JPH0844724A - Handwriting input character processor - Google Patents

Abstract

(57) [Summary] [Purpose] It is necessary to input a correct sentence by issuing a warning that there is a possibility of typographical errors due to erroneous recognition / input by handwriting input and generating correction candidates. To aim. [Constitution] A candidate is generated by a phrase segmenting means 2 for dividing an input sentence into phrases, a collating means 4 for collating them with a kana-kanji conversion dictionary 5, and a candidate generating means 12 for generating a candidate when they do not match. , By displaying to the user by the display means 6, and by making corrections,
A correct sentence can be input by issuing a warning that there is a possibility of typographical error due to erroneous recognition and erroneous input by handwriting input and by generating correction candidates.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text processing device in a device having handwriting input means and an OCR (Optical Character Reader), such as a word processor, a computer, an electronic notebook and a personal digital assistant.

[0002]

2. Description of the Related Art FIG. 7 is an example of a portable information terminal, and FIG. 8 is a screen diagram showing an example of handwriting input processing. There is a handwriting entry frame C on the screen, and when a character is handwritten and input using the input pen B (FIG. 7) in the frame, the character is pattern-recognized, and as a result of the recognition, the obtained character is input line D. Is displayed. By cyclically using a plurality of (four in this case) entry frames, it is possible to continuously write characters without interrupting thinking.

However, when recognizing an input character as a pattern, it may be recognized as a character different from the character requested by the input person. A conventional technique for such character recognition is disclosed in Japanese Patent Laid-Open No. 59-188783.
The recognition rate of the read word was raised by comparing the optically read word with the dictionary which ranked in order of high probability. Japanese Patent Laid-Open No. 61-131809 makes it easy to correct an erroneously determined character by handwriting input or a character that is difficult to determine by displaying a similar character from a similar character table and allowing the user to select from the similar characters. , Was to shorten the correction time.

In order to make an erroneously recognized character into a target character, if the wrong character on the input line is touched with a pen, a candidate for character recognition for that character is displayed in the candidate character display frame E. If you select and touch it, it will be replaced with the correct character.

[0005]

[Problems to be Solved by the Invention]

[Problem in Claim 1] When a character is directly input by handwriting, especially in the case of Kanji, if the input character is a recognizable character, even if it is a typographical error, it is recognized and input. To be done. In the case of inputting from a conventional keyboard, in most cases, the input is performed primarily in hiragana using the keyboard and then input using the kana-kanji conversion function. As a problem of kana-to-kanji conversion that converts hiragana to kanji, there were mistakes due to homonyms such as "collation" and "title", but it was due to the mistyping of hiragana, for example, "shogou" If you try to enter "Shogoi" and convert it to Kana-Kanji, it will be converted to "noon doctor",
Since it is displayed, it can be clearly seen that there is an input error.

However, in the case of handwriting input, the input person may think that the input character is correct and the input character may be erroneously recognized, causing a problem. With the conventional keyboard input method, an expert who is familiar with keyboards can find most of the mistakes such as erroneous conversions and typos at the time of input because they can input while looking at the screen without looking at the keyboard. However, in the handwriting input method, the user inputs by looking at the frame for input, but the recognition result is often output on the main screen, and there is a time lag from input to recognition, so multiple input frames are required. In order to perform successive conversion, the user's line of sight naturally concentrates on the entry frame, which makes it difficult to confirm mistakes due to erroneous recognition at the time of input.

Also, some handwriting recognition systems collate the stroke order (stroke order) of Chinese characters, so if you remember the wrong stroke order and input it as it is, it may be recognized as a different Chinese character. There is also. Further, in a system that uses the number of strokes for recognition, if a portion that must be accurately written in two strokes is written in a single stroke, it may not be possible to accurately recognize it.
Also, if a word that does not consist of a word is recognized, it can be input. For example, also intend to enter "reason", sometimes be input remains had been recognized as "management field". In this case as well, since the person intends to input "reason", there is a high possibility that the user does not notice when inputting. In addition, there are relatively many people who mistakenly remember the kanji, for example, the word " wai " for "perfect" as a " wall ". After you have entered in hiragana as "perfect" in the conventional kana-to-kanji conversion system, [conversion] always to convert, such as in the kanji key out the Chinese characters of "perfect", is converted to a Chinese character of "complete wall" It never happened. However, in the pen input method, even a word that does not hold is displayed. That will be output as it is, even if the "his acting was a complete wall" had entered and, even though it appears in the "complete wall" screen, because the person in question do not think it's a mistake, mistake The user does not notice.

