JPH0469768A - Document editing support method - Google Patents

Abstract

PURPOSE:To detect an error related to a grammatical rule and restrictions for a syntax, and to present its amendment by analyzing an input sentence, and thereafter, comparing result a result of analysis and the grammatical rule including a verb type and finding out an error. CONSTITUTION:A dictionary related to a word and an idiom, a first file in which how to use correctly words and phrases is stored, and a second file in which how to use erroneously words and phrases and a comment corresponding to how to use it erroneously are stored are provided in advance, and a sentence is inputted, a part of speech is given to each word of the sentence by using a dictionary, and a sentence style of the sentence is analyzed, based on a list of the parts of speech given to the sentence. In such a state, when words and phrases used in the sentence correspond to one of how to use erroneously stored in a second file, an erroneous part of the inputted sentence corresponding to the corresponding how to use erroneously is displayed, and in accordance with the displayed erroneous part, a comment corresponding to the corresponding to how to use erroneously is outputted. In such a way, an error related to a grammatical rule and restrictions for a syntax is detected, and its amendment is presented.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to a method for supporting document elaboration.

[Conventional technology]

One of the conventional technologies that supports document elaboration is an English syntax checker based on rHPSG,” Information Processing Society of Japan Research Group Report,
Natural Language Processing Research Group Material 722 (May 3, 1989-
9 El) discussed on pages 1 to 8. This is one of the syntactic analysis techniques, HPSG (Il
ead-drj-ven Rhrase Struct
urCGrammar) parser is used to continue parsing even if some features (pronounced as ``sousei'', whose meaning is the syntactical and semantic characteristics of the words) are mismatched between words ( This method is characterized by identifying the word that caused the mismatch.This method intentionally loosens some of the grammatical rules and syntactic constraints from the beginning.Therefore, it allows the input sentence to be corrected. This method has the advantage that it can be parsed only once, compared to the more traditional method of relaxing these rules and constraints when the syntax cannot be parsed.

[Problem to be solved by the invention]

- In the prior art, if a sentence containing an error regarding rules or constraints that cannot be relaxed is input during syntax analysis, the syntax analysis fails and elaboration is impossible. In particular, rules and constraints regarding verb types (the order of the parts of speech of the subsequent word dominated by a verb) are essential for syntactic analysis, and errors cannot be diagnosed. The reason for this is that the HII)SG cannot hold information about what caused the parsing to fail. This - as an example,
The above literature states that it is not possible to make an appropriate diagnosis regarding errors caused by confusion between intransitive verbs and transitive verbs.

The object of the present invention is to provide an elaboration support method that detects errors related to grammatical rules and syntactic constraints and presents correction plans.

[Means to solve the problem]

In order to achieve the above object, the present invention employs a method of detecting errors by comparing the analysis results with grammar rules including verb types after analyzing an input sentence. The analysis of the input sentence in the present invention will be referred to as sentence pattern analysis to distinguish it from the syntactic analysis used in the prior art. In other words, sentence pattern analysis is defined as finding the subject and predicate verb in an input sentence, and then finding the verb type from the sequence of parts of speech of the word following the predicate verb.

More specifically, the technical means will be described below.

(1) Focus on the predicate verb of the input sentence and perform sentence pattern analysis.

Namely. Find the subject corresponding to this predicate verb and the sequence of parts of speech (verb type) of the subsequent word dominated by this predicate verb.

(2) Detect errors in word usage by comparing word usage patterns such as verb types obtained through sentence pattern analysis with correct word usage patterns or incorrect word usage patterns that have been created in advance. do.

(3) If an error is detected, search for revised messages that have been classified in advance based on grammatical rules and word constraints, and output the corresponding revised message.

In addition, in realizing the above (1) for English texts, the present invention particularly devises the following points.

(a) When it cannot be determined whether a verb is used as a predicate or a past participle based on the spelling, these can be identified from the pattern of the parts of speech before and after the verb.

(b) When there are multiple subject candidates corresponding to the predicate,
Scores are determined in advance according to the arrangement pattern of the parts of speech before and after the subject candidate, and the combined H1 score of the subject candidate in the input sentence is calculated based on these scores. As a result, the subject candidate with the highest total score is selected as the subject.

[Effect]

In the present invention, each word of the read input sentence is looked up in a dictionary and morphologically analyzed. When a word is on the multi-item side,
A part-of-speech selection rule that preferentially selects one part of speech from a sequence of parts of speech is provided to eliminate multiple parts of speech. If it is not possible to narrow down the parts of speech to one even when using part of speech selection rules, it is possible to assign one part of speech to each word by selecting the part of speech that is used most frequently.

