JPS6174068A - Translation system - Google Patents

Abstract

PURPOSE:To obtain a correct translation result quickly by outputting translated sentences from an input sentence to be translated in accordance with plural mutually different relations and selecting one of respective translated sentences. CONSTITUTION:When a source language is inputted, a dictionary is consulted, a morpheme is analyzed and assembling of a part of speech is initially set. A row of a part of speech initially set is analyzed for sentence structure at a sentence structure analyzing part 52. It is judged whether or not sentence structure analysis is successful, when successful, the language is converted at a language converting part 53, a language is generated at a target language generating part 54 and a translation result by the first word used in a transla tion of a part of speech used for sentence structure analysis is generated. The result is displayed by the display device, it is judged whether or not the next candidate is called and further, it is judged whether or not the next sentence is translated. 51: dictionary consulting morpheme analyzing part, a: target language, b: source language.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to an interactive machine translation system in which a machine and a large book cooperate to produce correct translation results.

Background Art In general, in the case of Machine II translation, since the natural language itself has matching properties, vI! It is inevitable that lt's nagi will appear. Therefore, multiple l]! It is necessary to display translation candidates so that the user can select from among the translation candidates and come up with a more accurate translation. As an example, I bout a ear wit
h 200 dollars.

Consider the case of translating the English sentence into Japanese. If we define the dependency relationship in syntax as dependency, this sentence has two possibilities for dependency, as shown by arrows P and Q in FIG. Also, Car has the meaning of 0 cars and ``basket'', and buy has the meaning of ``to buy 1 to acquire 1.''

In other words, there are eight types of translations as shown in Table 1.

The correct number in Table 1 is the 811th translation candidate. If the next candidates are presented in order from the first translation candidate to the 811th translation candidate, in this case, the 8th translation candidate will be selected. Translation candidates 1st to 4th among these 8 types
The translation candidates are examples related to "with 200 dollars" as shown by the arrow P shown in Figure tJ&9. If the dependencies of the entire sentence in this case are displayed, they will look like the arrows P1R and S shown in FIG. 10. This is a difference in dependency. The user
I noticed this and pressed the key for the next candidate for syntax analysis,
After that, if the word candidate key is continued to be pressed, the correct translation will be displayed as the first translation candidate, the fifth translation candidate, the tIS6 translation candidate, the seventh translation candidate, and the eighth translation candidate.

Problems to be Solved by the Invention As mentioned above, in the case of machine translation, since natural language itself has a tendency to refine, it is inevitable that translations of PM and rlL will appear. Therefore, an object of the present invention is to provide a translation method that allows the user to select from a plurality of translation candidates and produce more correct r, +, (shi().

Means for Solving the Problems The present invention is characterized in that an input sentence to be translated is output as translated sentences according to a plurality of mutually different dependencies, and one of the translated sentences is selected. This is a translation method that uses

According to the present invention, the input sentence to be translated is outputted as translated sentences according to a plurality of mutually different dependencies, and one of the translated sentences is selected, so that correct translation results can be obtained quickly. I can do it.

Embodiment FIG. 1 is a block diagram showing the configuration of a translation device in which the present invention can be implemented. In connection with a medium-temperature processing unit (CPU) 1 that performs processing such as translation processing, there is a main memory 2 in which programs and the like for processing by the CPU i are stored. ＃Display for displaying the necessary information for translation processing *ff1 (CRT) 3, translation projector (keyboard 4 for inputting information necessary for processing, and translation software for storing information necessary for translation process) 5 is provided.

The tlS2 diagram shows one embodiment of the translation monoyl 5.

R,? tmonyl 5 contains 5 buffers A and B.

A table 6 having dictionaries for C, DIE and translation, grammar rules, and main structure conversion rules is included. These buffers A to E store data at each analysis level in the translation process, as will be described later.

General machinery here! ! ! Calm processing will be explained.

