JP2003108553A - Machine translation device, machine translation method and machine translation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve translation accuracy even in e-mail and a Web notice board few in information obtained from an individual document in translation using context information. SOLUTION: The application 10 outputs text data to a related document information acquiring part 6 via a control part 2. The related document information acquiring part 6 makes access to a document data storage part 9, acquires reference series document data related to the text data as reference document information, and outputs the data to a translation environment setting part 7. The translation environment setting part 7 stores information for facilitating a translation among the reference document information in a translation environment storage part 8 as translation environment information. A translation part 3 translates the text data by using knowledge of a dictionary part 4 and the translation environment information of the translation environment storage part 8. The reference document information includes information on a plurality of reference series document data related to the text data, and even when information required for the translation is few in the text data, the translation environment information for facilitating the translation can be obtained by using the reference document information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device, a machine translation method, and a machine translation program suitable for translating between a first language and a second language in an electronic mail transmitting / receiving tool, a Web browser, and the like. .

[0002]

2. Description of the Related Art In recent years, the popularity of personal computers has been high, and as the Internet environment has been improved, e-mail and We
The exchange of information with foreign countries using b (world wide web) sites has become popular. When exchanging information with foreign countries, it is necessary to consider language differences. The translation accuracy of machine translation software has been improved, and it is also possible to call machine translation software from e-mail software or Web browser software to perform translation processing. Also, the number of sites providing translation services on the Web is increasing.

By the way, in order to improve the translation accuracy,
A translation method using context information may be adopted.
In this document translation technology, the entire document is analyzed in advance to extract various information, the extracted contents are used for translation, and the information of the translated sentence is used for translation of the later sentence while the document is being translated. The method is adopted.

However, in the case of electronic mail,
Since the length of one sentence is short and each sentence is often in a simple style, it is difficult to obtain sufficient information for proper translation even when trying to obtain information from the entire email.

Further, recently, a bulletin board system on the Web is sometimes used as a place for exchanging information on the Internet. The document size handled by this bulletin board system is also comparatively small, and an extremely short article exists in a series of exchanges for a common topic. In addition, as a bulletin board configuration, there is also a type of bulletin board in which individual articles are listed in time series (hereinafter referred to as sequential display), instead of a threaded display that collectively displays interactions with related topics. In the case of such a sequential display type, a plurality of topics are randomly and independently described in one page,
It is extremely difficult to extract only the information that is suitable for translating articles having individual relationships.

[0006]

As described above, in the past, when sending and receiving electronic information such as electronic mails and Web bulletin boards, there is often little information available for translation processing in individual documents, and context information is available. There is a problem that sufficient translation accuracy cannot be obtained even if the translation method using is applied.

The present invention utilizes the characteristics of electronic information exchange such as electronic mail and electronic bulletin boards to extract information that can be used for translation processing from a document having a relationship other than the document to be translated, thereby performing translation. An object is to provide a machine translation device, a machine translation method, and a machine translation program that can improve accuracy.

[0008]

A machine translation apparatus according to claim 1 of the present invention accesses a document data holding means in which a plurality of document data is stored, and relates to the original text data among the plurality of document data. Related document information acquisition means for acquiring document data of a reference sequence having as reference document information,
A translation environment setting unit that obtains translation environment information that can be used for translation of the original text data from the reference document information obtained by the related document information obtaining unit and stores the translation environment information in the translation environment holding unit, a first language and a second language. The present invention further comprises a translation means for translating the original text data by utilizing a dictionary having knowledge information necessary for translating and translation environment information held by the translation environment holding means. The machine translation method according to Item 13 accesses a document data holding unit in which a plurality of document data is stored, and acquires reference series of document data among the plurality of document data as reference document information that is related to the original text data. Related document information acquisition process,
A translation environment setting process for obtaining translation environment information that can be used for translation of the original text data from the reference document information obtained in the related document information obtaining process, and causing the translation environment holding unit to hold the translation environment information, a first language and a second language. The present invention comprises a translation process for translating the original text data by using a dictionary having knowledge information necessary for translating and translation environment information held by the translation environment holding means. The machine translation program according to Item 14 accesses a document data holding unit in which a plurality of pieces of document data are stored in a computer, and refers to the reference series of document data among the plurality of pieces of document data as a reference document. The related document information acquisition process acquired as information, and the translation available from the reference document information acquired in the related document information acquisition process for translation of the original text data. A translation environment setting process for acquiring environment information and retaining it in the translation environment retaining means, a dictionary having knowledge information necessary for translation between the first language and the second language, and translation environment information retained in the translation environment retaining means. Is used to execute a translation process for translating the original text data.