After selecting one part of speech for each word, we focus on the predicate and separate the sentence immediately after the predicate in order to find the subject that corresponds to the predicate. If a predicate and a participle cannot be distinguished by morphological analysis (orthographically), the predicate can be found by using a rule that determines the predicate or participle from the sequence of parts of speech. Next, in each of the above categories, find the subject that corresponds to the predicate. If there are multiple subject candidates, the subject can be found by using an elimination rule that finds candidates with a low probability of becoming the subject from the arrangement of parts of speech. Note that for embedded sentences in which the related side is omitted, the interval A1 determined by a certain predicate v1
, not only the subject S1 of the predicate V, but also the next interval AI+
The predicate Vl+, of □ may contain the subject S,+. For such a subject SI+1, it is possible to deal with embedded sentences by treating it as the subject of v1+□ when all subject candidates of ■ and +□ are removed by the elimination rule in the subsequent interval A1+. becomes. After determining the subject and predicate in this way, the sentence type is determined from the sequence of parts of speech following the predicate. This makes it possible to clarify how the words in the input sentence are used.

The usage of words found from the sentence analysis results (for example, the sequence of parts of speech dominated by the words) is compared with the correct usage of words created in advance. When these usages match, it is determined that the usage of the word in the input sentence is correct.

This determination prevents the common belief that the usage of a phrase is incorrect even though it is correct. When there is a discrepancy in the usage of a word, compare it to patterns in which the word is often used incorrectly. When this pattern matches, it is determined that the word usage in the input sentence is incorrect. This determination can prevent "overlooking" errors in the usage of words. When the usage of a word in the input sentence does not match both the correct and incorrect usage of a word created in advance, the correctness or wrongness of the usage of this word is unknown, and a message to that effect is output.

〔Example〕

A first embodiment of the present invention will be described below with reference to FIG.

In the following embodiments, a method for supporting the revision of a document created in English will be mainly explained.

In step 10, an input sentence to be refined is read.

In step 20, dictionary 1.100 (see FIG. 4) is consulted to determine the part of speech for each word in the input sentence. Generally, English words are on the multi-component side, so in step 20, processing for eliminating the multi-component side is also performed. The details will be described later using FIG. 3. After assigning one part of speech to each word of the input sentence, in step 30, the sentence pattern is analyzed from the arrangement of these parts of speech. When performing this analysis, the processes of determining the predicate verb, determining the subject corresponding to this predicate, and determining the verb type are executed. The details will be described later using FIG. 6. In step 40, it is determined whether the usage of the phrase in the analyzed sentence pattern is the correct usage of the phrase created in advance. In this embodiment, the type (verb type) in which the verb dominates the subsequent part of speech is correctly registered for each word in advance in Table 1, . 500 (see FIG. 14) and the verb type of the input sentence (see column 1010 in FIG. 2) obtained from the result of the sentence pattern analysis in step 3o. Note that the verb type used in the input sentence is simply called the verb type, and is predefined in Table 1.50.
The correct verb type to be registered as 0 is sometimes called the Hornby verb type. Hornby's verb types can be found in one document, ``English Patterns and Usage (2nd Edition)'' (Oxford Publishing, 1977).
7) and is widely known throughout the world.

To explain step 40 in detail, take the following input sentence as an example.

“We are proceeding, ng 5ever
al 5tepsrequjredby the br
ethren,” The sentence pattern analysis result (step 30) of this input sentence is stored in the sentence pattern analysis table 1000. Sentence pattern analysis table 1
000 is a multidimensional array, as shown in FIG. The array number column 1o01 determines the horizontal size of the table 1001, and the number of arrays may be greater than the expected number of words in one sentence. For example, ]-oo is generally sufficient. The input text field 1002 is entered in step 1.
(1) This is an area where words and symbols are stored in order from the first word of the input sentence to the last symbol of the sentence, corresponding to the array number. The part-of-speech code field 1003 stores the part-of-speech code of each word in step 20 in the dictionary 1.1. This is an area where the part-of-speech code field 11-03 in OO is searched and the results are stored. To explain in more detail, step 2o, as shown in FIG. Store in.

Step 201 searches the headword column ]-1, O]- of the dictionary 1100, taking into account the inflection of the ending of the word (see FIG. 4). For example, the word “proce” in the example sentence above
edjng" in the dictionary]]-00 headword field 1]o]-
When searching with , a symbol N representing a noun is obtained from the part-of-speech column 1103 as a part-of-speech code. On the other hand, it is well known that one of the inflections of English words is "jng".

Therefore, in step 201, the headword column 11 is rewritten with "proceed" with the inflection JngI+ removed.
Search o1. As a result, a symbol V representing a verb is obtained from the part of speech column 11o3. However, to indicate that itjngn has been removed, in step 201, V N G is selected as the part of speech code. N and VNG obtained in this way are stored in the corresponding position of "proceedj ng" in the part-of-speech code column 1003 in the array element number 3) of the array element number 1I11001 in step 20]-. In addition, the original spelling from which spellings indicating inflection in the ending have been removed is stored immediately after the part-of-speech code.

In the example of “proceedjng”, part of speech code field 10
"V NG (pro
ceecl)" is stored (see FIG. 2). In this way, it can be read that the original spelling of "proceeding" when used as VNG is "proceed."

