JP2001325254A - Expression conversion method and expression conversion device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an expression conversion device in compact constitution capable of performing a processing at a high speed and outputting a result for correctly telling an intention even when a recognition error is generated. SOLUTION: This device is provided with a dependency relation analysis means 2 for selecting important words from sentences for the respective sentences of a corpus 1, specifying the combination of the important words with co-occurrence relation from the prescribed number of the combinations of the important words and making the specified combination of the important words and an expression after the conversion of the sentence which is the object of selecting the important word correspond beforehand and an important word extraction means 5 and an example sentence selection means 7 for selecting the important words decided beforehand from an inputted sentence, combining them, comparing the combination of the combined important words with the combination of the important words specified beforehand of the respective sentences, selecting one, two or more sentences whose compared results match or are highly similar and outputting the expression after the conversion of the sentence.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an expression conversion method, an expression conversion device, and a program for converting and outputting an input sentence expression. The present invention relates to an expression conversion method, an expression conversion device, and a program for performing, for example, a sentence pattern conversion that converts a sentence into a spoken language, a summary sentence that summarizes and outputs a complicated or redundant sentence.

[0002]

2. Description of the Related Art A conventional technique will be described below using translator software as an example.

[0003] The interpreter software comprises speech recognition means and language translation means. Speech recognition for converting an utterance sentence input as an acoustic signal into an output sentence displayed as a word text sequence, and word interpretation. Interpretation is realized by sequentially executing a language translation as an expression conversion for inputting a sentence displayed in a column and translating the sentence into another language sentence.

Further, the language translating means for executing the language translation as described above includes a language analyzing means for analyzing a syntactic or semantic structure of an input sentence, and a language converting means for converting to a different language based on the analysis result. And an output sentence generating means for generating a natural output sentence from the translation result.

[0005] Interpreting software and some translation software do not carry out linguistic analysis in accordance with grammatical rules in order to correctly translate frank expressions that are frequently found in spoken language and deviate from grammatical rules. In many cases, a technique of learning an example sentence similar to a sentence actually input and performing language analysis by searching for a similar example sentence is used. Less than,
An example of such conventional interpreter software will be described with reference to the example of FIG.

Here, a case where a Japanese utterance example is translated into an English utterance example will be described. The language before the translation is called the source language, and the language after the translation is called the target language.

Before translating, utterance rules are extracted from the bilingual corpus 1 in FIG. In the bilingual corpus 1, a plurality of bilingual utterance sentences, which are pairs of a Japanese utterance sentence example and an English utterance sentence example, are described. FIG. 9A shows an example of a bilingual utterance sentence described in the bilingual corpus 1 as a bilingual utterance sentence example 70.

Here, taking into account that some words are uttered with recognition errors or omissions, the utterance sentence example is a minimum unit as a semantic unit (hereinafter referred to as a phrase).
And rules in phrases and dependency rules between phrases are created.

First, the phrase determination means 61 divides a bilingual utterance sentence example into phrase units. FIG. 9B shows the bilingual phrase divided in this way as bilingual phrase (A) 7.
1, the translated phrase (B) 72 is shown.

Next, a bilingual phrase dictionary creating section 62 creates a corresponding phrase dictionary 62 in which the content words in the phrase are converted into variables.

For example, in the bilingual utterance sentence example 70 shown in FIG. 9A, “I would like to request a room reservation.”
I'd like to reserve a room "is described, and when these are divided into phrases, the translated phrase (A) 71 becomes (A)" reservation of room "and" rese
rve a room ”as a bilingual phrase (B) (B)“ I'd like to ask, ”“ I'd like to ”(A)
(B) It is divided into two bilingual phrases.

Content words such as "room", "reservation", and "request" are represented by variables XYZ, respectively, using a classification word table 64 prepared in advance as shown in FIG. 9 (e). . Here, the classification word meaning table 64 is a list in which content words that can be taken as values of each variable are described. For example, variable X
Takes values such as "room", "meeting room", "car"
The content word “room” is a value that the variable X can take. Therefore, the content word “room” in the bilingual phrase (A) 71 is replaced with the variable X.

In this way, (A) "Y of X""Y
Two bilingual phrase rules “X” and (B) “I want to do Z” and “I'd like to” are described in the bilingual phrase dictionary 62.

In order to regularize the order relation of each phrase, "(A) is converted to (B)" and "(B) (A)" as shown in FIG. Are stored in the inter-phrase rule table 63. Such processing is performed for all utterance sentences included in the bilingual corpus 1.

At the time of interpreting, first, the uttered source language voice is input to the voice recognition means 64. The speech recognition unit 64 extracts, for example, an acoustical word from a word string that may be predicted from a word string described as a phrase in the bilingual phrase dictionary 62 and a phrase string described in the inter-phrase rule 63. Is output as a speech recognition candidate.

The language translation means 65 inputs the continuous word string recognized in this way, converts the input continuous word string into a phrase string described in the bilingual phrase dictionary 62, and corresponds to each phrase string. The inter-phrase rule 63 is searched. Then, the input source language recognition result sentence is converted into the target language sentence based on the target language phrase that is a bilingual translation of each phrase and the inter-phrase rule of the target language.

The converted target language sentence is output to output sentence generating means 6.
6 to correct grammatical unnaturalness. For example, optimization such as third personization, pluralization, past tense, and the like of pronouns, verbs, and auxiliary verbs, and optimization of the structure of the entire sentence are performed. The corrected target language translation result sentence is output, for example, as text.

[0018]

However, conventional interpreting software has the advantage of being able to handle input sentences that deviate from grammatical rules, but on the other hand, many different bilingual phrases and their combinations are directly described as rules. Therefore, there is a problem that the conversion rules are complicated and enormous, and the processing time is long.

The grammatical check performed by the output sentence generation unit also requires complicated rules. In particular, there is no technique that can completely and completely correct zero pronouns, and sometimes erroneous completion is performed. The problem is that

When a partially incorrect sentence is input to the language translating unit due to a speech recognition error or the like, language conversion is performed based on rules within and between erroneous phrases, so that a translation that does not convey any intention is not transmitted at all. There is a problem of outputting a result.

An object of the present invention is to provide an expression conversion method, an expression conversion device, and a program which can be processed at high speed with a compact configuration in consideration of the above problems.

Further, in consideration of the above-mentioned problems, the present invention provides a result that even if a recognition error occurs in an input sentence other than an important word due to a speech recognition error or the like, the intention is correctly transmitted without adversely affecting the quality of the output sentence. It is an object of the present invention to provide an expression conversion method, an expression conversion device, and a program, which can output.

In addition, the present invention has been made in consideration of the above-described problem, and avoids outputting a result in which the sentence is not transmitted as in the related art even if a recognition error occurs in an input sentence due to a speech recognition error or the like. It is an object of the present invention to provide an expression conversion method, an expression conversion device, and a program capable of performing the following.

[0024]

In order to solve the above-mentioned problems, a first aspect of the present invention (corresponding to claim 1) is that, for each sentence of a corpus, an important word is selected from the sentence and the important words are selected. Of the words, a combination of key words having a co-occurrence relationship is specified from a predetermined number of combinations, and the combination of the specified key word and the sentence from which the key word is selected are converted. Expressions are made to correspond in advance, and a predetermined important word is selected from the input sentence, combined, and a combination of the combined important word and a combination of the specified important word of each sentence are determined. In contrast,
This is an expression conversion method in which one or more sentences having a high degree of similarity or similarity are selected and the converted expression of the sentence is output.

