JP2002351494A - Sentence recognition device, sentence recognition method, program, and medium - Google Patents

Abstract

(57) [Summary] [Problem] It has been difficult to perform appropriate sentence recognition using speech recognition or text sentence recognition. Kind Code: A1 A database for storing a plurality of predetermined standard content word pairs, each of which is composed of a plurality of predetermined content words, and a voice for recognizing an input sentence composed of a plurality of words. Recognition means 102
A content word selecting means 103 for selecting a content word from a plurality of words constituting a recognized sentence; and a standard content in which an arbitrary content word pair composed of the selected content words is stored in a database. Determining means 104 for determining whether or not the word pair is a word pair; misrecognized content word determining means 105 for determining, based on a result of the determination, a misrecognized content word having an incorrect recognition from among the selected content words; It is a sentence recognition device provided with.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sentence recognition apparatus, a sentence recognition method,
Program and media.

[0002]

2. Description of the Related Art The prior art will be described with reference to voice recognition means as an example.

[0003] In the speech recognition means, if the recognition processing is not perfect and an error occurs, and the error is output as an erroneous problem, it is a serious problem in practical use.

Conventionally, in order to solve this problem, when the recognition score of the first candidate of the recognition result is smaller than a certain value or less than the recognition score of the second candidate or less, the reliability of the recognition result is reduced. A method of rejecting a recognition result sentence deemed to be low or prompting re-input has been proposed.

[0005] This example is described by, for example, one-pass, n-best, which are typical search means of continuous speech recognition means.
This will be described more specifically with an example using search.

[0006] A speech DB for learning the acoustic characteristics of each phoneme in advance
, And the connection probability between words represented as a sequence of left-hand phonemes is calculated using a text DB. At the time of recognition, the acoustic feature per unit time of the input speech is analyzed, and the time series of this feature amount is compared with the acoustic feature amount of each phoneme learned in advance. An acoustic score is calculated as a certain probability.

[0007] According to the phoneme sequence of the words described in the word dictionary, the acoustic scores are added in chronological order, and the added value becomes the acoustic score at each time point. If a search space for all phoneme strings cannot be obtained, the process proceeds while leaving only the top N results with high scores.

If the input speech includes a plurality of words, the words are connected with reference to the connection probabilities that have been learned in advance, and at the time of connection, the connection probability between the words is added to the above-mentioned acoustic score. (Referred to as the language score) is added to the score.

When the difference between the scores of the first candidate and the second candidate in the recognition scores of the top N candidates calculated in this way is equal to or less than a certain value, it is considered that the reliability of the result of the first candidate is low. (For example, Jitsuhiro et al .: “Rejection by reliability measure based on likelihood difference between alliphones” IEICE technical report SP97-
76, pp. 1-7 (1997)).

[0010]

However, the above recognition score indicates the similarity between the input speech and the previously learned acoustic model or language model, and may differ depending on the input speaker and the utterance content. At present, even if correct recognition is performed, the value is greatly different. For this reason, it is very difficult to determine a score ratio threshold value for rejection, and sometimes a correct recognition result is rejected, or an incorrect recognition result is judged to be correct and output.

As described above, it has been difficult to perform appropriate sentence recognition using speech recognition and text sentence recognition.

The present invention provides a sentence recognition apparatus, a sentence recognition method, a program, and a medium capable of performing appropriate sentence recognition using speech recognition and text sentence recognition in consideration of the above conventional problems. It is intended to provide.

[0013]

Means for Solving the Problems The first invention (claim 1)
Is a database that stores a plurality of predetermined standard specific word pairs, each of which is composed of a plurality of predetermined specific words, and a sentence that recognizes an input sentence composed of a plurality of words. Recognition means, specific word selection means for selecting the specific word from among a plurality of words constituting the recognized sentence, and an arbitrary specific word pair composed of the selected specific words are stored in the database. Determining means for determining whether or not the word is a standard specific word pair that has been determined; and determining an erroneously recognized specific word in which the recognition is erroneous from the selected specific words based on a result of the determination. This is a sentence recognition device including a recognition specific word determination unit.

According to a second aspect of the present invention (corresponding to claim 2), the erroneously recognized specific word determining means includes an arbitrary specific word pair among the arbitrary specific word pairs determined to be not the standard specific word pair stored in the database. A sentence recognition apparatus according to a first aspect of the present invention for determining a specific word common to a predetermined number or more of arbitrary specific word pairs as the misrecognized specific word.

According to a third aspect of the present invention (corresponding to claim 3), when there is the misrecognized specific word, (1) re-input of the specific word corresponding to the misrecognized specific word, or (2) A sentence recognition apparatus according to the first or second aspect of the present invention, further comprising a re-input requesting unit for requesting re-input of the input sentence.

According to a fourth aspect of the present invention (corresponding to claim 4), when the erroneously recognized specific word is present, the first or second aspect of the present invention comprises a notifying means for notifying a user of the fact. It is a sentence recognition device.

A fifth aspect of the present invention (corresponding to claim 5) comprises a plurality of predetermined specific words.
A database storing a plurality of predetermined standard specific word pairs, a sentence recognizing means for recognizing an input sentence composed of a plurality of words, and a plurality of words constituting the recognized sentence. Specific word selecting means for selecting a specific word, and determining means for determining whether or not an arbitrary specific word pair composed of the selected specific words is a standard specific word pair stored in the database, A sentence recognition apparatus comprising: a sentence misrecognition determining unit that determines whether the input sentence is incorrectly recognized based on a result of the determination.

A sixth aspect of the present invention (corresponding to claim 6) is a fifth aspect of the present invention, comprising: a sentence re-entry requesting means for requesting re-entry of the inputted sentence when the erroneous recognition is made. 1 is a sentence recognition device of the invention.

A seventh aspect of the present invention (corresponding to claim 7) is the sentence recognition apparatus according to the fifth aspect of the present invention comprising a notifying means for notifying a user of the occurrence of the erroneous recognition. .

According to an eighth aspect of the present invention (corresponding to claim 8), a correspondence relationship between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong is stored. One database, a second database configured from the specific word classes, and storing a plurality of predetermined standard specific word class pairs, and a sentence recognizing an input sentence composed of a plurality of words Recognizing means, specific word selecting means for selecting the specific word from among a plurality of words constituting the recognized sentence, and using the correspondence stored in the first database, A specific word class determining means for determining the specific word class to which the specific word belongs; and an arbitrary specific word class pair comprising the determined specific word classes. Determining means for determining whether or not the pair is a standard specific word class pair stored in a second database; and, based on a result of the determination, erroneous recognition in which the recognition is incorrect from among the selected specific words. This is a sentence recognizing device including an erroneously recognized specific word determining means for determining a specific word.

