JP2922701B2 - Language conversion method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a language conversion method, and more particularly, to an interactive language conversion method in a machine translation device.

[0002]

2. Description of the Related Art Current machine translation systems do not have sufficient knowledge to automatically perform correct translation. Therefore, an interactive method for compensating for the lack of knowledge by inquiring the user is being studied in various places. In the mainstream method of conventional machine translation, the result (tree structure) obtained by analyzing the source language is converted into a tree structure of the target language for each part, and the target result is obtained based on the obtained result. A language has been generated (hereinafter referred to as a forward-looking structure conversion method), but several methods have been announced that add ambiguity resolution to user inquiries to this method ("Question-answer type Japanese-English machine"). "A method of presenting deep case ambiguity using example sentences in a translation system" by Minoru Ashizawa and one other user-cooperative Japanese language generation system by Akira Kumano, all of which are in the 42nd National Convention of IPSJ. In addition, some methods have been proposed in which all information is obtained by inquiring the user without using the source language input in a pre-set form ("Automatic creation of foreign language letters" by Hiroaki Saito One, Natural Language Processing Technology Symposium, November 1984).

[0003] Further, in the conversion using the forward structure conversion method, it has been proposed to use knowledge of a target language for selecting a translated word ("translation word selection using knowledge of a target language").
Hiroshi Nomiyama, IPSJ 42nd National Convention). In addition, a number of bilingual example sentences are prepared, a bilingual pair including the source language sentence closest to the source language input is searched, and only the difference between the source language input and the source language sentence in the translated pair is converted. A method has also been proposed in which a translation in a parallel translation is corrected according to the result of the conversion.

In the conventional mainstream system, an analysis (tree structure) of a source language is converted into a tree structure of a target language for each part to generate a target language. With this method, even if the lack of knowledge is compensated for by dialogue with the user, the structure tends to be closer to the source language. It becomes an unnatural sentence pattern as the target language,
It becomes an unnatural expression. On the other hand, in the method of acquiring all information by inquiring the user, it is necessary to prepare a very large amount of data in order to correspond to a wide range of documents, and since the work is machine-driven from the beginning, Give the user a sense of control. In addition, since information necessary for the target language is often missing in the source language sentence, an incomplete output may be generated by the method of the present invention excluding the interaction with the user. Very much. In some cases, information required in the target language does not exist in the original text. In a conventional method, this necessary information is often supplemented by default processing in order to complete a translated sentence. However, erroneous information is often inserted or necessary information is dropped due to insufficient default processing, and an unnatural translated sentence tends to be generated. On the other hand, in the translation method using a bilingual example sentence,
To handle a wide range of documents, a very large number of bilingual examples are required.

[0005]

[Purpose] The present invention has been made in view of the above-mentioned situation, and compared with a method of generating a proper and natural translation from a conventional mainstream forward structure conversion method and acquiring all information by inquiring a user. To provide a language conversion method that requires less data to be prepared, does not give the user a sense of domination, and does not generate incomplete output.
To provide a language conversion method that requires a smaller amount of data than that of a translation method using a translation example sentence,
In addition, it has a pattern of the target language expressed efficiently,
To provide a language conversion method that can flexibly cope with the importance of additional information; furthermore, a means of evaluating the degree of conformity between the expression pattern of the target language and the input of the source language; The purpose of the present invention is to provide a language conversion method that makes it easy to modify the means for searching and identifying the information required by a pattern from the input of the source language and to easily improve the performance of the language conversion system over time. It was done.

[0006]

