JPS62121570A - Continued clause conversion processing system based on connection probability - Google Patents

Abstract

PURPOSE:To improve operability by providing a means which extracts the connection probability of each word converted from a probability information storing means and performs a probability calculation, and the means which regulates the priority order of a conversion based on a calculated result. CONSTITUTION:A bit of connection probability information which represents the easiness of a connection between each group is stored in advance in a form of matrix table reaccessed with a preceding bit of group specification information and a succeeding bit of group specification information at a connection probability information storing part 16. A probability calculation part 17 finds the bit of connection probability information of words adjacent with each other from each group specification of the words resulted from the retrieval of a dictionary, and performs the probability calculation, such as the calculation of a probability as a whole in which each probability is multiplied with each other, or that of a forecasted probability, and selects and displays preferentially the word with a high probability. Thereby, the easiness in the connection of the words can be evaluated from the probability point of view, and an appropriate expression can be selected, therefore, a correct expression can be obtained in a short time, thereby the operability being improved.

Description

[Detailed Description of the Invention] [1 Already Required] The continuous clause conversion processing method based on the connection probability of the present invention is used in a Japanese document creation device that performs continuous clause conversion to convert words, which are units of conversion, into grammatical elements and semantics. Elements that have the same properties are grouped together, and probability information indicating the ease of connection between each group is stored in a matrix for each combination of groups around 1, and the dictionary search results are By determining the connection probability information of adjacent words, words with a high probability are prioritized and selected as conversion targets, so that correct expressions can be quickly obtained.

[Industrial application field]

The present invention is a so-called Japanese word processor that edits and creates a single document by inputting Japanese pronunciations containing multiple words, searching a dictionary, and converting them into sentences containing kanji.
The present invention relates to a continuous clause conversion processing method based on connection probability in a Japanese document creation device, which allows a correct expression to be quickly obtained regarding continuous clause conversion.

[Conventional technology]

In a Japanese document creation device, the pronunciation is input by human input of kana or Roman characters, conversion of kana to kanji is performed with reference to a dictionary, and a sentence containing kanji is created. Recently, devices have begun to be used that are capable of converting not only words and clauses, but also compound words and consecutive clauses consisting of a plurality of clauses. In this explanation, compound word conversion will also be explained as being included in continuous clause conversion.

Conventional Japanese document creation devices that perform continuous clause conversion perform a dictionary search using the so-called longest-lost method, which divides the readings that yield the longest conversion unit, and displays sentences containing kanji based on the search results. By inputting a conversion key or the like, conversion results corresponding to the mth first-order longest conversion are displayed in sequence.

[Problem that the invention seeks to solve]

In the conventional method described above, if you try to convert long readings at once, many expressions will be created depending on the way they are separated or the combination of homophones, and among them, the appropriate expression will be displayed with priority. Since it is not always the case, there are operations to select the correct expression.

The problem was that it was complicated.

The present invention aims to solve the above problems, and even when converting long readings at once, appropriate expressions are prioritized and selected, allowing the desired result to be quickly obtained, thereby improving operability. The purpose is to

[Failure to solve the problem]

FIG. 1 is a block diagram showing the basic configuration of the present invention.

In FIG. 1, 10 is a reading key, a conversion instruction key, etc. 11 is an input control unit that controls input from the keyboard 10; 12 is a reading division unit that divides the readings of the input continuous clause; 13 is a dictionary that searches the dictionary based on the divisions of readings divided by the reading division unit 12; Search section, 1
4 is a dictionary in which readings and corresponding kanji code strings are registered.

Reference numeral 15 denotes a group type assigning unit that provides predetermined group type information to which the word belongs for each word to be converted;
Reference numeral 6 denotes a connection probability information storage unit that stores probability information regarding the ease of connection between group types to which adjacent words belong in reading conversion.

Reference numeral 17 denotes a probability calculation unit that refers to the contact 'ff and probability information storage unit 16 and performs probability calculation regarding the appropriateness of the conversion according to the group type information for each reading division based on the conversion result of the dictionary search unit 13, and 18 a probability calculation unit. 19 is a display control unit that performs control to display the conversion result selected by the conversion selection unit 18; 20; represents the display where reading and conversion results are displayed.

In the case of the present invention, words (words or phrases) that are units of conversion are grouped in advance into groups that have the same properties in terms of grammatical and semantic elements.

Note that, if necessary, the frequency of appearance may also be considered as a basis for grouping. Group type information for each word is
At the time of dictionary creation, each word corresponding to the pronunciation is registered in the dictionary 14 in advance. Furthermore, the group type for items that are not registered as they are in the dictionary 14 and are processed as nine clauses is assigned by the group type assigning unit 15 to Guinamisoku based on the association with particles and the like.

Connection probability information indicating ease of connection between each group is stored in the connection probability information storage unit 16 in advance in the form of a matrix table accessed by previous group type information and subsequent group type information. tα is done.

