JP3447955B2 - Machine translation system and machine translation method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation system and a machine translation method for automatically performing translation.

[0002]

2. Description of the Related Art In recent years, with the downsizing of computers and the spread of the Internet, it has become possible for anyone to easily transmit and collect information through a homepage or the like. For example, about 80% of homepages are homepages in foreign languages including English, and general users are more and more likely to come into contact with foreign languages. Also, in various other situations, the number of cases requiring translation is rapidly increasing.

Under such circumstances, there is an increasing demand for translation systems, and many translation systems have been developed, sold and distributed in order to meet such demand. However, many conventional translation systems have been developed for technical documents. For example, non-grammatical sentences in which verbs and articles are omitted, such as title sentences that frequently appear in news, etc. Or it was difficult to translate properly.

[0004]

Conventionally, when a sentence to be translated has a grammatically special form, it has been difficult to accurately or properly translate such a sentence. The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a machine translation system and a machine translation method capable of more accurately or appropriately translating a sentence having a grammatically special form. And

[0005]

A machine translation system according to the present invention comprises input means for inputting an original sentence to be translated,
A translation unit that translates the original sentence to generate a translated sentence, a first extracting unit that extracts an incomplete sentence from the original sentence, and a second extracting unit that extracts a similar sentence similar to the incomplete sentence from the original sentence. And a complementing unit that generates a translated sentence for the incomplete sentence by using the similar sentence, and the complementing unit configures a tree structure generated by performing a predetermined structural analysis on the similar sentence. A node generated from nodes having a phrase corresponding to a phrase included in the incomplete sentence to a root node among the nodes to be used as a translated sentence for the incomplete sentence. To do. Further, the machine translation system according to the present invention includes an input unit for inputting an original sentence to be translated, a translation unit for translating the original sentence to generate a translated sentence, and a first extraction for extracting an incomplete sentence from the original sentence. Means, second extracting means for extracting a similar sentence similar to the incomplete sentence from another predetermined document described in the same language as the original sentence, and a translated sentence for the incomplete sentence using the similar sentence Complementing means for generating, the complementing means, among the nodes constituting the tree structure generated by performing a predetermined structural analysis on the similar sentence,
It is characterized in that a sentence generated by using nodes from a node having a phrase corresponding to a phrase included in the incomplete sentence to a root node is a translated sentence for the incomplete sentence.

Further, the machine translation system according to the present invention is
An input unit for inputting an original sentence to be translated, a translation unit for translating the original sentence to generate a translated sentence, a first extracting unit for extracting an incomplete sentence from the original sentence, and an incomplete sentence from the original sentence. Second extraction means for extracting similar similar sentences;
Complementary means for generating a translated sentence for the incomplete sentence using the similar sentence, wherein the complementing means, among the nodes constituting the tree structure of the concept dependent structure data generated for the similar sentence, It is characterized in that a sentence generated by using nodes from a node having a phrase corresponding to a phrase included in the incomplete sentence to a root node is a translated sentence for the incomplete sentence. Further, the machine translation system according to the present invention includes an input unit for inputting an original sentence to be translated, a translation unit for translating the original sentence to generate a translated sentence, and a first extraction for extracting an incomplete sentence from the original sentence. Means, and second extracting means for extracting a similar sentence similar to the incomplete sentence from another predetermined document described in the same language as the original sentence,
Complementary means for generating a translated sentence for the incomplete sentence using the similar sentence, wherein the complementing means, among the nodes constituting the tree structure of the concept dependent structure data generated for the similar sentence, It is characterized in that a sentence generated by using nodes from a node having a phrase corresponding to a phrase included in the incomplete sentence to a root node is a translated sentence for the incomplete sentence.

Preferably, the first extracting means may extract a sentence satisfying a predetermined condition as an incomplete sentence. Preferably, the incomplete sentence may be a title sentence.

[0008] Preferably, the second extracting means, among the sentences included in the original sentence, includes the largest number of phrases corresponding to phrases other than the function words included in the incomplete sentence as the similar sentence. You may make it extract.

[0009]

Preferably, only the translated sentence of the incomplete sentence generated by the complementing means is stored, or the translated sentence of the incomplete sentence generated by the translating means and the incomplete sentence generated by the complementing means. The translation of the sentence may be stored together.

Preferably, an incomplete sentence is extracted from the original sentence before the translation by the translation means, and if the incomplete sentence is extracted, the incomplete sentence may not be translated. Good.

[0012] Preferably, the apparatus further comprises means for outputting the translated sentence, and in order to inform that the translated sentence of the incomplete sentence generated by the complementing means is the sentence generated by the complementing means. The data may be added and output.

Preferably, the apparatus further comprises means for presenting the translated sentence of the incomplete sentence generated by the translating means and the translated sentence of the incomplete sentence generated by the complementing means at the same time or in sequence. May be.

