JP2718966B2 - Error detection device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an error detection device that detects an error in an input Japanese sentence.

2. Description of the Related Art Conventionally, there is a "text error detection method" disclosed in Japanese Patent Application Laid-Open No. Sho 62-150463, which detects an incompatibility of a predicate expression corresponding to an adverb as an error. This is to roughly detect an error using the semantic classification of the adverb and the predicate (word at the end of the sentence) required by the adverb.

However, since such a method focuses only on the binary relation of an adverb and a predicate, in the case of the following sentence example, it cannot be correctly handled, and the accuracy is low. It is bad.

For example, In the case of the sentence example, according to the above-described method, by describing that "Karakushi" co-occurs with a negative expression, as in this example, since "Da" is conclusive, it is detected as an error. If you do not look at not only the word at the end of the sentence but also the meaning of the adjective verb "Dame" that constitutes the predicate immediately before it, you cannot test correctly.

Also, for example, In the sentence example, there is a negative "no", but "no"
Is semantically negative, so the predicate as a whole represents a positive meaning. Even in the case of this sentence example, it cannot be detected by the above method.

Further, for example, Is an error because it does not co-occur with this predicate "slightly" because of the meaning of the independent word of the predicate. However, according to the above method, the adverb "slightly" co-occurs with negation, so that the sentence example is determined to be correct.

Means for Solving the Problems An adverb co-occurrence dictionary storing adverbs and semantic classifications of predicates co-occurring with adverbs, a predicate semantic dictionary storing semantic classifications of words constituting the predicate, and words before and after A semantic operation table that stores semantic classifications expressed by combinations of the semantic classifications, and a semantic storage unit that stores the searched semantic classifications. An error detection processing unit including a predicate semantic operation unit and a predicate determination unit connected to the part semantic dictionary, the semantic operation table, and the semantic storage unit is provided.

The semantic classification of all words constituting the predicate expression in the input Japanese sentence is calculated by the predicate semantic operation unit using the predicate semantic dictionary, the semantic operation table and the semantic storage unit. And
Whether the adverb and the predicate expression are unmatched depends on whether the semantic classification of the entire predicate expression is co-occurred with the adverb extracted by the adverb detection unit using the adverb co-occurrence dictionary. Detected by the unit. That is, regarding the predicate expression co-occurring with the adverb, not only the word at the end of the predicate but also the meanings of all the words constituting the predicate are checked and the meanings are calculated, so that the accuracy of error detection is high. It will be.

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

First, FIG. 1 is a block diagram showing a configuration example of a Japanese sentence error detecting device. The error detection processing unit 1 of this embodiment includes an adverb detection unit 2, a predicate semantic operation unit 3, and a predicate determination unit 4, and is connected between the input unit 5 and the output unit 6. The adverb co-occurrence dictionary 7 is connected to the adverb detection unit 2. The adverb co-occurrence dictionary 7 stores, for example, as shown in FIG. 2, the notation of the adverb and the semantic classification of the predicate co-occurring with the adverb as a pair. For example, adverbs such as "karakushi" and "zenzen" are given a co-occurrence predicate meaning classification of "negative".
That is, the “semantic classification” in the present invention means a classification such as “negative” or “affirmative”. Also, the predicate semantic operation unit 3 includes a predicate semantic dictionary 8, a semantic operation table 9, and a semantic storage unit 10.
And are connected. The predicate meaning dictionary 8 stores, for example, the notation of words that can form a predicate, the part of speech, and the semantic classification, as shown in FIG. For example, for the predicate notation “none”, it is registered that the part of speech is “auxiliary verb” and the semantic classification is “denial”. For example, as shown in FIG. 4, the semantic operation table 9 takes a semantic classification of a word located relatively before a predicate in a row and a semantic classification of a word located relatively behind a predicate in a column. , Which describes the semantic classification represented by two words at a position orthogonal to a row and a column. For example, if the combination of "affirmation" and "assertion", "affirmation"
Is stored as a list.
For example, as shown in FIG. 5, the semantic storage unit 10 is for storing the semantic classification searched for operation, and in the illustrated example, the searched semantic classification “No” is stored.

