JP2003202893A - Corpus processor for generating statistical language model, and method and program thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To prevent secret information included in a database or existent corpus from leaking out. <P>SOLUTION: A text analysis part 102 outputs a morpheme analytic result of a text. A storage part 104 stores a mask rule in which information for masking a person's name is described. A secret information mask part 103 masks the morpheme analytic result according to the mask rule. Consequently, no secret information is included in the morpheme analytic result from the secret information mask part 103. A corpus totaling part 105 totals morpheme analytic results to generate a corpus excluding the secret information. According to the corpus, a statistical language model is generated. Accordingly, no secret information leaks out in speech recognition processing using the statistical language model. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a corpus processing device, method and program for creating a statistical language model for preventing leakage of personal information or the like to be kept secret during voice recognition processing.

[0002]

2. Description of the Related Art In recent years, a Japanese dictation system capable of inputting Japanese sentences by voice has been put into practical use. This system realizes high recognition performance by using not only acoustic correctness, but also information on the appearance probability of word pairs called statistical language model. Here, the occurrence probabilities of the word pairs that make up the statistical language model are
It is obtained by aggregating statistical information using a statistical method for a large amount of text called a corpus.

Therefore, the probability of occurrence of a word set in a statistical language model strongly depends on the content of sentences in the corpus (text) used to create the language model. That is, in the dictation system, it is easy to recognize expressions that appear in sentences in the text used to create the statistical language model, and conversely, it is difficult to recognize expressions that do not appear. Therefore, when the user inputs a sentence that does not appear in the text used for statistical language model creation by voice, it may be erroneously recognized as a notation string acoustically similar to the word set that appears in the text. Not a few.

These techniques are described in detail in the following documents.

"Speech recognition by probabilistic model" Seiichi Nakagawa The Institute of Electronics, Information and Communication Engineers {ISBN4-88552-072-X} "Spoken Language Processing" Kenji Kita Morikita Publishing {ISBN4-627-8
2380-0} "Probabilistic language model" Kenji Kita The University of Tokyo Press {I
SBN4-13-065404-7}

[0006]

As described above, all the expressions appearing in the corpus (text) are directly reflected in the appearance probability of the word set of the statistical language model. Therefore, in the dictation system, the expression that appears in the corpus (text) is likely to appear as a recognition result.

For this reason, when the corpus (text) used to create the statistical language model contains private personal information or the like, the content of the personal information may appear as an erroneous recognition result. Yes, personal information may be exposed to others.

For example, in the case of performing medical-related voice recognition processing, the recognition accuracy can be improved by using a medical-related corpus. However, there is a problem that personal information is often included in a medical corpus such as a medical record, and the personal information may be exposed depending on the result of the voice recognition process.

According to the present invention, in the process of collecting texts for a statistical language model or in the process of creating a statistical language model from the collected texts, the notation information relating to the confidential information in the texts used for statistical information aggregation is removed. By doing so, it is an object of the present invention to provide a corpus processing device, method, and program for creating a statistical language model that can prevent private personal information from being erroneously output in a recognition result.

[0010]

A corpus processing device for creating a statistical language model according to claim 1 of the present invention includes a text analysis unit for morphologically analyzing text data and outputting a morphological analysis result, and a secret According to a predetermined mask rule for masking words included in information, a secret information mask unit that masks the morphological analysis result, and a corpus accumulation unit that accumulates the morphological analysis result masked by the secret information mask unit as a corpus. And a corpus statistical aggregation unit that collects statistical information from the corpus accumulated by the corpus accumulating unit, and the corpus processing device for creating the statistical language model according to claim 2 of the present invention comprises: A text analysis unit that morphologically analyzes text data and outputs a morphological analysis result, and a word included in confidential information According to a predetermined replacement rule for replacing with a word, a secret information replacement unit that replaces the morpheme analysis result, a corpus accumulation unit that accumulates the morpheme analysis result replaced by the secret information substitution unit as a corpus, and the corpus. And a corpus statistical aggregation unit that collects statistical information from the corpus accumulated by the accumulation unit.

