JP2013041421A - Input character string error detection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an input character string error detection device that detects an error of an input character string and does not perform unnecessary error detection on an input character string with no errors fewer errors.SOLUTION: A correct notation determination part 2 determines whether an input KATAKANA (square form of Japanese syllabary) character string includes a specific character. A confirmation candidate extraction part 3 extracts the specific character from the KATAKANA character string only when the correct notation determination part 2 determines that the specific character is not included therein. A confirmation candidate presentation part 4 mutually associates the input KATAKANA character string with the place extracted by the confirmation candidate extraction part 3 and presents it to a user.

Description

  The present invention relates to an input character string error detection device that detects an error in an input character string.

  Detecting an error in the input character string is performed in various cases. For example, it is necessary when creating a dictionary for a speech synthesizer that inputs and reads Japanese text composed of kana characters. As a dictionary for the speech synthesizer, a spoken character dictionary in which an incorrect character string is corrected to a correct character string so that it can be heard correctly when the input kana character string is read, and correspondence between an incorrect kana character string and a correct character string There is a correspondence dictionary and so on. When these dictionaries are created, it is necessary to detect errors existing in the input kana character string. In addition, Japanese text composed of kana characters is a copy of a ledger created when, for example, a computer was unable to handle kanji, and when checking against a copy of a ledger created based on this copy, It is input to the speech synthesizer and read out.

  As a technique for detecting an error in an input character string, the following technique is known (see, for example, Patent Document 1). In the technique described in Patent Document 1, a katakana character string is extracted from an input Japanese character string, and it is checked whether or not the extracted katakana character string is listed in a Japanese dictionary. If it is an unknown word not included in the Japanese dictionary, it is determined that there is an error in the katakana character string and presented to the user.

JP-A-3-242755

  Since the technique described in Patent Document 1 always performs error detection on a katakana character string extracted from a Japanese character string, it is useless for a katakana character string with no errors or few errors. There is a problem that error detection is performed.

  The following errors exist in the input kana character string.

(a) Long sounds are not taken into account (for example, “Tokyo” is used to describe what you want to say “Tokyo”)
(b) Improper writing of stuttering (such as writing “Kyo” for what you want to say “Kyo”)
(c) Improper way of writing sound (for example, “hattori” is used to describe what you want to say)
(d) Improper way of writing muddy / semi-voiced sound
(e) Ambiguity in reading simple characters (such as writing what you want to say "Hi" in Hiragana "he")

[Object of the invention]
SUMMARY OF THE INVENTION An object of the present invention is to provide an input character string error detection apparatus that solves the problem that an error detection may be performed unnecessarily for an input character string having no error or few errors. is there.

The first input character string error detection device according to the present invention is:
Determining means for determining whether or not a specific character is included in the input character string;
Only when the determination unit determines that the specific character is not included in the input character string, the detection unit includes a detection unit that detects a location where the specific character is missing from the input character string. .

An input character string error detection method according to the present invention includes:
An input character string error detection method executed by a computer including a determination unit and a detection unit,
The determination means determines whether or not the input character string includes a specific character;
Only when the determination means determines that the specific character is not included in the input character string, the detection means detects a location where the specific character is missing from the input character string. .

The program according to the present invention is:
A program for causing a computer to function as an input character string error detection device,
The computer,
Determination means for determining whether or not a specific character is included in the input character string;
Only when the determination unit determines that the specific character is not included in the input character string, the input character string functions as a detection unit that detects a portion where the specific character is missing from the input character string. .

  According to the present invention, error detection can be prevented from being performed unnecessarily for an input character string having no or few errors.

It is a block diagram which shows the structural example of the 1st Embodiment of this invention. It is a flowchart which shows the process example of the correct notation determination part 2. FIG. 5 is a flowchart illustrating a processing example of a confirmation candidate extraction unit 3. FIG. 10 is a diagram for explaining an example of an operation of a confirmation candidate presenting unit 4. 10 is a diagram for explaining another example of the operation of the confirmation candidate presenting unit 4. FIG. It is a figure for demonstrating the 2nd Embodiment of this invention. It is a block diagram which shows the structural example of the 3rd Embodiment of this invention.

  Next, embodiments of the present invention will be described in detail with reference to the drawings.

[First embodiment of the present invention]
Referring to FIG. 1, a first embodiment of an input character string error detection device according to the present invention is a kana character string input unit 1, a correct answer notation determination unit 2, a confirmation candidate extraction unit 3, and a confirmation candidate presentation. And a storage device 5 including a correct notation determination rule storage unit 51 and a confirmation candidate extraction rule storage unit 52.

  The kana character string input unit 1 has a function of inputting a kana character string from a dictionary or database (not shown). Here, the kana character string is a character string composed of hiragana or katakana and includes full-width and half-width variations depending on the character code. In the present embodiment, the kana character string input unit 1 inputs a kana character string in a predetermined input unit (for example, a term or a word unit).

