JPH034364A - Machine translating system - Google Patents

Abstract

PURPOSE:To execute the selection of a translation word more easily by using a co-occurence data in the limited capacity of a memory by generating and storing the co-occurence data in automatic learning type in the working process of edit, etc. CONSTITUTION:When a user selects an appropriate translation word out of plural translation word candidates by using a translation word selection key in an input part 1, it is recognized with an edit control part 9. Next, the grammatical relation of the translation word to be selected with another translation word is detected at a grammatical relation judging part 11, and the acquirement of coincidence with set grammatical relation is judged. When it is judged that coincidence is obtained, the control part 9 sends the information of the translation word to be selected and that of another translation word to a storage part 12 after making them to a pair. In such a way, the pair of the translation word to be selected and another translation word can be stored as the cogenerative data.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) The present invention relates to a machine translation system, and particularly to a machine translation system having a function of facilitating the selection of translated words.

(Prior Art) Machine translation systems that automatically perform translation processing using computer technology are being actively developed. In conventional machine translation systems, when multiple translation candidates exist for the same source word, a method is adopted in which the user selects an appropriate translation candidate from among them. When there are a large number of translation candidates, it is sufficient if the translation desired by the user is displayed first, but if this is not the case, the other translation candidates must be displayed one after another by operating the next candidate key, which takes time to select. It takes.

Therefore, we focused on the ease of semantic combination of two words, created so-called co-occurrence data that pairs two words that are easy to combine, and prepared a co-occurrence data that accumulates a large number of them. When multiple translation candidates occur, methods are being considered to prioritize and display the ones that are co-occurring or to automatically select them.

If there is no corresponding word pair in the co-occurrence, the procedure is as usual.

By using such a method of using co-occurrence, for example, if you register co-occurrence data pairing "hotel" and "boy" in co-occurrence, you can register the co-occurrence data that pairs "hotel" and "boy" in co-occurrence.
It is possible to display "Boy", which gives the greatest possibility of being related to "Hotel", at the top from among "Boy", etc., or to automatically select "Boy".

In the conventionally considered method of using co-occurrence, the co-occurrence must be stored in the machine translation system in advance. Here, if the number of words registered in the dictionary is 100,000 words, the pair of two words is simply calculated as 100,000 words x 100,000 words -
It will be 10 billion bears. Among these, there are far fewer co-occurring relationships, but it is thought that there are still millions to thousands of blade bears. It is almost impossible to register all such many pairs in advance as co-occurrence data in co-occurrence data.

By the way, dictionaries in machine translation systems are assumed to be used by an unspecified number of users, so
It is necessary to register a large number of words, 10,000 words, but the number of words actually used is far smaller, only around 10,000 to 20,000, if limited to -10,000 users or one department. I know that.

However, just as a dictionary targets an unspecified number of people, if co-registration is prepared in advance, it would have to target an unspecified number of people. This makes it difficult to collect co-occurrence data and create a co-occurrence cloth, and at the same time requires a huge amount of memory, which is not practical.

(Issue 8 to be solved by the invention) As mentioned above, in the conventional method of using co-occurrence for machine translation, there is a limit to the number of co-occurrence data that can be registered in memory as co-occurrence. In terms of meaning, there was a problem in that the effect of facilitating the selection of translation words was small.

An object of the present invention is to solve such problems and provide a machine translation system that can more easily select translated words using co-occurrence data with limited memory capacity.

[Structure of the Invention] (Means for Solving the Three Problems) In order to achieve the above-mentioned problems, the present invention collects co-occurrence data while a user is performing work such as human input and editing of the original text for machine translation. It is characterized by being created and memorized automatically.

That is, the machine translation system of the present invention determines whether a selected translation word selected from a plurality of translation word candidates and other translation words in the translated text have a specific grammatical relationship during machine translation, and Information indicating each of the selected translation word and another translation word that has been determined to have a grammatical relationship is stored as a set of co-occurrence data.

In addition, more simply, information indicating a uniquely determined selected translation word selected from a plurality of translation word candidates during machine translation, and information indicating other uniquely determined translation words are all stored. Either the grammatical relations between the two are stored as co-occurrence data, or the information representing the selected translation word and the information representing at least one of the translation words immediately before and after the selected translation word in the translation are stored in a grammatical manner. They may be grouped and stored as co-occurrence data regardless of the physical relationship.

(Operation) In this way, in the present invention, co-occurrence data is created and recorded in the process of machine translation processing, thereby accumulating co-occurrence co-occurrence, so there is no need to create co-occurrence co-occurrence in advance.

The co-pronunciations accumulated in this way are different from conventional ones prepared for an unspecified number of users, as they reflect the results of learning the word usage trends of one or a few specific users. , the selection of translation words becomes easier.

In addition, the amount of money used by a particular user is uneven, and it is normal for it to be within the tens of thousands of words, so even though the amount of co-occurrence data that can be accumulated as a co-occurrence table is very small. , which has a great effect in making it easier to select translation words.