Further, the machine translation method according to the present invention includes a step of inputting an original sentence to be translated, a step of storing the input original sentence in an original sentence storage means, and a translation means for translating the original sentence into a translated sentence. A step of generating, a step of storing the generated translated sentence in a translated sentence storage means, a step of extracting an incomplete sentence from the original sentence by a first extracting means, and a step of extracting an incomplete sentence from the original sentence by a second extracting means. And a complementing step of generating a translated sentence for the incomplete sentence by using the similar sentence by a complementing means, in the complementing step, the complementing means includes the similar sentence. Among the nodes that make up the tree structure generated by performing a predetermined structural analysis on the The statement generated using node up over preparative node, characterized in that the translation for the incomplete sentence. Further, the machine translation method according to the present invention includes a step of inputting an original sentence to be translated, a step of storing the input original sentence in an original sentence storage means, and a step of translating the original sentence by the translation means to generate a translated sentence. A step of storing the generated translated text in a translated text storage means;
Extracting the incomplete sentence from the original sentence by the extracting means, inputting another predetermined document described in the same language as the original sentence, and second extracting means from the other predetermined document by the second extracting means. There is a step of extracting a similar sentence similar to an incomplete sentence, and a complementing step of generating a translated sentence for the incomplete sentence by using a similar sentence by a complementing means, and in the complementing step, the complementing means, Of the nodes forming the tree structure generated by performing a predetermined structural analysis on the similar sentence, the nodes from the node having the phrase corresponding to the phrase included in the incomplete sentence to the root node are used. The sentence generated by the above is used as a translated sentence for the incomplete sentence.

Further, the machine translation method according to the present invention includes the steps of inputting an original sentence to be translated, storing the input original sentence in an original sentence storage means, and translating the original sentence by the translation means to obtain a translated sentence. A step of generating, a step of storing the generated translated sentence in a translated sentence storage means, a step of extracting an incomplete sentence from the original sentence by a first extracting means, and a step of extracting an incomplete sentence from the original sentence by a second extracting means. And a complementing step of generating a translated sentence for the incomplete sentence by using the similar sentence by a complementing means, in the complementing step, the complementing means includes the similar sentence. Among the nodes forming the tree structure of the concept-dependent structure data generated for, the node having the word corresponding to the word contained in the incomplete sentence is selected as the root node. A sentence node generated using the up, characterized in that the translation for the incomplete sentence.
Further, the machine translation method according to the present invention includes a step of inputting an original sentence to be translated, a step of storing the input original sentence in an original sentence storage means, and a step of translating the original sentence by the translation means to generate a translated sentence. A step of storing the generated translated sentence in the translated sentence storage means, a step of extracting an incomplete sentence from the original sentence by the first extracting means, and inputting another predetermined document described in the same language as the original sentence. And a step of extracting a similar sentence similar to the incomplete sentence from the other predetermined document by the second extracting means, and a translated sentence for the incomplete sentence using the similar sentence by the complementing means. Completing step, wherein in the complementing step, among the nodes forming the tree structure of the concept-dependent structure data generated for the similar sentence, Were generated using the node to the root node from the node with the phrase corresponding to the words contained in the full text sentence, characterized in that the translation for the incomplete sentence.

Further, according to the present invention, a computer is caused to input an original sentence to be translated, the inputted original sentence is stored in an original sentence storage means, and the original sentence is translated by the translation means to generate a translated sentence, and the translated sentence is generated. The translated sentence is stored in the translated sentence storage unit, the first extracting unit extracts the incomplete sentence from the original sentence, and the second extracting unit extracts the similar sentence similar to the incomplete sentence from the original sentence, and complements the sentence. When generating the translated sentence for the incomplete sentence by using the similar sentence by the means and generating the translated sentence by the complementing means, a tree generated by performing a predetermined structural analysis on the similar sentence. Among the nodes forming the structure, the sentence generated by using the nodes from the node having the phrase corresponding to the phrase included in the incomplete sentence to the root node is A computer-readable recording medium recording a program for causing a translation for a statement. Further, the present invention causes a computer to input an original sentence to be translated, store the input original sentence in an original sentence storage means, translate the original sentence by the translation means to generate a translated sentence, and generate the translated sentence. Is stored in the translated sentence storage means, an incomplete sentence is extracted from the original sentence by the first extracting means, another predetermined document described in the same language as the original sentence is input, and the other is written by the second extracting means. When a similar sentence similar to the incomplete sentence is extracted from the predetermined document, the complementing unit uses the similar sentence to generate a translated sentence for the incomplete sentence, and the complementing unit generates the translated sentence. From among the nodes forming the tree structure generated by performing the predetermined structural analysis on the similar sentence, the node having the phrase corresponding to the phrase included in the incomplete sentence The statement generated using node up over preparative node is a computer-readable recording medium recording a program for causing a translation for the incomplete sentence.