Some characters have almost the same type of kanji and katakana. For example, the Chinese character “evening” in the word “evening” and the katakana “ta” are almost the same type. Some pen input devices recognize correctly if the Chinese character "Yu" is written in a size of 70% or more of the entire frame and the Katakana "Ta" is written in a size of 70% or less. Such rules are not well known in general, and there are many cases where you do not notice even if you make a typo.
In other words, there was no means to confirm erroneous input.

[Problem in Claim 2] Regarding an input error that occurs in the above problem, it is very troublesome to input again, and there is a possibility that even if it is input again, it may be erroneously recognized again. . In addition, if the user cannot remember the correct stroke order or the correct kanji, or if the kanji and katakana are almost the same type as in the above example, input is not possible, so change to another input method, for example, romaji-kana conversion mode. After that, it is necessary to input the hiragana characters in Roman letters and convert them into Kanji. That is, a method of correcting without re-inputting is desirable.

[0010]

[Means for Solving the Problems]

[Means for Resolving Claims] Handwriting input means for inputting a sentence by handwriting input means, clause delimiter for dividing a sentence input by the input means into clauses, and notation of clauses delimited by the clause delimiter is a pseudonym. A collation collation unit that collates with the notation of the Kanji conversion dictionary, and as a result, if there is no collation in the Kana-Kanji conversion dictionary, the collation unit outputs the fact and points out to the user.

[Solution for Claim 2] With respect to the word pointed out by the output means, a candidate generating means for generating a candidate word from a word similar to each character forming the word, and the candidate generating means. Display means for displaying the selected candidates, and replacement means for replacing the misrecognized word by selecting the word displayed by the display means.

[Means for Resolving Claim 3] A function for registering a word pointed out by the output means in a user dictionary by the user's selection is provided.

[Means for Solving Claims] In the collating means, collation is performed not only with a dictionary prepared in advance in the system but also with a user dictionary.

[0014]

[Action]

[Operation of Claim 1] With respect to a sentence input by direct input of handwriting, one phrase is judged by the phrase segmenting means and cut out for each word. The word segmentation means is used so that each segmented segment can be collated with the kana-kanji conversion dictionary. With this word segmentation means, with respect to words that are not used, particles are excluded and prefixes and suffixes are deleted. For a word that has a conjugation, generate its final form. Each word obtained by the word slicing means is collated with the notation of the kana-kanji conversion dictionary by the collating means.
Regarding this fluctuation of the sending kana,
After matching, the word is judged to match. For words that do not match here, the word of the original phrase is issued by means of reverse display or the like.

[Operation of Claim 2] With respect to the word judged to be inconsistent by the matching means of Claim 1, the similar word dictionary is searched. This dictionary of similar words is a collection of those with similar glyphs and those that are prone to typographical errors, and mainly consists of kanji. All the combinations of the word drawn from the similar word dictionary and the original unmatched word are generated by the candidate generation means to generate the candidate word. This word is collated with the mark in the Kana-Kanji conversion dictionary by the collating means, and if there is a match, it is displayed by the display means. By selecting the one displayed by the display means, the selected one is replaced and the correction is performed.

[Operation of Claim 3] Even when the matching means of Claim 1 determines that they do not match, it may be possible to input a word that is not actually registered in the dictionary, such as a proper noun. In that case, even if a warning is issued by the present invention, it is ignored, but in that case, by registering the word in the user dictionary, when the same word is input next time, it is checked with the user dictionary. , No unnecessary warning is issued.

[Operation of Claim 4] In the matching means of Claim 1, not only the dictionary prepared in advance in the system, but also the user dictionary is the target of matching. by this,
According to the third aspect, it is possible to collate with a word that is automatically registered, and an unnecessary warning is not issued.

[0018]

【Example】

[Claims 1 and 3 and 4] Based on FIG.
The present invention will be described in detail. The present invention is not limited to this embodiment. FIG. 2 is an example of a mobile information terminal that inputs data to the display unit A with the pen B. A liquid crystal panel (LCD) or the like is used for the display unit A, and an input sentence or the like is displayed. Pen B is an input pen. As a pen input method, there is a switch at the tip of the pen, which detects the position based on the electric charge generated between it and the electrode provided on the display unit, or a pressure-sensitive tablet,
There is something that detects the position by magnetism from a pen.