According to a ninth aspect of the present invention (corresponding to claim 9), the erroneously-recognized specific word determining means includes an arbitrary specific word determined to be not a standard specific word class pair stored in the second database. The sentence recognition apparatus according to the eighth aspect of the present invention, wherein a specific word belonging to a specific word class common to a predetermined number or more of arbitrary specific word class pairs in the class pair is determined as the misrecognized specific word.

According to a tenth aspect of the present invention (corresponding to claim 10), when there is the misrecognized specific word, (1) re-input of the specific word corresponding to the misrecognized specific word, or (2) Eighth or ninth aspects of the sentence recognition apparatus of the present invention comprising re-entry request means for requesting re-entry of the input sentence.

The eleventh invention (corresponding to claim 11) provides:
The sentence recognition device according to the eighth or ninth aspect of the present invention, further comprising a notifying unit for notifying the user of the occurrence of the erroneously recognized specific word.

The twelfth invention (corresponding to claim 12) provides:
A first database that stores a correspondence relationship between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong; and a predetermined database including the specific word classes. A second database storing a plurality of standard specific word class pairs, a sentence recognizing means for recognizing an input sentence composed of a plurality of words, and a plurality of words constituting the recognized sentence. A specific word selecting means for selecting the specific word from a specific word class, and a specific word class determining means for determining the specific word class to which the selected specific word belongs by using the correspondence stored in the first database And a standard specific word in which an arbitrary specific word class pair composed of the determined specific word classes is stored in the second database. A sentence recognition apparatus comprising: a determination unit that determines whether a pair is a Lass pair; and a sentence misrecognition determination unit that determines whether the input sentence is misrecognized based on a result of the determination. is there.

A thirteenth aspect of the present invention (corresponding to claim 13) is:
A twelfth sentence recognition apparatus according to the present invention, further comprising a sentence re-input requesting means for requesting re-input of the input sentence when the erroneous recognition is made.

The fourteenth invention (corresponding to claim 14) is:
A twelfth sentence recognition apparatus according to the present invention including a notifying unit that notifies a user of the occurrence of the erroneous recognition.

The fifteenth invention (corresponding to claim 15) is:
A storage step of storing a plurality of predetermined standard specific word pairs composed of a plurality of predetermined specific words in a database, and a sentence recognition step of recognizing an input sentence composed of a plurality of words And a specific word selecting step of selecting the specific word from a plurality of words constituting the recognized sentence; and an arbitrary specific word pair composed of the selected specific words stored in the database. A determination step of determining whether or not the word is a standard specific word pair, and an erroneous recognition specification for determining an erroneously recognized specific word in which the recognition is erroneous from the selected specific words based on a result of the determination. This is a sentence recognition method including a word determination step.

A sixteenth aspect of the present invention (corresponding to claim 16) is:
A storage step of storing a plurality of predetermined standard specific word pairs composed of a plurality of predetermined specific words in a database, and a sentence recognition step of recognizing an input sentence composed of a plurality of words And a specific word selecting step of selecting the specific word from a plurality of words constituting the recognized sentence; and an arbitrary specific word pair composed of the selected specific words stored in the database. A sentence erroneous recognition determining step of determining whether or not the input sentence is erroneously recognized based on the result of the determination. It is a recognition method.

A seventeenth aspect of the present invention (corresponding to claim 17) is:
A first storage step of storing in a first database a correspondence between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong, and from the specific word classes. A second storage step of storing a plurality of predetermined standard specific word class pairs in a second database, a sentence recognition step of recognizing an input sentence composed of a plurality of words, A specific word selecting step of selecting the specific word from a plurality of words constituting the recognized sentence, and using the correspondence stored in the first database, to which the selected specific word belongs A specific word class determining step of determining the specific word class; and an arbitrary specific word class pair composed of the determined specific word classes. A determination step of determining whether or not the pair is a standard specific word class pair stored in the second database; and based on a result of the determination, the recognition is incorrect from among the selected specific words. And an erroneously recognized specific word determining step of determining an erroneously recognized specific word.

An eighteenth aspect of the present invention (corresponding to claim 18) is:
A first storage step of storing in a first database a correspondence between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong, and from the specific word classes. A second storage step of storing a plurality of predetermined standard specific word class pairs in a second database, a sentence recognition step of recognizing an input sentence composed of a plurality of words, A specific word selecting step of selecting the specific word from a plurality of words constituting the recognized sentence, and using the correspondence stored in the first database, to which the selected specific word belongs A specific word class determining step of determining the specific word class; and an arbitrary specific word class pair composed of the determined specific word classes. A determining step of determining whether or not the pair is a standard specific word class pair stored in the second database; and determining whether or not the input sentence is erroneously recognized based on a result of the determination. This is a sentence recognition method including a sentence erroneous recognition determination step.

The nineteenth invention (corresponding to claim 19) is:
In the sentence recognition method according to the fifteenth aspect of the present invention, a storage step of storing a plurality of predetermined standard specific word pairs in a database, the storage step including a plurality of predetermined specific words, and a plurality of words. Sentence recognizing step of recognizing the inputted sentence, a specific word selecting step of selecting the specific word from among a plurality of words constituting the recognized sentence, and the selected specific words. A determining step of determining whether or not the specified specific word pair is a standard specific word pair stored in the database; and, based on a result of the determination, determining that the recognition is incorrect from among the selected specific words. This is a program for causing a computer to execute all or a part of an erroneously recognized specific word determining step of determining an erroneously recognized specific word.

The twentieth invention (corresponding to claim 20) is:
In the sentence recognition method of the sixteenth aspect, a storage step of storing a plurality of predetermined standard specific word pairs in a database, the storage step including a plurality of predetermined specific words, and a plurality of words. Sentence recognizing step of recognizing the inputted sentence, a specific word selecting step of selecting the specific word from among a plurality of words constituting the recognized sentence, and the selected specific words. Determining whether the arbitrary specific word pair is a standard specific word pair stored in the database; and, based on a result of the determination, determining whether the input sentence is erroneously recognized. This is a program for causing a computer to execute all or a part of the sentence error recognition determining step to be determined.

The twenty-first invention (corresponding to claim 21)
In the sentence recognition method of the seventeenth aspect of the present invention, the correspondence between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong is stored in the first database. A first storage step for storing, a second storage step for storing a plurality of predetermined standard specific word class pairs in the second database, the second storage step being configured from the specific word classes, and a plurality of words. Sentence recognizing step of recognizing an inputted sentence, a specific word selecting step of selecting the specific word from a plurality of words constituting the recognized sentence, and a correspondence stored in the first database. A specific word class determining step of determining the specific word class to which the selected specific word belongs using a relationship; A determining step of determining whether an arbitrary specific word class pair composed of the specified specific word class pair is a standard specific word class pair stored in the second database; and, based on a result of the determination, determining the selected specific word class pair. This is a program for causing a computer to execute all or part of an erroneously recognized specific word determining step of determining an erroneously recognized specific word in which the recognition is incorrect from words.