Further, according to the present invention, a computer is made to input an original sentence to be translated, the inputted original sentence is stored in an original sentence storage means, and the original sentence is translated by the translation means to generate a translated sentence. The translated sentence is stored in the translated sentence storage unit, the first extracting unit extracts the incomplete sentence from the original sentence, and the second extracting unit extracts the similar sentence similar to the incomplete sentence from the original sentence, and complements the sentence. When generating the translated sentence for the incomplete sentence by using the similar sentence by the means, and generating the translated sentence by the complementing means, the tree structure of the concept-dependent structure data generated for the similar sentence is configured. Among the nodes to be generated, the sentence generated by using the nodes from the node having the phrase corresponding to the phrase included in the incomplete sentence to the root node is A computer-readable recording medium recording a program for a sentence. Further, the present invention causes a computer to input an original sentence to be translated, store the input original sentence in an original sentence storage means, translate the original sentence by the translation means to generate a translated sentence, and generate the translated sentence. Is stored in the translated sentence storage means, and the incomplete sentence is extracted from the original sentence by the first extracting means,
The other predetermined document described in the same language as the original sentence is input, the second extracting unit extracts a similar sentence similar to the incomplete sentence from the other predetermined document, and the complementing unit extracts the similar sentence. While generating a translated sentence for the incomplete sentence by using, and generating the translated sentence by the complementing means, among the nodes constituting the tree structure of the concept-dependent structure data generated for the similar sentence, A computer-readable recording medium recording a program for converting a sentence generated using nodes from a node having a phrase corresponding to a phrase included in the incomplete sentence to a root node into a translated sentence for the incomplete sentence. Is.

It should be noted that the present invention relating to the apparatus also holds as the invention relating to the method, and the present invention relating to the method also holds as the invention relating to the apparatus. Further, the present invention relating to an apparatus or a method is provided for causing a computer to execute a procedure corresponding to the present invention (or for causing a computer to function as means corresponding to the present invention, or for a computer to have a function corresponding to the present invention. It also holds as a computer-readable recording medium in which a program (for realizing it) is recorded.

According to the present invention, a translated sentence of a sentence that is similar to an incomplete sentence (a sentence that is not an incomplete sentence) is generated by using an original sentence or a translation result, and a more accurately translated sentence is generated. It is possible to improve the translation accuracy of incomplete sentences. As a result, the readability of the translated text can be improved, which contributes to quicker understanding of the text by the user.

[0020]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a configuration example of a machine translation system according to an embodiment of the present invention. As shown in FIG. 1, the machine translation system includes an input unit 101, a translation unit 102, a control unit 103, an incomplete sentence recognition complementing unit 104, and an output unit 105.

The input unit 101 is for translating documents and
It is for inputting an instruction or the like from the user. For inputting a document to be translated, for example, a keyboard (key input), a recording medium driving device (reading from a recording medium such as a magnetic disk, a magnetic tape, or an optical disk), OCR (optical reading), a network connection device (communication) Can be used). Also, any of a plurality of types of input forms may be usable. When OCR is used, a function of recognizing a read character image and generating a code string is used.

A keyboard, a mouse or the like is used for inputting an instruction (eg, a translation instruction) from the user. The control unit 103 causes the input unit 1 to input a document to be translated, sends the input document to the translation unit 102 for processing, and sends the translation result by the translation unit 102 to the incomplete sentence recognition complementing unit 104 for processing. The entire translation system is controlled by sending the translation result and the like obtained by the translation unit 102 and the incomplete sentence recognition complementation unit 104 to the output unit 105 for output.

The translation unit 102 translates the document data passed from the control unit 103 and returns the obtained result to the control unit 103. The incomplete sentence recognition complementing unit 104 receives the translation result by the translation unit 102 from the control unit 103, recognizes the incomplete sentence in the input document, and if the incomplete sentence is extracted, the incomplete sentence is extracted. The optimization process is performed and the result is returned to the control unit 103.

Here, the "incomplete sentence" corresponds to, for example, a sentence composed only of noun phrases, a sentence that is not grammatically correct, or a sentence with high ambiguity. What kind of incomplete sentence is targeted by the system can be appropriately determined. As a method of detecting an incomplete sentence, for example, a condition that characterizes an incomplete sentence that is a target in the system is set in advance, and for each sentence of the input text, it is checked whether the sentence satisfies the above condition, and the condition is determined. If the sentence is satisfied, the sentence may be regarded as an incomplete sentence.

In the optimization process, a sentence similar to the incomplete sentence (hereinafter referred to as a similar sentence) is extracted from the input document, and this similar sentence is used to determine that it is more appropriate for the incomplete sentence. Generate the expected translation. The similar sentence is preferably selected from sentences that are not incomplete sentences.

The output unit 105 is for outputting a translated sentence or the like. As the output form of the translated text, for example,
Various forms are conceivable, such as outputting only a translated sentence or outputting an input document and its translated sentence together.

