JPS6188368A - Translation system - Google Patents

Abstract

PURPOSE:To correct a part of speech, to translate an input sentence in accordance with the pat of speech after correction and to obtain a precise translation result with a simple operation by displaying the input sentence to be translated and the part of speech corresponding to respective words of the input sentence. CONSTITUTION:To a processing device 1a of a translation system, a keyboard 2a, which key-inputs the data, a memory 3a which stores the data related to the processing device 1a and a display device 4a, which displays the data processed at the processing device 1a, are connected. Further, to the processing device 1a, a translation module 5, which has an original, a consulting of the dictionary, sentence structure, a sentence structure analysis, a sentence structure generation and result buffers 51a-55a and a dictionary table 56a, is connected. In accordance with the operation of the keyboard 2a, the input sentence translated by the processing device 1a and the part of speech corresponding to respective words are displayed at the display device 4a. In accordance with the operation of the keyboard 2a, the part of speech is corrected by the translation module 5, translation is executed in accordance with the pat of speech after correction and the precise translation is performed with an easy operation.

Description

[Detailed Description of the Invention] When the field of industrial application was in its infancy, the idea of an interactive one-person translation system in which machines and humans worked together to create correct translations was popular.

BACKGROUND TECHNOLOGY □ In general, machine cutting is performed through a process as shown in FIG. The original text in the input language to be translated needs to be analyzed during the translation process. There are three stages of analysis: form analysis, syntactic analysis, and semantic analysis. Form element analysis refers to a dictionary for machine translation, extracts grammatical information such as part of speech, and translation information for each word.
This involves analyzing the person, number, tense of a sentence, etc. In syntactic analysis, the structure of a sentence is analyzed by checking the subordination relationships between each word. What is semantic analysis?
The purpose is to determine what is correct and what is not from multiple syntax analysis results. Machine translation performs analysis to any of the three levels to obtain the internal structure of the input language, then converts it into the same level internal structure of the output language that makes up the translated sentence according to that structure, and then converts it to the same level internal structure of the output language that makes up the translated sentence. is generated. The sophistication of machine translation differs depending on the depth of this analysis level. Those that only perform morphological analysis are unable to translate sentence by sentence, but only by word by word, as typified by so-called electronic translators. Although K, which performs syntactic analysis, outputs all grammatically correct interpretations, it ends up outputting a wide variety of t+++J == AM results, leaving humans with no choice but to determine the correct answer.
increases. Those who go as far as interpreting meaning! ♀: Although theoretically it is possible to output only one correct W translation result, it requires the machine to colorize a huge amount of information in Vi, which is practically impossible.

The machine translation that is the object of the present invention is defined as at least the level of syntactic analysis. In other words, the sentence unit tl
In translation is possible, grammatically correct interpretation can be outputted,
Semantic analysis may also be performed, but it is not complete and represents a machine + i, ji translation that does not narrow down the output to a unique one.

FIG. 8 is a block diagram showing the configuration of a prior art translation device. A keyboard 2 for manually inputting data related to the processing device 1, a memory 3 for storing data related to the processing device 1, a display device 4 for displaying the results processed by the processing device 1, and a tin J The 1h monthly translation module 5 in which data regarding < is stored is t'm. The translation module 5 includes an original text buffer 511iII? that stores the input text. A static subtraction buffer 52 that stores the result of writing, a syntax generation buffer 54 that stores the result of converting the contents of the syntax analysis buffer 53 that stores the input word and 1A (sentence analysis result) into output reading and use.
It includes a result buffer 55 for storing the results of generating the output six-fold form, and a table 56 consisting of stationary, sentence rules, etc. for machine translation.

Figure 9 shows that when the input sentence is translated,
row.

, the contents stored in each buffer 51 to 55 in the prior art are shown. FIG. 9 (l; indicates the same contents of the original text buffer 51 and ^f: confession buffer 52.