(Example) Hereinafter, the present invention will be described in detail with reference to the drawings.

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

In FIG. 1, the input unit 1 is, for example, a keyboard;
Enter the original text to be translated (e.g. Japanese text),
This is for inputting commands for various editing. The display unit 2 displays the input original text, a translated text (eg, English text) as a result of machine translation, a list of translation candidates, and other various editing information.

The syntactic analysis unit 3 performs syntactic analysis of the input original text, and the translation unit 4 translates the structure of the original text obtained as a result of the syntactic analysis into words and structures of the target language of translation, for example, English. The translation generation unit 5 uses the translation result of the translation unit 4 to generate a translation in the target language. The parsing grammar/dictionary 6 is used by the syntactic analysis section 3, the bilingual dictionary 7 is used by the translation section 4, and the generation grammar 8 is used by the translation generation section 5 for their respective processing.

The editing control unit 9 controls the entire editing process including the machine translation process, and in this embodiment, the editing control unit 9 also creates co-occurrence data, as will be described later.

When one word is selected from a plurality of translation word candidates, the grammatical relationship determination unit 11 uses the sentence analysis unit 4 to analyze the grammatical relationship (mainly relationships of dependency) between the selected translation word and other translated words in the translated sentence. The detection is performed using the results, and it is determined whether the detected grammatical relationship matches a specific relationship, that is, one or more grammatical relationships preset by a grammatical relationship table.

The judgment result of the grammatical relationship judgment section 11 is given to the editing control section 9. The editing control unit 9 creates co-occurrence data according to this determination result.

The co-occurrence data storage section 12 accumulates a co-occurrence table by storing the co-occurrence data created by the editing control section 9.

Next, the procedure for creating and storing co-occurrence data in this embodiment will be explained using the flowchart shown in FIG. Note that FIG. 2 shows the processing when machine translation is performed on a certain original text. As a result of machine translation, a display as shown in FIG. 3 is displayed on the display unit 2, for example. In FIG. 3, the input original text is displayed in the original text display area 31, and the translated text obtained as a result of translation is displayed in the translated text display area 32. Furthermore, various editing information such as a translation candidate list is displayed in the editing area 33.

If the translated text initially obtained by machine translation displays a translation word that is not the user's intended translation ("boy" in the example in Figure 3), the "next candidate key" provided in input unit 1 is displayed.
If you operate , other translation candidates will be displayed. Further, for example, when the user operates the "translation candidate-collective display key" provided in the input unit 1, a list of translation candidates is displayed in the editing area 33 as shown in FIG.

In any of these cases, when the user selects an appropriate translation from among multiple translation candidates using the "translation selection key" provided in the input unit 1, the editing control unit 9 recognizes it (step SL). Next, the grammatical relationship determination unit 11 detects the grammatical relationship between the selected word (selected translation word) and other translated words in the translation (especially, the translated word that has been arbitrarily determined), and then It is determined whether the detected grammatical relationship matches a grammatical relationship preset by the grammatical relationship table (steps 82 to S3).

As a result of the determination in step S3, if it is determined that the grammatical relationship between the selected translation word and another translation word matches the preset grammatical relationship, the editing control unit 9 The information and the information of other translated words are combined and sent to the co-occurrence data storage section 12. As a result, the co-occurrence data storage unit 12 stores a pair of the selected translation word and another translation word as co-occurrence data (step S4).

This will be explained more specifically using the example shown in FIG.

Suppose that the user selects "boy" as an appropriate translation for "boy", which is the original word for the part displayed as "boy" in the machine-translated text. In other words, suppose we have obtained the translation ``I call the boy from the hotel.''

The structure of this translated text is shown in FIG. As shown in the figure, the grammatical relationships between the translated words that make up this translated sentence are as follows. First of all, "boy" means "hotel's"
It is modified by the noun phrase ``to call,'' and is also the object of the verb ``to call.'' In other words, the grammatical relationship (dependency relationship) between the selected translation word ``boy'' in this case and the other uniquely determined translation words ``hotel'' and ``call'' in the same translation are as follows: The relationship between the modified word and the modifier, and the relationship between the object and the verb, respectively. Note that it is possible to determine whether a translated word is uniquely determined in a translated text by detecting a "translated word determination flag" that is temporarily set for each translated word during machine translation.

The grammatical relationship determination unit 11 selects the selected translation word "boy" and other uniquely determined translation words "hotel" and "karu".
Detects the grammatical relationship between the two and determines whether this has a specific preset relationship. In this case, it is assumed that all of these grammatical relationships are set in advance in the grammatical relationship table. When the editing control unit 9 receives the judgment result from the grammatical relationship judgment unit 11, it shares the pair of “boy” and “hotel” (boy, hotel) and the pair of “boy” and “call” (boy, call). The co-occurrence data is stored in the co-occurrence data storage unit 12 as co-occurrence data.

When storing in the co-occurrence data storage unit 12, character codes of two words constituting the co-occurrence data may be stored as a set, but an identification number called a dictionary ID attached to the character code may be stored as a set. It is desirable to do so.