Further, the second invention (corresponding to claim 2)
For each sentence in the corpus, an important word is selected from each sentence using classifying information in which the important words are classified into classes based on predetermined properties and the class names are given. Among the classes to which the class belongs, a combination of classes having a co-occurrence relationship is specified from a predetermined number of combinations, and the converted combination of the specified class combination and the sentence from which the important word is selected. Are selected in advance from the input sentence, a class to which each of the important words belongs is combined, the combination of the combined class, and the identification of each sentence in advance are selected. And a sentence with a high degree of similarity or high similarity.
This is an expression conversion method that outputs the converted expression of the sentence selected above.

The third invention (corresponding to claim 3)
For each sentence in the corpus, select important words from the sentence, identify a combination of important words that have a co-occurrence relationship from a predetermined number of combinations of those important words,
The specified combination of important words and the converted expression of the sentence from which the important word is selected are made to correspond in advance, and the important words are classified into classes based on predetermined properties, and the class names are given. By using the classifying information provided, the combination of the identified important words is made to correspond to the combination of the classes, so that the combination of the classes of the sentence is specified, and the important word determined in advance from the input sentence is determined. Is selected, the classes to which the respective important words belong are combined, and the combination of the combined classes is compared with the combination of the previously specified classes of each sentence,
This is an expression conversion method in which one or more sentences having a high degree of similarity or similarity are selected and the converted expression of the sentence is output.

The fourth invention (corresponding to claim 4)
For each sentence in the corpus, select important words from the sentence, identify a combination of important words that have a co-occurrence relationship from a predetermined number of combinations of those important words,
Associating means for associating the specified combination of important words with the converted expression of the sentence from which the important word is selected, and selecting a predetermined important word from the input sentence In the combination, a combination of the combined important words and a combination of the previously specified important words of each sentence are compared, and one or more sentences having a high matching or similarity in the comparison result are selected. A conversion unit that outputs a converted expression of a sentence.

The fifth invention (corresponding to claim 5)
When the comparison result has a high degree of similarity, the conversion means is an important word belonging to the combination of the important word that is not matched and is not included in the combination of the important word that matches. For the important word, the converted part is removed from the selected expression and the expression is output.
Is a representation conversion device according to the present invention.

The sixth invention (corresponding to claim 6)
Is the expression conversion device according to the fourth aspect of the present invention, wherein the converted expression is composed of only important words or bilingual words of the important words.

The seventh invention (corresponding to claim 7)
For each sentence in the corpus, an important word is selected from each sentence using classifying information in which the important words are classified into classes based on predetermined properties and the class names are given. Among the classes to which the class belongs, a combination of classes having a co-occurrence relationship is specified from a predetermined number of combinations, and the converted combination of the specified class combination and the sentence from which the important word is selected. A pre-determined important word is selected from the input sentence, a class to which the important word belongs is combined, a combination of the combined class, and a A conversion means for comparing a combination of the classes specified in advance, selecting one or more sentences having a high degree of matching or similarity, and outputting a converted expression of the sentence; Example was a representation conversion device.

An eighth aspect of the present invention (corresponding to claim 8)
For each sentence in the corpus, select important words from the sentence, identify a combination of important words that have a co-occurrence relationship from a predetermined number of combinations of those important words,
The specified combination of important words and the converted expression of the sentence from which the important word is selected are made to correspond in advance, and the important words are classified into classes based on predetermined properties, and the class names are given. Utilizing the classifying information provided, the specified combination of important words is made to correspond to the combination of classes, so that the association means for specifying the combination of the classes of the sentence and the sentence inputted are determined in advance. Are selected, the classes to which the respective important words belong are combined, and the combination of the combined classes is compared with the combination of the previously specified classes of each sentence. And a conversion means for selecting one or more sentences having a high sentence and outputting a converted expression of the sentence.

The ninth invention (corresponding to claim 9)
When the comparison result has a high degree of similarity, the conversion unit determines, for a class belonging to the non-matching class combination and a class belonging to the matching one that is not included in the class combination, , Removing the converted part from the selected expression and outputting the expression.
Or 8 is the expression conversion device according to the present invention.

A tenth aspect of the present invention (corresponding to claim 10) is the expression converting apparatus according to the seventh or eighth aspect of the present invention, wherein the converted expression is composed of only class terms.

According to an eleventh aspect of the present invention (corresponding to claim 11), for each sentence of the corpus of the expression conversion apparatus according to the fourth aspect of the present invention, an important word is selected from the sentence and the important words are selected. Of the words, a combination of key words having a co-occurrence relationship is specified from a predetermined number of combinations, and the combination of the specified key word and the sentence from which the key word is selected are converted. Means for associating expressions with the expressions in advance; selecting and combining predetermined important words from input sentences; combinations of the combined important words; and the previously specified important words of each sentence And a computer that functions as all or a part of a conversion unit that outputs a converted expression of the sentence by selecting one or more sentences that have a high degree of similarity or similarity as a result of the comparison. Which is the program.

According to a twelfth invention (corresponding to claim 12), in the expression conversion apparatus according to the seventh invention, important words are classified into classes based on predetermined characteristics in advance and given class names. For each sentence in the corpus, the important words are selected from the sentence using the classified information, and the classes having a co-occurrence relationship are selected from a predetermined number of combinations of the classes to which the respective important words belong. A combination of the specified class, and a correspondence unit that previously associates the combination of the specified class with the converted expression of the sentence for which the important word is selected, and a predetermined combination based on the input sentence. Are selected, the classes to which the respective important words belong are combined, and the combination of the combined classes is compared with the combination of the previously specified classes of the respective sentences, and the comparison result is one. Or a high degree of similarity statement 1 or 2 or more Pick a program for causing a computer to function as all or part of the conversion means for outputting a representation of the converted of the statement.

According to a thirteenth aspect of the present invention (corresponding to claim 13), for each sentence of the corpus, an important word is selected from the sentence of the expression conversion apparatus according to the eighth aspect of the present invention, and the important words are selected. Of the words, a combination of key words having a co-occurrence relationship is specified from a predetermined number of combinations, and the combination of the specified key word and the sentence from which the key word is selected are converted. Expressions are previously associated with each other, and the combinations of the identified important words are made to correspond to the combinations of the classes by using classifying information in which the important words are classified into classes based on predetermined properties and the class names are given in advance. By this means, a matching means for specifying a combination of the classes of the sentence, and a predetermined important word selected from the input sentence, and a class to which each of the important words belongs is combined, and the combination is selected. The combination of the matched classes is compared with the combination of the classes previously specified in each sentence, and one or more sentences having a high degree of matching or similarity are selected, and the converted expression of the sentence is selected. This is a program for causing a computer to function as all or a part of the output conversion means.

Next, the operation of the present invention will be described.

