JP3085394B2 - Translated word selection method in multi-sentence translation and machine translation system using the same - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a method of selecting a translated word and a method of supporting extraction of translated word selection information in a machine translation system. More specifically, the present invention relates to a method of performing translation word selection and translation word selection information extraction support using information obtained from the entire text when translating a text composed of a plurality of sentences.

Further, the present invention relates to a machine translation system using these methods.

[Conventional technology]

When performing translation, there is a many-to-many relationship between words in the source language and words in the target language. Therefore, it is difficult for a machine translation system to determine a translation of a certain word.
As a method for solving this, there is a method of selecting a translated word using co-occurrence information between words in a sentence to be translated, as disclosed in Japanese Patent Application Laid-Open No. 60-74081. Further, as disclosed in Japanese Patent Application Laid-Open No. 63-278174, there is a method of automatically determining a text field and selecting a translation using a field code assigned to a translation in a dictionary.

As a method for extracting the information for selecting these translated words, there is a method for extracting co-occurrence information between words as disclosed in Japanese Patent Application Laid-Open No. 1-41971.

[Problems to be solved by the invention]

However, in the method of selecting a translated word using co-occurrence information between words in one sentence, as in Japanese Patent Application Laid-Open No. 60-74081, if there is no co-occurring word in the same sentence, the translated word selection is performed successfully. Could not do.

Also, in the case of a method of selecting a translated word using a field of text as disclosed in Japanese Patent Application Laid-Open No. 63-278174, it is said that coarse information like the field cannot respond to a user's request for translating words in detail. There was a problem. Further, there is a problem that it is difficult to determine a ratio of each word appearing in each field.

As a method for extracting co-occurrence information, there is a method as disclosed in Japanese Patent Laid-Open No. 1-41971, but in order to extract co-occurrence information, many example sentences that are not related to the original translation work are prepared. There was no function to assist the user in extracting co-occurrence information during the translation process.

An object of the present invention is to provide a method for selecting an appropriate translated word even when it is not possible to select an appropriate translated word by processing one sentence at a time, by using a text composed of a plurality of sentences as a processing target. It is in.

Another object of the present invention is to provide a method for selecting a more appropriate translation word by setting priorities to a plurality of translation word selection steps in advance and giving priority to the translation word selected by the step with the higher priority. is there.

Still another object of the present invention is to provide a method for supporting extraction of information for selecting a translation word such as co-occurrence information or field information using a user's instruction during or after translation processing. Is to do.

Still another object of the present invention is to provide a machine translation system using a translated word selection method according to the gist of the present invention.

[Means for solving the problem]

In order to achieve the above object, according to the present invention, when translating a text composed of a plurality of sentences, a step of extracting information for selecting a translation from the entire text, and using the extracted information for selecting a translation, Is provided.

In order to achieve the above object, a plurality of steps for extracting information for selecting a translation word are provided, and priorities are given to the respective steps in advance. I do.

In order to achieve still another object, there is provided a step in which the user selects correct translation word selection information from candidates of a step of selecting a translation word and a step of extracting candidates of translation word selection information.

[Action]

In the present invention, by extracting information for selecting a translation word from the entire text, in the conventional processing for each sentence,
Even when a translation cannot be selected, an appropriate translation can be selected.

Also, by preparing a plurality of methods for extracting information for selecting a translated word and assigning priorities to the respective means in advance, an appropriate translated word can be selected.

Further, by providing a step for the user to select a translation, a step for extracting candidates for translation word selection information, and a step for the user to select translation word selection information from the candidates,
Translated word selection information such as co-occurrence information and field information can be easily extracted during the translation process.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited by this.

FIG. 2 is a block diagram of a machine translation system for translating a plurality of Japanese sentences into English sentences according to the first embodiment of the present invention.

In the figure, 1 is a display device, 2 is a keyboard, 3 is an original file, 4 is a translation file, 5 is a word dictionary, 6 is a compound word dictionary, 7 is a memory, and 8 is a processing device. The memory 7 includes a text constituent word table 71, a compound word constituent word table 72, and a field information table 73.

The process of performing the translation process by the machine translation system will be described with reference to the processing flow of FIG.

