JP3622841B2 - Kana-kanji conversion device and kana-kanji conversion method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a kana-kanji conversion device and a kana-kanji conversion method, and more particularly to a kana-kanji conversion device and a kana-kanji conversion method used for inputting Japanese into a computer system.
[0002]
[Prior art]
A personal computer that inputs a kana character string corresponding to the kanji character string you want to input from the keyboard as a device for inputting Japanese character strings, and converts the kana character string to a kana kanji character string in response to the input of the kanji conversion key. Conventionally, kana-kanji conversion devices such as word processors are known. In order to convert a kana character string input to this apparatus into a kana-kanji character string, a specific one or more keys for kanji conversion are pressed in combination to display kana-kanji character string candidates. It is also possible to display candidates in succession. In this case, when calling the immediately preceding candidate, it is possible to input a sentence by pressing the previous candidate key or the like. A procedure similar to the above is performed when a kana character string is converted into a katakana character string or when converted into a Roman character string.
[0003]
In a kana-kanji conversion apparatus that performs kana-kanji conversion on an input character string, conversion accuracy is improved by referring to morpheme and analysis information of each frame in order to determine a kanji corresponding to the character string. Here, a morpheme is the smallest linguistic unit that has a meaning consisting of one or more phonemes. In the morpheme analysis, a morpheme break in a character string is recognized and a morpheme part of speech is recognized. .
[0004]
In case frame analysis, a semantic connection relationship between words included in a character string is expressed by the concept of “case grammar”.
[0005]
Conventional Kana-Kanji conversion devices often use morphological analysis based on the part of speech as described above and case frame analysis.
[0006]
[Problems to be solved by the invention]
However, in the conventional Kana-Kanji conversion method described above, there is a language phenomenon that is difficult to convert accurately. In addition, in the case of the traditional Kana-Kanji conversion method, it is inexplicable for users who do not use such special terms if the expressions corresponding to the character strings are widely accepted so that special vocabulary and expressions can be converted as easily as possible. There is a problem that side effects such as incorrect word continuation and erroneous conversion / learning of synonym synonyms with a weak co-occurrence relationship occur.
[0007]
On the other hand, trigrams are known as language models used in speech recognition systems. This trigram has the merit of being able to deal with morphological analysis based on parts of speech and language phenomena that cannot be accurately converted with case frames.
[0008]
However, since the calculation amount of the trigram is extremely large, there is a problem that it is very difficult to apply the trigram as it is to a system that requires high speed such as kana-kanji conversion.
[0009]
The present invention has been made in view of such problems, and an object of the present invention is to provide a kana-kanji conversion apparatus and a kana-kanji conversion method capable of suppressing side effects such as erroneous conversion that have occurred in conventional kana-kanji conversion. It is to provide.
[0010]
[Means for Solving the Problems]
In order to achieve the above object, the present invention provides a dictionary that describes characters that are candidates for kana-kanji conversion and the priority of the characters, and the priority of connection of two parts of speech. A connection table of parts of speech to be described, a case frame dictionary that describes the semantic relationship of words contained in a character string to other words, and N (N ≧ 3) words in a text corpus containing actual text A storage device that stores the N-gram data describing the probability of occurrence of the chain, and the dictionary, the part-of-speech connection table, the case frame dictionary, and the N-gram data stored in the storage device. A kana-kanji conversion device comprising a data processing device for converting a kana character string input from a device into a kana-kanji character string, wherein the data processing device matches the input kana character string Based on the dictionary priority means for extracting characters from the dictionary, the priority of the characters, and the priority of connection of the parts of speech, the priority of the characters and the priority of connection of the parts of speech Based on the case frame dictionary and the first candidate based on the morpheme analysis unit that combines the characters extracted by the subtracting unit to create a first candidate for a kana-kanji character string whose possibility is greater than a predetermined criterion. Based on the case frame analysis means for rearranging the priorities of the candidates for the included Kana-Kanji character strings to create a second candidate, and the probability of occurrence of N chains of the word, the second candidate N-gram analyzing means for rearranging the priorities of the candidates of the included kana-kanji character strings.
[0011]
The invention according to claim 2 is the kana-kanji conversion device according to claim 1, wherein the storage device stores a roman character kana conversion table describing a kana character corresponding to a roman character, and the data processing device includes: Romaji-kana conversion means for converting a Roman character input from the input device into the kana character string based on a Roman-kana conversion table, and the dictionary lookup means converts the Roman character to the kana character string converted by the Roman character-kana conversion means. The matching character is extracted from the dictionary.