  In the correct notation determination rule storage unit 51, correct answer notation determination rules for determining whether or not the correct answer notation for each notation type is included in the kana character string input from the kana character string input unit 1 are recorded. ing. More specifically, if a specific character predetermined for each notation type is included in the input kana character string, it is determined that the correct notation for the notation type is included in the kana character string, Otherwise, a correct notation determination rule for determining that the correct answer notation for the notation type is not included in the kana character string is recorded. In the present embodiment, the correct notation determination rule for the notation type “long sound, stuttering, prompting sound, and muddy / semi-voiced sound” is recorded in the correct notation determination rule storage unit 51.

  In the confirmation candidate extraction rule storage unit 52, for each notation type, to extract from the input kana character string a confirmation candidate consisting of a combination of characters or characters that allows the user to confirm whether or not to indicate the notation type. The confirmation candidate extraction rule is recorded. In other words, in the confirmation candidate extraction rule storage unit 52, for each notation type, a rule for detecting from the input kana character string a location where a specific character defined for the notation type is missing. Is recorded. In this embodiment, the confirmation candidate extraction rule for the notation type “long sound, stuttering, prompting sound, and muddy / semi-voiced sound” is recorded in the confirmation candidate extraction rule storage unit 52.

  The correct notation determination unit 2 has a function of determining, for each notation type, whether or not there is a correct notation in the input kana character string according to the correct notation determination rule. Furthermore, when there is a notation type that does not include the correct answer notation in the kana character string, the correct answer notifying determination unit 2 may include an error notation regarding the notation type in the kana character string. It has a function of judging and passing the information indicating the notation type and the kana character string to the confirmation candidate extraction unit 3. In addition, as for the notation type in which the correct answer notation is included in the kana character string, the correct notation is considered when creating the above kana character string (when recording in the dictionary or database), and there is no possibility of incorrect notation Therefore, the process for notifying the confirmation candidate extraction unit 3 of the notation type is not performed. If the kana character string includes correct notations for all the notation types, the kana character processing is immediately terminated without passing the kana character string to the confirmation candidate extraction unit 3.

  The confirmation candidate extraction unit 3 extracts a confirmation candidate for the notation type passed from the correct notation determination unit 2 from the kana character string passed from the correct notation determination unit 2 according to the confirmation candidate extraction rule, and the extraction result (for example, , Extraction position) and the kana character string are passed to the confirmation candidate presentation unit 4.

  The confirmation candidate presenting unit 4 confirms the kana character string and the kana character string by displaying them on a display unit (not shown) based on the kana character string and the extraction result passed from the confirmation candidate extracting unit 3. It has a function of presenting candidates to the user.

  The input character string error detection device having such a function can be realized by a computer. When realized by a computer, a disk, a semiconductor memory, and other recording media on which a program for causing the computer to function as an input character string error detecting device is prepared are prepared, and the computer is caused to read the program. The computer controls its own operation according to the read program, thereby realizing a kana character string input unit 1, a correct notation determination unit 2, a confirmation candidate extraction unit 3, and a confirmation candidate presentation unit 4 on the computer.

  Next, the operation of the present embodiment will be described in detail. In the following description, each rule is explained by unifying katakana, but the same applies to hiragana, and katakana and hiragana may be regarded as the same.

  The kana character string input unit 1 inputs a kana character string from a database or an electronic dictionary (not shown). In the present embodiment, the kana character string input unit 1 inputs a kana character string in terms of terms.

  The correct answer notation determination unit 2 executes the process shown in the flowchart of FIG. 2 every time a kana character string is input from the kana character string input unit 1. First, the correct notation determination unit 2 determines whether or not the input kana character string includes a long sound notation according to the correct notation determination rule recorded in the correct notation determination rule storage unit 51 (step S201). .

  Here, the correct notation determination rule for long sound notation (sometimes referred to as correct answer determination rule 1) includes correct notation when one or more long sound characters “-” are included in the kana character string. If not, it is determined that the correct answer notation is not included. In addition, instead of determining a long sound character alone, a set of possible kana characters and long sound characters such as “A” and “E” is prepared, and a rule for determining whether or not any of them is included. It is also possible to adopt. Depending on the application, a rule may be adopted in which half-width long sound symbols, tilde symbols “˜”, hyphen characters “-”, etc. are identified as long sound characters.

  If the long sound notation is not included (No in step S201), it is determined that there is a possibility that the long sound notation is erroneously included, and the storage device 5 or the like may include an erroneous notation. “Long sound notation” is recorded as a characteristic notation type (step S202), and then it is determined whether or not stuttering notation is included in the input kana character string according to the correct notation determination rule (step S203). On the other hand, when the long sound notation is included (step S201 is Yes), it is determined that there is a low possibility that the long sound notation includes an incorrect notation, and the next determination item is determined according to the correct notation determination rule. It is determined whether the stuttering notation is included in the input kana character string (step S203).

  Here, the correct answer notation determination rule related to stuttering notation (sometimes referred to as correct answer notifying determination rule 2) includes any one of stuttering characters “a”, “yu”, “yo”, and “ヮ” in the kana character string. Is determined to include the correct answer notation, otherwise, the correct answer notation is not included. Instead of judging the roaring character alone, it is also possible to prepare a set of possible roaring notation and adopt a rule for judging whether any of them is included. In this case, for example, it may be determined that “ca”, “ju”, and “ku” are in stuttering, but “yo” and “key” are not in stuttering.