(Step 11) Text data is extracted from the original text file 3 and input to the machine translation system. Text data input may be performed directly by an input device such as a keyboard.
FIG. 3 shows an example of the input original text.

(Step 12) In the input text, specify a range to be treated as a set of text. For example, as input data, 1
If all books have been input, chapters and sections may be specified as a group of texts. If not specified, treat all input text as a single text. Hereinafter, this group of texts is referred to as a translation unit. In the following, a case where one translation unit is processed will be described by taking as an example a case where the entire original sentence in FIG. 3 is used as a translation unit. If the input text is composed of a plurality of translation units, the processing for one translation unit may be repeated.

(Step 13) Specify the sentence to be translated. This may be designated in a range by designating the first sentence and the last sentence, or all the sentences to be translated may be designated directly by numbers. If there is no instruction, all sentences in the translation unit may be translated, or only untranslated sentences in the translation unit may be translated. this is,
It is set in advance as a default value.

(Step 14) Extract information for selecting a translated word from the entire translation unit. This step will be described in detail with reference to FIG.

The following steps 141 to 143 are text constituent word extraction processing for extracting words and compound words that appear in the translation unit as information for selecting a translated word.

(Step 141) All words and compound words in the translation unit are cut out and morphological processing is performed. Since the method disclosed in Japanese Patent Application Laid-Open No. 61-40671 may be used for extracting words and compound words and for processing morphemes, detailed description thereof will be omitted.

(Step 142) A list of words and compound words appearing in the translation unit is created. Examine all words and compound words in the translation unit and their appearance frequencies, and text component word table in order of frequency
Store in 71. However, as for words, only content words such as nouns and verbs are stored, excluding functional words such as particles and auxiliary verbs. Hereinafter, these words and compound words are referred to as text constituent words. FIG. 5A shows an example of the text constituent word table 71 at the time when the first sentence of the example sentence shown in FIG. 3 has been processed.
Shown in the figure. The text constituent word table 71 includes a text constituent word, an appearance frequency, a keyword flag, and a compound word flag. Text constituent words represent content words or compound words that appear in the sentences in the translation unit. The appearance frequency indicates the frequency at which each text constituent word appears in the translation unit, and is all 1 at this time. The keyword flag is a flag used to determine whether or not to use as a translation word selection condition, and is initially set to all off. The compound word flag is a flag indicating whether the text constituent word is a compound word or a word. If the compound word is a compound word, it is turned on, and if it is a word, it is turned off. For example, the fifth
In (a), only the expert system is a compound word, and the compound flag is on, and the rest is off. Hereinafter, processing is performed for all sentences in the text, and words that appear repeatedly are counted in the frequency of appearance,
Create 71.

(Step 143) A predetermined threshold in the text constituent word table 71
Turn on the keyword flag of a text constituent word having an appearance frequency of Thr1 or more. In this embodiment, when the threshold is 1,
That is, a case will be described in which all kiss words are used as translation word selection conditions. In this case, the keyword flags of all the text constituent words are turned on. FIG. 5 (b) shows the text constituent word table after the above processing.

When the threshold value is increased, the keyword flag is turned on for a text component having a high appearance frequency. By increasing the threshold, the number of words used as the translation word selection condition decreases, so that the processing time can be shortened. Also, text constituent words having a high appearance frequency are considered to be keywords that characterize the text, and therefore, by increasing the threshold value, the translation word selection accuracy does not decrease significantly. Further, in the case where a keyword for searching for a text is previously assigned to the text in a text database or the like, the keyword may be used instead of extracting a word from the text.

The next step 144 is a compound word extraction process for extracting the words constituting the compound as the information for selecting the translation and the translation.

(Step 144) The compound word dictionary 6 for all words whose compound word flag is on in the text constituent word table 71, that is, for all compound words
Is stored in the compound word table 72, and the words forming the heading of each compound word and the corresponding translations are stored.

