JPH11238062A - Machine translating method/device and machine-readable medium to record program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To permit a user to easily judge what a head word means even if the head word is not registered in a machine translation device. SOLUTION: A head word recognition part 4 recognizes a head word appearing in an input sentence being a translation object. An original word retrieval part 5 retrieves the word of a full spelling, which corresponds to the head word, from the input sentence, sets the retrieved word of the full spelling and the head word as a group and stores it in a retrieval result storage part 6. A translation part 2 not only translates the word of the full spelling which corresponds to the head word in the input sentence, but inserts the spelling of the word of the full spelling into a translated sentence by associating the word with its translation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation technology, and more particularly, to a machine translation technology that enables a user to easily understand what an acronym (acronym) means.

[0002]

2. Description of the Related Art In a language represented by an alphabet such as English, a compound word is often abbreviated as an acronym in which initials of constituent words are arranged. Read Only Memory is abbreviated as ROM or Information Retr.
IEval (information search) is abbreviated as IR, for example.

When such an acronym is translated in its original form at the time of translation, a result cannot be understood by some users who understand the meaning in advance except what the acronym means. Becomes In conventional machine translation systems, such acronyms are registered in a translation dictionary,
This issue was addressed by translating full spells as needed. For example, Japanese Patent Application Laid-Open No. 5-282358 discloses a machine translation system in which an abbreviation and its full spelling are registered in a bilingual dictionary, and a user can specify whether to translate the abbreviation as an abbreviation or to translate it with a full spelling. The device is shown. For example, in the bilingual dictionary, when the translated word description for the headword “OS” is “OS, operating system”, “O
The word "S" can be translated as "OS" or as "operating system".

[0004]

The prior art described above has a limitation that acronyms and their translations must be registered in a bilingual dictionary. If the acronym is not included in the heading of the bilingual dictionary, a general dictionary unregistered word is processed, so the acronym is translated as it is in the original language, that is, as an acronym. Users who do not have prior knowledge have to do the difficult task of guessing what the acronym means using meaning and contextual knowledge. For example, if the following English sentence (original sentence) shown in FIG. 20 is translated by a conventional machine translation system, the following result (translation example) will be obtained. However, the acronyms "FAA" and "NTSB"
Is not registered in the bilingual dictionary. The following is a quotation from FIG.

[0005] The Federal Aviation Administrat
ion ordered an immediate change in fuel pumps and
the wiring.Allow the director of the FAA's airc
raft certification service said there is no eviden
ce, the agency believed it was important to make th
e changes.The National Transportation Safety Board
hasconcluded that the flight was destroyed when t
he center wing tank ignited.The NTSB has not recom
mended the changes announced Wednesday.In fact, it
has been pushing for a far more ambitious action.

[0006] The Federal Aviation Administration ordered immediate changes in pumps and wiring. The director of the FAA's aircraft certification service said there was no evidence, but the agency believes it is important to change. The National Transportation Safety Board concluded that the flight was destroyed when the central wing tank ignited. The NTSB does not recommend changes announced on Wednesday. In fact, it calls for more ambitious behavior.

In this translation, it is difficult for a user who first comes into contact with the acronyms "FAA" and "NTSB" to understand what this means from the translation. In practice, "FAA" is the acronym for Federal Aviation Administration, translated as Federal Aviation Administration, and "NTSB" is the National Transport Safety Board (translated as "National Transport Safety Board"). Although they are acronyms, it is difficult to reliably recognize the correspondence between these words unless the user has some prior knowledge.

If the number of acronyms is small, all of them need to be registered in a dictionary. However, acronyms are often used to indicate technical terms, standards, organization names, etc. associated with new technical fields. Therefore, acronyms are newly created or disappear with the change of these technologies and organizations, and the metabolism is very intense, and it is virtually impossible to deal with all by dictionary registration. Since the acronym is often an important keyword that represents a key technology of a new technology, a new organization, and the like, it is extremely important that the acronym is indicated to the user so that it is easy to determine what it stands for.

As described above, in the prior art machine translation apparatus, when an acronym not registered in the dictionary is present in the original text, it is difficult for the user to determine what that means from the translated text. There is a problem that there is.

Therefore, an object of the present invention is to output an acronym not registered in a dictionary in a form that is easy for a user to understand, even in the case of an acronym that is not registered in a dictionary. Is to do so.

[0011]

In order to solve the above-mentioned problems of the prior art, a machine translation method according to the present invention uses an acronym recognition process for determining an acronym existing in an input sentence to be translated, and an acronym recognition process. An original word search process for searching the input sentence for a full spelling word corresponding to the acronym recognized in the word recognition process, and a full spelling word searched for in the original word search process when generating a translated sentence for the input sentence The original spelling of
And a translation process for outputting the word in association with the translation of the word.