To achieve the above object, the present invention provides (1)
In a system that uses a computer to convert a source language into a target language, a database storing expression patterns of the target language, which includes information indicating restrictions and requirements in the target language.
And the central phrase that is semantically central to the input of the source language
Stores the expression pattern information of the target language in the database.
Representation of target language based on similarity with held central phrase
Goodness-of-fit evaluation that evaluates goodness of match between pattern and source language input
The original language according to the evaluation means and the fitness by the fitness evaluation means.
Select a target language expression pattern that matches the word input
Selection means and the target language to generate the output of the target language
From the input sentence of the source language
Search / identification means for searching and identifying, and selection by the selection means
The information required by the expression pattern of the target language
Source language input elements identified by search and identification means
Output means for generating an output of the target language by applying
The information needed to generate the output in the target language.
If it cannot be determined, the information is unknown
Generating incomplete target language output as is,
Further, (2) the analysis structure of the source language sentence is used as the input of the source language, and a combination of a surface phrase as a search key or a normal phrase such as a grammatical function or a semantic function and a surface phrase is used. that as a search key may perform search and identification, or (3) in a system which performs conversion to a target language using a computer from the source language, and the database of expression patterns of the target language, in the database
The match between the target language expression pattern and the source language input
A fitness evaluation means for evaluating, a search / identification means for searching and identifying information required by an expression pattern of the target language from an input of the source language to generate an output of the target language, and generating an output of the target language Query means for acquiring information necessary for the query by a question / response with the user, and selecting means for selecting an expression pattern of a target language suitable for input of the source language according to the fitness by the fitness evaluation means ; means
For the expression pattern of the target language selected in the above, search / identification of the input of the source language for the necessary information by the search / identification means, and question /
Output means for generating an output of the target language in combination with the response; and (4) information necessary for generating the output of the target language cannot be obtained by searching / identifying from the input of the source language. In such a case, a question is immediately issued to the user to compensate for the information, and (5) all the information necessary for generating the target language output is obtained by searching and identifying from the input of the source language. First, if there is information that could not be obtained, a question is issued to the user to compensate for the information, or (6) To convert from the source language to the target language using a computer, A database of expression patterns of the target language, a fitness evaluation unit for evaluating the fitness between the expression patterns of the target language and the input of the source language, and an expression parameter of the target language for generating an output of the target language. Turn required That information and a search and identification means to identify and search from the input of the original language, from the database of the source language
Select a target language expression pattern that matches the input, and
Search for the information required by the expression pattern
Search / identify from source language input by
Generates target language output from user inquiries
Means for performing the processing, wherein the database has a table of the target language.
Not only required elements but also optional elements are included in the current pattern.
Only the optional elements
Are different as one pattern
It has to allow the handle, further, (7) in the (6), and adds information indicating a degree of optional properties to the components of the pattern of the target language, depending on the information, The system includes a behavior determination unit that determines the behavior of the system when the component cannot be obtained from a source language input. (8) In (7), omitting each component of the target language pattern (9) The method further comprises: behavior determining means for determining a behavior of the system in a case where the component cannot be obtained from the input of the source language nor the inquiry to the user according to the degree of possibility. In (7) or (8), a behavior that determines a behavior of a system when the component cannot be obtained, according to a degree of omission of each component of the pattern of the target language. A plurality of deciding means, one of which is selectively used in accordance with a user's designation, etc. (10) In the above (7), the constituent elements of the pattern of the target language Is not obtained from the source language input, depending on the degree of omission of the component, the user is queried in some cases, and in some cases, the system continues processing with the component unknown. (11) In the above (7) or (8), when a component of the pattern of the target language cannot be finally obtained in (7) or (8), Depending on the degree of omission, a system may be used to generate a flagship that clearly indicates that a component is missing in some cases, and to generate an output ignoring the component in other cases. Have a behavior determining means for determining the behavior of the beam, or (1
To perform using a calculator conversion from 2) the source language to the target language, and information for evaluating the expression pattern of the target language, the adaptation of the input of the expression pattern and the original language of eye language, target language Means for searching and identifying information required by an object language pattern from an input of a source language in order to generate an output , and means for interpreting and applying the information described in the database Selection of the expression pattern of the target language that matches the input of the source language by the interpretation / application, and search / search for necessary information from the input of the source language.
Generating the output of the target language by performing the identification. (13) In the step (12), the information required for searching / identifying the information required by the expression pattern of the target language from the input of the source language. Alternatively or additionally, it has a database in which information for inquiring the user is described, and means for interpreting and applying the information. (14) Inputting the expression pattern of the target language and the source language Information to evaluate the relevance of the target language, and information to search and identify the information required by the target language pattern from the source language input to generate the target language output, or to query the user. In addition, it has a database in which information indicating the degree of omission of each component in the target language pattern is described. It is obtained by further comprising a behavior determining means for determining a household. Hereinafter, a description will be given based on examples of the present invention.

FIG. 1 is a diagram showing an embodiment of a language conversion system according to the present invention. In the figure, 1 is a control mechanism, 2 is a source language analysis mechanism, 3 is a fitness evaluation mechanism, and 4 is a search / identification mechanism. Reference numeral 5 denotes a target language sentence generation mechanism, and reference numeral 6 denotes a target language pattern database. The target language pattern database 6 stores expression patterns of the target language, and the conformity evaluation mechanism 3 evaluates the degree of conformity between the target language pattern in the target language pattern database 6 and the source language input. Select some target language patterns with high relevance. Next, the information required by the selected target language pattern for generating the target language output is searched and identified by the search and identification mechanism 4 from the source language input. It is conceivable that the source language analysis mechanism 2 is provided inside and the source language sentence character string is used as the source language input, or the source language analysis processing is performed externally and the analysis result is used as the source language input. Similarly, it is conceivable to use the target language sentence character string as the target language output by internally including the target language sentence generation mechanism 5, or to execute the target language sentence generation processing externally using the target language structure as the target language output. .