FIG. 6 shows an example of the compound word dictionary 6. The compound word dictionary 6 is
It consists of a heading part, a translated word part, and an additional information part. The headline and the translation are each composed of at least two words. The additional information section includes a field code and a constituent word correspondence. In the field code, a field in which this compound word is easily used is described by the code. If it cannot be limited to a specific field, the code "general" is assigned. The constituent word correspondence holds a relation between a unit forming a compound word heading and a word forming a translation of the compound word. For example, in the case of FIG. 6, the compound word “knowledge expression” is composed of the words “knowledge” and “expression”, and the translated word is “knowledge representation”.
owledge "," representation ", and information that it is used in the field of knowledge engineering.
Information that the translation of "knowledge" is "knowledge" and the translation of "expression" is "representation" is stored in the dictionary.

Here, for example, when the compound word “knowledge expression” is present in the text, the words “knowledge” and “expression” and the words “knowledge” and “representation” are stored in the compound word table 72. . FIG. 7 shows an example of the compound word table 72. As a result of applying the above processing to compound words such as “expert system”, “rule base”, “fuzzy inference”, and “knowledge expression” in the text constituent word table 71 shown in FIG. The words to be translated and their translations are stored in the compound word table 72 as shown in FIG.

The following steps 145 to 149 are field determination processing for determining a field of a translation unit as information for selecting a translation word.

(Step 145) With reference to the text constituent word table 71, the compound word dictionary 6 is searched for all compound words extracted in step 141, and field codes are extracted.

(Step 146) If the type of the field code included in each compound word is equal to or less than the predetermined threshold Thr2, the value of the field corresponding to the field code extracted in step 145 in the field information table 73 shown in FIG. Is increased by one. However, if there is a word to which the code “general” is assigned, nothing is performed. For example, in the case of “knowledge expression” in FIG. 6, the type of the field code is one type of “knowledge engineering”. Here, when the threshold value Thr2 is 2, the value of the column “knowledge engineering” is increased by one.

(Step 147) With reference to the text constituent word table 71, the text dictionary is searched for the text constituent words that are not compound words, that is, the field codes for field determination of all the words not included in the compound word. Examples of word dictionaries are shown in FIGS. 9 (a) and 9 (b). The word dictionary is composed of a Japanese part and a translated part. In the Japanese language part, the part of speech that the headline and the word can take, information unique to the part of speech, and the field code for field determination are described. For example, in the case of FIG. 9 (a), the heading is “construct”
It is. The part of speech includes a verb (described by #V) and a noun (described by #N), and it is described that the verb is to be utilized. Further, a general term for which a field cannot be specified is described in the field code for field determination (described by #NAR).

The translation section is composed of one or more translation groups. Each translated word group corresponds to one meaning. For example, there are two types of translation of "construct", "build" and "construct", each of which constitutes one translation group. Then, each translated word group is a derivative of each translated word (# for each part of speech)
V, #N, etc.), co-occurrence condition (describe with #CO), translation selection field code (describe with #EAR), statistical priority (#ST
). As the co-occurrence condition, a word closely related to a certain translated word is described. Also, the translation word selection field code describes a field closely related to a certain translation word. As for the statistical priorities, higher priorities (smaller numbers) are assigned in the order of statistical use.

For example, in the case of “construction” in FIG. 9A, the following is described. With two translation groups,
The second translation group has statistical priority 2 and the second translation group has statistical priority 1. The first translation group includes the verbs "build" and "building", which co-occur with the compound word "expert system", and whose field is "general", and cannot be particularly limited. The second translation group includes the verb "construct", the noun "construct", and the adjective "co
nstructive ", has no co-occurrence conditions, and the field is" general ", that is, it cannot be limited to a specific field.

(Step 148) The type of field code for field determination is a predetermined threshold Th
If it is equal to or less than r2, the field information table 73 shown in FIG.
The value of the column corresponding to the field code of the word is increased by one. However, if the code “general” is assigned, nothing is performed. For example, in the case of “construction” in FIG. 9A, the type of the field determination field code is one, but since the code is “general”, the field information table 73 is not changed.

(Step 149) The values in the fields corresponding to the respective fields are compared, and the field selection flag of the field in which the number of words equal to or larger than the preset threshold appears is turned on, and the field is selected as the text field.

Hereinafter, steps 15 to 18 are executed by
This is a step of translating each sentence in ascending order of the sentence number.