Note that the original spelling is not shown immediately after the part-of-speech code N of "proceedjng", but this is the dictionary 11.
When subtracting “proceedj, ng” with 00 (4th
(see figure), this is because nothing is entered in the original form column 1102. On the other hand, "N S (brether)" is stored in the position of array number 8 of the part-of-speech code [1003] as the part-of-speech code for "brethren" (see FIG. 2). This is Dictionary 1100"
brethren” (see Figure 4),
“B r o t h e r ” in the original form column 1102
This is due to the fact that it is filled in. As mentioned in section 2,
The original spelling is stored in the part-of-speech code field 1103.
These are words that have been registered in the original form column 1102 of the dictionary 1100 from the beginning as words with inflectional endings peculiar to English words or irregularly inflected words.

The part-of-speech selection column 1004 is an area for storing the result of the multi-part side elimination process in step 202. Step 202 is
Refer to the part-of-speech selection rule table 1.200 shown in FIG. This table], 200 is the rule activation word W1
201. It consists of a rule end word column J202 and a part of speech selection rule column 1203. Here, the rule starting word is a word that starts the application of the part of speech selection rule stored in the part of speech selection rule column 1203, and the rule ending word is the word that ends the application of the part of speech selection rule. (This is a word that becomes a rigger. In step 202, the part-of-speech code stored in the part-of-speech code ff41003 and the word stored in the input sentence field 1002 are searched in order of arrangement, and these part-of-speech codes and words are used as rule activation words. Column 1101-
Determine whether the content matches the content stored in . If there is a match, select one part of speech by applying the part of speech selection rule until the rule end word is encountered. If they do not match, the most frequently used part of speech is selected from among the parts of speech written in the speech book 11100. Dictionary 1100 part of speech column 1]
If we decide to list the parts of speech in order of frequency of use in 03, the part of speech that is used most frequently will be the part of speech listed first in the part of speech column 1103 (hereinafter referred to as the first part of speech). ) should be selected. To repeat, selecting the first part of speech is equivalent to selecting the most frequently used part of speech.

Here, we will summarize and explain the main part-of-speech codes used in the following explanation. S is the part of speech code that represents the subject, BE is the verb be, N is the singular noun, NS is the plural noun, ■ is the base form of the verb, ■S is the third person singular present tense of the verb, VD is the past tense of the verb, VN is the verb The past participle form of , VNG is the verb in
g form, A D J is (I5) adjective base form, ADV is wood tone base form, ■) is preposition, A R
T represents an article, PR and N represent pronouns, and CN J represents a conjunction. When using other part-of-speech codes, we will explain each time.

Now, the multi-product side elimination process in step 202 will be specifically explained using an example. In step 202, first, the part-of-speech code ``'S'' of array number 1 in the part-of-speech code column 1103 is read, and the part-of-speech code ``'S'' of array number 1 is read in
Search whether 1L S It is included in 1. As a result, in the fifth place in the same table 1200, rr S I
It can be seen that there is t (see Figure 6). It can be read from the rule end word column 1202 that this rule is applied to the immediately following word. However, in this example, the part-of-speech code of the word immediately after 'S' is found to be only "BE" from the part-of-speech code column 1003, and since it does not have a multi-part-of-speech code, this rule is not applied and the next array number is Read the part-of-speech code from the part-of-speech code column], OO3.That is, in step 202, read the part-of-speech code from the part-of-speech code column 1., 103, array number 1.
Read the part-of-speech code It B E 11 for -, and enter the part-of-speech (1
G) In the rule activation word column 1201 of selection table]-200
``BE'' is included.As a result, it can be seen that the second rule in the same table has a rule related to It B E It (see Figure 6).Part of speech selection rule @12
According to the second rule of 03, the part-of-speech code of "proceediB", which is the word immediately after "BE", is "V
NG” is selected and stored in the corresponding position of part of speech selection [1004. This rule ends with the word “proceeding” immediately after “BE” according to the rule end word column 1202. For the next word “5everal” Since there is no rule to apply in the part-of-speech selection table: +-200, the 11th part of speech in the part-of-speech code field 1003 is "A D J
" is selected. Similarly, in 11stepsI+, "N S''' is selected according to the first rule of the part-of-speech selection rule table 1200, and "required" is selected according to the rule No. 3 L1 of the part-of-speech selection rule table 1200.
T+ is selected and stored in the corresponding position of the part of speech selection column 10o4. Note that if there is only one part-of-speech code in the part-of-speech code field 1003, that code is stored as is in the part-of-speech selection field 1004 in step 202.

In step 202, information on the idiom $41104 in the dictionary 1100 is further utilized to resolve multi-part speech.

Even if a word has multiple parts of speech, the part-of-speech code can sometimes be ignored if the word is one of the components of an idiom. For example, in Bunchu Hisakata, there is a VD called “granted”.
, VN. granted
For example, the idiom "take...for gran"
When used as part of "ted", the only information required for sentence pattern analysis is "for" with P and CN J, and "granted" with VD and VN.
” can be ignored, i.e., step 20
In step 2, check the headword field 110 of dictionary 1-100 to see that there is an idiom "take-for granted" in the sentence.
1 and phrase column 11.04 (see Figure 5). Since the part of speech code 1 for this idiom or the part of speech column of the idiom column 1.104 is only written as "take (V)", give the part of speech code ■ only to "take" and write LLfO.
For the part-of-speech codes of rI+ and "granted", a symbol to be ignored, for example, ¥O, is stored in the corresponding position of the part-of-speech code field 10o3. Instead of ignoring the part-of-speech code, "take", , in the idiom element column 1505 of the sentence pattern analysis table.
The same numbers indicating that they are the same phrase components are stored in the corresponding positions of ITOrI+, 11g,, anl, and cd+1. This number is used to show which word corresponds to which compound word when there are multiple compound words in one sentence.For example, you can give numbers starting from 1 in the order in which the compound word is found. Bye.

