JP3084864B2 - Text input device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text input device such as a voice input word processor and a text reading device.

[0002]

2. Description of the Related Art Conventionally, in the above-mentioned sentence input device, when converting a kana character string containing ambiguity into a kana-kanji character, an enormous number of kana-kanji-mixed candidates are created. I was narrowing down. Then, for example, in a voice input word processor, when converting the speech recognition result of a phrase-uttered voice into Kana-Kanji conversion, not only grammatical characteristics in a phrase but also syntax information in a sentence unit and previous context information are used. To choose more plausible candidates. As a technique for improving conversion accuracy using statistical information, it has been performed to preferentially select a candidate that includes a high-frequency word using the frequency of appearance of the word.

[0003]

However, in the conventional text input device as described above, when selecting candidates using syntax information, it is difficult to accurately perform syntax analysis or context analysis. However, the correct answer candidate was not always selected in the higher rank. Further, even when selecting a candidate using statistical information, syntactic information is not reflected only in frequency information, so that it is not very accurate. For example, a phrase uttered at the beginning of a sentence and a phrase uttered at the end of a sentence should have different appearance frequencies for the same word.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and the accuracy of kana-kanji conversion is improved by reflecting syntactic information on statistical information by a relatively simple method. An object is to provide a high text input device.

[0005]

In order to achieve this object, a sentence input device according to the present invention comprises a kana character recognition means for recognizing a sentence input in units of kana characters, and an ambiguous character recognized by the character recognition means. A kana-kanji conversion means for converting a kana-character string containing kana into a kana-kanji-mixed sentence, and a word dictionary to be considered when the kana-kanji conversion means performs the conversion, wherein the word dictionary includes a word reading and a kana-kanji conversion means. , The word notation, and the distance from the beginning of the sentence in the sentence of the word
Storing the frequency information of the releasing each further include: a position determining means for determining the distance from the beginning of a sentence within a sentence of the kana character string, from the beginning of a sentence within a sentence of the kana character string obtained by said position determining means A frequency information reading unit that reads frequency information of the conversion candidate word of the kana character string from the word dictionary based on the distance; and the conversion based on the frequency information of the conversion candidate word read by the frequency information reading unit. Priority determining means for determining a priority of a candidate word, wherein the kana-kanji conversion means converts the kana character string into a kana-kanji mixed sentence according to the priority determined by the priority determining means. . The distance from the beginning of the sentence may be the number of words from the beginning of the sentence or the number of phrases from the beginning of the sentence.

[0006]

In the sentence input device of the present invention having the above-described configuration, when converting a kana character string containing ambiguity into kana-kanji characters, the kana-kanji conversion means creates kana-kanji mixed candidates with reference to a word dictionary. At this time, the position determining means obtains a distance from the beginning of the sentence in the sentence of the kana character string,
The frequency information reading means reads the frequency information of the conversion candidate word of the kana character string from the word dictionary based on the distance from the beginning of the sentence of the kana character string obtained by the position determining means, and The determining means obtains the priority of the conversion candidate word based on the frequency information of the conversion candidate word read by the frequency information reading means, and the kana-kanji conversion means according to the priority determined by the priority determining means. The kana character string is converted into a sentence mixed with kana-kanji. In the text input device of the present invention, the position determining means
Is the number of words from the beginning of the sentence in the sentence of the kana character string
Or, the number of phrases from the beginning of the sentence is obtained, and the frequency information reading means
In the sentence of the kana character string obtained by the position determining means
Based on the number of words from the beginning of the sentence or the number of phrases from the beginning of the sentence.
And the frequency information of the word of the kana character string conversion candidate is
Read from the dictionary, and the priority determining means
Based on the frequency information of the conversion candidate words read by the
Then, the priority of the conversion candidate word is obtained,
The kanji conversion means converts the priority determined by the priority determination means
Therefore, the kana character string is converted to a sentence mixed with kana-kanji.
You may do so.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment in which the present invention is embodied as a speech input word processor will be described below with reference to the drawings.

First, the overall configuration will be described with reference to FIG. The voice recognition result receiving unit 10 is a kana-kanji conversion unit C
It is connected to a PU (Central Processing Unit) 12. The CPU 12 has a word dictionary 14 to be taken into account when converting the speech recognition result into kana-kanji characters, a RAM (random access memory) 16 as a working area, and a ROM (read-only memory) 1 storing kana-kanji conversion procedures.
8 are connected. A display device 20 for displaying kana-kanji mixed candidates (hereinafter simply referred to as candidates) converted to kana-kanji is connected to the CPU 12.

When the voice recognition result receiving unit 10 receives a kana character string containing ambiguity, which is the voice recognition result, it passes it to the CPU 12. The CPU 12 converts kana-kanji character strings containing ambiguity into kana-kanji characters according to the procedure stored in the ROM 18 while referring to the word dictionary 14, and creates a plurality of kana-kanji mixed candidates. At this time, for each word included in the kana-kanji mixture candidate, the CPU 12 reads the frequency from the word dictionary 14 according to the position of the word in the sentence, calculates the priority of the candidate, and associates the candidate with the priority. To the RAM 16. The procedure for calculating the priority will be described later with reference to the flowchart in FIG. After all the candidates have been created, the CPU 12 sorts the plurality of kana-kanji mixed candidates according to their priorities and displays them on the display device 20.

