiment, the system automatically performs the processing of word transformation / replacement of the bilingual sentence of the search target sentence according to the word of the search input sentence performed by the word transformation unit 10 and the word substitution unit 11. However, the word correspondence detection unit 9
When the correspondence between the word in the search input sentence and the word in the bilingual sentence is detected, the display unit 3 presents the correspondence detected by the system, and the user selects whether or not to perform transformation / replacement processing. You can let me do it.

Further, the word transformation unit 10 and the word replacement unit 1
Whenever the bilingual sentence of the search target sentence is transformed or replaced by 1,
Instead of displaying the result, for each search target sentence that is judged to be similar to the search input sentence, for example, the sum of the similarities is calculated as the evaluation value, and the corresponding search target sentence is arranged in descending order of the evaluation value. The transformation / replacement results of the bilingual sentence may be displayed side by side so that the user can select the result. To this end, a search process for the search input sentence (steps S4 to S8)
Processing) is repeatedly performed for all search target sentences in the parallel translation storage unit 12, all search target sentences that are determined to be similar to the search input sentence are held, and thereafter, a search with a large evaluation value is performed. Transformation of bilingual sentences in the order of target sentences
It is better to make a replacement. Further, only for the search target sentence having the largest evaluation value, the steps S9 to S1 are performed.
The process 3 may be performed.

In short, the present invention can be implemented with various modifications without departing from the scope of the invention. [Second Embodiment] FIG. 14 is a block diagram of a document retrieval apparatus showing a second embodiment of the present invention.

The document retrieval apparatus shown in FIG. 14 has an input unit 21, a control unit 22, a display unit 23, a sentence dividing unit 24, and a priority control unit 2.
5, a similarity calculation unit 26, an optimum combination calculation unit 27, a parallel translation storage unit 28, and a search unit 29.

The input unit 21 is for inputting a character string and various control codes (commands, etc.), and is an input device composed of, for example, a keyboard or OCR. Control unit 22
Controls the entire device.

The display unit 23 is an output means for displaying a character string, a character string to be searched, etc., and is, for example, a CRT display device or a liquid crystal display device. Sentence division unit 24
Is to analyze an input character string (sentence) and divide it into fixed sentence constituent units.

The priority control unit 25 selects and outputs a priority one from the search results by the search unit 29. The similarity calculation unit 26 compares the sentence constituent units of the character strings (sentences) in the same language (first language or second language) to calculate the similarity.

The optimum combination calculation unit 27 obtains a combination of sentence constituent units that are most similar to each other from the calculation result of the similarity calculation unit 26. The parallel translation storage unit 28 is for storing and storing a sentence in the first language and a sentence in the second language, which is the parallel translation thereof, as a set so that they can be associated with each other.

The search section 29 searches the parallel translation storage section 8 for similar sentences. Next, an outline of the operation of the configuration of FIG. 14 will be described. First, a character string in the first language, which is a search input sentence, is input from the input unit 21.
This search input sentence is sent to the sentence dividing unit 24 via the control unit 22, and a fixed sentence constituent unit (hereinafter, simply referred to as a unit).
Is divided into FIG. 15 is a diagram showing a situation in which a character string is analyzed and divided into units (here, divided into clauses). At this time, the sentence dividing unit 24 adds information such as a part of speech and an inflection to the divided units.

Next, the result of unit division is sent to the search unit 29. The search unit 29 extracts one sentence to be searched from the parallel translation storage unit 28. In the parallel translation storage unit 28, sentences in the first language and the second language are stored as a set. The sentences in the first language and the sentences in the second language are each divided into units,
Correspondence can be taken for each unit. Parallel translation storage unit 2
16 shows an example of the contents stored in FIG.

The search unit 29 sends the search target sentence extracted from the parallel translation storage unit 28 to the similarity calculation unit 26. The similarity calculation unit 26 calculates the similarity by comparing one unit of the search input sentence with one unit of the search target sentence, and calculates the similarity by (number of units of search input sentence) × (number of units of search target sentence) Will be brute force.
Based on this calculation result, the optimum combination calculation unit 27
Find the most similar combination for the whole sentence. The search unit 29 determines from the combination obtained by the optimum combination calculation unit 27 whether or not the corresponding search target sentence is similar to the search input sentence, and if similar, accepts the search target sentence as a search result. .