The twenty-second invention (corresponding to claim 22)
Stores, in the first database, a correspondence relationship between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong, in the eighteenth sentence recognition method of the present invention. A first storage step for storing, a second storage step for storing a plurality of predetermined standard specific word class pairs in the second database, the second storage step being configured from the specific word classes, and a plurality of words. Sentence recognizing step of recognizing an inputted sentence, a specific word selecting step of selecting the specific word from a plurality of words constituting the recognized sentence, and a correspondence stored in the first database. A specific word class determining step of determining the specific word class to which the selected specific word belongs using a relationship; A determining step of determining whether or not an arbitrary specific word class pair composed of the standard specific word class pair stored in the second database; and, based on a result of the determination, the input sentence Is a program for causing a computer to execute all or part of a sentence misrecognition determining step of determining whether or not a misrecognition has been performed.

Twenty-third invention (corresponding to claim 23)
Is a medium that carries the nineteenth program of the present invention, and is a medium that can be processed by a computer.

The twenty-fourth invention (corresponding to claim 24)
Is a medium that carries the twentieth program of the present invention, and is a medium that can be processed by a computer.

Twenty-fifth invention (corresponding to claim 25)
Is a medium that carries the program of the twenty-first aspect of the present invention, and is a medium that can be processed by a computer.

The 26th invention (corresponding to claim 26)
Is a medium that carries the program according to the twenty-second aspect of the present invention, and is a medium that can be processed by a computer.

(1) A speech recognition means characterized in that a misrecognized word is estimated from a relationship between specific words included in a recognition result sentence, and the estimation result is reflected and output in the recognition result sentence. A result rejecting unit or a re-input requesting unit, which requests the user to re-input when all or many of the words used for estimating the erroneously recognized word are estimated to be erroneously recognized; (2) Specific words or word class pairs having a dependency relationship or a co-occurrence relationship are learned in advance, and when the same word pair is not included in the recognition result sentence at all or is small, the user is required to input again. The result rejecting means or re-input requesting means described above is also included in the invention.

Such a reject means is, for example,
Continuous speech recognition means for recognizing utterances composed of multiple words, important word extraction means for extracting specific words from recognition results, and checking dependencies or co-occurrence relations between extracted words to trust recognition results It is composed of a reliability calculating means for judging the reliability, a reject determining means for rejecting the result when the result is not reliable, and an output sentence generating means for generating a re-input request sentence when rejected.

In such a reject means, a specific word is extracted from the recognition result sentence, and a word pair having a dependency or a co-occurrence relationship is searched from the extracted word. By rejecting the recognition result, it is possible to stably reject an incorrect result even if the speaker or the utterance content changes.

The above-described result rejecting means or re-input requesting means, characterized by using a word class determined by using a general thesaurus dictionary and a relation between words included in a learning sentence set, Included in the invention.

Such a reject means is, for example,
A word class determining means for classifying important words, a relation table between word classes for describing a relation between the preceding word classes,
Continuous speech recognition means for recognizing utterances composed of multiple words, important word extraction means for extracting specific words from recognition results, and checking dependencies or co-occurrence relations between extracted words to trust recognition results It is composed of a reliability calculating means for judging the reliability, a reject determining means for rejecting the result when the result is not reliable, and an output sentence generating means for generating a re-input request sentence when rejected.

In such a reject means, words are optimally classified in advance, and the dependency or co-occurrence relationship between word classes is checked and described in a table. At the time of recognition, a specific word is extracted from the recognition result sentence, and a word table having a dependency relationship or a co-occurrence relationship is searched for from the extracted words using a relationship table describing the dependency relationship or the co-occurrence relationship.
By rejecting the recognition result when the left word pair does not exist or is small, it is possible to stably reject an erroneous result even if the speaker or the utterance content changes. Furthermore, even if a word that does not exist in the sentence set for learning the relationship between words is input during recognition, rejection or re-input request processing can be performed.

[0045]

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

(Embodiment 1) First, referring to FIG. 1 which is a block diagram showing the configuration of the re-entry requesting means according to Embodiment 1 of the present invention, The operation will be described.

Here, a case where the recognition result is rejected by the voice recognition means will be described as an example.

Before recognition, the dependency structure analysis is performed for each phrase of a corpus with a morphological analysis (hereinafter referred to as a tagged corpus) using a dependency analysis unit in advance, and the result is included in each phrase. As a dependency relationship between content words to be performed, a word pair having a dependency relationship is described in a table.
The dependency structure analysis means can be clarified by, for example, performing syntax analysis between clauses using case grammar rules.

At the time of recognition, first, the input speech is recognized, and a word string of the recognition result is output. As a recognition method at this time, for example, in a continuous speech recognition method using one-pass n-best search, it is possible to output a word string as a recognition candidate as a result. Next, only a specific word is extracted from the word string that is the output recognition result. Here, a word (referred to as an important word) necessary for understanding the intention of the sentence may be a specific word, and specifically, a content word may be a specific word. Next, erroneously recognized words are estimated by examining the relationship between the specified important words. This means, for example,
In the key words extracted from the recognition result, it is checked whether there is a pair of words having a dependency analyzed in advance, and an important word that does not form a word pair corresponding to any other key word is erroneously recognized. Consider In this means, when all the words are regarded as misrecognized words, the reliability of the recognition result is considered to be extremely low, the rejection is determined, and a message sentence prompting the speaker to re-input is generated. To display on the screen. The sentence generated in this case is "I could not hear. Please speak again." If some words are not presumed to be misrecognized, a sentence consisting only of a phrase including an important word that is not misrecognized is generated and output to the screen. In this reject decision means,
If all words were misrecognized, the rejection was determined.For example, if a certain percentage or more of the important words extracted from the recognition result were regarded as misrecognized words, It may be determined to be rejected, and the recognition result may be output if the ratio is less than a certain ratio.

In this embodiment, the dependency between words is extracted in advance, and the dependency is compared with an important word included in the recognition result to estimate a misrecognized word included in the recognition result. If all or a certain percentage or more of the recognized words are considered to be misrecognized words, the recognition result is rejected and the speaker is prompted to re-enter, so that the speaker or the content of the utterance changes. It is possible to stably reject an incorrect result and request re-entry effectively.

Therefore, misrecognized words are estimated from the dependencies or cooccurrence relationships between important words included in the recognition result sentence, and when many of the words are estimated to be misrecognized words, the user is required to re-input. By doing so, even if there is a change in the speaker or utterance content, an incorrect result can be stably rejected, and a re-input can be requested effectively.