According to the present invention, an important word is extracted from an input sentence, and the extracted important word is converted into a standard or simplified expression sentence having the same sentence, and the converted sentence is output. The generation rule can be made compact, and the expression conversion can be performed by simple processing. In addition, even if an incorrect sentence is input in a part other than the important words, it can be converted to an expression that can understand the meaning correctly, and it is possible to solve the problem that the expression conversion result is output as it is incorrect as in the past Becomes

Further, according to the present invention, the words of some content words included in the input sentence or the converted words are extracted as important words, and are determined by the meaning of the important words and the sentence estimated from the input sentence. By generating a standard or simplified expression sentence composed of a combination with an expression, a conversion rule or a sentence generation rule can be made compact, and expression conversion can be performed by simple processing. In addition, even if an incorrect sentence is input in a part other than the important words, it can be converted to an expression that can understand the meaning correctly, and it is possible to solve the problem that the expression conversion result is output as it is incorrect as in the past Becomes

Further, according to the present invention, an important word is extracted from an input sentence, a sentence meaning is estimated from a co-occurrence relationship or a co-occurrence relationship of the extracted important word, and the important word or its translated word is estimated. By generating standard or simple expressions from combinations of only words that are predetermined from the intention, conversion rules and sentence generation rules can be made compact, and expression conversion can be performed with simple processing To In addition, even if an incorrect sentence is input in a part other than the important words, it can be converted to an expression that can understand the meaning correctly, and it is possible to solve the problem that the expression conversion result is output as it is incorrect as in the past Becomes

Further, according to the present invention, a standard or simplified expression pattern to be output is associated in advance with an important word group applicable to the pattern and a co-occurrence relationship or a co-occurrence relationship between the important word groups. Create an example sentence database, extract important words from the input sentence, select an expression pattern having the most similar keyword group from the extracted important words from the example sentence database, and select By outputting an example sentence, in addition to the effect of the present invention, it is possible to accurately and accurately perform expression conversion on the type, domain, and sentence type of a sentence that is actually input.

Further, according to the present invention, the effect of the present invention can be further enhanced by the fact that the expression pattern described in the example sentence database is composed only of important words or bilingual words of important words. It is possible.

Further, according to the present invention, the description of the expression pattern, the important word group, and the co-occurrence relation between the important word groups or the important words of the co-occurrence relation described in the example sentence database is made to the word class including the important word. In addition to the effects of the present invention described above, even if an important word that is not included in the example sentence database is input, a valid example sentence can be selected, and expression conversion that can respond to a wider variety of input sentences Becomes possible.

According to the present invention, a group of important words is extracted from the input sentence, a word that is an input error is estimated from the relation between the extracted important words, and the remaining important words excluding the estimated word are removed. By estimating the sentence meaning from and generating a standard or simplified expression from a combination of words determined by the estimated sentence meaning, in addition to the effects described in the present invention, even if the important word is incorrect, Depending on the severity of the error, it is possible to convert the expression to an expression that can understand the meaning correctly, or to notify the user that the meaning was not understood, and to output the expression conversion result as it was in the past as it is erroneously. Can be solved.

[0045]

Embodiments of the present invention will be described below with reference to the drawings.

(First Embodiment) First, a first embodiment will be described. In the first embodiment, a source language sentence input by voice (hereinafter, a sentence to be expressed is referred to as an original language sentence) is converted into a target language sentence, which is another language (hereinafter, a sentence of an expression-converted sentence). This is referred to as a target language sentence).

FIG. 1 shows the configuration of an interpreter according to an embodiment of the present invention.

The interpreter of this embodiment comprises a corpus 1 with a tag, a dependency analysis unit 2, an example DB 3, a speech recognition unit 4, an important word extraction unit 5, an example sentence selection unit 7, an output sentence generation unit 8, It is composed of a bilingual dictionary 6.

The tagged corpus 1 has an intention tag added to each translated sentence of the bilingual corpus.

The dependency analysis means 2 is a means for creating an example DB 3 by analyzing the co-occurrence relation of important words in each parallel translation of the corpus 1 with a tag.

The example DB 3 is a database that stores example sentences in which an important word pair indicating a co-occurrence relationship of important words in a source language sentence is associated with a target language sentence.

The speech recognition means 4 is means for recognizing speech input as a source language sentence and outputting word string candidates.

The important word extracting means 5 is a means for inputting a word string candidate output from the speech recognition means 4 and extracting a predetermined important word from the word string candidate.

The example sentence selection means 7 compares an important word pair created by combining important words in the input sentence with an important word pair in each example sentence of the example DB 3, and the input sentence contains the most important word pairs. This is a means for selecting and outputting the used example sentence as an example sentence having the same sentence as the input sentence.

The output sentence generating means 8 is a means for correcting the grammatical unnaturalness of the output example sentence and outputting it as a target language sentence, for example, as text or voice.

The important word bilingual dictionary 6 stores important words in a source language (hereinafter, a language of a sentence to be expressed is referred to as an original language) and a target language of the important word (hereinafter, a language of a sentence subjected to the expression conversion). (Called a language).

Next, the operation of this embodiment will be described.

The interpreting apparatus of this embodiment determines the correspondence between the contents of important words, the group of important words, and the target language expression pattern before interpreting in advance, and translates the important words in the target language into an important word bilingual dictionary. 6 and the correspondence between the important word group and the expression pattern of the target language needs to be described in the example DB3.

For this purpose, first, the developer may manually determine an important word representing an intention and an expression pattern using the same for each sentence meaning, or may specify an intention tag for each parallel sentence in the bilingual corpus. We categorize the bilingual corpus for each intention, determine the words that are commonly used in each sentence as important word candidates, and determine the important words and expression patterns semi-automatically by checking them by the developer. Good. The above sentence means one or a plurality of different sentences expressing the same intention combined into one. The bilingual corpus is a database of sentence examples, and stores a large number of bilingual sentences in which sentences in the source language are associated with sentences in the target language.

An important word bilingual dictionary and a conversion example DB are created from the important words and expression patterns determined by any of the above methods. FIG. 2A shows an important case where the source language is Japanese and the target language is English, that is, when the interpreter of the present embodiment translates Japanese speech into English speech. Example of word translation dictionary 6 and example DB3
Here is an example.

In the example of the important word bilingual dictionary 6 shown in FIG. 2A, the English “cof” is translated as the Japanese “coffee”.
"feel" is described, and "milk" in English is described as a parallel translation of "milk" in Japanese. In addition, since Japanese "requests" cannot be expressed as a single word in English, "*" is described in the English part as a translation of Japanese "requests".

In the example of the example DB3 shown in FIG. 2A, one important word group “coffee” and “request”
The expression pattern in the target language is "I'd like
e to coffee please. "Indicates that they correspond. In addition, one of the important word groups “cold”, “milk” and “is” has an expression pattern “Do you have a” in the target language.
cold milk? "Indicates that they correspond.

Further, in the example of the example DB3 in FIG. 2A, important words such as (coffee → milk) are described in pairs in each important word group. Thus, each important word is always described in the example DB3 in pairs. Such an important word pair represents a co-occurrence relationship of important words, and is created by the dependency analysis unit 2 as described below.