Further, in order to solve the above-mentioned problem of the prior art, the machine translation apparatus of the present invention comprises an acronym recognition section for certifying an acronym present in an input sentence to be translated, and an acronym recognition section for locating the acronym. Full-spelled words that correspond to certified acronyms,
An original word search unit that searches from the input sentence, and a translation unit that, when generating a translated sentence for the input sentence, outputs the original spelling of the full spelled word searched by the original word search unit in association with the translated word of the word And

In this configuration, an acronym appearing in an input sentence is recognized by an acronym recognition unit, and an original word search unit checks whether a corresponding full-spelled word appears in the input sentence. When a full-spelled word appears, it not only translates it but also outputs the full-spelled spelling of the original language. By looking at the full spelling of this source word, the user can see what the acronym stands for,
You can see at a glance what translation was translated.

An example will be described with reference to FIG. Here, the acronym "FAA" appears in the original text. Suppose that this is an acronym, which has been certified by the Acronym Certification Department. On the other hand, "Feder
al Aviation Administration "appears elsewhere. It is likely that the latter is the full-spelling word corresponding to the former, because the first letter of each latter word is capitalized, and Can be easily determined based on clues such as the former being arranged, and so on. The original word search unit makes such a determination. Various determination methods are conceivable, but it is necessary to register in a dictionary in advance. Any method may be used as long as there is no method.
When the above determination is made, in the present invention, not only the full spelling word is translated but also the full spelling in the original language is included in the translation to create a translation. In this example,
The full spelled original spelling is output in parentheses immediately after the translated word. "NTSB" and "National Transporta"
The same determination and processing are performed for the "Action Safety Board" to obtain, for example, the translation shown in FIG. 13B. The following is quoted from FIG. 13B.

[0015] Federal Aviation Administra
option) ordered immediate changes in pumps and wiring.
The director of the FAA's aircraft certification service said there was no evidence, but the agency believes it is important to change. National Transporta Board
The Action Safety Board concluded that the flight was destroyed when the central wing tank ignited. The NTSB does not recommend changes announced on Wednesday. In fact, it calls for more ambitious behavior.

[0016] From this translation, the word "FAA" translated as "Federal Aviation Administration" (Federal Aviation Administration)
Acronym for "NTSB" and the word "NTSB" translated as "National Transportation Safety Board"
It is clear that the acronym corresponds to the Safety Board, and users without prior knowledge can easily understand the acronym.

Here, it should be noted that the system merely gives clues by indicating the translated word and the original full spell in association with each other. It is up to the user to determine these correspondences. For this reason, even if the system incorrectly recognizes the correspondence between the acronym and the corresponding full spelling word, only the full spelling corresponding to the extra word is indicated, without causing an error in the translation itself. This robustness is one of the characteristic effects of the present invention.

In general, it is not a problem specific to translation that the meaning of an acronym is difficult to understand. Even when reading the original text directly, if the acronym suddenly appears, it is natural that the meaning can not be understood except for some readers. Therefore, when articles such as newspapers and magazines are intended for an unspecified number of readers, it is customary to show the user abbreviations when using acronyms. For example, use the full spelling at the first appearance in the article and use the acronym from the second time onward, or use the acronym from the beginning, but write the full spelling with the same case or parentheses at the first occurrence. It is normal that there is. With this in mind, the reader of the original text can easily understand the meaning of the acronym. However, once this is translated into another language, such considerations become meaningless.
Because the full spelling words themselves are translated and lose spelling information, the correspondence with the acronyms cannot be understood. The machine translator according to the present invention, when translating a full spelling word corresponding to an acronym for an original sentence written with such considerations, also includes the spelling in the original language in the translation, so that the full spelling and acronym It is intended to make it easy for readers to find correspondences between words not only in the original text but also in the translated text.

Further, the machine translation apparatus of the present invention prevents the full spelling of a very common acronym from being inserted into a translation and makes it difficult to see the translation, and shortens the time required for the translation process. The certification unit stores an acronym pattern storage unit in which conditions for being an acronym are stored, an acronym suppression list in which words that should not be treated as acronyms are stored, and is stored in the acronym pattern storage unit. An acronym pattern application engine that satisfies the conditions described above and that recognizes words that are not stored in the acronym suppression list as acronyms.

In this configuration, even if the word satisfies the condition stored in the acronym pattern storage, the word stored in the acronym suppression list is not recognized as an acronym. Therefore, by storing words such as "USA" and "TV" which are very common in the acronym suppression list, they appear in the original sentence, and the corresponding full-spelled words appear in the original sentence. Even in this case, the full spell is inserted in the translation, and the translation is not difficult to see. Further, for "USA", "TV", and the like, the full spelling search processing is not performed, so that the time required for the translation processing can be shortened.

Further, the machine translation apparatus of the present invention includes an acronym recognition section for certifying an acronym existing in an input sentence to be translated, and an acronym recognition section for using the existing translation section. Full-spelled words that correspond to certified acronyms,
An original word search unit to be searched from the input sentence, and a full spelled word searched by the original word search unit is inserted into a translation in association with a translation of the word when the input sentence is translated. An input sentence editing unit for pre-editing the input sentence,
A translation unit for translating the input sentence after being preedited by the input sentence editing unit.

In this configuration, the input sentence editing unit inputs the full spelled word searched by the original word search unit so as to be inserted into the translated sentence in association with the translated word of the input sentence when the input sentence is translated. Pre-edit the statement.

