JPH01156866A - Processing system for japanese sentence - Google Patents

Abstract

PURPOSE:To minimize influence given to the analysis of a Japanese sentence as a whole by providing an analysis-unable character string processing part to give further analysis to a character string to which the analysis is impossible due to a fact that the Japanese sentence includes such words that are not included in a dictionary or have errors. CONSTITUTION:An analysis-unable character string processing part 40 analyzes as much as possible an analysis-unable character string detected by a morpheme analyzing part 20. Thus the analysis-unable characters are limited and specified and therefore the analyzing efficiency of Japanese sentences is improved. Furthermore, the correct parts of speech of the extracted analysis-unable character string can be limited up to a degree where no problem is produced even though said parts of speech are estimated as a single part of speech. Thus the analysis- unable character string can be estimated as proper parts of speech and processed. As a result, such a sentence including the words contained in a dictionary or having errors, etc., is also allowed for analysis of Japanese sentences.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a method for analyzing Japanese sentences using a computer, and more specifically, it improves the accuracy of analysis even for Japanese sentences that include words not found in dictionaries or misspellings. It concerns the Japanese sentence processing method.

[Conventional technology]

FIG. 5 shows a schematic diagram of a conventional computer-based method for analyzing Japanese sentences. In FIG. 5, 10 is a data input section, 20 is a morphological analysis section, and 50 is an output editing section.

The data input unit 10 consists of a Kanji OCR, Bentouch, tablet, keyboard, etc., and reads Japanese character strings to be analyzed. The morphological analysis unit 20 performs morphological analysis on the read Japanese character string using a grammar dictionary and a word dictionary prepared in advance, and performs recognition of words such as segmentation into one segment. The output editing section 50 edits and outputs the Japanese character string according to the processing result of the morphological analysis section 2o. In the case of Japanese sentences that include words that are not found in the dictionary or spelling errors, the morphological analysis unit 20 detects the entire clause as an analysis unit as an unanalyzable character string, and the output editing unit 50 converts the unanalyzable character string into its own. Well, it will be output in units of clauses.

[Problem that the invention seeks to solve]

In conventional technology, when a Japanese sentence is analyzed by a computer, if there are errors such as input errors in the Japanese sentence, or there are words that are not in the dictionary, the impact cannot be limited.
The phrase or entire sentence that contained that part was an unparsable character string. For this reason, 1. For example, in a voice output system that analyzes Japanese sentences, certifies them in word units, adds pronunciations, accents, and pauses, and reads them out using synthesized speech, even if there is actually one unanalyzable character, Because the characters could not be identified, the reading of clauses and entire sentences became incorrect, reducing the rate of correct continuation.

Furthermore, automatic Japanese sentence error detection systems that take advantage of the fact that errors may exist where the analysis fails can only point out long character strings, such as when used to proofread newspaper articles. Since it is necessary to process a large amount of data in a short period of time, there is a problem in that the labor-saving effect is reduced due to the task of finding errors in long character strings.

An object of the present invention is to provide a Japanese sentence processing method that can limit and specify character strings as much as possible even in sentences that include words that are not found in the dictionary or errors, thereby minimizing the impact on the analysis of the entire sentence.

[Means for solving problems]

In the method of processing Japanese sentences using a computer, the present invention has a morphological analysis section that recognizes and divides the Japanese sentence to be processed into words, phrases, etc., and also uses a morphological analysis section that recognizes and divides the Japanese sentence to be processed into words, phrases, etc. An unanalyzable character string processing unit is provided that analyzes character strings more deeply.

[For production]

The unanalyzable character string processing unit analyzes the unanalyzable character string detected by the morphological analysis unit as much as possible from the beginning until it can be analyzed.

If the beginning of a character string that cannot be analyzed is a katakana or alphabetic character, the entire consecutive character string of the same character type is detected as an unanalyzable character string. In addition, in the case of other character types (Kanji, Hinagana), the next character after the first character is analyzed again, and if it cannot be successfully analyzed after that, it is skipped and not included in the analysis. Assume that one character is an unanalyzable target. If you are still unable to parse the following character string, skip the first character and start parsing from the next character. Continue this process until you reach the end of a clause or sentence, then skip the first character and start parsing from the next character. Detects strings as unparsable.

In this way, it is possible to further analyze the long unanalyzable character strings detected by the morphological analysis unit to identify the error location, and to limit the unanalyzable portions to short character strings.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the present invention, which is applied to an audio output system, and FIG. 2 is a diagram showing a specific example of its processing in association with each part of FIG. 1. .

In FIG. 1, 10 is a Japanese sentence data input section, 20 is a morphological analysis section, 30 is a reading assignment section, 40 is an unanalyzable character string processing section, 50 is an output editing section, and 60 is a synthesized speech generation section. The unanalyzable character string processing unit 40 is further divided into an unanalyzable character limiting unit 41, an unanalyzable character part-of-speech assigning unit 42, a reading assignment unit 43, and an accent/pause assigning unit 44.

Suppose that the Japanese sentence (Japanese character string) ``Light like Calvik'' is read from the data input unit 10 (
■). Now let's assume that ``Kyalbik'' is a word that is not in the dictionary.