In such a configuration, for the Japanese sentence input from the input unit 5, the adverb detection unit 2 detects the adverb using the adverb co-occurrence dictionary 7, and searches the semantic classification of the co-occurring predicate.
Next, the predicate semantic operation unit 3 searches for a word from the end of the sentence, and extracts words constituting the predicate and their semantic classification using the predicate semantic dictionary 8. When a predicate consists of more than one word,
The semantic classification is checked using the semantic operation table 9, and the semantic classification of the predicate is determined while the semantic classification is stored in the semantic storage unit 10. Thereafter, it is determined whether or not there is an error from the semantic classification co-occurring with the adverb and the semantic classification of the predicate. That is, the predicate discriminating unit 4 checks whether the semantic classification of the predicate co-occurs with the adverb, and if it does not co-occur with the adverb, the predicate is displayed as an error.

FIG. 6 shows an outline of such processing in the error detection processing section 1.

FIG. 7 shows the processing in the predicate semantic operation unit 3. This processing proceeds from the end of the sentence to the front. First, the part-of-speech and semantic classification of the word at the end of the sentence are examined using the predicate semantic dictionary 8. If the result is not an independent word, a semantic classification is extracted. At this time, if the semantic classification is stored in the semantic storage unit 10, the semantic classification in the semantic storage unit 10 and the semantic classification of the word being processed are checked using the semantic operation table 9, and the result is checked. Stored in 10 and proceed forward. If the semantic classification is not stored in the semantic storage unit 10, the searched semantic classification is stored in the semantic storage unit 10,
Advances processing forward. Such processing is repeated for the word of the predicate, and when the word becomes an independent word, the semantic classification of the independent word is extracted. At this time, if a word is stored in the meaning storage unit 10, the meaning classification of the independent word and the meaning classification in the meaning storage unit 10 are checked using the meaning operation table 9. Meaning storage unit 10
If no word is included in, the semantic classification of the independent word is used as the semantic classification of the predicate.

Now, a specific processing example will be described with reference to the flowcharts of FIGS. 6 and 7.

Example 1 First, the presence of the adverb "karakishi" in the sentence can be found by the processing of the adverb detection unit 2 using the adverb co-occurrence dictionary 7. Next, the process proceeds from the end of the sentence. The end of the sentence is the predicate semantic dictionary 8
Indicates that the auxiliary verb is "da", and this is not an independent word, so extracting the semantic classification results in "asserted" (see FIG. 3). At this time, since nothing has been registered in the meaning storage unit 10 yet, the meaning information “asserted” is stored in the meaning storage unit 10. Then, the process proceeds to the processing of the word “no use” immediately before. "No" is determined to be an adjective verb in the predicate meaning dictionary 8, and its meaning classification "negative" is extracted. At this time, since the semantic classification “assertion” is stored in the semantic storage unit 10, when the semantic classification “denial” of “no” and the semantic classification “assertive” of “da” are checked in the semantic operation table 9, “ No "(see FIG. 4). Therefore, since the semantic classification co-occurring with the adverb "karakishi" is "denial", the predicates co-occur and are determined to be error-free.

Example 2 First, the presence of the adverb "karakishi" in the sentence can be found by the processing of the adverb detection unit 2 using the adverb co-occurrence dictionary 7. Next, the process proceeds from the end of the sentence. The end of the sentence is known from the semantic dictionary 8 of the predicate to be an auxiliary verb "no". Since this is not an independent word, extracting the semantic classification results in "negative" (third.
See figure). At this time, since nothing has been registered in the meaning storage unit 10 yet, the meaning classification information “No” is stored in the meaning storage unit 10. Then, the process proceeds to the processing of the word “de” immediately before. Since "de" is not an independent word, the semantic classification "assertion" is extracted. At this time, since the semantic classification “deny” is stored in the semantic storage unit 10, the semantic classification “deny” of “de” and the semantic classification “deny” of “none” are checked in the semantic operation table 9, No "(see FIG. 4). This meaning classification “negation” is stored in the meaning storage unit 10. And
The process proceeds to the “nodame” immediately before that, but since the “nodame” is an independent word, its meaning classification “negation” is extracted.
At this time, since the semantic classification "denial" is stored in the semantic storage unit 10, the semantic operation table 9 is searched using this and the semantic classification "denial" of "no". Then, it turns out that it expresses the semantic classification “affirmation”, and it does not match the co-occurrence semantic classification of the adverb “karakishi”. Thus, the adverb and the predicate are displayed as errors.