[0023]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 shows a first embodiment of the present invention.
The first embodiment of the present invention shown in FIG. 1 includes an input unit 1 for inputting an original sentence to be translated, a translation unit 2 for translating the input sentence, an output unit 3 for outputting a translation result, An acronym recognition unit 4 that recognizes an acronym appearing in the input sentence; an original word search unit 5 that searches for a full spelling word corresponding to the acronym from the input sentence; and a search result storage that stores the full spelling original word corresponding to the acronym. 6. In FIG. 1, L12, L23, L24, L25, L46, L56, L62, L
65 indicates a communication line.

The operation of the first embodiment will be described with reference to FIGS.

When an original sentence is input to the input unit 1 (step S1 in FIG. 2), the acronym recognition unit 4 detects and recognizes an acronym in the input sentence and stores it in the search result storage unit 6 (step S2). ).

When the acronym recognition process is completed and all detected acronyms are stored in the search result storage unit 6, the original word search unit 5 is activated, and the input sentence is targeted for each acronym. A full spelling word string is searched (step S3).
The full spelling word string obtained as a search result is stored in the search result storage unit 6 in association with the acronym (step S4).

When the search processing is completed, translation is performed in the translation unit 2 (step S5). In the translation process, for the word string stored in the search result storage unit 6 in full spelling, not only a corresponding translated word is generated, but also a spelling of the original word before translation is generated immediately after the translated word.

When the translation process ends, the result is output from the output unit 3 (step S6), and the entire process ends.

Next, a second embodiment of the present invention will be described. Referring to FIG. 3, a second embodiment of the present invention is directed to an input device A1, a data processing device A2 including a computer, an output device A3, and a storage device A4.
And a recording medium A5 on which a machine translation program is recorded. The recording medium A5 is a magnetic disk, a magnetic tape, a semiconductor memory, or another recording medium.

The machine translation program is read from the recording medium A5 into the data processing device A2, and is read from the data processing device A2.
Control the operation of. The data processing device A2 performs the following processing under the control of the machine translation program.

When an input sentence is given from the input device A1,
The acronym recognition unit 4 in the storage device A4 detects and recognizes the acronym in the input sentence, and stores it in the search result storage unit 6 in the storage device A4.

When the acronym recognition process is completed and all the detected acronyms are stored in the search result storage unit 6 in the storage device A4, the original word search unit 5 in the storage device A4 is activated and the input sentence is input. Search for a full-spelled word string corresponding to each acronym. The full spelling word string obtained as a search result is stored in the search result storage unit 6 in association with the acronym.

When the search processing is completed, the translation unit 2 in the storage device A4 is started. In the translation process, the search result storage unit 6
For a word string stored in full spelling, a corresponding translated word is generated, and immediately after the translated word, a spelling in the original word before translation is generated.

When the translation process ends, the translation result is output from the output device A3, and the translation process of the input sentence ends.

[0036]

Next, an embodiment of the present invention will be described with reference to the drawings. The first embodiment of the present invention shown in FIG. 4 is realized as an English-Japanese machine translation system, and has an input unit 1 for receiving an English original sentence input, and an input for temporarily storing all input sentences inputted from the input unit 1. A sentence buffer 7, a translation unit 2, an output unit 3, an acronym recognition unit 4 for searching and certifying an acronym from the input sentence buffer 7, and a full spelling word string corresponding to the acronym from the input sentence buffer 7 is searched. An original word search unit 5;
A search result storage unit 6 stores the searched acronyms and full spelling word strings in association with each other.

The translation unit 2 further includes an English-Japanese bilingual dictionary 24 storing English-Japanese bilingual translations, an English-Japanese bilingual translation rule 25 storing grammatical rules required for English-Japanese translations, and an English-Japanese bilingual dictionary 2.
English morphological analyzer 2 that performs English morphological analysis with reference to 4
1 and an English-Japanese syntax conversion unit 22 responsible for the main processing of translation by English-Japanese syntax conversion, and a Japanese morpheme generation unit 23 for utilizing Japanese morphemes and generating a final translated sentence Have been.

Next, the operation of this embodiment will be described in detail.

FIG. 5 is a flowchart showing an operation example of this embodiment. First, an original sentence is input from the input unit 1 (step S
1). At this time, in this embodiment, all input sentences are temporarily stored in the input sentence buffer 7 via the communication line L17.

Next, the acronym recognition unit 4 is activated. Thus, the acronym recognition unit 4 searches and recognizes the acronym in the original sentence stored in the input sentence buffer 7 via the communication line L74, and stores the acronym in the search result storage unit 6 via the communication line L46. Is stored (step S2).

When this process ends, the original word search section 5 is started. As a result, the original word search unit 5 reads each acronym stored in the search result storage unit 6 via the communication line L65, and stores a full spelling word string for each read acronym in the communication line L75. (Step S)
3). If the search is successful, the selected full spelling original word is associated with the acronym and stored in the search result storage unit 6 via the communication line L56 (step S4).