In claim 1 of the present invention, the document data holding means stores a plurality of document data. The related document information acquisition means acquires, as reference document information, reference series of document data among the plurality of pieces of document data stored in the document data holding means, which is related to the original text data. As a result, the reference document information includes a plurality of document data of the reference series that are related to the original text data, and even if the original text data has a small amount of information necessary for improving the translation accuracy, the reference document information has the translation accuracy. It contains a lot of information needed to improve The translation environment setting means acquires the translation environment information that can be used to translate the original text data from the acquired reference document information, and causes the translation environment holding means to hold the translation environment information. The dictionary has knowledge information necessary for translating the first language and the second language.
The translation means translates the original text data using the dictionary and the held translation environment information.

In the thirteenth aspect of the present invention, the document data of the reference series which is related to the original text data among the plurality of document data stored in the document data holding means is acquired as the reference document information. Translation environment information that can be used to translate the original text data is acquired from the acquired reference document information, and is stored in the translation environment holding means. During the translation process of the original text data, not only the dictionary having the knowledge information necessary for the translation between the first language and the second language but also the translation environment information held in the translation environment holding means is used, whereby the translation accuracy is improved. improves.

In the fourteenth aspect of the present invention, the computer is made to access the document data holding means in which a plurality of document data is stored, and the document data of the reference series having a relation with the original text data among the plurality of document data is referred to. Related document information acquisition processing acquired as document information and translation environment setting processing for acquiring translation environment information that can be used for translation of original text data from the reference document information acquired in the related document information acquisition processing and holding it in the translation environment holding means. And first language and second
Using a dictionary having knowledge information necessary for translation with a language and the translation environment information held by the translation environment holding means, a translation process for translating the original text data is executed.

[0012]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing a machine translation device according to an embodiment of the present invention.

The present embodiment collects a chain (document set) of mutually related documents from a plurality of documents having a small amount of information contained in each document. By using the information contained in each document in the chain to perform translation, the translation accuracy is improved.

In the present embodiment, for example, an example of application to the exchange of electronic information via electronic mail or the exchange of electronic information using a website will be described.

In FIG. 1, the machine translation apparatus includes a control section 2 for controlling the entire apparatus, an input section 1 and an output section 5 for interfacing with the outside for input and output, and an input obtained via the input section 1. A translation unit 3 for translating a document, a dictionary unit 4 that stores various translation knowledge used by the translation unit 3 for translation, a related document information acquisition unit 6, a translation environment setting unit 7, and a translation environment storage unit 8. .

The machine translation device is activated by various applications 10. The configuration of each unit of the machine translation apparatus in FIG. 1 can be realized by a program. In this case, the machine translation device corresponds to a machine translation program, and the various applications correspond to application programs that activate the machine translation program. The application program may have a machine translation program incorporated therein. For example, the application program includes an electronic mail transmission / reception program, an electronic bulletin board program, etc., and is not limited to a stand-alone software,
It may be a program activated on a web browser.

The document data storage unit 9 stores a plurality of document data. For example, the document data storage unit 9 stores electronic information transmitted and received by electronic mail software, electronic information transmitted and received by Web access, and the like. That is, the document data storage unit 9 has only to be accessible by the application 10, and may be configured by, for example, a hard disk device on the Internet.

The application 10 causes the document data storage unit 9 to store the document data to be transmitted / received via a predetermined transmission path (not shown), and reads and uses the document data stored in the document data storage unit 9.

When the application 10 is an electronic mail transmission / reception program, information such as the location and structure in the document data storage unit 9 for storing the document data to be processed is stored in advance in the related document information acquisition unit 6 which will be described later. Register Further, when the application 10 is software or the like that uses a bulletin board system on the Web, basically, the storage location can be acquired from the source information of the page, so there is no need to register in advance. As a result, the application 10 can reliably access each document data in the document data storage unit 9.

In the present embodiment, the related document information acquisition unit 6 is controlled by the control unit 2 as will be described later, and information (corresponding to each document constituting each chain from the translation target) is associated with each other. Reference relationship information) is acquired. Further, the related document information acquisition unit 6 accesses the document data storage unit 9, collects chains based on the reference relation information, and acquires each document data in the chain as reference document information.

The related document information acquisition unit 6 supplies the acquired reference document information to the translation environment setting unit 7. Translation environment setting section 7
As will be described later, extracts information useful for translation processing (hereinafter referred to as translation environment information) from the reference document information and outputs the extracted translation environment information to the translation environment storage unit 8. The translation environment storage unit 8 stores the translation environment information extracted and acquired by the translation environment setting unit 7.