In addition, in step 202, when a plurality of part-of-speech selection rules act on a certain word, a priority order of application is set among the rules. Since this priority has nothing to do with the essence of the present invention, the explanation will be omitted.The explanation of step 20 is now complete, and next, the sentence pattern analysis process of step 30 will be explained. The detailed diagram of step 30 is shown in the 7th page.
As shown in the figure. Step 301 creates a verb phrase 2 name phrase and a prepositional phrase, and
Store it in the corresponding column of 07. The verb phrases are A, UX (auxiliary verb) +V, BE+VNG, etc., and the noun phrases are ART1N, ADJ1N, ART1N+N, etc. The prepositional phrase is a P ten noun phrase. The symbol for the phrase is the verb phrase V.

They are a noun phrase NP and a prepositional phrase PP. The same symbol as the part-of-speech code is used as the phrase symbol, such as S representing the subject and VN representing the past participle, respectively. Also, NP and NIP for noun-word 2 pronoun-word and verb-word, respectively.

We will use VP's phrase symbols. When using phrase symbols other than these, explain each time. The phrase elements are stored in the phrase structure element column of the sentence pattern analysis table 1000], OO5. The same digits are stored in elements of the same phrase. In this way, even if there are multiple phrases, it is possible to read which word is a component of which phrase. In step 302, it is determined whether there is a "predicate participle" in the input sentence. A predicative participle is a word in which it is difficult to tell whether it is a predicate verb or a past participle. As mentioned above, in sentence pattern analysis, the analysis focuses on predicate verbs, so determining predicate participles is important. The predicate verb (V,
When it cannot be determined whether the word is a past participle (VS, VD) or a past participle (VN), the symbol VX is entered in the corresponding position in the part-of-speech selection column 1004. In this way, in step 302, the predicate participle can be easily determined by checking whether vX is present in the part-of-speech selection column 1o04. In step 303, if there is a predicate participle, VP or VN is selected according to the predicate participle determination rule and stored in the phrase structure column 1-007. The predicate participle determination rules are as shown in FIG.
-300 is stored in advance. Predicate participle order column 1301
is an area that stores the arrangement of phrase symbols before and after VX. The judgment condition column 1-302 is an area for storing conditions held by VX, and the 1 judgment rule column 1303 is an area for storing rules for determining whether it is a predicate or a participle.

In step 303, it is determined whether the pattern of the arrangement of phrasal symbols before and after vX in the input sentence matches the pattern stored in the predicate participle arrangement column 1301, and if they match,
It is determined whether the predicate participle is a predicate or a participle according to the description in the judgment condition column 1-302 and the judgment rule column] 303. If the predicate participle cannot be determined even if the predicate participle determination rule is used, one or the other can be decided. Also, steps 303a to 30 in FIG.
As shown in 3d, the user can also give instructions.

Step 304 focuses on the VP stored in the phrase structure column 1007 and stores a number in the position corresponding to the vP in the sentence classification column OO8. This number is multiple V
When there is P, 1, 2, 3, etc. are selected in order from the closest to the beginning of the sentence. All you have to do is assign a number like... This has the meaning of dividing sentences by focusing on vP.

In step 305, subject candidates corresponding to the predicate verb are found in each segment into which the sentence is divided. As subject candidates, NP, VNG, TV (to-infinitive phrase), etc. can be selected. Subject candidates can be easily found by searching phrase structure column 1007, and the results are stored in subject candidate analysis table 3000, described immediately below. In step 306, a subject is selected according to the subject candidate selection rule, and the symbol S and classification number N of the subject are stored in the subject and predicate column 1009 of the sentence analysis table 1-000. The subject candidate selection rules are stored in a subject candidate selection table 1400, as shown in FIG. The subject candidate condition section (X,) 14.01 of the same table 1400 is an area that stores the sequence of executive phrases of subject candidates.

The processing unit (Y) 1402 is an area that also stores the contents of the subject candidate selection process. In step 306, the phrase structure N1
The sequence of subject candidate phrases of o07 is the subject candidate condition part (X,
) column 1401, and if so, the processing unit (Y) column, i-4,
Execute the process stored in 02. That is, I
Processing may be performed according to the format F (X), T II E N (Y). Specifically, the content of the process to be executed is to use the subject candidate analysis table 3000 to calculate the score of each subject candidate. As shown in Figure 1, the subject candidate analysis table 3000 is
001゜Subject candidate array number 3002. Score column 3003
Consisting of The total score obtained as a result of the processing contents stored in the processing unit (Y) column 1402 is displayed in the score column 300.
It is stored in 3. In this way, in each category, the one with the highest score is selected as the subject from among the subject candidates with a score of O or less. If there is no subject that satisfies the above subject conditions in that category, then the subject candidate in the category before that category,
Choose something that is not the subject. This is to go beyond division and deal with linguistic phenomena in which subjects and predicate verbs correspond. Even if the subject candidate selection rule is used, if the i: word candidates cannot be narrowed down to one, the subject candidate located closest to the predicate verb can be selected in step 30G. Further, as shown in steps 306a to 306g in FIG. 1-2, the user can also specify a subject or a subject and a predicate.