  If it is determined in step S203 that the input kana character string does not include the stuttering notation, it is determined that there is a possibility that the stuttering notation is included, and the storage device 5 or the like has an incorrect notation. “Stuttering notation” is recorded as the notation type that may be included (step S204), and then it is determined whether or not the input kana character string includes a prompt sound notation according to the correct notation determination rule (step S205). ). On the other hand, when the stuttering notation is included (Yes in step S203), it is determined that there is a low possibility that the stuttering notation is included, and it is determined whether or not the prompting notation is included (step S205). ).

  Here, the correct notation determination rule (sometimes referred to as correct answer determination rule 3) for the prompt sound notation includes the correct answer notation if it includes one or more prompt sound characters “tsu”, otherwise it is the correct answer. It is determined that the notation is not included. Instead of determining the prompting characters alone, a set of possible prompting notations may be prepared and a rule for determining whether any of them is included may be adopted. In this case, for example, “A” and “GA” are determined to be in long sound notation, but “A” is determined not to be in long sound.

  If the input kana character string does not include a prompting notation (No in step S205), it is determined that there is a possibility that the prompting notation is erroneously included, and the storage device 5 or the like has an incorrect notation. “Sound prompt notation” is recorded as a notation type that may be included (step S206), and then it is determined whether or not the input kana character string includes muddy / semi-voiced notation according to the correct answer notation determination rule. (Step S207). On the other hand, when the prompt sound notation is included (Yes in step S205), it is determined that there is a low possibility that the prompt sound notation includes an incorrect notation, and the next determination item is the muddy / semi-voiced sound. Is included in the input kana character string (step S207).

  Here, the correct answer notation judgment rule (sometimes referred to as the correct answer notation judgment rule 4) for the muddy / semi-voiced sound is correct when the kana character string includes one or more characters constituting the muddy or semi-voiced sound. If notation is included, otherwise, it is determined that the correct answer notation is not included. The characters constituting the muddy sound are kana characters in the “Ga”, “The”, “Da”, and “B” lines. The characters composing the semi-voiced sound are kana characters in the “pa” line. The muddy character “” and the semi-voiced character “°” may be included here. In this case, you may employ | adopt the rule which confirms that the character string combined with the character preceding those characters is a character string which comprises a muddy sound or a semi-voiced sound. In this case, for example, “D” is determined to be a muffled sound notation character string, but “A” is determined not to be a muddy sound notation character string.

  If the input kana character string does not include the muddy / semi-voiced notation (No in step S207), it is determined that there is a possibility that a mis-sound / semi-voiced notation is included, and the storage device “Muddy / semi-voiced sound” is recorded as a notation type that may include an erroneous notation in 5 (step S208), and then the process of step S209 is performed. On the other hand, if the input kana character string includes a muddy / semi-voiced sound (Yes in step S207), it is determined that there is a low possibility that the muddy / semi-voiced sound contains an erroneous notation. The process of S209 is performed.

  In step S209, the correct notation determination unit 2 refers to the storage device 5 and checks whether or not a notation type that may contain an incorrect notation is recorded. And when it is not recorded (step S209 is No), the process is complete | finished. On the other hand, if it is recorded (Yes in step S209), the recorded notation type and the kana character string input from the kana character string input unit 1 are passed to the confirmation candidate extraction unit 3 (step) Thereafter, the notation type recorded in the storage device 5 is erased (step S211), and the process is terminated.

  When the notation type and kana character string that may contain an incorrect notation are passed from the correct notation determination unit 2, the confirmation candidate extraction unit 3 receives the notation type of the passed notation type as shown in the flowchart of FIG. Paying attention to one, according to the confirmation candidate extraction rule for the notation type, a confirmation candidate is extracted from the kana character string (steps S31 and S33). Then, when the above processing is performed for all the notation types passed from the correct notation determination unit 2 (Yes in step S32), the kana character string and the extraction result (for example, the extraction position for each notation type) are confirmed. It passes to the candidate presentation part 4 (step S34), and the process is complete | finished. The confirmation candidate presenting unit 4 displays the kana character string and the confirmation in the kana character string by displaying them on a display unit (not shown) based on the kana character string and the extraction result passed from the confirmation candidate extracting unit 3. Candidates are presented to the user. At this time, the confirmation candidate is displayed in a form that can be understood as to which notation type the confirmation candidate is.

  Here, the confirmation candidate extraction rules for each of the notation types “long sound notation”, “stuttering notation”, “prompting sound notation”, and “turbid / semi-voice notation” will be described in detail.

(1) Confirmation candidate extraction rule for “long sound notation”
(1-1) “Confirmation candidate extraction rule 1-1” is a chain of kana characters and single vowels, and a portion where the vowel part of the preceding kana character and the subsequent single vowel are the same is extracted as a confirmation candidate.
(1-2) “Confirmation candidate extraction rule 1-2”: This is a chain of kana characters and single vowels, and the part where the vowel part of the preceding kana character and the subsequent single vowel form “A” or “O” is confirmed. Extracted as a candidate.
(1-3) “Confirmation candidate extraction rule 1-3”: A kana character chain “n” is extracted as a confirmation candidate.

  A description will be given of the confirmation candidate extraction rule 1-1. Specifically, the rule can be broken down into the following. Each example is shown.