The input unit 1 supplies the translation target document sent from the application 10 to the control unit 2. The control unit 2
When the translation target document is input, the related document information acquisition unit 6 is caused to execute the reference document information acquisition process.

The information sent from the application 10 may be the document itself consisting of character codes, or only the document name and storage location information. Even if only the document name and the storage location information are input, the control unit 2 can acquire the original text information itself composed of character data from the document data storage unit 9 via the related document information acquisition unit 6. In addition, even when the document itself including the character code is sent from the application 10, the storage location information (or file information) of the original data may be received together with the character code data, or based on the character data. The original data location may be searched through the related document information acquisition unit 6.

When the reference document information acquisition processing is executed, the related document information acquisition unit 6 refers to the information in the translation environment storage unit 8 so as not to acquire the previously acquired information again. The translation environment storage unit 8 returns an end signal to the control unit 2 when the storage of the translation environment information is completed.

When the control unit 2 receives a signal indicating that the translation environment setting process (storage of translation environment information) is completed,
The document to be translated is sent to the translation unit 3 to be translated. The translation unit 3 uses the translation knowledge and the translation environment information stored in the dictionary unit 4 to convert the translation target document described in the first language into a sentence in the second language to control the control unit 2
Output to. The control unit 2 is adapted to send the translation result from the translation unit 3 to the application 10 via the output unit 5.

The dictionary unit 4 stores the translation knowledge necessary for the translation process of the translation unit 3. For example, assuming that the translation unit 3 performs bilingual translation processing between English and Japanese, the dictionary unit 4 has the following configuration for each translation direction.

4-1. Basic dictionary part (a) Vocabulary part Utilization information, meaning information, for each vocabulary of the first language,
Translation information, etc.

(B) Morphological analysis rule Knowledge for morphological analysis of input sentences in the first language.

(C) Syntax / Semantic Analysis Rule Knowledge for performing syntactic / semantic analysis after morphological analysis of the input sentence of the first language.

(D) The conversion rule syntax / semantic analysis result is the second semantic structure of the first language.
Knowledge to translate into the semantic structure of language.

(E) Syntax generation rule Knowledge for generating a word string of the second language from the semantic structure of the second language.

(F) Morphological Generation Rule Knowledge for outputting the final translated sentence, which reflects the use of words in the second language.

4-2. Technical term dictionary section A plurality of field-specific dictionaries are prepared, and the dictionary used for translation can be selected according to the contents of the input document. It consists only of vocabulary.

4-3. User dictionary section A dictionary for user definition. The user can create a plurality according to the document to be translated. There are (a) and (b) for each.

(A) Vocabulary section Utilization information, meaning information, translated word information, etc. regarding terms newly registered by the user.

(B) Translated word learning unit Stores information on translated words learned by the user with respect to a certain phrase in the first language.

The dictionary unit 4 may be configured to include a translation pattern dictionary in which original sentence patterns and translation patterns are accumulated.

Next, the operation of the embodiment thus constructed will be described with reference to the explanatory view of FIG. 2 and the flow charts of FIGS. FIG. 2 shows an example of storage of information in the translation environment storage unit 8.

Now, assuming that the application 10 is an electronic mail transmission / reception program, one or more users who use the first language and one or more users who use the second language receive a plurality of times for a predetermined title. It is assumed that electronic information is exchanged. In addition, an application 10 is provided for the document data storage unit 9 that stores these electronic information (e-mail transmission data and reception data).
Can be accessed by an operation of a user who uses the first language (hereinafter, simply referred to as a user).

The application 10 supplies the original text data of the first language by the user to the control section 2 from the input section 1 as a translation target document (translation original text). For example, when the application 10 is an e-mail transmission / reception program, by embedding a program for calling the execution of translation processing in this program, the user can perform a predetermined operation on the program to send the original text data to be transmitted. It is possible to execute a translation process.

When the original text data is input via the input unit 1, the control unit 3 activates the related document information acquisition unit 6. FIG. 3 shows the flow of processing from the input of the original text data from the application 10 to the control unit 2 to the output of the translated text. In step S301 of FIG. 3, the related document information acquisition unit 6 analyzes the information attached to the original text data and acquires the reference relation information.

Now, it is assumed that the original text data input via the input unit 1 is composed of data of a set of a document to be translated and auxiliary information. The attached information is the header information of the email, subject information, sender, destination,
Corresponds to various information such as Cc destination. In this case, the reference relationship information needed to collect the document chain is:
The subject information, sender, destination, Cc destination, message ID, etc. can be considered. Since these pieces of information are incorporated in the original text data in a fixed format, they can be easily obtained by searching a character string, for example.