[0012]
Further, the invention according to claim 3 is a dictionary that describes characters that are candidates for kana-kanji conversion stored in a storage device and priority of the characters, and connection of parts of speech that describes the priority of connection of two parts of speech. N (N ≧ 3) word chains appear in tables, case frame dictionaries that describe the semantic relationship of words contained in a string to other words, and text corpora containing actual text A kana-kanji conversion method for converting a kana character string input from an input device into a kana-kanji character string based on N-gram data describing a probability, wherein the character that matches the input kana character string is retrieved from the dictionary. a dictionary step for extracting, the priority of the text, based on the priority of the connection of the parts of speech, a combination of the character extracted in the dictionary step, possibly of a predetermined A morphological analysis step of creating a first candidate for the kana-kanji character string is quasi more, based on the rated frame dictionary, rearrange the priority order of the candidate of the kana-kanji character string contained in the first candidate, the A case frame analysis step for generating two candidates, and an N-gram analysis for rearranging the priorities of the kana-kanji character string candidates included in the second candidate based on a probability that N chains of the word appear. And a step.
[0013]
The invention according to claim 4 is the kana-kanji conversion method according to claim 3, wherein the storage device stores a romaji kana conversion table describing kana characters corresponding to romaji, and is stored in the input device. A Romaji-kana conversion step of converting the input Romaji into the Kana character string based on a Romaji-Kana conversion table, wherein the dictionary lookup step matches the Kana character string converted in the Romaji-Kana conversion step. Is extracted from the dictionary.
[0014]
According to this method, the accuracy of kana-kanji conversion can be improved without requiring an extremely long time and a large disk capacity.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described below in detail with reference to the drawings.
[0016]
FIG. 1 is a functional block diagram of the kana-kanji conversion apparatus according to this embodiment.
As shown in the example of FIG. 1, the kana-kanji conversion device of the present embodiment includes an input device 101, a display device 103, a data processing device 105, and a storage device 117. The input device 101 is configured by a keyboard or the like for inputting a kana character string for kana-kanji conversion, or performing various instructions such as conversion and determination.
[0017]
The display device 103 is specifically composed of a CRT, LCD, or the like, and displays a character string or the like input by the input device 101.
[0018]
The data processing unit 105 includes a basic processing unit (BPU) and a control unit that read and execute instructions constituting a computer program in a central processing unit (CPU). , A dictionary lookup unit 109, a morpheme analysis unit 111, a case frame analysis unit 113, and a trigram analysis unit 115.
[0019]
The Romaji-kana conversion means 107 performs a process of converting a Roman character inputted from the outside into a Kana character, that is, hiragana or katakana. The dictionary lookup unit 109 finds a kanji character corresponding to the reading of the kana character converted by the roman character kana conversion unit 107 from the dictionary stored in the kana-kanji conversion device for the purpose of performing kana-kanji conversion.
[0020]
When a kana character is directly input from the input device 101, the process by the dictionary lookup unit 109 is performed without the process by the Roman character kana conversion unit 107. Whether Roman characters or Kana characters are input from the input device 101 is determined by the character input mode in the Kana-Kanji conversion device.
[0021]
The morpheme analyzing means 111 uses the connection information of the part of speech and the priority of the words for words included when the input character string is converted into a kana-kanji mixed character string, that is, a kana-kanji character string. Determine. The case frame analysis means 113 changes the priority order of the character string candidates in the order that the character string candidates mixed with kana-kanji are considered more semantically correct based on the relationship between the verb and its subject and object.
[0022]
The trigram analyzing unit 115 rearranges the character string candidates ranked by the case frame analyzing unit 113 using a set of three words (trigram) extracted from the text corpus.
[0023]
The storage device 117 is constituted by a main storage device or the like for storing data, and stores data accessed by the CPU in relation to the present invention. In the present embodiment, the storage device 117 stores a Romaji-kana conversion table 119, a dictionary 121, a part-of-speech connection table 123, a case frame dictionary 125, and trigram data 127 as data to be referred to in kana-kanji conversion. .