(Embodiment 2) First, referring to FIG. 2 which is a block diagram showing the configuration of the re-entry requesting means according to the second embodiment of the present invention, The operation will be described.

Here, a case where the recognition result is rejected by the voice recognition means will be described as an example.

Before the recognition, the dependency structure analysis is performed for each phrase of the corpus with morphological analysis (hereinafter referred to as a tagged corpus) using the dependency analysis means in advance, and the result is included in each phrase. As a dependency relationship between content words to be performed, a word pair having a dependency relationship is described in a table.
The method is the same as in the previous embodiment. Next, important words are classified into classes in order to cope with a case where a word that does not exist in the corpus is input. In accordance with a commonly used thesaurus dictionary, a classifying means for classifying important words belonging to the same category in the thesaurus in the same class may be used. It would be possible to do this. First, all important words are classified into classes according to the semantic codes of the thesaurus, and this is set as an initial class. Words that do not depend on any of the words belonging to the same class and that do not depend on the same word are excluded from the word class from the previously analyzed dependent word pair, and the word alone forms one class. With respect to all the word pairs, a result of searching for a word to be excluded from the class under the above condition is determined as a word class. When the word class is determined, the previously extracted dependent word pair is expressed as a word class pair and described in a table.

At the time of recognition, continuous speech recognition and important word extraction are performed in the same manner as in the previous embodiment, and a misrecognized word is estimated by comparing with a pre-analyzed dependent word class pair.
Determine the output of rejects and re-entry requests.

In this embodiment, dependencies between words are extracted in advance, word classes are formed using the thesaurus dictionary and the preceding dependencies, and the preceding dependencies are expressed as dependencies between the word classes. It is described in. By comparing the dependency between the word classes and the important words included in the recognition result, the misrecognized words included in the recognition result are estimated, and all or a certain percentage or more of the important words in the recognition result are misrecognized words. When it is determined that the recognition result has been recognized, the recognition result is rejected, and a request for re-input is prompted to the speaker. Thereby, similar to the previous embodiment,
In addition to the effect of rejecting incorrect results stably and effectively requesting re-entry even if the speaker or utterance changes, in addition to incorrect recognition even when key words that are not in the corpus are entered, There is an effect that the above processing such as estimation and rejection determination can be performed.

Therefore, the important words are classified into classes using the thesaurus dictionary and the corpus in advance, and the misrecognized words included in the recognition result are estimated using the dependencies between the preceding important word classes. By requiring the user to re-enter when it is estimated to be a word, it is possible to stably reject erroneous results and effectively request re-entry even if the speaker or utterance changes. Even if an important word or sentence expression that is not included in the corpus is input, it is possible to perform the above processing such as misrecognition estimation and rejection determination.

In Embodiments 1 and 2 described above,
Instead of setting a reject threshold with the recognition score as an index, use the naturalness of the sentence of the recognition result sentence as an index, check the dependency or co-occurrence relationship between specific words included in the recognition result, and check that all words By rejecting when the relationship is weak, it is possible to stably reject an erroneous result and request re-entry effectively even if the speaker or utterance changes.

(Embodiment 3) First, the configuration of the speech recognition apparatus according to the present embodiment will be described with reference to FIG. 3 which is a block diagram showing the configuration of the speech recognition apparatus according to Embodiment 3 of the present invention. .

The speech recognition apparatus according to the present embodiment includes a database 101, a speech recognition means 102, a content word selection means 103, a determination means 104, an erroneously recognized content word determination means 105, and a content word re-input request means. 106.

The database 101 corresponds to the database of the present invention, the speech recognition means 102 corresponds to the sentence recognition means of the present invention, the content word selecting means 103 corresponds to the specific word selecting means of the present invention, 104 corresponds to the determining means of the present invention, the misrecognized content word determining means 105 corresponds to the misrecognized specific word determining means of the present invention, and the content word re-input request means 106 corresponds to the re-input request means of the present invention. The speech recognition device of the present embodiment corresponds to the sentence recognition device of the present invention.
Further, the content words of the present embodiment correspond to the specific words of the present invention.

Next, the configuration of the speech recognition apparatus according to the present embodiment will be described in more detail with reference to FIG. 4, which is an explanatory diagram of data stored in the database 101 according to the third embodiment of the present invention.

The database 101 is a standard content word pair having a predetermined keyword dependency, which is composed of content words learned in advance.
This is a means for storing “(hot, coffee)”, “(coffee, drink)” and the like (see FIG. 4).

The voice recognition means 102 is a means for voice recognition of a sentence composed of words and voice input.

The content word selecting means 103 looks at the content word dictionary (not shown) in which the learned content words are described, and writes the content words in the content word dictionary from the words constituting the speech-recognized sentence. This is a means for selecting the content word that is present.

The determination means 104 is a means for determining whether or not an arbitrary content word pair composed of the selected content words is a standard content word pair stored in the database 101.

The erroneously-recognized content word determining means 105 determines a content word common to two or more arbitrary content word pairs among the arbitrary content word pairs stored in the database 101 and determined not to be a standard content word pair. This is a means for determining as a misrecognized content word.

The content word re-input request means 106 is a means for requesting re-input of a content word corresponding to the erroneously recognized content word when there is an erroneously recognized content word.

Next, FIG. 5A which is an explanatory diagram (part 1) for explaining the operation of the speech recognition apparatus according to the third embodiment of the present invention, and the speech recognition apparatus according to the third embodiment of the present invention. For explaining the operation of (2)
The operation of the speech recognition apparatus according to the present embodiment will be described with reference to FIG. Note that while describing the operation of the speech recognition apparatus in the present embodiment, an embodiment of the sentence recognition method of the present invention will also be described.

The speech recognition means 102 (erroneously) recognizes the sentence "Do you drink hot coffee?" As the recognition sentence "Do you drink blue coffee?"

The content word selecting means 103 selects one of the words “blue”, “coffee”, “wo”, “drink”, “mas”, and “ka” constituting the recognition sentence “do you drink blue coffee?”. , The content words "blue", "coffee", and "drink" are selected (see FIG. 5A). Since the content word of the inflected word is selected as the end form (dictionary heading form), “drinking (continuous form)” is selected as “drinking (end form)”.

The judging means 104 selects a total of three arbitrary content word pairs “(blue, coffee)”, “(blue, drink)” and “(coffee, drink)” composed of the selected content words. "(Coffee, drink)" is Database 1
01 is a standard content word pair stored in "01 (blue,
"Coffee)" and "(blue, drink)" are database 10
It is determined that the pair is not the standard content word pair stored in No. 1 (see FIG. 5B).