That is, first, the dependency analysis means 2 performs dependency structure analysis on the source language sentence of the tagged corpus 1 to clarify the dependency structure for each phrase. If there is an important word pair that is co-occurring in this dependency structure relationship, the information is added to the corresponding important word and expression pattern pair in the example DB3. Specifically, the key word “coffee” is added to the source language sentence of the bilingual corpus “coffee request”.
Since the request and the "request" are in a co-occurrence relationship, a co-occurrence relationship such as "(coffee → request)" is added as shown in FIG.

In this way, an important word bilingual word dictionary 6 and an example DB 3 as shown in FIG. 2A are created from the tagged corpus 1, and the co-occurrence of each important word is stored in the example DB 3. Relationships are added.

Next, an explanation will be given of the operation when interpreting using the example DB 3 and the important word bilingual dictionary 6 created in advance as described above.

At the time of interpreting, first, the voice recognition means 4
The speech recognition of the input source language speech is performed, and a word string candidate as a recognition result is output. Then, the recognition result is input to the important word extracting means 5. For example, if the input source language speech is "Do you have cold coffee?", The recognition result sentence "Do you have cold coffee?" Is output to the key word extraction means as a result of the speech recognition.

Next, the important word extracting means 5 extracts a predetermined important word from the recognition result sentence, and outputs the extracted important word to the example sentence selecting means 7. For example, for a recognition result sentence "Do you have cold coffee?"
And three important words, “coffee” and “there”.

Then, the example sentence selection unit 7 combines the important words in the input sentence output from the important word extraction unit 5 to create a pair of important words in the input sentence. Then, the key word pair in the input sentence is compared with the key word pair in each example sentence of the example DB3, and the example sentence containing the most important word pairs in the input sentence is determined as the input sentence and the sentence meaning. Are selected as similar example sentences and output.

For example, the three important words “cold”, “coffee”, and “present” in the input sentence are combined, and “(cold → coffee)”, “(coffee → present)”, “(cold”) → Yes) "are created.

Then, a comparison is made with the key word pair of the first example sentence of the example DB3. Since the key word pair in this example sentence is “(coffee → request)”, no key word pair in the input sentence is included.

Next, a comparison is made with the key word pair of the second example sentence in the example DB3. There are two key word pairs such as “(cold → milk)” and “milk → present” in this example sentence. The key word pair in the input sentence does not include “(cold → present)” but includes “(cold → milk)” and “(milk → present)”. Therefore, the second example sentence includes two pairs of important words in the input sentence.

In this way, the pair of important words in the input sentence
It is assumed that the second example sentence is the example sentence containing the most important word pairs of the input sentence as a result of comparison with the important word pairs of all the example sentences in the example DB3. Then, the second example sentence is selected and output, assuming that it is the example sentence most similar in sentence to the input sentence. That is, the example sentence selection unit 7 outputs the expression pattern of the target language “Do you”.
Have a cold milk? Is output.

Although FIG. 1 shows the output sentence generating means 8, the output sentence generating means 8 does not need to be provided, and the target language expression pattern output from the above-mentioned example sentence selecting means 7. Can be output to the outside as it is to function as an interpreter.

As described above, in the present embodiment, the standard or simplified expression pattern of the target language sentence to be output is associated in advance with the key word group corresponding to the pattern and the co-occurrence relation between the key words. An example DB3 is created, an important word is extracted from the input sentence, and an expression pattern having an important word group most similar to the extracted important word group is selected from the example DB3 and output. Therefore, if the output sentence generation means 8 is provided, the output sentence generation means 8 uses conversion rules and sentence generation rules to optimize third-person, plural, and past tense pronouns, verbs, and auxiliary verbs. , Even when correcting grammatical unnaturalness such as interpolation of zero pronouns or optimization of the whole sentence structure, the conversion rules and sentence generation rules can be configured compactly and simplified processing Can perform high-speed expression conversion.

Further, even when an erroneous sentence is input in a part other than an important word, it can be converted into an expression that can understand the meaning of the sentence correctly, and the problem that the expression conversion result is output without being mistaken as in the related art can be solved. It becomes possible. For example, even if a sentence containing the wrong part other than the key words, such as "Utto" or "No", such as "Do you have cold milk um?" Or "Do you have cold milk?" Do you have a cold mil
k? ] Can be converted into an expression that can understand the meaning correctly.

In this embodiment, the case where only one example sentence is selected by the example sentence selecting means 3 has been described. However, a pair of an extracted important word and a pair of an important word of an example sentence in the example DB 3 are used. As a result of the comparison, when the similarities of a plurality of example sentences are equal, the plurality of example sentences may be output.

If the expression pattern described in the example DB3 created in advance is composed of only important words as shown in FIG. 3, it is necessary to provide the output sentence generation means 8 or not. Alternatively, when the output sentence generating means 8 is provided and the conversion rule and the sentence generation rule are used, the conversion rule and the sentence generation rule can be made more compact, and a highly effective interpreter can be realized. Becomes possible.

(Second Embodiment) Next, a second embodiment will be described.

FIG. 4 shows the configuration of an interpreter according to an embodiment of the present invention.

The example DB 11 is different from the example DB of the first embodiment in that important words are classified into classes and replaced with meaning codes representing classes.

The classification word meaning table 12 is a list in which semantic codes indicating which classes each important word belongs to are assigned to each important word.

The word classifying means 13 further converts the important words in the example DB 11, the pair of important words forming a co-occurrence relationship, and the important words included in the conversion rule into the meaning codes according to the meaning codes of the classification word table 12. This is a means of classifying the important words by replacing each important word with a class, and replacing the important word extracted by the important word extracting means 5 with the meaning code according to the meaning code of the classification word table.

The example sentence selecting means 14 compares the important word class in the input sentence with the important word class pair in the example sentence, and determines the example sentence containing the most important word class pair in the input sentence.
This is a means of selecting an example sentence that is most similar in sentence to the sentence sentence.

The output generating means 15 is means for returning the important word class of the selected example sentence to the inputted important word included in the same class and outputting the same.

The other parts are the same as in the first embodiment.

Next, the operation of this embodiment will be described.

As in the case of the first embodiment, the interpreting apparatus of this embodiment determines the correspondence between the contents of important words, important word groups, and target language expression patterns before interpreting them in advance. This is described in the bilingual dictionary 16 dictionary and the example DB 11. Then, the dependency analysis unit 2 also adds an important word pair that is co-occurring in the dependency structure relationship to the expression pattern pair.

Further, the word classifying means 13 converts important words in the example DB 11, key word pairs forming a co-occurrence relationship, and important words included in the conversion rule into meaning codes in accordance with the meaning codes of the classified word table 12. Replace each key word and classify it.

FIG. 5A shows an example of the classified word table 12, and FIG. 5B shows an example of the example DB 11.

In the classification word table 12 shown in FIG.
The meaning code of a word indicating a drink such as "coffee" is 10
It is 0, and the meaning code indicating the state of the drink such as “cold” is 200. As described above, each important word in the categorized word table 12 is classified into classes by being given a meaning code.

Further, in the example of the example DB 11 shown in FIG. 5B, the respective meaning codes assigned to the respective important words in the classification word table 12 are used to explain the first embodiment shown in FIG.
Important words that appear in the example of the example DB3 are represented by semantic codes.

Next, a description will be given of an operation when interpreting is performed using the example DB 11 and the important word translation dictionary 16 created in advance as described above.