[0024]
Further, in an area (not shown) of the storage device 117, an input character string, a searched kanji candidate, and an execution instruction of a computer program executed by the CPU including the data processing device 105 are stored. Then, the CPU can directly access the contents and put an instruction or data into a register to execute a program, perform an operation on the data, or perform an operation based on the data.
[0025]
2-6 is a figure for demonstrating the flow of the kana-kanji conversion by the kana-kanji conversion apparatus based on this embodiment about the character string "Hon-san, Sansai-yoda". Hereinafter, the operation of the kana-kanji conversion apparatus according to this embodiment will be described.
[0026]
First, in FIG. 2, when a roman character “honwoni, sanstuyononda” is input from the input device 107, the romaji kana conversion unit 107 selects a hiragana corresponding to the reading of the roman character. The romaji-kana conversion means 119 refers to the romaji-kana conversion table 119,
From a correspondence such as ho → hon → nonwo →, a hiragana string “honsa ni san san sai san” is created.
[0027]
Next, the dictionary lookup unit 109 extracts from the dictionary 121 a character string that matches the reading of the hiragana string thus converted. Specifically, “ho”, “ho”, “sail”, “ho” are words corresponding to the reading “ho”, and “book”, “translation” are words corresponding to the reading “hon”. Characters have been extracted and this process continues until the end of the hiragana string.
[0028]
Subsequently, as shown in FIG. 4, the morpheme analyzing unit 111 creates character string candidates including kanji characters based on the part-of-speech connection table 123 for the set of words found by the dictionary searching unit 109. Prioritize candidates. In this embodiment, the character string candidates are given priority to each of the words included in the dictionary 121, and given priority for each type of part-of-speech connection in the part-of-speech connection table 123. Are ranked so that a character string with a low total score is given priority.
[0029]
For example, in the case of the character string “similar to a book and called three volumes”, based on the dictionary 121,
(Book) + (do) + (similar) + (,) + (three) + (book) + (call) + (do)
The score of 10 + 5 + 10 + 5 + 10 + 2 + 20 + 5 = 85 points is calculated as the word priority. This string is
Noun-Particulate-One-level inflection verb-Reading-Numeric-Numerical-Ba-line five-tier inflection-End of speech connection based on part-of-speech connection table 121,
(Noun-Participant) + (Participant-One-stage Inflection Particle) + (One-Step Inflection Particle-Reading Point) + (Reading-Numerical) + (Numerical-Numerical Word) + (Numerical Word-Five-Line Inflection) + (Bain Five-Line Inflection) -Ending)
As a priority by connecting parts of speech,
A score of 30 + 20 + 30 + 40 + 10 + 40 + 30 = 200 points is obtained. Then, the priority of the word and the priority based on the part-of-speech connection are summed, and the priority is calculated as 285 points.
[0030]
By performing the same calculation, 290 kana kanji strings that read “similar books, read three books”, 295 points for “call two or three books”, and “read two or three books” A score of 300 points is calculated for a kana character string. Therefore, the priority of these four candidates is
1. I called the book three times in the same way 2. I read three books in the same way. I called two or three books. I read a couple of books. If the score indicating priority is equal to or higher than a reference value, it is excluded as a candidate that is difficult to connect, that is, a candidate with low possibility. Various combinations of words are possible, for example, “Similar translations, called three volumes”. However, for such Kana-Kanji character strings, the priority score is high as a result of the calculation, so Excluded.
[0031]
Subsequently, as shown in FIG. 5, the case frame analysis unit 113 rearranges the candidates that have been prioritized and narrowed down as described above. The case frame analysis unit 113 refers to the case frame dictionary 125 and performs the following determination process.
[0032]
For example, according to the case frame dictionary 125, when the particle “ga” is positioned before the word “reading”, the character string is determined to be semantically correct if the main case is a person. In addition, when the particle “O” is positioned before “Reading”, it is determined that it is semantically correct if the target case is “Book”. Similarly, if the particle “ga” is located before the word “call”, the main case is “person”, and if “wa” is located, the case is semantically human. Judged to be correct. Therefore, in the case frame analysis means 113, candidates such as “similar book read three books” and “read two books three books” “similar books, called three books” and “two books read”. It is judged that it is more suitable than the three volumes, "that is, semantically correct, and the priority is higher.
[0033]
As a result of such ranking, the priority is 1. I read three books in the same way. 2. I read a few books. Called three books, similar 4 If you call two or three books.