The misrecognized content word determining means 105 selects an arbitrary content word pair “(blue, coffee)” or “(blue, drink)” determined to be not a standard content word pair stored in the database 101. The content word "blue" common to two or more arbitrary content word pairs is determined to be a misrecognized content word.

The content word re-input request means 106 requests re-input of the content word corresponding to the misrecognized content word "blue".

(Embodiment 4) First, the configuration of the speech recognition apparatus according to the present embodiment will be described with reference to FIG. 6 which is a block diagram showing the configuration of the speech recognition apparatus according to Embodiment 4 of the present invention. .

The speech recognition apparatus according to the present embodiment includes a database 101, a speech recognition means 102, a content word selection means 103, a judgment means 104, a sentence error recognition determination means 105 ', and a sentence re-input request means 106. '.

The database 101 corresponds to the database of the present invention, the voice recognition means 102 corresponds to the sentence recognition means of the present invention, the content word selecting means 103 corresponds to the specific word selecting means of the present invention, 104 corresponds to the determination means of the present invention, the sentence error recognition determining means 105 'corresponds to the sentence error recognition determining means of the present invention, and the sentence re-input request means 106' corresponds to the sentence re-input request means of the present invention. The speech recognition device of the present embodiment corresponds to the sentence recognition device of the present invention. Further, the content words of the present embodiment correspond to the specific words of the present invention.

The speech recognition apparatus according to the present embodiment has a configuration similar to the configuration of the speech recognition apparatus according to the third embodiment described above, but the erroneous sentence recognition determining means 105 '.
And a sentence re-input request means 106 '.

The sentence erroneous recognition determining means 105 'and the sentence re-input request means 106' will be described in more detail.

The sentence misrecognition determining means 105 'determines a content word that is not included in any arbitrary content word pair determined to be a standard content word pair stored in the database 101 as a misrecognized content word, and selects Between the number of content words X and the number of misrecognized content words Y

[0081]

## EQU1 ## If the relation of Y> 0.4 × X holds, the sentence is a means for determining that the sentence is erroneously recognized.

The sentence re-input requesting means 106 'is a means for requesting re-input of the input sentence when an erroneous recognition is made.

Next, the operation of the speech recognition apparatus according to the present embodiment will be described. Note that while describing the operation of the speech recognition apparatus in the present embodiment, an embodiment of the sentence recognition method of the present invention will also be described.

The voice recognizing means 102 (erroneously) recognizes the voice-input sentence "Do you not drink local beer?" As a recognition sentence "Call local beer?" .

The content word selecting means 103 generates the words “local”, which constitute the recognition sentence “call local beer, Ikesenka?”
"No", "beer", "wo", "call", "de",
Select the content words "local", "beer", "call", "pond", and "1,000" from "pond", "thousand", and "ka". In addition,
As in the case of the above-described third embodiment, since the content word of the inflected word is selected as the end form (dictionary heading form), “call (continuous form)” is selected as “call (end form)”.

The determination means 104 selects a total of 60 arbitrary content word pairs “(local, beer)”, “(local, call)”, “(local, pond)”,
Of “(local, thousand)”, “(local, beer)” is a pair of standard content words stored in the database 101, and other “(local, call)”, “(local, pond)” "
It is determined that “(local, thousand)”,... Are not a standard content word pair stored in the database 101.

The sentence misrecognition determining means 105 'is a content word "call" which is not included in the arbitrary content word pair "(local, beer)" determined to be a standard content word pair stored in the database 101. , “Pond”, and “thousand” are determined to be misrecognized content words. In addition, the sentence misrecognition determining means 105 ′ selects the selected content words “local”, “beer”, “call”,
The number of “ponds” and “thousands” X = 5 and the misrecognized content words “call”,
Since the relationship of (Equation 1) is established between the number of “ponds” and “1,000” Y = 3, it is determined that the input sentence “Do you not drink local beer?” Was incorrectly recognized. I do.

Then, the sentence re-input requesting means 106 'inputs the sentence "Do you not drink local beer?"
Request to re-enter.

(Embodiment 5) First, the configuration of a character recognition apparatus according to the present embodiment will be described with reference to FIG. 7 which is a block diagram showing the configuration of a character recognition apparatus according to Embodiment 5 of the present invention. .

The character recognition device according to the present embodiment includes a database 201 to 202, a character recognition unit 203,
Content word selecting means 204 and content word class determining means 205
Determining means 206 and erroneously recognized content word determining means 207
And a content word re-input requesting means 208.

The database 201 corresponds to the first database of the present invention, the database 202 corresponds to the second database of the present invention, the character recognizing means 203 corresponds to the sentence recognizing means of the present invention, The selecting means 204 corresponds to the specific word selecting means of the present invention, the content word class determining means 205 corresponds to the specific word class determining means of the present invention, and the judging means 206 corresponds to the judging means of the present invention. The word determining means 207 corresponds to the misrecognized specific word determining means of the present invention, the content word re-input request means 208 corresponds to the re-input request means of the present invention, and the character recognition device of the present embodiment is a Corresponds to the recognition device. Further, the content words of the present embodiment correspond to the specific words of the present invention, and the content word classes of the present embodiment correspond to the specific word classes of the present invention.

Next, FIG. 8A illustrating the data stored in the database 201 according to the fifth embodiment of the present invention, and the data stored in the database 202 according to the fifth embodiment of the present invention. The configuration of the character recognition device according to the present embodiment will be described in more detail with reference to FIG.

The database 201 stores “beer-100” and “beer-100” as correspondences between the content words learned in advance and the predetermined content word classes to which the content words belong.
"Coffee-100", "Local-200", "Hot-2"
00, “drink-300”, “call-400”, and the like. Here, “90” is a semantic code of the content word class for the category “building”, and is “10”.
“0” is a semantic code of a content word class for the category “beverage”, “200” is a semantic code of a content word class for the category “property of the beverage”, and “300” is a content for the category “operation related to the beverage”. This is the meaning code of the word class, and “400” is the meaning code of the content word class for the category “operation related to voice”.

The database 202 stores “(100, 200)”, “(100, 200)” as standard content word class pairs having predetermined co-occurrence relations composed of content word classes.
This is a means for storing “(100, 300)” or the like.

The character recognizing means 203 is an OCR (optical character register) composed of words.
ader) This is a means for character recognition of the input sentence.

The content word selection means 204 looks at the content word dictionary (not shown) in which the learned content words are described, and writes the content-recognized words from the words constituting the character-recognized sentence. This is a means for selecting the content word that is present.

The content word class determining means 205 is a means for determining the content word class to which the selected content word belongs by utilizing the correspondence between the content word stored in the database 201 and the content word class.

The determination means 206 is a means for determining whether or not an arbitrary content word class pair composed of the determined content word classes is a standard content word class pair stored in the database 202.