(Step 15) Perform analysis processing. Morphological analysis and syntax analysis of the input Japanese sentence to obtain an intermediate representation.

(Step 16) The intermediate expression obtained in step 5 is subjected to a conversion process to convert the intermediate expression from Japanese into an intermediate expression from English.

The processing of step 15 and step 16 are described in
The method disclosed in JP-A-61-163467 may be used.

(Step 17) If each word constituting the sentence has a plurality of translation groups, an appropriate translation group is selected. Steps 171 to 179 for all words constituting the sentence
The process up to is repeated. Hereinafter, a case where the present invention is applied to the i-th word in a sentence will be described with reference to FIG.

(Step 171) A translated word group is determined using the constraints in the sentence.
Hereinafter, this step will be described in detail with reference to FIG.

(Step 1711) j is set to 1.

(Step 1712) The dictionary is searched to extract a word described as a co-occurrence condition of the j-th translation group.

(Step 1713) It is checked whether or not the word currently extracted is included in the currently translated sentence. This can be realized by referring to a table used for translation processing. If it is included, the process moves to step 1714. If not included, the process moves to step 1715.

(Step 1714) The j-th translation group is selected, and the process proceeds to the next word.

(Step 1715) It is checked whether all the translation groups have been processed. If all processes have been performed, the process ends with the translation word group undecided.
If there is an unprocessed word group, the process proceeds to step 1716.

(Step 1716) j is incremented by 1, and the process returns to Step 1712.

In the example sentence of FIG. 3, when translating the 16th sentence, there is an "expert system" in the same sentence.
The translation group of "build" is selected as the translation group of.

(Step 172) If the translation word group has been determined in step 171, the process ends, and the process moves to the next word. If undecided, proceed to step 173.

(Step 173) A translated word group is determined by using the compound word table 72. Hereinafter, this step will be described in detail with reference to FIG.

(Step 1731) The Japanese column of the compound word table 72 is searched to determine whether the i-th word is included. If it is not included, the process ends.If it is included, step
Move to 1732.

(Step 1732) The translated word of the word found in Step 1731 is extracted from the translated word column of the compound word table 72.

(Step 1733) j is set to 1.

(Step 1734) The word dictionary is searched to determine whether or not the translated word extracted in Step 1732 exists in the j-th translated word group. If it is included, the process moves to step 1735. If not included, the process moves to step 1736.

(Step 1735) The j-th translation group is selected, and the process ends.

(Step 1736) It is checked whether all the translation groups have been processed. If all processes have been performed, the process ends with the translation word group undecided.
If there is a translation group that has not been checked, the process proceeds to step 1737.

(Step 1737) j is incremented by 1 and the routine goes to Step 1734.

For example, the determination of the translation group of “expression” in the 13th sentence of the example sentence in FIG. 3 is performed as follows. First, "expression" is searched for in the Japanese column of the compound word table 72, and "representation" is extracted as a translation from the translation column. Next, when the word dictionary is searched, since the first translation group has "representation", the first translation group is selected as the translation group of "expression".

(Step 174) If the translation word group has been determined in Step 173, the process ends, and the process moves to the next word. If undecided, proceed to step 175.

(Step 175) The translated word group is determined using the text constituent word table 71. Hereinafter, this step will be described in detail with reference to FIG.

(Step 1751) The variables j and k are set to 1.

(Step 1752) The j-th word is extracted from the text constituent word table 71.

(Step 1753) The i-th word is searched in the dictionary, and the word described as the co-occurrence condition is extracted from the k-th translation word group.

(Step 1754) It is checked whether the word extracted in step 1752 is equal to the word extracted in step 1753. If they are equal, the process moves to step 1755. If they are not equal, go to step 1756.

(Step 1755) The k-th translated word group is selected as a correct translated word group, and the process ends.

(Step 1756) Check whether all translation groups have been processed. If all processing has been completed, the process moves to step 1757. If there is an unprocessed word group, the process proceeds to step 1758.

(Step 1757) It is checked whether all the words in the text constituent word table 71 have been processed. If all processes have been performed, the process ends without determining the translation word group. If there is a word that has not been processed, the process proceeds to step 1759.