As the output unit 105, an optical display (presented to the user), a printer (presented to the user), a network connection device (output to a printer connected via the network, transfer to another computer via the network, etc.) ), A recording medium driving device (storing in a recording medium such as a magnetic disk, a magnetic tape, or an optical disk) can be used. Further, it is possible to use arbitrary plural kinds of output forms.

In the following, the present embodiment will be described in more detail using a specific example. In this embodiment, the case where an English sentence is translated into a Japanese sentence will be described as an example. In addition, in the present embodiment, “title sentence” will be used as an example of an incomplete sentence.

FIG. 2 shows an example of an English text (original text) obtained by the input unit 101. This input text is
It consists of multiple sentences, the first sentence is "ANDERS"
ON HEADS FOR MIDDLE EAS
T "and the second sentence is" U. S. Secretary
of State Tom Anderson set
out on his first Middle
East trip Tuesday. ", The third sentence is" He said he hopped to ...
", And so on.

The input text obtained by the input unit 101 is passed to the translation unit 102 under the control of the control unit 103. The translation unit 102 translates the input text, and translates the translation result (parallel correspondence between the original sentence and the translated sentence).
In a format of As a translation method in the translation unit 102, a general method can be used.

Further, the translation unit 102 returns the analysis result of each sentence (for example, the analysis result by syntax analysis) to the control unit 103 in the form of concept dependent structure data together with the above-mentioned parallel translation data. The concept-dependent structure data may be obtained in the process of translation processing, or may be obtained by performing analysis separately from the translation processing.

FIG. 3 shows an example of parallel translation data, and FIG. 4 shows an example of concept-dependent structure data obtained by syntax analysis. Note that the pointer to the concept-dependent structure data provided for each sentence in FIG. 3 indicates a link to the concept-dependent structure data in FIG. 4 corresponding to the sentence.

When the translation unit 102 generates the parallel translation data and the concept-dependent structure data, these data are then controlled by the control unit 103 to complete the incomplete sentence recognition complementation unit 104.
Passed to.

In the incomplete sentence recognition complementing unit 104, the title sentence (that is, the incomplete sentence in this example) is selected from the parallel translation data.
Is detected, the detected title sentence is subjected to optimization processing to correct the corresponding translated sentence, and then returned to the control unit 103.
Hereinafter, the processing in the incomplete sentence recognition complementing unit 104 will be described in detail.

FIG. 5 shows an example of the processing procedure of the incomplete sentence recognition complementing unit 104. First, in step S402, a title sentence is detected from the input bilingual data.

As a method of detecting the title sentence, for example,
One or more combinations of language phenomena often found in title sentences such as "the original sentence is written in all capital letters" and "the sentence is only a noun phrase in the first line" may be used. If the input text is written in a hypertext description language such as HTML, the title sentence will contain
Such additional information may be used because it is tagged or has a larger character size than other sentences.

Here, assuming that the sentence in which all the characters are written in capital letters is recognized as the title sentence, the sentence of sentence number 1 in the input text of FIG. 3, that is, "ANDERSON HEADS FOR MID"
"DLE EAST" is recognized as a title sentence and taken out.

The retrieved title sentence is translated by the translation unit 102.
The original form of the word is restored by performing the morphological analysis used in. In this example, "Anderson",
"Head", "for", "middle", "ea"
st ”is extracted. A general method can be used for this morphological analysis.

Next, in steps S403 to S408,
For sentences other than the title sentence, the similarities with the title sentence are examined, and the most similar sentence is set as the title similar sentence. The degree of similarity with the title sentence can be evaluated by, for example, a method such as how many original forms of words constituting the title sentence are included in each sentence. Here, as an example, the degree of similarity includes the number of original forms of words excluding function words (prepositions, etc.) among the words included in the title sentence.

First, the first sentence is extracted in step S404, but since it is a title sentence, nothing is done in step S405 and the similarity is set to 0. Next, in step S407,
The next sentence is extracted in S404, and the similarity with the title sentence is calculated in step S405. In this embodiment,
Since the degree of similarity is defined as how many original forms of words excluding functional words are included in the sentence included in the title sentence, morphological analysis is performed on the extracted sentence to obtain the original form of the word.

Next, the words are "under
"son", "head", "middle", "eas"
It is checked whether they match t ", and the number of matches is written in the similarity column of the parallel translation data in FIG. 3 as the similarity.

After writing the similarity, step S406.
If it is not the last sentence, the process proceeds to the next sentence in steps S407 and S404, and in step S4.
At 05, the degree of similarity is calculated again.

If it is the last sentence, step S408.
Then, the sentence having the highest degree of similarity of the bilingual data is set as the title similar sentence, and the process proceeds to the title reflection process of step S409. FIG. 6 shows an example of parallel translation data in which information on the degree of similarity is added to the example of FIG.

Here, in step S408, the sentence of sentence number 2 in FIG. 6, that is, "US Secreta" is displayed.
ry of State Tom Anderson
set out on his first Medd
le East trip on Tuesday. ”
Is selected as the title similar sentence.