Based on the 4L text input to the original text buffer 151, the dictionary lookup buffer 52 retrieves grammatical information such as part of speech and translation information corresponding to each word. The parsing buffer 53 shown in FIG. 9(2) parses the input sentence based on part-of-speech information given to each word. FIG. 9 (3) K is the syntax generation buffer 54 and the result buffer 5
5 is shown. The result of narrowing the contents of the syntax analysis buffer 53 for N output words is stored in the syntax generation buffer 54, and the translation result is stored in the result buffer 55 based on the generated syntax.

Fig. 1θ shows a display image j by the display device 4 in the prior art.
FIG. As shown in Figure 10 (1), the input sentence to be d F is input manually.
I was seen by someone. When the iI+1 translation key is operated, the
(2) ↑ 1-translation result is displayed as shown in 1st O-ward It61. Each display-plane 1θ figure (2) ~'5Sl
The translated sentence shown in Figure 16) is the result of translating each word of the input sentence by recognizing it as having each part of speech as shown in Table 1 below.

(Margin below) Table 1 (Margin below) These are all grammatically correct interpretations.

Of course, humans can judge that the interpretation shown in Figure 1θ (6) is correct. If the semantic analysis is complete, the translated sentence will be as shown in Figure 10 (61).For that reason, (a) Arrows and flies are not similar.

[b) Arrows have no ability to change time.

(c)! 5:11 does not like arrows.

Is it necessary to remind Kayo of this wisdom? It is obvious that it is impossible for a machine to memorize all such knowledge about the real world.

Therefore, it is obvious that there are solutions that should not be compatible at the conventional level of machine translation.

As described above, the conventional @machinetranslation system has a drawback in that, since there are many solutions, the translation key must be pressed many times until the correct answer is obtained. Furthermore, in machine translation, it takes a long time to get the correct answer because it has to search for all kinds of possibilities and find all the possible solutions. For this reason, a method is adopted in which the solutions are output in order, starting with those found.

Therefore, when the first solution is output in FIG. 10 (2), how many remaining solutions exist Vi,! l! IJ is not clear. It is not possible to display things like the same sound numbers in kana to kanji conversion. For this reason, not only does it have the disadvantage of having to press the execution key many times, but it also has the disadvantage of not being able to get the correct answer even after pressing the translation key a maximum of times. Therefore: 1. The author's mental burden is quite large. From this point of view, the total number of solutions can be reduced to 1 with simple operations.
This means that we are encouraged to reduce it at all times.

Problems to be Solved by the Invention In the above translation results, the mistranslations shown in Figures 10 (2) to (5) are caused by incorrect part-of-speech recognition of multi-product@words. I understand that. That is, in the example of FIG. 1θ (2), the part of speech is incorrectly recognized as rrliesJ as a noun. Also, Figure 10 (3) and! Figure 10 (
In the example 4), the part of speech is incorrectly recognized as ``TimeJ is a # word, and the part of speech is incorrectly recognized as ka rfliesJ is a name Th'1.Furthermore, in the example in Figure 10 (5), `` They incorrectly identified the part of speech for "fliesJ" as a noun, and incorrectly determined the part of speech for "l"'1ikej as a verb.

However, such errors can be easily noticed by a person who has the ability to judge the success or failure of a translation by looking at it.

The purpose of the present invention is to eliminate the variability of translation results in machine translation, and to allow a person to quickly give instructions.
This provides a translation method that can obtain correct translation results.

Means to Solve the Problem The present invention displays the input sentence to be translated, displays the quality of each vulgar word -J in correspondence with the 41 valley words of the input sentence, and calculates the quality of tJ11. This translation method is characterized in that the input sentence is translated according to the corrected sentence sequence.

Operation According to the present invention, the input sentence to be translated is displayed and l[
It is possible to display the part of speech of each word in correspondence with each word in the J entry blade sentence, and modify the part of speech.
Since the input sentence is translated according to 4iJ, correct translation results can be obtained with simple operations.