(Step 1758) Increment k by 1 and move to Step 1753.

(Step 1759) j is incremented by 1 and the routine goes to Step 1752.

For example, a description will be given of the determination of the translated word group of "construction" when translating the second sentence in FIG. "Construct" is searched in the word dictionary, and "expert system" is extracted as a co-occurrence condition of the first translated word group. Since this is included in the text constituent word table 71, the first translation word group is selected, and the translation word group assignment ends.

(Step 176) If the translated word group has been determined in step 175, the process is terminated and the process moves to the next word. If undecided, proceed to step 177.

(Step 177) The translated word group is determined using the field information table 73. Hereinafter, this step will be described in detail with reference to FIG.

(Step 1771) j is set to 1.

(Step 1772) From the j-th translation group, a field code for selecting a translation is extracted.

(Step 1773) With reference to the field information table 73, it is checked whether or not the field selection flag of the field indicated by the field code extracted in step 1772 is on. If it is on, the process moves to step 1774; if it is off, the process moves to step 1775.

(Step 1774) The j-th translated word group is selected, and the translated word group selection process ends.

(Step 1775) It is checked whether all translated word groups have been processed. If all processes have been performed, the process ends without determining the translation word group.
If there is an unprocessed word group, the process proceeds to step 146.

(Step 1776) j is incremented by one, and the routine goes to Step 1772.

For example, in the case of "conflict" in the ninth sentence in FIG. 3, the translated word group is determined as follows. By referring to the dictionary shown in FIG. 9 (b), it can be understood that the translation selection field code of the second translation group is "knowledge engineering". Then, referring to the field information table, since the field selection flag of "knowledge engineering" is turned on, the translated word "conflict" is selected.

(Step 178) If the translation word group has been determined in step 177, the process ends, and the process moves to the next word. If undecided, proceed to step 179.

(Step 179) If the translation word group is not selected in the processing from step 171 to step 177, the translation word group that is statistically likely to appear is selected as a clue. In this process, the statistical priority of the dictionary is checked, and the translation group having the highest priority is selected.

(Step 18) An English sentence is generated from the intermediate expression of the h-th sentence and the determined translation group. For the generation of English sentences, see
Since the method disclosed in JP-A-163467 may be used, the description is omitted.

(Step 19) It is checked whether or not all the sentences for which translation has been instructed have been translated. If there are untranslated sentences, the process goes to Step 15, and if all the translations have been completed, the process ends.

The above is one embodiment of the machine translation system for translating a plurality of Japanese sentences into English sentences according to the present invention.

As a second embodiment, an example in which the present invention is applied to an interactive machine translation system for translating a plurality of Japanese sentences into English sentences and a translation word selection information extraction support system will be described.

FIG. 15 is a block diagram of an interactive machine translation system for translating a plurality of Japanese sentences into English sentences according to one embodiment of the present invention.

In the figure, 1 is a display device, 2 is a keyboard, 3 is an original file, 4 is a translated file, 5 is a word dictionary, 6 is a compound word dictionary, 7 is a memory, and 8 is a processing device. The memory 7 includes a text constituent word table 71, a compound word constituent word table 72, a field information table 73, a user information table 74, a selection means table 75, and a translation state table 76.

The process of performing the translation process by the machine translation system will be described with reference to the processing flow of FIG.

(Step 201) Text data is input to the machine translation system.

(Step 202) In the input text, a range to be treated as a set of text is designated.

(Step 203) Specify a sentence to be translated.

Steps 201 to 203 are the same as steps 11 to 13 in the first embodiment, respectively, and thus detailed description is omitted.

(Step 204) Information for selecting a translated word is extracted from the entire translation unit. In this step, text constituent word extraction processing, compound word constituent word extraction processing, and field determination processing are performed as in step 14 described in the first embodiment. However, the following points are different from the first embodiment. In the second embodiment, when the text constituent word table 71 is created in the text constituent word extraction processing, all the appearance positions are stored together with the appearance frequency. The appearance position is represented by a pair of a sentence number in which each word appears and a position in the sentence. The position in the sentence is a number sequentially assigned to a text constituent word in the sentence from the beginning of the sentence. Except for storing the appearance position, it is the same as step 14, and a detailed description is omitted. FIG. 17 shows an example of the text constituent word table in the second embodiment. Here, for example, “select” appears only once in the translation unit,
It can be seen that the appearance position is the tenth in the eighth sentence.

