JPH0682364B2 - Japanese sentence processing method - Google Patents

Description

TECHNICAL FIELD The present invention relates to a Japanese sentence analysis processing method by a computer, and in particular, improves the analysis accuracy even for Japanese sentences including words or typographical errors that are not in the dictionary. It relates to the Japanese sentence processing method.

[Conventional technology]

FIG. 5 shows a conventional schematic configuration diagram of a Japanese sentence analysis processing system by a computer. In FIG. 5, 10 is a data input section,
Reference numeral 20 is a morphological analysis unit, and 50 is an output editing unit.

The data input unit 10 is Kanji OCR, pen touch, tablet,
It consists of a keyboard, etc., and reads the Japanese character string to be analyzed. With respect to the read Japanese character string, the morphological analysis unit 20 performs morphological analysis using a grammar dictionary or word dictionary prepared in advance, and carries out word recognition, segmentation, and the like. The output editing unit 50 edits and outputs the Japanese character string according to the processing result of the morphological analysis unit 20. Here, in the case of Japanese sentences containing words or typographical errors that are not in the dictionary, the morphological analysis unit 20
In (1), the entire phrase in the parse unit is detected as an unparseable character string, and the output editing unit 50 outputs the unparseable character string in the bunsetsu unit.

[Problems to be solved by the invention]

In the prior art, when analyzing a Japanese sentence by a computer, if there is an error such as a typographical error in the Japanese sentence, or if there is a word that is not in the dictionary, the effect cannot be limited,
The clause including the part and the whole sentence were the unparseable character string. Therefore, for example, in a voice output system that analyzes a Japanese sentence, certifies it in word units, adds readings and accents / pauses and reads aloud with synthetic speech, even if one character cannot be analyzed, Since it was not possible to identify the phrase, the reading of the phrase or the whole sentence was incorrect, and the correct reading rate was reduced.

In addition, even in the Japanese sentence error automatic detection system that uses the fact that there may be an error where the analysis does not go well, only a long unit of character string can be pointed out, and it is used for reviewing newspaper articles etc. Needs to process a large amount of data in a short time, and there is a problem in that the labor-saving effect is reduced due to the work of finding an error from a long character string.

An object of the present invention is to provide a Japanese sentence processing method capable of limiting the character string as much as possible even in a sentence including a word or an error which is not in the dictionary and specifying it to minimize the influence on the analysis of the entire sentence.

[Means for solving problems]

The present invention, in a method of processing Japanese sentences by a computer, in addition to a morphological analysis unit for accrediting and dividing the Japanese sentence to be processed into words, clauses, etc., it is impossible to analyze because there are words or errors in the Japanese sentence that are not in the dictionary. An unanalyzable character string processing unit is provided for deeper analysis of the character string.

The unparsable character string processing unit analyzes the unparsable character string detected by the morpheme analysis unit as much as possible from the beginning until it can be analyzed. If the beginning of the character string that cannot be analyzed by any means is katakana or alphabet, the entire continuous character string of the same character type is detected as an unanalyzable character string. For other character types (Kanji, Hiragana), the character next to the first character is analyzed again, and after that, one character that could not be analyzed successfully and was skipped was not analyzed. Is a target that cannot be analyzed. If you still can not parse the subsequent character string, skip the first character further and start the analysis from the next character, perform this processing until the end of the phrase or the end of the sentence, or if you can parse successfully Detects the column as unparsable.

[Action]

According to the present invention, since an unanalyzed long character string detected by the morphological analysis unit is further analyzed, an error location can be specified, and an unanalyzable portion can be limited to a character string of a short unit.

〔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 a voice output system, and FIG. 2 is a diagram showing a specific processing example thereof in association with each part of FIG. .

In FIG. 1, 10 is a Japanese text data input unit, 20 is a morphological analysis unit, 30 is a reading addition unit, 40 is an unanalyzable character string processing unit, 50 is an output editing unit, and 60 is a synthetic speech generation unit. The unparseable character string processing unit 40 is a further unparseable character limiting unit.
41, an unanalyzable character part-of-speech assigning unit 42, an attitudinal reading imparting unit 43, and an accent / pose imparting unit 44.

Now "lighter like a Calbic" than the data input section 10
Suppose that the Japanese sentence (Japanese character string) is read (). Here, suppose "Calbic" is a word that is not in the dictionary.

In the morphological analysis unit 20, using a grammar dictionary and a word dictionary prepared in advance, "Calbic-like light" is word-segmented and phrase-segmented. However, since "Calbic" is not in the dictionary, "Calbic-like" is used. The segment of is a non-parsable character string, and "" is recognized as an adjective word "". For the word "light" that has been recognized by the morphological analysis unit 20, the reading addition unit 30 adds reading and appropriate accent and pose information ().