Machine Il+ translation generally has levels as shown in Figure 3. Source from top left:! When the various information is input, the digit promotion/reduction of ■, the morpheme M fold of ■, and the AV syntax analysis of ■ proceed. Depending on this level of analysis, the gt*xtp translation is roughly 2
It can be divided into One is to analyze concepts that do not depend on each of the three words in ■ (intermediate words), and from there, create tags) f
This is a pivot method that generates fM. The other is
It is called a transformer 77 one-way type and is in the shape of ■! After performing any of the following analyzes: a-element analysis, ■ syntactic analysis, ■ semantic analysis, and ■ context analysis, we obtain the internal grooves of the source language and convert them to the same level structure of the tarded language. It generates a tarded language. Here, the contents of each analysis are listed in Table 2.

(The following is a blank space.) Table 2 In the present invention, it is assumed that at least the level of syntactic analysis (2) is carried out. That is, in the present invention, as shown in FIG. ! 11! consists of four parts: elementary analysis section 51, syntax analysis section 53, language conversion section 53, and tarded language 54 generation section. Thinking about translation monoyl.

FIG. 5 shows the contents of each buffer in FIG.
s t pen.

This example shows how to translate an English sentence into Japanese. First, the original text is read into the buffer 7A of the $2 diagram as shown in tIS5 diagram (1). Necessary information for each word is extracted by the dictionary lookup morphological analysis unit 51 shown in FIG. 4 and placed in the buffer 7B shown in FIG. The part-of-speech information that is part of it is shown in FIG. 5 (2).

This is a multi-part speech word, which is uniquely determined by the parsing unit 52 shown in FIG. to go into. At this time, an analysis as shown in Table 3 is performed based on the grammar rules of Table 6 in FIG.

(Margins below) Table 3 This rule represents, for example, that "a sentence consists of a subject and a predicate." In the language conversion unit 53 of FIG. The analysis is performed according to the rules, and the results are stored in buffer D in FIG. 2 as shown in FIG. 5(4). Tarded M word generator 5 in Figure 4
4, appropriate particles and auxiliary verbs are added to the Japanese form, and the result is placed in buffer 7E in Figure 2, as shown in Figure 5 (5).

This becomes the output of Translation Mononeal 5.

FIG. 6 is a 70-chart of the present invention. First, in step n1, before inputting one sentence consisting of the source language, the next candidate 7 blowfish is reset. This flag is set to
It is used to issue an error when no solution is found during syntax analysis.Then, when the source language is entered in step 12, step n3 is used to look up the dictionary and form!
! In step n4 of the 0th order in which elemental analysis is performed, M combinations of the first parts of speech of each multipart speech word appropriately selected corresponding to each word are set for the first time. In other words, Time flies
An arrow.

Considering the English sentence, as shown in Table 4 below,
There are several combinations of parts of speech.

Table 4 For example, in this example, noun, noun, conjunction, U, word,
The part-of-speech sequence is initialized, such as noun.

The part-of-speech sequence initialized in this way is parsed at step n5. Then, in step n6, it is determined whether the parsing was successful. If successful, step n7
The process progresses to language conversion and generation of 13 TARDEF words in step n8, and in step o9, a translation result is generated using the first translated word of the part of speech used in the parsing. step n
In IO, the result is displayed on the display wave rI13.The primary step n11C calls the next candidate and it is determined whether it is 1, and in step n12, it is determined whether or not to translate the next sentence a. When the zero-order sentence in which is vqwi+ is not translated, the translation process ends.

When the next candidate is called in step n11, it is determined in step n13 whether or not the cursor is on a word. If the cursor is on a word (the method for specifying the next candidate will be described in detail later), the process moves to step n14, where it is determined whether or not there is a translation for that word, and if there is, step n
15, a translation result of the word dog is generated. If there is no translation for dai, return to step n9. If the cursor is not on a word, such as at the end of a sentence, the process moves to step 116, sets the next candidate flag mentioned above, becomes the next syntax candidate, and returns to step n5 for syntax analysis.