Embodiment FIG. 1 is a block diagram of a &) J board device constructed in accordance with the present invention. Associated with the processing device 1aK are a keyboard 2a for manually inputting data, and a memory 3a for storing data reaching the processing device IaglP1.

A display device 1i4a for displaying the results processed by the processing device ILa' and a translation module 5a in which data for translation are stored are installed. Ar1? :
The module 5aK stores the result of dictionary lookup in the original text buffer 51a1 which stores the input text.

Writing buffer 52a1 Syntax analysis buffer 53a1 that stores the result of manual syntax analysis
Contains a syntax generation buffer 54a1 that stores the results of converting the contents of a into the output language; a result buffer 55a1 that stores the results of morphological element generation of the output d-word; a table 55a that includes a dictionary for machine translation, grammar rules, etc. .

FIG. 2 is in accordance with the invention! 1iIJ is a flowchart for explaining the work. FIG. 3 also shows the display LL! of the display device 4・a according to the present invention. An example of 11 songs is shown. First, in step n1, the next candidate flag is cleared prior to inputting the ii+In statement. This flag is
As will be described later, even if η'' sentences are analyzed for all part-of-speech combinations, an error message will be displayed if no solution is obtained.

Next, when the operator operates the key 6 of the keyboard 2a, it is determined in step n2 whether to input the 41st message of the input sentence or input the translation key for instructing translation. When inputting an obscene word, the process moves to step n3, where the word is displayed on the display means 4a via the processing device 1a, and at the same time, the character code is sent from the processing device 1a to the one-translation module 5a. At step n4, the character code input to the translation module 5a is stored in the syntax buffer 51a, and the dictionary of the table 56a is looked up to perform morphological analysis.

When the morphological analysis is performed and the process moves to step n5, the analysis result is stored in the dictionary lookup buffer 52a. While inputting the original text continues in this way, step n2. n3
．． n4. The processing operation n5 is repeated.

When the input of the original text is completed and the operator issues a translation instruction by operating the translation key in step n2, the process moves to step n6. FIG. 3 (1) V'C shows a state in which the original text for which word input has been completed is displayed.

In step n6, the combination of parts of speech is initialized. In step n7, syntax analysis is performed with reference to the parsing rules in the table 56a, and the analysis results are stored in the syntax analysis buffer 56.
Stored in a. In step n8, it is determined whether or not the analysis was successful. If the sentence analysis is successful, the process moves to step n9. In step n9, the translation module 5a
Structural conversion is performed while referring to the conversion rules and generation rules in table 56a. Then step nlO
Then, syntax generation is performed and the syntax generation buffer 54a is completed. And in step nil, the result buffer 55
a is completed, and the result is displayed on the display device 4aK in step n12. In step n13, the part of speech of each column in the sentence is displayed with reference to the contents stored in the horizontal sentence analysis buffer 53a. FIG. 3+2+Vi shows a state in which the part of speech of each word and its translation are displayed corresponding to the original text.

Moving on to step n14, it is determined whether the operator has given an instruction to accept or correctly correct the part of speech of each vulgar word! 141 cut off. If there is no instruction, the process moves to step n18, and if an instruction is given, the process moves to step n15, where the part of speech for which the acceptance or correction instruction was given is displayed. Figure 3+31K shows a state in which "TimeJ" was displayed as a noun in step n13, and the operator operated key 7 in step n14 to instruct acceptance. For each of the 41 words, the parts of speech other than the part of speech for which the acceptance instruction was given are cut from the dictionary lookup buffer 52a.In the embodiment shown in FIG. 3, the part of speech of "Time" is cut. Next, in step n17, the combination of parts of speech of each word is reset according to the contents of the operator's instructions. In step n18, it is determined whether or not the operator inputs the next candidate key 9 to generate the next candidate when the translation result is being asked. If the next candidate key 90 has been input, the process moves to step n19 and the next candidate flag is set. Like this step n
In step 7, after the syntax analysis is successful even once and the translation result is displayed, when the operator requests another translation result, the next candidate flag is set. step n
From 19 onwards, syntactic analysis progresses again in step n7, but the syntactic analysis buffer 53a stores the previous combination of parts of speech and the result of syntactic analysis, and if the combination of parts of speech is unchanged from the previous one, a different syntactic analysis result from the previous one is sent. It has the ability to explore. At this time, if there is no other solution for the same combination of parts of speech, the analysis will fail and the process will proceed from step n8 to step n20.