Further, when replying an electronic mail, a sentence used last time is often quoted, and a tag indicating a quoted sentence is often added to the text. Therefore, the related document information acquisition unit 6 can also determine that the tag indicating the quote and the content of the quote are the same as the reference series document data.

When the analysis and information acquisition in step S301 are completed, the related document information acquisition unit 6 next refers to the translation environment storage unit 8 in step S302, and the translation original text in the input data has already been translated. Check whether the document is registered as a translation environment by referring to the reference document. If the identification information of the original translation is already in
If it is included in the MessageID, the process is step S3.
The process moves from 03 to step S306 to execute the translation process.

On the other hand, if the identification information of the translated original sentence is not included in any of the MessageIDs of any of the categories, the reference document acquisition process for this translated original sentence has not been performed. Therefore, the process proceeds to step S304. , Obtains reference document information.

That is, the related document information acquisition unit 6 uses the reference relation information acquired in step S301 as a basis, and refers to the document referred to when the input translation source text is created and the translation source text after the source text is created. A document created by referring to the translated original sentence is acquired from the document data storage unit 9. In this case, if there is a document further referred to by the document acquired as the reference document information, the related document information acquisition unit 6 traces back to the starting point or the end point of the document chain (reference sequence) to acquire the document.

The following method can be considered as a method of acquiring the reference document information by the related document information acquisition unit 6.

That is, the same subject and the items of the subject to which "Re:", "RE:", etc. are added are searched and obtained in a time-series manner in the sent folder, the receiving folder, and other storage folders. There is a way. If the message ID is complete in the program, the reference sequence can be acquired using this message ID as a key. Every time one mail is obtained by tracing the reference sequence, the Message in any of the categories in the translation environment storage unit 8
It is checked whether or not it matches the ID, and if there is a match, it can be determined that the destination has already been referred to in the past, and the reference data acquisition processing ends at this point.
However, when a new reference sequence is found in time series, only the detected reference document information is added without omission.

Further, even if the message ID does not match, the process can be terminated in the following cases. For example, even when the sender, the receiver, and the Cc destination are clearly discriminated from each other even in the same subject, the reference sequence detection processing is terminated by regarding it as a sequence delimiter. Further, even if the mail is sent by the same program, but the In-Reply-to information is lost, it can be considered as the starting point of another series, so the search ends. Other than this, it is possible to devise a search method as appropriate using various information.

When the acquisition of the reference document is completed, the control unit 2 shifts the processing to step S305 and activates the translation environment setting unit 7. The translation environment setting unit 7 uses the data of the acquired reference document information to determine the translation environment suitable for the translation original sentence.

FIG. 4 specifically shows the process of step S305 in FIG. First, the translation environment setting unit 7 extracts content words in the entire reference document information in step S401. Next, the translation environment setting unit 7 uses the content word extracted in step S402 to estimate the content (field) of the series of documents. For this estimation process,
The methods detailed in the reference “Dictionary Management Method and Dictionary Management System and Translation Method and Translation System” can be applied. Note that the estimation process described in this document defines a set of related words for each specific field, and estimates the field of the document based on the matching status between the defined word and the extracted word. Is adopted.

When the field estimation process is completed, the translation environment setting unit 7 shifts the process to step S403 to select a technical term dictionary to be used at the time of translation. As this method, a correspondence table indicating which technical term dictionary should be used according to the field estimated in step S402 is prepared,
The selection should be made accordingly. Table 1 below shows an example of this correspondence table.

[0053] When the selection of the specialized dictionary is completed, the translation environment setting unit 7 proceeds to the next step S404, where steps S304, S401 to S40 are performed.
Processing for reflecting the information obtained in 3 in the translation environment storage unit 8 is performed. In this case, if one of the categories in the translation environment storage unit 8 matches the reference sequence of the translation original sentence and is already described, it is added to it, and if it is new, a new category is added. To be done. The content words extracted in step S401 are listed in the "content word E" and "content word J" columns, respectively, depending on whether they are English documents or Japanese documents. In the case of a new item, subject information is described in Subject. The referenced message identification information is listed in MessageID. The estimated field information is written in the “field” column, and the type name of the selected technical term dictionary is written in the “used technical dictionary” column.

FIG. 2 shows the translation environment information stored by the update processing of the translation environment storage unit 8 in step S404. In Figure 2, "Subject:", "Message ID:",
“Field:”, “specialized dictionary for use:”, “content word E:” and “content word J:” indicate the following contents, respectively.

Subject: In the present embodiment, the common subject is one document set (chain).
Recognize as. Here, the subject name is stored.

Depending on the method of implementation, the reference relationship information may be obtained from other than the subject name (when the application is an electronic bulletin board, etc.). In that case, a plurality of subject names are stored.