The above operation is repeated for all the search target sentences in the parallel translation storage unit 28. The search result obtained by the search unit 29 is a set of sentences in the first language and the second language, and has a sentence in the second language as a parallel translation.

The search unit 29 uses the sentence in the second language as the parallel translation of the search result (searched sentence to be searched) as the input sentence, this time using the second language to search in the previous first language. Similar to the case, the search target sentence stored in the parallel translation storage unit 28 is searched. After this search, if an unsearched sentence to be searched for in the second language remains in the parallel translation storage unit 28, one unsearched sentence is extracted from it and the above operation is repeated.

When outputting the search result obtained by the search for the second language, the control unit 22 sends the result to the priority control unit 25 to make it possible to determine which search result has priority. The result is output to the display unit 23. The priority control unit 25 determines which is to be prioritized in consideration of the result and the situation of the search for the first language and the search for the second language.

The above is the outline of the operation of the configuration of FIG. Next, regarding the details of the operation of the configuration of FIG. 14, FIG.
This will be described with reference to the flowchart in FIG.

In this embodiment, when a search input sentence is input from the input unit 21, an operation of searching the parallel translation storage unit 28 for a search target sentence that is similar to the search input sentence as a whole sentence is executed. .

First, when a search input sentence is input from the input unit 21 (step S41), the control unit 22 displays the search input sentence on the display screen of the display unit 23 and outputs it to the sentence dividing unit 24. Here, it is assumed that, for example, a Japanese character string (character string in the first language) "Please submit the document as soon as possible" is input as the search input sentence.

The sentence dividing unit 24 analyzes the character string of the input search input sentence and divides it into units (step S42). Specifically, as shown in FIG. 15, the sentence segmentation unit 24 sets the search input sentence “submit the document as soon as possible” to
Divide into certain units (here, clauses) such as "that", "documents", "one day", "early", "submit", and use parts of speech, conjugations, adjuncts, etc. Add information.

The control unit 22 outputs this information to the search unit 29. When the search unit 29 internally holds the information output from the control unit 22 (search input sentence divided by the sentence dividing unit 24), any one sentence is retrieved from the parallel translation storage unit 28 as a search target sentence ( Step S43). The search target sentence stored in the parallel translation storage unit 28 is divided into units by the function equivalent to that of the sentence dividing unit 24 in advance. It is assumed that, for example, a sentence "I want to read your paper as soon as possible" is retrieved from the parallel translation storage unit 28 as a search target sentence. In this case, as shown in FIG. 16, the parallel translation storage unit 28 stores units such as "Kimi no", "paper", "one day", "early", and "I want to read". , Divided by each unit, "I d like to read you
The bilingual information "r article assoon as possible" is also stored. The parallel translation storage unit 28 also stores (information indicating) a correspondence relationship between Japanese (first language) and English (second language) for each unit.

When the retrieval unit 29 completes the retrieval of the retrieval target sentence, the control unit 22 activates the similarity calculation unit 26. The similarity calculation unit 26 stores one unit (for example, a phrase) of the search input sentence obtained by the sentence dividing unit 24 and one unit (for example, a phrase) of the search target sentence extracted from the parallel translation storage unit 28 by the search unit 29. The comparison is performed and the similarity is calculated (step S44). This similarity calculation is performed by a brute force of (the number of units of the search input sentence) × (the number of units of the search target sentence), and is repeated for all combinations of units (step S45).

The optimum combination calculation unit 27 obtains a combination of units that are most similar in the entire sentence between the search input sentence and the search target sentence, based on the similarity calculation result of the similarity calculation unit 26 (step S46). Here, FIG.
As shown in FIG. 8, a combination of associating “all day” of the search input sentence with “all day of the search target sentence” and “early” of the search input sentence with “early” of the search target sentence is optimal. Is calculated.

The retrieving unit 29 uses the optimum combination calculating unit 27.
When the optimum combination of the unit of the search input sentence and the unit of the search target sentence is obtained by the above, it is determined from the optimum combination whether or not the search input sentence and the search target sentence are similar (step S47). In order to simplify the description, the judgment criterion in step S47 is a combination including at least one unit that is similar (for example, the degree of similarity is not zero) as in the case of the first embodiment. , It is assumed that the search input sentence and the search target sentence are similar. Here, it is determined that the search target sentence “I want to read your paper as soon as possible” retrieved in step S43 is similar to the search input sentence “submit the document as soon as possible”. I shall.