・ A chain of kana characters belonging to the dan and the character “a” (eg, substring “kaa” is a confirmation candidate for the long sound “car”).
・ A chain of kana characters belonging to the dan and the letter “I” (eg, substring “Shii” is a candidate for confirming the long sound “See”).
・ A chain of kana characters belonging to the dan and the character “u” (eg, substring “Tu” is a confirmation candidate for the long sound “Tu”).
・ A chain of kana characters belonging to Edan and the letter “e” (eg, substring “hee” is a confirmation candidate for the long sound “hee”).
・ A chain of kana characters belonging to the dan and the character “o” (eg, substring “ho” is a confirmation candidate for the long sound “ho”).

  Further, the confirmation candidate extraction rule 1-2 will be described. Specifically, the rule can be broken down into the following. Each example is shown.

・ A chain of kana characters belonging to Edan and the letter “I” (eg, substring “Bay” is a confirmation candidate for the long sound “Bay”).
・ A chain of kana characters belonging to the dan and the character “U” (eg, substring “PO” is a confirmation candidate for the long sound “PO”).

  The confirmation candidate extraction rule 1-3 has only one type. A corresponding example is shown.

-Chain of characters “Nun” (eg, confirmation candidate for the long sound “Nun”).

  In addition to these confirmation candidate extraction rules 1-1 to 1-3, when the preceding character is a small kana character, it can be handled in the same manner together with the preceding character. For example, an arbitrary character, a character “a”, and a character “a” are chained (for example, the substring “Care” is a confirmation candidate for the long sound “Care”). In addition, as a subsequent character, a small kana character can be treated as the same. For example, it is a chain of characters belonging to the row and the character “a” (for example, the substring “ka” is a confirmation candidate for the long sound “car”).

  As a result, a confirmation candidate for confirming whether or not to use long sound notation from the kana character string is extracted.

(2) Confirmation candidate extraction rule for “stuttering”
(2-1) “Confirmation candidate extraction rule 2-1” is a chain of two kana characters, the vowel part of the preceding kana character is “a” and the succeeding kana character is “ya” or “yu” or “yo”. Is extracted as a confirmation candidate.
(2-2) “Confirmation candidate extraction rule 2-2”: This is a chain of two kana characters, and a portion where the vowel part of the preceding kana character is “u” and the succeeding kana character is “wa” is extracted as a confirmation candidate.

  The confirmation candidate extraction rule 2-1 will be described. Specifically, the rule can be broken down into the following. Each example is shown.

・ A chain of kana characters belonging to the dan and the character “ya” (eg, substring “ki” is a confirmation candidate for the roar “ca”).
・ A chain of kana characters belonging to the dan and the character “yu” (eg, substring “Jiyu” is a confirmation candidate for the roar “ju”).
・ A chain of kana characters belonging to the dan and the character “yo” (example: the substring “piyo” is a confirmation candidate for the roar “piyo”).

  The confirmation candidate extraction rule 2-2 has only one type. A corresponding example is shown.

・ A chain of kana characters belonging to the dan and the character “wa” (eg, the substring “kwa” is a confirmation candidate for the roar “ku”).

  In addition to these rules, if the leading kana character is a vowel (e.g. `` Year '') or the leading kana character is a small kana character (e.g. May be excluded.

  As a result, a confirmation candidate for confirming whether or not to use stuttering is extracted from the kana character string.

(3) Confirmation candidate extraction rule for “sound prompt notation”
(3-1) “Confirmation candidate extraction rule 3-1”: a chain of three kana characters, with the central kana character “tsu” being a confirmation candidate.

  The confirmation candidate extraction rule 3-1 has only one type. A corresponding example is shown.

-Arbitrary kana characters, the character “tsu”, and the chain of arbitrary kana characters (eg, substring “Atta” is a confirmation candidate for the prompt sound “Atta”).

  In addition to the above confirmation candidate extraction rules, if the preceding kana character is a small kana character (eg "Ituta"), the following kana characters are "a", "i", "u", "e", "o", "wa", "wo" In any case of "" or "" (e.g., "Atsua"), a rule may be adopted that excludes the candidate and determines that the sound is not configured.

  As a result, a confirmation candidate for confirming whether or not the phonetic sound should be expressed from the kana character string is extracted.

(4) Confirmation candidate extraction rule for “sound / semi-sound notation”
(4-1) “Confirmation candidate extraction rule 4-1”..., Line, line, line, and kana characters in the line are extracted as confirmation candidates.

  The confirmation candidate extraction rule 4-1 will be described. Specifically, the rule can be broken down into the following. Each example is shown.

・ Kana characters in the ka, sa, and ta lines may be muddy (eg, substring “K” is a candidate for checking muddy sound “Ga”).
・ The kana character in the line may be a cloudy sound and a semi-sounding sound (for example, the substring “hi” is a confirmation candidate for the cloudy sound “bi” and the semi-sounding sound “pi”).