Message ID: Document identification information managed by the application. Depending on the embodiment, it may be a number or a file name.

Field: in which field the translation original sentence was judged by the translation unit 3 at the time of the previous translation, or
The translation environment setting unit 7 stores information about which field the reference document information is determined to be a document.

Specialized dictionary to be used: The special dictionary automatically selected by the translation unit 3 based on which dictionary the user specified to use at the time of the previous translation, or the field information in the translation environment storage unit 8 immediately before translation. Information about what the term dictionary was, or what kind of specialized dictionary corresponds to the field determined by the translation environment setting unit 7 for the reference document information is stored.

Content word E: In the reference document history, a list of content words originally generated in the document written in English is stored. In the example of FIG. 2, only simple enumeration is performed, but frequency information may be added and used for controlling the use in the translation unit 3 later.

Content word J: A list of content words originally generated in a document written in Japanese is stored in the reference document history. Same as content word E except for language differences.

When the updating of the translation environment storage unit 8 is completed, the processing shifts to step S306 in FIG. 3 and the translation unit 3 is activated. FIG. 5 shows a specific flow of step S306 of FIG.
Shows the translation process of. For the translation of the entire input translation source text, the flow of FIG. 5 is repeated until the end of the document.

When the translation source sentence is sent from the control unit 2, the translation unit 3 first determines in step S501 whether the translation source sentence is the first sentence of a new block. When the application 10 is an electronic mail transmission / reception program, one block corresponds to one message. Therefore, in this case, only the first sentence of the message is determined as the "new block". If it is the first sentence of the block, the translation unit 3 proceeds to step S502 to read the translation environment information of the corresponding section in the translation environment storage unit 8.
For the sentences other than the beginning of the block, the same translation environment information as the environment used in the previous sentence is used without newly reading the translation environment information. In this way, control is performed so that the translation environment information corresponding to each block is used.

When the translation environment is ready, in the next step S503, the translation unit 3 uses the following knowledge of the dictionary unit 4 to perform dictionary lookup and morphological analysis processing.

[0065] In addition, 4-2. The type of dictionary used in the technical term dictionary section is the dictionary set in the translation environment information. User dictionary 4-3. With respect to the above, since it is not associated with the translation environment information, a dictionary specified separately by the user should be used. Even in this case, by providing an item of the user dictionary as one of the items stored in the translation environment storage unit 8,
The translation accuracy can be further improved.

Next, in step S504, the translation unit 3 4-1. (c) The syntax and semantic analysis of the input source text is performed using the syntax and semantic analysis rules. At this stage, the structure of the analysis result of the input language is constructed. Next, the translation unit 3 shifts the processing to step S505, and 4-1. (a) Vocabulary part, (d) Conversion rule, 4-2. Vocabulary section, 4-3. Using the knowledge of (a) vocabulary part and (b) translated word learning part, processing for converting the structure of the input language into the structure of the output language is performed. At this stage, the translation of the input word is determined along with the linguistic structure of the output language.

In the present embodiment, when determining the translated word, the information of the field name, the technical term dictionary, and the content words E and J set in the translation environment information is used by the following method.

* If a compound word or the like registered in the technical term dictionary but not in the basic dictionary appears in the input text of the technical term dictionary, step S50.
In step 3, the registered words in the technical term dictionary are looked up, and their translated words are selected here. Also, when there is a word in the input sentence that is registered in both the basic dictionary and the technical term dictionary,
The translation terms defined in the technical term dictionary are adopted. In this way, the translation will be performed in a translation word that matches the specialized field. (The same applies to user dictionaries.) * Some translations in the field name basic dictionary have field information.
An example of "base" is shown below. In the field of baseball, it is translated as "base", in the field of military it is translated as "base", and in other cases it is translated as "basic". Therefore,
The same headword can be translated into different translations according to the field set in the translation environment information.

(Example) base: Basic, base (field = baseball), base (field = military) * Content words E and J Now, suppose that a document with MessageID = 000015 is being translated. In that document, "The program featured wor
Suppose there is a sentence "ks by Picasso." If the registered contents of “work” in the basic dictionary section are as follows, the first definition “work” is usually adopted as the translation of “work”. However, when listing the words in the content word J as translation environment information, there is the word "work", which is "wor
Matches one of the translation definition groups of "k". In such a case, the matched translated word is preferentially used and controlled. On the other hand, when translating a document with MessageID = 000010, “work” in the content word J is prioritized with respect to the same “work”. This will help maintain consistent translations through email and message exchanges through translations.