When the search unit 29 determines in step S47 that they are similar to each other, it accepts and holds the corresponding search target sentence as a search result (search result of the original sentence) (step S48), and then proceeds to step S49. . When the search unit 29 determines in step S47 that they are not similar to each other, the process directly proceeds to step S49.

The search unit 29, in step S49,
It is checked whether or not an unsearched search target sentence remains in the parallel translation storage unit 28, and if there is any, a series of search processing for the above-mentioned search input sentence (original sentence) (step S43-).
S48) is repeated.

In this way, when the search process for the search input sentence (original sentence) is performed for all the search target sentences in the parallel translation storage unit 28, the search unit 29 proceeds from step S49 to step S50.

As described above, the parallel translation storage unit 28 also stores the second language which is the parallel translation of the search target sentence (see FIG. 16). Here, `` I d like to read your
The English sentence "article as soon as possible" was determined to be similar to the search input sentence in step S47, "I want to read your paper as soon as possible" (held as a search result) It is stored as a parallel translation of (search target sentence).

The search unit 29, in step S50,
The English sentence of this bilingual translation, that is, the bilingual translation of the search result is taken out, and this also obtains information for each unit stored in the bilingual translation storage unit 28.

The search unit 29 uses the parallel translation sentence (the parallel translation sentence of the search target sentence similar to the search input sentence) of the search result extracted from the parallel translation storage unit 28 as the search input sentence in English (second language). In the same manner as that performed for Japanese in steps S43 to S48, an English search is also performed using the similarity calculation unit 26 and the optimum combination calculation unit 27.

That is, the retrieval unit 29 first takes out a bilingual translation of an arbitrary sentence as a retrieval target sentence from the bilingual storage unit 28 (step S51). The similarity calculation unit 26 calculates one unit (for example, a word) of the parallel translation sentence (English search input sentence) of the search result and the translation sentence (the English search target sentence) of the search target sentence retrieved from the parallel translation storage unit 28 by the search unit 29. ) Is compared with one unit (for example, a word) to calculate similarity (step S5).
2). The similarity calculation unit 26 repeats this similarity calculation for the total number of (the number of units of the search input sentence) × (the number of units of the search target sentence) (step S53).

The optimum combination calculation unit 27, based on the similarity calculation result of the similarity calculation unit 26, a combination of units that are most similar as a whole sentence between the English search input sentence and the English search target sentence. Is calculated (step S54).

The search unit 29 uses the optimum combination calculation unit 27.
When the optimal combination of the unit of the English search input sentence and the unit of the English search target sentence is determined by, the optimum combination is used to determine whether the English search input sentence and the English search target sentence are similar. It is determined (step S54). If it is determined that they are similar, the search unit 29 internally holds the Japanese search target sentence corresponding to the English search target sentence as a search result (step S56).

If the search unit 29 determines in step S55 that the new English search input sentence is similar to the English search target sentence, it executes step S56 and then determines that they are not similar. Skips step S56,
It is checked whether or not the unsearched search target sentence for the new English search input sentence remains in the parallel translation storage unit 28 (step S57).
The subsequent processing is repeated until there are no unsearched search target sentences.

As a result of the above, for example, the original search input sentence (search input sentence of the original Japanese) "Please submit the document as soon as possible""Your paper as soon as possible""I d like to re"
19 (a) as a search result in English (second language) for the new search input sentence when "ad your article as soon as possible" is set as the new search input sentence.
It is assumed that the search result as shown in FIG.

The control unit 22 receives the search result obtained by the search unit 29, outputs the search result to the priority control unit 25, and activates the priority control unit 25. The priority control unit 25 determines what is to be preferentially output from the received search result (step S58). Either the first procedure (method) shown in the flowchart of FIG. 20 (a) or the second procedure (method) shown in the flowchart of FIG. 20 (b) can be applied to the determination processing in step S58. However, the specific procedure will be described later.

The determination result of the priority control unit 25 is returned to the control unit 22. When the control unit 22 receives the determination result of the priority control unit 25, the control unit 22 preferentially outputs the sentence of the search result determined to have priority to the display screen of the display unit 23 (step S59). When the output processing of the search result by 22 is performed, the unprocessed search result (that is, the search result used for the search with the parallel translation as a new search input sentence) for the original search input sentence (Japanese search input sentence) It is checked whether or not it remains (step S60), and if it remains, the process returns to step S50. In this way, the processing from step S50 is repeated until there are no unprocessed search results.