[0034]
Subsequently, the trigram analyzing unit 115 refers to the trigram data 127 and rearranges the candidates ranked by the case frame analyzing unit 113 as shown in FIG.
[0035]
Trigram data 127 describes the probability that a chain of three words will appear in the text, and this probability is created from the actual text corpus. That is, the trigram data 127 is created by dividing the text into words for a large amount of text data included in the actual text corpus and determining the probability that three collocations will appear.
[0036]
When the trigram data 127 is referred to, there is a high probability that an arrangement of the words “2”, “,”, “three” will appear (in other words, “2”, “,” “three” are included in the actual text). Therefore, the character string candidates are rearranged so that a candidate including a lot of word chains of “two”, “,” “three” is prioritized. In the example shown in FIG. 6, for example, word chains with extremely low probabilities such as the arrangement of the words “similar”, “,” and “three” are omitted.
[0037]
Therefore, the result of the rearrangement of the priority order by the trigram analyzing means 115 is
1. I read a few books. 2. I called two or three books. I read three books in the same way. It looks like a book and three books are called.
[0038]
Hereinafter, the result of the experiment using the kana-kanji conversion apparatus according to this embodiment will be described.
[0039]
The probability that a character in a character string is converted to a correct character by one Kana-Kanji conversion process (hereinafter referred to as “char rate”) is 94.09%, and the entire character string is converted to a correct character string by one Kana-Kanji conversion process. Trigram using approximately 80 MB of trigram data created based on an actual text corpus of hundreds of MB for a conventional Kana-Kanji conversion processing device with a probability of being played (hereinafter referred to as “sence rate”) of 46.05% As a result of applying the analysis processing according to the above, char rate was 95.03% and sentence rate was 52.68%. That is, it has been confirmed that the probability of conversion to a correct character string by a single conversion process increases both in character units and sentence units.
[0040]
The preferred embodiment of the present invention has been described above, but the present invention is not limited to this and can be implemented in various other forms.
[0041]
For example, in the above-described embodiment, the trigram data describing the probability that a chain of three words appears in the text included in the actual text corpus is generated, but the number of word chains is three. The number of words is not limited, and a chain of arbitrary N (N ≧ 2) words may be used. In this case, the storage device of the Kana-Kanji conversion device stores N-gram data describing the probability that a chain of N words will appear in the text corpus. That is, the N-gram data is created by dividing the text into words for a large amount of text data included in the actual text corpus and determining the probability that N collocations will appear.
[0042]
Then, instead of the trigram analysis means in the above-described embodiment, the N-gram analysis means refers to the N-gram data, and further rearranges the priority order of the character string candidates rearranged by the case frame analysis means. .
[0043]
【The invention's effect】
As described above, according to the present invention, a dictionary that describes characters that are candidates for Kana-Kanji conversion and the priority of the characters, a part-of-speech connection table that describes the priority of connection of two parts of speech, and a character string A case frame dictionary that describes the semantic relationship of other words to other words, and an N-gram that describes the probability that N (N ≧ 2) chains of words will appear in the text corpus containing the actual text Based on the storage device for storing data, the dictionary stored in the storage device, the part-of-speech connection table, the case frame dictionary, and the N-gram data, a kana-kanji character is input from the input device. A kana-kanji conversion device comprising a data processing device for converting into a string, wherein the data processing device extracts from the dictionary the characters that match the input kana character string And can means, the priority of the text, based on the priority of the connection of the parts of speech, a combination of the character extracted by the dictionary means, the possibility of kana-kanji character string is equal to or greater than a predetermined reference a morphological analysis means for creating a first candidate, on the basis of the case frames dictionary, the sorted priority candidate kana-kanji character string included in the first candidate, case frame analysis to create a second candidate And N-gram analyzing means for rearranging the priorities of the kana-kanji character string candidates included in the second candidate based on the probability that N chains of the word appear. By applying a trigram only to candidates output by Kana-Kanji conversion, conversion accuracy can be improved while reducing the amount of calculation.
[0044]
The storage device stores a Roman-kana conversion table that describes a kana character corresponding to a Roman character, and the data processing device converts the Roman character input from the input device into the kana character string based on the Roman-kana conversion table. Romaji-kana conversion means for converting to kana, and the dictionary lookup means extracts from the dictionary the characters that match the kana character string converted by the romaji-kana conversion means. In any case, the accuracy of kana-kanji conversion can be improved without requiring an extremely long time and a large disk capacity.
