JPH0433072B2 - - Google Patents

Description

[Detailed description of the invention]

[Technical Field of the Invention] The present invention relates to a machine translation device that can improve the efficiency of translation processing and obtain highly natural translated sentences. [Technical background of the invention and its problems] Recently, machine translation devices that use a computer to automatically translate an input original text and obtain a translated text have been attracting attention. For example, various attempts have been made to develop natural language machine translation devices that input a Japanese sentence and obtain its English translation, or conversely input an English sentence and obtain its Japanese translation. This type of machine translation device basically performs morphological analysis, syntactic analysis, and semantic analysis of the input original text, classifies the input original text into predetermined language processing units, such as words (phrases), and performs translation processing. The translation dictionary that stores the necessary knowledge information is referred to, and the translated words (translated phrases) are obtained for each language processing unit, and these translated words (translated phrases) are combined according to predetermined translation rules to create a translated sentence for the original sentence. It is configured to generate . However, the expression styles of natural language are extremely diverse, and it is difficult to say that analysis techniques have been sufficiently established. That is, syntactic analysis basically begins with selecting one part of speech from a list of parts of speech of the original language that constitutes the input original text. However, when there are a huge number of candidates for part-of-speech sequences, the time required for processing them becomes extremely long. On the other hand, conventionally, if the above-mentioned translation process fails, the original sentence is divided at appropriate and appropriate positions, the partial translations required for each division unit are output, and the partial translations are combined under the operator's instructions. It is recommended that the translation be completed by However, in order to perform such partial translation processing,
It is necessary to wait for a long time until the above-described translation process for the input original text is completed and its failure is revealed. Such waiting time imposes a large psychological and physical burden on the operator, resulting in a reduction in translation processing efficiency. Further, the waiting time (time required for translation processing) is influenced by the complexity of the syntactic structure of the input original text, the length of the sentence, etc. For this reason, there is also a problem in uniquely setting the processing time that can be spent on the translation process and discontinuing the translation process on the input original text when the set time is reached. [Object of the Invention] The present invention has been made in consideration of the above circumstances, and its purpose is to reduce unnecessary translation processing by limiting the time spent on translation processing depending on the input original text. The object of the present invention is to provide a machine translation device that can improve translation efficiency. [Summary of the Invention] The present invention provides a machine translation device that calculates a translated text by translating an original text given from an input unit using knowledge information stored in a translation dictionary, which measures the translation processing time for the input original text. , when this translation processing elapsed time reaches a predetermined set time, the translation processing form is changed, and in particular, the set time is changed according to the input original text.
For example, it is determined according to the number of words constituting the input original text and the number of parts of speech that each word has, thereby reducing the time required for translation processing for the input original text. [Effects of the Invention] Thus, according to the present invention, the time spent on translation processing for an input original text is limited depending on the complexity of the input original text, and if the translation processing is not completed within the time limit, the translation processing form is changed. Therefore, it is possible to reduce the waste of translation processing and, as a result, obtain the translation result in a short time. Therefore, it is possible to perform translation processing efficiently and obtain translation results in a short period of time, without having to spend an unnecessarily long time translating an input original text with a complex structure, which has many practical benefits. This produces a certain effect. [Embodiment of the Invention] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In this embodiment, an English sentence is input and it is machine translated into a Japanese sentence. FIG. 1 is a schematic diagram of the apparatus of this embodiment. In FIG. 1, reference numeral 1 denotes an input section consisting of a keyboard and the like. An English sentence consisting of a character string or the like input through the input unit 1 is stored in the original text storage unit 2 as an original text to be subjected to translation processing. The translation unit 4, under the control of the editing control unit 3, creates a translation dictionary 5 in which knowledge information necessary for translation processing is stored in advance.
With reference to , the original text stored in the original text storage section 2 is subjected to machine translation processing in a predetermined processing unit. The translated text obtained by this translation section 4 (Japanese here)
is managed and stored in the translation storage unit 6 in association with the obtained original text. The knowledge information stored in the translation dictionary 5 is as follows:
For example, regular/irregular change dictionary 5a, word (translation)
It consists of a dictionary 5b, an analysis grammar 5c, a conversion grammar 5d, a generative grammar 5e, and a morpheme generative grammar 5f. The editing control unit 3 drives the display control unit 7 to associate the English sentence (original text) stored in the original text storage unit 2 and the Japanese sentence (translation text) stored in the translation storage unit 6 with each other. The information is displayed on a display unit 8 such as a CRT display. By displaying the original text and the translated text on the display unit 8, post-editing processing is performed on the translated text. This post-editing process is performed, for example, in accordance with control information provided from the input section 1 and by referring to knowledge information stored in the translation dictionary 5. The printing unit 9 is for outputting a hard copy of the translated text (Japanese text) completed by the above-mentioned post-editing process. By the way, the keyboard constituting the input section 1 is constructed as shown in FIG. 2, for example. That is, this keyboard includes, for example, in addition to a key group 1a for inputting character data, a key 1b for translation instructions;
Editing key group 1c, function control key group 1d, and cursor control key group 1e for the display section 8
It is composed of the following. Further, FIG. 3 shows an example of a screen configuration on the display section 8. As shown in FIG. Here, the translation editing area 8a is at the top of the display screen, and the original text display area 8a is at the left side of the screen.
b, and a translation display area 8c is set on the right side of the screen to configure the display screen. The original text display area 8b is an area for displaying the input original text (English text) stored in the original text storage section 2. The translation display area 8c is an area for displaying the translations (Japanese sentences) stored in the translation storage section 6 in correspondence with the original texts from which the translations were obtained. Further, the translation editing area 8a is an area for displaying knowledge information necessary for translation processing, such as translation word candidates, searched from the translation dictionary 5. FIG. 4 shows the basic operation sequence of the embodiment apparatus configured as described above. The editing control unit 3 determines the translation completion information provided from the translation unit 4 and various key information input from the input unit 1 according to the operation sequence shown in FIG. 4, and interactively processes the information. Controls the translation/editing process. That is, the editing control section 3 monitors the translation processing status in the translation section 4 (step A), and when the translation section 4 completes the translation processing for a certain original text, the editing control section 3 monitors the translation processing state in the translation section 4, and when the translation processing section 4 completes the translation processing for a certain original text, the editing control section 3 displays the translated text obtained by the translation processing. The translated text is stored in the translated text storage section 6, and the translated text is displayed on the display section 8 (Step B). If a translation completion signal is not obtained from the translation section 4, the key information provided from the input section 1 is determined (steps C, D, E, F). Depending on the determination result of the key information, the following processes are controlled. Specifically, the input key information is "translation instruction key"