Next, the details of the determination processing in step S58 will be described in the case where the first procedure shown in the flowchart of FIG. 20 (a) is applied. For example, the original search input sentence is "Please submit the document as soon as possible."
And, as a result of the search, "I want to read your paper as soon as possible" is obtained. In addition, as a parallel translation of this search result, `` I d like to read your articleas
"as soon as possible", and the result of searching it as an input sentence is 191a to 19a in FIG.
It is assumed that there are 5 sentences indicated by 5a. Furthermore, among them, sentences 191a, 192a, and 194a are "as (soon) as
The part of "possible" is similar, and sentences 193a, 19
5a is similar in the "would like to" portion.

The priority control section 25 first extracts one sentence from the search result of the search section 29 received from the control section 22 (step S60), and checks information about which part is similar (step S61). Sentence 191a, "I ll do it a
"s soon as possible" means "as soon as possible"
You can see that the part of is similar.

Next, the priority control unit 25 uses the information that the Japanese sentence "as soon as possible" corresponds to the English sentence "as soon as possible", and determines whether the similar portions are the same. It is determined (step S62) that if similar parts are the same, priority is given (step S63), and if they are different, priority is not given. In the above example, sentence 191a has priority, but sentence 1
93a's "I d like to know where the money came fr
"om" is different from "as soon as possible" above, but "woul
Since "d like to" is similar, it is not prioritized.

By the above procedure, the priority control unit 25 checks whether or not to give priority to each of all the search results received from the control unit 22 (step S64). As a result, in the above example, sentences 191a, 192a, 194
It is determined that a has priority and sentences 193a and 195a do not have priority.

Next, the details of the determination processing in step S58 will be described for the case where the second procedure shown in the flowchart of FIG. 20 (b) is applied. First, in addition to "I want to read your paper as soon as possible" as the search result of the original search input sentence "Please submit the document as soon as possible", Should be computerized as soon as possible ".
The search result in this case is indicated by reference numeral 191 in FIG.
b, 192b. In addition, as shown in Fig. 19 (a), a parallel translation of "I want to read your paper as soon as possible""I d like to read your article as soon as possibl
In the search results using "e" as input, a translation of "We mus should be computerized in this department as soon as possible"
t computarize this department as soon as possible
"It is included.

First, the priority control unit 25 takes out one sentence from the search result as shown in FIG. 19 (a) (the result of searching the original search input sentence by using the parallel translation of the search result as the input sentence) (step S70). , And the search result in the original Japanese as shown in FIG. 19B is compared (step S71). Then, the priority control unit 25 checks whether or not the retrieved one sentence is in the search result in Japanese (step S72), and if there is, gives priority (step S7).
3) If not, there is no priority.

Through the above procedure, the priority control unit 25 retrieves the search result of the original search input sentence as the input sentence and retrieves the retrieval result shown in the 19th (a) (all retrievals received from the control unit 22. It is checked whether or not each of the results) has priority (step S74). As a result, in the above example, it is determined that the sentence 192a has priority and the others do not have priority.

The present invention is not limited to the above embodiments. In this embodiment, Japanese is assumed as the first language and English is assumed as the second language, but they may be reversed. It may be another language other than Japanese and English.

Further, the priority control section 25 may give priority to the output result instead of giving priority to the output result. In addition, the sequence of the search in the first language → the retrieval of the parallel translation from the search result → the search in the second language (parallel translation) may be performed according to an instruction from the user, or the system may be used. May be done automatically. In short, the present invention can be variously modified and implemented without departing from the scope of the invention.

[0112]

As described above, according to the present invention, a document retrieval method and apparatus for retrieving a retrieval target sentence similar to the retrieval input sentence in the first language is a parallel translation of the retrieved retrieval target sentence. Presenting a reference sentence for simply translating the search input sentence by arranging the word of the sentence of the second language by replacing the translated word of the word corresponding to the word with the corresponding search input sentence Instead, it is possible to present a sentence in which the search input sentence is corrected to a form close to the translated sentence, and thus to present a more useful example sentence to a user who wants to translate the search input sentence.