FIG. 2 is a flowchart for explaining a language conversion method according to the present invention. Hereinafter, the steps will be sequentially described. step1: First, input the structure of the analysis result of the source language. step 2: Determine the central phrase. step 3: Search for a matching target language pattern. step 4: Using the searched target language pattern, search, identify, and convert information necessary for generation from the source language input. Step 5: For information necessary for generation that is not obtained, substitute general substitute symbols in the target language pattern. step6: Output the target language structure.

Here, a case is shown in which a source language analyzing mechanism and a target language sentence generating mechanism are externally provided. Using the analysis result structure of the original sentence as input, a word or phrase that is semantically central in the original sentence is determined. Based on this central phrase, a matching target language pattern is searched from the target language pattern database. Then, using the pattern obtained by the search, the information required by the target language pattern is searched for from the structure of the original sentence.
Identify and translate into target language structure. If necessary information cannot be obtained, it is possible to respond to information loss by post-processing, for example, by changing the active voice to the passive voice. It tends to spoil. If information is missing, it is easier to understand if the target language is the native language, and it is easier to output an incomplete but natural syntax while leaving the information unknown. If the user is not good at the target language, the user can be conscious of the lack of information and can learn the information required by the target language. In this method, the missing information is substituted by a general substitute symbol in the target language pattern (an example of claim 1 ).

FIGS. 3A to 3E are diagrams showing examples of conversion. FIG. 3A shows an example of an input sentence and a conventional system, and FIG. 3B shows a target pattern of the present system (which can be seen by the user's eyes). (C) is a diagram illustrating an example of an intermediate output of the present method, FIG. (D) is a diagram illustrating an example of an analysis structure of a source language sentence, and FIG. (E) is a diagram illustrating an example of a target language pattern database. In this example, the source language is Japanese and the target language is English. Hereinafter, the operation of this method will be described based on this example. When the source language sentence “tested for substations” is parsed, a structure as shown in FIG. As shown by this structure, the word "do" is syntactically central in the original sentence, but semantically the phrase "do test" seems to be central. In this method, such a semantic center is called a central phrase. Determine the central phrase from the input structure.

Then, a target language pattern having the highest degree of matching with the determined central phrase is searched from the target language pattern database shown in FIG. In this case,
The target pattern shown in Fig. (B) [(Someone) conduc
ts a test of (something) {for (some purpose)}.]. Terms written in parentheses (), for example (Someon
e) or (something) is called a general substitute symbol in this method. The general substitute symbol is information necessary for generating a target language output based on this pattern. Omitting the phrase described in curly braces {} can produce a syntactically correct target sentence, but not omitting it can create a more natural completed target language sentence.

A search key for searching and identifying necessary information from an analysis structure of a source language sentence which is a source language input is described in a target language pattern database as shown in FIG. Information is retrieved using the key. The target language pattern corresponding to "test" is
conducts a test. In addition, one of the necessary information is to directly find the [sem substation] using the search key [surface phrase (object) -superstructure-function name (obj)] with the object and convert it to the structure of the object language. . On the other hand, the subject (Subj) is also necessary information, but since the corresponding information does not exist in the structure of the source language, a general substitute symbol is employed as it is without a translation corresponding to Someone. The output result is a pattern [(Someone) conducted a test of a
transformerstation {for (some purpose)}.].

When a language structure is searched, a direct partial structure is sequentially processed by using (grammatically or semantically) function names such as "sem" and "wo obj" which are partial structure labels. In general, the language conversion method according to the present invention can also use the search method.
The data description is much easier if the surface words such as can be specified as the search key. In this case, if multiple search keys are ordered as in the example shown above, it is necessary to set the partial structure obtained by searching with the first search key as a new search range and search with the remaining search keys. It is repeated sequentially (claim 3).

FIG. 4 is a block diagram for explaining an embodiment of another language conversion system according to the present invention. In FIG. 4, reference numeral 7 denotes an inquiry mechanism, and other parts which operate in the same manner as in FIG. The code is attached. Target language pattern database 6
Stores the expression pattern of the target language, and the conformity evaluation mechanism 3 evaluates the conformity between the target language pattern in the target language pattern database 6 and the source language input. Select some target language patterns. Next, the information required by the selected target language pattern for generating the output of the target language is searched / identified from the source language input by the search / identification mechanism 4 or the user is inquired by the inquiry mechanism 7. It is conceivable that the source language analysis mechanism 2 is provided inside and the source language sentence character string is used as the source language input, or the source language analysis processing is performed externally and the analysis result is used as the source language input. Similarly, it is conceivable to use the target language sentence character string as the target language output by internally including the target language sentence generation mechanism 5, or to execute the target language sentence generation processing externally using the target language structure as the target language output. .