Next, with reference to FIG. 1, frequency information for each position in a sentence of a word stored in the word dictionary 14 will be described. The frequency of words is obtained in advance using a large number of sentences as learning data. At this time, the frequency is obtained for each distance from the beginning of the sentence (for example, the number of words), and the value of the frequency is normalized so that the maximum frequency becomes 1. FIG. 1 is a graph obtained by approximating the frequency value with a multidimensional function using the distance from the beginning of the sentence as input and the frequency value as output. For example, the graph on the left side of FIG. 1 indicates that the frequency is highest when the word A is at the beginning of the sentence, and decreases as the distance from the beginning of the sentence increases. The graph on the right side of FIG. 1 shows that the frequency is low when the word B is at the beginning of the sentence, and that the frequency peak is located far from the beginning of the sentence.

A procedure for calculating the priority of a candidate using the frequency of each position in the text will be described with reference to the flowchart of FIG. In the kana-kanji conversion, each time one kana-kanji mixture candidate is created, the priority of the candidate is calculated according to the following procedure. Assuming that the priority of a candidate is p, p is first set to an initial value 0 (Step 1, hereinafter referred to as S1). Next, one word constituting the candidate is extracted (S2). Then, the frequency h of the word (not the frequency for each position in the text, but simply the number of occurrences in the learning data and normalized to be 0 to 1) is read out from the word dictionary 14 ( S3).

Further, the number of words from the beginning of the word is calculated (S4), and the frequency h1 of each word in the phrase is read out from the word dictionary 14 with the number of words from the beginning of the word as an input (S5). ). Here, the frequency h is calculated from the graph described above.
Seeking 1 Then, the frequency h and the frequency h1 for each position in the text are added to the priority p of the candidate (S6). It is checked whether or not there is a word whose frequency has not been read out among the candidates (S
7) If there is, one word whose frequency has not been read out is extracted (S8), and the process returns to S3. If the frequency has been read out for all the words in the candidate, the priority p is divided by the number of words in the candidate in order to convert the frequency value to the value per word, and finally the priority p of the candidate is finally divided. Degree p (S
9).

For example, when "I am" is uttered at the beginning of the sentence,
It is assumed that as a result of voice recognition and Kana-Kanji conversion, candidates such as “I am” and “I passed” are created. At this time, the noun “I” and the particle “ha” have a higher frequency at the beginning of the sentence than the verbs “pass” and “ta”, so “I am” has a higher priority than “passed”. Thus, "I am" is output to the display device 20 with priority. Conversely, when "uttered" is uttered at the end of the sentence, it is assumed that "I", "handed", etc. are created as candidates as in the previous case. This time, the verb "passing" and the auxiliary verb "ta" have a higher frequency at the end of the sentence than the noun "me" and the auxiliary particle "ha", so the priority of "passing" is higher than "I am". It becomes higher, and “passed” to the display device 20 is output with priority.

The present invention is not limited to the above-described embodiment. For example, the number of words from the beginning of a sentence is used as a position in a sentence.
The frequency of each word may be calculated for each number of phrases from the beginning of the sentence. Also, the priority of the candidate is not only frequency,
Priority obtained from part of speech information, semantic information, etc. may be added.

[0015]

As is apparent from the above description, in the sentence input device of the present invention, when converting a kana-kanji character string containing ambiguity, the kana-kanji conversion means selects kana-kanji mixed candidates while referring to a word dictionary. create. At this time,
The position determining means determines the distance from the beginning of the sentence in the sentence of the kana character string, and the frequency information reading means determines the distance based on the distance from the beginning of the sentence in the sentence of the kana character string determined by the position determining means. The frequency information of the conversion candidate word of the kana character string is read out from the word dictionary, and the priority determination unit determines the priority of the conversion candidate word based on the frequency information of the conversion candidate word read by the frequency information reading unit. The kana-kanji conversion means converts the kana character string into a sentence containing kana-kanji characters according to the priority determined by the priority determination means. Thus, there is an effect that the syntactic information is reflected on the statistical information by a relatively simple method, and the accuracy of the kana-kanji conversion is further increased. Also, the above sentence
Is the number of words from the beginning of the sentence or the number of phrases from the beginning of the sentence.
In the event that a
Higher accuracy of Kana-Kanji conversion by reflecting literary information
Has the effect of doing

[Brief description of the drawings]

FIG. 1 is a graph showing frequency information for each position in a sentence of a word.

FIG. 2 is a block diagram showing the overall configuration of the embodiment.

FIG. 3 is a flowchart showing a procedure for calculating the priority of kana-kanji mixed candidates.

[Explanation of symbols]

 Reference Signs List 10 speech recognition result receiving unit 12 CPU 14 word dictionary 16 RAM 18 ROM 20 display device

Claims (2)

(57) [Claims] 1. A kana-kanji character recognizing means for recognizing a sentence input in a kana character unit, a kana-kanji conversion means for converting a kana character string including ambiguity recognized by the character recognizing means into a kana-kanji mixed sentence, In a sentence input device provided with a word dictionary to be taken into account when performing conversion by a kana-kanji conversion means, the word dictionary includes word reading, word notation, and frequency information for each distance from the head of the sentence in the word sentence. Further, position determining means for determining the distance from the beginning of the sentence in the sentence of the kana character string, based on the distance from the beginning of the sentence in the sentence of the kana character string determined by the position determining means, A frequency information reading unit that reads frequency information of the conversion candidate word of the kana character string from the word dictionary; and a frequency of the conversion candidate word read by the frequency information reading unit. Priority determining means for determining the priority of the word of the conversion candidate based on the information, wherein the kana-kanji conversion means converts the kana character string into a kana-kanji mixed sentence according to the priority determined by the priority determining means. A sentence input device characterized by conversion. 2. The distance from the head of the sentence is a word from the head of the sentence.
Claims characterized by the number or the number of phrases from the beginning of the sentence
The sentence input device according to 1.