JPS6174069A - Translation system - Google Patents

Abstract

PURPOSE:To delete a multi-meaning characteristic of translation results and to obtain correct translation results by indicating a part of speech of respective words composing an input sentence to be translated and translating the input sentence with the indicated part of speech. CONSTITUTION:When a word is inputted, the word is displayed through a processing device 1 to a display means 4. The processing device 1 sends the character code to a translation module 5. The translation module 5 accommodates the inputted code to an original sentence buffer 51 and analyzes a dictionary consulting morpheme. Consequently, when the word is inputted, an operator can indicate a part of speech. When the operator indicates a display, a name of the indicated part of speech is displayed at the display means 4 and only the part of speech designated from a dictionary consulting result is not excluded.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to an interactive translation method in which a machine and a human cooperate to create a correct translated sentence.

BACKGROUND TECHNOLOGY Generally, machine translation is performed through a process as shown in FIG. The source text consisting of the input language to be translated needs to be analyzed during the translation process. There are three stages of analysis: morphological analysis, syntactic analysis, and semantic analysis.

Morphological analysis refers to looking up a dictionary for machine translation, extracting grammatical information such as part of speech and translation information for each word, and
It involves analyzing numbers, sentence tenses, etc. Syntactic analysis means examining the dependency relationships between words and analyzing the structure of the sentence. Semantic analysis is the process of determining what is correct and what is not from the results of multiple sentence analysis.0 Machine translation performs analysis to one of three levels to obtain the internal structure of the input language. According to this structure, the internal structure of the output language that constitutes the In translation is converted to the same level internal structure, and then the output language is generated.

A single sentence of machine translation differs depending on the depth of this analysis level.

shape! ! ! Those that only perform elemental analysis cannot translate sentence units, but are limited to word unit translations, as typified by so-called electronic translation methods. Those that perform syntactic analysis output all grammatically correct interpretations, but this results in outputting a wide variety of translation results, which increases the amount of time it takes for humans to determine the correct answer. Those who go as far as interpreting the meaning are
Although FA can output only one correct translation result, in order to do so, it is necessary for the machine to store a huge amount of information, which is almost impossible in reality. Translation is defined as at least the level of syntactic analysis. In other words, machine translation is capable of sentence-by-sentence translation, outputting all grammatically correct interpretations, and even meaning analysis, but this is not perfect and machine translation cannot narrow down the output to a unique one.

FIG. 1 is a block diagram showing the configuration of a flat type Ii translation device. Kihebor 1/2 connects to the processing device 1 and manually inputs data, a memory 3 stores data related to the processing device 1, and a display device 4 displays the results processed by the processing device 1. A translation monochrome 5 is provided in which data related to translation is stored. The translation monoyl 5 includes an original text buffer 74M that stores the input sentence system;
A dictionary lookup buffer 52 that stores the result of dictionary lookup, and a configuration analysis buffer 5 that stores the result of syntax analysis of the input language.
A syntax generation buffer 754 that stores the result of converting the internal expression of 3 for the output Ftg, a result buffer 55 that stores the result of the morpheme principal composition of the output W word, a dictionary for translation, a grammar 1111, etc. A table 5G consisting of is included.

Figure 8 1 When translated, the input writing ability f T ise r l ies l ike
An arrow.

, the contents stored in each pub 7751-55 in the prior art are shown. FIG. 8 (1) shows the contents of the original text buffer 51 and the dictionary lookup buff 7752. According to the single dr1 input to the original text buffer 5, the dictionary lookup buffer 52 reads grammatical information such as parts of speech corresponding to each word. Translation information is extracted. The parsing buffer 53 shown in FIG. 8(2) parses the input sentence based on part-of-speech information given to each word. FIG. 8(3) shows a sentence generation buffer 54 and a result buffer 5.
5 is displayed, and the result of converting the contents of the parsing buffer 53 to the output language is stored in the 61 sentence generation buffer 54! Firu.