In claim 1 of the present invention, the result of morphological analysis by the text analysis unit is given to the secret information mask unit. The secret information masking unit masks the morphological analysis result according to a predetermined mask rule for masking the words included in the secret information. The corpus accumulating unit accumulates the morphological analysis result in which the words included in the secret information are masked as a corpus. The corpus statistics aggregating unit collects statistical information based on the accumulated corpus. The confidential information is masked in the statistical information, and the confidential information is prevented from leaking in the voice recognition processing using the statistical information.

In claim 2 of the present invention, the result of morphological analysis by the text analysis unit is given to the secret information replacement unit. The secret information replacement unit replaces the morphological analysis result according to a predetermined replacement rule for replacing a word included in the secret information with another word. The corpus accumulating unit accumulates the morphological analysis result in which the words included in the secret information are replaced as a corpus. The corpus statistics aggregating unit collects statistical information based on the accumulated corpus. In the statistical information, the confidential information is replaced with another word, and the confidential information is prevented from leaking in the voice recognition process using the statistical information.

It should be noted that the present invention relating to the apparatus also holds true as an invention relating to the method.

The present invention relating to the apparatus is also realized as a program for causing a computer to execute the processing corresponding to the present invention.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 shows the first of the present invention.
2 is a block diagram showing a corpus processing device for creating a statistical language model according to the embodiment of FIG.

In this embodiment, in the process of collecting texts for a statistical language model, notation information about confidential information (hereinafter referred to as confidential information) such as personal information in the texts used for statistical information aggregation is displayed. By removing in advance, private personal information or the like is prevented from being erroneously output in the recognition result of the dictation system or the like.

Various kinds of data are stored in the database 100. For example, in the medical field such as a hospital, as the information accumulated in the database 100, information such as a medical record and a radiographic result can be considered. The information in the database 100 also includes sentence data including text necessary for creating the statistical language model. In the text data, information that should be kept secret, for example, notation information about personal information may be described.

The text extraction unit 101 extracts text information that can be used as a corpus from the database 100 in which data is stored and supplies it to the text analysis unit 102.

The text analysis unit 102 outputs the result of morphological analysis of the text extracted by the text extraction unit 101. For example, the text analysis unit 102 divides a sentence in a text into words, and obtains a morphological analysis result composed of pairs of the divided words (headings) and their parts of speech. The morphological analysis method used here may be any known method for dividing the text into morphemes composed of pairs of headings and parts of speech.

The analysis result of the text analysis unit 102 is supplied to the secret information mask unit 103. Secret information mask unit 10
3 reads the mask rule stored in the storage unit 104, and according to the read mask rule, the text analysis unit 1
The result of morphological analysis from No. 02 is masked. The mask rule describes information for deleting confidential information such as personal information when creating a corpus.

FIG. 2 is an explanatory diagram for explaining the mask rules stored in the storage unit 104.

In the example of FIG. 2, the mask rule is stored as information on a set of a part of speech and a headline. FIG. 2 shows an example in which two records are accumulated. The * (asterisk) in FIG. 2 indicates that there is no constraint condition for all headings, that is, all words included in the part of speech are masked. In the example of FIG. 2, the part of speech masks all the headlines (words) of the surname: surname and the part of speech represents the person's name:
It shows a process of masking all headings (words) of a name.

For example, by writing "Tanaka" in the heading column instead of an asterisk, it is possible to mask only the word "Tanaka" in the part of speech of the person's name: surname.

Although the example of FIG. 2 shows an example in which the storage unit 104 stores the mask rule described based on the part-of-speech information of the morpheme, the mask rule is expressed by a headline or a combination of the headline and the part-of-speech. It is also possible.

The secret information masking unit 103 masks the morphological analysis result masked in accordance with the mask rule, and outputs the result to the corpus accumulation unit 10.
Output to 5. The corpus accumulating unit 105 gives the input morphological analysis result after the masking to the storage unit 106 and accumulates it as a secret information excluded corpus for creating a statistical language model. In this way, the storage unit 106 stores the corpus from which the secret information is removed according to the mask rule. The corpus format stored in the storage unit 106 may be a morphological analysis result or a simple text format.

The corpus statistics totaling unit 107 includes a storage unit 10.
The corpus accumulated in 6 is read out to create a statistical language model and output to the storage unit 108. The corpus statistics aggregating unit 107 converts the statistical language model (N-
Any known method may be used as a method for creating a gram). The storage unit 108 stores the statistical language model.