When this process is completed for all acronyms in the search result storage unit 6, the translation unit 2 is started (step S5). In the translation unit 2, first, the English morphological analysis unit 21
Inputs an original sentence from the input sentence buffer 7 via the communication line L72 and performs morphological analysis (step S51). The analysis result is sent to the English-Japanese syntax conversion unit 22 via the communication line L2122. Thereby, the English-Japanese syntax conversion unit 22 performs English-Japanese syntax conversion with reference to the English-Japanese bilingual dictionary 24 and the English-Japanese translation rule 25 (step S52), and transmits the result to the Japanese morpheme via the communication line L2223. It is sent to the generation unit 23. The Japanese morpheme generation unit 23 performs a Japanese morpheme generation process and a process of inserting a full spell into a translated sentence (step S53), and when the translated sentence is completed, outputs it from the output unit 3 (step S6).

Next, the configuration and operation of the acronym recognition unit 4 will be described in detail with reference to FIGS.

As shown in FIG. 6, an acronym recognition unit 4 in this embodiment includes an acronym candidate storage unit 41 storing a character string (word) to be checked for an acronym, and an acronym candidate storage unit 41. The acronym pattern storage unit 42 storing an acronym pattern that describes a condition for becoming as a regular expression, and the condition stored in the acronym pattern storage unit 42 is satisfied, but is not actually an acronym Alternatively, for the words stored in the acronym candidate storage unit 41 and the acronym suppression list 44 that stores a list of words that should not be treated as acronyms, the regular expression pattern in the acronym pattern storage unit 42 is used. An acronym pattern for determining whether an acronym can be recognized by determining whether or not the word is present in the acronym suppression list 44 if the matching is successful. It is configured to include a use engine 43.

A communication line L74 extends from the input sentence buffer 7 to the acronym candidate storage unit 41, and the words in the input sentence buffer 7 are sequentially transmitted via the communication line L74. Further, a communication line L46 extends from the acronym pattern application engine 43 to the search result storage unit 6, and the word finally identified as an acronym is stored in the search result storage unit 6 via the communication line L46. Sent and stored.

FIG. 7 shows a part of the contents of the acronym pattern storage section 42. Here, two patterns are described, both of which are regular expression patterns for the surface character string. The first pattern matches a character string in which a set of one uppercase letter and one period is repeated twice or more. The second pattern matches a pattern in which uppercase letters repeat two or more characters.

Next, the operation of these components will be described with reference to FIG.

The acronym recognition section 4 acquires words one by one from the input sentence buffer 7 and repeats the process of determining whether or not the word is an acronym (step S200). When the determination process is performed on all the words, the entire process ends.

First, the words acquired from the input sentence buffer 7 via the communication line L74 are stored in the acronym candidate storage section 41 (step S201).

Next, the acronym pattern application engine 43 is started, and the words in the acronym candidate storage unit 41 obtained via the communication line L4143 and the acronym pattern storage unit 42 obtained via the communication line L4243 are stored in the acronym pattern storage unit 42. Matching of the stored acronym pattern is performed (step S202). The acronym pattern application engine 43 is realized as a general regular expression matcher for a character string.

If the matching fails (step S20)
No in 3), the word is not considered an acronym, and the next loop is started. If the matching is successful (Yes in step S203), the word is compared with the contents of the acronym suppression list 44 (step S204).

In the acronym suppression list 44, words that do not want to be processed even if they match the acronym pattern, for example,
Words that are abbreviations such as “TV”, but cannot be considered as acronyms at present, or that are so common as “USA” that the full spelling of the original word exists, but spelling is indicated. Contains a list of words that would not be better.

As a result of the comparison, if it matches any of the words in the acronym suppression list 44 (Yes in step S204)
s) Start the next iteration to process the next word without considering it as an acronym. If it does not match any of the words (No in step S204), the word in the acronym candidate storage unit 41 is recognized as an acronym, and sent to the search result storage unit 6 via the communication line L46 and stored (step S204). S20
5). Thereafter, the process enters the next repetition of step S200.

FIG. 9 is a block diagram showing an example of the configuration of the original word search section 5. As shown in FIG.
Is a search control unit 51 that performs overall control, an acronym buffer 52 that stores an acronym serving as a search key, a sentence buffer 53 that temporarily stores a sentence to be matched,
A focused character position pointer 54 indicating which character in the acronym buffer 52 is focused on during the matching process, and a focus which holds from which position in the word string in the sentence buffer 53 the matching is started. It is configured to include a word start position pointer 55 and an attention word current position pointer 56 which holds the word position of the current attention target.

The acronym buffer 52 stores the search result storage 6
Communication line L6 for receiving an acronym as a search key
5, the sentence buffer 53 is connected to the input sentence buffer 7 storing the input sentence via the communication line L75, and transmits the searched full spelling original word string to the search result storage unit 6. It has a line L56.

Next, the operation of the original word search section 5 will be described in detail with reference to FIG. The purpose of the original word search unit 5 is to find a word string that is the full spelling of the given acronym in the input sentence. For this purpose, the following processing is repeated for each acronym stored in the search result storage unit 6 in step S2 of FIG. 5 (FIG. 10, step S300).