FIG. 9 is a diagram showing a display screen of the display device 4 in the prior art. When the input sentence to be translated is input as shown in 9th [M (11), the translation result as shown in M91J (2-19 (6)) is displayed. The translated sentences shown in Figures 9 (2) to 9 (6) on each display screen can be translated to This is the result of the translation.

Table 1 These are grammatically correct interpretations.

Of course, human beings can judge that the interpretation shown in Figure 9 (6) is the correct answer.If semantic analysis is the basis, then the translated sentence is as shown in Figure 9 (6). However, in order to do this, (a) the arrow and the fly do not eat together, (b) the arrow has the ability to measure the river, and (c) the fly There is a necessity to have the 8!1 machine record the knowledge that it will never fall in love with the arrow. It is obvious that it is impossible for a machine to memorize all of this knowledge about the real world. Therefore, it can be seen that it is inevitable that there are solutions that should not be compatible at the conventional level of machine translation. As described above, the conventional machine translation system has a drawback in that, since there are a large number of solutions, the translation key must be operated many times until the correct answer is obtained. Furthermore, in machine R translation, all kinds of possibilities are searched and all solutions are found, so it takes a long time to obtain the correct answer. For this reason, a method is adopted in which the solutions are output in order, starting with those found. Therefore, 1jSS diagram (2)
When the first solution is output and compared, it is not known how many remaining solutions exist, and it is not possible to display things like the number of remaining homophones in kana-kanji conversion. For this reason, it not only has the disadvantage of having to press the translation key many times, but also has the disadvantage that the correct answer cannot be obtained even after pressing the translation key many times. Therefore, the mental burden on the operator is quite large, and from this point of view 11!1. l
I hope that even one of the total number of solutions will be destroyed by X manipulation.
That's why.

The problem that the invention is trying to solve, α In the above translation results, Figures 9 (2) to 9 (5)
It can be seen that mistranslations like the one shown in Figure 2 are caused by incorrect identification of parts of speech in multipart-of-speech words. However, such an error can be immediately recognized by someone who has the ability to judge the r# and anzu by looking at the translation result.

The purpose of the present invention is to solve the problem of % productivity of translation results in machine translation and to provide a translation method that can obtain correct translation results by giving simple instructions from a human.

Means for Solving Problem 7 Hiroshi The present invention is a translation method characterized by specifying the part of speech of each word constituting an input sentence to be translated, and translating the input jJ sentence using the specified part of speech. It is a method.

+￥ JII The part of speech of each word in the input sentence to be translated is specified, and the input sentence is translated using the part of speech that is proportionate to the specified part of speech, so the number of translation candidates can be reduced.

Embodiment 11 FIG. 11 is a block diagram showing the configuration of an If translation device capable of carrying out the present invention. l to processing device 1! 111
! ! a keyboard 2 for manually inputting data; a memory 3 for storing data related to the processing device 1; and a display device 4 for displaying the results processed by the processing device 1.
A translation monochrome 5 in which data regarding translation is stored is also provided. The translation monologue 5 outputs the contents of the original text buffer 51 that stores the input sentence, the dictionary lookup buffer 52 that stores the dictionary lookup result, the syntax analysis buffer 53 that stores the syntax analysis result of the input language, and the syntax analysis buffer 753. It includes a syntax generation buffer 54 that stores M-buttons converted for M, a result buffer 55 that stores the results of morpheme generation of the output language, and a table 56 that includes a dictionary for machine translation, grammar rules, and the like.

FIG. 2 is a 7-coach diagram for explaining the operation according to the present invention.+First, in step n1, the next candidate flag is cleared prior to inputting the translation original text. This flag is used to display an error message when a single solution cannot be obtained even after performing syntactic analysis on all combinations of parts of speech, as will be described later. Next, when one word is input from the key human power means 2, it is determined whether the input is a word or a translation key for instructing translation.

When a single Sh input is made, the process moves to step n3 and the word is displayed on the display means 4 via the processing device 1. The processing device 1 sends the sentence 7: Food to the translation monoyl 5. Let's go to step n4! The character code into which 1 translation monoyl 5 has been input is stored in the original text buffer 51, and the character code is stored in the table 5.
According to the present invention, the operator can specify the part of speech when inputting a word. At input step 15, whether or not the part of speech is displayed by the operator is determined at t9.
If the request is rejected and there is no such instruction, the process moves to step n8.