At the time of interpreting, first, the source language voice is input to the voice recognition means 4, and the voice recognition means 4 performs voice recognition of the input source language voice, and outputs a word string candidate as a recognition result. I do. Then, the recognition result is input to the important word extracting means 5.

Next, the important word extracting means 5 extracts a predetermined important word from the recognition result sentence. Then, the word classifying means 13 replaces the extracted important words with the meaning codes according to the meaning codes of the classification word table 12,
Make important words into word classes.

Next, the example sentence selecting unit 14 combines the classes of the important words in the input sentence that has been word-classified by the word classifying unit 13 to create a pair of important word classes in the input sentence. Then, a pair of key word classes in the input sentence,
The example sentence in which the input sentence includes the most important word class pairs is compared with the class pair of the important word in each example sentence in the example DB11. And select it.

The output sentence generating means 15 is used for the example sentence generating means 1.
The key word class of the example sentence selected by 4 is returned to the input key word included in the same class, and is replaced with the input key word translation using the key word bilingual dictionary 16 and output.

Further, in this embodiment, the case where only one example sentence is selected by the example sentence selection means 21 has been described. However, the pair of the extracted important word class and the class of the example sentence class in the example DB 3 are used. If a plurality of example sentences have the same similarity as a result of the comparison with the pair, a plurality of example sentences may be output.

As described above, according to the present embodiment,
Example DB1 that associates a standard or simple expression pattern of an output target language sentence with an important word class group corresponding to the pattern and a co-occurrence relationship between the important word class groups.
1, an important word is extracted from the input sentence, the extracted important word is classified into classes according to the classification word table 12, and an expression having an important word class group most similar to the classified important word group is prepared. By selecting a pattern from the example sentence DB 11 and outputting the classified important words back to the original important words, the conversion rules and the sentence generation rules are compactly configured as in the first embodiment. And expression conversion can be performed by simple processing.

Further, even when an incorrect sentence is input in a part other than the important word, the expression can be converted into an expression whose meaning can be understood correctly, and the problem that the expression conversion result is output as it is incorrect as in the prior art is solved. It becomes possible.

Further, by making word classes, it is possible to output a correct translation result even for a new input sentence that is not in an example sentence, so that a large number of input sentences can be handled with a small example sentence database. An interpreter can be realized.

In the present embodiment, before interpreting, the correspondence between the contents of the important words, the group of important words, and the target language expression pattern is determined, and the important word bilingual dictionary 16 dictionary and the example DB 11
And the key word pairs that are co-occurring in the dependency structure relationship are similarly added to the expression pattern pair.
Is the example DB1 according to the semantic code of the classification word table 12.
It has been described that the important words in No. 1, the important word pairs forming a co-occurrence relationship, and the important words included in the conversion rule are replaced with semantic codes to classify each important word, but the present invention is not limited to this. First of all, for each sentence in the tagged corpus 1, important words are replaced with semantic codes from the sentence using the classification word table 12 before interpreting in advance, and a predetermined number of combinations of the semantic codes are used. A combination of semantic codes having a co-occurrence relationship may be specified from among them, and the specified combination of the semantic codes may be associated with the converted expression of the sentence from which the semantic code is selected in advance. Absent.

(Third Embodiment) Next, a third embodiment will be described. In the third embodiment, an explanation will be given of an interpreter as an example, as in the first and second embodiments.

FIG. 6 shows the configuration of the interpreter of this embodiment.

The example sentence selecting means 21 has the function of the example sentence selecting means 7 of the first embodiment, and further has an example DB3.
Key word pairs and the extracted key word groups are compared, and among the extracted key words, key words that do not become key word pairs due to co-occurrence with any of the other key words are estimated as misrecognized words. This is a means for selecting an example sentence.

The other parts are the same as in the first embodiment.

Next, the operation of the present embodiment will be described.

The interpreter according to the present embodiment determines the correspondence between the contents of important words, important word groups, and target language expression patterns before interpreting in the same manner as in the first embodiment. The word bilingual dictionary 6 is described in the example DB3, and the co-occurrence relation of important words is similarly added to the example DB3.

Next, the operation at the time of interpretation will be described.

At the time of interpreting, first, the source language speech is input to the speech recognition means 4, and the speech recognition means 4 performs speech recognition of the source language speech, and outputs a word string candidate as a recognition result.

The important word extraction means 5 receives the recognition result output from the speech recognition means 4 and extracts a predetermined important word from the recognition result sentence.

Next, the example sentence selecting means 21 uses the example sentence DB 3
Key word pairs and the extracted key word groups are compared, and among the extracted key words, key words that do not become key word pairs due to co-occurrence with any of the other key words are estimated as misrecognized words. I do.

Then, the example sentence selecting means 21 uses the remaining important words excluding the word regarded as misrecognized,
As in the first embodiment, an example sentence whose intention is similar to the input sentence is selected and output.

FIG. 7 shows an example. This is an example in which the voice recognition unit 4 erroneously recognizes “blue milk” even though “hot milk” is input to the voice recognition unit 4. In other words, voice recognition means 4 says, "Do you have hot milk?"
Is input, but the recognition result sentence "Do you have blue milk" is output because the voice recognition means 4 has erroneously recognized.

In the example of the example DB3 in FIG. 7, one important word group of “coffee” and “request” is a pair of important words “(coffee → request)”, and the expression pattern in the target language is used. "Coffee plea
se. "Indicates that they correspond. In addition, one of the important words "hot" and "milk" and "Yes"
Has a pair of important words “hot → milk” and “milk → present”, and is an expression pattern “D
o you have a hot milk? "Indicates that they correspond.

If the speech recognition means 4 makes an erroneous recognition as described above, the key word extraction means 5 inputs the recognition result sentence "Do you have blue milk", and as key words "blue", "milk"? , And “exist” are extracted.

In such a case, the example sentence selecting means 21
"Blue", "milk",
Combining the three key words "Yes", create a pair of three key words "(blue → milk)", "(blue → high)", and "(milk → high)", and use these in the example DB3. Compare with important words.

That is, first, “(coffee → coffee → keyword)” is a key word pair of the first example sentence described in the example DB3.
Please) and three key word pairs combined. As a result, there is no key word pair that matches “(coffee → request)” which is the key word pair of the first example sentence.

Next, three important words combined with “(hot → milk)” and “(milk → present)” which are important word pairs of the second example sentence described in the example DB3. Compare with pair. As a result, “(blue → milk)” and “(blue → present)” among the key word pairs combined
Does not match any of the key word pairs in the second example sentence, but “(milk → present)” matches the key word pair in the second example sentence.

That is, the key word pair of the second example sentence is more significant than the key word pair of the first example sentence of the example DB3.
High similarity with key word pairs combined. Therefore, the second example sentence is selected.

Further, an important word belonging to the pair of important words, which is inconsistent with the pair of important words in the pair of important words in the second example sentence, matches with the pair of important words combined. For the important word that is not included in the important word pair, the second example sentence is output after removing the converted part of the important error from the second example sentence. That is,
Since "hot" is such an important word, by referring to the important word bilingual dictionary 6, "Any hot mil"
k? "H" which is the expression of the conversion word corresponding to "hot"
ot ”is removed and a sentence“ Any milk? ”is output.