The misrecognized content word determining means 207 is common to two or more arbitrary content word class pairs among the arbitrary content word class pairs stored in the database 202 and determined not to be the standard content word class pairs. This is a means for determining a content word belonging to the content word class as a misrecognized content word.

The content word re-input request means 208 is a means for requesting re-input of a content word corresponding to the erroneously recognized content word when there is an erroneously recognized content word.

Next, FIG. 9A which is an explanatory diagram (part 1) for explaining the operation of the speech recognition apparatus according to the fifth embodiment of the present invention, and the speech recognition apparatus according to the fifth embodiment of the present invention. For explaining the operation of (2)
The operation of the character recognition device according to the present embodiment will be described with reference to FIG. Note that while describing the operation of the character recognition device in the present embodiment, an embodiment of the sentence recognition method of the present invention will also be described.

The character recognition means 203 (erroneously) recognizes the sentence "Do you drink local beer?" As the recognition sentence "Do you call local beer?"

The content word selecting means 204 generates the words "local" constituting the recognition sentence "Do you call a local beer?"
"No", "beer", "wo", "call", "masu",
Content words "local", "beer", "call" from "ka"
Select Note that, as in the case of the above-described third embodiment, since the content word of the inflected word is selected as the end form (dictionary head form), “call (continuous form)” is selected as “call (end form)”. .

The content word class determining means 205 utilizes the correspondence between the content words stored in the database 201 and the content word classes to select the selected content words “local”, “beer”, and “call”. The content word classes to which they belong are "2
00 "," 100 ", and" 400 ".

The determination means 206 determines a total of three arbitrary content word class pairs “(100, 200)”, “(100, 400)”,
"(100, 200)" in "(200, 400)"
Are standard content word class pairs stored in the database 202, "(100, 400)", "(200, 4
00) ”is not a standard content word class pair stored in the database 202.

The misrecognized content word determining means 207 determines that the arbitrary content word class pair "(100, 40) which is determined not to be the standard content word class pair stored in the database 202.
0) ”and“ (200, 400) ”, a content word“ call ”belonging to a content word class“ 400 ”common to two or more arbitrary content word class pairs is determined to be a misrecognized content word.

The content word re-input request means 208 requests re-input of a content word corresponding to the misrecognized content word "call".

(Embodiment 6) First, the configuration of a character recognition apparatus according to the present embodiment will be described with reference to FIG. 10 which is a block diagram showing the configuration of a character recognition apparatus according to Embodiment 6 of the present invention. .

The character recognition device according to the present embodiment includes a database 201 to 202, a character recognition unit 203,
Content word selecting means 204 and content word class determining means 205
Determining means 206, erroneous sentence recognition determining means 207 ',
Sentence re-input request means 208 '.

The database 201 corresponds to the first database of the present invention, the database 202 corresponds to the second database of the present invention, the character recognizing means 203 corresponds to the sentence recognizing means of the present invention, The selecting means 204 corresponds to the specific word selecting means of the present invention, the content word class determining means 205 corresponds to the specific word class determining means of the present invention, and the determining means 206 corresponds to the determining means of the present invention. The deciding means 207 'corresponds to the sentence misrecognition deciding means of the present invention, the sentence re-input request means 208' corresponds to the sentence re-input request means of the present invention, and the character recognizing device of the present embodiment is the Corresponds to the recognition device. Further, the content words of the present embodiment correspond to the specific words of the present invention, and the content word classes of the present embodiment correspond to the specific word classes of the present invention.

The speech recognition apparatus according to the present embodiment has a configuration similar to the configuration of the speech recognition apparatus according to the third embodiment described above. ′.

The sentence error recognition determining means 207 'and the sentence re-input request means 208' will be described in more detail.

The sentence misrecognition determining means 207 'determines, as a misrecognized content word, a content word that is not included in any arbitrary content word class pair determined to be a standard content word class pair stored in the database 202. , Between the number of selected content words X and the number of misrecognized content words Y

[0114]

## EQU2 ## If the relation of Y> 0.5 × X is established, the input sentence is a means for determining that the sentence is erroneously recognized.

The sentence re-input requesting means 208 'is a means for requesting re-input of the input sentence when there is an erroneous recognition.

Next, the operation of the character recognition device according to the present embodiment will be described with reference to FIG. 10 which is a block diagram showing the configuration of the speech recognition device according to the sixth embodiment of the present invention. Note that while describing the operation of the character recognition device in the present embodiment, an embodiment of the sentence recognition method of the present invention will also be described.

The character recognizing means 203 (erroneously) recognizes the sentence "Would you like to drink a local beer?" As the recognition sentence "Would you like to drink a local building?"

The content word selecting means 204 generates the words "local", which constitute the recognition sentence "Do you not call a local building?"
"No", "Building", "O", "Call", "Nase",
Content words "local", "building",
Select "Call". Note that, as in the case of the above-described third embodiment, since the content word of the inflected word is selected as the end form (dictionary heading form), “call (continuous form)” is selected as “call (end form)”. .

The content word class determining means 205 uses the correspondence between the content words stored in the database 201 and the content word classes to select the selected content words “local”, “building”, and “call”. The content word classes to which they belong are "20
0 "," 90 ", and" 400 ".

The determination means 206 determines a total of three arbitrary content word class pairs “(90, 200)”, “(90, 400)”, “(2
00, 400) ", there is no standard content word class pair stored in the database 202 at all.

The sentence misrecognition determining means 207 'determines that the selected content words "local", "building", and "call" are all misrecognized content words. Also, the sentence misrecognition determining means 207 '
Is (number) between the number of selected content words “local”, “building” and “call” X = 3 and the number of misrecognized content words “local”, “building” and “call” Y = 3. Since the relationship of 2) is established, it is determined that the input sentence "Do you not drink local beer?" Has been incorrectly recognized.

Then, the sentence re-input requesting means 208 'requests re-input of the input sentence "Would you like to drink a local beer?"

In the above, Embodiments 1 to 6 have been described in detail.

In the third and fifth embodiments described above, when there is a misrecognized specific word, the re-input request means of the present invention re-inputs a specific word corresponding to the misrecognized specific word. Although the request has been made, the present invention is not limited to this. For example, when there is a misrecognized specific word, a re-input of the input sentence may be requested.

Of course, when there is a misrecognized specific word (when there is a misrecognition), a notifying means for notifying the user to that effect may be provided. For example, using a voice or text to notify the user of a message such as "The part of the name could not be heard (partly)" has an effect similar to the above-described re-entry request. is there.

The present invention is a program for causing a computer to execute the functions of all or a part of the above-described sentence recognition apparatus of the present invention (or an apparatus, an element, a circuit, a unit, or the like). Includes programs that operate in cooperation with a computer. Of course, the computer is not limited to pure hardware such as a CPU, but may include firmware, an OS, and peripheral devices.

