JPH0262660A - Japanese sentence processor - Google Patents

Abstract

PURPOSE:To give a proper meaning code to a compound noun by using a semantic code rule dictionary where the rules are registered to obtain the semantic codes of compound nouns and a meaning code adding device which obtains the semantic codes of compound nouns by reference to the semantic code rule dictionary. CONSTITUTION:A evaluating device 107 evaluates the analysis result candidates and decides a compound noun when detecting a part where two or more nouns or affixes are continuous among the analysis result candidates. A semantic code adding device 112 obtains the semantic code of the compound noun. Then the device 112 receives the information on the words forming a compound word from the device 107 and obtains the semantic codes. The semantic code of a compound noun is shown in a combination of the last word and the semantic code of the part excluding the last word. For example, a meaning code 'HITO' is added to a compound noun including the last word 'SHI'. Then a semantic code 'JIKAN' is added to a compound noun including a word 'CHU' added to a noun having a semantic code 'KOI'.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a Japanese text processing device that analyzes input text.

(Prior Art) Many devices that process Japanese sentences, such as Japanese word processors, analyze what words the target sentence consists of and how they are combined. In the process of analysis, it is necessary to select one from many analysis result candidates. Therefore, candidates are evaluated using some method, and the candidate with the highest evaluation is used as the final analysis result. The evaluation method is the two-clause longest match method, which uses the reading length of two consecutive phrases as a measure ([Separation of solid text and kana-kanji conversion], Makino et al., Information Processing Society of Japan Transactions Vol. 1.20. No. , 4), or the minimum clause count method that prioritizes the analysis result with the minimum number of clauses ([Morphological analysis of solid Japanese sentences using the minimum clause count method 1, Yoshimura et al., Information Processing Society of Japan Journal Vol. 1. 24. No. 1
). In addition, there is also a method of evaluation using syntactic pattern information of predicates ([Removal of ambiguity in kana-kanji conversion using case grammar], Oshima et al., Information Processing Society of Japan Transactions Vol.
o1.27. (See No. 7). These evaluation methods are not necessarily used alone, but may be used in combination as necessary.

A kana-kanji conversion device using syntactic pattern information will be described below as an example of a conventional technique. FIG. 5 is a block diagram showing an outline of a device according to an embodiment of the prior art. In FIG. 5, 101 inputs sentences in the form of character strings or sounds;
An input device outputs the text in the form of a corresponding character code; 102 is a large buffer that temporarily stores sentences inputted by the input device 101; and 103 is a word dictionary in which information necessary for analyzing sentences for each word is registered. , 104 is a grammar dictionary in which rules for composing sentences are described, 10
5 is a morphological analysis device that divides a sentence into words and combines them grammatically correctly; 106 is a morphological analysis device 105;
507 is an evaluation device that evaluates the results of morphological analysis and outputs the most appropriate analysis result; 108 is an output buffer that temporarily stores the output of the evaluation device 507; 109 is the content of the output buffer 108; An output device 110 is a control device that controls the operation of the entire device.

The input device 101 can be realized by a keyboard, a voice recognition device, etc., and the output device can be realized by a 109 display, a printer, a voice synthesis device, etc.

FIG. 7 is a conceptual diagram showing an example of the structure of the word dictionary 103.

In this example, pronunciation, spelling, part of speech, meaning code, and syntactic pattern information are registered for each word. The semantic code in this example is a code that divides nouns and affixes into several types based on meaning and usage. Furthermore, the syntactic pattern information in this example is given to a predicate, and the syntax in which the predicate is used is expressed by a combination of a case particle and a semantic code.

FIG. 8 is a conceptual diagram showing an example of the structure of the grammar dictionary 104.

In the example of FIG. 8, the grammar dictionary 104 takes the form of a matrix, and when a word whose part of speech is b can be connected immediately after a word whose part of speech is a, the element (a, b) is 1. become,
If connection is not possible, the value is 0. Furthermore, in this example, whether or not the beginning of a sentence, the end of a sentence, a bunsetsu break, and words of other parts of speech can be connected are also registered in the same way.

Here, the operation of the conventional example will be explained by taking as an example the case where the sentence ``Koiwotsuta'' is converted into a sentence containing kanji and kana. However, the meaning code and syntax pattern information necessary in the explanation are those shown in FIG. 7.

FIG. 6 is a flowchart outlining the operation of the prior art.

First, the sentence ``J'' is input from the input device 101 and stored in the input buffer 102 (601
). Then, when the input user instructs conversion to a sentence containing kanji and kana, or when predetermined conditions such as the number of input characters and punctuation marks are met, the control device 110
instructs the morphological analysis device 105 to start morphological analysis.