FIG. 5 is another flowchart for explaining the language conversion method according to the present invention. Hereinafter, the steps will be sequentially described. step1: First, input the structure of the analysis result of the source language. step 2: Determine the central phrase. step 3: Search for a matching target language pattern. step 4: Using the searched target language pattern, search, identify, and convert information necessary for generation from the source language input. step5: Acquire missing information by dialogue with the user. step 6: If the information required for generation is not acquired, substitute general substitute symbols in the target language pattern. step7: Output the target language structure.

Here, a case is shown in which a source language analysis mechanism and a target language sentence generation mechanism are externally provided. Using the analysis result structure of the original sentence as input, a word or phrase that is semantically central in the original sentence is determined. Based on this central phrase, a matching target language pattern is searched from the target language pattern database. Then, using the target language pattern obtained by the search, information necessary for the target language pattern is obtained from the structure of the original sentence by searching / identifying or inquiring the user, and is converted into the structure of the target language. If necessary information is still not obtained, substitute general substitute symbols in the target language pattern.

There are various ways to combine the search / identification from the source language input and the inquiry to the user.
If the inquiry to the user is given priority, the certainty of the information to be obtained is increased, but the user is forced to perform complicated work. Therefore, the search / identification from the source language input is given priority to reduce unnecessary queries. Should. When necessary information is not obtained by the search / identification from the source language input, even if the user is inquired each time (claim 4 ), the contents of the inquiry are stored and the next search / identification work is performed. To make inquiries at the end
5 ) Useless inquiries can be reduced compared to inquiring first. According to the method of the fourth aspect, there is no need to store the contents to be queried and perform the next search / identification work, so that the method is efficient and the processing is simple. On the other hand, in the method of claim 5 , at the time of inquiry, all information obtained by search / identification from the source language input (and all information obtained by previous queries) can be used. Unnecessary inquiries can be further reduced, or inquiries can be made with question sentences that are easy for the user to answer.

FIGS. 6A to 6E show examples of conversion. FIG. 6A shows an example of an input sentence and a conventional method, and FIG. 6B shows a target pattern of the present method (which can be seen by the user's eyes). (C) is a diagram illustrating an example of an intermediate output of the present method, FIG. (D) is a diagram illustrating an example of an analysis structure of a source language sentence, and FIG. (E) is a diagram illustrating an example of a target language pattern database. In this example, the source language is Japanese and the target language is English. Hereinafter, the operation of this method will be described based on this example. When the source language sentence “tested for substations” is parsed, a structure as shown in FIG. As shown by this structure, the word "do" is syntactically central in the original sentence, but semantically the phrase "do test" seems to be central. In this method, such a semantic center is called a central phrase. Determine the central phrase from the input structure.

Then, a target language pattern having the highest matching degree with the determined central phrase is searched from the target language pattern database shown in FIG. In this case,
The target pattern shown in Fig. (B) [(Someone) conduc
ts a test of (something) {for (some purpose)}.]. Terms written in parentheses (), for example (Someon
e) or (something) is called a general substitute symbol in this method. The general substitute symbol is information necessary for generating a target language output based on this pattern. Omitting the phrase described in curly braces {} can produce a syntactically correct target sentence, but not omitting it can create a more natural completed target language sentence.

A search key for searching and identifying necessary information from an analysis structure of a source language sentence which is a source language input is described in a target language pattern database as shown in FIG. Searches using that key.
The target language pattern corresponding to "test" is con
ducts a test. Also, one of the necessary information is the search key directly with the object.

Find [sem substation] using [surface phrase (target)-higher-level structure-function name (obj)] and convert it to the structure of the target language.
On the other hand, the subject (Subj) is also necessary information, but since the corresponding information does not exist in the structure of the source language, by inquiring the user, it becomes “the government a
gency ”.

[0021] The question in this case is, in the method of claim 4, for example, "translation candidate" (Someone) conducts a test of
(something) [for (some purpose )]. part of the subject of "(Someone) I do not know. Who is" to test "is given?" and, in the method of claim 5, for example, "translation candidate “(Someone) conducted a test of (a) transfor
merstation [for (some purpose)]. "(Someone) in the subject part is not known. Who is" test substation "?" Next, “transformer stati
Query for the article "on" to get the final output result.
The general substitute symbol is used as it is for the missing information that has not been obtained or has not been inquired. The output result is, as shown in FIG. 6C, a pattern [The government agency conducted a te
st of the transformer station [for (some purpos
e)].] is the target language sentence.