When a plurality of similar sentences having the maximum degree of similarity are obtained, one similar sentence is selected based on another predetermined criterion. For example, a method that selects the one that has the closest position (closest sentence number) to the incomplete sentence, a method that selects the largest number of displayable nodes described later, a method that selects the largest number of words, etc. Various methods are possible.

Next, in the title reflection process of step S409, the translated sentence of the title-similar sentence is processed so as to be appropriate as a title, the bilingual translation data of FIG. 6 is corrected, and this is returned to the control unit 103, and the process is executed. finish.

FIG. 7 shows an example of the procedure of the title reflection processing in step S409. First, in step S602, the syntactic structure of the title-similar sentence is extracted from the concept-dependent structure data using the concept-dependent structure pointer of the bilingual data in FIG. Although the concept-dependent structure obtained by the syntactic analysis is used here as the structure analysis result, the result obtained by another analysis method such as semantic analysis may be used.

FIG. 8 shows an example of the concept-dependent structure of the title-similar sentence obtained from the parallel translation data of FIG. Next, in steps S603 to S607, the node (displayable node) to be displayed is selected from the nodes forming the concept-dependent structure, and the displayable flag is set.

First, in step S603, the variable p is set to 1
And take the p-th word from the beginning of the title sentence,
Search for that is included as a node of the title similar sentence. If it is included, in step S604, the parse tree is traced from that node to display all the nodes up to the root node, and the process proceeds to step S606. On the other hand, if it is not included, nothing is done and the process proceeds to step S606.

Next, in step S606, it is determined whether the p-th word is the end-of-sentence word, and if it is not the end-of-sentence,
The variable p is incremented by 1, and the process is repeated from step S604. On the other hand, if it is the end of the sentence, step S608.
Proceed to.

In the example of FIGS. 6 and 8, first, in step S604, "ANDERSON" is selected,
In the concept-dependent structure of FIG. 8, "Anderson" which is a translation of "ANDERSON" can be displayed. This state is shown in FIG.

Next, by the processing of step S605, up to the root node can be displayed. This state is shown in FIG. Next, in steps S606 and S607, the next word is viewed, but since "HEAD" does not appear in the title-similar sentence, nothing is done, and steps S606 and S607 are executed.
The processing proceeds to 607 and S604.

By repeating the above processing, FIG.
A tree structure as shown by is generated. Next, in step S608, the translated sentence of the title sentence (that is, incomplete sentence) is reconstructed using the displayable node of the generated tree structure.

The sentence generation process used in the translation unit 101 may be used to reconstruct the translated sentence. A general method can be used for this sentence generation processing. Next, the translated sentence in the title sentence of the bilingual data in FIG. 6 is rewritten with the generated translated sentence, and the process ends.

In the example of FIG. 11, "Tom Anderson departed for the Middle East trip" is generated as the translated sentence, the translated sentence of the bilingual data in FIG. 6 is rewritten with this translated sentence, and the process ends.

FIG. 12 shows an example of rewritten parallel translation data. When the parallel translation data is corrected by the incomplete sentence recognition complementing unit 104, the data to be output is passed to the output unit 105 under the control of the control unit 103, and the output unit 105 outputs the data.

According to the present embodiment, a sentence similar to an incomplete sentence is used to generate a translated sentence of the incomplete sentence, a more accurately translated sentence can be obtained, and the readability of the translated sentence is improved. be able to.

In the above description, the data is transferred to each unit via the control unit. However, actual data may be transferred, or pointer information such as an address where the actual data is stored may be transferred. You may do it.

Further, in the above description, the translated sentence of the title-similar sentence is used so as to be suitable as a title, but the translated sentence of the title-similar sentence may be used as it is as the title.
Further, in the above, the concept dependent structure data is generated for all sentences, but if the concept dependent structure data is not obtained in the process of translation, but is obtained by performing analysis separately from translation, The concept-dependent structure data may be generated only for the “similar sentence” that is the basis for generating the translated sentence of the incomplete sentence in the incomplete sentence recognition complementing unit.
In this case, the incomplete sentence recognition complementing unit may generate concept dependent structure data.

In the above description, the translated sentence of the incomplete sentence by the translation unit is rewritten with the translated sentence obtained by the incomplete sentence recognition complementing unit, but both the translated sentence of the former and the latter sentence may be stored together. Good. Alternatively, the incomplete sentence may be first extracted from the input document, and the incomplete sentence may not be translated by the translation unit, and only the translated sentence may be generated by the incomplete sentence recognition complementing unit.

When both the translated sentence of the incomplete sentence by the translation unit and the translated sentence of the incomplete sentence by the incomplete sentence recognition complementing unit are stored, the translated sentence as the translation result is output to the outside. For, it is possible to output both translated sentences at the same time. Further, for example, when displaying a translated sentence on an optical display, for an incomplete sentence, first, the translated sentence by the incomplete sentence recognition complementing unit is presented, and the translated sentence by the translation unit is presented when an instruction from the user is given. Or both translations can be presented at the same time.