By using the statistical language model stored in the storage unit 108, the voice recognition process is performed. In this case, since the secret information is not included in the corpus that is the source of the creation of the statistical language model, it is unlikely that the secret information accumulated in the database 100 appears in the speech recognition result.

Next, the operation of the embodiment thus configured will be described with reference to the explanatory views of FIGS.

Now, it is assumed that the database 100 stores medical information. The text extraction unit 101 extracts text from various information stored in the database 100 and outputs it to the text analysis unit 102. The text analysis unit 102 obtains a morphological analysis result composed of a set of a headline and its part of speech from the text by morphological analysis.

Now, it is assumed that a part of the text extracted by the text extracting unit 101 is as shown in FIG. FIG. 3 shows that in the extracted text, there are notations indicating that Mr. Ichiro Nihon has stomach cancer, Mr. Jiro Nihon has gastric ulcer, and Mr. Saburo Nihon has lung cancer.

In this case, the text analysis unit 102
For example, the analysis result shown in FIG. 4 is obtained. That is, as shown in FIG. 4, each extracted word string in FIG. 3 is divided into words and the part of speech is added. For example, “Nihon Ichiro: Gastric cancer” in FIG. 3 is made up of four words: Japan / Ichiro /: / gastric cancer. “Japan” has a part of speech as a surname: “Ichiro” has a part of speech as a name: ":" Is a sign of part of speech, and "stomach cancer" is a noun of part of speech.

The analysis result of the text analysis unit 102 is given to the secret information mask unit 103. Secret information mask unit 10
3 reads the mask rule from the storage unit 104. For example,
It is assumed that the rule shown in FIG. 2 is described as the mask rule. In this case, the secret information mask unit 103
Among the analysis results of the text analysis unit 102, all the words whose part of speech is a person's name: surname and all words whose part of speech is a person's name: first name are masked, and the morphological analysis result after masking is output to the corpus accumulation unit 105. .

For example, if a word masked by the secret information masking unit 103 according to a mask rule is represented by a circle or a black circle, the secret information masking unit 103 outputs the morphological analysis result shown in FIG.

As is clear from the comparison between FIG. 5 and FIGS. 3 and 4, the words whose part-of-speech is person name: surname and person name: name in the morphological analysis result are masked by ◯ or (black circle), respectively, and concealed. It is output from the information mask unit 103. That is, in the example of FIG. 3, personal information to be concealed, such as a person's name, out of medical-related information included in a medical record of a hospital is prevented from being output as a morphological analysis result.

The corpus accumulating unit 105 stores the morphological analysis result masked by the secret information masking unit 103 in the storage unit 106. In this way, the storage unit 106 stores the corpus from which the secret information is excluded. The corpus statistical aggregation unit 107 creates a statistical language model based on the accumulated corpus and stores it in the storage unit 108.

Speech recognition processing is performed using the statistical language model accumulated in the storage unit 108. The statistical language model of the storage unit 108 is created based on the corpus from which the secret information is excluded, and prevents the secret information from appearing in the speech recognition result.

As described above, in the present embodiment, confidential information is excluded when creating a corpus based on the text extracted from the database. Therefore, when the created statistical language model is used, Can prevent the confidential information from appearing in the voice recognition result. That is, even when using a database of data such as personal information, which must be strictly managed in terms of security due to contracts or legal reasons, leakage of confidential information can be completely prevented.

FIG. 6 is a block diagram showing a second embodiment of the present invention. 6, the same components as those in FIG. 1 are designated by the same reference numerals and the description thereof will be omitted.

In the first embodiment, the confidential information is masked to prevent leakage of the confidential information. As shown in FIG. 5, the originally named portion is masked and the part-of-speech information is also deleted. That is, since all the confidential information such as the name including the part-of-speech information is not used as the morphological analysis result, it may be difficult to accurately convert the name and the like in the voice recognition process. Therefore, in the present embodiment, the secret information is used as a morphological analysis result by replacing it with another typical word.

The present embodiment is different from the first embodiment in that the secret information replacing unit 110 and the storage unit 111 are adopted instead of the secret information masking unit 103 and the storage unit 104, respectively.

The storage unit 111 stores the replacement rule. The secret information replacement unit 110 replaces the word of the secret information in the morphological analysis result from the text analysis unit 102 with another typical word according to the replacement rule of the storage unit 111, and outputs it to the corpus accumulation unit 105. There is.