First, one acronym to be processed is obtained from the search result storage unit 6 via the communication line L65, and is stored in the acronym buffer 52 (step S301). next,
The following processing is repeated for each sentence in the input sentence buffer 7 (step S302).

First, one sentence is obtained from the input sentence buffer 7 via the communication line L75 and stored in the sentence buffer 53 (step S303). Next, the noticed word start position pointer 55 is initialized to 0, and points to the first word of the sentence in the sentence buffer 53. Then, while incrementing the attention word start position pointer 55 by one, it is checked whether or not the word string starting from this position is the full spelling original word string corresponding to the acronym (step S304).

In step S305, it is checked whether or not the word string starting from the position of the attention word start position pointer 55 can be regarded as the full spelling of the acronym. As a result of the comparison, when it is determined that the full spell can be considered (step S
(Yes in 306), a search result storage unit via the communication line L56 for a word string in the range starting from the target word start position and regarded as full spelling (up to the word immediately before the target word current position pointer 56 at the end of comparison). 6 (Step S30)
7). Then, the process for the acronym is completed, and the process for the next acronym (the next repetition of step S300) is performed. On the other hand, if the comparison result is No in step S306, the next repetition of step S304 is performed, and the target word start position pointer 55 is incremented by 1 to compare the word string starting from the right adjacent position.

FIG. 11 is a flowchart showing a detailed processing example of the step of comparing a word string with an acronym in step S305. In this step, in the sentence stored in the sentence buffer 53, it is determined whether the word string starting from the position pointed to by the attention word start position pointer 55 is a full spelling word string corresponding to the acronym stored in the acronym buffer 52. Is determined. Specifically, it is checked whether or not the first characters of the words in the word string are aligned with the acronym character string in order. However, periods in acronyms are ignored during the check.

First, as a preparation, in step S30601, the target word start position indicated by the target word start position pointer 55 is substituted for the target word current position pointer 56, and the target character position pointer 54 is initialized to 0. Make the comparison start at the beginning of both.

Next, it is checked whether the target character position pointer 54 is beyond the end of the acronym character string in the acronym buffer 52 (step S30602). If it exceeds (Yes in step S30602), it means that the comparison has been successfully completed for all of the acronym character strings, and the process returns as success (step S30607). If not exceeded (No in step S30602), the comparison is continued because characters to be compared remain.

First, it is checked whether or not the character in the acronym indicated by the target character position pointer 54 is a period. If it is a period (Yes in step S30603), since it is the next character of the acronym to be compared, the target character position pointer 54 is incremented by 1 (step S30606), and the process returns to step S30602.

On the other hand, if it is not a period (step S30)
(No at 603), an actual comparison is made. Specifically, it is checked whether or not the first character of the word pointed by the current word pointer 56 of interest and the character indicated by the current character position pointer 54 match, distinguishing between uppercase and lowercase characters (step S30604). .

If they match (Yes in step S30604)
s) Since the comparison of the character is successful, the target character position pointer 54 and the target word current position pointer 5
6 is incremented by 1 (step S30605), and step S30605 is executed.
Return to 30602 to continue the comparison. On the other hand, if they do not match (No in step S30604), the process returns as a comparison failure (step S30608).

FIG. 12 is a flowchart showing a detailed processing example of the Japanese morpheme generation processing performed in step S53 shown in FIG. In the Japanese morpheme generation processing, processing such as performing inflection transformation so that the Japanese morpheme constituting the translated sentence is connected to the preceding and succeeding morphemes while observing the constraints is performed. Also, a process of outputting the spelling of the original word to the portion translated from the original word of the full spelling stored in the translation together with the translated sentence is performed.

Specifically, after the inflection processing (step S5302) associated with a general morphological connection rule, it is checked whether or not the corresponding original word is a word stored in the search result storage unit 6. (Step S5303), and if so (Yes in Step S5304), the original word is enclosed in parentheses and connected as a character string immediately after the translated word (Step S53).
05). As a result, this original word character string is translated immediately after the original Japanese word string. Furthermore, in order to prevent this processing from being repeated again when the same word appears at another position, the original word is deleted from the search result storage unit 6 (step S5306). This process is not essential, but if the full spelling of the same word in the original word occurs more than once, the translated sentence becomes difficult to read. On the other hand, step S5304
In the case of No, the processing of the next morpheme is continued without doing anything. The morpheme strings generated in this way are connected and output from the output unit 3 to become a translated sentence.

As described above, the full-spelled original word corresponding to the acronym stored in the search result storage unit 6 appears when the translated word first appears in the translated sentence, in relation to the translated word. As a result, the user can easily understand what spelling the acronym was made from.

Next, the processing from the input sentence of FIG. 13A to the output of FIG. 13B will be described.

First, the original sentence is stored in the input sentence buffer 7. Next, the acronym recognition unit 4 finds an acronym in the input sentence in step S2 of FIG. Here, "FAA" and "NTS"
The two words "B" match the second pattern "continuation of two or more capital letters" in the acronym pattern storage unit 42,
Since it is not listed on the acronym suppression list 44, it is recognized as an acronym and is sent to the search result storage unit 6 and stored. At the time when the processing of the acronym recognition unit 4 (step S2) is completed, the contents of the search result storage unit 6 are such that "FAA" and "NTSB" are stored in the acronym storage buffer unit as shown in FIG. It has become. At this point, the corresponding original word (full spelling word string corresponding to the acronym) has not been stored in the original word storage buffer unit.