(Example) work: work, work, labor, work When the structure of the output language is constructed in this way, the translation unit 3 shifts the processing to step S506, and 4-1. (e) Convert a two-dimensional structure into a one-dimensional structure using the syntax generation rule. Finally, in step S507, the translation unit 3 4-
1. (f) The surface morphology of each word is generated using the morpheme generation rule, and the final translated sentence is output.

The translated text output from the translation unit 3 is the control unit 2
Sent to. The control unit 2 outputs the translated text to the application 10 via the output unit 5 in step S307 of FIG. In this way, the translated sentence in the second language with respect to the translated original sentence can be obtained on the electronic mail transmission / reception program which is the application 10.

Next, the operation will be described when the application 10 is a bulletin board system on the Web. To simplify the description, only the differences based on the different applications will be described. It should be noted that a user who uses the first language is assigned to the application 1 by using the document data storage unit 9 on the Internet that stores electronic information (document data displayed on the bulletin board) of the bulletin board system.
It is assumed that 0 is used for access.

The first difference is the additional information analysis process in step S301 of FIG. 3, and the second difference is the reference document information acquisition process in step S304 of FIG. 3, which is the third difference. The point is that blocks are considered in step S501.

First, the first difference will be described.

On the bulletin board on the Web, it is conceivable that the translation object and the document stored in the document data storage unit 9 are html source files. In this case, various structural information such as tags other than the original text data embedded in the source file corresponds to the attached information.

By the way, the structure of the bulletin board differs for each Web site. For example, there is a bulletin board in which posting messages are listed in a time series on one page and a link button for reply is not prepared (hereinafter referred to as “type A”). There is also a bulletin board in a form (hereinafter, referred to as "type B") in which information indicating which message a particular message is a reply to is displayed with a link. Furthermore, there is a bulletin board in which the reference relationship is expressed using a branching tree or indentation (hereinafter referred to as "type C"). These types are identified in advance in step S301 by the related document information acquisition unit 6 determining how to use the table in the page.

In the "type A" bulletin board, each message is managed by a number, and it is usual to take a display form such that a message number referred to a subject name is added as information indicating a reply relationship. is there. Therefore, in this case, the related document information acquisition unit 6 performs step S301.
In, the reference message number, subject, poster, etc. of each message are acquired as reference relationship information.

For the "type B" bulletin board, the related document information acquisition unit 6 determines in step S301 the location of the data that is the source of the reply, the file name, etc. in addition to the subject of the message and the poster information. get. The following display method (Pattern A) that is usually used in bulletin boards
1, A2) is stored and the pattern is input
By performing a search in the ml file, the data that is the source of the reply can be detected. It should be noted that these patterns can be configured so that the user can easily add them.

Pattern A1: This is ($ variable) 's ($
This is a reply to (variable_reference) Pattern A2: This comment is a reply to ($ variable_reference);; Similarly, the following pattern B1 is also used for detecting a message created by referring to the message. You can get messages while following the link relations.

Pattern B1: See reply to this message;; For the "type C" bulletin board, it is possible to easily detect the display portion of the reference relation information by analyzing the html format structure. The following display example is an example of the display. Often expressed using a table format, the subject part is usually a link to the actual message data. From these, at the simplest level, the usage of "Re", "RE", "re", etc. in the subject is used as a key, and the location of the message data in the referenced relationship and the file name etc. are acquired as reference relationship information. do it. Furthermore,
By analyzing the structure such as the nesting relation in the table,
Even if the subject name is changed, the reference / reference relationship can be detected.

Example> No. Speaking order ;; 5 5 About purchasing software Ichiro Hanada 2001-06-10 6 6 About collecting user opinions Hiroshi Suzuki 2001-06-11 7 ・ Re: About collecting user opinions Taka Sato 2001-06-11 9 ・ ・ Re: Collection of user opinions Tomoko Yamada 2001-06-13 7 8 Warning display Satoru Inoue 2001-06-12 ;; For both of these types, the input is the address of the text data, In the case where only the location and the file name are present, it is necessary to start the process from the place where the original data is acquired by accessing the document data storage unit 9 before analyzing the attached information.

Next, the second difference, that is, how to search for reference document information will be described.

In the case of a type A Web bulletin board,
Since there is a possibility that a document belonging to the reference sequence exists in the page, first, the reference sequence is searched in the page for each message on the page. The search on the page is completed,
If there is no message (document) belonging to the reference sequence, the previous page is referred to. In addition, in order to detect the data of the previous page, you can easily find the data of the previous page by searching the place where the character string such as "previous page" or "next page" is a button or link in html. Can be detected. When the data on the previous page is acquired, similarly, the message that matches the message number (reference message number) of the reference sequence is sequentially searched, and the reference document information is acquired in sequence while tracing to the starting point of the reference sequence.