Step 301 described above-1-. A specific example of the series of processes from to 306 will be explained using sentence pattern analysis table 10.00 in FIG. The input sentences used in this explanation are as follows.

”A good book read when on
e was a child can 1nfulenc
e one through one life.

The input sentence is stored in the input sentence column (step 10).

The analysis result of step 20 is stored in the part-of-speech selection column]-〇〇4. Here, II readl+ has part-of-speech codes V, Vl), and VN, but since the part-of-speech selection rule cannot determine any of these, the predicate participle symbol vX
is stored. In step 301, “A goo
d book" is stored as N I) in the phrase structure column 1007. Similarly, PR,N is stored as N P "a chi
ld" to N P, can jnflucnce" to V
It is stored as p. Other part-of-speech symbols are stored as phrase symbols in the phrase structure field. It is found in step 302 that there is a predicate participle, and a determination is made in step 303. In step 303, a search is made to find that the phrase symbol sequence in the phrase structure field 1007 matches the third position in the predicate participle sequence column 130 of the predicate participle determination rule table 1300 shown in FIG. According to the conditions and the rules in the determination rule field 1303, Llread++ is determined to be a participle, and as a result, VN is determined as "" in the phrase structure field 1007.
f'ead". Step 30
4, 1 is stored in the position of classification column 1008 corresponding to phrase symbol BE, and 2 is stored in the position of classification column 1008 corresponding to phrase symbol VP. In step 305, two subject candidates, NP corresponding to array number 2 and NP corresponding to array number 5, are found as subject candidates for category J, and the results are used in the subject candidate analysis table shown in FIG. 3000, is stored in the subject candidate array number column 3002. Similarly, for section 2, NP of array array number 8 is stored. In step 306, subject candidate selection rule table]-40
According to the fifth rule of 0, the NP with array number 5 has a score of 1.
give.

Similarly, according to the rule at the top of the table 1400, a score (-2) is given to Nl) of array number 8, and these results are stored in the score column 3003. Further step 305
Then, from among the subject candidates of category 1, the one with the highest score of 0 or more is selected as the NP (word "
On the other hand, the subject candidate for category 2 is (-2).
Since only those with a score of Select NP 2 (word 1+book+!).

This result is stored in the corresponding position of the subject and predicate column]-009. In other words, as can be seen from Figure 12,
A symbol S2 indicating that it is the subject of category 2 is stored in the position corresponding to ``ook'', and a symbol S2 indicating that it is the subject of category 1 is stored in the position corresponding to 11O1e'' of array number 5.
1 is stored. Similarly, for predicate verbs, Vl and V2 are stored corresponding to the classification.

The above steps 301- to 306 have shown that the subject and predicate can be selected from the input sentence.

In step 307, the verb type is determined from the sequence of parts of speech following the predicate verb, and the verb type is stored in the sentence pattern analysis table 1000. That is, in step 307, phrase structure column 1
It is determined whether the sequence of phrase symbols following vp 007 matches the verb type stored in the verb correct usage type table 1500 shown in FIG. The table 15oO consists of a verb type 1-501-, a composition field 1502, and a verb type 1503 by Hornby. The verb type 15o1 is an area for storing names given to grammatically permissible verb types. For the name, Alphabella I- may be used, for example, A, B, C, etc. Configuration column 1-
502 is stored in the verb type 15o1. A list of parts of speech corresponding to verb type names is stored. Hornby's verb type 15o3 stores Hornby's verb type corresponding to the arrangement of parts of speech in the composition column 1502. Note that vP used in Hornby's verb type is an acronym for verppatt; ern, and does not mean a verb phrase (vervpattern). Further, the description of the verb-type constituent elements stored in the composition field 1502 is not necessarily based on parts of speech, but may use symbols representing phrases. For example, the description of verb type B is
In the diagram, it looks like "verb ten preposition + (pronoun) noun", but this can be called "verb phrase ten prepositional phrase" or rVP + P
It may also be written as PJ.

The names obtained by analyzing in step 307 are the sentence pattern analysis table], the verb type in the input sentence of OOO]010
is stored in The process of step 307 above will be explained using the example shown in FIG.

In step 307, first, the verb type after the verb phrase ■P in the phrase structure column 1007 is rVP+NPJrVP+NP.
Search for +VNJ or rVP+NP+PPJ. This is because it is usually sufficient to consider the number of verb types as 2 or 3 as the number of phrases. The last verb type r
In VP+NP+PPJ, VN located next to NP
The reason for dropping this is to take into consideration the fact that when this VN modifies the immediately preceding NP, rNP+VNJ is often regarded as an NP as a whole. Therefore, the verb form]−
By performing pattern matching with the stored contents of 502, the verb types A, , C are obtained, and this value is stored in the input verb type 1010. Just one,
Due to space constraints, the verb type in Figure 1-4 is
502 is not shown.