Further, in the above, the translated sentence of the incomplete sentence is generated by the incomplete sentence recognition complementing unit using the similar sentence most similar to the incomplete sentence, but some similar sentences similar to the incomplete sentence are generated. A plurality of translated sentences may be generated by utilizing each.

When outputting the translated text data,
Information that informs that the translation of the incomplete sentence was obtained by the incomplete sentence recognition complementation unit (for example, data such as characters that express that fact, or the characters that make up the translated sentence). Character decoration information such as color (a value that allows the recognition unit to distinguish the translated text), or the like may be added. Further, information such as the content of the similar sentence or the sentence number that is the source of the translated sentence generation may be added.

In the above description, a similar sentence similar to an incomplete sentence is extracted from the same input document and is used to generate a translated sentence of the incomplete sentence. Instead of the document containing the incomplete sentence, or together with the document containing the incomplete sentence, or when there is no similar sentence in the document containing the incomplete sentence, the document containing the incomplete sentence. Use different documents, eg, texts that have been translated and stored in the system in the past, existing bilingual corpus, such as one or more other texts when multiple texts are to be translated. You may A document different from the document containing the incomplete sentence can be read from the input unit, for example.

Further, in the present embodiment, the title sentence is treated as an example of the incomplete sentence, but of course the present invention is not limited to this.
Various other forms can be treated as incomplete sentences, for example, a sentence consisting of only noun phrases can be treated as an incomplete sentence.

It is also possible to handle a plurality of types of incomplete sentences. For example, the "title sentence" and the "sentence consisting only of other noun phrases (for example, headline sentence)" may be incomplete sentences. In this case, depending on the type of incomplete sentence,
The contents of the optimization process may be different.

Further, the types of incomplete sentences that can be handled by the system are presented to the user, and only one or more types of incomplete sentences selected by the user are generated by the incomplete sentence recognition complementing unit to generate translated sentences. May be performed.

In the above description, the case where the original sentence is an English sentence and the translated sentence is a Japanese sentence has been described as an example. It can also be applied when translating between English and other languages.

Although one embodiment of the present invention has been described above, each function described above can be realized by using hardware or software. Further, the machine translation system according to the present embodiment can be realized by using a general-purpose computer or a dedicated machine.

Further, the present embodiment is a computer recording a program for causing a computer to execute a predetermined procedure (or for causing the computer to function as a predetermined means or for causing the computer to realize a predetermined function). It can also be implemented as a readable recording medium. The present invention is not limited to the above-described embodiments, but can be implemented with various modifications within the technical scope thereof.

[0071]

According to the present invention, a sentence that is similar to an incomplete sentence can be used to generate a translated sentence of the incomplete sentence, and a more accurately translated sentence can be obtained, thus improving readability of the translated sentence. Can be made.

[Brief description of drawings]

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

FIG. 2 is a diagram showing an example of input text.

FIG. 3 is a diagram showing an example of parallel translation data.

FIG. 4 is a diagram showing an example of concept-dependent structure data.

FIG. 5 is a flowchart showing an example of a procedure of incomplete sentence recognition complementing processing.

FIG. 6 is a diagram showing an example of parallel translation data with similarity added.

FIG. 7 is a flowchart showing an example of a procedure of title reflection processing.

FIG. 8 is a diagram showing an example of a concept dependency structure of a title similar sentence.

FIG. 9 is a diagram showing an example of a concept dependency structure of a title similar sentence.

FIG. 10 is a diagram showing an example of a concept dependency structure of a title similar sentence.

FIG. 11 is a diagram showing an example of a concept dependency structure of a title similar sentence.

FIG. 12 is a diagram showing an example of finally rewritten bilingual data.

[Explanation of symbols]

101 ... Input section 102 ... Translation Department 103 ... Control unit 104 ... Incomplete sentence recognition complementing unit 105 ... Output unit

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References Maki Murata and Makoto Nagao, Completion of verb abbreviations using surface expressions and examples, IPSJ Research Report 96-SLP-14-4, Japan, December 1996. 13th, Vol. 96, No. 123, p. 23-p. 30 Maki Murata, Makoto Nagao, Complementation of verb abbreviations using surface expressions and examples, IEICE Technical Report SP96-72, Japan, December 12, 1996, Vol. 96, No. 421, p. 23-p. 30 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/28

Claims (19)