In step n20, it is determined whether all parts of speech have been combined in the Vi dictionary lookup buffer 52a. However, at this stage, all the items in the dictionary lookup buffer 52a have not been combined for "jlj", so step n2
1, a new combination of parts of speech is set, and the syntactic analysis of step n7 is performed again. As a result, if the syntax analysis is successful, processing operations from step n8 to steps n9 to n17 are performed. If the translation result is also incorrect, the process moves to step n18 or last step n19 according to the operator's next candidate instruction, and then moves to step n7 for syntax analysis. At this time, if there is another solution for the combination K74 of the same part of speech, the processing operations of steps n8 to n17 are performed. If the result is again an error, the process moves from step n18 to step n19 according to the operator's instruction, and the syntax analysis in step n7 is performed again. A detailed explanation of the previous work after this will be omitted as it will be complicated, but generally speaking, if there is another solution for the same combination of parts of speech, step n19゜n
7. n8. n9. nlO, nil, n12. nl 3,
n 14. n 15, n 16. n 17 is performed, the solution is displayed, and the operator's decision is awaited. If no other solution exists, step n19
．． n7. A processing operation is performed at step n8, and it is determined at step n20 whether or not all part-of-speech combinations have been completed. If not completed, a new combination of parts of speech is selected in step n21, and the process returns to step n7. A new combination of parts of speech is used to construct the syntax + p If it fails, move from step n8 to step n20,
A check is made to see if all products have been matched.

As shown in Figure 3 (6), when syntactic analysis is successful for a combination of parts of speech and the resulting translated sentence is correct,
Step n18 to step n2 depending on the operator's judgment.
Moving on to step 4, it is determined whether or not to input the next Bunka.
If the next sentence is to be input, the process moves to step n1, and if the next sentence is to be inputted, the translation ends.

When all the parts of speech have been combined in step n20, the process moves to step n22, where it is determined whether the next lord M7 lag is set or not. If the successive first flag is set, the process moves to step n7, and if no cent has been placed, the process moves to step n23. In this case step n7. n8. When processing operations are performed in a loop of n20°n21, the next candidate flag remains unset. At this time, an error message is displayed in step n23. In the embodiment shown in FIG.
From the part of speech of “me”! l! 11 #) has been cut, so f21. in Table 1 shown in the prior art. (31 solutions were cut, resulting in 2 fewer keystrokes to obtain accuracy.

Similar to FIG. 3, FIGS. 4 and 5 show the display means 4a.
A display screen at K is shown. FIG. 4(1) and FIG. 5 t1+ show a state in which input of the original text to be translated has been completed and it is displayed on the display screen. Fourth
Figure (2) and Figure 5 (2) illustrate the step n13 described above.
The part of speech of each word in the initially set sentence is displayed together with the translated sentence. Figure 4 (3) is
[This is a case where the part of speech of "fliesJ" is displayed as a noun, but the operator operates the key 8 in step n14 to instruct a correction to indicate that it is a verb. Figure 5 (3)
This is a case where "1ike" is displayed as a preposition and the operator instructs to accept it in step n14. l'-fliesJ as shown in Figure 4! ! When the correction instruction is given as ``lJ'', the solution is uniquely determined as shown in FIG.
4) , (The solution of 5i is cut and a positive value can be obtained with the first candidate.