The probability calculation unit F obtains connection probability information of each adjacent word from each group type of words in the dictionary search result, and calculates the probability such as the overall probability or predicted probability by multiplying each probability. The probability calculation unit 17 selects the one with higher probability and displays it as one.

[Effect]

According to the present invention, the ease with which words are connected is probabilistically evaluated for the conversion result based on reading divisions, so accurate expressions are prioritized.

You will be able to obtain correct expressions early on. In addition.

If the probabilities are the same, it is preferable to determine the priority among them by using the longest-formation method or the minimum clause method with a small number of clauses in total.

C Embodiment] FIG. 2 is a diagram for explaining assignment of group types, FIG. 3 is a diagram for explaining the storage contents of the connection probability information storage unit, and FIG. 4 is according to an embodiment of the present invention. FIG. 5 is a diagram illustrating probability calculation, and FIG. 5 is a diagram illustrating conversion according to another embodiment of the present invention.

For grouping words and phrases according to the invention.

The part of speech is taken into consideration, such as nouns, verbs, adjectives/adjective verbs, adverbs, etc., and then they are prefixes.

If there are 9 nouns that can be used as suffixes, there are 9 names of people and places.

Elements such as the semantic elements of numbers, words that are likely to become compound words, and words that are difficult to form into compound words are taken into consideration.

Verbs, adjectives/adjective verbs, etc. are classified based on their conjugation forms, such as adjunctive case, adjunctive case, and final form, and clauses containing nine nouns are also categorized into nominative case, objective case, adjunctive case, and adjunctive case. , final form, etc., and a group ID indicating the group type is assigned to each group.

This group ID information is stored in advance in the dictionary 14 for each word corresponding to each reading, for example, as shown in FIG. 2(A). As for phrases, as shown in FIG. 2(b), the group type assigning unit 15 assigns them to Guinamisoku based on predetermined standards of grammatical elements.

As shown in FIG. 3, the connection probability information storage unit 16 stores adjacent front group types (ID=G1°G2.G3.
) and the adjacent rear group type (ID=G
1. Probability pH, R12, . . . indicating ease of connection that is predetermined grammatically or statistically corresponding to each combination of G2, 03, . . .

R55... is registered.

Next, connection probability calculation for determining the priority order when converting the readings (RO, R1, . . . R5) shown in FIG. 4 will be explained. The group IDs of related words are as shown in Figure 2 (a) and (b), and the contact CGtl rate between each group is determined as shown in Figure 3. .

While sequentially cutting the backward readings (RO, R1,...R5), we search the dictionary 14 in the same way as the conventional longest-form processing and search for all convertible expressions. Expressions (a) to f) shown in FIG. 4 are obtained.

In the case of the expression in Figure 4(a), the readings are RO-R3 and R4+
The group ID of the first half of the clause WO is G.
1. The group ID of the second half word W1 is G3. For 01 on the front side and 03 on the rear side, the connection probability is calculated by referring to the connection probability information storage unit 16 shown in FIG.
The overall connection probability P is R13. In the case of the expression shown in FIG. 4(b).

The group ID of the next word W2 is G4.

The connection probability P is R14.

If the reading is divided into three or more parts, the probability of the result obtained by multiplying all the reading probabilities together is the overall connection probability.

For example, in the case of the expression in FIG. 4(d), the probability P = P56 · R63, which is the product of the connection probability P56 between G5 and 06 and the connection probability P63 between G6 and G3, is the overall connection probability. It is said that

In this way, if expression candidates are selected in descending order of the calculated probability P, the most appropriate expressions will be selected in order.

By the way, as the reading gets longer, the number of ways to divide the reading also increases, so if we calculated the overall probability after finding all possible expressions, it would take two hours, which would be inconvenient. may occur. Therefore, first, as explained with reference to Figure 5, the final probability is predicted at each node that searches for one expression, and the predicted probability of the connection probability is dynamically calculated, and the one with the highest predicted probability is By proceeding with the search, it is possible to quickly select an appropriate expression.

In the example shown in FIG.
Convert the reading.

The expected probability is determined as follows.

(Expected probability) = PaXPbxPc Pa: Connection probability to the currently focused word.

Pb: Average probability when the word is the first half word.

PC: Probability for remaining reading length.

The last term in this equation, ``Probability PcJ for the length of the remaining reading,'' is a predetermined numerical value corresponding to the length so that the longer the remaining reading, the smaller the probability.
a is "1" for the first word (word or phrase)
becomes. For the second word, it is the connection probability between the first word and the second word. For the nth word,
Pa is.

The product is the sum of all connection probabilities up to the nth one.

The average probability pb is calculated from PL, P2. ....

P5 . . . is stored in correspondence with each previous group type in the connection probability information storage unit 16, and has a value that takes into account the average connection probability in that row.

In the 0th search for the pronunciation of "Kyohayoitenkides", the first group is "sect".

The second group: ``Today is 2 kyō, sutra, evil, appearance.''
The third group "Today Ikyo, Sutra, Kyoku", the fourth group "Strength", etc. are searched.