Next, in step S 3 of FIG. 5, the original word search unit 5 finds full spelled original words corresponding to the acronyms “FAA” and “NTSB” stored in the search result storage unit 6.
First, “FAA” is loaded into the acronym buffer 52 of the original word search unit 5 in step S301 of FIG. Next, the first sentence is first loaded into the sentence buffer 53 in step S303. Step S305 is executed with the target word position set to 0, but the comparison fails because the first word of the first sentence is "The" while the first character of the acronym is "F", and the process proceeds to No in Step S306. .

Next, the attention word start position pointer 55 is set to 1
The comparison step of step S305 is called with the second word "Federal" of the first sentence pointed to. Then, the word string becomes “Federal Aviation Administration orde
red an… ”. If the first characters are arranged in order,“ FAAo
a… ”, whereas the acronym is“ FAA ”, the comparison is successful and the program returns.
Then, the process branches to Yes, and the partial word string that has been successfully compared, and the first to third words are transmitted to the search result storage unit 6 as the corresponding original word strings.

Similarly, for "NTSB", the comparison from the second word of the third sentence succeeds. As a result, when the processing of the original word search section 5 (step S3) is completed, the contents of the search result storage section 6 correspond to the acronym "FAA" and the original word string "Federal Aviation" as shown in FIG. Administration "corresponds to the acronym" NTSB "and the original word string" National Transportati
on Safety Board "is stored.

In this state, the English morphological analysis of FIG. 5 (step S51) and the English-Japanese translation (step S52) are performed.
The Japanese morpheme generation (step S53) is activated.

In step S53, "Federal Aviation
When processing the translation "Federal Aviation Administration" corresponding to "Administration", this original "Federal Aviation Admin
It is checked that "istration" is stored in the search result storage 6 as the original word string for "FAA".
The original word “Federal Aviation Adm” immediately follows “Federal Aviation Administration”.
Inistration "is enclosed in parentheses and concatenated as a character string. As a result, the original Japanese character string" Federal Aviation Administration "will appear in the translated text as" Federal Aviation Administration ". Become.

"National Transportation Safety Board"
Similarly, for the National Transport Safety Board,
The surface character strings are concatenated to form the "National Transport Safety Board (Nation
al Transportation Safety Board) "
The output sentence shown in FIG. 3 (b) is obtained. In this translation, it is quite clear to the user what abbreviations "FAA" and "NTSB" are and which translations they correspond to.

Next, a second embodiment of the present invention will be described with reference to FIG. The difference between the second embodiment and the first embodiment is that the second embodiment has an input sentence editing unit 8 which does not exist in the first embodiment, the search result storage unit 6 and the input sentence A communication line L68 connecting the editing unit 8; a communication line L78 connecting the input sentence buffer 7 and the input sentence editing unit 8; a translation unit 2A instead of the translation unit 2; The point is that the search result storage unit 6 and the Japanese morpheme generation unit 23 are not connected by the communication line L62. Further, the translation unit 2A
Can be specified as a unit and a non-translated section by pre-editing the original sentence. The difference is that in the first embodiment, the Japanese morpheme generation unit 23 performs the process of generating the spelling of the full-spelled original word corresponding to the acronym in the translated sentence, whereas the second embodiment This is because the input sentence editing unit 8 performs the pre-editing of the input sentence. Specifically, a marker is inserted around the target word string, and the target is achieved by specifying the unit and non-translation.

The operation of this embodiment is almost the same as that of the first embodiment, as shown in the flowchart of FIG. The difference is that the full spelling word sequence corresponding to the acronym is searched,
Immediately after step S4 stored in the search result storage unit 6 and immediately before translation step S5, step S4A for rewriting the input sentence using the information of the full spelling word is provided. Regarding the operation of this input rewriting step,
This will be described in detail with reference to FIG. FIG. 13 as input
It is assumed that the English sentence shown in FIG.

When the step S4A is started, the input sentence buffer 7 contains the English sentence as it is, and the search result storage section 6 stores the contents shown in FIG. The processing of S4A01 and subsequent steps is repeated for each of the original word strings stored in the search result storage unit 6.

First, the original word string is referred to as “Federal Aviation Admi”.
This word string is retrieved from the input sentence buffer 7 as "nistration" (step S4A02). The first word found is the position of the second word in the first sentence. A "designation of unity" which is a symbol for pre-editing is inserted (step S4A0).
3). Further, immediately after this word string, a copy of the word string sandwiched between non-translated symbols and further enclosed in parentheses is inserted (step S4A04).