Next, the operation of the embodiment thus constructed will be described with reference to the explanatory views of FIGS. 3, 7 and 8. 7 and 8 correspond to FIGS. 2 and 4, respectively.

FIG. 7 shows an example of the replacement rule stored in the storage unit 111. In the example of FIG. 7, the replacement rule is stored as information on the set of part of speech, headline, and typical word. FIG. 7 shows an example in which two records are accumulated. Also in FIG. 7, * (asterisk) indicates that there is no constraint condition for all headings, that is, all words included in the part of speech are masked.
In addition, the typical word in FIG. 7 indicates a word to be replaced when the word of the morpheme analysis result is the word defined by the part of speech and the headline.

In the example of FIG. 7, for example, for all headings (words) whose part of speech is person name: surname, the word “Suzuki” is replaced and the part of speech is changed to person name: surname. Although the replacement rule based on the morpheme part-of-speech information is described in FIG. 7, it can be expressed by a headline or a combination of a headline and a part-of-speech.

Now, it is assumed that a part of the text extracted by the text extracting unit 101 is as shown in FIG.

The morpheme analysis result in this case is as shown in FIG. 4, for example, and the analysis result is the secret information replacing unit 110.
Given to. The secret information replacement unit 110 reads the replacement rule from the storage unit 111. When the replacement rule is as shown in FIG. 7, the confidential information replacement unit 110 replaces all words whose part of speech is personal name: surname in the analysis result of the text analysis unit 102 with “Suzuki” and whose part of speech is personal name. : My surname,
Part-of-speech is a person's name: All words with the name are replaced with “Taro”, and the part-of-speech is made a person's name: name, and the morphological analysis result after the replacement is output to the corpus accumulation unit 105.

In this way, the secret information replacing unit 110 outputs the morphological analysis result shown in FIG. As is clear from the comparison between FIG. 4 and FIG. 8, the part of speech in the result of morphological analysis is replaced with the typical words “Suzuki” and “Taro” as words whose names are: surname and surname: name. It is output from the secret information replacing unit 110. That is, for example, of the medical-related information included in the medical record of the hospital, the personal information that should be kept secret, such as the personal name, is replaced with the typical personal name, for example, "Taro Suzuki", instead of the original personal name. It is possible to prevent personal information from being output as a morphological analysis result. In addition, since the same part-of-speech information as that of the word before replacement is also added to the word after replacement, it is possible to utilize the part-of-speech information in the voice recognition process.

Other functions are similar to those of the first embodiment.

As described above, in the present embodiment, the confidential information is replaced with a typical word and is output as a morphological analysis result. Therefore, even when the created statistical language model is used, the confidential information is kept secret. It is possible to prevent the information from appearing in the voice recognition result. Moreover, since the part-of-speech information is not lost even when the word is replaced, it is possible to perform the process using the part-of-speech information in the voice recognition process using the statistical language model.

FIG. 9 is a block diagram showing a third embodiment of the present invention. 9, the same components as those in FIG. 6 are designated by the same reference numerals and the description thereof will be omitted.

In the second embodiment, the secret information is prevented from leaking by replacing the secret information with a typical word. Similarly, in the present embodiment, the secret information is replaced with an appropriate hidden character to prevent leakage of the secret information. When replacing, the same part-of-speech information as the part-of-speech information before the replacement is added to the blank character.

The present embodiment differs from the second embodiment in that a storage unit 121 is adopted instead of the storage unit 111. The storage unit 121 stores a replacement rule for replacing with a hiding character having part-of-speech information.

Next, the operation of the embodiment configured as described above will be described with reference to the explanatory views of FIGS. 10 to 13. 10 to 13 correspond to FIGS. 2 to 5, respectively.

FIG. 10 shows an example of the replacement rule stored in the storage unit 121. In the example of FIG. 10, the replacement rule is stored as information on a set of a part of speech, a headline, and a mask word. FIG. 10 shows an example in which 11 records are accumulated. In Figure 10, * (asterisk)
Means all words, for example, * prefecture indicates all headings whose last character is “prefecture”. The mask word in FIG. 10 indicates a word to be replaced with a hidden character when the word as a result of morphological analysis is a word defined by a part of speech and a headline.