Thus, “(blue → milk)”,
When the three key word pairs of “(blue → present)” and “(milk → present)” are compared with the co-occurrence relationship of the example DB3, that is, the important word pair, they have a co-occurrence relationship with “hot”. Since the word does not exist in the recognition result, it is presumed that "hot" is erroneously recognized, and the sentence "Any milk?" Is output as described above.

Further, in the present embodiment, the case where only one example sentence is selected by the example sentence selection means 21 has been described. However, a pair of an extracted important word and a pair of an important word of an example sentence in the example DB 3 are used. As a result of the comparison, when the similarities of a plurality of example sentences are equal, the plurality of example sentences may be output.

The function of the example sentence selecting means 21 in the present embodiment is not limited to the above, and the function of the example sentence selecting means 21 can be replaced with a two-stage function of an erroneous recognition selecting means and an example sentence selecting means. You can divide them. In this case, the erroneously recognized word estimating means uses the important word group extracted from the recognition result and the example DB.
3 is a means for estimating an erroneously recognized word by comparing with a pair of important words described in Example 3. The example sentence selecting means has the same function as the example sentence selecting means 7 of the first embodiment. Means.

In this case, at the time of interpretation, the misrecognized word estimating means 21 compares the created important word pair with the important word pair of each example sentence in the example DB3. Then, an important word that is not an important word due to a co-occurrence relationship with any of the other important words in the extracted important words is estimated as a misrecognized word. Next, the example sentence selecting means uses the remaining important words excluding the words deemed to be misrecognized by the misrecognized word estimating means, and determines the input sentence and the intention most like the first embodiment. Select and output similar example sentences. With this configuration, as described above, the recognition result of the voice recognition unit 4 includes the erroneous recognition, and even when the voice recognition unit 4 outputs the recognition result of “Are there blue milk”, the erroneous recognition is performed. By being processed by the word estimating means and the example sentence selecting means, "Anymilk?" Can be output as in the case described above.

In each of the first to third embodiments, when the source language sentence is a Japanese sentence and the target language sentence is an English sentence, that is, the interpreter converts the Japanese sentence into an English sentence. Has been described, but the source language sentence and the target language sentence may be sentences in other languages.

The following describes differences between the above-described embodiment in the case where the interpreter translates an English sentence into a Japanese sentence and the case where the interpreter translates a Chinese sentence into a Japanese sentence. I will explain mainly.

First, when the interpreter translates an English sentence into a Japanese sentence, the translator translates the sentence shown in FIG.
And FIG. 7 can be replaced with FIGS. 10, 11, and 12, respectively.

That is, FIG. 10A is an example of the important word bilingual word dictionary 6 and the example DB 3 in this case, and FIG. 10B is an example of the tagged corpus 1 in this case.

FIG. 11A shows an example of the classification word table 12 in this case, and FIG. 11B shows an example of the example DB 11 in this case.

FIG. 12 shows an example of the example DB3 in this case.

As described above, by replacing FIGS. 2, 5 and 7 with FIGS. 10, 11 and 12, respectively, the above-mentioned embodiments are able to replace English sentences with Japanese sentences. It is clear that the present invention can also be applied to the case of translating into.

When the interpreter translates a Chinese sentence into a Japanese sentence, FIGS. 2, 5, and 7 described in the above embodiment show FIGS. 13, 14, and 15, respectively.
Can be replaced with

That is, FIG. 13A shows an example of the important word bilingual word dictionary 6 and the example DB 3 in this case, and FIG. 13B shows an example of the tagged corpus 1 in this case.

FIG. 14 (a) shows an example of the classification word table 12 in this case, and FIG. 14 (b) shows an example of the example DB 11 in this case.

FIG. 15 shows an example of the example DB3 in this case.

As described above, by replacing FIG. 2, FIG. 5, and FIG. 7 with FIG. 13, FIG. 14, and FIG. Obviously, it can also be applied to translating sentences.

Furthermore, in the present embodiment, the dependency analysis means 2 has been described assuming that the key words having a co-occurrence relation are a pair of important words, but the present invention is not limited to this.
May be used as a pair of important words that have a dependency.

Note that the corpus 1 with a tag according to the present embodiment 1
Is an example of a corpus of the present invention, an important word pair of the present embodiment is an example of a combination of important words of the present invention, and the dependency analysis means 2 of the present embodiment is It is an example, the important word extracting means 5 and the example sentence selecting means 7 of the present embodiment are examples of the converting means of the present invention, and the classified word table 12 of the present embodiment is an example of the classifying information of the present invention. Yes,
The dependency analysis means 2 and the word classifying means 13 of the present embodiment are examples of the association means of the present invention, and the important word extracting means 5 and the example sentence selecting means 14 of the present embodiment are conversion means of the present invention. Is an example of the key word extracting means 5 of the present embodiment.
The example sentence selection unit 21 is an example of the conversion unit of the present invention, and the meaning code of the present embodiment is an example of the class name of the present invention.

Further, the predetermined number of the present invention is not limited to two in the present embodiment. The case where the predetermined number is 1 may be used. In this case, instead of combining important words and comparing them with a pair of important words, the comparison is made using only the important words. Further, the predetermined number may be three. In this case, the comparison is performed using a combination of three important words. Further, the predetermined number may be two or three. In this case, the comparison is performed using both the pair of important words and the combination of three important words. In short, the predetermined number of the present invention may be any positive integer or a plurality of different positive integers.

Furthermore, the expression conversion device of the present invention is not limited to the interpreter in the present embodiment, but converts input text into a different language such as inputting Japanese text, converting it into English text, and outputting it. A translation device that converts and outputs
In short, the expression conversion device of the present invention, such as a sentence pattern conversion device that converts a written word into a spoken word or a summary sentence creation device that summarizes and outputs a complicated or redundant sentence, is even a device that converts an input sentence into a different expression. do it.

Furthermore, the expression conversion method of the present invention is not limited to the method of translating according to the present embodiment, but may be applied to a different language such as inputting Japanese text, converting it into English text, and outputting it. In short, the expression conversion method of the present invention, such as translation for converting and outputting text, sentence conversion for converting written words to spoken words, and summarizing sentences for summarizing and outputting complex or redundant sentences, use different expressions for input sentences. It just needs to be a method of converting to.

As described above, an example sentence database is created in which the standard or simplified expression patterns of the output target language sentence are associated in advance with the key word groups corresponding to the patterns and the co-occurrence relationships between the key word groups. In addition, conversion rules and sentence generation rules can be compacted by selecting important words from the input sentence and selecting and outputting from the example sentence database an expression pattern having an important word group most similar to the extracted important word group. This makes it possible to perform expression conversion by simple processing.

Further, even when an incorrect sentence is input in a part other than the important word, the expression can be converted into an expression whose meaning can be understood correctly, and the problem that the expression conversion result is output without being mistaken as in the related art is solved. It becomes possible.

Further, even if an important word is incorrect, it is possible to convert the meaning into an expression that can understand the meaning correctly or to notify the user that the meaning was not understood, depending on the degree of the error. Thus, it is possible to solve the problem that the expression conversion result is output without being mistaken as in the related art.