Further, according to the present invention, after a search target sentence similar to the search input sentence in the first language is searched, a sentence in the second language, which is a parallel translation of the search target sentence, is newly input as a search input. By configuring the sentence to be searched again, it is possible to provide the user with useful information that makes use of the linguistic characteristics of the second language. In particular, at the time of another search, priority is given to presenting results that are similar to the portion determined to be similar at the time of searching the first language, or among the search results of sentences in the second language, By preferentially presenting the same search result in one language, more useful information can be provided to the user.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of a document search device showing a first embodiment of the present invention.

FIG. 2 is a diagram showing an example of a search input sentence, and FIG.
Is a word and word information of the search input sentence divided by the word dividing unit 4, FIG. 2B is a phrase of the search input sentence synthesized from the words by the phrase synthesizing unit 5, and FIG. 2C is a search input. The correspondence between the words in the sentence and the translated words searched by the word search unit 8 is shown.

FIG. 3 is a diagram showing an example of contents stored in a parallel translation storage unit 12,
FIG. 3A shows the words and word information of the search target sentence as shown in FIG.
3 (b) shows the clauses of the search target sentence, FIG. 3 (c) shows the words and word information of the translation target sentence, and FIG. 3 (d) shows the words of the search target sentence and the corresponding translation words.

4 is a flowchart showing a processing flow of the configuration of FIG.

5 is for explaining the details of the process of step S13 in FIG. 4, FIG. 5 (a) is a word of a search input sentence, and a method of transforming / replaceing the corresponding word of the search target sentence is applied. FIG. 5B is a flowchart to which a method for performing transformation according to syntax attributes is applied.

FIG. 6 is a diagram showing a combination of a phrase of a search input sentence and a phrase of a search target sentence, which are obtained by an optimum combination calculation unit 7.

FIG. 7 is a diagram showing the words of the search input sentence, their translations, the words of the search target sentence, and the bilingual words of the search target sentence, which are detected by the word correspondence detection unit, in association with each other.

FIG. 8 is a diagram showing a display example of search and word replacement in the configuration of FIG.

FIG. 9 is a diagram showing a combination of a phrase of a search input sentence and a phrase of a search target sentence, which are obtained by an optimum combination calculating unit 7.

FIG. 10 is a diagram showing words in a search target sentence and corresponding bilingual words.

FIG. 11 is a diagram showing the words of the search input sentence, their translations, the words of the search target sentence, and the words of the search target sentence, which are detected by the word correspondence detection unit, in association with each other.

FIG. 12 is a diagram showing syntax attributes.

13 is a diagram showing a display example of search and word replacement in the configuration of FIG.

FIG. 14 is a block configuration diagram of a document search device showing a second embodiment of the present invention.

FIG. 15 is a diagram showing a unit divided by a sentence dividing unit and information added to the unit.

16 is a diagram showing an example of a search target sentence stored in a parallel translation storage unit 28. FIG.

FIG. 17 is a flowchart showing a processing flow of the configuration of FIG.

FIG. 18 is a diagram showing a combination of phrases obtained by an optimum combination calculation unit 27.

FIG. 19 is a diagram showing an example of a result of a search in a second language.

20 is a diagram for explaining the details of the process of step S58 in FIG. 18, and FIG. 20 (a) shows a translation word corresponding to a unit determined to be similar in the search by the original search input sentence. FIG. 20B is a flowchart showing a procedure for giving priority to a similar part in the search by the translated word. FIG. 20B shows the result of the search for the translated word for the original search input sentence. The flowchart which showed the procedure in the case of giving priority to what is also contained in the search result.

[Explanation of symbols]

1, 21 ... Input unit, 2, 22 ... Control unit, 3, 23 ... Display unit, 4 ... Word division unit, 5 ... Phrase synthesis unit, 6 ... Phrase similarity calculation unit, 7, 27 ... Optimal combination calculation unit, 8 ... Word search unit, 9 ... Word correspondence detection unit, 10 ... Word transformation unit, 11 ...
Word replacement section, 12, 28 ... Parallel translation storage section, 13 ... Translated word storage section (word dictionary), 24 ... Sentence division section, 25 ... Priority control section (similar part determination means, search result determination means), 26 ... Similarity calculation Part (first similarity calculation means, second similarity calculation means), 29 ... Search unit (first search means, second search means).