If there is a display instruction from the operator, step n
6, the designated part of speech name is displayed on the display means 4, and in step n7, words other than the designated part of speech are removed from the dictionary lookup result. In step n8, when there is no part of speech displayed in step n5, all the dictionary lookup results 7'L are stored in the dictionary lookup buffer 52, and only the designated part of speech is stored in the dictionary lookup buffer 52. While the input of the original text continues in this way, if there is no part of speech instruction by the operator, step n2゜n3 +n4 +05. n
After 8th step, the processing operation 1 step pro 2 is performed again, and if the part of speech instruction by the operator is 7t, step n2 is also performed.
, n3 +n4, n5, n6 +n? , n8
Then, the processing operation of step n2 is performed again. When inputting the original text is completed and the operator issues a translation instruction by operating the translation key in step o2, the process moves to step n9. entering step n9
Now, the combination of parts of speech is first M, then step nlO.
Move to. At step nln, syntax analysis is performed with reference to the analysis rules of the 1t table 56, and the syntax analysis buffer 53
is completed. In step n11f', it is determined whether the analysis was successful or not. If the 61 sentence analysis is successful, the process moves to step n12, where structural conversion is performed in translation monoyl 5 while referring to the conversion rule # and the current rule of table 56. Be done. Next, in step o13, syntax generation is performed, and the sentence generation buffer 54 is completed, and in step 'm'n+4, the result is completed, and the result is transferred to step o13. It is displayed on the display means 4 at the mouth 15 . In step n16, if the n translation result is incorrect, it is determined whether the next candidate key was manually input or not in order to generate the next candidate. When the next candidate key is entered manually, the process moves to step n17, where the next candidate flag is set.In step n+0, the syntax parsing is completed 7 times and the translation result is displayed, and after t, the operator further When requesting another translation result, the next candidate flag is set. From step n+7, the syntactic analysis starts again at step nlO, but the 6η sentence M analysis buffer 53 stores the previous combination of parts of speech) and the result of the syntactic analysis, and if the combination of parts of speech is unchanged from the Rjf times, it is There is a subsection that searches for different parsing results. At this time, if there is no other solution for the combination of the same parts of speech, the analysis fails (1. The process moves from step n11 to step n18. In step n18, it is determined whether or not all the combinations of parts R have been completed. Only ( , at this stage, all parts of speech in the dictionary lookup buffer 752 are
1 (Step n19 since the mediation has not been completed)
At step 89, a new combination of parts of speech is set, and the inscription analysis in step nl□ is performed again. As a result, if the parsing is successful, step nil to step n12
~n15 processing operations are performed. If the translation result is also incorrect, step n is given by the operator's next candidate instruction.
16, the process moves to step n17, and the process moves to step port 10 again. At this time, if there is another solution for the same combination of parts of speech, the processing operations of steps n11 to n15 are performed. If the result is again an error, the process moves from step n16 to step n17 according to the operator's instruction, and the syntax analysis in step n1O is repeated. The detailed operation after this will be complicated, so a detailed explanation will be omitted, but in general, if there is another solution for the same part-of-speech combination ν1, step nl7. nl O, nl 1. nl 2
．． nl 3. nl 4. nl 5 and place 1! l! An action is taken and the solution is displayed for the operator to make a decision.

If no alternative solution exists, steps nl 7 , nl
The processing operations O+nl 1 and nl 8 are performed, and whether or not the combinations of all parts of speech have been completed is 1! (Select a new combination of parts of speech in step 119 and return to step nlo. If the parsing is successful with the new combination of parts of speech, step nll, i12゜nl 3 + nl 4,
A processing operation is performed with nl 5 , the solution is displayed, and the operator waits for nl to be disconnected. If failure r, step n
The process moves from il to step n18, where it is checked whether all parts of speech have been combined.