FIG. 5 shows a part of the bilingual dictionary 7 in FIG. 1, which stores a character string in English as the original language, a character string in Japanese as the translated word, and a dictionary ID as a set. Furthermore, since there are multiple Japanese translation candidates corresponding to the same English character string, each translation candidate is assigned a translation number.

Here, when storing the pair (boy, hotel) as co-occurrence data, as shown in Figure 6, the dictionary ID and translation number indicating "boy" and the dictionary ID indicating "hotel" are combined. and memorize it. Dictionary ID and translation number have far fewer bits than character code, so dictionary ID
When co-occurrence data is stored using character codes and translated word numbers, the storage capacity of the co-occurrence data storage unit 12 becomes smaller than when co-occurrence data is stored using character codes.

In addition, the two dictionary IDs and translation word N that make up the co-occurrence data
A combination of ``o'' and ``dyadic relationship name'', which is information indicating a grammatical relationship between the two, may be stored as co-occurrence data.

Next, during machine translation, a source word that provides multiple translation word candidates is input, and a combination of one of the translation word candidates and another uniquely determined translation word in the translated text is stored in the co-occurrence data storage unit. 12 as co-occurrence data.

In such a case, the translation candidate appears first in the translation display as the one that provides the highest possibility. Also,
In this case, if a translation candidate list as shown in FIG. 3 is displayed, the translation candidates stored as co-occurrence data are displayed at the top. Therefore, the user can easily select an appropriate translation word from among a plurality of translation word candidates.

Alternatively, the translation word candidates stored as co-occurrence data may not be used as candidates, but may be automatically selected.

The present invention is not limited to the above embodiments, and can be implemented with various modifications. For example, in the above embodiment, two words are combined as co-occurrence data, but three or more words may be combined and stored as co-occurrence data. For example, according to the above example, a set of "boy", "hotel", and "call" (boy, hotel, call) can be stored as co-occurrence data.

In addition, in the above example, in order to increase the reliability of the co-occurrence data to be learned, the selected translation word selected from multiple translation word candidates and other uniquely determined translation words in the translated text initially obtained by machine translation are used. We detected the grammatical relationship between the selected translated word and the other translated word that has a specific grammatical relationship, and used only pairs of the selected translated word and other translated words that have a specific grammatical relationship as co-occurrence data. It is not necessary to use co-occurrence data, and a set of a uniquely determined selected translation word and another translation word that have all grammatical relationships may be used as co-occurrence data. In addition, without determining such grammatical relationships, the selected translation is automatically paired with a uniquely determined translation immediately before or after it, or with a uniquely determined translation both immediately before and after it. It may also be used as co-occurrence data.

In addition, the present invention can be implemented with various modifications without departing from the scope.

[Effects of the Invention] According to the present invention, by learning translated words that have co-occurrence relationships in the process of machine translation and storing the co-occurrence data, a large number of co-occurrence data can be stored in advance by accumulating a co-occurrence table. By preparing a co-occurrence table in a large memory, you can easily select translation words.

Furthermore, since the co-occurrence table accumulated by the present invention strongly reflects the results of learning the word usage tendencies of users who actually use the machine translation system, the number of co-occurrence data stored is at least The effect is great.

Furthermore, the machine translation system of the present invention has the advantage that co-occurrence data is accumulated as translation words are selected, and the performance improves as the system is used.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of a machine translation system according to an embodiment of the present invention, and FIG. 2 is a flowchart for explaining the procedure for creating and storing co-occurrence data in the embodiment.
Figure 3 is a diagram showing an example of the display on the screen during machine translation in the same example, Figure 4 is a diagram showing the structure of a translated sentence obtained by the same example, and Figure 5 is a diagram showing the co-occurrence in the same example. FIG. 6 is a diagram showing a part of the dictionary from which data is created, and FIG. 6 is a diagram showing a specific example of co-occurrence data in the same embodiment. 1... Input unit 3... Syntactic analysis unit 5... Translation generation unit 7... Bilingual dictionary 9... Edit control unit 11... Grammatical relationship determination unit 12... Co-occurrence data storage Part 2... Display part 4... Translation part 6... Parsing grammar/dictionary 8... Generation grammar

Claims (3)

[Claims] (1) A determining means for determining whether a selected translation word selected from a plurality of translation word candidates during machine translation has a specific grammatical relationship with other translated words in the translated text; A machine translation system comprising: storage means for storing a set of information indicating each of the selected translation word determined to have a grammatical relationship and the other uniquely determined translation word. (2) Information indicating a uniquely determined translation word selected from multiple translation word candidates during machine translation and information indicating other uniquely determined translation words in all grammatical relationships. A machine translation system characterized by comprising a storage means for storing things in pairs. (3) Information indicating a selected translation word selected from a plurality of translation word candidates during machine translation, and information indicating at least one translation word immediately before or after the selected translation word in the translated text are stored in pairs. A machine translation system characterized by comprising a storage means.