  Although not described in the above-described embodiment, a section in which a confirmation candidate has already been extracted by a confirmation candidate extraction rule for a preceding notation type is a confirmation candidate in a confirmation candidate extraction rule for a subsequent type notation. It is possible not to extract. In the above-described embodiment, four types of correct notation determination rules and confirmation candidate extraction rules are sequentially processed, but they may be processed in parallel. In addition, although not described in the above-described embodiment, a confirmation candidate extraction rule that extracts “he” as a confirmation candidate for “e” may be used. In the above-described embodiment, the correct notation determination rule and the confirmation candidate extraction rule are recorded in the storage device 5, and the correct notation determination rule 2 and the confirmation candidate extraction unit 3 are recorded in the storage device 5. In addition, the determination process of whether or not the normal notation is included and the extraction process of the confirmation candidate are performed using the confirmation candidate extraction rule, but the determination process and the extraction are performed in the correct notation determination unit 2 and the confirmation candidate extraction unit 3 You may make it incorporate the logic which performs a process.

  Next, the operation of the present embodiment will be described by taking as an example the case where the kana character string “Nyu Ueno Ishii Katsuf” is input from the kana character string input unit 1.

  When the kana character string is input from the kana character string input unit 1, the correct answer notation determination unit 2 executes the process shown in the flowchart of FIG. 2. In the case of this example, it is determined that the above-described correct notation determination rule 1 does not include a long sound notation (step S201), and the correct answer notation determination rule 2 determines that no stuttering notation is included (step S203). It is determined by the notation determination rule 3 that the prompt sound notation is not included (step S205), and it is determined by the correct answer notation determination rule 4 that the muddy / semi-turbid sound notation is not included (step S207). Therefore, the correct answer notation determination unit 2 passes “long sound notation”, “stuttering notation”, “promotion sound notation”, and “turbid sound / semi-turbid sound” to the confirmation candidate extracting unit 3 as notation types that may not include the correct answer notation. The kana character string is passed to the confirmation candidate extraction unit 3.

  As a result, the confirmation candidate extraction unit 3 executes the processing shown in the flowchart of FIG. First, the confirmation candidate extraction unit 3 pays attention to one of the four notation types passed from the correct notation determination unit 2, and according to the confirmation candidate extraction rule for the noticed notation type, the confirmation candidate is extracted from the kana character string. Extract (steps S31 and S33). Now, for example, if attention is paid to “long sound notation”, the confirmation candidate extraction unit 3 follows the above-described confirmation candidate extraction rules 1-1 to 1-3 as the confirmation candidates for long sound notation as the character strings “Yu” and “good”. As shown below, “SHI” is extracted, and the confirmation candidate, the extraction result (extraction position), and the correction candidate for the confirmation candidate are associated with each other and recorded in the storage device 5 or the like. The extraction result includes at least the extraction position of the confirmation candidate. Further, the correction candidate can be created by rewriting the second character of the confirmation candidate with a long sound character.

・ "Yu" (position = 2-3rd character, correction candidate = "you")
・ `` Good '' (position = 6-7 characters, correction candidate = `` E '')
・ `` Shii '' (position = 8th-9th characters, long sound candidate = `` sea '')

  Thereafter, the confirmation candidate extraction unit 3 pays attention to one of the not-noted notation types (for example, “stuttering notation”) (step S31). Then, according to the confirmation candidate extraction rules 2-1 and 2-2 for the stuttering notation, the character string “Nyu” is extracted as the confirmation candidate for the stuttering notation. The extraction position and the correction candidate for the confirmation candidate are associated and recorded in the storage device 5 or the like (step S33). The correction candidate can be created by rewriting the second character of the confirmation candidate into a stuttering notation.

  ・ "Nyu" (position = 1-2 characters, correction candidate = "nu")

  Next, the confirmation candidate extraction unit 3 pays attention to one of the not-noted notation types (for example, “prompting sound notation”) (step S31). Then, in accordance with the confirmation candidate extraction rule 3-1 for the sound notation, the character string “Katsuf” is extracted as the confirmation candidate for the sound notation, and the confirmation candidate, its extraction position, And a correction candidate for the confirmation candidate are recorded in association with the storage device 5 (step S33). Note that the correction candidate can be created by rewriting the second character of the confirmation candidate into a prompt sound expression.

・ "Katsufu" (position = 10-12 characters, correction candidate = "Cuff")

  Thereafter, the confirmation candidate extraction unit 3 pays attention to one of the not-notated notation types (for example, “turbid sound / semi-turbid sound”) (step S31). Then, according to Confirmation Candidate Extraction Rule 4-1 for muddy sound / semi-turbid sound, “shi”, “ka”, “tsu”, “fu” are extracted as confirmation candidates for muddy sound / semi-turbid sound notation. As information about the semi-voiced sound notation, a confirmation candidate, its extraction position, and a correction candidate for the confirmation candidate are recorded in association with the storage device 5 (step S33). In addition, the correction candidate of a muddy sound and a semi-turbid sound can be created by attaching a muddy point and a semi-turbid point to a confirmation candidate. However, semi-turbid sound correction candidates are created only when the confirmation candidate is a kana character in a line.