A: If the parsing is successful for the combination of J and M, and the resulting annotations and Kumon are correct, the operator moves from step n+6 to step n22, and it is determined whether or not to manually edit the dog's text. If the input of the sentence ``dog'' is invalid, the process moves to step r11, and if the input of a different sentence is invalid, the translation ends. When the combination of all parts of speech is completed in step n18, step n
Moving on to 20, the weather! It is determined whether or not the No. 41 flag is set. If the zero-order candidate flag is set, the process moves to step nlO, and if it is not set, the process moves to step n20.

In this case, steps nl O, nl 1, nl 8
, nl 9 , n10, the weather M7 lag remains unset. At this time, an error message is displayed in step n21.

The display screen of the display means 4 is shown in FIG. Pt53 to FIG. FIG. 3 shows "in step n5"
This shows the case where "Time" is specified as a noun.

FIG. 4 shows a case where ``flies'' is specified as a verb, and FIG. 5 shows a case where ``1ike'' is specified as a preposition. This example shows the case where the part of speech of one word is specified at the time of input. For example, "f l ies" as shown in the fB4 diagram.
If you indicate that the part of speech is the part of speech, the solution is uniquely determined, and the correct answer can be obtained with the first candidate. In the present invention, it is only necessary to specify the part of speech of one word when using manual input, but it is of course not limited to Al single::n, and instructions may be given for a plurality of words. For example, it is also possible to designate "Timp" as a noun and designate "1ike" as R7f e R. In the case of 1 and 2, the total number of solutions in the flit translation result is smaller than when the part of speech of one word is designated.

FIG. 6 shows the contents stored in each of the 7A's 51 to 56 in the present invention. ! - The input sentence to be written is input to the original text buffer 51, and as shown in the fIS4 diagram, n
When the part of speech of ``1ike'' is specified as the !IIJ word, the part of speech of ``Ni
There is no noun part in the contents of the ``ji'' column, and there is only a verb part.At this time, in step n9 described above, ``Tl1le'' is a noun, ``flies'' is a verb, ``like'' is a preposition, When the combination of parts of speech such as ``an'' as an article and ``l1rroIu'' as a noun is set as the truncated side, Translation Mononym 5 refers to the parsing rules in Table 56 and parses the syntax as shown in Figure 6 (2). will be carried out. In this case, the first composition level includes nouns, 9JJ words, prepositions, articles, and nouns; the second composition level includes subjects, verbs, and phrases; and the third composition level includes subject parts. , it is analyzed as if it were a predicate and constituted a sentence. FIG. 6(3) shows the contents of the syntax generation buffer 54 and the result buffer 55.Based on the parsed contents, the inscription analysis buffer 54 contains:
The sentence is treated as a subject and a predicate at the first composition level, and as a noun phrase, adverb phrase, and verb at the second level, and the content of the sentence is stored in the result buffer 55 based on the composition. . In this way, in accordance with the operation shown in FIG. 2, in the present invention, when an original text to be translated is input, a part of speech is specified for each word that makes up the original text, and the part of speech is translated according to the specified part of speech. A translated sentence is generated.

Effects of the Invention As described above, according to the present invention, it is possible to solve the problem of the high quality of translation results in fi machines and translations, and to obtain correct translation results by giving simple instructions from a human.

[Brief explanation of the drawing]

Figure 1 is a block diagram showing the configuration of the translation device, 2nd CK
7a-chart for explaining the movement according to the present invention, FIGS. 3 to 5 are views showing both sides of the display in the display means 4, and FIG. 6 is a diagram showing each bread 7751 to 5 in the present invention.
6 is a diagram showing the contents stored in the buffers 51 to 56 in the prior art. FIG. 9 is a diagram showing both sides of the display by the display device 4 in the prior art. 5...Translation monoyl, 51-55...Buffer 7.
5G...Table agent or J〒74IIt Raw--Department Fig. 1 Fig. 3 Fig. 4 Fig. 6 Fig. 7

Claims (1)

[Claims] A translation method characterized by specifying the part of speech of each word constituting an input sentence to be translated, and translating the input sentence using the specified part of speech.