In the example of FIG. 10, for example, for all headings (words) whose part of speech has "prefecture" as the last character in the place name, the word "□ prefecture" is replaced and the part of speech is used as the place name. Shows. Note that the replacement rule in FIG. 10 is executed, for example, from the top of the record in order, and, for example, rather than a headline whose part of speech is represented by * of a place name,
The rules described in the upper records such as * city, * road, ... Are executed first.

Also, in FIG. 10, the replacement rule based on the morpheme part-of-speech information is described, but it can be expressed by a headline or a combination of a headline and a part-of-speech.

Now, description will be given assuming that a part of the text extracted by the text extracting unit 101 is that shown in FIG. FIG. 11 shows that the address of Ichiro Nihon is “1-2-345 Shibaura, Minato-ku, Tokyo” in the extracted text, and the address of Jiro Nippon is “6-7 Komukai, Kouki-ku, Kawasaki-shi, Kanagawa Prefecture”. -8 "and that there is a notation indicating that the address of Mr. Saburo Nihon is" 9-1 Sachimachi, Kawasaki-ku, Kanagawa ".

In this case, the text analysis unit 102
For example, the analysis result shown in FIG. 12 is obtained. That is, as shown in FIG. 12, each extracted word string in FIG. 11 is divided into words and the part of speech is added. For example, “Ichiro Nippon: 1-2-345 Shibaura, Minato-ku, Tokyo” in FIG. 11 is Japan / Ichiro /: / Tokyo / Minato-ku / 1 /-/ 2 /-/ 3/4 /
It consists of 12 words of 5, "Japan" is the part of speech of the person: surname, "Ichiro" is the part of speech of the person: first name, ":" is the symbol of the part of speech, and "Tokyo" is the place of the part of speech. "Minato Ward" is a place of speech part name, "1" is a part of speech number, "-" is a part of speech symbol, "2" is a part of speech number, "-" is a part of speech symbol,
"3" is a part of speech number, "4" is a part of speech number, "5"
It can be obtained from the analysis result that the part of speech is a number.

These morphological analysis results are given to the secret information replacing section 110. The secret information replacement unit 110 is the storage unit 1.
The replacement rule is read from 21 and the secret information is replaced according to the replacement rule. When the replacement rule is as shown in FIG. 10, the confidential information replacement unit 110 replaces all words whose part of speech is a personal name: surname, in the analysis result of the text analysis unit 102, with “○”. : The surname and the part-of-speech is the person's name: All words in the name are replaced with "(black circles)" The part-of-speech is the person's name: All words with the part-of-speech are numbers are replaced with "◇" The place name is also replaced according to the replacement rule of FIG. In this way, the secret information replacing unit 110 outputs the morpheme analysis result shown in FIG.

As is clear from the comparison between FIG. 12 and FIG. 13, the words defined by the replacement rules in the morphological analysis results are replaced with characters of the circle, (black circle), ◇, (black diamond) shape. , Or □, (black square) and “city”, “road”,
It is replaced with a word added with the letters "fu", "prefecture", "city", "ward", and "town". For example, 6-7 Komukai, Saiwai-ku, Kawasaki-shi, Kanagawa, which is the address of "Jiro Nihon" in FIG.
-8 ”is“ □ prefecture (black square) city (black square) ward (black rhombus) ◇
-◇-◇◇◇ ". That is, for example, personal information to be concealed, such as a person's name or a place name, out of medical-related information included in a medical record of a hospital or the like, is replaced with a hiding letter to which information of a part of speech is added. Output is prevented. Moreover, since the same part-of-speech information as that of the word before replacement is added to the blank character, it is possible to utilize the part-of-speech information in the voice recognition process.

Other functions are similar to those of the second embodiment.

As described above, also in this embodiment, the same effect as that of the second embodiment can be obtained.

FIG. 14 is a block diagram showing a fourth embodiment of the present invention. 14, the same components as those of FIG. 6 are designated by the same reference numerals and the description thereof will be omitted.

In the embodiments shown in FIGS. 2 and 3, the secret information is prevented from leaking by replacing the secret information with a predetermined typical word or a hidden letter. However, since the secret information word is replaced with a hiding letter or a typical word, the morpheme analysis result does not include an individual name, and when a statistical language model created based on this morpheme analysis result is used. May make it difficult to correctly perform voice recognition processing on a personal name or the like. Therefore, the present embodiment prevents leakage of confidential information by replacing the confidential information word with a random word having the same part of speech, and
This makes it possible to correctly perform voice recognition processing even for an individual name or the like.

