JPS62219165A - Processing system for equivalent selection in machine translation - Google Patents

Abstract

PURPOSE:To improve exactness and flexibility in a machine translation, by enabling the registration of the optimum equivalent for a verb in advance with a bit of conditional information in which the equivalent is used, and performing an equivalent selection only in the translation of the verb. CONSTITUTION:A verb conversion table 16 is provided for the equivalent selection regarding the verb. In the verb conversion table 16, the pattern of a noun, etc., used being attached on each verb, and a bit of equivalent information for the verb corresponding to the pattern, for example, by every verb of the Japanese, are registered in advance by a registration processing part 11. The verb conversion table 16 is retrieved by a verb equivalent selecting part 19 when a translation processing part 17 inputs a text to be translated, and outputs a translated text. For example, when a verb concept B included in the text to be translated, is used with the way of using a pattern B2, b2 is selected as the equivalent of the verb concept from the retrieved result of the verb conversion table 16.

Description

[Detailed Description of the Invention] [Summary] A user who uses a machine translation processing device can register in advance an optimal translation for a verb along with information on the conditions under which the translation will be used, and By selecting a limited number of translation words, the accuracy and flexibility of machine translation are improved.

[Industrial application field]

The present invention relates to a machine translation processing device that performs processing such as translating an input Japanese sentence into English using a device equipped with a central processing unit and a memory, and in particular, the present invention relates to a machine translation processing device that performs processing such as translating an input Japanese sentence into an English sentence. This is related to the processing method.

[Conventional technology]

In the machine translation processing device, the translated word and its translated word are
When there is a relationship of l to n (n≧2).

It is necessary to select the most suitable translation from among them. In the conventional system, the translation word selection function was built into the grammar of the machine system and was fixed.

[Problem that the invention seeks to solve]

Therefore, in the conventional method, when selecting a translated word in one machine translation processing device, there was a problem in that it was difficult to make the selection in accordance with each application field. In particular, only those who are familiar with the internals of an a-machine translation processing device can change the translation selection function, so ordinary translators cannot tune it to select translations that suit them. Since the translation word selection process is complicated, there is a problem in that the computer cost for selecting the translation words is high.

The present invention aims to solve the above problems and reduces computer costs by limiting translation selection to verbs only.

The purpose is to provide an easy-to-use machine translation processing device by providing a registration means that is as easy as possible for translators to memorize translated words.

[Means for solving problems]

FIG. 1 shows an example of the basic configuration of the present invention.

In FIG. 1, 10 is a processing device consisting of a CPU and memory, etc., 11 is a registration processing unit that registers translation selection information regarding verbs, and 12 is a storage device that serves as a model containing the verb to be registered in order to register translation selection information. Translated sentence analysis section that inputs and analyzes translated phrases.

Reference numeral 13 denotes a translated phrase analysis unit that inputs and analyzes a correct translated phrase of the translated phrase as a model; 14 a pattern matching unit that performs a process of classifying each of the translated phrase and the translated phrase into predetermined patterns; and 15; A registration unit that registers translated word selection information according to the matching result of the pattern matching unit 14 is shown.

In addition, 16 is a verb conversion in which the verb concept included in the sentence to be translated is used as a heading, and for each word that has a specific semantic relationship with the verb concept, translation information regarding the optimal verb corresponding to each word is stored in advance. It's a table. 17 is a translation processing unit that performs processing such as converting a Japanese sentence into English; 18 is a translated sentence analysis unit that inputs and analyzes a translated sentence; 19 is a verb included in the translated sentence based on the verb conversion table 16; 20 is a translated sentence generation unit that generates a translated sentence according to a predetermined grammar; 21 to 23 are input information; 24 is output information; 25 is a verb conversion table 16
represents an external storage device such as a magnetic disk device in which the contents of the file are stored.

In the present invention, for selecting translations regarding verbs.

A verb conversion table 16 is provided. The verb conversion table 16 includes, for example, for each Japanese verb, a pattern of a noun or the like used accompanying each verb, and translation information of the verb corresponding to that pattern.

It is registered in advance by the registration processing unit 11.

When the translation processing unit 17 inputs a translated sentence and outputs a translated sentence, the verb translation selection unit 19 searches the verb conversion table 16. for example.

When verb concept B included in the sentence to be translated is used in the usage of pattern B2, as a translation of the verb concept from the search results of the verb conversion table 16.

b2 will be selected.

[Effect]

FIG. 2 is a diagram for explaining the present invention in detail.

A machine translation processing device that performs Japanese-English translation will be described below as an example. For example, in the Japanese sentence ``to take a shower,'' the English translation of ``to take a shower'' is ``take.''
is optimal. On the other hand, in English, "under fire" is used to say "to be exposed to gunfire."