When the processing of this word is completed, the same processing is repeated for another original word string “National Transportation Safety Board”. As a result, the contents of the input sentence buffer 7 at the end of step S4A are as shown in FIG. "FederalAviation Administration" in the first sentence
Was pre-edited and [[Federal Aviation Admini
stration << (Federal Aviation Administration) >>]]
And the third sentence "National Transpor
tation Safety Board "is pre-edited and [[Nati
onal Transportation Safety Board << (National Transp
ortation Safety Board) >>]] Please be careful. here,
The section enclosed by [[…]] is a symbol indicating that the section enclosed by <<… >> is to be output as it is in the original notation without translation, and the section enclosed by <<… >> Are also inserted in step S4A. By performing the translation process in accordance with the designation of the unit and the untranslated symbols, the sentence shown in FIG. 13B is finally output.

The present invention can be implemented in various modifications other than those described above.

First, in addition to the method shown in the embodiment (showing the full spelling of the original word enclosed in parentheses immediately after the full spelling word), for example, the method of indicating the correspondence between full spelling words and acronyms Indicate the spelling of the original language with * in the footnote by adding a symbol such as * to the footnote. As an option for selecting the translation of the full spelling word, show the original spelling when the user specifies the translation display, the original spelling as the ruby of the translation Various modifications, such as showing the translated word and the original spelling in association with each other and displaying the translated word in a window different from the translation display window, can be easily implemented without affecting the gist of the invention.

In the embodiment, the method of identifying acronyms has been described using a regular expression pattern of character strings and a suppression list. However, an attribute other than a character string, such as a part of speech, may be used for pattern description, or a general program may be used. It can be implemented using various recognition methods such as expressing as.

Further, in the embodiment, the search for the full spelling original word string corresponding to the recognition of the acronym is performed on the character string of the input sentence. Data with estimated grammatical attributes such as part of speech,
By processing the parsed data, it is easy to implement the acronym recognition and the original word search and recognition with higher accuracy. For example, as a full spelling original word, a modification such as allowing only a unit recognized as a compound noun phrase in syntactic analysis can be considered, but this can also be easily implemented within the framework of conventional machine translation technology .

Further, the case where the acronym is registered in the bilingual dictionary is not described in the embodiment, but the acronym registered in the bilingual dictionary is translated based on the dictionary, or , Translate based on a bilingual dictionary,
Modifications such as the user being able to freely select whether to use the method of the present invention or the like can also be easily implemented. As described above, the present invention can be variously modified and implemented within the scope of the invention.

[0087]

According to the present invention, even when an acronym unknown to the user is included in a translation, if a full-spelled word corresponding to the acronym appears in the original text, the user can delete the translation. You can easily understand what the acronym means simply by reading. Even if the acronym is not registered in the dictionary, the feature of the present invention that it is effective is a great advantage compared with the conventional technology that requires dictionary registration, in view of the fact that the metabolism of the acronym is very intense. I have.

Further, as another advantage and effect of the present invention, there is a robustness that a translated sentence is not destroyed even when a correspondence between an acronym and a full spelling word is incorrectly recognized. In this case, in the present invention, only the extra words that are not related to the acronym are output with the original sentence notation, and the translated text itself is no different from that output by the machine translation unit that is the base of the present invention. Because there is no. In other words, only unnecessary clues are shown, and the translation itself does not contain errors. The user can ignore such cues and interpret the acronym correctly. This advantage essentially comes from the inventive feature of devising the output of full-spelled words corresponding to acronyms. On the other hand, in the case of a method of devising the translation of an acronym, such as a system that reads and translates an acronym with the full spelling of the original word and outputs it, erroneous recognition of this correspondence leads to an incorrect translation immediately. This leaves the user with no room for correct interpretation. As described above, the present invention has a great advantage and effect that it is robust against an error in identifying the relation between an acronym and an original word.

Further, the machine translation apparatus of the present invention recognizes a word stored in the acronym pattern storage unit that satisfies the condition for becoming an acronym and is not stored in the acronym suppression list as an acronym. As a result, it is possible to prevent the full spelling of an acronym such as "USA", which is very common, from being inserted into the translated sentence, thereby making it difficult to see the translated sentence, and to shorten the time required for the translation process. .

Further, the machine translation apparatus according to the present invention is arranged such that the full spelled word searched for in the original word search section is inserted into the translation in association with the translation of the input sentence when the input sentence is translated. Since an input sentence editing unit for pre-editing a sentence is provided, an existing translation unit can be used.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a machine translation device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing processing of the machine translation device according to the first embodiment of the present invention.

FIG. 3 is a block diagram illustrating a configuration of a machine translation device according to a second embodiment of the present invention.

FIG. 4 is a block diagram illustrating a configuration of a machine translation apparatus according to the first embodiment of the present invention.

FIG. 5 is a flowchart showing processing of the machine translation apparatus according to the first embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration example of an acronym recognition unit 4.

FIG. 7 is a diagram showing an example of contents of an acronym pattern storage unit 42.

FIG. 8 is a flowchart showing a processing example of the acronym recognition unit 4;

FIG. 9 is a block diagram illustrating a configuration example of an original word search unit 5;

FIG. 10 is a flowchart showing a processing example of the entire original word search unit 5;

FIG. 11 is a flowchart showing a detailed processing example of a word string and acronym comparison step performed by an original word search unit 5;

FIG. 12 is a flowchart showing a processing example of a Japanese morpheme generation unit 23;

FIG. 13 is a diagram illustrating an example of input and output of the machine translation device. (A) shows an input sentence, and (b) shows an output sentence.