(57) [Claims] 1. Input means for inputting an original sentence to be translated, translating means for translating the original sentence to generate a translated sentence, first extracting means for extracting an incomplete sentence from the original sentence, and from the original sentence second extracting means for extracting a similar sentence similar to the incomplete sentence, the example Bei and complementing means for generating a translation for said utilizing similar sentence incomplete sentence, said complementary means, the similar sentence Predetermined structural analysis for
Of the nodes that make up the tree structure generated by applying
Of these, have a phrase that corresponds to the phrase included in the incomplete sentence.
Generated using nodes from one node to the root node
A machine translation system, wherein the translated sentence is a translated sentence for the incomplete sentence . 2. An input unit for inputting an original sentence to be translated, a translation unit for translating the original sentence to generate a translated sentence, a first extracting unit for extracting an incomplete sentence from the original sentence, the original sentence, Second extracting means for extracting a similar sentence similar to the incomplete sentence from another predetermined document described in the same language; and complementing means for generating a translated sentence for the incomplete sentence using the similar sentence Bei example, said complementary means, the predetermined structural analysis with respect to the similar sentence
Of the nodes that make up the tree structure generated by applying
Of these, have a phrase that corresponds to the phrase included in the incomplete sentence.
Generated using nodes from one node to the root node
A machine translation system, wherein the translated sentence is a translated sentence for the incomplete sentence . 3. Input means for inputting an original sentence to be translated
When extracts a translation means for generating a translated sentence by translating the original text, first extracting means for extracting an incomplete sentence from the original sentence, a similar sentence similar to the incomplete sentence from the original sentence
Generate a translated sentence for the incomplete sentence using the second extraction means and the similar sentence.
Complementing means for performing the concept-dependent generation generated for the similar sentence.
Of the nodes that make up the tree structure of existing structure data,
From a node that has a phrase corresponding to a phrase included in a complete sentence
The sentence generated using the nodes up to the root node is
Characterized by being a translation of an incomplete sentence
Machine translation system. 4. Input means for inputting an original sentence to be translated
When a translation means for generating a translated sentence by translating the original text, before the first extraction means and other predetermined document written in the source and the same language to extract an incomplete sentence from the original sentence
Second extracting means for extracting a similar sentence similar to the incomplete sentence
And generate a translated sentence for the incomplete sentence using the similar sentence.
Complementing means for performing the concept-dependent generation generated for the similar sentence.
Of the nodes that make up the tree structure of existing structure data,
From a node that has a phrase corresponding to a phrase included in a complete sentence
The sentence generated using the nodes up to the root node is
Characterized by being a translation of an incomplete sentence
Machine translation system. 5. The first extracting means extracts a sentence satisfying a predetermined condition as an incomplete sentence, according to any one of claims 1 to 4 . Machine translation system. 6. A machine translation system according to any one of claims 1 to 4, wherein the incomplete sentence is a title sentence. 7. The second extraction means extracts, as the similar sentence, one of the sentences included in the original sentence, which includes the largest number of phrases corresponding to phrases other than the functional words included in the incomplete sentence. The invention according to claim 1 or 2 , characterized in that
4. The machine translation system according to any one of 4 above. 8. A translation sentence of the incomplete sentence generated by the complementing means is stored, or a translated sentence of the incomplete sentence generated by the translating means and the incomplete sentence generated by the complementing means. The machine translation system according to any one of claims 1 to 4, wherein the machine translation system further stores the translated text. 9. An incomplete sentence is extracted from the original sentence prior to the translation by the translation means, and when the incomplete sentence is extracted, the incomplete sentence is not translated by the translation means. The machine translation system according to any one of claims 1 to 4 . 10. A means for outputting the translated text, further comprising: for informing that the translated text of the incomplete sentence generated by the complementing means is the sentence generated by the complementing means. The machine translation system according to any one of claims 1 to 4, wherein data is added and output. 11. A means for presenting the translated text of the incomplete sentence generated by the translating means and the translated text of the incomplete sentence generated by the complementing means at the same time or in order. Claims 1 to 4 characterized by
The machine translation system according to item 1 . 12. A step of inputting an original sentence to be translated.
And a step of storing the input original text in the original text storage means
And a step of translating the original text to generate a translated text by the translation means.
And a step of storing the generated translated text in the translated text storage means.
When, to extract the incomplete sentence from the original sentence by the first extraction means
A step, a step of extracting a similar sentence similar to the incomplete sentence from the original by the second extracting means, and a complementary step of generating translation for the incomplete sentence by using the similar sentence by complementing unit And in the complementing step, the complementing means adds to the similar sentence.
A tree generated by applying a predetermined structural analysis to
Of the nodes that make up the structure, included in the incomplete sentence
From the node that has the word corresponding to the word to the root node
The sentence generated by using the node in
A machine translation method characterized in that it is translated. 13. A step of inputting an original sentence to be translated.
And a step of storing the input original text in the original text storage means
And a step of translating the original text to generate a translated text by the translation means.
And a step of storing the generated translated text in the translated text storage means.
When, to extract the incomplete sentence from the original sentence by the first extraction means
Input step and other specified document described in the same language as the original text