[Brief description of the drawings]
FIG. 1 is a functional block diagram of a kana-kanji conversion apparatus according to an embodiment.
FIG. 2 is a diagram for explaining a flow of kana-kanji conversion by the kana-kanji conversion apparatus according to the present embodiment.
FIG. 3 is a diagram for explaining a flow of kana-kanji conversion by the kana-kanji conversion apparatus according to the present embodiment.
FIG. 4 is a diagram for explaining a flow of kana-kanji conversion by the kana-kanji conversion apparatus according to the present embodiment.
FIG. 5 is a diagram for explaining a flow of kana-kanji conversion by the kana-kanji conversion apparatus according to the present embodiment.
FIG. 6 is a diagram for explaining a flow of kana-kanji conversion by the kana-kanji conversion apparatus according to the present embodiment.
[Explanation of symbols]
101 Input Device 103 Display Device 105 Data Processing Device 107 Romaji Kana Conversion Unit 109 Dictionary Lookup Unit 111 Morphological Analysis Unit 113 Case Frame Analysis Unit 115 Trigram Analysis Unit 117 Storage Device 119 Romaji Kana Conversion Table 121 Dictionary 123 Part-of-Speech Connection Table 125 Case Frame dictionary 127 Trigram data

Claims (4)

A dictionary that describes characters that are candidates for Kana-Kanji conversion and the priority of the characters, a part-of-speech connection table that describes the priority of the connection of two parts of speech, and the semantic relationship of other words contained in the character string to other words A case frame dictionary that describes and a storage device that stores N-gram data that describes the probability of N (N ≧ 2) chains of words appearing in a text corpus that contains the actual text;
A data processing device for converting a kana character string input from an input device into a kana-kanji character string based on the dictionary, the part-of-speech connection table, the case frame dictionary, and the N-gram data stored in the storage device; A kana-kanji conversion device comprising the data processing device,
Dictionary lookup means for extracting from the dictionary the characters that match the input kana character string;
Based on the priority of the character and the priority of connection of the part of speech, combining the characters extracted by the dictionary lookup unit , a first candidate for a kana-kanji character string whose possibility is equal to or greater than a predetermined criterion is obtained. Morphological analysis means to create,
Based on the case frame dictionary, case frame analysis means for rearranging the priorities of the kana-kanji character string candidates included in the first candidate to create a second candidate;
Kana-Kanji conversion comprising: N-gram analyzing means for rearranging the priorities of the Kana-Kanji character string candidates included in the second candidate based on a probability that N chains of the word appear. apparatus. The storage device stores a romaji kana conversion table describing kana characters corresponding to romaji, and the data processing device converts the romaji input from the input device into the kana character string based on the romaji kana conversion table. 2. The kana-kanji character according to claim 1, further comprising: a romaji kana conversion unit configured to extract the characters that match the kana character string converted by the romaji kana conversion unit. Conversion device. A dictionary that describes characters that are candidates for kana-kanji conversion and the priority of the characters stored in the storage device, a part-of-speech connection table that describes the connection priority of two parts of speech, and a phrase included in the character string Based on case frame dictionaries that describe semantic relationships to other phrases, and N-gram data that describes the probability of N (N ≧ 2) chains of words appearing in a text corpus containing actual text, A kana-kanji conversion method for converting a kana character string input from an input device into a kana-kanji character string,
A dictionary lookup step of extracting from the dictionary the characters that match the input kana character string;
Based on the priority of the character and the priority of connection of the part of speech, combining the characters extracted in the dictionary lookup step, a first candidate for a kana-kanji character string whose possibility is equal to or higher than a predetermined criterion is obtained. A morphological analysis step to create,
Based on the case frame dictionary, a case frame analysis step of rearranging the priorities of the kana-kanji character string candidates included in the first candidate to create a second candidate;
A kana-kanji conversion device comprising: an N-gram analysis step of rearranging the priorities of the kana-kanji character string candidates included in the second candidate based on a probability that N chains of the word appear. Method. The storage device stores a romaji kana conversion table describing kana characters corresponding to romaji, and a romaji kana conversion step for converting romaji input from the input device into the kana character string based on the romaji kana conversion table. 4. The kana-kanji conversion method according to claim 3, wherein the dictionary lookup step extracts, from the dictionary, the characters that match the kana character string converted in the Roman character kana conversion step.

Images