This embodiment differs from the second embodiment in that the secret information replacing unit 110 and the storage unit 111 are replaced with a secret information replacing unit 130 and a storage unit 131, respectively.
The storage unit 131 stores a replacement rule for replacing the confidential information with another random word having the same part of speech.

Next, the operation of the embodiment thus configured will be described with reference to the explanatory views of FIGS. 15 to 18 correspond to FIGS. 2 to 5, respectively.

FIG. 15 shows an example of the replacement rule stored in the storage unit 131. In the example of FIG. 15, the replacement rule is stored as information on a set of a part of speech, a headline, and a random word. FIG. 15 shows an example in which 11 records are accumulated. In Figure 15, * (asterisk)
Means all words. The random word in Figure 15 is
When the word of the morphological analysis result is a word defined by a part of speech and a headline, it indicates a word to be replaced.
d () indicates that the word is randomly extracted.

In the example of FIG. 15, for example, the part of speech is the number 0.
-9 indicates that each part of speech is replaced by a random number 0-9. Also, in FIG. 15, the replacement rule based on the morpheme part-of-speech information is described, but it can be expressed by a headline or a combination of a headline and a part-of-speech.

Now, description will be given assuming that a part of the text extracted by the text extracting unit 101 is as shown in FIG. In Fig. 16, some numerical value related to Mr. Ichiro Nihon is "123-1234" in the extracted text.
That is, some numerical value regarding Mr. Jiro Nihon is "1.
23-4567 ”, and that there is a notation indicating that some numerical value regarding Mr. Saburo Nippon is“ 999-9999 ”. In addition, these numerical values are various information such as various measured values, amount of money, and age.

In this case, the text analysis unit 102
For example, the analysis result shown in FIG. 17 is obtained. That is, as shown in FIG. 17, each extracted word string in FIG. 16 is divided into words and the part of speech is added. For example, “Ichiro Nihon: 123-1234” in FIG. 16 is Japan / Ichiro /: /
It consists of 12 words 1/2/3 /-/ 1/2/3/4/5. "Japan" has a part of speech as a surname: "Ichiro" has a part of speech as a personal name: name, and ":". Is a symbol for the part of speech, "1" ~
"3" is a part of speech number, "-" is a part of speech symbol, "1"
It can be obtained from the analysis result that the part of speech of "4" is a number.

These morphological analysis results are given to the secret information replacing unit 130. The secret information replacing unit 130 is the storage unit 1.
The replacement rule is read from 31 and the secret information is replaced according to the replacement rule. When the replacement rule is as shown in FIG. 15, the confidential information replacement unit 130 has the part of speech whose personal name is: surname and surname and personal name in the analysis result of the text analysis unit 102.
All words of the name are replaced with other words of the same part of speech that are randomly extracted, and the same part of speech is attached, and for all words of which the part of speech is a number, the randomly extracted part of speech is replaced with another number of a number. In this way, the secret information replacing unit 130 outputs the morphological analysis result shown in FIG.

As is clear from the comparison between FIG. 17 and FIG. 18, the word defined by the replacement rule in the morphological analysis result is replaced with another word of the same part of speech that is randomly extracted. For example, “123-1234” of “Nihon Ichiro” in FIG. 17 is “313-” of “Ryoko Suzuki”.
6924 ". That is, for example, personal information to be concealed such as a person's name and a numerical value among medical-related information included in a medical record of a hospital or the like is replaced with another randomly extracted word having the same part of speech, so that the personal information is morpheme. It is prevented from being output as the analysis result. Further, since a word with the same part of speech extracted at random is used as the word after replacement, the morphological analysis result after replacement includes information such as a name and a numerical value. Therefore, when using a statistical language model created based on this morphological analysis result,
In the voice recognition process, it is possible to surely recognize the name, the numerical value and the like.

Other functions are similar to those of the second embodiment.