If so (step C), the editing control unit 3 retrieves one of the input original texts stored in the original text storage unit 2 and supplies it to the translation unit 4 to start its translation process (step G). . In this translation process, for example, as shown in the processing sequence in FIG. Dictionary 5
Analyze using a (step Q). Through this morphological analysis, for example, a language that has a conjugation change or an inflection change is converted into its original form (basic form). Specifically, it converts languages expressed in the past tense, progressive tense, and past participle form into the present tense, and converts languages expressed in the comparative and superlative forms into their original forms (original words). Next, for each original word of the original text that has been morphologically analyzed as described above, information on its translation word and part of speech is obtained using the translation dictionary 5b (Step R). This process is performed using the translation dictionary 5b using the original word as a headword.
This is done by searching for . Thereafter, the input original text is parsed (step T) under control according to the above-mentioned parsing grammar 5c (step S). This parsing is repeated until the parsing of the original text is successful. Then, through this syntactic analysis, the arrangement structure of the parts of speech of the original language that constitutes the original sentence, their dependency relationships, the mode of tense, etc. are determined. Thereafter, the structure of the parsed original sentence is converted into the syntactic structure of a translated sentence using the conversion grammar 5d (step U). That is, the conversion grammar 5d is used to convert the syntactic structure of English (original text) into the syntactic structure of Japanese (translated text). If this conversion of the syntactic structure fails, for example, it is determined that there is an error in the syntactic analysis of the original text, and the process of parsing the original text is restarted. After that, the translation word candidates obtained for each source word of the original text are calculated based on the syntactic structure of the translation text.
The words are rearranged in the order of the words according to the language form of the translated text to obtain translation candidates for the original text (Step V).
This syntax generation process is performed according to the generation grammar 5f described above. Then, each translation candidate (basic form) constituting the translation candidate is subjected to conjugation transformation processing and ending transformation processing according to the syntactic analysis result of the original sentence and the morpheme generation grammar 5f, and the translation is converted into an appropriate linguistic expression. Complete it (Step W). Through such a series of translation processes, a translated text for the original text is obtained. Incidentally, when a plurality of translations are required for one original text, it is convenient to display only one of them and at the same time display the existence of other translations. On the other hand, returning to FIG. 4, if the input key information is a "character key" (step D), the character code indicated by the character key is stored in the input buffer (step H), and the character code is The character pattern is stored in the original text storage section 2 and displayed on the display section 8 (Step I). By displaying each character pattern of the character code stored in this input buffer, the input original text consisting of the character string input from the input section 1 is displayed. If the input key information is an "editing key" (step E), editing processing corresponding to the editing key is executed on the translation (step J). Similarly, if the input key information is a "function key" (step F), the process corresponding to the function key is executed (step K). If no key information is input, or if the input key information is other than the above-mentioned "key", other processing (step L) is performed, for example, a hard copy of the translation obtained in the translation storage section 6 is performed. Output etc. are performed. According to such an operation sequence of the editing control unit 3, for example, when an operator inputs a character by operating the character input key group 1a of the keyboard, the information of the input character is sequentially set in the input buffer and used for translation processing. The original text is sequentially stored in the original text storage unit 2 (steps D and H). The input original text is then displayed on the original text display section 8b of the display section 8 (Step I). However, if you operate the translation instruction key 1b at any point in character input, for example, at the end of inputting one sentence,
Translation processing for the input original text stored in the input buffer is started in accordance with the input key information (steps C and G). When the translation process is completed, the translation thus obtained is displayed in the translation display area 8c of the display section 8 (steps A and B). In addition, if it is necessary to modify the input original text, while inputting the original text by operating the character input key group 1a,
For example, the editing process is performed by operating the cursor control key group 1e to move the cursor to the corrected part and operating the editing key group 1c for correction, insertion, deletion, etc. Further, by operating the modification key, modification candidates other than the word or phrase indicated by the cursor are displayed using auxiliary information, for example. Further, editing processing for the translated text is similarly performed using function keys. These function keys can be used, for example, to display the translated word for a word in the translated text using the ``translation display key'', to display the dictionary for a word in the original text using the ``dictionary display key'', and to register new words and phrases in the dictionary using the ``dictionary registration key''. , new words and phrases registered in the dictionary are deleted from the dictionary using the ``dictionary deletion key,'' and partial translations of original sentences for which translation processing has failed using the ``partial translation key'' are displayed. In this way, the basic operations of machine translation processing are controlled. By the way, the translation section 4 is schematically configured as shown in FIG. 6. The time setting unit 4a sets a time limit that can be spent on translation processing based on a sentence of the input original text to be translated. The translation processing unit 4b translates the input original text by referring to the translation dictionary 5. This translation processing unit 4b first properly translates the original sentence according to the procedure described above, and when the translation process fails or the time spent on the translation process reaches the set time. At this time, the original text is divided into predetermined translation processing units, such as clauses and phrases, in cooperation with the original text division processing unit 4d, and translation candidates are determined and sequentially output for each of the divided translation processing units. ing. The post-editing process for this partial translation process result is executed through interactive processing by the operator to complete the translated text. Thus, the elapsed time measurement unit 4c measures the elapsed time of translation processing for one sentence of the input original text by the translation processing unit 4b, and determines whether or not the processing elapsed time has reached the above-mentioned set time. .
If the translation processing is not completed within the set time, the above-mentioned regular translation processing by the translation processing section 4b is stopped, and instead of this, the original text division processing section 4d is started and the translation processing section 4b is started. It is designed to perform partial translation processing. Now, the time setting section 4a sets the translation processing time according to the input original text, as shown in FIG. 7, for example. That is, when a sentence to be translated is given, the number N of words constituting the sentence is counted (step a). The count of this number of words N is
For example, this is done by cutting out the input character string into words according to spaces and delimiters, and counting the number of words. After that, the counted number of words N is sequentially compared with predetermined set values l1, l2, l3, and the length of the input sentence is determined from the magnitude relationship of the values (step b,
c, d). The above setting values l1, l2, l3 are, for example, (60),
(40), (20). As a result, long sentences with a word count of (60) or more, moderately long sentences with a word count of less than (60) but more than (40), and normal length sentences with a word count of less than (40) but more than (20). The length of the input original sentence is determined into four types: , and short sentences with the number of words less than (20). Then, depending on the judgment result of the length of this sentence (number of words), the time S1 that can be spent on translating the sentence,
S2, S3, S4 are, for example, 180 seconds, 120 seconds, 90 seconds, 60 seconds
They are set as seconds (steps e, f, g, h). In other words, the translation processing time is set according to the number of words N, which is the length of the input original text, as shown in the following table.