In the above description, when the conversion from the source language to the target language is performed by using a computer, the database of expression patterns of the target language and the evaluation of the degree of conformity between the expression pattern of the target language and the input of the source language. Means, and means for searching and identifying information required by the pattern of the target language from the input of the source language to generate an output of the target language, and selecting a pattern of the target language that matches the input of the source language. The present invention relates to a language conversion system for generating an output of a target language by searching and identifying an input of a source language of necessary information or inquiring a user.

When trying to express certain information in a language,
Additional information that is indispensable for expressing in the language, important additional information that is not indispensable but that is missing and unnatural or difficult to understand, especially though not required from the language side Additional information whose meaning can be refined by this, additional information of low importance that can be added but does not make much difference even if it is added, There is additional information such as extra additional information that may be unnatural or difficult to understand. For example, when trying to express "testing for substations" in English, it is indispensable that "who performed the test" and "why did the test"? Is important for understandability. On the other hand, the distinction of politeness in Japanese, such as “performed” or “performed”, can be considered extra additional information. The conversion system according to the target language led described above, by generating the target language output began to acquire the information target language pattern is requested from the source language input or user, a low additional information or extra additional information importance By removing it, a natural output can be obtained as the target language. However, there are many combinations of additional information that is not indispensable as a target language expression and what is not added, and if each of them is a separate pattern, it is inefficient from both capacity efficiency and processing efficiency. In addition to being economical, it is difficult to flexibly respond to the importance of the additional information.

The following embodiment solves such a point. Since the basic configuration and processing flow are the same as those described above, an example of conversion by this system will be described here. Consider a case where Japanese is the source language and English is the target language, and the source language input of "Testing was conducted for substations." The English pattern in this case is basically “<SOMEONE> products at
est of <SOMETHING>, but this pattern includes "<FOR SOME PURPOS".
E> ”,“ <SOMEHOW> ”,“ <SOMETIM ”
E>"can be added. If each of these added patterns is treated as a separate pattern, the number of patterns increases explosively, so that the present system treats them as one pattern including components that can be omitted. (Claim
6 ).

In addition, even with the same optional component, "<FO
In the absence of “R SOME PURPOSE>”, an unstable feeling is given, whereas there is no “<SOMETIME>” etc., and the stability is obtained.
ONE> ”, including the essential elements such as ONE>”
(Such as "<SOMEONE>"), preferred
(Such as “<FOR SOME PURPOSE>”),
It can be set to three stages of optional (such as "<SOMETIME>"). The behavior of the system when the pattern component could not be acquired may be asking the user, continuing the process with the element unknown, or ignoring the element.
The behavior can be selected according to the omissibility of the element. In this example, the only component that can be obtained from the source language input is “substation” corresponding to “<SOMETHING>”, but the processing for the other components that could not be obtained is determined according to the possibility of omission. . For example, queried required, the user for preferred elements as in FIG. 7, optional elements will continue processing while the unknown (claim 7, claim 1
0 ). Here, it is also possible to inquire only about required ones, and to continue the process with unknown the preferred and optional ones, and to select which one is selected by the user (Claim 9 ).

Even if some components cannot be finally obtained, depending on the possibility of omitting the components,
Regarding the queried and preferred elements, "<S
OMEONE> ”and the like, and outputs are generated while leaving the general substitute symbols, ignoring optional elements (claims 7 and 10 ). Again, only required ones are explicitly indicated as missing information, and preferred elements are specified. Can be ignored.

In the above embodiment, the means for evaluating the degree of conformity between the expression pattern of the target language and the input of the source language and the information required by the pattern of the target language in order to generate the output of the target language are input to the source language. The means for searching and identifying from the database is realized by an easily modifiable database and its interpretation / application means. It has also been proposed to add information indicating the possibility of omitting each component to the expression pattern of the target language. However, there is a means for evaluating the degree of conformity between the expression pattern of the target language and the input of the source language, and a means for searching and identifying information required by the pattern of the target language from the input of the source language in order to generate an output of the target language. If the modification is difficult, it is difficult to gradually improve the performance of the language conversion system. If information indicating the possibility of omitting a component is included in the expression pattern word of the target language, the same data is separately stored in different expression patterns representing almost the same meaning, which is space-efficient and uneconomical. It is difficult to modify the data.

The following embodiment solves this problem. Since the basic configuration and processing flow are the same as those described above, an example of conversion by this system will be given here. Japanese is the source language and English is the target language. In the present invention, the information for evaluating the conformity between the expression pattern of the target language and the input of the source language and the information required by the target language pattern to generate the output of the target language are searched and identified from the input of the source language. Table 1 shows an example of data in which information for performing the operation is described.