In the case of the type B bulletin board, the data linked to the "$ variable_reference" of the part found by the search of the pattern is acquired, and similarly, until the reference disappears, the message Get reference series.

In the case of the type C bulletin board, information on a set of referenced / referenced messages can be acquired from the structure of the display portion of the reference relationship, so the messages in the acquired information can be sequentially acquired. .

Similar to types A, B and C, the subsequent reference data acquisition processing is omitted when a message in which the MessageID has been described is found in any of the categories of the translation environment storage unit 8.

Next, the third difference, that is, the case where the input translated document is composed of a plurality of blocks will be described.

In the present embodiment, the bulletin board system defines a block as a set of messages that are normally expressed in a group using a table format. A document handled by a bulletin board system on the Web may be composed of a plurality of blocks like a type A bulletin board.
In this case, the translation unit 3 moves the process from step S501 to step S502 if the sentence is the head sentence of the block, and reads the translation environment information of the corresponding section in the translation environment storage unit 8. Except for the beginning of the block, the same translation environment information as the environment used in the previous sentence is used without newly reading the translation environment information. In this way, when there are a plurality of messages on one page, it is possible to control to use each translation environment information.

In the case of the type B bulletin board, the entire input document may be regarded as one block, and whether or not it is the first sentence may be determined in step S501.

Other operations are the same as when the application 10 is a mail transmission / reception program, and the translation environment information is set for each block of the message using the information of the acquired reference document, and the translation is performed based on the translation environment information. Process and output the translated document.

As described above, in the present embodiment, the reference relationship information corresponding to the translation target is used to detect the related reference document information from the document data storage unit, and the translation is assisted based on the reference document information. The translation environment information is set, and the translation process is performed using the set translation environment information, so that the translation accuracy can be improved. As a result, even when translating an e-mail of a relatively short document or a message on an electronic bulletin board, it is possible to use the reply history and the like to obtain translation environment information for assisting translation from the document in the history and effectively use it. it can. Further, even in a bulletin board system in which a plurality of topics are posted in parallel on one page, it is possible to obtain a translation result with high translation accuracy by using the translation environment information suitable for each topic.

FIG. 6 is a flow chart showing a translation environment setting process adopted in another embodiment of the present invention. 6, the same steps as those in FIG. 4 are designated by the same reference numerals, and the description thereof will be omitted.

The present embodiment is different only in the processing of the translation environment setting unit 7 from the embodiment of FIG. The translation environment setting unit 7 creates bilingual terms in step S412 of FIG. That is, the translation environment setting unit 7 obtains an optimal translated word as a translated word of the collected words in the chain by the translation process using the context information for the entire reference document information or a part thereof.

FIG. 7 shows the contents of the translation environment storage unit 8 updated in step S404 of FIG. In FIG. 7, description of the same storage contents as in FIG. 2 will be omitted.

In FIG. 7, the parallel translation term is step S412.
It was acquired in. A typical means for determining the optimum second language translation for a certain word in the first language is 4-
1 (d) There are some conversion rules. For example, “plane” has two different meanings, “airplane” and “plane”. In the 4-1 (d) conversion rules, a large number of rules for determining translation as follows are accumulated as knowledge for deciding which translation is appropriate.

Plane (subject) + bomb / crash / explode (main verb) → plane = airplane (Meaning: used as the subject of bomb, crash, explode
(Plane is translated as “airplane”) By selecting one of these rules and determining the translation, it is possible to create parallel translation information “plane = airplane”.

The bilingual terms in FIG. 7 are determined to be optimal as the translated words of the documents in the chain that belong to the original translation. Therefore, the translation unit 3 gives the highest priority to the translation term of FIG.

Other configurations and operations are similar to those of the embodiment shown in FIG.

As described above, FIG. 1 is also used in this embodiment.
It is possible to obtain the same effect as that of the above embodiment.

The information used in the above embodiments is merely an example, and any information can be used without departing from the scope of the claims. Further, in accordance with the detailing of the data information handled by the application program, various forms can be used for the means for acquiring the history information. Also,
The above embodiments do not limit the detailed specifications of acquisition. Further, although the application program is assumed to be the electronic mail transmission / reception program or the electronic bulletin board program in the above-mentioned embodiment, it can be applied to any communication program for exchanging natural language sentence data through translation.

That is, the present invention can be applied to the exchange of messages through translation. In this case, since the pair of words in the first language and the second language is stored as the translation environment information, the translation is not performed. As a result, it becomes easier to avoid the problem that the conversation becomes like a completely different meaning.