[Table] Therefore, in this example, if the number of words N in one sentence is (48), the time limit for translation processing for that sentence is set as 120 seconds. Note that the complexity of the input original text is not necessarily determined only by the length of the sentence based on the number of words. For example, each word typically has multiple parts of speech,
Many interpretations arise due to the different parts of speech,
The number of possible translations for this will also increase. Naturally, therefore, the time required for the translation process increases. Therefore, as shown in FIG. 8, the number of parts of speech that each word constituting a sentence has may be checked, and the total number of parts of speech may be calculated to set the translation processing time. That is, data t indicating the number of combinations of parts of speech is initialized to (1) (step i), and the number n of parts of speech that each word has from the first word to the last word of a sentence is sequentially checked (steps j, k). Then, according to the number n of the parts of speech, t=t×n is calculated (step 1), and the total number of combinations of the parts of speech is determined as t. This data t is multiplied by the unit time of translation processing to set a time limit T for translation processing of one sentence (step m). Specifically, for example, when the original sentence ``I saw a girl in the park.'' is input, a search in the translation dictionary 5 first determines that the part of speech for the word () is (noun; I). Since there are only types, t=(t;1)×(n;1)=(1) is obtained as data t by the processing in step 1. For the next word (saw), there are three parts of speech: see (verb; past of seeing) saw (noun; saw) saw (verb; past of sawing).
Through the process, t=(t;1)×(n;3)=(3) is obtained. As for the next word (girl), there is only one type (noun; girl), so t = (t; 3) x (n; 1) = (3). Similarly, for the last word (park), there are two words: (noun; park) and (verb; to park).
Since there are three parts of speech, t=(t;3)×(n;2)=(6) is obtained by the processing in step 1. As a result, t = (6) is obtained as part-of-speech combination data indicating the complexity of the input original text, and if the translation processing unit time is set to, for example, 10 seconds, the translation processing time for the input original text is 60 seconds. It is possible to set it as . In this way, a limit is set on the translation processing time according to the complexity of the input original text (number of words and parts of speech), and when the translation processing is not completed within the time limit, the translation processing format is changed to, for example, partial translation processing. According to this device that switches to translation processing, it is no longer necessary to spend an unnecessarily long time on translation processing. Moreover, the translation processing time is limited depending on the complexity of the input source text.
It becomes possible to proceed with translation processing efficiently. Therefore, the problem of having to translate short and simple sentences over a long period of time is eliminated. Furthermore, the problem of not being able to obtain correct translation results for long and complex sentences due to the setting time being too short is avoided. If a result is not obtained within the processing time corresponding to the input original text, it is possible to change the translation processing form at an appropriate timing indicated by the set time and proceed with the translation processing for the input original text. Become. Therefore, the response waiting time from inputting the original text to obtaining the translation result can be significantly reduced, making it possible to efficiently perform interactive translation processing. At the same time, since the responsiveness can be improved, great practical effects such as being able to reduce the psychological and physical burden on the operator can be achieved. Note that the present invention is not limited to the embodiments described above. Although we have explained the translation from English to Japanese here, it is also possible to translate it in the opposite direction.
Furthermore, the present invention can be similarly applied to translation processing between other languages. Further, the translation process prepared in the translation unit may be, for example, a mode in which the number of part-of-speech sequence candidates that are tried for one sentence is limited and the translation process is performed. Further, the translation processing time may be set depending on the structural complexity of the input original text. In addition, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