The present invention also provides a program for causing a computer to execute all or some of the steps (or steps, operations, actions, etc.) of the above-described sentence recognition method of the present invention. Includes programs that work in conjunction with.

It should be noted that some of the means (or devices, elements, circuits, parts, etc.) of the present invention, and some of the steps (or steps, operations, actions, etc.) of the present invention may be described by a plurality of means or It may mean some means or steps within a step, or some functions or some operations within one means or step.

Further, some devices (or elements, circuits, units, etc.) of the present invention mean some of the plurality of devices, or some of the means in one device. (Or an element, a circuit, a part, or the like) or a part of the function of one means.

The present invention also includes a computer-readable recording medium on which the program of the present invention is recorded. Further, one usage form of the program of the present invention may be a form in which the program is recorded on a computer-readable recording medium and operates in cooperation with the computer. 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, and sound waves.

Note that the configuration of the present invention may be realized by software or hardware.

The present invention also provides a medium having a program for causing a computer to execute all or a part of the functions of all or part of the above-described sentence recognition apparatus of the present invention, and is readable by a computer. Further, a medium in which the read program cooperates with the computer to execute the function is included.

The present invention also relates to a medium carrying a program for causing a computer to execute all or some of the operations of all or some of the steps of the above-described sentence recognition method of the present invention. A medium is included that can execute the operation in cooperation with the computer when the program that can be read is executed.

The entire disclosures of the above-mentioned documents are unified by quoting (referencing) here as it is.

[0135]

As is apparent from the above description,
The present invention has an advantage that appropriate sentence recognition can be performed using speech recognition or text sentence recognition.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a re-input requesting unit according to Embodiment 1 of the present invention.

FIG. 2 is a block diagram showing a configuration of a re-input request unit according to Embodiment 2 of the present invention.

FIG. 3 is a block diagram illustrating a configuration of a speech recognition device according to a third embodiment of the present invention.

FIG. 4 is a database 1 according to the third embodiment of the present invention.
FIG. 4 is an explanatory diagram of data stored in No. 01.

FIG. 5A is an explanatory diagram (part 1) for explaining the operation of the speech recognition device according to the third embodiment of the present invention; (B) It is explanatory drawing (the 2) for demonstrating operation | movement of the speech recognition apparatus in Embodiment 3 of this invention.

FIG. 6 is a block diagram illustrating a configuration of a speech recognition device according to a fourth embodiment of the present invention.

FIG. 7 is a block diagram illustrating a configuration of a speech recognition device according to a fifth embodiment of the present invention.

FIG. 8A is an explanatory diagram of data stored in a database 201 according to the fifth embodiment of the present invention. (B) Database 20 according to Embodiment 5 of the present invention
FIG. 3 is an explanatory diagram of data stored in a second example.

FIG. 9A is an explanatory view (1) for explaining the operation of the speech recognition device according to the fifth embodiment of the present invention. (B) It is explanatory drawing (the 2) for demonstrating the operation | movement of the speech recognition apparatus in Embodiment 5 of this invention.

FIG. 10 is a block diagram illustrating a configuration of a speech recognition device according to a sixth embodiment of the present invention.

[Explanation of symbols]

 1 Tagged Corpus 2 Dependency Analysis 3 Key Word Dependency Table 4 Key Word Dictionary 5 Speech Recognition 6 Key Word Extraction 7 False Recognition Word Estimation 8 Reject Decision 9 Result Sentence Generation 10 Word Class Determination 11 Key Word Class Dependency Table 12 Thesaurus dictionary

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 15/22 G10L 3/00 561D 537G

Claims (26)