Further, the present invention is a program for causing a computer to execute the functions of all or a part of the above-described expression conversion device of the present invention (or the device, element, circuit, unit, etc.), A program that operates in cooperation with a computer.

Furthermore, the present invention provides all or a part of the steps (or steps,
(Operation, operation, etc.) by a computer, and is a program that operates in cooperation with the computer.

Some of the means (or devices, elements, circuits, units, etc.) of the present invention, and some of the steps (or steps, operations, functions, etc.) of the present invention include a plurality of these means. Or means of several means or steps of a step, or means of one means or steps
It means some functions or some operations.

Further, the present invention also includes a computer-readable recording medium on which the program of the present invention is recorded.

[0150] An embodiment of the program of the present invention may be a mode in which the program is recorded on a computer-readable recording medium and operates in cooperation with the computer.

[0151] One usage of the program of the present invention may be a mode in which the program is transmitted through a transmission medium, read by a computer, and operates in cooperation with the computer.

The recording medium includes a ROM and the like, and the transmission medium includes a transmission medium such as the Internet,
Light, radio waves, sound waves, etc. are included.

The computer of the present invention described above
It is not limited to pure hardware such as a CPU, but may include firmware, an OS, and peripheral devices.

As described above, the configuration of the present invention may be realized by software or hardware.

As described in detail above, according to the present embodiment, an important word is extracted from an input sentence, and is converted into a standard or simplified expression sentence expressing the same sentence using the extracted important word. By outputting, the conversion rules and the sentence generation rules can be made compact, and the expression conversion can be performed by simple processing. In addition, even if an incorrect sentence is input in a part other than the important words, it can be converted to an expression that can understand the meaning correctly, and it is possible to solve the problem that the expression conversion result is output as it is incorrect as in the past Becomes

Also, the words of some content words included in the input sentence or the converted words are extracted as important words,
By generating a standard or simplified expression sentence consisting of a combination of this important word and an expression determined by the sentence meaning estimated from the input sentence, it is possible to configure conversion rules and sentence generation rules in a compact manner. Enables representation conversion in processing.

Further, even when an erroneous sentence is input in a part other than an important word, it can be converted into an expression whose meaning can be understood correctly, and the problem that the expression conversion result is output without being erroneous as in the related art can be solved. It becomes possible.

Further, an important word is extracted from the input sentence, a sentence meaning is estimated from the co-occurrence dependency or co-occurrence of the extracted important word, and a key word or its bilingual word is determined in advance. By generating a standard or simple expression from a combination of only words that are present, conversion rules and sentence generation rules can be made compact, and expression conversion can be performed by simple processing.

Further, even when an incorrect sentence is input in a part other than the important word, the expression can be converted into an expression whose meaning can be understood correctly, and the problem that the expression conversion result is output without being mistaken as in the related art is solved. It becomes possible.

In addition, an example DB is created in advance, in which standard or simplified expression patterns to be output are associated with important word groups corresponding to the patterns and the dependency relationships or co-occurrence relationships between the important word groups. By extracting an important word group from an input sentence, selecting an expression pattern having an important word group most similar to the extracted important word group from the example DB, and outputting the selected example sentence, the above effect is obtained. In addition to this, it is possible to accurately and accurately perform expression conversion on the type, domain, and sentence type of a sentence that is actually input.

Further, since the expression pattern described in the example DB is composed of only important words or bilingual words of important words, it is possible to further strengthen the expression patterns.

Further, the description of the expression patterns, important words, and the dependencies or the co-occurrence relationships between the important words in the example DB can be described as the word class including the important words. In addition to the effects, even if an important word not included in the example sentence DB is input, a valid example sentence can be selected,
Expression conversion that can respond to a wider variety of input sentences becomes possible.

Further, a group of important words is extracted from the input sentence, and a word that is an input error is estimated from the relation between the extracted important words.
In addition to the above effects, by estimating sentence meaning from the remaining important words excluding the estimated word and generating a standard or simplified expression from a combination of words determined by the estimated meaning, Is incorrect, depending on the severity of the error, convert it to an expression that understands the meaning correctly,
It is possible to notify the user that the meaning cannot be understood, and it is possible to solve the problem that the expression conversion result is output without being mistaken as in the related art.

[0164]

According to the present invention, it is possible to provide an expression conversion method, an expression conversion device, and a program which have a compact configuration and can be processed at high speed.

Further, the present invention provides a method of expression conversion that can output a result that correctly conveys an intention without adversely affecting the quality of an output sentence even if a recognition error occurs in an input sentence other than an important word due to an error in speech recognition or the like. , An expression conversion device, and a program.

Also, the present invention provides a method for converting expressions that can avoid outputting a result in which the sentence is not conveyed at all, even if a recognition error occurs in an input sentence due to an error in speech recognition or the like. An expression conversion device and a program can be provided.

[Brief description of the drawings]

FIG. 1 is a diagram showing an interpreter according to a first embodiment of the present invention;

FIG. 2A is a diagram showing an example of an important word bilingual dictionary and an example DB used in the first embodiment of the present invention. FIG. 2B is a corpus with a tag used in the first embodiment of the present invention. Figure showing an example of

FIG. 3 is an example DB used in the first embodiment of the present invention.
Figure showing

FIG. 4 is a diagram showing an interpreter according to a second embodiment of the present invention;

FIG. 5A is a diagram showing a classification word table used in the second embodiment of the present invention. FIG. 5B is a diagram showing an example DB used in the second embodiment of the present invention.

FIG. 6 is a diagram showing an interpreter according to a third embodiment of the present invention.

FIG. 7 is an example DB used in the third embodiment of the present invention.
Figure showing

FIG. 8 is a diagram showing a configuration of a conventional interpreting apparatus.

9A is a diagram showing an example of a conventional bilingual utterance sentence example. FIG. 9B is a diagram showing an example of a conventional bilingual phrase. FIG. 9C is a diagram showing an example of a conventional bilingual phrase dictionary. (E) Diagram showing an example of a conventional classification word table

10A is a diagram showing an example of an important word bilingual dictionary and an example DB used when translating an English sentence into a Japanese sentence according to the first embodiment of the present invention; FIG. Of an example of a corpus with a tag used when translating an English sentence into a Japanese sentence in the first embodiment of the present invention.

FIG. 11 (a) is a diagram showing a classification word table used when translating an English sentence into a Japanese sentence in the second embodiment of the present invention. (B) A second embodiment of the present invention. Figure showing an example DB used when translating English sentences into Japanese sentences

FIG. 12 is a diagram showing an example DB used when translating an English sentence into a Japanese sentence according to the third embodiment of the present invention.