・ `` Shi '' (position = 8th character, muddy sound correction candidate = `` di '', no semi-turbid sound correction candidate)
・ "K" (position = 10th character, muddy sound correction candidate = "ga", no muddy sound correction candidate)
・ `` Tsu '' (position = 11th character, muddy sound correction candidate = `` ヅ '', no muddy sound correction candidate)
・ "F" (position = 12th character, muddy sound correction candidate = "bu", semi-turbid sound correction candidate = "p")

  When the confirmation candidate extraction unit 3 performs the above-described processing for all the notation types passed from the correct notation determination unit 2 (Yes in step S32), information on each notation type recorded in the storage device 5 ( (Confirmation candidate, extraction position, correction candidate) and the kana character string “Nyu Ueno Ishii Katsuf” are passed to the confirmation candidate presenting unit 4 (step S34).

  The confirmation candidate presentation unit 4 displays the confirmation screen shown in FIG. 4 on a display unit (not shown) according to the information passed from the confirmation candidate extraction unit 3.

  In the “kana character string” column, the kana character string “Niyu Ueno Ishii Katsu” inputted from the confirmation candidate extraction unit 3 is displayed. In the “long sound candidate” column, a horizontal line indicating the position of the confirmation candidate for the long sound notation and a correction candidate for the confirmation candidate are displayed. The horizontal line indicating the position of the confirmation candidate is displayed based on the extraction position of the confirmation candidate passed from the confirmation candidate extraction unit 3. For example, when the extraction position of the confirmation candidate is “2-3rd character”, a horizontal line is displayed at a position corresponding to the second and third characters “Yu” of the kana character string. In addition, correction candidates are displayed in association with the horizontal line.

  In the “stuttering candidate” column, a horizontal line indicating the position of the confirmation candidate for stuttering and a correction candidate for the confirmation candidate are displayed. In the “prompting sound candidate” column, a horizontal line indicating the position of the confirmation candidate for the prompt sound notation and a correction candidate for the confirmation candidate are displayed. In the “Muddy sound candidate” column, a horizontal line indicating the position of the confirmation candidate for the muddy sound notation and a correction candidate for the confirmation candidate are displayed. Correction candidates for the confirmation candidates are displayed in the “semi-voice sound candidate” field.

  The user sees the horizontal line indicating the confirmation candidate displayed on the confirmation screen shown in FIG. 4, the correction candidate, and the like, and the correct answer notation for the inputted kana character string “Nyuuei Ishii Katsuf” is “New NUC Cup” Recognize that Then, the input kana character string “Nyu Ueno Ishii Katsuf” and the correct notation “New NC Cup” are recorded in association with the dictionary for the speech synthesizer. Alternatively, the reading character string dictionary is created by directly rewriting the corresponding part of the database or dictionary to which the kana character string is inputted with the correct notation.

  Another example of the confirmation screen displayed by the confirmation candidate presenting unit 4 is shown in FIG. In this case, the confirmation candidate presenting unit 4 displays the kana character string passed from the confirmation candidate extraction unit 3 on the display unit, and displays an underline in the section where the confirmation candidate exists. In addition, when all the sections where the confirmation candidates of the other notation type exist are included in the section where the confirmation candidates of the other notation type exist, only the underline indicating the section where the confirmation candidate of the other notation type exists. indicate. When the mouse pointer is underlined and a section is selected, the confirmation candidate presenting unit 4 obtains and displays a set of correction candidates that can be taken in the section. A set of correction candidates that can be taken in the corresponding section is created based on the correction candidates related to the corresponding section among the correction candidates delivered from the confirmation candidate extraction unit 3.

  In the above-described embodiment, the input character string is a kana character string, but may be a character string mixed with kanji. For example, a patent publication is used as an input character string, a rule for determining whether or not a specific character “registered trademark” is included in the input character string as a correct notation determination rule, and the above specification as a confirmation candidate extraction rule It is also possible to adopt a rule for extracting a portion where the character “registered trademark” is missing from the input character string.

[Effect of the first embodiment]
According to the present embodiment, error detection can be prevented from being performed unnecessarily for an input kana character string having no or few errors. The reason is that when the input kana character string includes a specific character, the input kana character string is created in consideration of the specific character. The possibility of existing is considered small. Therefore, only when the correct notation determination unit 2 (determination unit) determines that a specific character is not included in the input kana character string, the confirmation candidate extraction unit 3 (detection unit) lacks the specific character. By detecting the existing portion from the input kana character string, it is possible to prevent the error detection from being performed unnecessarily for the input kana character string having no or few errors.

[Second embodiment of the present invention]
Next, a second embodiment of the present invention will be described. The present embodiment is characterized in that confirmation candidates to be presented to the user differ depending on the level of the dictionary for the speech synthesizer to be created. When creating a dictionary for a speech synthesizer, for example, when creating a dictionary that does not cause a mistake even if there is a sense of incongruity, or when creating a dictionary that does not cause a sense of incongruity or a mistake The correction part for the kana character string is different. In the present embodiment, the confirmation candidates to be presented to the user are made different depending on the level of the dictionary to be created, so that the amount of work when creating the dictionary depends on the level of the dictionary to be created. .

  The input character string error detection apparatus according to the present embodiment is realized by adding the following functions to the confirmation candidate extraction unit 3 and the confirmation candidate presentation unit 4 in the first embodiment shown in FIG. Is done. The confirmation candidate extraction unit 3 is added with a function of giving a confirmation priority extracted according to the kana character string to the confirmation candidate extracted from the input kana character string, or a task priority according to a sense of discomfort when listening to the confirmation candidate as speech. The confirmation candidate presenting unit 4 has a function of displaying the input kana character string in association with the confirmation candidates of the work priority or higher specified by the user among the confirmation candidates extracted by the confirmation candidate extracting unit 3 To do. Note that the input character string error detection apparatus according to the present embodiment can also be realized by a computer in the same manner as the input character string error detection apparatus according to the first embodiment.