As described above, also in this embodiment, the same effects as those of the second and third embodiments can be obtained. Furthermore, since the words of the same part of speech that are randomly extracted are used as the words after replacement, it is possible to create a statistical language model that includes information such as names and numerical values that are actually used, and speech recognition accuracy is reduced. Can be prevented. Further, by extracting the word after the replacement from the database according to the usage pattern, a statistical language model suitable for the usage pattern can be created, and the voice recognition accuracy can be further improved.

FIG. 19 is a block diagram showing the fifth embodiment of the present invention. 19, the same components as those in FIG. 1 are designated by the same reference numerals and the description thereof will be omitted.

In the above-described first to fourth embodiments, in the process of collecting the text for the statistical language model,
The notational information about confidential information in the text used for statistical information aggregation was removed in advance. On the other hand, in the present embodiment, in the process of creating the statistical language model from the existing corpus (text after collecting the text for the statistical language model), the notation information about the secret information in the text used for the statistical information aggregation Is to be removed in advance.

The present embodiment differs from the embodiment shown in FIG. 1 in that a corpus 140 is adopted in place of the database 100 and a corpus input unit 141 is adopted in place of the text extraction unit 101.

The corpus 140 is a collection of texts collected for creating a statistical language model. The corpus 140 may include confidential information.
The corpus input unit 140 sets the text in the corpus 141 to the text analysis unit 1 with the corpus 141 as a processing target.
Output to 02.

In the embodiment configured as described above, the text is extracted from the existing corpus 140 already created and output to the text analysis unit 102. Accordingly, thereafter, by the same method as that of the first embodiment, it is possible to obtain a morphological analysis result in which the secret information is masked and accumulate the corpus from which the secret information is excluded.

Other operations are similar to those of the embodiment shown in FIG.

As described above, in the present embodiment, even when the existing corpus is used, it is possible to convert the corpus to the secret information excluded, and the statistical language model is used without leaking the secret information. Speech recognition processing can also be enabled.

The fifth embodiment is the same as the second to fourth embodiments.
It can also be applied to the embodiment. 20 to 22 are block diagrams when the fifth embodiment is applied to the second to fourth embodiments shown in FIGS. 6, 9 and 14.

20 to 22, the corpus 140 is used in place of the database 100, and the corpus input unit 141 is adopted in place of the text extraction unit 101, which is a difference from FIGS. 6, 9, and 14. .

Other configurations and operations are the same as those of the second to fourth embodiments, respectively.

As described above, also in the examples shown in FIGS. 20 to 22, the same effects as those of the first to fifth embodiments can be obtained.

[0086]

As described above, according to the present invention, in the process of collecting texts for a statistical language model or in the process of creating a statistical language model from the collected texts, the texts used for statistical information aggregation By removing the notation information regarding the personal information therein, it is possible to prevent the private personal information from being erroneously output in the recognition result.

[Brief description of drawings]

FIG. 1 is a block diagram showing a corpus processing device for creating a statistical language model according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram for explaining mask rules stored in a storage unit 104 in FIG.

FIG. 3 is an explanatory diagram for explaining the operation of the first embodiment.

FIG. 4 is an explanatory diagram for explaining the operation of the first embodiment.

FIG. 5 is an explanatory diagram for explaining the operation of the first embodiment.

FIG. 6 is a block diagram showing a second embodiment of the present invention.

7 is an explanatory diagram for explaining a replacement rule stored in a storage unit 111 in FIG.

FIG. 8 is an explanatory diagram for explaining the operation of the second embodiment.

FIG. 9 is a block diagram showing a third embodiment of the present invention.

FIG. 10 is an explanatory diagram illustrating replacement rules stored in a storage unit 121 in FIG. 9.

FIG. 11 is an explanatory diagram for explaining the operation of the third embodiment.

FIG. 12 is an explanatory diagram for explaining the operation of the third embodiment.

FIG. 13 is an explanatory diagram for explaining the operation of the third embodiment.

FIG. 14 is a block diagram showing a fourth embodiment of the present invention.

FIG. 15 is an explanatory diagram illustrating replacement rules stored in a storage unit 131 in FIG.

FIG. 16 is an explanatory diagram for explaining the operation of the fourth embodiment.

FIG. 17 is an explanatory diagram for explaining the operation of the fourth embodiment.

FIG. 18 is an explanatory diagram for explaining the operation of the fourth embodiment.

FIG. 19 is a block diagram showing a fifth embodiment of the present invention.