(Step 205) Perform analysis processing.

(Step 206) A conversion process is performed.

Steps 205 and 206 can use the method disclosed in Japanese Patent Application Laid-Open No. 61-163467, similarly to Steps 15 and 16 in the first embodiment, and a detailed description thereof will be omitted.

(Step 207) A translated word group is selected. First, with reference to the user information table 74 shown in FIG. 18, it is checked whether or not a translation group instruction from the user is stored. The user information table 74 stores the appearance position of the word for which the user has designated a translation and the translation group designated by the user as one or more numbers.

If the user's instruction is stored in the user information table 74, the translated word group is selected. If there is no instruction, a process similar to step 17 in the first embodiment is performed to select a translated word group. However,
When a translated word group of each word is selected, a numeral indicating which means has been used to select the translated word group is stored in the selecting means table 75 for each text constituent word. An example of the selection means table 75 is shown in FIG. Here, “1” is stored when a selection is made based on information in one sentence, “2” is stored when a selection is made using the compound word table, and the text word table is used. When the selection is made, “3” is stored. When the selection is made using the field information table, “4” is stored. When the selection is made using the statistical information, “5” is stored. Store. If there is only one translation group, "0" is stored. For example,
In the example sentence of FIG. 3, "realize" of the fifth sentence of the first sentence is "realize" selected according to the statistical priority and is the fifth word. Therefore, the sentence number in the selection means table 75 of FIG. 1,
“5” is stored in the column of word number 5. In the fourth sentence "construct", since the "expert system" exists in the same sentence, the selecting means table 75 in FIG.
"1" is stored in the column of sentence number 16 and position 4 in the sentence.

(Step 208) An English sentence is generated from the converted intermediate expression and the determined translation group. As for the generation of the English sentence, as in the first embodiment, the method described in Japanese Patent Application Laid-Open No. 61-163467 may be used, and a description thereof will be omitted.

(Step 209) The user changes the translated word group for all words constituting the sentence translated in steps 205 to 208.
Hereinafter, the case where the process is performed on the i-th word in the sentence will be described in detail with reference to FIG.

(Step 2091) Select priority table of translation word selection means for each word selection means table
Look at 75. If the priority is higher than the predetermined threshold Thr3, the process proceeds to the next word for which the processing has been completed. Otherwise, go to step 2092.

In the following, a case where 5 is set as the threshold Thr3 in advance will be described by taking as an example the translation of "construction" in the second sentence and "conflict" in the ninth sentence in FIG. In the case where the word dictionaries shown in FIGS. 21 (a) and 21 (b) are used, the co-occurrence condition and the field code are not added, so that in step 207, the translated word is selected according to the statistical priority. This result is stored in the selection means table 75. Here, referring to the selection means table 75, the priority 5 of the selection means of "construction" is obtained. Since the threshold value Thr3 is 5, the process proceeds to step 2092.

(Step 2092) Referring to the text constituent word table 71, if the frequency of occurrence of the word is smaller than the predetermined value Thr4, the processing is terminated and the next word is advanced. If so, step 2093
Move to Here, a case where Thr4 is 2 will be described. At this time, the appearance frequency of “construct” is three times, and the appearance frequency of “conflict” is two times, so that the process proceeds to step 2093.

(Step 2093) The word is displayed, a warning is issued, and the routine proceeds to Step 2094. In the case of the example, a warning is issued as shown in FIG. 22 for the translation of the word “construct”.

(Step 2094) The user instructs whether or not the translation is acceptable. If it is, the process ends and proceeds to the next word. If not, go to step 2095. Here, "built" means "bu
Since the translation of “conflict” is appropriate for “ild” and “conflict”, the user issues an instruction that the translation is not good, and proceeds to step 2095.

(Step 2095) Display all the translations of the word. As shown in FIG. 23, all translation groups are displayed.