  Next, the operation of the present embodiment will be described. Here, the operation will be described by taking as an example the case where the following three Kana character strings (a) to (c) are input.

(a) Nieuwe Ishii Katsuf
(b) Super computer
(c) Cutting sheet

  The correct notation determination unit 2 performs a determination process on each kana character string (a) to (c) using the correct answer notation determination rule, a kana character string including a notation type for which there is no correct notation, and a correct answer notation. A notation type that does not exist is passed to the confirmation candidate extraction unit 3 (see FIG. 2). In the case of this example, the following information is passed to the confirmation candidate presenting unit 4.

・ A set of kana character string (a) and notation type “long sound notation, stuttering notation, prompt sound notation, muddy / semi-notated notation”.
・ A set of kana character string (b) and notation type “long sound notation, prompt sound notation”.
-A combination of the Kana character string (c) and the notation type “stuttering notation, prompting notation”.

  Each time a pair of a kana character string and a notation type is input, the confirmation candidate extraction unit 3 extracts a confirmation candidate using a confirmation candidate extraction rule (steps S31 and S33 in FIG. 3). In step S33, in the present embodiment, the processing priority of each extracted confirmation candidate is determined according to the notation type, and is also recorded in the storage device 5 in association with the confirmation candidate. Here, the work priority is defined as the lower priority, the higher the priority.When listening as a voice, the work priority of the confirmation candidate for the stuttering notation or muddy / semi-voice notation is likely to be mistaken. The priority of the confirmation candidate for the prompt notation that is easy to recall the correct answer but with a sense of discomfort is “2”, and the priority of the confirmation candidate for the long notation with a small discomfort is “3”. To do. Furthermore, in step S33, in this embodiment, the priority of the kana character string to be processed is determined, and recording processing is also performed in the storage device 5. The work priority of the kana character string is the highest work priority among the work priorities of each confirmation candidate extracted from the kana character string.

  In this example, the work priorities of the kana character strings (a) to (c) are as follows.

・ Kana character string (a)… There is a confirmation candidate for the kana character string (a) whose notation type is “long sound notation, stuttering notation, prompt sound notation, muddy / semi-notated notation”. The work priority is set to “1” as the highest work priority among the work priorities.
・ Kana character string (b) ... Since the kana character string (b) has only the confirmation candidate whose notation type is “long sound notation”, the work priority of the kana character string (b) is the confirmation candidate for the long sound notation. The work priority is “3”.
・ Kana character string (c): Since there is only a confirmation candidate whose notation type is “prompting notation” in the kana character string (c), the work priority of the kana character string (c) is the confirmation candidate of the prompt sound notation. The work priority is “2”.

  Information regarding these work priorities is passed to the confirmation candidate presenting unit 4 together with a kana character string or the like when the kana character string or the like is handed to the confirmation candidate presenting unit 4 in step S34.

  The confirmation candidate presenting unit 4 displays a confirmation screen as shown in FIG. 6 according to the information passed from the confirmation candidate extracting unit 3. This confirmation screen is provided with an input field for work priority, and the user can input the work priority using a keyboard (not shown). The confirmation candidate presenting unit 4 displays a kana character string and a confirmation candidate having a priority higher than the work priority input from the work priority input field when displaying a kana character string or a confirmation candidate on the display unit. Display only. For example, when the work priority “2” is input from the input field, the confirmation candidate presenting unit 4 has the kana character string (a) with the work priority “1” and the work priority “2”. The kana character string (c) is displayed, but the kana character string (b) having the work priority “3” is not displayed. In other words, the input kana character string for which there is no confirmation candidate higher than the work priority specified by the user among the confirmation candidates extracted by the confirmation candidate extraction unit 3 is not presented to the user. In addition, for kana character strings (a) and (c), confirmation candidates with work priority “1” or “2” are displayed, but confirmation candidates for long sound notation with work priority “3” are displayed. Not done. In FIG. 6, confirmation candidates with work priority “1” are indicated by solid lines, and confirmation candidates with work priority “2” are indicated by dotted lines. Can be recognized.

  In addition, in FIG. 6, the candidate types are displayed together with each candidate type, such as “Muddy sound?”, “Stuttering sound?”, And “Sounding sound?”. It is also possible to generate and display correct notation that can be guessed by the candidate extraction unit 3. Also, here, work priorities were assigned for each type of candidate, but more finely, for example, the ease of hearing mistakes between each rule of long sound candidates is quantified by the appearance frequency of bigrams etc. in the corpus, It can also be assigned to work priority. Further, here, the work priority of the confirmation candidate is determined based on the notation type, but it may be determined based on the confirmation candidate extraction rule from which the confirmation candidate is extracted. For example, the priority “1” is given to the confirmation candidate extracted by the confirmation candidate extraction rule 1-1 described in the first embodiment, and the priority “3” is given to the part extracted by the confirmation candidate extraction rule 1-2. "Is set. As a result, it is possible to take more detailed measures for listening errors and discomfort.