This completes the explanation of the sentence pattern analysis process in step 30, and next step 40 will be explained. In step 40, it is determined whether the usage of the word/phrase found from the result of the sentence pattern analysis is the correct usage of the word/phrase created in advance. That is, the verb having the verb type stored in the verb type 1010 in the input sentence is entered in the part-of-speech code field 1003 or in the input sentence field 1002.
Search from.

To describe in more detail, first, using the array number where the verb type stored in the verb type temperature 1010 is located as a search key,
A part-of-speech symbol v** (** is any character string including a space) having the same array number as the search key in the part-of-speech code field 1003 is found, and immediately after that, it is determined whether the original form of the verb is stored in parentheses. If it is stored, it is the original form of the desired action; if it is not stored, it is the input text field 10.
The word with the same array number as the search key in 02 is the original form of the desired verb. Next, the correct verb type of the original form of this verb is searched using the verb correct usage type table ]-600 shown in FIG. In the example in FIG. 2, from the part-of-speech code column 1003, the base form of the verb is
ed” is searched and the headword column 16 of the table 1600 is searched.
By searching for "proceed" in 01, the verb type [correct verb type of "proceed" stored in 1602, VP2A, VP3A].

You can search for VF6. In step 40, the verb form [On]5
Search the verb type 1503 by Hornby from the verb types stored in 01, and if the correct verb type found earlier is included in the search results for verb type 1503, the verb type in the input sentence is correct. If this is not the case, it is determined that the verb type in the input sentence contains an error. In the example in Figure 2, "proceed" in the input sentence
The verb type of is A, ni, C, which is VP6A,6
B.

VF24. A, 24B, 24. The verb type 11503 can be searched to correspond to C, VF14. On the other hand, the correct verb type is VP2A according to the verb type in table 1600.
It can be seen that they are VP3A,,VF6. Since there is nothing in common between the verb type in the input sentence and the correct verb type, "proceed" is determined to be used incorrectly.

Next, step 50, which is executed when there is an error in the usage of a phrase, will be explained. In step 50, it is determined whether the usage of the phrase found from the result of the sentence pattern analysis is an incorrect usage of a previously created phrase. Step 50 will be explained in more detail using the example of "proceed" described above. Step 50 uses the verb type stored in the verb type 1010 in the input sentence as a search key, and searches the verb incorrect usage type table 1.70.
701 is searched. If the search is successful, the verb type 702 of the verb is searched, and it is determined whether the verb type stored in this column 1702 is included in the correct verb types found earlier. do. If it is included, it is determined that there is an error in the word or phrase in the input sentence, and the process proceeds to step 60. Otherwise, proceed to step 60. In the example of FIG. 2, it is already known from the verb type [010] in the input sentence that one of the verb types of "proceed" used in the input sentence is A. Therefore, for this A, the verb type i in the verb incorrect usage type table 1700! Search j 1.702, V
Obtain P3A.

“proceed” is already a valid verb type in VP3A
Therefore, it is determined that the use of "proceed" is incorrect, and the process proceeds to step 60.

In step 60, message column 1 of table 1700
The message number stored in 704 is retrieved and stored in the corresponding position of the message number column of the sentence pattern analysis table 1000, and the message number stored in the message column 17 is stored in the output device 100, which will be described later.
04 message contents are output.

On the other hand, in step 70, a message is output to the effect that it cannot be determined whether the words used in the input sentence are clearly correct or incorrect.

FIG. 17 shows an example of the screen immediately before displaying the revised message on the output device 100.
1-8 shows an example in which the user views the content and makes corrections in the editor area at the bottom of the screen.

In this example (first example), a method for supporting the elaboration of verb types has been described. It is clear that elaboration support for verb types (sequences) can also be supported using a method similar to that for verb types.

The hardware configuration for realizing this embodiment is described in the 19th section.
As shown in the figure. The input sentence is input from the input device 80, the elaboration support process is executed by the CI) U 90, and the elaboration result is displayed by the output device 100. The storage device 150 is roughly divided into an analysis table 110 and an analysis rule table 1.
20, a usage dictionary 130, a dictionary 3-100, and an area for storing a revision program 140. Second
FIG. 0 shows that the analysis table 11-O includes a sentence pattern analysis table 100 and a subject candidate analysis table 3000. FIG. 21 shows the analysis rule table 120.
Part of speech selection rule table 1.200.

Predicate participle determination rule table 1300. This indicates that a subject candidate selection rule table 1400 is stored. FIG. 22 shows a verb type table 150. in the usage dictionary. Table of correct usage of verbs], 600.

A verb incorrect usage type table 1-800 is included.

In addition, article incorrect usage type table 1800. Subject-predicate match incorrect usage type table 1900. The subject-predicate match exception table is also included here.

A method for supporting article elaboration, which is a second embodiment of the present invention, will be described below with reference to FIG. In FIG. 23, step 1-1 to step 3]- is the step 10 in FIG.
- Since it is the same as step 30, the explanation will be omitted. To explain step 41, take as an example the following input sentence containing an article error.