Claims (14)

[Claims] 1. A search input sentence in the first language and a search target sentence are compared for each certain sentence constituent unit to calculate a similarity of the sentence constituent unit, and it is determined that one sentence as a whole is similar. In the document search method for obtaining a search target sentence, a parallel translation storage unit for storing a search target sentence in the first language and a sentence in the second language, which is a parallel translation thereof, in association with each other; Has a word dictionary that is a set of words or word strings in the second language that are translations for words in the first language, and translates words from the word dictionary using the words included in the search input sentence in the first language as a key When a word in the second language is searched for, and in the obtained search target sentence, there is a word corresponding to the word of the search input sentence in the expression similar to the search input sentence, the obtained Corresponding to the sentence in the second language that is paired with the search target sentence. Document search method a word, and replaces it by the search as the translation of the word in the input sentence word or word string. 2. A parallel translation storage means for storing the input means for inputting a character string, the search target sentence which is the first language, and the sentence which is the parallel translation of the second language as a set in association with each other. And a translated word storage means for storing a word in the first language and a word or word string in the second language, which is a translated word, in combination, and a search input sentence in the first language is input from the input means. And a similarity calculation means for calculating the similarity between the search input sentence and an arbitrary search target sentence in the parallel translation storage means for each fixed sentence constituent unit, and the calculation of the similarity calculation means An optimum combination calculating means for obtaining a combination of sentence composition units that are most similar between the search input sentence and the search target sentence based on the result; and a second translation that is a translation of a word included in the search input sentence. Of words in different languages from the translated word storage means And a word in the second language corresponding to the word in the search input sentence from the combination obtained by the means, and a word in the second language corresponding to the word in the search input sentence. Of the word in the second language sentence corresponding to the word in the sentence to be searched detected by the word correspondence detecting unit, and the word correspondence detecting unit searches for the word in the second language corresponding to the word in the search target sentence detected by the word correspondence detecting unit. And a word replacement unit that replaces with a word or a word string in the second language that is a translation of a word of the search input sentence that has been obtained, so as to obtain a corrected bilingual translation of the search target sentence similar to the search input sentence A document retrieval device characterized by the above. 3. A search input sentence in the first language and a search target sentence are compared for each fixed sentence constituent unit to calculate a similarity of the sentence constituent unit, and it is determined that one sentence as a whole is similar. In the document search method for obtaining a search target sentence, a parallel translation storage unit for storing a search target sentence in the first language and a sentence in the second language, which is a parallel translation thereof, in association with each other; Has a word dictionary that is a set of words or word strings in the second language that are translations for words in the first language, and translates words from the word dictionary using the words included in the search input sentence in the first language as a key And searching for a word in a second language that is, and in the obtained search target sentence, there is a word corresponding to the word of the search input sentence in an expression similar to the search input sentence, the obtained Corresponds to the sentence in the second language that is paired with the search target sentence Term and document search method characterized by replacing by transforming the word or word string as the translation of the words in the search input sentence. 4. When replacing a corresponding word of a sentence of a second language, which is paired with the obtained sentence to be searched, with a word or a word string that is a translation of the word of the search input sentence, the second word
4. The document search method according to claim 3, wherein the translated word or word string is modified and replaced according to the word attribute of the corresponding word of the sentence of the language. 5. When replacing a corresponding word of a sentence of a second language that forms a pair with the obtained search target sentence with a word or a word string that is a translation of the word of the search input sentence, the second word
4. The document search method according to claim 3, wherein the translated word or word string is modified and replaced according to the syntax attribute of the corresponding word of the sentence of the language. 6. A parallel translation storage means for storing the input means for inputting a character string, the search target sentence of the first language, and the parallel sentence of the second language as a set in association with each other. And a translated word storage means for storing a word in the first language and a word or word string in the second language, which is a translated word, in combination, and a search input sentence in the first language is input from the input means. And a similarity calculation means for calculating the similarity between the search input sentence and an arbitrary search target sentence in the parallel translation storage means for each fixed sentence constituent unit, and the calculation of the similarity calculation means An optimum combination calculating means for obtaining a combination of sentence composition units that are most similar between the search input sentence and the search target sentence based on the result; and a second translation that is a translation of a word included in the search input sentence. Of words in different languages from the translated word storage means And a word in the second language corresponding to the word in the search input sentence from the combination obtained by the means, and a word in the second language corresponding to the word in the search input sentence. Word correspondence detecting means for detecting the word from the parallel translation storage means, and word transforming means for transforming a word or word string in the second language which is a translation of the word of the search input sentence searched by the translation search means. A word replacement unit that replaces the word of the sentence in the second language corresponding to the word of the search target sentence detected by the word correspondence detection unit with the word or word string transformed by the word transforming unit. A document search device, wherein a corrected bilingual translation of a search target sentence similar to the search input sentence is obtained. 