JPS63136264A - Mechanical translating device - Google Patents

Abstract

PURPOSE:To easily look through overall working and to shorten working time with a mechanical translating device by retrieving collectively all unknown words in an input sentence before a translation process is started and registering those words totally into a user dictionary. CONSTITUTION:One of the input sentences displayed on a CRT 3 is read into a processing buffer and unknown words are retrieved. A single word is specified when the words to be retrieved are not through. The specified word is collated with the contents of each type of translation dictionaries selected previously by an operator for execution of the dictionary consulting and changing processes. If the specific word is not contained in the translation dictionary, this word is added to a list of unknown words or not treated as an unknown word by the state of a flag. If the flag is ON, the word is treated as an unknown word and as a proper noun. While the word is directly decided as an unknown word when the flag is OFF. Then the spelling of the word and the information on the emerging position of the word are written to a result buffer together with insertion of an unknown word mark. When the processing is through with all sentences, the contents of the result buffer are sorted and arranged for easy-to-see edition. Thus a list of unknown words is produced and displayed on the CRT 3.

Description

[Detailed Description of the Invention] Technical Field> The present invention relates to a machine translation device, and before starting translation processing,
Search for unknown words (words not listed in the translation dictionary, including words spelled incorrectly due to input errors) that exist in the input sentence, and as a result, calculate the total number of unknown words,
The present invention relates to a device that outputs a list of unknown words edited together with information such as appearance positions.

<Prior Art> Conventional translation devices have been able to detect unknown words present in an input sentence only when the input sentence (original text) is sequentially subjected to translation processing. For this reason, ■ When translating in an interactive mode, the translation process is interrupted every time an unknown word is detected, and it is necessary to register each word in a user dictionary, etc., which slows down the flow of translation work and makes it difficult to operate. A loss occurs.

- When translating in conversational format, input sentences containing unknown words may not be translated accurately or remain untranslated at all, and the translation process must be performed again after checking the results. No.

There were drawbacks such as.

To further discuss ■, in the case of interactive translation,
When a translation instruction is issued for an arbitrary sentence, translation processing is executed, and unknown words are detected at the dictionary lookup and morphological analysis stages. Therefore, it becomes impossible to continue the process after that, and the translation process stops. The operator must take measures against this, such as registering the unknown word in the user dictionary, and then issue the translation instruction again.

To further discuss ■, in the case of translation in the -talk format, translation processing is performed for each sentence in the sentence for which a translation instruction has been issued, but for sentences containing unknown words, translation processing is performed first. For the same reason as in interactive translation, the translation of that sentence is stopped and the translation of the next sentence is started. After the batch translation is completed, the operator registers the unknown words contained in the text in the user dictionary one by one, searches for sentences that could not be translated due to unknown words, and avoids issuing translation instructions again. I shouldn't have.

As mentioned above, the time loss caused by using the translation instruction sound again and the time loss caused by searching for sentences that were not translated due to unknown words impede the smooth flow of translation work by the operator. Ta.

<Purpose of the Invention> The present invention provides a machine translation device that detects all unknown words existing in an input sentence before starting translation processing, and makes it possible to register all unknown words in a user dictionary at once, thereby simplifying the entire work. The purpose is to make it easier to see and to shorten the working time.

<Example> Hereinafter, the configuration of the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a translation device that can implement the present invention. A central processing unit (C, PU) that performs processing such as translation processing.
A main memory 2 in which programs and the like for processing with I are stored, a display device (CRT) 3 for displaying 2 necessary for translation processing, and a keyboard 4 for performing input necessary for translation processing are provided. There is. Depending on the system, an optical character reader (OCR) may also be used as an input device. Furthermore, a translation module 5 in which information necessary for the translation process is stored, a dictionary 1 for translation, grammatical rules, and a structure conversion rule table 6 are provided.

The translation dictionary includes a basic word dictionary containing basic words, a technical term dictionary containing specialized words, and a user dictionary in which users have registered their own words.

Next, a general procedure for machine translation processing will be explained.

FIG. 2 is an example of a block diagram of main parts of machine translation processing. The process generally proceeds in the direction of Sl→S2→s3→S4→S5. In dictionary lookup and morphological analysis (Sl), a translation dictionary is looked up to obtain grammatical information such as part of speech for each word, translation, and analysis of tense, person, number, etc. In syntactic analysis (S2), the input sentence written in the French language is analyzed based on grammatical rules, etc., including dependencies between words.
Determine the structure of the sentence. There may be multiple solutions depending on the sentence.