[Claims] 1. A database storing a plurality of predetermined standard specific word pairs, each including a plurality of predetermined specific words, and a sentence recognizing an input sentence including a plurality of words. Recognizing means, specific word selecting means for selecting the specific word from among a plurality of words constituting the recognized sentence, and an arbitrary specific word pair composed of the selected specific words are stored in the database. Determining means for determining whether or not the specified specific word pair is correct; and determining an erroneously recognized specific word in which the recognition is erroneous from the selected specific words based on a result of the determination. A sentence recognition device including a recognition specific word determination unit. 2. The method according to claim 1, wherein the erroneously recognized specific word determining unit is configured to specify a specific common word which is not a standard specific word pair stored in the database and which is common to at least a predetermined number of arbitrary specific word pairs. The sentence recognition device according to claim 1, wherein a word is determined as the misrecognized specific word. 3. When there is the misrecognized specific word,
3. The sentence recognition apparatus according to claim 1, further comprising: (1) a re-input requesting unit for requesting re-input of a specific word corresponding to the erroneously recognized specific word or (2) re-input of the input sentence. 4. When there is the misrecognized specific word,
3. The sentence recognition device according to claim 1, further comprising a notification unit that notifies a user of the fact. 5. A database storing a plurality of predetermined standard specific word pairs composed of a plurality of predetermined specific words, and a sentence recognizing an input sentence composed of a plurality of words. Recognizing means, specific word selecting means for selecting the specific word from among a plurality of words constituting the recognized sentence, and an arbitrary specific word pair composed of the selected specific words are stored in the database. Determining means for determining whether the input sentence is a standard specific word pair, and sentence erroneous recognition determining means for determining whether or not the input sentence is erroneously recognized based on the result of the determination. Sentence recognition device. 6. The sentence recognition apparatus according to claim 5, further comprising a sentence re-input requesting unit for requesting re-input of the input sentence when the erroneous recognition is made. 7. The sentence recognition device according to claim 5, further comprising a notification unit for notifying a user of the erroneous recognition when the erroneous recognition is performed. 8. A first database for storing a correspondence relationship between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong, and comprising the specific word classes. A second database storing a plurality of predetermined standard specific word class pairs, a sentence recognizing means for recognizing an input sentence composed of a plurality of words, and comprising the recognized sentence Using a specific word selecting means for selecting the specific word from among a plurality of words, and using the correspondence stored in the first database, determining the specific word class to which the selected specific word belongs. A specific word class determining means, and an arbitrary specific word class pair composed of the determined specific word classes are stored in the second database. Determining means for determining whether or not the pair is a standard specific word class pair, and erroneous recognition for determining an erroneously recognized specific word in which the recognition is erroneous from among the selected specific words based on a result of the determination. A sentence recognition device including a specific word determining means. 9. The erroneously-recognized specific word determining means, wherein at least a predetermined number of arbitrary specific words in the arbitrary specific word class pairs determined as not being the standard specific word class pairs stored in the second database. The sentence recognition device according to claim 8, wherein a specific word belonging to a specific word class common to the class pair is determined as the misrecognized specific word. 10. When there is the erroneously recognized specific word, (1) re-input of a specific word corresponding to the erroneously recognized specific word, or (2) a re-input requesting re-input of the input sentence. The sentence recognition device according to claim 8, further comprising an input request unit. 11. The sentence recognition apparatus according to claim 8, further comprising a notifying unit that notifies a user of the occurrence of the erroneously recognized specific word. 12. A first database storing a correspondence relationship between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong, and comprising the specific word classes. A second database storing a plurality of predetermined standard specific word class pairs, a sentence recognizing means for recognizing an input sentence composed of a plurality of words, and comprising the recognized sentence Using a specific word selecting means for selecting the specific word from among a plurality of words, and using the correspondence stored in the first database, determining the specific word class to which the selected specific word belongs. A specific word class determining means, and an arbitrary specific word class pair composed of the determined specific word classes are stored in the second database. A standard specific word class pair, and a sentence erroneous recognition determining unit that determines whether the input sentence is erroneously recognized based on the result of the determination. Sentence recognition device. 13. The sentence recognition apparatus according to claim 12, further comprising a sentence re-input requesting unit for requesting re-input of the input sentence when the erroneous recognition is made. 14. The sentence recognition apparatus according to claim 12, further comprising a notifying means for notifying a user of the occurrence of the erroneous recognition. 15. A storage step of storing a plurality of predetermined standard specific word pairs composed of a plurality of predetermined specific words in a database, and a step of storing an input sentence composed of a plurality of words. A recognizing sentence recognizing step, a specific word selecting step of selecting the specific word from among a plurality of words constituting the recognized sentence, and an arbitrary specific word pair composed of the selected specific words is A determining step of determining whether or not the pair is a standard specific word pair stored in a database; based on a result of the determination, an erroneously recognized specific word in which the recognition is incorrect among the selected specific words. A step of determining an erroneously recognized specific word to be determined. 16. A storage step of storing a plurality of predetermined standard specific word pairs composed of a plurality of predetermined specific words in a database, and a step of storing an input sentence composed of a plurality of words. A recognizing sentence recognizing step, a specific word selecting step of selecting the specific word from among a plurality of words constituting the recognized sentence, and an arbitrary specific word pair composed of the selected specific words is A determining step of determining whether the input sentence is a standard specific word pair stored in a database; and a sentence error recognition determining step of determining whether the input sentence is incorrectly recognized based on a result of the determination. Sentence recognition method comprising: 17. A first storing step of storing, in a first database, a correspondence relationship between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong; A second storing step of storing a plurality of predetermined standard specific word class pairs composed of specific word classes in a second database; and a sentence recognizing an input sentence composed of a plurality of words. A recognition step, a specific word selection step of selecting the specific word from among a plurality of words constituting the recognized sentence, and the correspondence selected using a correspondence stored in the first database. A specific word class determining step of determining the specific word class to which the specific word belongs; A determining step of determining whether the specific word class pair is a standard specific word class pair stored in the second database; and, based on a result of the determination, A wrong recognition specific word determining step of determining a wrong recognition specific word whose recognition is incorrect. 18. A first storing step of storing, in a first database, a correspondence relationship between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong; A second storing step of storing a plurality of predetermined standard specific word class pairs composed of specific word classes in a second database; and a sentence recognizing an input sentence composed of a plurality of words. A recognition step, a specific word selection step of selecting the specific word from among a plurality of words constituting the recognized sentence, and the correspondence selected using a correspondence stored in the first database. A specific word class determining step of determining the specific word class to which the specific word belongs; A determining step of determining whether the specific word class pair is a standard specific word class pair stored in the second database; and, based on a result of the determination, whether the input sentence is erroneously recognized. A sentence recognition method, comprising: 19. The sentence recognition method according to claim 15, comprising a plurality of predetermined specific words.
A storage step of storing a plurality of predetermined standard specific word pairs in a database, a sentence recognition step of recognizing an input sentence composed of a plurality of words, and a plurality of words of the plurality of words constituting the recognized sentence. A specific word selecting step of selecting the specific word from within, and determining whether or not an arbitrary specific word pair composed of the selected specific words is a standard specific word pair stored in the database. Steps, based on the result of the determination,
A program for causing a computer to execute all or part of an erroneously recognized specific word determining step of determining an erroneously recognized specific word whose recognition is incorrect from among the selected specific words. 20. The sentence recognition method according to claim 16, comprising a plurality of predetermined specific words.
A storage step of storing a plurality of predetermined standard specific word pairs in a database, a sentence recognition step of recognizing an input sentence composed of a plurality of words, and a plurality of words of the plurality of words constituting the recognized sentence. A specific word selecting step of selecting the specific word from within, and determining whether or not an arbitrary specific word pair composed of the selected specific words is a standard specific word pair stored in the database. Steps, based on the result of the determination,
A program for causing a computer to execute all or a part of a sentence misrecognition determining step of determining whether or not the input sentence is incorrectly recognized. 21. The first database according to claim 17, wherein the correspondence between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong is stored in the first database. A first storing step of storing, a second storing step of storing a plurality of predetermined standard specific word class pairs in the second database, which is composed of the specific word classes, and a plurality of words Sentence recognition step of recognizing the input sentence, and a specific word selection step of selecting the specific word from among a plurality of words constituting the recognized sentence,
A specific word class determining step of determining the specific word class to which the selected specific word belongs using the correspondence stored in the first database; and Determining whether the specified specific word class pair is a standard specific word class pair stored in the second database; and, based on a result of the determination, A program for causing a computer to execute all or a part of an erroneously recognized specific word determining step of determining an erroneously recognized specific word whose recognition is incorrect. 22. The first database according to claim 18, wherein a correspondence between a plurality of predetermined specific words and a plurality of predetermined specific word classes to which the specific words belong is stored in the first database. A first storing step of storing, a second storing step of storing a plurality of predetermined standard specific word class pairs in the second database, which is composed of the specific word classes, and a plurality of words Sentence recognition step of recognizing the input sentence, and a specific word selection step of selecting the specific word from among a plurality of words constituting the recognized sentence,
A specific word class determining step of determining the specific word class to which the selected specific word belongs using the correspondence stored in the first database; and Determining whether the arbitrary specific word class pair is a standard specific word class pair stored in the second database; and erroneously recognizing the input sentence based on the result of the determination. A program for causing a computer to execute all or a part of a sentence misrecognition determining step of determining whether or not the recognition has been performed. 23. A medium carrying the program according to claim 19, wherein the medium can be processed by a computer. 24. A medium carrying the program according to claim 20, wherein the medium can be processed by a computer. 25. A medium carrying the program according to claim 21, wherein the medium can be processed by a computer. 26. A medium carrying the program according to claim 22, wherein the medium can be processed by a computer.