(Step 2096) The user selects an appropriate one from the displayed translated word groups.

(Step 2097) A translation word selection condition is extracted for the translation word group selected by the user. Hereinafter, this step will be described with reference to FIG.

(Step 20971) Select whether to extract co-occurrence information. If not, the process ends and the process moves to the next word. When extracting, the process moves to step 20972. Here, a case where the co-occurrence information extraction of “construction” is selected will be described.

(Step 20972) The words in the compound word table 72 and the words whose appearance frequency is higher than the threshold Thr5 preset in the text word table are displayed as shown in FIG.

(Step 20973) The user selects 0 or more words from the displayed words. If the number of selected words is zero, that is, if no word is selected, the process proceeds to step 20975. If one or more words have been selected, the process moves to step 20974. In FIG. 25, “expert system” is selected as a word co-occurring with “construct”, and the routine goes to step 20974.

(Step 20974) The word selected in Step 20973 is added to the co-occurrence condition column of the translation group selected by the user in Step 2096. Here, the first translation group (“bu
“expert system” is stored in the dictionary as co-occurrence information of “ild” group).

(Step 20975) Select whether to extract the field code. If not, the process ends and the process moves to the next word. When extracting, the process moves to step 20976. Here, a case will be described in which the field word extraction for selecting the word of "competition" is selected.

(Step 20976) By referring to the field information table 73, fields whose field selection flag is ON are displayed as shown in FIG.

(Step 20977) The user selects zero or more fields from the displayed fields. If the number of selected words is zero, that is, if no word is selected, the process ends. If one or more words have been selected, go to step 20978. No. 2
In FIG. 6, “knowledge engineering” is selected, and the process proceeds to step 20974.

(Step 20978) The word selected in Step 20977 is added to the translation word selection field code column of the translation group selected by the user in Step 2096. Here, "knowledge engineering" is stored in the dictionary as the field of the second translation group of "conflict" (group of "conflict").

(Step 2098) It is determined whether or not to use the translation group selected in Step 208 as a translation group of the same word present in another sentence. Hereinafter, this step will be described in detail with reference to FIG.

(Step 20981) Whether to change the translated word group of the same word existing in the translation unit being processed to the translated word group selected by the user is (1) Replace all, (2) Enter a threshold value, and give priority to a lower value than the threshold value Replace only the word selected by the means having the ranking, or (3) select from not replacing. Here, a case will be described in which the user changes the translation of “construct” in the second sentence of FIG. 3 and inputs “4” as the priority threshold Thr6.

(Step 20982) When (1) is selected, the flow proceeds to Step 20984. Otherwise, go to step 20983.

(Step 20983) When (2) is selected, the priority threshold Thr6 is set by the user, and the process proceeds to Step 20985. Otherwise, it is determined not to be applied, and the process ends.

(Step 20984) The priority threshold Thr6 is set to 0, and the process proceeds to Step 20985.

(Step 20985) The appearance position column of the corresponding word in the text constituent word table 71 is checked, and the processing of steps 20986 and 20987 is performed for all occurrence positions. In the example shown in FIG.
In addition to the sentence, "construction" exists in the 16th and 17th sentences,
Processing is performed on these two.

(Step 20986) Selection means table corresponding to the i-th occurrence position of the word
The position in 75 is searched to determine whether the priority of the selection means is lower than the priority of the user instruction. If it is smaller, the process moves to step 20987, and if it is not smaller, the process ends.
In the example, in the case of “construction” in the 16th sentence, the priority is 1 for the selection means, and thus the processing is terminated. In the case of “construction” in the seventeenth sentence, the priority order of the selection means is 5, and therefore the flow proceeds to step 20987.

(Step 20987) In the translation state table 76 shown in FIG. 28, the translation state of the sentence including the corresponding word is set to “translation waiting”. As a result, the user can know the sentence including the word for which the user has specified the translated word. If the user gives a retranslation instruction as needed, the user can use the user information table 74 to execute step 2096.
Can reflect the result of selecting the translated word group. It is also easy to automatically re-translate a sentence in the "waiting for translation" state.

(Step 210) An English sentence is generated again using the translation group designated by the user and output.