“It: hank you, nded for
], ove1. y andpl, easant ca
l, endar, ``Step 31 for this input sentence
Sentence analysis table 1 which is the output result of . OOO second
Shown in Figure 4.

Regarding this sentence pattern analysis table 1000, the second
Since the explanation has been made using figures, the explanation will be omitted here.

In step 4]-, it is determined whether the usage of the phrase found from the result of the sentence pattern analysis is an incorrect usage of a previously created phrase. In this embodiment, the article elaboration support method is concerned, so the part-of-speech selection column of the sentence pattern analysis table 1000 ] 004
Search for part-of-speech codes N and NS. N or NS
When searching for , input text column +002 is searched using the array number as a search key. The resulting word is
This becomes the target word for elaboration. The article type of this word to be revised is searched for in the article type column 1.105 of the dictionary]1-00 shown in FIG. Note that the article type column 1105 is added after the dictionary 1-], OO's idiom column], , ], 04 shown in FIG. 4 for elaboration of articles. The article type is a group name of a group created by classifying nouns based on their attributes. We considered three attributes of nouns: singular, plural, and uncountable, and classified nouns based on which of these attributes a noun can have.

The relationship between article type and noun classification is as follows.

Article type 1: Has all three attributes.

Article type 2: Does not have an attribute that is treated as uncountable.

Article type 3: Does not have attributes that can be treated as plural. It has two other attributes.

Article type 4: Does not have attributes that are treated as Qi numbers. It has two other attributes.

Article type 5: Has only the attribute of being treated as a number of shares.

Article type 6: Only has attributes that are treated as plural.

Article type 7: Only has attributes that are treated as uncountable.

The principles for adding articles to singular nouns, plural nouns, and uncountable nouns are as follows.

(1) A singular countable noun can be accompanied by an indefinite article, but cannot be used without an article.

(2) Plural countable nouns can be used without articles, but
It cannot be accompanied by an indefinite article.

(3) Uncountable nouns can be used without an article, but cannot be accompanied by an indefinite article.

When classified from a different perspective than how to attach articles, that is, from the perspective of inflection and number, uncountable nouns do not have a plural form and are
It differs from a plural countable name in that the verbs treated as t numbers correspond to each other. Countable nouns are used in the singular form and correspond to singular verbs.

Now, Step 4] - refers to the incorrect article usage type table 1800, so first, this table, +-80
The configuration of 0 will be explained. Article type [1801 stores the article type, ilt number handling column]-802 stores the symbol X for the article type that cannot be treated as singular. .

Similarly, the plural column] 803 stores X in article types that cannot be treated as plural, and the uncountable column 1804 stores X in article types that cannot be treated as uncountable.

The message number column 1805 is a sentence pattern analysis table],
The message number corresponding to message content column]-806 of 000 is stored. In step 41, a noun with article type n (n=j to 7) is placed in column 1-7 in the input sentence.
If it is used in the article type of
Message number column 1011 of sentence pattern analysis table 1000
and outputs the message content in the message content column 1806 to the output device 1-00.

In the example shown in FIG. 24, the part-of-speech code field 1003 is searched to find that the array number storing N is 8, and the input sentence field 10o2 is searched.
Search for the singular "calendar" in array number 8, and use this as the word to be refined.

In order to find the article type in which ゞ'calendar'' is used in the input sentence, depending on whether N or NS is stored in the part-of-speech selection field 1004, it can be either (singular countable or uncountable) or plural. You can see that it is handled well.

Next, the phrase structure element field 1006 is searched for phrase elements that include nouns, and if there is no article among the elements, they are treated as uncountable.
If there is an indefinite article, it is determined to be countable. In this way, the type in which the noun is used (columns 1802 to 1-804.
), and on the other hand, the article type (lines 1 to 7 of column 1-801) pulled from dictionary 1.000 is determined, and Table 1
If the symbol X is stored at the intersection of the column and row 800, there is an error in either the article 2 inflection or the number, so the message in the message column 1806 is output.

In the example of FIG. 24, it can be seen from the part-of-speech selection field 1004 and the phrase structure element field that the number of words is treated as uncountable. On the other hand, from the dictionary 1000, the article type of "calendar" is 2.
, and when the position where column 1804 and row 2 intersect is detected, it is found that the symbol X is stored. As a result, in step 61, an elaboration message saying "This type of noun cannot be used without an article" is output to the output device 100.

Next, the third example regarding subject-predicate agreement will be explained in the second example.
This will be explained according to Figure 7. Steps 12 to 32 are the same as steps 10 to 30 in FIG.
The explanation will be omitted here.

In step 62, the subject and predicate columns], OO9 of the sentence pattern analysis table 1000 are searched to find the subject and predicate. Next, select part of speech column 1. Search for OO4 and determine the number of subjects and predicates from the part-of-speech code.