The syntactic analysis unit 52 has a function of searching for the original solution and another dependency solution when the part-of-speech order has not changed. If there is a solution, repeat the same thing as the first time. If it fails, move to step n17 and judge whether all possible combinations of parts of speech have been completed. If there is a combination, enter step n18. The combination of parts of speech is changed in step n5, and if there are no more combinations of 0 parts of speech to be parsed again in step n5, the process moves from step n17 to step n19.
0 to determine whether the next candidate flag is set
If the next candidate flag is set, the process returns to step n4, and only if the major candidate 72 flag is not set, that is, only if there is no solution for parsing, step n4 is executed.
20, step 2-display is performed, and step n1
Return to

FIG. 7 shows an example of a method for indicating the next candidate when the source language is jA language.

As a first example, when the cursor is placed on a word as shown by ■ in FIG. 7(1), a translation candidate for that word is displayed. When it is at the beginning of a sentence, as in (2), a cursor movement key and a major candidate key are provided as keys to display the syntactic next candidate. As a second example, when the cursor is at the positions of ■ and 1■ in Figure 7 (2), the next translation candidate will be displayed, and as shown by *ta, ■, it will include words above a period (?, !, etc.) ), or when the cursor is at the end of the sentence as shown by ■, the syntax next candidate is displayed. The keys include a cursor movement key and a next candidate key. As an example of Pt53, a cursor movement key, a syntax next candidate key, and a word translation candidate key are provided, and when the syntax next candidate key is suppressed, the syntax next candidate is suppressed, and when the word translation candidate key is pressed, it is suppressed. , displays the next translation candidate for the word where the cursor is located, the fourth
In this example, as shown in Fig. 7 (3), cursor movement keys for each word are provided in addition to the general cursor movement keys, and the cursor on the word in the first example is It is made to move for each word.

In Figure tIS8, dependencies show how relationships are displayed. Figures 8 (1) and 8 (3) are stored in buffer C in Figure 1142, as shown in Figure 5 (3). This is the result of syntax analysis. From now on, Figure 8 (2) and Figure 8 (
In order to display the display as shown in 4), a rule II7 is prepared in advance that specifies what kind of tree should be displayed and how it should be displayed, and this rule is used. For example, when the tree ``noun phrase → noun phrase, prepositional phrase'' used in Figure 148 (1) is completed, 7 lines are added separately to the noun phrase part and the prepositional phrase part, and Make a rule to draw the arrow from the 8th point toward the noun phrase as shown by the reference mark F in (2). Apply these rules to the partial parts, and use the modification to create an indication. I can do it. By displaying the relationship in the modification in this way, it is easy to see where the mistake has been made, which is advantageous for future correction work. Furthermore, by dividing the major candidates into two stages, fewer keystrokes are required to arrive at the correct translation, and the display of snakes and dependencies has the effect of being similar to that of a junior high school student.

Effects of the Invention As described above, according to the present invention, a user can
It is possible to easily select the correct n translation candidates from among the candidates, and it is possible to produce the correct translation more quickly than in the prior art.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a translation device capable of implementing the present invention, FIG. 2 is a block diagram showing an embodiment of the IT+ translation module 5, and ttS3 diagram explains machine translation processing. Figure 4 is for Il! Figure 5 is a diagram showing the contents stored in each buffer A-E, Figure 6 is a 70-chart for explaining the present invention in detail, and Figure 7 is a diagram showing the military structure of the translation monoyl. A diagram to explain how to indicate the next candidate when the source language is English,
ttS8 is a diagram showing a method of displaying relationships, and FIGS. 9 and 30 are diagrams showing modifications of one sentence. 5...Hot translation monoyl, 6...Table, Δ~E
···buffer

Claims (1)

[Claims] A translation method characterized in that an input sentence to be translated is output as translated sentences according to a plurality of mutually different dependencies, and one of the translated sentences is selected.