In addition, if rlikej is accepted as a 00 position as shown in Figure 5, the translated text of Table 1 (6) shown in the prior art as shown in Figures 5 (4) to 5 (6) has been cut, resulting in one less key operation. Figures 3 to 5
In any case shown in the figure, errors in translation candidates can be reduced compared to the prior art, and accuracy can be obtained quickly. In the present invention, when displaying the translation result, it is only necessary to instruct the acceptance or correction of the part of speech of one word, but of course it is not limited to J of the part of speech of one base word, and the part of speech of multiple words can be indicated. may be done. For example, accept "T i m e" as a noun, rlikej
Table 1 (31, i4) when used as a preposition to indicate acceptance
, [5) j+j'g is cut, key 7.8.9
This operation is omitted three times, and the first candidate is selected four times.

FIG. 6 shows the contents stored in each buffer 51a to 55a in the present invention. Figure 6 fil is the original text buffer 5
1a and the contents of the stationary pull buffer 52a. Based on the words input to the original text buffer 51a, the 11+¥ book reference buffer 52a retrieves grammatical information such as part of speech, translation, and 1-word information corresponding to each word. The parsing buffer 53a shown in FIG. 6(2) parses the input sentence based on part-of-speech information given to each word. FIG. 6(3) shows the contents of the syntax generation buffer 54a and the result buffer 55a. Based on the parsed content, the syntax generation buffer 54a generates a syntax consisting of three output words, and based on the structure, the syntax is generated in the result buffer 55a.
The contents of the K7=liN statement are stored.

Effects of the Invention As described above, according to the present invention, it is possible to solve the problem of the large volume of translation results in machine translation, and to obtain correct translation results by giving simple instructions from a human.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of the Fi translation device, FIG. 2 is a 70-chart for explaining the operation according to the present invention, FIGS. 3 to 5 are diagrams showing the display screen of the display means 48, and FIG. 7 is a diagram showing the contents stored in each buffer 51a to 55a in the present invention, FIG. 7 is a diagram for explaining the process of machine translation, and FIG. 8 is a block diagram showing the configuration of a prior art translation device. FIG. 9 is a diagram showing the contents stored in each buffer 51 to 56 in the prior art, and FIG. 10 is a diagram showing a display screen by the display device 4 in the prior art. 5.5a -- Translation module, 51-56.51・a
~56a...Buffer, 56.56a...Table agent Patent attorney Kei Nishi Engraving of some drawings (no changes to the content) /mS 1
1-r ('J-)
N no 11-
/S's
GuN,I
-Yuno! $6 Figure 116 Figure 7 FA Figure 8 Coro's. 7 NF) 1'1 edition (no changes in content) 11~ Maetomi
-〇一ノ -ノ
- Procedural amendment (method) 1985 2p J13 Patent application 1985-210205 2. Name of the invention If! Translation method 3, relationship with the name to be amended Applicant address 22-22, Aiike-cho, Abeno-ku, Osaka City Name (50
4) Sharp Co., Ltd. Representative name: Saeki Asahi 4, Agent address: 1-13-38 Nishihonmachi, Nishi-ku, Osaka Shinko Sangyo Building Country Equipment EX 0525-5985 INTAPT
J International FAX GI [I&GIl (06)538
-0247 Phone (06) 538-0263 (Representative) Full name,! Ya (7555) 'f! Part of 4 teachings－゛－゛：・
, 45. Date of amendment order 6, drawings to be amended 7, content of amendments: engravings of drawings 3, 4, tjS5, and 10 (no changes in content). that's all

Claims (1)

[Claims] An input sentence to be translated is displayed, a part of speech of each word is displayed in correspondence with each word of the input sentence, the part of speech can be corrected, and the part of speech after the correction is displayed. A translation method characterized in that the input sentence is translated according to the following.