On the other hand, the unparseable character string “Like Calbic” is sent to the unparseable character string processing unit 40. First, the unparsable characters are limited 41 to limit the unparsable characters. In this case, if "Calbic" starting with the Katakana character is a non-parsable character of one unit, "like you" can be analyzed into the case particle "no" and auxiliary verb "like" (). Next, in order to add accents and poses to the unparseable character string as naturally as possible, the unparseable character part-of-speech assigning unit 42 creates an appropriate part-of-speech (general noun in this case) that can be connected to the case particle "no". After giving () to "Calbic" and making it like a general noun in the dictionary, an appropriate reading (in this case katakana reading because it is katakana character) and accent are added in the reading reading adding part 43 and accent / pose adding part 44.・
Add a pose (,).

The output editing unit 50 edits the outputs of the reading adding unit 30 and the accent / pose adding unit 44 together (). The editing result of the output editing unit 50 is sent to the synthesized voice generating unit 60 to be a voice.

In this way, by limiting unparsable characters, even if a sentence containing a character string that cannot be parsed is input, the part of speech of that part can be estimated appropriately, so the pronunciation should be output in a form that is as natural as possible. You can

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

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

Now, it is assumed that the Japanese sentence "I just don't understand everything." Is read from the data input unit 10 and that "ta" in the "Kota just" is a typographical error (). The Japanese sentence is analyzed by the morphological analysis unit 10 and is limited to the adverb "all" and the phrase "not understandable kotada da" which cannot be analyzed (). Of these, the unparsed character string “I just knew it” is sent to the unparsed character string processing unit 40. The unanalyzable character limiting unit 41 of the processing unit 40 manages to further analyze even the characters for which the analysis fails, and the failure leading character "ta"
It skips (there is no "ta" that can be connected to the ka-inflectional verb "ko"), analyzes it again, and succeeds in parsing "simple" (). By sending the analysis result of the unparsable character limiting unit 41 to the error detecting unit 70, one character "ta" can be detected as an error (). By limiting unparsable characters in this way, error characters can be limited.

[Effects of the Invention] As described above, when there is a sentence including a word or a typographical error that is not in the conventional dictionary in the analysis of character sentences, the entire minimum unit of processing such as bunsetsu must be an unparseable character string. However, according to the present invention, it becomes possible to limit and specify the unparsable characters among them, so that the analysis efficiency of Japanese sentences is improved. Furthermore, it is possible to limit the correct part-of-speech of the extracted unparsable character string to the extent that there is no problem even if it is estimated as a single part-of-speech, and it is possible to estimate the unparsable character string to an appropriate part-of-speech and process it. There is an advantage that it is possible to analyze Japanese sentences by allowing sentences including typographical errors.

The general sentence used in actual business often has a sentence including a word or a typographical error that is not in the dictionary, and the effect of the present invention is significant. For example, if it is applied to a Japanese sentence voice output system, it is possible to accurately read the reading even if there is a character string that cannot be analyzed. Also, the Japanese sentence error detection system can detect an error in a limited range.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a diagram showing a concrete processing example of FIG. 1, FIG. 3 is a block diagram of another embodiment of the present invention, and FIG. The figure which shows the concrete example of processing of Figure 3, the 5th
The figure is a block diagram of a conventional method. 10 ... Data input section, 20 ... Morphological analysis section, 40 ... Unanalyzable character string processing section.

 ─────────────────────────────────────────────────── --- Continuation of the front page (72) Inventor Satoru Ikehara 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation (56) Reference JP-A-61-134877 (JP, A) JP 62-90760 (JP, A) JP 62-209659 (JP, A)

Claims (1)

[Claims] 1. A method for processing Japanese sentences by a computer, a morphological analysis unit for dividing a Japanese sentence to be processed into words, clauses, etc. by using a grammar dictionary, word dictionary, etc., and a word not included in the Japanese sentence in the dictionary. For a character string that cannot be analyzed by the morphological analysis unit due to an error or an error, if the first character type of the character string is Katakana or the alphabet, the consecutive character strings of the same character type are detected as unanalyzable character strings. However, when the first character type of the character string that cannot be parsed is other than katakana or alphabet, the character next to the first character is analyzed again as the head of the character string to exclude the first character from the analysis, and the result is the result. If it cannot be parsed, the next character is parsed sequentially as the first character, and if it can be parsed, the character string excluded from the analysis is detected as an unparseable character string. Text processing system date and having a 析不 ability character string processing unit.