The morphological analysis device 105 reads a sentence from the power buffer 102 according to instructions from the control device 110 and reads the sentence from the word dictionary 103.
Extract words using . In this example, it is assumed that the following words have been extracted (parts of speech are shown in parentheses).

koi (noun), intentional (noun), individual (suffix), thick (adjective stem), shot (1st verb stem), ku (adjective conjugation ending),
wo (case particle), fishing (action word stem), product (ma action word stem), tsu (action conjugation ending), ta (auxiliary verb), and ta (noun).Furthermore, using the grammar dictionary 104, Perform a test and select the correct combination. The results of the grammar test are then output and stored in the intermediate buffer 106 (602).

FIG. 9 is a conceptual diagram showing possible connection relationships between words in this example. Note that the numbers assigned to each word in FIG. 9 correspond to the word numbers in FIG. 10, which will be described below. FIG. 10 is a conceptual diagram showing an example of the output of the morphological analysis device 105. In Figure 10, word number A is the number of the word being output, word information B is information obtained from the word dictionary such as pronunciation and spelling, and pointer C is the number A that indicates the word connected after that word. It is.

The control device 110 monitors the morphological analysis device 105, and when the morphological analysis is completed to the end of the sentence, the evaluation device 507
instructs to start operation.

Upon receiving the instruction from the control device 110, the evaluation device 507 evaluates the morphological analysis results (603). FIG. 2 is a flowchart showing an example of the evaluation procedure.

The evaluation device 507 reads the data shown in FIG. 10 from the intermediate buffer 106 (201), traces the word pointer from the beginning of the sentence to the end of the sentence, and selects ``carp'' as a candidate for the analysis result. Create (202
). Here, "/)+ represents a word break, +1, 11
represents a break between phrases. This candidate ■ is compared with the syntactic pattern information for "fishing" [person] is [animal]. Since the meaning code for "carp" is [animal], the [carp l o] part of candidate ■ matches the [[animal] j part of the syntactic pattern information. Therefore, the evaluation value of candidate ■ is increased. (203). From the data in Figure 10, we can obtain another candidate: ``Purpose ``; As in the case of candidate ■, the syntactic pattern information is compared with candidate ■. [The semantic code of intentional J is [abstract], so
Candidate ■ and syntactic pattern information do not match. Therefore, the evaluation value of candidate ■ is lowered. In this example, since there are no other candidates, the evaluation values of candidate ■ and candidate ■ are compared, and candidate ■ with a higher evaluation value is output from the evaluation device 501, and the output buffer 1
08 (204).

Finally, the output device 109 outputs the sentence "I caught a carp" mixed with kanji and kana (304).

In this way, sentences can be analyzed by combining the syntactic pattern information of predicates and the semantic classification of nouns.

(Problem to be solved by the invention) Some nouns are compound nouns made by combining several nouns or affixes. Because there are so many compound nouns and new ones are sometimes created, some of them are not registered in word dictionaries. As a result, conventional techniques could not be used as is because such compound nouns were not given semantic codes.

As a method of giving a meaning code to a compound noun in order to take advantage of the conventional technology, there is a method of using the meaning code of the last word constituting the compound noun. For example, [experiment device]
is given the semantic code [object] of ``device'', and the semantic code [location] of [room] is given to ``principal's room''. Many compound nouns can be given the correct semantic code in this way. However, for some compound nouns, the above method cannot give a correct semantic code, which impairs the effectiveness of the evaluation method using syntactic pattern information.

As an example, consider converting the sentence ``Taikichu wo Hikosuru'' into kana-kanji. FIG. 11 is a conceptual diagram showing an example of the result of morphological analysis of this sentence. Assuming that the results shown in FIG. 11 are obtained, the following two analysis candidates are obtained from this data.

■ [Flying in the atmosphere] These two candidates are evaluated using syntactic pattern information and semantic codes. However, the meaning codes of each word are: atmosphere: [object 1] waiting: [action] junior high school [place],
[The syntactic pattern information of flying J is [Object 1 is [location]
Suppose that . To evaluate the two candidates, we need the semantic codes of ``1 in waiting l'' and ``J in atmosphere l.'' Using the semantic code of the last word, both compound nouns have the following meanings:
[Location], which is a semantic code of "inside", is given. As a result, both candidates match the ``location 1'' portion of the syntax pattern information and obtain the same evaluation value. As a result, this method determines which candidate is output.
- Unable to make decisions voluntarily. In this way, conventional methods cannot give correct semantic codes to compound words, which impairs the effectiveness of syntactic pattern information.