When a machine translation processing device is tuned and amplified to select such different translation words.

:L-zaba, translation selection information is registered in advance by the registration processing unit 11 as follows.

■ As the input information 21, input the translated phrase "take a shower."

■ As input information 22, take a shower
Enter the correct English for the translated phrase you entered in ■, ``.

■ Similarly, as the input information 21, the translated phrase ``to be exposed to gunfire'' is input.

■ Next, as the input information 22, “under fir
Enter the correct English of the phrase to be translated that you entered with ■, "e".

The Japanese phrase to be translated entered in ■ and ■ above is.

It is analyzed by the translated phrase analysis unit 12, and a conceptual structure as shown in FIG. 2(b) is generated. Also.

The English translated phrases input in ■ and ■ are processed by the translated phrase analysis unit 13.
The conceptual structure shown in FIG. 2(c) is generated by the analysis. It may be considered that the conversion rules of the conceptual structures shown in FIG. 2(B) and (C) are registered in the verb conversion table 16, as shown in FIG. 2(A).

When translating Le 11, if the verb ``to take a shower'' is used with the verb ``to shower'', the English translation is ``take'' by searching the verb conversion table 16.
If "a shower" is selected and used with the word "under fire"
is selected.

The user can register verb translation selection information using the registration processing unit 11 at any time as needed, and how to do this is simply by giving a registration instruction and activating the registration processing unit 11. It is easy because all you need to do is input the Japanese word containing the verb to be registered and the corresponding English word.

During translation, the translation processing unit 17 only needs to use the verb translation selection unit 19 to search the verb conversion table 16 for one verb included in the sentence to be translated, and check the constraint conditions as necessary. , the processing load related to translation word selection is small.

〔Example〕

FIG. 3 is a diagram for explaining the registration process for selecting translation words according to the present invention. FIG. 4 is a diagram for explaining the translation process according to the present invention.

I gave a brief explanation of the second drawing, but the Japanese words for ``take a shower'' and
” in English at the time of registration.
This will be explained according to the diagram.

The Japanese input sentence "take a shower" is decomposed into segments by searching two Japanese dictionaries. The results are shown in Figure 3.

“Shower (noun) / wo (particle) / bathe (verb) / ru (
final form/adjunctive form). This is compared with a predetermined pattern on the two grammars.

There are four types of this pattern: Jl to J4.

It is as follows.

Jl: ~ is a ten-final verb with an adnominal form (e.g. - to throw a spoon) J2: ~ is a ten-final verb with an adnominal form (e.g. - bone ga oreru) J3: ~ is a ten-final verb with a sub-adjunction form (e.g. - to throw a spoon) J4: Verbs in which ~ is a subjunctive form (e.g. - gagsakemomo) In the case of "taking a shower", the above pattern Jl is matched.

Similarly, "take a shower" according to the English dictionary.

Break it down into segments. “take (verb)” and “
shower (noun)” (“a” is omitted).

There are five types of grammatical patterns in English: first order.

El: “verb” – object (example – read book)
E2: Adjective (e.g. hard) E3: Prepositional noun (e.g. under fire) E
4: Adverbs (e.g. devotedly, never)
E5: Prepositional nouns (e.g. with one's head held h
i8b=[shaking one's hand]) "take a shower" in this example corresponds to pattern El. In addition, 9 Japanese patterns J
l.

j2 corresponds to the set of English patterns E1 to E3.

Japanese pattern J3. J4 is the English pattern E4, E
Corresponds to group 5.

Based on the correspondence between pattern J1 and pattern E1, registration as shown in FIG. 2(a) is performed. As a result, when the object is "shower" and the verb is "to take a bath,""take" is selected as the translation of the verb.

In this example there are no constraints, but in some cases.

Restrictive conditions are attached regarding the applicability of translation word selection. For example, the condition is quadratic.

b) When the English pattern is an adjective 9 Modifiers should not be attached to Japanese words. for example.

If a modifier (of the shoulder) is added, such as ``My shoulder bone hurts,'' it becomes inapplicable.

(ii) When the English pattern is an adverb 2 A modifier must not be attached to the Japanese object.

(iii) When the Japanese patterns are Jl and J2, the first Japanese side is connected in parallel, such as in the final form, in the adjunct form, or in ``～shite～''.

(1v) Japanese pattern is J3. At the time of J4, the relationship on the 9 Japanese side must be modal.

In other words, as in ``I spare no time to study (study).''
Must be a phrase that expresses a state of action.

In the present invention, translation processing is executed as shown in FIG. 4 using the verb conversion table 16 set in two or more ways. The numbers ■ to [phase] in the following description correspond to the process numbers ■ to [phase] shown in FIG.

