JPH05189416A - Word processor - Google Patents

Abstract

PURPOSE:To attain the neuro-learning in a wide range without increasing the necessary capacity with a document processor which performs the KANA (Japanese syllabary)/KANJI (Chinese character) conversion with use of a neural network (neuro-learning function). CONSTITUTION:The conceptual information showing the meanings of conversion words is stored in a KANA/KANJI dictionary 15. Then the conversion word retrieved out of the dictionary 15 and the conceptual information on the word are stored in a homonym buffer 18 when the KANA/KANJI conversion is carried out. Meanwhile the neuro-information defining the conceptual information as a node is stored in a neuro-table 17. Then the output control of the conversion word stored in the buffer 18 is carried out based on the active value of the neuro-information. Furthermore the active value of the neuro-information is updated based on the conceptual information when a conversion word is decided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document creating device for performing kana-kanji conversion using a neural network (neuro learning function).

[0002]

2. Description of the Related Art Conventionally, kana-kanji conversion using a neuro-learning function has been studied as a means for performing kana-kanji conversion while grasping the topic of an input document in a document creation device. This is more than the words in the documents created so far,
Focusing on the topicality of the input document, the words that have been topically similar to the words that have appeared so far are preferentially output. Therefore, we construct a neural network with each word registered in the kana-kanji conversion dictionary as a node, and automatically imitate the human association on the network as follows.

That is, a word is associated with an activity value indicating the degree to which the word is associated with the context at any given time. And give a high activity value to the word that appeared recently,
The activity value is controlled through the link in some way so as to increase the activity value of other words with a strength according to the weight of the link.

For example, in response to the reading "Shikou", FIG.
In the case where the neural network as shown in Fig. 2 is configured, after inputting the document "The president intends to promulgate it immediately if the bill passes", then input the document "What is it?" Then, "enforcement" is preferentially output as a conversion word of "shikou" at that time. This is because many words related to law such as “bill” and “promulgation” are used in the document, so that the activation value of “enforcement” in the neural network is different from other homophones (“trial”, “oriented”, “ Thought ”). Note that in FIG. 8, the numbers attached between the nodes indicate the link weights.

[0005]

By the way, conventionally, in the Kana-Kanji conversion using the neural network as described above, each node in the network uniquely corresponds to each word in the dictionary. Here, the number of words registered in the dictionary is enormous, and if a network is formed with all the words as nodes, the capacity required for neurolearning increases. For this reason, there is no choice but to form a network with words in a limited range as nodes, and it has been impossible to perform neuro learning over a wide range.

The present invention has been made in view of the above points, and it is an object of the present invention to provide a document creation apparatus capable of realizing neuro learning over a wide range without increasing the capacity required for neuro learning.

[0007]

A document creating apparatus of the present invention is a concept for storing a conversion dictionary storing conversion word information corresponding to reading information, and concept information of the conversion word information stored in the dictionary. Information storage means, neuro information storage means for storing neuro information having the conceptual information stored in the conceptual information storage means as a node, and the dictionary is used based on the neuro information stored in the neuro information storage means. And a control means for performing conversion.

[0008]

According to the above structure, the neuro information is used as a method of searching the dictionary for the conversion word information corresponding to the reading information and determining the output rank of the conversion word information. In this case, since the neuro information is formed by using the conceptual information as a node, the memory capacity required for the neuro learning can be significantly reduced as compared with the conventional neuro learning in which the conversion word information is used as a node, and the neuro information over a wide range can be obtained. Learning can be realized, and the conversion word desired by the user can be output more accurately and preferentially.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A document creating apparatus according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the arrangement of a document creating apparatus having a kana-kanji conversion function. In FIG. 1, an input unit 11 is a keyboard for inputting various kinds of information necessary for document creation and the like. The input unit 11 is provided with input keys for kana characters, English characters, numbers, etc., and here, a "conversion instruction" key 11a for instructing conversion of kana-kanji characters. The input control unit 12 controls input of various information input by the input unit 11. The kana-kanji conversion control unit 13 performs kana-kanji conversion using the kana-kanji conversion processing unit 14 and the neuro processing unit 16.

The kana-kanji conversion processing section 14 refers to the kana-kanji conversion dictionary 15 to create conversion word candidates corresponding to the input reading information in the homophone word buffer 18. The dictionary 15 is composed of, for example, a ROM, and preliminarily stores readings (reading information) and conversion words corresponding to the readings, as shown in FIG. In addition, in the same embodiment, corresponding to each conversion word, conceptual information indicating the meaning of the word is stored. In the example of FIG.
For the reading "koi", the conversion words "love", "carp", "willfulness", and "begging", and corresponding to each conversion word, "love", "fish", "meaning", " The conceptual information of "request" is stored.

The neuro processing unit 16 uses the neuro table 17 to perform neuro learning processing. The neuro table 17 is composed of, for example, a RAM, and stores neuro information indicating the appearance frequency of conversion words as shown in FIG. This neuro information is composed of link information and active value. The link information indicates the connection between the nodes on the neural network having the conceptual information as the node as shown in FIG. The activity value represents the degree to which the word is associated with the current context in the currently created document, and is updated with the strength according to the link weight. Figure 3
In the example, for the conversion word "love", "human",
The link information such as “social relations”, “grief”, and “mind” and the activity value “1.3” are stored.