In the pen-based input method, keys are displayed on the screen such as "Japanese syllabary mode" (Fig. 9) and "Typewriter mode" (Fig. 10), and the key portion is touched with the pen. , There is a mode to input.

On the other hand, the "handwriting recognition mode" is shown in FIG.
Characters that you want to input directly into the entry frame are input by handwriting and recognized. There are usually a plurality of entry boxes C1 to C4 (four in FIG. 8), and characters to be entered may be written in any box. It doesn't matter where you write it, it starts recognizing characters in the order you write them. The recognized characters are displayed on the input line D and disappear from the entry frame C, so that the characters can be sequentially input.

As shown in FIG. 3, "Mr. Shida was walking with a big bag at Osaka station. ”In the“ handwriting recognition mode ”. In this case, since all characters in this input sentence are within the first and second levels of JIS, all the characters are recognized and the result is displayed on the input line D. This input sentence "Mr. Shida was walking with a big bag at Osaka station. ]
As shown in Fig. 3 [2], the phrase segmentation means 2 "At Osaka Station / Mr. Shida / Large / Luggage / Holding /
I was walking. It is divided into 6 clauses.

The phrase segmentation means performed here does not perform processing only by collating with the word dictionary, because the sentence that may be erroneously input or erroneously recognized is the target sentence. A method of recognizing a part-of-speech is known by the type of character (kanji, katakana, hiragana, symbol), particle, and inflection ending. In this method, basically, the kanji / katakana part is not collated with the dictionary, and only the hiragana part is collated with the dictionary. This dictionary is provided separately from the kana-kanji conversion dictionary as the phrase segmentation means dictionary 14. Since the dictionary is only hiragana, the size of the dictionary is very small. The part of speech of the kanji part is recognized from the information of the hiragana part. For example, if there is an input sentence "Use this method.", The part "" is Hiragana, so it is collated with the dictionary to obtain part-of-speech information. The next "method" portion is simply recognized as "two-character kanji string" without collating the dictionary. After that, "o" is looked up in the dictionary and recognized as "particle". From this, it can be seen that the "two-character kanji string" connected before is a noun. Similarly, it can be seen that “use” + “do” are sa verbs. By using this method, even if there is an erroneous input of kanji, katakana, etc., segmentation and part-of-speech recognition can be accurately performed.

The input sentence is subjected to phrase segmentation processing, and if there is a particle for each of the divided bunsetsu, the word segmentation means 3 takes a particle by particle processing (FIG. 2S3). For example, the final form is generated by the final form processing. Particle processing is the processing of taking a particle to extract only a noun and match it with a dictionary when a particle such as "ha" or "ga" is attached after an independent word that is not used, such as a noun. is there. Termination processing determines the type of usage (five-step utilization, upper-first-step utilization, etc.) and inflectional shape (coupling, concatenation, etc.) based on the word stem and inflection ending, and terminates it to match it with the dictionary entry word. This is the process of generating a shape. Since the part-of-speech information is obtained by the phrase segmentation means described above, the result is used to perform particle processing and final form processing (FIG. 2S).
4).

If there is a suffix or a prefix, they are also removed for dictionary matching (FIG. 2S5), and a notation that can be matched with the notation of the Kana-Kanji conversion dictionary is generated (FIG. 3).
[3]). Regarding this, there are cases in which the prefix / suffix is recognized as part-of-speech information by the phrase segmentation means,
You may have a prefix / prefix dictionary externally and collate with that dictionary.

In the phrase "at Osaka Station" in the example sentence of FIG.
Since "de" is a particle, the above particle processing is performed,
It will be "Osaka Station". Since this "Osaka station" is a noun, it is not used and there is no prefix or suffix, so "Osaka station" is the target word for the Kana-Kanji conversion dictionary. Similarly, "ga" in "Shita-san ga" also corresponds to a particle, and as a result of particle processing, it becomes a noun "Mr. Shita". However, since "Mr." is a suffix, "Shida" is processed by prefix / suffix processing. Is the matching target word. The next clause "big" has no particle. However, because it is an adjective, there is a practical use, so the final form processing is performed, and "big"
Becomes There is no prefix or suffix, and the processing target word is "large". Similarly, "luggage" is the particle "wo"
Is removed, and the processing target words are "luggage", "take",
Since both "walking" were verbs, the final form processing is performed, and the final forms "having" and "walking" are to be collated with the kana-kanji conversion dictionary.