■ For example, enter a sentence to be translated, such as "This is exhausting work."

■Use the 2-Japanese dictionary to input the translated sentence.

Split words. The division result is 2, for example, "This/ha/bone/ga/I/ru/work/da/."

■ Compose two clauses according to Japanese analysis grammar.

Analyze dependencies. "My bones are broken" means "work."
It can be seen that ``this'' refers to all of the following.

■ Generate a conceptual structure based on the analysis results of process ■. In the diagram, word concepts are represented by nodes, and relationships between concepts are represented by arcs. <3j> refers to the main concept.

■ Verb conversion table 16 for the verb concept “oreru”
Search for. "" corresponding to pattern J4 shown in FIG.
You can find the conceptual pattern of "bone".

From this, it can be seen that the corresponding English word is "hard".

(2) From the constraint index set in the verb conversion table 16, refer to the constraint in (2) and match whether or not "ore" is an adnominal modification. In this example, "work" is modified, so one condition is satisfied.

■ Assign the translation concept of “hard” to the concept of “bone”.

■ By searching an English dictionary, you can learn about each concept.
Convert to English words. The processes after this process (1) can be considered to be the same as the conventional process.

(2) Perform English generation processing as shown in FIG. 4 (B') according to the English generation grammar. Therefore, first it is determined whether the predicate is a noun. In this example, the predicate “”
Since "work" is applicable, the result of determination 2 is YES.

[Phase] If the determination result is No, it is determined whether the predicate is a verb. Then, □ processing is performed according to the 1 determination result.

■ If the predicate is a noun, a one-noun syntax is generated. Therefore, first, it is determined whether it is a title. If it is not a title, perform the following process.

@ Generate a subject゛. In this example, "this" is generated.

o Generate BE verbs. “'is” is added to “this”.

■ Generate predicates. Overall, “this is
ahard work” will be generated.

■ Generation of periods as post-processing for English generation.

Performs processing such as capitalizing the beginning of sentences. Finally “Th
1s is a hard work,'' will be generated.

[Phase] “”Fhis is a hard work
, ” is output as a translated sentence of ``This is exhausting work.''

〔Effect of the invention〕

As explained above, according to the present invention, the selection of translation words is limited to verbs only, so the computer cost for translation can be reduced, and the accuracy of translation can be improved relatively easily.
Flexibility can be improved.

For example, a translator memorizes the Japanese phrase ``take a shower'' as if it were a single word, and also memorizes its translation as a sequence, ``take a shower.'' Registration of translation selection information according to the present invention is similar to how translators memorize information, so it is easy to input.

Moreover, unlike if you simply registered one word, ``take a shower,'' it would say ``I don't take a shower.''

It is also effective for transformations such as "I took a shower yesterday."

[Brief explanation of drawings]

FIG. 1 is an example of the basic configuration of the present invention, FIG. 2 is an explanatory diagram of the operation of the present invention, FIG. 3 is a diagram for explaining the translation selection registration process according to the present invention, and FIG. 4 is a diagram according to the present invention. A diagram for explaining translation processing is shown. In the figure, 10 is a processing device, 11 is a registration processing unit, 12 is a translated sentence analysis unit, 13 is a translated phrase analysis unit, 14 is a pattern matching unit, 15 is a registration unit, 16 is a verb conversion table, 1
Reference numeral 7 represents a translation processing unit, 18 represents a translated sentence analysis unit, 19 represents a verb translation selection unit, 20 represents a translated sentence generation unit, 21 to 23 represent input information, 24 represents output information, and 25 represents an external Hito a device. Patent Applicant: Fujitsu Ltd. Representative Patent Attorney Hiroshi Mori 1) (1 other person) Pregnant G-Suri T is 1 month and 8 months, 4 calms (A)

Claims (1)

[Claims] A device comprising a central processing unit and a memory, inputting a sentence to be translated,
A machine translation processing device that generates a translated sentence based on the analysis results uses the verb concept included in the translated sentence as a heading, and uses the verb concept included in the translated sentence as a heading to correspond to each word that has a specific semantic relationship with the verb concept. A verb conversion table (16) in which translation information regarding the optimal verb concept is stored, and when registering translations related to verbs, input translated phrases containing verbs and their correct translated phrases, and input the translated phrases and translated phrases. The connections between the words constituting the words are classified into predetermined patterns, and the correspondence information is created in the above verb conversion table (16) with the verb concept to be translated as the heading.
) and a processing means (11) that is registered in the translation process, searches the verb conversion table (16) for the verb concept appearing in the sentence to be translated, and matches the word pattern that causes a dependency with the verb. A translation word selection processing method in machine translation, comprising a translation word selection means (19) for selecting a translation word concept.