[0029]

[Table 1]

Hereinafter, this data is referred to as a Japanese-English required information pattern. This is: Means for searching for an English pattern that is more suitable for a phrase that is the semantic center of Japanese input. It is used as a means to search and identify information necessary for English generation from Japanese input. The structure type of the information type for each means will be described. 1. In order to search for an English pattern that matches the central phrase, there is a partial structure with a label <English-Patterns>. Each of these has a cross-reference of the pattern in the English sentence pattern dictionary, and <English-P
There are also conditions for selecting the most applicable molecules in atterns>. For example, using the numerator <v-test-002>, the verb phrase “conduct a test” is generated, but in Japanese input, the verb “test” is nounized, and It is considered appropriate when it becomes a verb phrase. This is described in <Restrictions> for <v-test-002>. In other words, there is a condition of (＠Nounization surface phrase (do) -superstructure-function name (obj)) or a condition of BINARY_CONTRAST test.

2. <Required> and <Preferred>
There are two molecules>. The molecules in <Required> are considered grammatically essential. For example, the subject of <Subj-AGT> or the object of <DirObj-PATIENT>. <
Even though the molecules in Preferred> are not essential, they are considered necessary to fully convey the meaning of the central phrase.
For example, most of the time we test, we have already decided what to test. In English, conduct a te
When we say st, we often say for (some reason). In order to demonstrate this phenomenon, or to teach users who are using Japanese,
Preferred> molecule is adopted. To search for this information from Japanese input, there is a molecular structure called <Japanese>. In this structure, a Japanese structure corresponding to object information, that is, a subject / object, etc., is described. For example, the object for "test" is mainly "object" and "X
Test on X. There are many sentences like Table 2 shows the format of the English sentence pattern dictionary.

[0032]

[Table 2]

A dictionary for generating an English output using information searched and identified by the English necessary information pattern database. Describe the target pattern of the English sentence <
There are two molecules, <Pattern> and <Component> composed of the sentence. For example, <Pattern> of <v-test-002>
Is (conduct <n-test-002>) or (conduct <n-test-002
003>) and <Pattern> of <n-test-002> is (test
(<P-of-001>) (<vinf-to determine-002>).
The (<p-of-001>) of <Pattern> of <n-test-002> means a prepositional phrase (of + noun phrase). The object noun phrase of the preposition is represented by the numerator of <Component> in the form of (X1X
1) and the corresponding grammar function is <DirObj-PATIENT>
It is going to be described. Similarly, <Patt of <n-test-002>
ern>(<vinf−−todetermine−−002>)
It means a verb phrase.

FIG. 8 shows an example of conversion using this format.
FIG. 8A shows an original Japanese sentence and a representative English translation obtained by a conventional method. FIG. 8B is an example of an English target sentence pattern for generating a sentence using this format.
FIG. 8C shows an example of an intermediate output generated by a method using this format. FIG. 8D shows an example of an analysis structure of a Japanese sentence. By parsing the source language sentence “Testing was performed on the substation,” a structure as shown in FIG. 8D is obtained. As shown by this structure, the word "do" is syntactically central in the original sentence, but semantically the phrase "do test" seems to be central. In this form, such a semantic center is called a central phrase. Determine the central phrase from the input structure, parse it into the composed parts, and use <Engl
ish-Patterns> to search for the inquiring candidate. In this example sentence, further analysis of the phrase "test" is the central phrase, "verify" and "verify" the verb "do"
To obtain a structure composed of two parts. The entry in FIG. 8 is searched using “test” as a key.
Next, <Restri of <English-Patterns> of the entry
ctions>, (＠nounization surface phrase (do)-higher-order structure-function name (obj))
Make <v-test-002> the target pattern.

After the pattern has been searched, it is confirmed whether or not the information required to generate the English output exists in the original Japanese sentence. <Subj-AGT> by structural molecule <Required>
Is required. Due to the molecule <Japanese>, the "ga" case in the original text corresponds to <Subj-AGT>. The parsing structure of the original text has no case, so necessary information is omitted.
Substitute the last candidate in the molecule <Alternates>. In this example, SOMEONE is a general substitute symbol. Similarly, the following <R
Equired> search for the molecule <DirObj-PATIENT>. In this case, since “corresponding to OO” is described as (corresponding to {targeting}) as the corresponding Japanese, the original substation is detected, and the English <DirObj-PATIE
NT>. Next, <Preferred> information of the pattern is searched from the original text. None of the three molecules described in this example are converted because they are not in the original text. However, here, the information necessary to inquire the user, such as an instruction to ask the user if necessary and a question text (ask
If it is in the necessary information in Table 1 as in "What is the purpose of the test?", Ask the user.