The generated notation is collated with the notation of the Kana-Kanji conversion dictionary 5 by the collating means 4 (FIG. 2S6) (see FIG. 2).
S7), if there is a matching word, nothing is done (S8 in FIG. 2). Doing nothing means that the process of the present invention is not performed, but the process is performed normally. If there is no matching word, FIG. 3 [4] shows the word of the original segment that has been cut out by the display unit 6.
To reversely display (FIG. 3 [5], FIG. 2S9). If you compare the 6 words "Osaka Station", "Shita", "Large", "Luggage", and "Walk" with the Kana-Kanji conversion dictionary, there is no match for "Shita", and all the rest are match words. (Fig. 3 [4]).

"Shita", which had no matching words in the original input sentence, is displayed in reverse video as shown in Fig. 3 [5] and a warning to the effect that "it may be wrong" Give out. In this way, when issuing a warning for a word that is not used, put it in the part where the particle, the prefix, and the suffix are taken. In addition, when a word that has been used becomes a warning target, the one phrase that has been used in the original sentence is highlighted (for example, “have”).
If it becomes the target, "Hold" is highlighted.

Similarly, as shown in FIG. 4, "He performed perfectly. Because "if you've entered a" complete wall "when you enter the, the" complete wall "does not exist in the kana-to-kanji conversion dictionary,
A warning will be displayed.

In the collation processing with the Kana-Kanji conversion dictionary, the collation with the user dictionary created by the user is naturally performed. Also, if you issue a warning once and adopt (confirm) without changing,
In other words, a warning was issued for the word "Shita", but when adopted (determined), the word is automatically registered in the user dictionary by selection, and meaningless warning is suppressed from the second time. It is also possible to use not only this system but also kana-kanji conversion by quickly registering words such as proper nouns (especially person names, place names) that are often not found in ordinary system dictionaries in the user dictionary. Is possible.

[Embodiment of claim 2] According to the embodiment of claim 1, "Shibata" is displayed in contrast to "Shita" highlighted.
If the user notices that it is wrong, you can enter "Shiba" instead of "Purple" at that stage, but when you try to make a correction candidate word, you can use the inverted "Shita" with a pen By touching, a window for candidate words is opened, and the candidate words are displayed therein (FIG. 5 [9]).

If there is a correct answer in these candidates, touching the word with a pen replaces the highlighted word (warned word) with the selected word. If it is not among the candidates, touch "Cancel" in the window to cancel it and the window will disappear. When there are many candidates, by touching △ or ▽, the words in the window are scrolled up and down and all candidates can be viewed.

The candidate words displayed in this window are generated as follows. The non-matching word processing means 7 is applied to "Shita" which is reversed as a post-mismatch in FIG. 3 [5].
(Fig. 6 S11), 2 for each character "purple" and "field"
It is divided into two, and the similar word dictionary 10 is searched for each (S12 in FIG. 6), and each candidate is retrieved ([6] and [7] in FIG. 5). In the similar word dictionary, words with similar glyphs and easily erroneous words are collected as a group. For example, as shown in [6] and [7] of FIG.
For the word "purple", it is "frame, dye, pear, mulberry, tsuki, beam,"
10 words of "proposed, accumulating, ligating, prosperous" to "Ta"
4 words of "Yu, Shin, Ko" are registered as a group. The word generated from the similar word dictionary and the original word are generated by the candidate generating means 12 as words in all combinations (FIG. 5 [8], FIG. 6S13). In case of "Shita", Fig. 5
As shown in [8], (Shita, Shiyu, ...
54 words of "prosperity" (11 words including "purple" and "purple"
Characters, “Ta” and 5 similar words of “Ta”, 11 × 5 = 55, minus one of “Shita” are generated.