7. A search input sentence in the first language and a search target sentence are compared for each certain sentence constituent unit to calculate a similarity of the sentence constituent unit, and it is determined that one sentence as a whole is similar. In the document search method for obtaining a search target sentence, a parallel translation storage unit that stores a search target sentence in the first language and a sentence in a second language, which is a parallel translation thereof, in association with each other; Has a word dictionary that is a set of words or word strings in the second language that are translations for words in the first language, and translates words from the word dictionary using the words included in the search input sentence in the first language as a key When a word in the second language is searched for, and there is a word in the expression similar to the search input sentence that corresponds to the word of the search input sentence in the obtained search target sentence, the search target sentence Transform the word in the corresponding search input sentence At the same time, the word corresponding to the sentence of the second language, which is paired with the word of the search target sentence, is replaced by transforming the word or word string that is the translation of the word of the search input sentence. Document search method. 8. A parallel translation storage means for storing the input means for inputting a character string, the search target sentence which is the first language, and the sentence which is the parallel translation of the second language as a set in association with each other. And a translated word storage means for storing a word in the first language and a word or word string in the second language, which is a translated word, in combination, and a search input sentence in the first language is input from the input means. And a similarity calculation means for calculating the similarity between the search input sentence and an arbitrary search target sentence in the parallel translation storage means for each fixed sentence constituent unit, and the calculation of the similarity calculation means An optimum combination calculating means for obtaining a combination of sentence constituent units that are most similar between the search input sentence and the search target sentence based on the result; and a second translation that is a translation of a word included in the search input sentence Of words in different languages from the translated word storage means And a word in the second language corresponding to the word in the search input sentence from the combination obtained by the means, and a word in the second language corresponding to the word in the search input sentence. Word correspondence detecting means for detecting the word in the parallel translation storage means, and a word or word string in the second language which is a translation of the word in the search input sentence and the word in the search input sentence searched by the translation word searching means. Replacing the word of the search target sentence detected by the word correspondence detection unit with the word of the search input sentence modified by the word transformation unit, and detecting by the word correspondence detection unit The word of the sentence of the second language corresponding to the searched word of the search target sentence is a translation of the word of the search input sentence transformed by the word transforming means. ; And a word replacement means for replacing in the language of the word or word string,
A document search device, wherein a corrected parallel translation of a search target sentence similar to the search input sentence is obtained. 9. A one-sentence sentence is calculated by applying a search method of calculating a similarity between a given sentence constituent unit between a search input sentence and a search target sentence and finding a combination of the most similar sentence constituent units. In a document search method for searching a search target sentence that is determined to be similar as a whole, a bilingual translation in which a sentence in a first language and a sentence in a second language, which is a translation thereof, are associated and stored With respect to a search input sentence having a storage means, a search input sentence in a first language is searched by a search method for a first language from a set of sentences stored in the storage means. A document search method characterized in that a sentence of a second language corresponding to the sentence is used as an input sentence, and a set of sentences stored in the storage means is searched by the search method for the second language. 10. An input means for inputting a character string, and a parallel translation storage means for storing a search target sentence of a first language and a sentence of a second language which is a parallel translation thereof as a set in association with each other. A first similarity calculation means for calculating a similarity between a search input sentence in the first language and a search target sentence for each fixed sentence constituent unit; and a search input sentence in the second language and a search target sentence. Second similarity calculation means for calculating the similarity for each fixed sentence constituent unit between the first and second similarity calculation means, and the first language or the second similarity calculation means based on the calculation result of the first or second similarity calculation means. An optimum combination calculation means for obtaining a combination of sentence constituent units that are most similar to a language; and when a search input sentence of the first language is input from the input means, the storage means for the search input sentence The first language from the set of sentences stored in A first search unit that applies the similarity calculation unit and the optimum combination calculation unit to perform a search, and a sentence in a second language that is a parallel translation of the result of the search target sentence searched by the first search unit. As a new search input sentence, a search is performed by applying the second similarity calculation unit and the optimum combination calculation unit for a second language from a set of sentences stored in the storage unit, with respect to the search input sentence. And a second search means for executing the document search apparatus. 