And a step of extracting a similar sentence similar to the incomplete sentence from the other predetermined document by the second extracting means, and a translation sentence for the incomplete sentence using the similar sentence by the complementing means. Complementing step of
A tree generated by applying a predetermined structural analysis to
Of the nodes that make up the structure, included in the incomplete sentence
From the node that has the word corresponding to the word to the root node
The sentence generated by using the node in
A machine translation method characterized in that it is translated. 14. A step of inputting an original sentence to be translated.
And a step of storing the input original text in the original text storage means
And a step of translating the original sentence by the translation means to generate a translated sentence.
And a step of storing the generated translated text in the translated text storage means.
And an incomplete sentence is extracted from the original sentence by the first extracting means.
Step and second extraction means to resemble the original sentence to the incomplete sentence
A similar sentence is extracted, and a complementary means is used to match the incomplete sentence using the similar sentence.
And a complementary step of generating a translated sentence.
Construct a tree structure of concept-dependent structural data generated for
Node that corresponds to the phrase included in the incomplete sentence.
From the node that has the phrase to the root node
The sentence generated by using the sentence can be used as a translation for the incomplete sentence.
A machine translation method characterized by. 15. A step of inputting an original sentence to be translated
And a step of storing the input original text in the original text storage means
And a step of translating the original sentence by the translation means to generate a translated sentence.
And a step of storing the generated translated text in the translated text storage means.
And an incomplete sentence is extracted from the original sentence by the first extracting means.
Input step and other specified document described in the same language as the original text
And a second extraction means for extracting the incompleteness from the other predetermined document.
A step of extracting a similar sentence similar to the whole sentence;
And a complementary step of generating a translated sentence.
Construct a tree structure of concept-dependent structural data generated for
Node that corresponds to the phrase included in the incomplete sentence.
From the node that has the phrase to the root node
The sentence generated by using the sentence can be used as the translated sentence for the incomplete sentence.
A machine translation method characterized by. 16. An original sentence to be translated is input to a computer, and the input original sentence is stored in an original sentence storage means.
The translation means translates the original text to generate a translated text, and the generated translated text is stored in the translated text storage means.
Incomplete sentences are extracted from the original sentence by the extraction unit 1 of
To extract similar sentence similar to the incomplete sentence from the original by the second extraction means and by using the similar sentence to produce a translation for the incomplete sentence by complementing unit together
In generating the translated text by the complementing means,
Is obtained by performing a predetermined structural analysis on the similar sentence.
Of the nodes that make up the tree structure generated by
From a node that has a phrase corresponding to a phrase included in a complete sentence
The sentence generated using the nodes up to the root node is
A computer-readable recording medium in which a program for translating an incomplete sentence is recorded. 17. An original sentence to be translated is input to a computer, and the input original sentence is stored in an original sentence storage means.
The translation means translates the original text to generate a translated text, and the generated translated text is stored in the translated text storage means.
Incomplete sentences are extracted from the original sentence by the extraction unit 1 of
Input another specified document described in the same language as the original text
Then, the second extraction unit extracts a similar sentence similar to the incomplete sentence from the other predetermined document, and the complementary unit extracts the similar sentence .
Wherein together to produce translation for said utilizing similar sentence incomplete statements Ri, when to generate the translation by the complementing unit
Is obtained by performing a predetermined structural analysis on the similar sentence.
Of the nodes that make up the tree structure generated by
From a node that has a phrase corresponding to a phrase included in a complete sentence
The sentence generated using the nodes up to the root node is
A computer-readable recording medium in which a program for translating an incomplete sentence is recorded. 18. An original sentence to be translated is input to a computer, and the input original sentence is stored in an original sentence storage means.
The translation means translates the original text to generate a translated text, and the generated translated text is stored in the translated text storage means.
Incomplete sentences are extracted from the original sentence by the extraction unit 1 of
To extract similar sentence similar to the incomplete sentence from the original by the second extraction means and by using the similar sentence to produce a translation for the incomplete sentence by complementing unit together
In generating the translated text by the complementing means,
Is the concept-dependent structural data generated for the similar sentence.
Of the nodes that make up the tree structure , included in the incomplete sentence
Root node from the node that has the phrase corresponding to
A sentence generated using nodes up to
A computer-readable recording medium in which a program for making a translation is recorded. 19. An original sentence to be translated is input to a computer, and the input original sentence is stored in an original sentence storage means.
The translation means translates the original text to generate a translated text, and the generated translated text is stored in the translated text storage means.
Incomplete sentences are extracted from the original sentence by the extraction unit 1 of
Input another specified document described in the same language as the original text
Then, the second extraction unit extracts a similar sentence similar to the incomplete sentence from the other predetermined document, and the complementary unit extracts the similar sentence .
Wherein together to produce translation for said utilizing similar sentence incomplete statements Ri, when to generate the translation by the complementing unit
Is the concept-dependent structural data generated for the similar sentence.
Of the nodes that make up the tree structure, included in the incomplete sentence
Root node from the node that has the phrase corresponding to
A sentence generated using nodes up to
A computer-readable recording medium in which a program for making a translation is recorded.