■[Standing on standby; flying] However, the meaning code for "Standing on" is [Time J,
[[Location] is appropriate as the meaning code for J in the atmosphere. If an appropriate meaning code is given in this way, only the candidate ■ matches the syntactic pattern information, and the result of kana-kanji conversion becomes [J flying in the atmosphere.

SUMMARY OF THE INVENTION An object of the present invention is to solve the conventional problems and provide a Japanese sentence processing device that can give appropriate meaning codes to compound nouns.

(Means for Solving the Problems) A Japanese text processing device according to the present invention includes an input device for inputting sentences, and a word dictionary that registers word pronunciations, spellings, parts of speech, as well as semantic codes and syntactic pattern information. a grammar dictionary in which rules for sentence structure are registered; a morphological analyzer that performs a morphological analysis on sentences input from the input device using the word dictionary and the grammar dictionary; Rules for determining the meaning code of a compound noun in a Japanese sentence processing device having an evaluation device that evaluates and determines an optimal analysis result, and an output device that outputs the analysis result determined by the evaluation device. and a semantic code addition device that receives information on word strings constituting a noun from the evaluation device and obtains a meaning code of a compound noun using the semantic code rule dictionary. shall be.

(Operation) In the present invention, rules for determining the meaning code of a compound noun from a word string constituting the compound noun are registered in advance in a meaning code rule dictionary. If a compound noun is found in the results of morphological analysis, a semantic code is determined using compound rules for the words that make up the compound noun. When evaluating the morphological analysis results, this semantic code is used to compare it with the syntactic pattern information, and the one with the highest evaluation is taken as the analysis result.

(Example) An example of the present invention will be described below. This embodiment is an example in which the present invention is applied to a kana-kanji conversion device. FIG. 1 is a block diagram showing an outline of an embodiment of the present invention. In FIG. 1, the same parts as in FIG. 5 are designated by the same reference numerals. Further, 107 is an evaluation device that inputs the results of morphological analysis from the intermediate buffer 106, evaluates them, and outputs the most appropriate analysis results, and 111 is a semantic code rule in which rules for obtaining the semantic code of a compound noun are registered. A dictionary 112 is a semantic code adding device that receives information on a word string constituting a noun from the evaluation device 107 and uses the semantic code rule dictionary 111 to obtain a semantic code for the noun.

Figure 3 (a) and (b) are meaning code rule dictionary 111-2.
FIG. 2 is a conceptual diagram showing two configuration examples. In the example of FIG. 3(a), the meaning code of a compound noun is expressed by a combination of the last word and the meaning code of the part excluding the last word. For example, a compound noun with the word ``Mr.'' at the end is given the meaning code [person], and a compound noun with the word ``F middle'' attached to a noun with the meaning code ``act 1'' is given [time]. Give the meaning code. Figure 3 (
In b), the same rules as in FIG. 3(a) are registered for the case where the meaning code of the compound noun is different from the last meaning code. In the case of FIG. 3(b), the meaning code of the last word is used as the meaning code of a compound word for which no rule is registered in the meaning code rule dictionary.

Next, the operation of the embodiment of the present invention will be explained.

The flow of operation of this embodiment is the same as that shown in FIG. Note that the description of devices that operate in the same manner as in the prior art example will be omitted here, and the operations of the evaluation device 107 and the meaning code addition device 12 will be described.

The evaluation device 107 evaluates candidates of analysis results, similar to the evaluation device 507 of the prior art embodiment described above.

However, if there is a part in which two or more consecutive nouns or affixes are included in the analysis result candidates, that part is regarded as a compound noun, and the meaning code addition device 112 obtains the meaning code of the compound noun.

The semantic code addition device 112 receives information on words forming a compound noun from the evaluation device 107 and obtains a semantic code. FIG. 4 is a flow chart showing an example of the operation of the meaning code adding device 112. Below, the meaning code addition device 11 when using the meaning code rule dictionary in the format shown in FIG. 3(a) will be explained.
An example of the second operation will be explained.

(a) The number of words composing a compound noun is n, and the number of words is Wl.

W2. −, Wn. (401) (b) Initialize variable i to 1. Further, the meaning code of wl is assigned to the variable S. (402) (c) If i is greater than or equal to n, the process ends. The value of S becomes the semantic code of the compound noun. (403) (d) Increase i by 1. (404) (e) Search the Wi compound rule from the meaning code rule dictionary. (405) (f) From the value of S, that is, the semantic code of the noun consisting of Wl, . . . , Wi-1, and the compound rule of Wi, Wl, .
・Find the meaning code of the noun consisting of Wi, and substitute it into S.

(g) Return to (c).