Example 3 First, the adverb co-occurrence dictionary 7 indicates that there is an adverb "slightly" in the sentence. Therefore, processing is performed for the end of the sentence.
The end of the sentence is identified by the predicate semantic dictionary 8 as an auxiliary verb "no". Since this is not an independent word, its meaning classification is extracted as "negative". At this time, since nothing has been registered in the meaning storage unit 10, the meaning classification "negative" is stored, and the process proceeds to the immediately preceding word "de". Since "de" is not an independent word, its semantic classification "assertion" is extracted. At this time, since there is “negation” in the semantic storage unit 10, the semantic operation table 10 is searched for the “negation” and “assertion”, and the semantic classification “negation” by the combination is stored. Then, the process proceeds to the immediately preceding "Japanese". Since "Japanese" is an independent word, its semantic classification "general noun" is extracted. At this time,
Since the semantic storage unit 10 has the semantic classification "negative", the semantic operation table 9 is searched based on this and the semantic classification "general noun" of "Japanese". Then, it can be seen that the combination represents the semantic classification “noun negation” and does not match the co-occurrence semantic classification of the adverb “slightly”. Therefore, the adverb and the predicate are displayed as errors.

Example 4 First, the adverb co-occurrence dictionary 7 indicates that there is an adverb "Zenzen" in the sentence. When processing is performed on the end of the sentence, the end of the sentence is identified by the predicate semantic dictionary 8 as the adjective "beautiful". Since this is an independent word, the semantic classification "adjective" is extracted, and the semantic storage unit 10 has the semantic classification. Find out if. Here, since the predicate is not stored yet, the meaning of the predicate is set to “adjective”, and it is checked whether or not the predicate co-occurs with the adverb “at all”. Since these do not co-occur, the adverb and the predicate are displayed as errors.

As described above, the present invention provides an adverb co-occurrence dictionary that stores an adverb and a semantic classification of a predicate that co-occurs with the adverb, a predicate semantic dictionary that stores a semantic classification of the words that constitute the predicate, and A semantic operation table storing semantic classifications expressed by a combination of semantic classifications of preceding and succeeding words, and a semantic storage unit storing a searched semantic classification, and an adverb detecting unit connected to the adverb co-occurrence dictionary An error detection processing unit including a predicate semantic operation unit and a predicate determination unit connected to a predicate semantic dictionary, a semantic operation table, and a semantic storage unit, and forms a predicate expression in an input Japanese sentence The semantic classification of all the words to be performed is calculated by the predicate semantic operation unit using the predicate semantic dictionary, the semantic operation table, and the semantic storage unit. Whether to co-occur with adverbs extracted by the adverb detector using a dictionary Since the unmatch between the adverb and the predicate expression is detected by the predicate determination unit, the predicate expression co-occurring with the adverb is not detected only by the last unit of the predicate.
Since the semantic classification of all the words constituting the predicate is also checked and the semantic classifications are calculated, the accuracy of error detection can be improved.

[Brief description of the drawings]

1 is a block diagram, FIG. 2 is a configuration diagram of an adverb co-occurrence dictionary, FIG. 3 is a configuration diagram of a predicate semantic dictionary, and FIG. 4 is a semantic operation. FIG. 5 is a block diagram of the meaning storage unit, FIG. 6 is a schematic flowchart of the whole, and FIG. 7 is a flowchart showing the processing of the predicate meaning calculation unit. 1 ... Error detection processing unit, 2 ... Adverb detection unit, 3 ... Predicate semantic calculation unit, 4 ... Predicate determination unit, 7 ... Adverb co-occurrence dictionary,
8 predicate semantic dictionary 9 semantic operation table 10 semantic storage

Claims (1)

(57) [Claims] 1. An adverb co-occurrence dictionary storing an adverb and a semantic classification of a predicate co-occurring with the adverb, a predicate semantic dictionary storing a semantic classification of a word constituting the predicate, and the meaning of preceding and following words A semantic operation table storing semantic classifications expressed by combinations of classifications, a semantic storage unit storing the searched semantic classifications, an adverb detecting unit connected to the adverb co-occurrence dictionary, and a semantic dictionary of the preceding description unit. An error detection processing unit including a predicate semantic operation unit and a predicate determination unit connected to the semantic operation table and the semantic storage unit, and all the words constituting the predicate expression in the input Japanese sentence Each semantic classification is calculated by the preceding description part semantic operation unit using the preceding description part semantic dictionary, the semantic operation table and the meaning storage unit,
Using the adverb co-occurrence dictionary to determine whether the adverb extracted by the adverb detection unit co-occurs in the semantic classification of the entire predicate expression calculated by the adverb detection unit, determines whether the adverb and the predicate expression correspond to each other. An error detection device, wherein the error is detected by a determination unit.