The morphological analysis unit 20 uses a grammar dictionary and a word dictionary prepared in advance to divide ``light like kyalbik'' into terms and phrases, but since ``kalbik'' is not in the dictionary, ” is an unparsable character string, and “light” is recognized as an adjective word (■)
. For the word "light" that has been recognized by the morphological analysis unit 20, the reading assignment unit 30 assigns information on the pronunciation and appropriate accents and poses (■).

On the other hand, the unanalyzable character string "Kyalbik ni ni" is sent to the unanalyzable character string processing section 40. First, the unanalyzable character limiting unit 41 limits the unanalyzable characters. In this case, if the katakana character ``karubik'' is an unparsable character, then ``ni'' is the case particle ``no''.
and can be analyzed into the auxiliary verb ``yo ni'' (■). Next, in order to add accents and poses to the unanalyzable character string as naturally as possible, the unanalyzable character part-of-speech adding unit 42 uses the case particle "no".
After giving an appropriate part of speech (in this case, a common noun) that can be connected to the unparsable character string "Kalbik" (■) and making it look like a common noun found in the dictionary, the Ate-yomi adding unit 43
, the accent/pose giving unit 44 gives an appropriate reading (in this case, the katakana reading since it is a katakana character) and an accent/pose (■, ■).

The output editing section 50 edits the outputs of the pronunciation adding section 30 and the accent/bose adding section 44 together (■). The editing result in the output editing section 50 is sent to the synthesized speech generating section 60 to become speech.

In this way, by limiting the characters that cannot be parsed, even if a sentence containing a string that cannot be parsed is input, the part of speech of that part can be appropriately estimated, so the reading can be output in a form as close to natural as possible. Can be done.

FIG. 3 is a block diagram of another embodiment of the present invention, in which the present invention is applied to an error detection system, and FIG. 4 shows a specific processing example thereof in correspondence with each part of FIG. 3. This is a diagram.

In FIG. 3, 10 is a data input section, 20 is a morphological analysis section, 40 is an unanalyzable character string processing section, and 70 is an error detection section. The unanalyzable character string processing section 40 includes an unanalyzable character limiting section 41.

Now, the Japanese sentence ``I don't understand everything.'' is read from the data input section 1o, and the Japanese sentence ``I don't understand everything.'' is read.
Suppose that "ta" is a typo (■).

This Japanese sentence is analyzed by the morphological analysis unit 10, and is recognized as the adverb "all" and the phrase "I don't understand" which is a string of characters that cannot be analyzed (■). Among these, the unanalyzable character string [I just don't understand it] is sent to the unanalyzable character string processing section 40.

The unanalyzable character limiting unit 41 of the processing unit 40 manages to analyze even the characters whose analysis fails, and skips the first character of failure, “ta” (there is no “ta” that can be connected to the force-transforming verb “ko”). 5. By sending the analysis results of the unanalyzable character limiting section 41 to the error detection section 70, the single character "ta" is treated as an error. Detectable (■) By limiting unanalyzable characters in this way, erroneous characters can be identified.

〔Effect of the invention〕

As explained above, in the conventional analysis of a single Japanese sentence, if there were words that were not found in the dictionary or sentences that contained misspellings, the entire minimum unit of processing, such as five clauses, had to be treated as an unanalyzable character string, but the present invention Furthermore, since it becomes possible to limit and specify characters that cannot be analyzed, the efficiency of analyzing Japanese sentences is improved. Furthermore, it is possible to limit the correct part of speech of extracted unparsable character strings to the extent that there is no problem even if the correct part of speech is estimated as a single part of speech, and it is possible to process unparsable character strings by inferring them to an appropriate part of speech. It has the advantage of allowing Japanese sentences to be analyzed while allowing sentences that contain typographical errors.

General sentences used in actual work often contain words that are not found in dictionaries or sentences that include misspellings, so the effects of the present invention are of great significance. For example, when applied to a Japanese speech output system, even if there is a character string that cannot be parsed, it will be possible to read the text close to the correct answer. Furthermore, the Japanese sentence error detection system can detect errors by limiting the range that includes them.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of one embodiment of the present invention, FIG. 2 is a diagram showing a specific processing example of FIG. 1, FIG. 3 is a block diagram of another embodiment of the present invention, and FIG. Figure 5 shows a specific processing example in Figure 3.
The figure is a block diagram of a conventional system. 10...Data input section, 20...Morphological analysis section. 40...Unanalyzable character string processing section. 1st L11 22nd ゛meh? vinegar.

Claims (1)

[Claims] (1) In the method of processing Japanese sentences using a computer, the Japanese sentences to be processed are processed using a grammar dictionary, word dictionary, etc.
A morphological analysis unit that recognizes and divides into phrases, etc., and a deeper analysis of character strings that cannot be analyzed by the morphological analysis unit due to words that are not found in the dictionary or errors in Japanese sentences, to limit the number of character strings that cannot be analyzed. A Japanese sentence processing method characterized by having an unanalyzable character string processing unit that performs specification.