JPS61240362A - Japanese language inputting device - Google Patents

Abstract

PURPOSE:To make the work of inputting easy by automatically registering in a word dictionary word strings assumed highly probable to be inputted therein again from among inputted ones. CONSTITUTION:Plural arrays of parts of speech frequently used as a compound word or used compound-wise in the Japanese language are registered in advance in the part of speech string table 105. A word string register part 104 examines whether or not the partial word string that has the same array of parts registered in the table 105 exists among the ones yet inputted. If the said string exists, it is registered in the word dictionary newly as one unit together with the part information.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a Japanese input device that recognizes input signals such as speech or handwritten characters and receives them as a string of words registered in a word dictionary.

(Prior art and its problems) If it were possible to input Japanese sentences using voice or handwritten characters on a word processor, etc., there would be no need to use a large number of keys on a keyboard, etc., and the input work would be extremely difficult, especially for beginners. becomes easier.

In the case of Japanese input using voice or handwritten characters, the voice signal input as a series of syllables or the handwritten character signal input as a series of characters is converted into the corresponding Japanese character code one syllable or character at a time. If conversion is possible, Japanese sentences can be input by accepting the obtained character code string as is or by performing kana-kanji conversion.

However, with current technology, if a voice signal is recognized syllable by syllable or a handwritten character signal is recognized character by character, 10
It is impossible to obtain a recognition rate of 0%, and it is inevitable that some degree of misrecognition will occur.

Therefore, as seen in Japanese Patent Application No. 58-55995 "Voice Recognition System" (Reference 1), word dictionaries containing words that appear in input Japanese sentences, and furthermore connections between the words, have been proposed. By preparing in advance grammatical knowledge such as a word connection table that describes the possibilities of , a method has been adopted that allows correct input of Japanese characters.

In such a method, first, word candidates for each part of the input signal are extracted from the word dictionary, and the possibility of connection between adjacent word candidates on the input signal is tested using the word connection table. Create column candidates. Generally, a plurality of such word string candidates exist. For example, the nouns in the word dictionary include "interpreter", [function], "yesterday", [
Suppose that ``induction'', ``symbol'', ``hope'', ``interpreter'', and ``arm J'' are registered, and the word connection table indicates that nouns can be connected. J with function
Suppose that we input the voice [TSUYAKUKINOUDEJ] to input the following. The word string candidates for this input voice "Tsuyakuki no de" are [J for interpreter function, "interpreter induction", "interpreter yesterday", [interpreter symbol], [J for interpreter request, "arm of interpreter"). exist. Therefore, the plurality of word string candidates are displayed in order of reliability, and the user selects the correct word string. Normally, the longest-dispersion concept is often used when searching word dictionaries, and word string candidates containing as many words as possible are considered more reliable. Therefore, in the above example, if the word dictionary does not contain the noun "interpreter function", the first option would be "interpreter's arm", which includes the longest word "interpreter", and the user would be able to find the correct word string. It is necessary to sequentially obtain the next candidate by pressing the next candidate key, etc. until it is displayed "with the interpreter function". On the other hand, if the word dictionary includes an "interpretation function", the "interpretation function" becomes the first choice, and the user can immediately obtain the correct word string.

That is, in order to improve recognition performance and to perform input easily and quickly, it is desirable that the longest possible subsequence of the input word string be registered in the word dictionary.

However, in Japanese, compound nouns such as [interpretation function J, U interpretation function investigation J, compound verbs such as ``read out J'', ``take out J'', prefixes and suffixes such as ``all functions'', ``study session'', etc. There are an infinite number of combinations of words and nouns, and it is impossible to register them all in a word dictionary in advance. Furthermore, since these compound words, especially compound nouns, are often used in each specific specialized field, words that are only used in other fields may be registered in the word dictionary when inputting sentences in one field. This is undesirable because it requires unnecessary storage capacity and processing time. In other words, it is desirable that as many words or word strings as possible that will appear in the sentence being created be registered in the word dictionary.

In general, there is a fact that a word string that appears once in a certain sentence has a high possibility of appearing again.

Therefore, conventionally, the user has performed the task of registering a word string that has appeared once in a word dictionary as a new word, thereby facilitating subsequent input operations. However, in this case, the user must specify the range to be registered and instruct the registration, which is very time-consuming.