For example, when the subject is N and the predicate is ■, it can be determined that the subject is singular and the predicate is third person plural present. Next, this subject-predicate pair is
900, it is determined whether or not the subject type field 1-901 is equal to the pair of the incorrectly used predicate field 1902. If they are not equal, it is determined that there is no error, and if they are equal, the process proceeds to step 82. The pair of N and ■ is shown in Table 19.
00 number] - can be determined to be an error. but,
Even if an error is determined in this table 1900, there are exceptional cases where it is not an error. In order to find this exception, in step 82, reference is made to the subject-predicate match exception table 2000, which stores the exception conditions of the table 1900. This table 2000 consists of a message number column 2001 that stores the numbers of messages with exceptions, an exception subject column 2002, and a message content column 2003 that stores messages that take exception conditions into consideration. In step 82, it is determined whether the error in the main axis and the predicate matches the exception condition, and if it matches, the message content column 2
The contents of 003 are output as comments 1 to 1 in step 70. If they do not match, the message in the message number column is output in step 62 as before. Step 92
Now, the error is corrected using the input device (old) device 80.

Next, a fourth embodiment will be described. The embodiments described so far have common steps 10 to 30 in FIG. Therefore, as shown in FIG. 30, if the common parts are grouped together and the steps corresponding to the details of the redaction process 53 to 93 are structured in parallel, the user can specify the redaction process in step 3. In this way, for example, it is possible to execute only the verb-type elaboration process 53 on all input sentences (step 5).

〔Effect of the invention〕

According to the present invention, it is possible to detect errors related to grammatical rules and syntax constraints and present correction plans, so it is possible to refine the created document and create a high-quality document.

[Brief explanation of the drawing]

Figure 1 shows the first embodiment of the present invention, Figure 3 is a sentence pattern analysis table, Figure 3 is a detailed diagram of step 20, and Figure 4 shows the first embodiment of the present invention.
5 shows the structure of the dictionary, FIG. 6 is a part-of-speech selection rule table, FIG. 7 is a detailed diagram of step 30, FIG. 8 is a predicate participle determination rule table, and FIG. 9 is step 30.
3, FIG. 10 is a subject candidate selection rule table, FIG. 1-1 is a subject candidate analysis table, FIG. 12 is a detailed diagram of step 306, and FIG. 13 is an example of a sentence pattern analysis table. The figure is a verb type table, Figure 15 is a table of correct verb usage, Figures 1 and 6 are tables of incorrect verb usage, and Figures 17 and 18 are Fig. 19 shows the hardware configuration of the embodiment, Fig. 20 shows the structure of the analysis table, Fig. 21 shows the structure of the analysis rule table, Fig. 22 shows the structure of the usage dictionary, and Fig. 23 shows the structure of the usage dictionary.
The figure shows the second embodiment, and Figure 24 shows the sentence pattern analysis table.
For example, Figure 25 is a table of incorrect usage of articles, Figure 26 is an example of a dictionary, Figure 27 is the third example, Figure 28 is a table of incorrect usage of subjects and predicates, and Figure 29 is a table of incorrect usage of articles. The figure shows an exception table for matching the subject and predicate, and FIG. 3o shows the fourth embodiment. 140...Elaboration program group, 1], O...Analysis table group, 120...Analysis rule table group, 13
0 Usage dictionary group, 1 ], OO Dictionary, 80... Input device, No. ? Ka〃Emuno3θ1 /3ρ2 /3ρ3 ・)゛bar〉d -medzu1 and lru jealousy diagram 3θθθ ■ ward first badai

Claims (1)

[Claims] 1. A dictionary for words and phrases, a first file that stores correct usage of words, and a second file that stores incorrect usage of words and comments corresponding to the incorrect usage. inputting a sentence, assigning a part of speech to each word of the sentence using the dictionary, analyzing the sentence pattern of the sentence based on the arrangement of the parts of speech assigned to the sentence, and determining the sentence type used in the sentence. If the word/phrase corresponds to any of the incorrect usages stored in the second file, display the erroneous part of the input sentence corresponding to the corresponding incorrect usage; A document editing method characterized in that the comment corresponding to the corresponding incorrect usage is output in correspondence with the displayed error part. 2. A document editing method according to claim 1, characterized in that the input sentence is displayed. 3. In the document editing method according to claim 2, the display of the error portion displays the error portion in the displayed inputted sentence in a manner different from the others. Elaboration method. 4. In the document editing method according to claim 1, if the words used in the sentence correspond to any of the correct usages of the words stored in the first file, then the A document editing method characterized by repeating the process of inputting sentences. 5. In the document editing method according to claim 1, the word or phrase used in the sentence does not correspond to any of the incorrect usages stored in the second file, and 1. If the document does not correspond to any of the correct usages stored in the file No. 1, a predetermined comment is output. 6. In the document editing method according to claim 1, further inputting correction contents corresponding to the error part of the displayed sentence, and correcting the input sentence according to the inputted correction contents. A document editing method characterized by: 7. The document editing method according to claim 6, further comprising repeating the process of adding the part of speech. 8. The document elaboration method according to claim 1, wherein the input of the sentence is input from a storage device that stores the document to be elaborated. 9. The document elaboration method according to claim 1, wherein the step of analyzing the sentence includes the step of having the user input at least one of a subject and a predicate.