[0102]

As described above, according to the present invention, it is possible to utilize the characteristic of electronic information exchange such as electronic mail and electronic bulletin board for the translation process from the related document other than the translation target document. By extracting such information, the translation accuracy can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a machine translation device according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a storage example of information in a translation environment storage unit 8.

3 is a flowchart showing a flow of processing from input of original text data from an application 10 to output of a translated text in a control unit 2 in FIG.

4 is a flowchart specifically showing a translation environment setting process in step S305 in FIG.

FIG. 5 is a flowchart specifically showing the translation process of step S306 in FIG.

FIG. 6 is a flowchart showing a translation environment setting process adopted in another embodiment of the present invention.

7 is an explanatory diagram showing the contents of a translation environment storage unit 8 updated in step S404 of FIG.

[Explanation of symbols]

2 ... control unit, 3 ... translation unit, 4 ... dictionary unit, 6 ... related document information acquisition unit, 7 ... translation environment setting unit, 8 ... translation environment storage unit,
9 ... Document data storage unit, 10 ... Application.

Claims (14)

[Claims] 1. Related document information for accessing, as reference document information, document data holding means in which a plurality of document data is stored, and reference sequence document data having a relation with original text data among the plurality of document data is acquired. An acquisition unit, a translation environment setting unit that acquires translation environment information that can be used to translate the original text data from the reference document information acquired by the related document information acquisition unit, and stores the translation environment information in the translation environment holding unit; A translation means for translating the original text data using a dictionary having knowledge information necessary for translation into the second language and the translation environment information held in the translation environment holding means. Machine translation device. 2. The related document information acquisition means obtains reference relationship information that correlates document data of the reference series by analyzing the original text data, and uses the reference relationship information to obtain the reference document. The machine translation apparatus according to claim 1, wherein the machine translation apparatus acquires information. 3. The machine translation device according to claim 2, wherein the related document information acquisition means obtains the reference relation information by analyzing information attached to the original text data. 4. The reference relationship information is at least one of subject information, a sender, a destination, a Cc destination, and a message ID when the original text data is transmission / reception data of an electronic mail. The machine translation device according to claim 2. 5. When the original text data is data posted on a bulletin board on the web, the reference relationship information includes a reference message number of each message, a subject, a contributor, and the location of the data that is the source of the reply. The machine translation device according to claim 2, wherein the machine translation device is at least one of file names. 6. The machine translation apparatus according to claim 1, wherein the document data of the reference sequence is document data of a response regarding the same content. 7. The related document information acquisition unit refers to the translation environment information stored in the translation environment storage unit, and does not acquire the reference document information corresponding to the registered translation environment information. The machine translation device according to claim 1. 8. The machine translation apparatus according to claim 1, wherein the related document information acquisition unit acquires the document data of the reference series in the order of creation. 9. The machine translation device according to claim 1, wherein the related document information acquisition unit acquires the document data of the reference sequence in a reverse order of the creation time. 10. The dictionary includes a plurality of pieces of knowledge information necessary for translation, and the translation environment information includes information for selecting knowledge information to be used for translation from a plurality of pieces of knowledge information included in the dictionary. The machine translation device according to claim 1, wherein: 11. The translation environment holding means stores the phrase of the first language and the phrase of the second language included in the reference document information of the reference sequence as translation environment information, and the translation means The machine translation apparatus according to claim 1, wherein the translation environment information is used when determining a translation word. 12. The translation environment holding means includes the first language and the second language included in the reference document information of the reference sequence.
2. The machine translation device according to claim 1, wherein the translation means stores a translation term as a translation environment information, and the translation unit uses the translation term when determining a translation word. 13. Related document information for accessing, as reference document information, document data holding means in which a plurality of document data is stored, and reference sequence document data having a relation with original text data among the plurality of document data is acquired. An acquisition process, a translation environment setting process for acquiring translation environment information that can be used for translation of the original text data from the reference document information acquired in the related document information acquisition process, and causing the translation environment holding unit to hold the translation environment information; A translation process for translating the original text data by using a dictionary having knowledge information necessary for translation into a second language and the translation environment information held in the translation environment holding means. Machine translation method. 14. A computer accesses a document data holding means in which a plurality of document data is stored, and acquires, as reference document information, document data of a reference series having a relation with original text data among the plurality of document data. A related document information acquisition process, a translation environment setting process for acquiring translation environment information that can be used for translation of the original text data from the reference document information acquired in the related document information acquisition process, and holding the translation environment information in a translation environment holding unit; To execute a translation process for translating the original text data by using a dictionary having knowledge information necessary for translation into one language and a second language and the translation environment information held in the translation environment holding means. Machine translation program.