After the above search is completed, an English output is generated using the English sentence pattern dictionary (Table 1) of the present invention. Using <v-test-002> that was searched before, (co
nduct <n-test-002>) or (conduct <n-test-003>
Search for the condition of>). Choose between <n-
Determine under the conditions of test-002 and <n-test-003>. In both cases, the condition is described in the molecule <Component>. For <n-test-002>, <Xnnn-BINARY_CO
There is a <Component> called NTRAST>, and <n-test-003>
In the case of <Component>, it was <Xnnn-BOOLEAN>. However, since neither exists in the Japanese original text, it cannot be selected. In English, either <Comp
Onent> allows you to generate a more natural output, so at this stage the first <Xnnn-BINARY_CONTRA
ST>. The generated output can be displayed to the user and completed by the user's judgment.

[0037]

As apparent from the above description, the present invention has the following effects. (1) Effect corresponding to the first aspect: It is easier to generate an appropriate and natural translated sentence than the forward structure conversion method. The required data amount is smaller than that of the translation method using the bilingual example sentence. If the target language is the user's native language, it is easy for the user to understand.If the user is not good at the target language, let the user be aware of missing information and learn about the information required by the target language. Can be encouraged. (2) Effect corresponding to claim 2 : Data description is much easier. (3) Effect corresponding to claim 3 : It is easier to generate an appropriate and natural translated sentence than the forward structure conversion method.
The required data amount is smaller than the method of acquiring all information by inquiring the user. Does not give the user a sense of control. Less likely to produce incomplete output. (4) Effect corresponding to claim 4 : Useless inquiries can be reduced. Since there is no need to store the contents to be queried and to perform the next search / identification work, it is efficient and the processing is simple. (5) Effect corresponding to claim 5 : Useless inquiries can be reduced. At the time of inquiry, all information obtained by searching / identifying from source language input (and all information obtained by previous queries) can be used, so that unnecessary queries are further reduced or used Can make inquiries using question sentences that are easy to answer. (6) An effect corresponding to claim 6 : By allowing a target language pattern to include not only essential elements but also optional elements, a plurality of patterns having different optional elements are treated as one pattern. Capacity efficiency and processing efficiency are improved. (7) Effects corresponding to claims 7 to 11 : Flexible processing can be performed according to the possibility of omitting each component in the target language pattern and the user's desire. (8) Advantages Corresponding to Claims 12 and 13 : The gradual improvement of the performance of the system is facilitated. (9) Effect corresponding to claim 14 : The gradual performance improvement of the system is facilitated, and the capacity efficiency is also improved.

[Brief description of the drawings]

FIG. 1 is a configuration diagram for explaining an embodiment of a language conversion system according to the present invention.

FIG. 2 is a flowchart illustrating a language conversion method according to the present invention.

FIG. 3 is a diagram showing a conversion example.

FIG. 4 is a diagram showing another embodiment of the language conversion system according to the present invention.

FIG. 5 is another flowchart illustrating a language conversion method according to the present invention.

FIG. 6 is a diagram showing another conversion example.

FIG. 7 is a diagram showing a display example.

FIG. 8 is a diagram showing still another conversion example.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 control mechanism 2 source language analysis mechanism 3 conformity evaluation mechanism 4 search / identification mechanism 5 target language sentence generation mechanism 6
... target language pattern database, 7 ... inquiry mechanism.

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-3-276367 (JP, A) JP-A-4-160473 (JP, A) Sumida, Iida, Yukiyama, "Based on the example for (1F-5) "Translation", IPSJ 40th (Early 1990) National Convention Lecture Papers (I), March 14, 1990, p. 411-412 (58) Fields surveyed (Int. . ⁶ , DB name) G06F 17/28

Claims (14)