[Effect of the second embodiment]
According to the present embodiment, the confirmation candidates presented to the user can be made different depending on the level of the dictionary for the speech synthesizer to be created. The reason is that the confirmation candidate extracted from the kana character string is given a degree of listening mistake when listening to the confirmation candidate as speech, or a work priority according to the sense of incongruity, and the confirmation candidate is presented to the user. This is because a confirmation candidate having a work priority or higher designated by the user is presented.

[Third embodiment of the present invention]
Next, a third embodiment of the present invention will be described. This embodiment is characterized in that the user can easily register the corrected kana character string in the dictionary for the speech synthesizer.

  FIG. 7 is a block diagram showing a configuration example of the input character string error detection apparatus according to the present embodiment. Differences from the input character string error detection apparatus in the first embodiment shown in FIG. It is the point provided with the confirmation candidate presentation part 4a instead of the confirmation candidate presentation part 4, the point provided with the dictionary registration part 6, and the point provided with the dictionary 7.

  In addition to the functions provided by the confirmation candidate presenting unit 4, the confirmation candidate presenting unit 4a corrects the kana character string displayed on the confirmation screen with the correction candidate selected by the user or according to the editing work performed by the user. And has a function of creating a corrected kana character string. The dictionary registration unit 6 has a function of registering the corrected kana character string created by the confirmation candidate presenting unit 4 a in the dictionary 7.

  Note that the input character string error detection apparatus of the present embodiment can also be realized by a computer, as in the first embodiment.

  Next, the operation of this embodiment will be described.

  Assume that the confirmation candidate presenting unit 4a displays a confirmation screen as shown in FIG. 4 on the display unit. When the user selects a correction candidate and directly edits the input kana character string, the confirmation candidate presenting unit 4a replaces the confirmation candidate with the selected correction candidate in the kana character string, etc. “NC Cup” is created and passed to the dictionary registration unit 6. The dictionary registration unit 6 registers the corrected kana character string “New NC Cup” and the input kana character string “Nyu Ueno Ishii Katsuf” in the dictionary 7 in association with each other. Further, instead of the dictionary 7, the corresponding part of the database or dictionary in which the input kana character string is stored may be rewritten with the corrected kana character string.

[Effect of the third embodiment]
According to the present embodiment, a dictionary for a speech synthesizer can be easily created. The reason is that it includes a confirmation candidate presenting unit 4a that presents a correction candidate to the user and generates a corrected kana character string obtained by correcting the input kana character string using the presentation candidate selected by the user.

  The present invention is preferably used when creating a dictionary for a speech synthesizer that receives a kana character string as an input.

DESCRIPTION OF SYMBOLS 1 ... Kana character string input part 2 ... Correct notation determination part 3 ... Confirmation candidate extraction part 4, 4a ... Confirmation candidate presentation part 5 ... Memory | storage device 51 ... Correct notation determination rule storage Unit 52 ... Confirmation candidate extraction rule storage unit 6 ... Dictionary registration unit 7 ... Dictionary

Claims (9)

Determining means for determining whether or not a specific character is included in the input character string;
Only when the determination unit determines that the specific character is not included in the input character string, the detection unit includes a detection unit that detects a location where the specific character is missing from the input character string. An input character string error detecting device characterized by that. The input character string error detection device according to claim 1,
The input character string error detecting device, wherein the specific character is a character representing a long sound. The input character string error detection device according to claim 1,
The input character string error detection device, wherein the specific character is a character representing stuttering. The input character string error detection device according to claim 1,
The input character string error detecting device, wherein the specific character is a character representing a prompt sound. The input character string error detection device according to claim 1,
The input character string error detecting device, wherein the specific character is a character representing a muddy sound and a semi-voiced sound. In the input character string error detection device according to any one of claims 1 to 5,
Presenting means for presenting the input character string and a correction candidate for the portion detected by the detecting means to the user and generating a corrected input character string in which the input character string is corrected by the correction candidate selected by the user When,
An input character string error detecting apparatus comprising: a registration unit that registers the input character string and the corrected input character string in association with each other. Determining means for determining whether or not the specific character is included in the input character string for each of the predetermined specific characters;
Only the specific characters that are determined not to be included in the input character string by the determining means, the location where the specific character is missing is detected from the input character string, and for each detected location, Detecting means for determining the degree of misunderstanding when listening to a part as sound, or the work priority according to the uncomfortable feeling;
A presentation unit that associates the input character string with a portion having a work priority that is equal to or higher than the work priority specified by the user among the portions detected by the detection unit; Characteristic input character string error detection device. An input character string error detection method executed by a computer including a determination unit and a detection unit,
The determination means determines whether or not the input character string includes a specific character;
Only when the determination means determines that the specific character is not included in the input character string, the detection means detects a location where the specific character is missing from the input character string. An input character string error detection method characterized by the above. A program for causing a computer to function as an input character string error detection device,
The computer,
Determination means for determining whether or not a specific character is included in the input character string;
Only when the determination unit determines that the specific character is not included in the input character string, the input character string functions as a detection unit that detects a portion where the specific character is missing from the input character string. Program for.

Images