11. A one-sentence sentence is calculated by applying a search method for calculating a similarity between a given sentence constituent unit between a search input sentence and a search target sentence and finding a combination of sentence constituent units that are most similar. In a document search method for searching a search target sentence that is determined to be similar as a whole, a bilingual translation in which a sentence in a first language and a sentence in a second language, which is a translation thereof, are associated and stored With respect to a search input sentence having a storage means, a search input sentence in a first language is searched by a search method for a first language from a set of sentences stored in the storage means. A sentence in the second language corresponding thereto is used as an input sentence, and a search is performed for the second language from the set of sentences stored in the storage means by the search method. Document search method section of the second language, also characterized by a priority as a search result which are similar in the search by the second language corresponding to the determined sentence structural units are. 12. An input means for inputting a character string, and a parallel translation storage means for storing a search target sentence in a first language and a sentence in a second language which is a parallel translation thereof as a pair in association with each other. A first similarity calculation means for calculating a similarity between a search input sentence in the first language and a search target sentence for each fixed sentence constituent unit; and a search input sentence in the second language and a search target sentence. Second similarity calculation means for calculating the similarity for each fixed sentence constituent unit between the first and second similarity calculation means, and the first language or the second similarity calculation means based on the calculation result of the first or second similarity calculation means. An optimum combination calculation means for obtaining a combination of sentence constituent units that are most similar to a language; and when a search input sentence of the first language is input from the input means, the storage means for the search input sentence The first language from the set of sentences stored in A first search unit that applies the similarity calculation unit and the optimum combination calculation unit to perform a search, and a sentence in a second language that is a parallel translation of the result of the search target sentence searched by the first search unit. As a new search input sentence, a search is performed by applying the second similarity calculation unit and the optimum combination calculation unit for a second language from a set of sentences stored in the storage unit, with respect to the search input sentence. That the second language portion corresponding to the sentence constituent unit determined to be similar in the first search means is similar in the second search means A document retrieval apparatus, comprising: a similar portion determination unit for determining whether or not it is determined, and giving priority to a result determined to be similar by the similar portion determination unit as a search result. 13. A one-sentence sentence is obtained by applying a search method that calculates a similarity between certain sentence constituent units between a search input sentence and a search target sentence and obtains a combination of the most similar sentence constituent units. In a document search method for searching a search target sentence that is determined to be similar as a whole, a bilingual translation in which a sentence in a first language and a sentence in a second language, which is a translation thereof, are associated and stored With respect to a search input sentence having a storage means, a search input sentence in a first language is searched by a search method for a first language from a set of sentences stored in the storage means. A sentence corresponding to the second language is used as an input sentence, and a search for the second language is further performed from the set of sentences stored in the storage unit by the search method. From Among results, document search method characterized by giving priority to those contained in the search results for the first language as a search result. 14. An input means for inputting a character string, and a parallel translation storage means for storing a search target sentence of a first language and a sentence of a second language which is a parallel translation thereof as a set in association with each other. A first similarity calculation means for calculating a similarity between a search input sentence in the first language and a search target sentence for each fixed sentence constituent unit; and a search input sentence in the second language and a search target sentence. Second similarity calculation means for calculating the similarity for each fixed sentence constituent unit between the first and second similarity calculation means, and the first language or the second similarity calculation means based on the calculation result of the first or second similarity calculation means. An optimum combination calculation means for obtaining a combination of sentence constituent units that are most similar to a language; and when a search input sentence of the first language is input from the input means, the storage means for the search input sentence The first language from the set of sentences stored in A first search unit that applies the similarity calculation unit and the optimum combination calculation unit to perform a search, and a sentence in a second language that is a parallel translation of the result of the search target sentence searched by the first search unit. As a new search input sentence, a search is performed by applying the second similarity calculation unit and the optimum combination calculation unit for a second language from a set of sentences stored in the storage unit, with respect to the search input sentence. And a search result determination unit that determines whether or not the search result obtained by the second search unit is included in the search result obtained by the first search unit. Then, the document search apparatus characterized in that the search result determined by the search result determination means is prioritized as a search result.