FIG. 20 is a block diagram showing a modified example of the fifth embodiment.

FIG. 21 is a block diagram showing a modification of the fifth embodiment.

FIG. 22 is a block diagram showing a modification of the fifth embodiment.

[Explanation of symbols]

100 ... Database, 101 ... Text extraction unit, 10
2 ... Text analysis unit, 103 ... Secret information mask unit, 10
4, 106, 108 ... Storage unit, 105 ... Corpus accumulation unit, 107 ... Corpus statistical aggregation unit.

─────────────────────────────────────────────────── ─── Continued Front Page (51) Int.Cl. ⁷ Identification Code FI Theme Coat (Reference) G10L 3/00 521C

Claims (12)

[Claims] 1. A text analysis unit for morphologically analyzing text data and outputting a morphological analysis result, and a secret information mask for masking the morphological analysis result according to a predetermined mask rule for masking a word included in the secret information. A corpus accumulating unit that accumulates the morphological analysis results masked by the secret information masking unit as a corpus, and a corpus statistical aggregation unit that collects statistical information from the corpus accumulated by the corpus accumulating unit. A corpus processing device for creating a characteristic statistical language model. 2. A text analysis unit that morphologically analyzes text data and outputs a morpheme analysis result, and replaces the morpheme analysis result according to a predetermined replacement rule for replacing a word included in confidential information with another word. A secret information replacing unit, a corpus accumulating unit that accumulates the morphological analysis results replaced by the secret information replacing unit as a corpus, and a corpus statistics aggregating unit that collects statistical information from the corpus accumulated by the corpus accumulating unit. A corpus processing device for creating a statistical language model characterized by being provided. 3. The statistical language model according to claim 1, wherein the text data given to the text analysis unit is extracted from a predetermined database. For corpus processing equipment. 4. The statistical language model according to claim 1, wherein the text data given to the text analysis unit is extracted from a predetermined corpus. For corpus processing equipment. 5. The corpus processing device for creating a statistical language model according to claim 2, wherein the secret information replacing unit replaces a word included in the secret information with a typical word. 6. The corpus for creating a statistical language model according to claim 2, wherein the secret information replacing unit replaces a word included in the secret information with a hiding letter to which part of speech information is added. Processing equipment. 7. The corpus processing device for creating a statistical language model according to claim 2, wherein the secret information replacing unit replaces a word included in the secret information with a random word. 8. The mask rule or the replacement rule is a rule for masking or replacing a word having a part of speech of at least one of a person's name, a place name and a number as a word included in the confidential information. A corpus processing device for creating the statistical language model according to claim 1. 9. A secret information mask for masking the morphological analysis result according to a text analysis procedure for morphologically analyzing text data and outputting a morphological analysis result, and a predetermined mask rule for masking a word included in the confidential information. A statistical language model is created which comprises a procedure, a corpus accumulation procedure for accumulating the masked morphological analysis results as a corpus, and a corpus statistical aggregation procedure for collecting statistical information from the accumulated corpus. Corpus processing method to do. 10. The morphological analysis result is replaced according to a text analysis procedure for morphologically analyzing text data and outputting a morphological analysis result, and a predetermined replacement rule for replacing a word included in confidential information with another word. A secret information replacement procedure, a corpus collection procedure for collecting the replaced morphological analysis results as a corpus, and a corpus statistics collection procedure for collecting statistical information from the collected corpus. A corpus processing method for creating a language model. 11. A morphological analysis process of morphologically analyzing text data and outputting a morphological analysis result to a computer, and masking the morphological analysis result according to a predetermined mask rule for masking a word included in confidential information. Create a statistical language model for executing confidential information mask processing, corpus accumulation processing that accumulates the masked morphological analysis results as a corpus, and corpus statistical aggregation processing that collects statistical information from the accumulated corpus Corpus processing program to do. 12. The morphological analysis according to a text analysis process of morphologically analyzing text data and outputting a morphological analysis result to a computer, and a predetermined replacement rule for replacing a word included in confidential information with another word. Statistical information for executing secret information replacement processing for replacing results, corpus accumulation processing for accumulating the replaced morphological analysis results as a corpus, and corpus statistical aggregation processing for collecting statistical information from the accumulated corpus A corpus processing program for creating a language model.