The generated word is collated with the notation of the kana-kanji conversion dictionary 5 by the collating means 4 (FIG. 6S14), and if there is a match (FIG. 6S15), it is displayed by the display means 6 as a candidate (FIG. 6S15). In S16 of 6, if there is no match, the display of "not applicable" is performed by the same means. When all 54 words generated from the words drawn with "purple" and "ta" from the similar word dictionary are compared with the kana-kanji conversion notation,
As shown in the result of FIG. 5 [9], “Shibata, Nashida,
The three words "Kuwata" match. If you touch "Shida" highlighted in the original input sentence with a pen, these three words will be displayed in the window as correction candidates. If there is a word intended by the user in the displayed words, touching and selecting it (FIG. 6S17) replaces the unmatched word with the selected word (FIG. 6S18). Touch "Shibata" here (Fig. 5 [9]), and select "Shibata".
Is replaced by the highlighted "Shita" and the correction ends. In other words, "Mr. Shida was walking with a big bag at Osaka station. 』Is" Mr. Shibata was walking with a big bag at Osaka station. ] Is corrected.

[0034] Similarly, "He was a complete wall of acting. In case you accidentally type "it is also a candidate word for the" complete wall "," perfect "is generated and displayed as a candidate.

[0035]

【The invention's effect】

[Effect of claim 1] There has been a problem that, conventionally, there is a problem that a handwritten character can be directly input even if it is a typographical error and a recognizable character can be input. In such a case, the user is assuming that it is correct and is inputting,
In many cases, the error was not noticed until it was pointed out from the outside, but by using a method such as word matching with the Kana-Kanji conversion dictionary according to the present invention, by displaying a warning such as display reversal, at the sentence creation stage. You are more likely to notice it.
Further, it becomes possible to support registration when a word intentionally used in the Kana-Kanji conversion dictionary is not used.

[Effect of Claim 2] Even if the possibility of a typographical error is pointed out in Claim 1, even if correction cannot be made because the correct word cannot be remembered, according to the present Claim, correction candidates are displayed. Therefore, it is possible to easily check and shorten the correction time without having to look up a dictionary or the like.

[Effect of claim 3] Depending on proper nouns such as a person's name and a place name, it may not be collated with the kana-kanji conversion dictionary, and it may be pointed out that even a word to be actually input may be a typographical error. However, this is avoided by registering in the user dictionary.

[Effect of Claim 4] Similar to the effect of Claim 3, even if a word that is not in the Kana-Kanji conversion dictionary is registered in the user dictionary in advance, the present invention can be effectively used. Become.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration in an exemplary embodiment of the present invention.

FIG. 2 is a diagram showing an example of a flowchart showing a process of an embodiment of the present invention.

FIG. 3 is a diagram showing an example of specific input data according to an embodiment of the present invention.

FIG. 4 is a diagram showing an example of specific input data according to an embodiment of the present invention.

5 is a diagram showing the data of one embodiment of the present invention in a form associated with the input data of FIG.

FIG. 6 is a diagram showing an example of a flowchart showing the process of one embodiment of the present invention.

FIG. 7 is a diagram showing an example of a portable information terminal having a pen input means.

FIG. 8 is a screen diagram showing an example of handwriting input processing.

FIG. 9 is a screen diagram in “50-sound mode” of the pen input method.

FIG. 10 is a screen diagram in “typewriter mode” of the pen input method.

[Explanation of symbols] A liquid crystal display section B handwriting input pen C entry frame D input line 1 document input means 2 phrase segmentation means 3 word cutout means 4 collation means 5 kana-kanji conversion dictionary 6 display means 7 wording processing after disagreement Means 10 Similar Post-Dictionary 12 Candidate Generation Means 13 User Dictionary 14 Phrase-Spacing Means Dictionary

Claims (4)

[Claims] 1. A handwriting input means for inputting a sentence by handwriting input means, and a sentence input by said input means,
If there is no matcher in the Kana-Kanji conversion dictionary, a matcher that separates the phrase by the phrase-breaker, a matching unit that matches the notation of the phrase separated by the phrase-breaker with the notation of the Kana-Kanji conversion dictionary, A handwritten input character processing device having output means for outputting that effect. 2. A candidate generation unit that generates a candidate word from a word similar to each character forming the word with respect to the word pointed out by the output unit, and a candidate generated by the candidate generation unit is displayed. 2. The handwritten input character processing device according to claim 1, further comprising a display unit and a replacement unit that replaces the indicated word by selecting the candidate displayed by the display unit. 3. The handwritten input character processing apparatus according to claim 1, further comprising means for registering a word pointed out by said output means in a user dictionary by selection. 4. The handwritten input character processing device according to claim 1, wherein the collating means performs collation with a user dictionary.