In the semantic analysis (S3), it is verified whether the result of the syntactic analysis is semantically correct. If there are multiple results of syntactic analysis, the most valid interpretation is selected by semantic analysis.

The level of semantic analysis differs depending on the contents of the semantic processing dictionary that the translation system has, the internal representation of the analysis results, etc. There are also methods that organically combine syntactic analysis and semantic analysis.

In the conversion (S4), the data structure (mostly structure) obtained by the analysis of the source language is converted into the data structure of the target language.

In generation (S5), a sentence expressed in the target language is generated based on the data structure of the target language created by the conversion unit. Generally, this sentence becomes the translated sentence obtained as a result of the translation process.

FIG. 8 is a flowchart of batch unknown word search processing.

The explanation will be given below according to the flowchart.

Here, it is assumed that the target input sentence has already been read into the machine translation device and its contents are displayed on the display device 3.

In step S, the position of the search start sentence is specified.

The above-mentioned position designation process is normally performed by the operator referring to the display content on the display device 3 and moving a cursor.

If the entire sentence is to be searched, place the cursor at the beginning of the sentence.

In the next step 82, a start key is input to start the following batch unknown word search function.

A function key or the like provided on the keyboard 4 is assigned to the start key.

Next, the main process begins at step S3, and it is determined (checked) whether the target input sentence has been completed.

If it is determined in step S3 that the input sentence has ended, the process advances to step S15. This part will be explained later.

On the other hand, if it is determined in step S3 that the input sentence is not completed and there are remaining parts, the process moves to step S4.

In step S4, one sentence is read into a processing buffer (READ).

Then, an unknown word search is performed on the one sentence.

Generally, the first word is checked against a translation dictionary in order.

In step S5, it is determined whether or not the search target word has been searched.

Here, if the word to be searched has ended, the process branches to step 83 to read the next sentence.

If it is determined in step S5 that the word does not end, the process moves to step S6 and one word to be searched is specified.

Next, in step S7, the identified word is compared with the contents of various translation dictionaries.

Although there may be a plurality of translation dictionaries, it is assumed that an appropriate variety of dictionaries have been selected by the operator in advance before starting this function.

A dictionary comparison process is described in steps S8 to Slo.

For example, the inflection process in step S8 includes English plurals,
Third person singular present tense, past tense, past participle.

It corresponds to dictionary matching that includes the basic form (original form) of a certain word, such as the present participle form, and even the original deformed forms. It is also possible to further deepen the level of processing and, for example, perform matching that includes prefixes and suffixes.

Step S9 and step 510 are processes related to proper noun notation in English, for example.

As a matter of course, proper nouns have a high probability of becoming unknown words.

For example, if you consider the word "Tanaka" and it does not exist in the translation dictionary, you can treat it as an unknown word and add it to the list of unknown words described below, or you can assume that it is a proper noun and treat it as an unknown word. There are two ways of thinking about whether or not it should be treated as such. In order to select one of these, a flag is provided and set by the operator before starting this function. If you choose the former, register it in the user dictionary as "Tanaka'k" during the day.This may be when a more detailed translation is required.On the other hand, if you choose the latter, enter 'T
It is considered that simply translating "anaka" as the noun "Tanaka" is sufficient for the purpose of the translation. Although the proper noun estimation method differs depending on the machine translation device, the proper noun estimation algorithm of each machine translation device can be used for the proper noun processing of this function.

In step S9, it is determined whether the flag is on or off, and if it is on (=Yes), a process is performed in which the word estimated to be a proper noun is treated as an unknown word (S+).

step).

If the flag is determined to be off (=No) in step S9, the process directly proceeds to step S11, where it is determined whether the word is an unknown word.

Depending on the judgment result, from step S12 to SI3 → S14
The process proceeds step by step, and from the 512 steps to the S
The process returns to Step 5 and branches to the process of starting the processing of the next target word.

In step S13, information such as the spelling and appearance position of the word is written into the result buffer, and in the next step S1+, an unknown word mark is inserted into the word in order to clearly indicate the unknown word in the input sentence.

The step S14 makes it easier to search for unknown words by displaying or printing out the input sentence, and it is of course possible to omit the step S14 depending on the machine translation device.

The above processing is continued, and if it is determined in step S3 that processing of all sentences has been completed, the process advances to step S15.