The homophone word buffer 18 is composed of, for example, a RAM, and stores conversion word candidates created by the kana-kanji conversion processing unit 14 and here conceptual information thereof. The display control unit 19 also performs display control on the display unit 20 under the control of the Kana-Kanji conversion control unit 13. Display unit 2
Reference numeral 0 is a display device for displaying various kinds of information, which is, for example, a liquid crystal display with a backlight of 40 characters and 20 lines, and the bottom line is an area dedicated to guide display (guide line).
Is set to. Next, the operation of the embodiment will be described.

As shown in the flow chart of FIG. 4, the user inputs, through the input unit 11, reading information such as "Ikekara, Mizuno Haneotogashi." (Step A1), and the "Conversion instruction of the input unit 11". When the Kana-Kanji conversion is instructed by operating the key 11a (step A2), the information is given to the Kana-Kanji conversion control unit 13 through the input control unit 12. As a result, under the control of the kana-kanji conversion control unit 13, first, the kana-kanji conversion processing unit 14 searches the dictionary 15 (step A3) to read all corresponding conversion words and their conceptual information for each reading, These are stored in the homophone word buffer 18 in order of priority in the dictionary 15 (step A
4). The homophone word buffer 18 for each conversion word at this time
Is shown in FIGS. 6 (a) to 6 (d).

When the conversion word candidates are created in the homophone word buffer 18 in this way, the neuro processing unit 16 refers to the neuro table 17 (step A5), and the neuro information stored in the neuro table 17 is obtained. Based on this, the conversion word candidates in the homophone word buffer 18 are rearranged in the order of high activity value (step A6). The display control unit 19
The conversion word candidates of the homophone word buffer 18 are sequentially output to the display unit 20 in accordance with the operation of the "next candidate" key or the like.

Here, as shown in the flow chart of FIG. 5, if the user determines that the conversion result is "the sound of the water bouncing from the pond" (step B1), the neuro processing unit 16 causes the homologous word buffer. 18 from "Pond",
The conceptual information “lake”, “water”, “concentration / foot movements”, and “sound” corresponding to the conversion words “water”, “bounce”, and “sound” are searched (step B2). Next, the neuro processing unit 16
Updates the active value of a node (concept information here) in the neuro table 17 and the active values of other nodes linked to that node according to the weight of each link based on this conceptual information (step B3). .. When the updating process of the active value is completed, the kana-kanji conversion control unit 13
The homophone word buffer 18 through the kana-kanji conversion processing unit 14.
To initialize and prepare for the next kana-kanji conversion (step B
4).

The state at this time is shown in FIG. Figure 7 (a)
Shows a neural network before the conversion result is fixed, and FIG. 7B shows the neural network after the conversion result is fixed. In the above example, "the sound of the water bouncing from the pond." As a result, it can be seen that the activity value related to seafood increased and the activity value related to romantic relationship did not change. Therefore, when the next reading, for example, "I wonder if there is a carp."
As shown in FIG. 6 (e), “love”, “carp”, “intentional”, and “begging” are stored as conversion word candidates for “koi” in FIG. Is output as a conversion result with priority. This is because the activity value of “fish”, which is the conceptual information of “carp”, was higher than the activity values of other words, not the activity value of the word “carp” itself.

[0018]

As described above, according to the present invention, neuro learning is performed based on the concept information of the conversion word. Therefore, as compared with the conventional neurolearning in which each of the conversion words in the dictionary is used as a node, the memory capacity required for the neurolearning can be significantly reduced, and the neurolearning over a wide range can be realized. It is possible to preferentially output the conversion word desired by the user.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a document creation device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration of a dictionary of the embodiment.

FIG. 3 is a diagram showing a configuration of a neuro table of the same embodiment.

FIG. 4 is a flowchart for explaining the operation of the embodiment.

FIG. 5 is a flowchart for explaining the operation of the embodiment.

FIG. 6 is a view showing the contents of a homophone word buffer according to the embodiment.

FIG. 7 is a diagram showing a configuration of a neural network of the same embodiment.

FIG. 8 is a diagram for explaining kana-kanji conversion using a neural network.

[Explanation of symbols]

11 ... Input unit, 11a ... "Conversion instruction" key, 12 ... Input control unit, 13 ... Kana-Kanji conversion control unit, 14 ... Kana-Kanji conversion processing unit, 15 ... Dictionary, 16 ... Neuro processing unit, 17 ...
Neurotable, 18 ... Homophone buffer, 19 ... Display control unit, 20 ... Display unit.

Claims (1)

[Claims] 1. A conversion dictionary that stores conversion word information corresponding to reading information, a concept information storage unit that stores concept information of the conversion word information stored in the dictionary, and a concept information storage unit. A neuro information storage means for storing neuro information having the stored conceptual information as a node, and a control means for performing conversion using the dictionary based on the neuro information stored in the neuro information storage means are provided. A document creation device characterized by the above.