(57) [Claims] In a system for performing conversion from a source language to a target language by using a computer, a target language expression pattern comprising information indicating restrictions and requirements in the target language is stored in a case.
And the semantic center of the input of the source language
And the expression pattern of the target language in the database
Based on the similarity to the central phrase stored in the
Assess the relevance between the expression pattern of the target language and the input of the source language.
Conformity evaluation means to be evaluated and conformance by the conformity evaluation means
Expression pattern of target language that matches input of source language according to degree
Selection means for selecting a language and generating output in the target language
The information required by the expression pattern of the target language
Searching / identifying means for searching and identifying from the input sentence of
The expression pattern of the target language selected by the selection means requires
The original language identified by the search / identification means
Output that applies the elements of the word input to produce the output in the target language
Power means for generating an output of the target language.
If the necessary information cannot be determined,
Output of an incomplete target language with the information of
Language conversion method which is characterized in that formed. 2. A parsing structure of a source language sentence is used as the input of the source language, and a combination of a surface phrase as a search key or a normal phrase such as a grammatical function or a semantic function and a surface phrase is searched. 2. The language conversion method according to claim 1, wherein search / identification can be performed as a key. 3. A system for performing conversion from a source language to a target language using a computer, wherein a database of expression patterns of the target language and a degree of conformity between the expression pattern of the target language in the database and the input of the source language. The fitness evaluation means to be evaluated and the target language to generate the output of the target language
Searching / identifying means for searching and identifying information required by the expression pattern from the input of the source language, querying means for acquiring information necessary for generating an output of the target language by a question / response with the user, and selection means for selecting the representation pattern <br/> over down the target language conforming to the input of the source language in accordance with the fitness by the fitness evaluation means, selected by said selection means eyes
Output for generating an output of a target language by a combination of a search / identification of an input of a source language for necessary information by the search / identification means and a question / response with a user by the inquiry means for an expression pattern of a target language And a language conversion method. 4. If the information necessary for generating the output of the target language is not obtained by searching / identifying from the input of the source language, a question is immediately issued to the user to supplement the information. The language conversion method according to claim 3, wherein 5. An attempt is first made to obtain all the information necessary for generating the target language output by searching and identifying from the input of the source language. If there is any information not obtained, a question is issued to the user. 4. The language conversion method according to claim 3, wherein information is compensated for by using the information. 6. A computer for converting a source language to a target language using a computer, comprising: a database of target language expression patterns; and a target language expression pattern and source language input.
A fitness evaluating means for evaluating the fitness, and the search and identification means for identifying and retrieving information requested by the expression pattern of the target language to produce an output in the target language from the input of the source language, the original from the database Purpose to adapt to language input
Selection of language expression pattern and expression pattern of the target language
Information required by the source language by the search / identification means.
Search / identify from user input or ask user
Means for generating the output of the target language
The database is required in the expression pattern of the target language
Enable to include not only elements but also optional elements
Multiple elements that differ only in optional elements
To be able to treat this pattern as one pattern
Language conversion method being characterized in that the. 7. A system according to claim 1, wherein information indicating the degree of omission is added to each component of said target language pattern, and said component is not obtained from a source language input according to said information. 7. The language conversion method according to claim 6, further comprising a behavior determining means for determining a behavior. 8. The behavior of the system when the component cannot be obtained from the input of the source language nor the inquiry to the user according to the degree of omission of each component of the pattern of the target language. The language conversion method according to claim 7, further comprising a behavior determining means for determining. 9. A system comprising a plurality of behavior determining means for determining the behavior of the system when the component cannot be obtained, according to the degree of omission of each component of the target language pattern, 9. The language conversion method according to claim 7, wherein one of the behavior determining means is selectively used from among them according to designation. 10. When a component of the pattern of the target language cannot be obtained from a source language input, an inquiry is made to a user in some cases according to the degree of omission of the component, and in another case, the configuration is made. 8. The language conversion method according to claim 7, further comprising a behavior determining means for determining a behavior of the system so that the processing is continued with the element unknown. 11. In a case where a component of the pattern of the target language cannot be finally obtained, depending on the degree of omission of the component, it may be determined that the component is missing in some cases. 9. The language conversion method according to claim 7, further comprising a behavior determination unit that determines a behavior of the system so as to generate a main component to be specified and, in some cases, generate an output ignoring the components. 12. In order to perform conversion from a source language to a target language by using a computer, an expression pattern of a target language, and a degree of conformity between an expression pattern of a target language and input of a source language are evaluated. That describes information for searching for and identifying information required by a target language pattern from an input of a source language in order to generate an output of the target language
Means for interpreting and applying the information described in the database , selecting a target language expression pattern that matches the input of the source language by the interpretation and application, and searching and identifying necessary information from the input of the source language A language conversion method characterized by generating an output of a target language by performing the following. 13. A database in which information for inquiring a user is described instead of or in addition to information for searching and identifying information required by the expression pattern of the target language from an input of a source language. 13. A means for interpreting and applying information.
Language conversion method described. 14. Information for evaluating the degree of conformity between the expression pattern of the target language and the input of the source language, and information required by the pattern of the target language to generate an output of the target language are obtained from the input of the source language. In addition to the information for searching and identifying or the information for inquiring of the user, it has a database describing information indicating the degree of omissibility of each component in the pattern of the target language. 14. The language conversion method according to claim 12, further comprising a behavior determining means for determining a behavior of the system according to the information.