(Object of the Invention) An object of the present invention is to eliminate the drawbacks of the prior art, and to automatically select a word string that is likely to be input again from among the input word strings without the user's instruction. To provide a Japanese language input device that registers words in a dictionary and facilitates subsequent input work.

(Structure of the Invention) In order to achieve the above object, the present invention has a Japanese word dictionary, and inputs input signals such as Japanese speech and handwritten characters to words consisting of one or more words in the word dictionary. In a Japanese input device that accepts a string, each word in the word dictionary is assigned a part of speech of the word, and the order of the parts of speech of a partial word string of the input word string is predetermined. If it is a sequence of parts of speech, the partial word string is
The present invention is characterized by comprising means for registering the word in the word dictionary.

(Examples) Hereinafter, the present invention will be described in detail according to examples with reference to the drawings. , FIG. 1 is a diagram showing an embodiment in which the present invention is applied to a Japanese language input device.

The speech recognition unit 101 inputs speech and outputs as a recognition result a word string consisting of words in the word dictionary 102 and intended by the user.

FIG. 2 is a diagram showing an embodiment of the speech recognition section 101.

Input audio is input from the microphone 201. Syllable recognition part 2
02 analyzes the audio signal received from the microphone, matches each syllable part in the signal with the syllable standard bangs in the syllable standard bang dictionary 203, and selects several syllable standards in order from the standard bang with the smallest matching distance. The syllable candidate storage unit 20 uses the syllable category of batan as a syllable candidate for the syllable part.
Output to 4. For example, if you input the voice uttered ``Tsu yaku kino de'' to input `` with the 6 interpreter function, the first syllable ``tsu'' will be replaced with ``tsu'' as a syllable candidate.
, [zuJ, "su", "ri", and "fu" are obtained, and several syllable candidates are similarly obtained for the following syllables.

As a result, the syllable candidate storage unit 204 stores a plurality of syllable candidates for each syllable in the input signal.

The word string candidate generation unit 205 retrieves syllable candidates for each syllable one by one from the syllable candidate storage unit 204, and extracts syllable candidates one by one for each syllable from the syllable candidate storage unit 204, and calculates a word string consisting of one or more words in the word dictionary 102 and the relationship between adjacent words. creates a word string that does not contradict the word continuation possibilities listed in the word connection table 206. Generally, there are a plurality of such word strings, but using the idea of longest dispersion, it is assumed that a word string that includes words with as long a word length as possible has a high degree of reliability. In the above example, the word dictionary contains the nouns "interpreter", "function", [yesterday], "induction", [symbol J, "hope", "interpreter", "arm J, with r as particle", [ ' is registered, and the word connection table states that nouns can be connected to each other. For the input voice 'Tsuyakukinoude J', the word string candidates are 'with the interpreter function', ' ``By interpretation induction'', ``Interpretation yesterday'', ``Interpretation symbol'', ``Interpretation hope'', ``Interpretation machine arm J'' can be obtained. Here, assuming that the noun "interpretation function" is not registered in the word dictionary 102, among these word string candidates, the one with the highest reliability 1i, which includes the longest word, "
He is said to have the same skill as an interpreter, and other candidates are said to have the same level of reliability.

The word string candidate creation unit 205 sends the created word string candidates one by one, starting from the one with the highest degree of reliability, to the word string candidate confirmation unit 20.
7, the part-of-speech information of each word in the word string candidate is output. The word string candidate confirmation section 207 displays the accepted word string candidates on the display section 208, and allows the user to respond as to whether or not the word string candidates are what the user intends. When the user answers "6 intended" from the keyboard 209, the word string candidate confirmation section 207 adds the word string candidate to the input sentence storage section 103 along with the part of speech information and processes it. Terminated. 6 Not intended by the user.”
If Uke receives the answer, the word string candidate confirmation unit 207 requests the word string candidate creation unit 205 to send the next candidate. In the above example, the word string candidate creation unit 205 first selects “the arm of the interpreter”, which has the highest reliability, to the word string candidate confirmation unit 205.
7, but since this is not what the user intended, a request is made to the word string candidate generation unit 205 for the next candidate. The word string candidate creation unit 205 creates word strings of "by interpreter symbol,""by interpreter request,""by interpreter function," [j by interpreter yesterday, and [by interpreter induction] one by one in an appropriate order as the next candidates. It is output to the candidate confirmation unit 207. The user sequentially checks these word string candidates on the display unit 208, and then enters the answer "That is what I intended" from the keyboard when J is displayed in the interpreter function, and as a result, the input The word string "with the interpreter function" is sent to the sentence storage unit 103, and the input of "tsuuyakukinoude" is completed. The contents of the input sentence storage section 103 are displayed on the input sentence display section 106 and shown to the user.