The group type of "sect" is (famous car).

This is a group of common nouns with low complexity. The group type of the second group [Nashukoha.

This indicates that it is a nominative noun clause. The group type (famous car) of the third group represents a group of common nouns that are used alone. The group type of "Strong" is (
@small), which is a group of prefixes that are generally used less frequently.

Now, the average probability pb of the group of (famous cars) is (1/4)
, the probability Pc of the length of the remaining reading ``Yoitenkides'' is (
3/4), then Pa is 1, so the expected connection probability of "sect" is.

P a x P b x P c = I X (1/4
)X(3/4)=3/16.

Assuming that the average probability of the second group [name head] is (3/4), the expected connection probability of the second group is:

P a x P b x P c = I X (3/
4) X (3/4) = 9/16.

By similar calculation of expected connection probabilities, the expected probability of the third group is (3/16), and the expected probability of the fourth group is (3/8).

Note that the probability Pc for the remaining reading length in the third group and the fourth group should be set smaller than that in the first and second groups, if the accuracy of probability calculation is to be increased. It is better.

Among the predicted probabilities up to now, the predicted probability of the second group is high, so this is selected and the 0th search is performed. The group "good, good, good" searched here has a group type of [form], which represents an adjective or an adjective verb, and the group "evening, drunkenness, kinzo" has a group type of [form]. (famous car). Although not shown.

There are also groups such as ``Se, Yoru, Yuichidai (Famous Cars)''.

Groups such as "four (number)" are searched.

When calculating the predicted probability for the group "good..." in the 0th search, first refer to the connection probability information storage unit 16,
The connection probability P with the following word group type and shape]
Find a (=2/4). Also, find the average probability P b (=3/4) of the group of [shape]. Assuming that the probability Pc regarding the remaining length is "1", the expected probability of this group is.

Pa xP bxPc=(2/4)X(3/4)X 1
= 3/8.

In addition, in the case of the group "Evening...", the average probability P
When b is (1/4) and the expected probability is calculated, (
1/8). Therefore, here, the "good..." group is selected with priority.

Similarly, in the 0th search, the connection probability of the group "It's weather, it's transcription", the probability of the group "turning point", etc. are determined. In the case of ``It's the weather, here's the transcription,'' the last reading is required.

This is not an expected probability but an actual connection probability, and the connection probability is the sum of all the probabilities of Pa up to that point. That is, (2/4) x (3/4) = 3
/ 8. Average probability Pb=1. Probability of remaining length P
It may be considered c-1. In the case of "turning point", the expected probability calculation result is (1/8).

Therefore, with the group ``Today is 1 quintillion...''.

There is a group of ``good, good, good'' and a group of ``weather.''

Expressions by connecting groups of ``transcription'' are selected preferentially, and among these, 1 is selected in order from the first one ordered by learning, for example. In the manner described above, appropriate expressions are selected with priority. In particular, according to this embodiment, since expected probabilities related to connection probabilities are handled, it is possible to quickly search for valid expressions.

〔Effect of the invention〕

As explained above, according to the present invention, a valid expression is selected by evaluating the relationship between the words or clauses before and after the second phrase in continuous clause conversion including compound words, so that the correct expression can be obtained quickly. This improves operability.

[Brief explanation of drawings]

FIG. 1 is a basic configuration diagram of the present invention, FIG. 2 is a diagram for explaining assignment of group types, FIG. 3 is a diagram for explaining the storage contents of the connection probability information storage unit, and FIG. FIG. 5 is a diagram for explaining probability calculation according to one embodiment of the invention, and FIG. 5 is a diagram for explaining conversion according to another embodiment of the invention. In the figure, 10 is a keyboard, and 11 is an input control unit. 12 is a reading division section, 13 is a dictionary search section, 14 is a dictionary, 1
5 is a group type assigning unit; 16 is a connection probability information storage unit;
17 is a probability calculation section, 18 is a conversion selection section, 19 is a display control section, and 20 is a display. Patent Applicant Fujitsu Ltd. Representative Patent Attorney Hiroshi Mori 1) (and 1 other person) Figure 1

Claims (1)

[Claims] A Japanese document creation device that edits and creates a document by inputting Japanese pronunciations containing multiple words, searching a dictionary, and converting them into sentences containing kanji. The words are grouped in advance into groups having the same properties in terms of grammatical and semantic elements, and means (14, 15) for providing each word with group type information for identifying the group to which the word belongs (14, 15). ), a probability information storage means (16) for storing probability information indicating the ease of word connection between each of the groups, corresponding to the combination of the preceding and succeeding groups; Obtain the group type information of the group to which the word to be converted corresponds to, and extract the connection probability of each word to be converted from the probability information storage means (16) based on the group type information, and calculate the probability. 1. A continuous clause conversion processing method based on connection probability, comprising: means (17) for performing the calculation, and means (18) for determining priority of conversion based on the calculation result by the probability calculation means (17).