The figure shows a device according to an embodiment of the present invention.
The figure is a schematic configuration diagram of the embodiment device, FIG. 2 is a diagram showing an example of the configuration of the input section, FIG. 3 is a diagram showing an example of the configuration of the display screen, and FIG. 4 is a diagram showing the basic operation sequence of the embodiment device. 5 is a diagram showing the processing procedure of translation processing, FIG. 6 is a diagram showing an example of the configuration of the translation section, and FIG.
8 and 8 are diagrams each showing a procedure for setting the translation processing time. 1... Input section, 2... Original text storage section, 3... Edit control section, 4... Translation section, 4a... Time setting section, 4b... Translation processing section, 4c... Elapsed time measurement section, 4d... Original text division processing section, 5... Translation dictionary, 6... Translation storage section, 7... Display control section, 8... Display section, 9... Printing section.

Claims (1)

[Claims] 1. A translation dictionary unit that stores knowledge information used in translation processing, a translation unit that uses the knowledge information to translate an original text given from an input unit, and a translated text obtained by this translation unit. A machine translation device is provided with a translation output unit that outputs a translated text, and the translation unit includes a measuring unit that measures the elapsed execution time of translation processing for the input original text, and a translation unit that measures the elapsed execution time of translation processing for the input original text, and a setting means for setting a predetermined time that can be spent on translation processing according to the constituent elements; and when the elapsed execution time reaches the predetermined time, the mode of translation processing for the input original text is set to partial translation processing or the number of part-of-speech sequence candidates. 1. A machine translation device comprising: a change means for changing a translation processing format such as a limited translation processing to a translation processing format that can shorten translation processing time. 2. The machine translation device according to claim 1, wherein the setting means sets a predetermined time that can be spent on translation processing according to the number of words forming the input original text. 3. The machine translation apparatus according to claim 1, wherein the setting means sets a predetermined time that can be spent on translation processing according to the total number of parts of speech of each word forming the input original text.