When the meaning code of the compound noun is obtained, the evaluation device 107 evaluates the candidate by comparing the candidate with the syntactic pattern information, and takes the candidate given the highest evaluation value as the analysis result.

As a specific example, we will show the operation of the evaluation device 107 and the semantic code addition device 112 when converting the above-mentioned sentence ``Taikichu wo Hikosuru J'' into a sentence containing kanji and kana.However, the semantic code and syntactic pattern information are In FIG. 7, it is assumed that the compound rules are registered as shown in FIG. 3(a).

First, it is assumed that the data shown in FIG. 11 is obtained as a result of morphological analysis. From this data, [waiting l in l;
A candidate (candidate ■) is created. Here, ゛(1) I represents a word break, and “f′ represents a phrase break.
In order to obtain the meaning code of "中", the meaning code adding device 11
Send information about "standby" and "medium" to 2.

The semantic code addition device 112 obtains a semantic code using the following procedure.

(a) Put the meaning code [act] of "wait" into variable S. (402) (b) Obtain the rule "F Medium" from the meaning code rule dictionary 111. (405) Obtain the meaning code [time] from the rule [in (C) E] and the value [act] of S, and put it in S. (406) (d) Since [middle] is the last word, the process is finished and [time] is set to [wait l].
The meaning code is for 1st year of middle school. (403) Since the meaning code for "waiting l" is found as [time], candidate ■ is compared with [person], which is syntactic pattern information for "flying", and [place 1]. Since candidate ■ does not match the syntactic pattern information, the evaluation value of candidate ■ is lowered. Subsequently, [l in the atmosphere
A sentence candidate (candidate ■) is created: "I will fly."

Using the semantic code addition device 112, obtain [location] as the semantic code for the clause [atmosphere I in J] using the same procedure, and compare it with the candidate ■ and the syntactic pattern information. Then [in the atmosphere]
Since the J part of the syntax pattern information [[location] matches J, the evaluation value of the candidate ■ is increased. Compare the evaluation values of the two sentence candidates and output the candidate ■ with the higher evaluation value to the buffer 108
Output to.

In the end, the sentence ``to fly a horse'' becomes ``
``Fly through the atmosphere.''

The present invention has been explained using Kashi-Kanji conversion as an example, but the present invention is also applicable to Kanji-Kana conversion, sentence-to-speech conversion, character recognition post-processing, voice recognition post-processing, and Japanese analysis stage of machine translation. It can be widely applied to devices that analyze Japanese sentences, such as

(Effects of the Invention) As described above, in the present invention, by providing a means for giving a meaning code to a compound noun, it is possible to give an appropriate meaning code to the compound noun. As a result, it is now possible to accurately analyze sentences that previously could not be analyzed correctly.

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram of an apparatus according to an embodiment of the invention. FIG. 2 is a flowchart showing an example of the evaluation procedure. FIG. 3 is a conceptual diagram showing an example of the structure of a meaning code rule dictionary. FIG. 4 is a flowchart showing an example of the operation of the meaning code adding device. FIG. 5 is a schematic block diagram of an apparatus according to an embodiment of the prior art. FIG. 6 is a flow chart showing an overview of the operation of the prior art. FIG. 7 is a conceptual diagram showing an example of the structure of a word dictionary. FIG. 8 is a conceptual diagram showing an example of the structure of a grammar dictionary. FIG. 9 is a conceptual diagram showing connection relationships between words. FIG. 10 is a conceptual diagram showing an example of output from the morphological analysis device. FIG. 11 is a conceptual diagram showing an example of the results of morphological analysis. In FIG. 1, 101...input device, 102...power buffer, 1
03...Word dictionary, 104...Grammar dictionary, 105.
...Morphological analyzer, 106...Intermediate buffer, 10
7...Evaluation device, 108...Output buffer, 109
. . . output device; 110 is a control device; 111 . . . meaning code rule dictionary; 112 . . . meaning code adding device.

Claims (1)

[Claims] An input device for entering sentences, reading and writing words,
In addition to parts of speech, there is also a word dictionary in which semantic codes and syntactic pattern information are registered, and a grammar dictionary in which rules for sentence structure are registered. A Japanese text comprising a morphological analysis device that performs morphological analysis, an evaluation device that evaluates the results of the morphological analysis and determines an optimal analysis result, and an output device that outputs the analysis result determined by the evaluation device. A processing device receives a meaning code rule dictionary in which rules for determining the meaning code of a compound noun are registered, and information on a word string constituting the compound noun from the evaluation device;
A Japanese text processing device comprising: a semantic code adding device that obtains a semantic code of a compound noun using the semantic code rule dictionary.