FIG. 14 is a diagram exemplifying the contents of a search result storage unit immediately after the processing of the acronym recognition unit 4 is completed.

FIG. 15 is a diagram exemplifying the contents of a search result storage unit immediately after the processing of the original word search unit 5 is completed.

FIG. 16 is a block diagram illustrating a configuration of a machine translation device according to a second embodiment of the present invention.

FIG. 17 is a flowchart illustrating a processing example of the machine translation device according to the second embodiment.

FIG. 18 is a flowchart showing a processing example of the input sentence editing unit 8;

FIG. 19 is a diagram showing an example of the contents of an input buffer 7;

FIG. 20 is a diagram illustrating an example of input and output in a conventional machine translation apparatus.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input part 2, 2A ... Translation part 3 ... Output part 4 ... Acronym recognition part 5 ... Original word search part 6 ... Search result storage part 7 ... Input sentence buffer 8 ... Input sentence editing part 21 ... English morphological analysis part 22 ... English-Japanese syntax conversion unit 23 ... Japanese morpheme generation unit 24 ... English-Japanese bilingual dictionary 25 ... English-Japanese translation rule 41 ... Acronym candidate storage unit 42 ... Acronym pattern storage unit 43 ... Acronym pattern application engine 44 ... Acronym Suppression list 51 Search control unit 52 Acronym buffer 53 Sentence buffer 54 Attention character position pointer 55 Attention word start position pointer 56 Attention word current position pointer A1 Input device A2 Data processing device A3 Output device A4 ... Storage device A5 ... Recording medium

Claims (9)

[Claims] 1. An acronym recognition process for recognizing an acronym existing in an input sentence to be translated, and an original word for searching a full spelling word corresponding to the acronym recognized in the acronym recognition process from the input sentence Generating a translated sentence for the input sentence, including a translation process of outputting the original spelling of the full spelled word searched for in the original word search process in association with the translated word of the word; Machine translation method. 2. The machine translation method according to claim 1, wherein in the translation processing, the original spelling of the full spelled word searched in the original word search processing is inserted immediately after the translated word of the word. . 3. An acronym recognition process for recognizing an acronym existing in an input sentence to be translated, and an original word for searching the input sentence for a full spelling word corresponding to the acronym recognized in the acronym recognition process. An input for pre-editing the input sentence so that the full-spelled word searched for in the original word search process is inserted into the translated sentence in translation of the input sentence when the input sentence is translated. A machine translation method comprising: a sentence editing process; and a translation process for translating an input sentence that has been preedited in the input sentence editing process. 4. An acronym recognition unit that recognizes an acronym existing in an input sentence to be translated, and an original word that searches the input sentence for a full spelling word corresponding to the acronym recognized by the acronym recognition unit. A search unit, and a translation unit that, when generating a translated sentence for the input sentence, outputs the original spelling of the full-spelled word searched for by the original word search unit in association with the translated word of the word. Machine translation equipment. 5. The apparatus according to claim 4, wherein the translating unit inserts the original spelling of the full-spelled word searched by the original word searching unit immediately after the translated word of the word. Machine translation device. 6. An acronym recognition unit, comprising: an acronym pattern storage unit in which a condition for being an acronym is stored; an acronym suppression list in which words that should not be treated as acronyms are stored; 5. An acronym pattern application engine that satisfies the conditions stored in the acronym pattern storage unit and identifies words that are not stored in the acronym suppression list as acronyms. Machine translation equipment. 7. An acronym recognition unit that recognizes an acronym present in an input sentence to be translated, and an original word that searches the input sentence for a full spelling word corresponding to the acronym recognized by the acronym recognition unit. A search unit; and an input for preediting the input sentence so that the full-spelled word searched for by the original word search unit is inserted into the translated sentence in translation of the input sentence in relation to the translated word. A machine translation device comprising: a sentence editing unit; and a translating unit that translates an input sentence that has been pre-edited by the input sentence editing unit. 8. An acronym recognition process for recognizing an acronym existing in an input sentence to be translated by a computer, and a full spelling word corresponding to the acronym recognized in the acronym recognition process is searched from the input sentence. An original word search process, when generating a translated sentence for the input sentence, a program for executing a translation process for outputting the original spelling of the full spelled word searched for in the original word search process in association with the translated word of the word A recorded machine-readable recording medium. 9. An acronym recognition process for recognizing an acronym existing in an input sentence to be translated by a computer, and a full spelling word corresponding to the acronym recognized in the acronym recognition process is searched from the input sentence. An original word search process, when generating a translation for the input sentence, a translation process for associating the original spelling of a full spelled word searched for in the original word search process with a translation of the word and inserting it into a translation, the input sentence Machine-readable recording medium storing a program for executing a translation process of inserting the original spelling of a full-spelled word searched for in the original word search process immediately after the translated word when generating a translated sentence for the word .