13A illustrates an example of an important word bilingual dictionary and an example DB used when translating a Chinese sentence into a Japanese sentence in the first embodiment of the present invention. FIG. FIG. 6 is a diagram showing an example of a tagged corpus used when translating a Chinese sentence into a Japanese sentence in the first embodiment of the invention;

FIG. 14 (a) is a diagram showing a classification word table used when translating a Chinese sentence into a Japanese sentence according to the second embodiment of the present invention. (B) A second embodiment of the present invention. Diagram showing an example DB used when translating a Chinese sentence into a Japanese sentence in the form

FIG. 15 is a diagram showing an example DB used when translating a Chinese sentence into a Japanese sentence according to the third embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Corpus with a tag 2 Dependency analysis means 3,11 Example database 4 Speech recognition means 5 Key word extraction means 6,16 Key word bilingual dictionary 7,14 Example sentence selection means 8,15 Output sentence generation means 12 Classification word table 13 words Classification means 21 False recognition word estimation means

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 15/10 G10L 3/00 551C 15/00

Claims (13)

[Claims] For each sentence in the corpus, an important word is selected from the sentence, and a combination of important words having a co-occurrence relationship is specified from a predetermined number of combinations of the important words. The combination of the identified important word and the converted expression of the sentence from which the important word is selected are made to correspond in advance, and a predetermined important word is selected from the input sentence, and the combination is selected. The combined key word combination is compared with the previously specified key word combination of each sentence, and one or two or more sentences whose comparison result is high in matching or similarity are selected. Expression conversion method that outputs the expression of. 2. For each sentence in the corpus, an important word is selected from each sentence by using classifying information in which important words are classified into classes based on predetermined properties and class names are given in advance. Among the classes to which the important words belong, a combination of classes having a co-occurrence relationship is specified from a predetermined number of combinations, and the conversion of the specified combination of the classes and the sentence for which the important word is selected. The following expressions are made to correspond in advance, a predetermined important word is selected from the input sentence, the classes to which the respective important words belong are combined, and the combination of the combined class and the sentence An expression conversion method for comparing a combination of classes specified in advance, selecting one or more sentences having a high degree of similarity or similarity, and outputting a converted expression of the sentence. 3. For each sentence in the corpus, an important word is selected from the sentence, and a combination of important words having a co-occurrence relationship is specified from a predetermined number of combinations of the important words. The specified combination of important words and the converted expression of the sentence from which the important word is selected are made to correspond in advance, and the important words are classified into classes based on predetermined properties, and a class name is given in advance. The combination of the identified important words is made to correspond to the combination of the classes by using the classifying information, and the combination of the classes of the sentence is specified, and the important words determined in advance from the input sentence are determined. Is selected, the classes to which the respective important words belong are combined, and the combination of the combined classes is compared with the combination of the previously specified classes of the sentences, and Result match or a high degree of similarity statement 1 or 2 or more Pick, representation conversion method for outputting a representation of the converted of the statement. 4. For each sentence of the corpus, an important word is selected from the sentence, and a combination of important words having a co-occurrence relationship is specified from a predetermined number of combinations of the important words. Associating means for associating the specified combination of important words with the converted expression of the sentence from which the important word is selected, and selecting a predetermined important word from the input sentence In the combination, a combination of the combined important words and a combination of the previously specified important words of each sentence are compared, and one or more sentences having a high matching or similarity in the comparison result are selected. A conversion unit that outputs a converted expression of a sentence. 5. The conversion means, when the comparison result has a high degree of similarity, includes the mismatched one of the important words belonging to the combination of important words and the matching one of the important words belonging to the combination of important words. The expression conversion device according to claim 4, wherein, for important words that have not been converted, the converted part is removed from the selected expression and the expression is output. 6. The expression conversion apparatus according to claim 4, wherein the converted expression is composed of only important words or bilingual words of the important words. 7. For each sentence in the corpus, an important word is selected from each sentence by using classifying information in which important words are classified into classes based on predetermined properties and class names are given in advance. Among the classes to which the important words belong, a combination of classes having a co-occurrence relationship is specified from a predetermined number of combinations, and the conversion of the specified combination of the classes and the sentence for which the important word is selected. A matching means for associating a later expression with a predetermined expression, and selecting a predetermined important word from the input sentence, combining the classes to which those important words belong, and combining the combined classes, A sentence in which each sentence is compared with the combination of the classes specified in advance, and one or more sentences in which the comparison result is high in matching or similarity is selected, and a converted expression of the sentence is output. Representation converter and means. 8. For each sentence of the corpus, an important word is selected from the sentence, and a combination of important words having a co-occurrence relationship is specified from a predetermined number of combinations of the important words. The specified combination of important words and the converted expression of the sentence from which the important word is selected are made to correspond in advance, and the important words are classified into classes based on predetermined properties, and a class name is given in advance. By using the classifying information obtained, the specified combination of important words is made to correspond to the combination of the classes, whereby the association means for specifying the combination of the classes of the sentence is determined in advance from the input sentence. Selected important words, combine the classes to which the important words belong, and combine the combination of the combined classes with the combination of the previously specified classes of each sentence. And, select the comparison result is a match or a high degree of similarity statement 1 or 2 or more, representing conversion device and a converting means for outputting a representation of the converted of the statement. 9. The conversion unit, when the comparison result has a high degree of similarity, the class belonging to the unmatched class combination and not included in the matching class combination. 9. The expression conversion device according to claim 7, wherein the expression is output after removing the converted part of the class from the selected expression. 10. 10. The expression conversion device according to claim 7, wherein the converted expression is composed of only classes. 11. An expression conversion apparatus according to claim 4, wherein, for each sentence of the corpus, an important word is selected from the sentence, and a co-occurrence relation of a co-occurrence relation is selected from a predetermined number of combinations of the important words. An association means for identifying a certain combination of important words and associating the combination of the identified important words with the converted expression of the sentence from which the important words are selected in advance, A predetermined important word is selected from a sentence, the combination is compared, and the combination of the combined important word is compared with the combination of the specified important words of each sentence in advance, and the comparison result has a high degree of agreement or similarity. A program for causing a computer to function as all or a part of conversion means for selecting one or more sentences and outputting a converted expression of the sentences. 12. A sentence of each sentence in a corpus, using classifying information in which an important word is classified into classes based on predetermined properties and a class name is assigned, of the expression conversion apparatus according to claim 7. From among the classes to which each of these important words belongs, specify a combination of classes having a co-occurrence relationship from a predetermined number of combinations, and specify the combination of the specified class and the important word. Means for associating in advance the converted expression of the sentence to be selected, and pre-determined key words from the input sentence, and classifying each of the key words to the class to which the key words belong. The combination, the combination of the combined classes, and the combination of the previously specified classes of each sentence are compared, and one or more sentences whose matching result is high in matching or similarity are selected. Program for causing a computer to function as all or part of the conversion means for outputting a representation of the converted of the statement. 13. An expression conversion apparatus according to claim 8, wherein for each sentence in the corpus, an important word is selected from the sentence, and a co-occurrence relation of a co-occurrence relation is selected from a predetermined number of combinations of the important words. A certain combination of important words is specified, and the combination of the specified important word and the converted expression of the sentence from which the important word is selected are corresponded in advance, and the important word is set in advance to a predetermined property. Correlation means for specifying the combination of the sentence classes by associating the specified combination of important words with the combination of the classes by using the classifying information in which the classes are classified based on the class names. , And a predetermined important word is selected from the input sentence, the classes to which the respective important words belong are combined, and the combination of the combined class and the previous The computer is used as a whole or a part of the conversion means for comparing the specified combination of classes and selecting one or more sentences having a high degree of matching or similarity with the comparison result and outputting a converted expression of the sentences. Program to make it work.