(Step 211) It is checked whether or not all the sentences for which translation has been instructed have been translated. If there are untranslated sentences, the process proceeds to step 205, and if all the translations have been completed, the process ends.

The above is an embodiment of the interactive machine translation system for translating a plurality of Japanese sentences into English sentences and the translated word selection information extraction support system according to the present invention.

〔The invention's effect〕

According to the present invention, by extracting information for selecting a translation from the entire text, an appropriate translation can be selected even if a translation cannot be selected by the conventional one-sentence processing. become.

Also, a plurality of steps for extracting information for selecting a plurality of translations are provided, priorities are given to the respective steps in advance, and a translation is selected by the step having the highest priority among applicable means, so that an appropriate translation is obtained. Can be selected.

Further, a step of extracting a candidate for translation word selection information is provided. When selecting a translation word in accordance with a user's instruction, the candidate translation word selection information is displayed, and information for selecting the translation word selected by the user is displayed. By selecting from among them, it becomes possible to support extraction of translated word selection information such as co-occurrence information and field information.

[Brief description of the drawings]

FIG. 1 is a basic flowchart of the first embodiment of the present invention, FIG. 2 is a hardware configuration diagram of the first embodiment of the present invention, FIG. 3 is an example of input text, and FIG. Of the translation word selection information extraction process from FIG. 5, FIG. 5 (a), FIG.
(B) is a conceptual diagram of a text constituent word table, FIG. 6 is a conceptual diagram of a compound word dictionary, FIG. 7 is a conceptual diagram of a compound word constituent word table, FIG. 8 is a conceptual diagram of a field information table, FIG.
9A and 9B are conceptual diagrams of a word dictionary, FIG. 10 is a flowchart of a translation word group selection process, FIG. 11 is a flowchart of a translation word group selection process based on intra-sentence information, and FIG. FIG. 13 is a flowchart of a translation word group selection process using a compound word composition word table, FIG. 13 is a flowchart of a translation word group selection process using a text composition word table, FIG. 14 is a flowchart of a translation word group selection process using a field information table, and FIG. FIG. 16 is a hardware configuration diagram of the second embodiment of the present invention, FIG. 16 is a basic flow diagram of the second embodiment of the present invention, FIG. 17 is a conceptual diagram of a text constituent word table in the second embodiment, 18 is a conceptual diagram of a user information table, FIG. 19 is a conceptual diagram of a selection means table, FIG. 20 is a flowchart of a translation word group selection process by a user, and FIGS. 21 (a) and 21 (b) are Conceptual diagram of word dictionary, Figures 22 and 23 show translated words Illustration of a screen in-option process, Figure 24 the flow diagram of translation selection information extraction support processing,
FIGS. 25 and 26 show examples of screens in the translation word selection information extraction support processing, FIG. 27 shows a flow chart of processing applied to the entire text specified by the user, and FIG. 28 shows a conceptual view of the translation state table, respectively. . 1 ... display device, 2 ... keyboard, 3 ... original file, 4 ... translated file, 5 ... word dictionary, 6
... compound word dictionary, 7 ... memory, 8 ... processing unit, 71 ...
... Text constituent word table, 72 ... Compound constituent word table, 73 ... Field information table, 74 ... User information table, 75 ... Selection means table, 76 ... Translation state table.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-1-314373 (JP, A) JP-A-63-278174 (JP, A) JP-A-63-293673 (JP, A) JP-A-63-27837 211067 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 17/28

Claims (1)

(57) [Claims] 1. A translation word selection method in a machine translation system including a display device, a keyboard, an original file, a word dictionary, a compound word dictionary, a memory, and a processing device. Inputting a text from the keyboard to the processing device; extracting, in the processing device, a compound word from the text with reference to the compound word dictionary, and storing information on a word constituting the compound word and a translation thereof in the memory; And translating the text with reference to the word dictionary and the memory. In the step of translating the text, there are a plurality of translation candidates of the words constituting the text in the word dictionary. In this case, the plurality of candidate translations are compared with the translations of the words constituting the compound words stored in the memory. ,
When there is a matching translation, a translation is selected from among the plurality of translation candidates as a translation.