In the S+s step, the contents of the result buffer (cumulative data of unknown words) are classified, organized, and edited into an easy-to-read format to create an unknown word list If. Possible ways to do this include counting the number of occurrences of words, clarifying the position of occurrence in a sentence, and sorting (classifying) spellings in alphabetical order.

In step SI6, the unknown word list' is output to the CRT display device 3. This output can also be done by a printing device.

The above-described flow completes a series of processes of this function.

FIG. 4 shows an example of a display when this function is executed.

In the display example in the same figure, the word "parchessi'" does not exist in the translation dictionary and is an unknown word.
The unknown word mark of step s+4 is surrounded by rice.

In addition, a comment ``There is only one word with rice in the dictionary'' is displayed in the translation area to clearly indicate the detection of an unknown word.

FIG. 5 shows another display example.

In this display example, "beginners" is a misspelled word (originally "beginners") and is therefore searched as an unknown word. Furthermore, as mentioned above, it goes without saying that RASIC'' in the example sentence can also be treated as an unknown word by using the proper noun processing flag.

FIG. 6 is an example of an unknown word list. The spellings are sorted in alphabetical order, and information such as the number of the word from the beginning of the sentence number 9 in which it appears, the total number of unknown words, the number of different words, etc. are restored. Also, depending on the purpose, it is of course possible to list in order of appearance (order of sentence number) or order of the largest number of items instead of alphabetical order.

If you register the user dictionary using this unknown word list,
It becomes possible to carry out highly efficient translation work. or,
If the unknown word list is called up when registering in a user dictionary and the English spelling part is used, manual input can be reduced and the burden of registration work can be reduced.

According to this embodiment, first, prior to translation processing, all unknown words inherent in the text are searched at once, and the results are edited together with useful information such as the number of unknown words and the location of the unknown words.
Since the operator is notified, countermeasures such as user dictionary registration can be taken all at once, making the work flow more efficient and making it easier to use.

The above embodiment was explained using English-Japanese translation as an example, but the concept of an unknown word for a machine translation device is common regardless of the source language, and is not limited to English-Japanese translation, but Japanese-English translation,
Furthermore, it is clear that it can be applied to other language bears as well.

Effects> As described above, according to the present invention, in a machine translation device that performs translation processing on an input sentence, before executing translation processing,
Since the present invention includes a collating means for collating each word constituting an input sentence with a translation dictionary, and an output means for outputting words that do not exist in the dictionary based on the collation result, the following effects can be obtained.

(b) Unknown words in the input sentence that do not exist in the translation dictionary can be registered in the user dictionary all at once, improving work efficiency.

(b) Since you can know in advance the location and number of unknown words,
It serves as a reference for determining translation words when registering in a user dictionary, and as a result, high-quality translations can be obtained.

H9: Since the work can be divided into parts such as batch unknown word search → user dictionary registration → translation processing, the work can be done by multiple people, i.e.
It becomes possible to divide labor according to the work level.

) When the input text is English, etc., by changing the function level of the batch unknown word search, proper nouns that include capital letters can be listed as unknown words, and by registering them as a user, detailed translations can be obtained. can get.

(b) If the unknown words restored as unknown words during the batch unknown word search are stored by some means, and the unknown words are called up at the time of user dictionary registration, the efficiency of registration will be improved.

(f) Since input errors also result in unknown words, mistakes can be corrected in advance.

[Brief explanation of the drawing]

Figure 1 is a block diagram of the machine translation device, Figure 2 is a diagram for explaining translation processing, Figure 3 is a flowchart showing batch unknown word search processing, and Figures 4 and 5 are display examples of unknown words. , FIG. 6 is a diagram showing a list of unknown words. 1...CPU, 2...Main memory, 3...C
RT. 4...Keyboard, 5...Translation module, 6...
·table. Agent Patent Attorney Takeshi Sugiyama (and others!) Figure 1 6 Figure 2 - Figure 4 Figure 5

Claims (1)

[Scope of Claims] 1. In a machine translation device that performs translation processing on an input sentence, before executing the translation processing, a collation means that collates each word constituting the input sentence with a translation dictionary; A machine translation device comprising: output means for outputting words that do not exist in the dictionary based on the matching result. 2. The machine translation device according to claim 1, wherein the output means is a display means. 3. The machine translation device according to claim 1, wherein the output means is a printing means.