Thereafter, the word string registration unit 104 registers a part of the word string that has been input so far that is likely to appear again in future input operations as a new word in the word dictionary 102. For this purpose, a plurality of sequences of parts of speech that are often used as compound words or in combination in Japanese are registered in advance in the part-of-speech sequence table 105.

Examples of such arrangement of parts of speech include repetition of nouns, repetition of verbs, arrangement of nouns and suffixes, and arrangement of prefixes and nouns. Such a sequence of parts of speech is often used as one word, so a word string that has been input once is likely to appear again.

The word string registration unit 104 checks whether there is a partial word string among the word strings input so far that has a part-of-speech arrangement that is the same as the part-of-speech arrangement registered in the part-of-speech string table 105. If there is such a partial word string, the partial word string is treated as a new word and is added to the word dictionary 10 along with the part of speech information.
Register for 2. In the above example, the input sentence storage unit 103
A new word string ``with the interpreter function'' has been stored, and the ``interpreter function'' part is a string of two nouns, ``interpreter J'' and ``function'', which are registered in the part-of-speech string table 105. Since this is a string of parts of speech, the word string registration unit 104
is a word string [word dictionary 1 with interpreter function J as a new word]
Register on 02. Furthermore, if the word [automatic j] was input just before inputting the word string ``with the interpreter function'' described above, the word string ``with the interpreter function'' was inputted in the input sentence storage unit 103. "With automatic translation function" will exist. As a result, the word string [automatic interpretation function] is also registered in the word dictionary 102 because it is a repetition of a noun. Further, at this time, the user may be informed of the newly registered word string part of the input sentence displayed on the input sentence display section 106 by highlighting it, underlining it, changing its color, or the like.

As a result of the new word string being registered in the word dictionary 102, in subsequent input operations, if you input by voice ``Tsuyakukinoude'', ``Tsuyakukinouga'', or ``Jidou Tsuuyakukinouwo'', the word string ``Interpretation function'' will be displayed. in"
, "Interpretation function", and "Automatic interpretation function" are very likely to be displayed on the display unit 208 as the first candidates, so the user hardly needs to check other word string candidates, and the input work is simplified. can be done easily and quickly.

Similarly, if you input J to import, initially 2
It is recognized as a sequence of two words ``take'' and ``put'', but
Since this word string is a repeated verb, "take in" is registered as one word in the word dictionary 102, and as a result, it becomes possible to easily input "take in" in subsequent input operations.

In this case, the conjugation form of "take in" is "put in".
Inherits the conjugation type of.

In the above embodiments, the case of a voice input device that recognizes syllable units has been described, but it can also be applied to a voice input device that recognizes word units. In this case, the word dictionary is a dictionary of the phonetic patterns of each word, and when the word string registration unit newly registers a word string in the word dictionary, the word string registration unit generates a voice that is a concatenation of the phonetic patterns of each word in the word string. The pattern will be registered in the word dictionary.

Further, the same device may be used for the word string candidate display section 208 and the input sentence display section 106. In that case, if the word string candidates are displayed at the end of the input sentence up to that point, the user can easily understand the input sentence and the word string. It becomes easier to see both candidates.

Further, if the voice recognition section in the above embodiment is replaced with a handwritten character recognition section, it can be used as a handwritten character input device.

(Effects of the Invention) As is clear from the above explanation, according to the present invention, word strings that are likely to be input again among input word strings are automatically changed into word strings without any instructions from the user. It is possible to provide a Japanese language input device that is registered in a dictionary and can perform subsequent input operations easily and quickly.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing one embodiment of the present invention, and FIG. 2 is a diagram showing one embodiment of a speech recognition section.

Claims (1)

[Claims] In a Japanese input device that has a Japanese word dictionary and receives an input signal such as Japanese speech or handwritten characters as a word string consisting of one or more words in the word dictionary, each word in the word dictionary is provided. is given the part of speech of the word,
In the case where the arrangement of the parts of speech of a partial word string of the input word string is a predetermined arrangement of parts of speech, the method further comprises means for registering the partial word string as one word in the word dictionary. Characteristic Japanese input device.